Shifting tasks from pharmacy to non-pharmacy personnel for providing antiretroviral therapy to people living with HIV: a systematic review and meta-analysis

Objectives Lay people or non-pharmacy health workers with training could dispense antiretroviral therapy (ART) in resource-constrained countries, freeing up time for pharmacists to focus on more technical tasks. We assessed the effectiveness of such task-shifting in low-income and middle-income countries. Method We conducted comprehensive searches of peer-reviewed and grey literature. Two authors independently screened search outputs, selected controlled trials, extracted data and resolved discrepancies by consensus. We performed random-effects meta-analysis and assessed certainty of evidence using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach. Results Three studies with 1993 participants met the inclusion criteria, including two cluster trials conducted in Kenya and Uganda and an individually randomised trial conducted in Brazil. We found very low certainty evidence regarding mortality due to the low number of events. Therefore, we are uncertain whether there is a true increase in mortality as the effect size suggests, or a reduction in mortality between pharmacy and non-pharmacy models of dispensing ART (risk ratio (RR) 1.86, 95% CI 0.44 to 7.95, n=1993, three trials, very low certainty evidence). There may be no differences between pharmacy and non-pharmacy models of dispensing ART on virological failure (risk ratio (RR) 0.92, 95% CI 0.73 to 1.15, n=1993, three trials, low certainty evidence) and loss to follow-up (RR 1.13, 95% CI 0.68 to 1.91, n=1993. three trials, low certainty evidence). We found some evidence that costs may be reduced for the patient and health system when task-shifting is undertaken. Conclusions The low certainty regarding the evidence implies a high likelihood that further research may find the effects of the intervention to be substantially different from our findings. If resource-constrained countries decide to shift ART dispensing and distribution from pharmacy to non-pharmacy personnel, this should be accompanied by robust monitoring and impact evaluation.

both at patient and population levels 3 4 . Scale up of ART in low and middle-income 98 countries has averted more than 5 million deaths; however, bottlenecks preventing 99 universal access to ART still exist. One challenge is the critical shortage of human 100 resources for health (HRH), including for delivery of essential pharmacy services. 101 As for pharmacists, although the World Health Organization (WHO) recommends a 102 minimum of one pharmacist per 2300 population 5  shortage, it is likely that there is an uneven distribution of pharmacists in such 105 settings, as is the case with other specialist healthcare workers who tend to 106 concentrate in urban areas and the private sector, further aggravating the HRH 107 shortage. For instance in South Africa, which is home to the largest number of PLHIV 108 in any country worldwide, in 2010 only 24% of registered pharmacists worked in the 109 public sector where 80% of the population received care 6 . 110

112
Studies and programmes report that involvement of pharmacy personnel in HIV care 113 results in improved patient outcomes 7 . For instance, in the USA, the use of a 114 multidisciplinary team approach with pharmacists assuming a central role in the 115 initiation, dispensing and adherence counselling improved treatment outcomes such as 116 viral load, patient retention and medication adherence 8 . 117 The work of pharmacists includes supply management, dispensing and distributing 118 medications, promoting adherence, identifying and preventing potential medication-119 related issues (such as over dosage, sub-therapeutic dosage, adverse drug reactions, 120 medication errors and untreated indications), and monitoring and reporting adverse 121 drug events 9 10 . In some settings, programmes have implemented alternative models 122 of pharmacy services that shift selected tasks from pharmacy to non-pharmacy 123 personnel. Such alternative models could potentially increase the number of health 124 workers involved in ART distribution, adherence counselling and patient education; 125 free more time for pharmacy personnel; support the integration of ART in primary 126 care settings; minimise the number of facility visits for ART collection; and, reduce 127 pharmacy queue waiting times for patients. 128 The specifics of shifting ART-related tasks from pharmacy personnel have not been 129 addressed in a systematic review. Available systematic reviews on task shifting focus 130 on clinical services where nurses and non-physician clinicians provide care 131 comparable to physicians 11 . We therefore synthesised the evidence for task shifting in 132 pharmacy personnel services, where non-pharmacy personnel undertake ART 133 distribution and medication adherence counselling. Table 1 provides definitions for 134 the pharmacy functions distribution and dispensing as they have been used in this 135 review. 136 Table 1: Definitions and descriptions of dispensing and distribution of 137 medications 138

Term Definition
Dispensing • Dispensing is a controlled act that authorises one to select, prepare and provide stock medication that has been prescribed to a patient or client (or his/her representative) for administration at a later time [12].
• To dispense is to prepare and supply to a patient a course of therapy on the basis of a prescription [11].
• Dispensing is preparation and distribution of a course of therapy to a patient, with appropriate instructions based on a prescription [11].

Distribution
• At facility level, drug distribution mainly refers to drugs dispensed by licensed practitioners, such as nurses, doctors, pharmacists or pharmacy assistants and collected by a patient.
• At community level, drug distribution refers to trained lay people collecting pre-packed medications from the facility and delivering them to HIV patients in the community 12 .

139
How task shifting using non-pharmacy personnel might 140 work 141 Within the last decade, several high HIV burden countries adopted task shifting 142 strategies where nurses and non-physician clinicians initiate and maintain ART. 143 Although this has undeniably expanded access to ART, it is also increasingly essential 144 when non-health professionals (lay people) delivered ART at the community level 13 . 149 In Mozambique, the use of PLHIV for distributing ART, monitoring adherence, 150 reporting outcomes and referring sick patients to health facilities yielded a retention 151 rate of 97.5% among stable patients on ART 13 . In a cluster randomised trial in 152 Uganda, the use of community health workers produced comparable results with the 153 facility-based ART programme in terms of patient retention, viral load suppression 154 and mortality rate 14 . Similar findings were also obtained in Kenya and some other 155 parts of Uganda when lay providers were engaged in ART delivery 14 15 . 156 Task shifting has therefore been seen as an achievable solution to the critical human 157 resource shortages affecting scale up of ART 16 . While it is imperative to increase the 158 rate of recruitment and training of health workers as well as improve working 159 conditions to reduce attrition and emigration, the HIV pandemic requires that all 160 possible options are considered to address the critical skills shortage 17 . Such 161 measures may include shifting selected tasks from pharmacy to non-pharmacy 162 personnel, particularly for patients stable on their ART. Task-shifting allows more 163 time for pharmacy personnel to focus on more technical functions such as supply 164 management, pharmacovigilance and patient consultation. 165 166 Why this review is important 167 168 Dependence on and shortage of pharmacists are key constraints on ART expansion, 169 but the specifics of task shifting for ART distribution from pharmacy to non-170 The aim of this review was to evaluate the efficacy and safety of shifting pharmacy 179 related tasks, including ART distribution and adherence assessment, from pharmacy 180 to non-pharmacy personnel. We used appropriate Medical Subject Heading (MeSH) terms, relevant keywords, and 215 validated terms for identifying reports of randomised controlled trials 19 . We provide 216 the details of the search strategy for each electronic database in Table 2. We also 217 searched conference abstract archives on the websites of the Conference on 218   Two review authors (NMM and OA) independently assessed the risk of bias in each 267 study using the Cochrane risk of bias tool which measures risk of bias in controlled 268 trials across seven domains: sequence generation, allocation concealment, blinding of 269 participants and personnel, blinding of outcome assessors, completeness of outcome 270 data, selective outcome reporting and other potential biases 19 . 271

Sequence generation (checking for selection bias) 272
We considered a study as having a low risk of selection bias due to sequence 273 generation if investigators described a random component in the sequence-generation 274 process, such as the use of random number tables; coin tossing, card or envelope 275 shuffling. We classified a study as having a high risk of bias if investigators described 276 we classified the study as having unclear risk of bias. 280

Allocation concealment (checking for selection bias) 281
We classified the study as having a low risk due to allocation concealment if 282 participants and investigators enrolling participants could not foresee assignment (e.g. 283 central allocation; or sequentially numbered, opaque, sealed envelopes). A study was 284 classified as having a high risk of selection bias if participants and investigators could 285 foresee assignment (e.g. an open, random-allocation schedule; a list of non-random 286 numbers; or use of unsealed, non-opaque, or not sequentially numbered envelopes. 287 We assigned a study as having an unclear risk of selection bias when the study 288 reported insufficient information to permit judgment of the allocation concealment or 289 the methods used to conceal allocation were not described. 290

Blinding (checking for performance and detection bias) 291
We determined that a study was at low risk of performance bias when there was 292 blinding of the participants and key study personnel, and it was unlikely that the 293 blinding could have been broken or there was no blinding in the situation but non-294 blinding was unlikely to introduce bias. A study with no blinding or incomplete 295 blinding when the outcome was likely to be influenced by lack of blinding was 296 classified as having a high risk of performance bias. In the absence of sufficient 297 information to permit judgment of adequacy of the blinding of participants and 298 investigators, the study was classified as having unclear risk of performance bias. 299 Regarding detection bias, we determined that a study was at low risk of bias when 300 there was blinding of outcome assessors. We classified a study as having a high risk 301  16 of detection bias if there was no blinding or the outcome was likely to be influenced 302 by lack of blinding of outcome assessors. In the absence of sufficient information to 303 permit judgment of adequacy of the blinding of outcome assessment, the study was 304 classified as having unclear risk of detection bias. 305 Incomplete outcome data (checking for attrition bias) 306 We classified a study as having a low risk of attrition bias if there was no missing 307 outcome data, reasons for missing outcome data were unlikely to be related to the 308 outcome, or the proportion of missing outcome data was balanced across groups. We 309 classified a study as having a high risk of attrition bias if there was imbalance in 310 proportions lost to follow-up across groups and the reason for missing outcome data 311 was likely to be related to the outcome. When there was insufficient reporting of 312 attrition or exclusions, we classified the study as having an unclear risk of attrition 313 bias. 314

Selective reporting (checking for reporting bias) 315
We determined a study as having a low risk of reporting bias if a protocol was 316 available and primary outcomes in the final trial report corresponded closely to those 317 presented in the protocol. High risk of selective reporting was considered to be 318 present when the primary outcomes differed between the protocol and final trial 319 report. When no trial protocol was available or there was insufficient reporting to 320 determine if selective reporting was present, we classified the study as having an 321 unclear risk of reporting bias. 322

Checking for other forms of bias 323
We classified a study as having a low risk of other forms of bias, when there was no 324 evidence of bias from other sources. A study was classified as having a high risk from 325 other sources when potential bias was present from other sources (e.g. early stopping 326 Measures of effect 330 We calculated and presented summary statistics for the risk ratio (RR) for 331 dichotomous outcomes and the weighted-mean difference for continuous outcomes, 332 with their 95% confidence intervals (CI). 333

Unit of analysis issues
The unit of analysis was the individual study participant for all the trials. 336 Dealing with missing data 337 338 We contacted study authors for one of the included trials to obtain information on the 339 composition of health professionals in the control groups involved in dispensing ART 340 in order to establish whether pharmacy personnel were part of the team. Unfortunately 341 the author could not remember details but assumed that health professionals also 342 included pharmacy personnel. 343

345
We used the I 2 statistic to measure heterogeneity among the trials. We planned to 346 explore substantial heterogeneity (I 2 >50%) by pre-specified subgroup analysis. 347 However, there was no evidence of serious heterogeneity in our review hence we did 348 not perform sub-group analyses. 349

351
We contacted study authors to provide any missing outcome data. However, only one 352 author from one included trial responded but did not provide any additional 353  18 information. If any meta-analysis in our review was to include 10 or more studies, we 354 planned to assess the potential for publication bias using a funnel plot 19 20 . We 355 minimised the potential for publication bias by using a comprehensive search strategy, 356 as described above. 357

359
We conducted meta-analysis, using Cochrane's Review Manager software 21 . We used 360 a random effects model considering the diverse settings of the included studies. 361 Where meta-analysis was not possible, for instance, for adherence and cost, a 362 narrative synthesis of the evidence was carried out. We summarised the quality (or 363 certainty) of evidence for each outcome using GRADE 22 23 . For GRADE we 364 independently considered the five factors for downgrading the evidence to assess the 365 certainty of a body of evidence as it relates to the studies that contribute data to the 366 meta-analyses for the pre-specified outcomes. The quality rating across studies has 367 four levels: high, moderate, low or very low. Randomised trials are considered to be 368 of high quality but can be downgraded for any of the following five reasons: risk of 369 bias; indirectness of evidence; unexplained heterogeneity or inconsistency of results; 370 imprecision of results; and, high probability of publication bias. Similarly, 371 observational studies are considered to be of low quality, but can be upgraded for any 372 of these three reasons: large magnitude of effect; all plausible confounding would 373 reduce a demonstrated effect; and, the presence of a dose-response gradient. 374

376
Due to the absence of significant heterogeneity, we did not perform subgroup 377 analyses. Additionally, sensitivity analyses were not conducted to investigate the 378

381
A total of 3557 records were identified. Following title and abstract screening, eight 382 studies were identified for full-text screening. The full-text review resulted in three 383 studies 15 24 25 that met the review inclusion criteria. The study selection process is 384 shown in Figure 1. The characteristics of included and excluded studies are shown in 385 Tables 3 and 4 respectively. In total, 1993 HIV-infected patients on ART were 386 enrolled in the three included studies. Of these, 1121 were assigned to non-pharmacy 387 personnel model of ART delivery while 872 were assigned to pharmacy personnel 388 model of ART delivery. However, 974 and 746 were available for analysis in the non-389 pharmacy personnel and pharmacy personnel groups respectively for all the three 390 studies. Overall, there were no differences in loss to follow up between non-pharmacy 391 and pharmacy personnel groups (3% versus 3%) p=0. 9. Two studies 15 , 24 included lay 392 people in the non-pharmacy personnel group as shown in Table 3 while the third 25 393 included nurses in the non-pharmacy personnel group.    401 We judged the three trials to have a low risk of selection bias as a result of adequate 402 randomisation and allocation concealment. The studies by Selke et al 15

and Jaffar et 403
al 24 were judged to have low risk of performance bias for blinding because the 404 reported outcomes were objective and unlikely to be influenced by lack of blinding. 405 However, the Selke trial 15 did not give sufficient information to permit judgment of 406 whether or not detection bias was present. The Silveira trial 25 was judged to have a 407 high risk of performance and detection bias due to lack of blinding for participants, 408 personnel and outcome assessment respectively for all study outcomes. The three 409 studies were judged to have a low risk of attrition bias as they did not have differential 410 or large numbers of losses to follow-up across the intervention arms. The risk of 411 selective reporting bias was judged to be low for the study by Jaffar et al 24

418
There was one reported death in the non-pharmacy personnel group in the Selke trial 419 15 during the study period. However, the authors reported the death as unrelated to 420 HIV but rather due to use of herbal abortifacients. The trial by Jaffar 24 in Uganda did 421 not find differences in mortality rates between the non-pharmacy and pharmacy 422 personnel groups: adjusted rate ratio RR 0.95 (95% CI 0.71 to 1.28). Five deaths were 423 reported in the non-pharmacy personnel group compared to none in the pharmacy 424 personnel group in the Silveira trial 25 . This difference was not statistically significant 425 (RR 11.00, 95% CI 0.61 to 197.35). When these data were pooled as shown in Figure  426 3, there was no difference in mortality between the groups cared for by non-pharmacy 427 personnel and those cared for by pharmacy personnel (RR 1.86; 95% CI 0.44 to 7.95), 428 with no significant heterogeneity (Chi 2 = 3.34; df = 2; p = 0.19; I 2 = 40%). The 429 certainty of the evidence for this outcome was low due to serious indirectness and 430 imprecision as shown in Table 5. We downgraded for indirectness because two of the 431 three trials (Selke 15 and Jaffar 24 ) compared complex interventions that included, but 432 were not limited to, pharmacy and non-pharmacy personnel as shown in Table 5. 433 There were few deaths in the trials and the effect estimate was imprecise, with wide 434 confidence intervals ranging from appreciable benefit to substantial harm as reported 435 in Table 5.  In a meta-analysis, we detected no differences in virological failure between the group 466 cared for by non-pharmacy personnel and that cared for by pharmacy personnel (RR 467 0.92; 95% CI: 0.73 to 1.15), with no significant heterogeneity detected between the 468 trials (Chi 2 = 0.24; df = 2; p=0.89; I 2 = 0%) as shown in Figure 4. The quality of the 469 evidence for this outcome was low, due to serious indirectness and imprecision as 470 shown in Table 5. 471

473
Selke 15 reported no differences in median duration (days) in the study for both groups 474 at six or 12 months (343 days for both groups at 12 months), p = 0.2. At study 475 closure, the proportion lost to follow up was similar between the two groups (5.2% for 476 the non-pharmacy personnel group and 4.5% for the pharmacy personnel group), p = 477 1.0. Jaffar 24 reported very low rates of loss to follow up (1% in the non-pharmacy 478 versus 2% in the pharmacy personnel group), however, 3% in the non-pharmacy 479 personnel group and 6% in the pharmacy personnel group respectively withdrew from 480 participation. Similarly, in the Silveira trial 25 , 4.8% (n = 8) participants were lost to 481 follow up in the non-pharmacy personnel group and 3% (n = 5) in the pharmacy 482 personnel group. Overall, there were no differences in loss to follow up between the 483 non-pharmacy personnel and the pharmacy personnel group with both having a 3% 484 loss to follow up, p=0.96. 485 In our meta-analysis, we detected no difference in loss to follow up between the 486 groups (RR 1.13; 95% CI: 0.68 to 1.91), with no significant heterogeneity (Chi 2 = 487 0.54; df = 2; p=0.76; I 2 = 0%) as shown in Figure 4. The quality of the evidence for 488 this outcome was low due to serious indirectness and imprecision as shown in Table  489 5. 490 Selke 15 reported significantly more clinic visits in the pharmacy compared to the 493 non-pharmacy personnel group (mean visits 12.6 versus 6.4), p < 0.001. Despite 494 fewer clinic visits for the non-pharmacy personnel group, clinical outcomes between 495 the two groups were comparable with an incidence of 13.6 HIV opportunistic 496 infections per 100 person-years in the non-pharmacy personnel group and 19.8 HIV 497 opportunistic infections per 100 person-years in the pharmacy personnel group, p = 498 0.42. Although the non-pharmacy personnel group was found to have fewer clinic 499 visits, the authors observed that this group undertook 64% more clinic visits than 500 originally scheduled. The study by Jaffar 24 found a high frequency of outpatient 501 attendance (15 242 visits) in the pharmacy compared to the non-pharmacy personnel 502 group (6691 visits). However, they found similar distribution of new diagnoses 503 between groups where more than 50% were infectious and parasitic infections. The 504 trial by Silveira 25 did not report on clinic visits. 505

507
Both groups in the Selke trial 15 demonstrated high levels of self-reported medication 508 adherence with no statistical differences between groups at six and 12 months (96% 509 versus 97%), p = 0.71 and (94% versus 97%), p = 0.47 respectively. In the Jaffar trial 510 24 , self-reported adherence at routine clinical and counselling reviews was high with 511 94% in the intervention group and 91% in the control group. In the Silveira trial 25 , 512 participants' self-reported adherence remained the same between the two groups after 513 controlling for potential confounders RR 1.05, 95% CI 0.95-1. 15, p=0.35. 514 Cost 515 In the trial by Jaffar 24 , a societal perspective economical analysis showed a higher 516 mean cost per patient per year in the pharmacy personnel group ($838) for health 517  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   30   services (staff, transport, drugs, laboratory and clinical services, sensitisation, training  518 and workshops, utilities, supervision and overhead capital) compared to non-519 pharmacy personnel group ($793). Similarly, total per patient costs to access care per 520 year were higher in the pharmacy personnel group ($54) compared to non-pharmacy 521 personnel group ($18). Costs to access care included transport, lunch, childcare and 522 lost work time. The outcome cost was not pre-specified in the protocol, however, this 523 provides additional data that may be relevant for decision makers and was therefore 524 included for consideration. 525 Acceptability to pharmacy personnel, non-pharmacy personnel and 526 patients and harm, including error rates 527 528 These outcomes were not reported in the included trials. 529 530

533
Two cluster randomised clinical trials and one non-blinded randomised controlled 534 trial were included in this review of effects of shifting responsibility from pharmacy 535 to non-pharmacy personnel for adherence assessment and dispensing antiretroviral 536 drugs to HIV-infected patients. 537 In our meta-analysis, we found low quality evidence that there may probably be no 538 difference in mortality, virological response, and loss to follow up between non-539 pharmacy personnel and pharmacy personnel distributing ART to patients. In 540 addition, we did not observe significant differences in adherence to treatment. Selke et 541 al 15 reports that the non-pharmacy personnel group had significantly fewer all-cause 542  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   31   sick visits to the clinic compared to the pharmacy personnel group. Although the main  543   difference between the Selke et al 15 and Jaffar et al 24 was the level of education of  544 the non-pharmacy personnel, it did not seem to affect their overall performance 15 24 . 545 Additionally, the Jaffar trial 24 found the non-pharmacy dispensing strategy to be 546 cost-effective and cheaper to run than the pharmacy dispensing strategy by almost 45 547 US dollars per patient per annum and this accounted for only 6% of the total cost of 548 healthcare service expenditure for the intervention group 24 . The patients who were 549 accessing their medications and other healthcare services from pharmacy personnel 550 incurred more costs in terms of transportation, lunch, childcare costs and lost work 551 time. In the first year of the study, the pharmacy personnel group incurred double the 552 cost of healthcare services per patient per annum compared to the non-pharmacy 553 personnel group. This is a challenge in most poor settings in Africa where many 554 cannot afford basic necessities and this further impedes their access to treatment 24 . 555 Lack of or inadequate financial resources are major factors in late presentation to 556 health facilities, poor access to HIV care and support, and low retention in care after 557 initiation of ART 31 32 . Although health services are provided for free in most public 558 institutions, the cost of accessing the services may serve as a hindrance to achieving 559 optimal care and support for HIV-infected patients 31 . The non-pharmacy personnel 560 dispensing ART could therefore considerably reduce the cost of services for both 561 patients and governments with potentially favourable clinical outcomes. 562 Although issues related to drug-drug interaction, medication errors, pre-screening for 563 ART and some special cases of adherence to therapy may be beyond the capacity of 564 non-pharmacy personnel, targeted pharmacy support to non-pharmacy personnel 565 could offer benefits to patients having challenges with their medications especially 566 those who are taking concomitant medications with ART 33 . 567  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y   32 This review and meta-analysis has shown that the use of non-pharmacy personnel 568 comprising either lay people or other health professionals such as nurses who are 569 given well-tailored and comprehensive short trainings with refreshers could 570 effectively support ART dispensing systems for HIV-infected patients. The use of 571 non-pharmacy personnel such as lay people who are known and trusted members of 572 the community in dispensing ART and monitoring patients' health would enable 573 identification of vital psychosocial features that might be overlooked by physicians, 574 nurses and pharmacists at the facility. Some of these psychosocial problems such as 575 gender-based violence, food insecurity and alcohol abuse negatively impact on 576 adherence and retention in care 15 . In the case of community-based non-pharmacy 577 personnel, they could also serve as a link between the patients and other healthcare 578 workers and facilitate smooth communication between the patients and the 579 pharmacists on issues such as changes in regimen and dosage of drugs 15 33 . Although it may not be possible to rule out publication and language biases in 584 systematic reviews and that this could potentially threaten the conduct of this review, 585 our trial search was not restricted to any language, setting or publication status. We 586 are of the opinion that we identified all the existing randomised controlled trials 587 relevant to our research question. However, there is a possibility that some 588 unpublished or published trials were not accessible by our search. 589 We identified three studies for inclusion in this review. The trials generally had small 590 sample sizes. Recruitment restrictions such as basing recruitment on location of 591 residence of the participants resulted in a small sample size in the Selke trial 15 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60   F  o  r  p  e  e  r  r  e  v  i  e  w  o  n  l  y 33 trials did not include some of the outcomes of interest for decision-making such as 593 acceptability to participants and feasibility of the intervention, for example, adherence 594 to treatment. 595

597
We assessed the certainty or quality of evidence using the Grading of 598 Recommendations Assessment, Development and Evaluation (GRADE) approach 34 , 599 which defines the certainty of evidence for each outcome as "the extent of our 600 confidence that the estimates of effect are correct" 35 . Evidence from this review 601 should be applied with caution considering its low quality. Three outcomes in this 602 review were found to be of low quality. We are therefore not confident enough to 603 state that the estimates lie close to the true value and it is possible that different and 604 further studies are needed to contribute to the body of knowledge to increase our 605 certainty in these review findings. 606

608
There is low-quality evidence that the use of non-pharmacy personnel including lay 609 people in dispensing and distribution of ART may be as effective as pharmacy 610 personnel performing these functions in resource-limited settings when accompanied 611 by continued capacity building and supervision by trained pharmacy personnel. The 612 introduction of non-pharmacy personnel is probably cost-effective for the patient and 613 health system. Strong referral systems are crucial in task-shifting of dispensing 614 responsibilities to non-pharmacy personnel to support patients that require advanced 615 medical attention. Due to the critical shortage of human personnel in the health sector, 616 most HIV programmes engaged other cadres of health personnel other than doctors to 617 prescribe and dispense ART at the onset of their programme implementation. This 618  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59                8 Eligibility criteria 6 Specify study characteristics (e.g., PICOS, length of follow-up) and report characteristics (e.g., years considered, language, publication status) used as criteria for eligibility, giving rationale.

8-9
Information sources 7 Describe all information sources (e.g., databases with dates of coverage, contact with study authors to identify additional studies) in the search and date last searched.

9
Search 8 Present full electronic search strategy for at least one database, including any limits used, such that it could be repeated.

10-11
Study selection 9 State the process for selecting studies (i.e., screening, eligibility, included in systematic review, and, if applicable, included in the meta-analysis).

12
Data collection process 10 Describe method of data extraction from reports (e.g., piloted forms, independently, in duplicate) and any processes for obtaining and confirming data from investigators.

12-13
Data items 11 List and define all variables for which data were sought (e.g., PICOS, funding sources) and any assumptions and simplifications made.

12-13
Risk of bias in individual studies 12 Describe methods used for assessing risk of bias of individual studies (including specification of whether this was done at the study or outcome level), and how this information is to be used in any data synthesis.

16
Synthesis of results 14 Describe the methods of handling data and combining results of studies, if done, including measures of consistency (e.g., I 2 ) for each meta-analysis.

Study selection
17 Give numbers of studies screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally with a flow diagram.

17-18
Study characteristics 18 For each study, present characteristics for which data were extracted (e.g., study size, PICOS, follow-up period) and provide the citations.

19-23
Risk of bias within studies 19 Present data on risk of bias of each study and, if available, any outcome level assessment (see item 12).

23-24
Results of individual studies 20 For all outcomes considered (benefits or harms), present, for each study: (a) simple summary data for each intervention group (b) effect estimates and confidence intervals, ideally with a forest plot.

24-29
Synthesis of results 21 Present results of each meta-analysis done, including confidence intervals and measures of consistency.

24-28
Risk of bias across studies 22 Present results of any assessment of risk of bias across studies (see Item 15).
n/a DISCUSSION Summary of evidence 24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers).

29-31
Limitations 25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias).

32
Conclusions 26 Provide a general interpretation of the results in the context of other evidence, and implications for future research.   (ART) tasks from specialised healthcare workers to those with shorter or less formal training has been implemented in resource-limited settings to alleviate critical shortages of human resources for health. However, the specifics of shifting ART dispensing from pharmacy to non-pharmacy personnel have not been addressed in a systematic review, although this can potentially increase access to ART. We will assess the effects of shifting dispensing and distribution of ART and adherence assessment from pharmacy to nonpharmacy personnel in low and middle-income countries.

31-32
Methods and analysis: We will search PubMed, CENTRAL, EMBASE, WHO Global Health Library and relevant grey literature for eligible controlled trials. Two authors will screen the search output, select eligible studies, assess risk of bias and extract data from included studies, resolving discrepancies by discussion and consensus. We will perform meta-analysis using both fixed and random effects models, investigate clinical and statistical heterogeneity, and assess our confidence in the overall evidence using standard Cochrane methods, including GRADE.
Ethics and dissemination: Only secondary data will be included in this review and ethics approval is not required. We will disseminate the review findings in various scientific fora, including peer-reviewed journals. The findings may help to inform policy makers in defining the scope of work of healthcare workers, and global recommendations for shifting the dispensing and distribution of ART from pharmacy to non-pharmacy personnel.

INTRODUCTION
Description of the condition By March 2015, 15 million (40.7%) of the estimated 36.9 million people living with HIV (PLHIV) globally were receiving antiretroviral therapy (ART). 1 Combination ART is effective for reducing HIV related morbidity and mortality as well as preventing HIV transmission. 2 Initiating ART early in the course of HIV infection has been associated with better health outcomes, both at patient and population levels. 3 4 Scale-up of ART in low and middle income countries has averted more than 5 million deaths; however, bottlenecks preventing universal access to ART still exist. One challenge is the critical shortage of human resources for health (HRH), including for the delivery of essential HIV related pharmacy services. The WHO recommends a minimum of one pharmacist per 2300 population, but most countries in low-resource settings such as sub-Saharan Africa have not yet met this target. 5 In addition to the absolute shortage, it is likely that there is an uneven distribution of pharmacists in such settings, as is the case with other specialist healthcare workers who Strengths and limitations of this study ▪ To our knowledge, this is the first published protocol of a systematic review that will investigate the effects of task shifting from pharmacy to non-pharmacy personnel for dispensing or distributing antiretroviral therapy to patients living with HIV. ▪ The protocol was written according to the PRISMA-P (Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols) recommendations. ▪ The review findings may help to inform antiretroviral therapy guidelines by the WHO. ▪ The possible weakness of the planned review would be the limitations of included studies, for example, high risk of bias and heterogeneity of settings, designs and effects. Description of the intervention Studies and programme reports indicate that involvement of pharmacy personnel in HIV care results in improved patient outcomes. For instance, in the USA, the use of a multidisciplinary team approach with pharmacists assuming a central role in ART initiation, dispensing and adherence counselling improved treatment outcomes such as viral load, patient retention and medication adherence. 8 The work of pharmacists includes supply management, dispensing and distributing medications, promoting adherence, identifying and preventing potential medication-related issues, and monitoring and reporting adverse events. In some settings, programmes have implemented alternative models of pharmacy services that shift selected tasks from pharmacy to non-pharmacy personnel. Such alternative models could potentially increase the number of health workers involved in ART distribution, adherence counselling and patient education, free more time for pharmacy personnel, support the integration of ART in primary care settings, minimise the number of facility visits for ART collection, and reduce pharmacy queue waiting times for patients. 9 However, the specifics of shifting ART related tasks from pharmacy to non-pharmacy personnel have not yet been addressed in a systematic review. We therefore plan to synthesise the evidence for task shifting in pharmacy services, where non-pharmacy personnel undertake ART dispensing and distribution and medication adherence counselling. For this systematic review, pharmacy personnel will include both pharmacists and pharmacy technicians. Pharmacy technicians constitute an important part of the pharmacy workforce in low and middleincome countries: a survey of 26 low and middle-income countries in 2011 revealed that pharmacy technicians constitute 10% (Nigeria) to 70% (Pakistan) of the pharmacy workforce. 10 How the intervention might work Within the last decade, several high HIV burden countries adopted task shifting strategies where nurses and non-physician clinicians initiate and maintain ART. 11 Although this has undeniably expanded access to ART, it is also increasingly essential that long facility waiting times and frequent facility visits to collect ART are addressed to alleviate the burden of care, both for patients and healthcare providers. 12 13 Recent studies in Uganda, Kenya and Mozambique have shown positive outcomes when non-health professionals (lay people) delivered ART at the community level. 14 In Mozambique the use of PLHIV for distributing ART, monitoring adherence, reporting outcomes and referring sick patients to health facilities yielded a retention rate of 97.5% among stable patients on ART. 14 In a cluster randomised trial in Uganda, the use of trained community health workers produced comparable results with facility-based ART programme in terms of patient retention, viral load suppression and mortality rate. 15 Similar findings were also obtained in Kenya and Uganda when lay providers were engaged in ART delivery. 15 16 Task shifting has therefore been seen as an achievable solution to the critical human resource shortages for scale-up of ART. 17 While it is imperative to increase the rate of recruitment and training of health workers as well as improve working conditions to reduce attrition and emigration, the HIV pandemic requires a more urgent measure to address the critical skills shortage. 18 Such measures may include shifting selected tasks (including dispensing and distributing ART and adherence counselling) from pharmacy to non-pharmacy personnel. The task shifting could free time for pharmacy personnel to focus on more technical functions such as supply management and pharmacovigilance.
Why this review is important Previous systematic reviews of task shifting for increasing ART access focused on clinical services where nurses and non-clinician physicians provide care. 11 Dependence on and shortages of pharmacists are also key constraints on ART expansion, but the specifics of task shifting for ART dispensing or distribution from pharmacy to non-pharmacy personnel have not been reviewed systematically. We will systematically review the scientific literature and assess the efficacy and safety of task shifting models that use non-pharmacy personnel in dispensing or distributing ART and assessing adherence to treatment of HIV infection.

OBJECTIVE
The aim of this review is to evaluate the efficacy and safety of shifting dispensing and distribution of ART as well as assessment of adherence from pharmacy to nonpharmacy personnel.

METHODS
This review protocol has been registered in the PROSPERO International Prospective Register of Systematic Reviews (http://www.crd.york.ac.uk/PROSPERO), registration number CRD42015017034.
Criteria for considering studies for this review Types of studies We will include randomised controlled trials (RCTs) and non-RCTs, irrespective of whether allocation to interventions occurred at the individual or cluster level.

Types of interventions
We will include studies that evaluate the shifting of selected tasks from pharmacy personnel to non-pharmacy personnel. The selected tasks include dispensing and distribution of ART and adherence assessment. Pharmacy personnel will include both pharmacists and pharmacy technicians. Non-pharmacy personnel may include (but are not limited to) nurses, non-physician clinicians, and lay providers such as patient peer groups, community volunteers, PLHIV and community health committees.

Types of outcome measures Primary outcomes
The primary outcome for this review is risk of death.

Secondary outcomes
Our secondary outcome measures include: ▸ Virological suppression ▸ Number of all-cause sick visits made to the health facility, including for adverse events ▸ Loss to follow-up ▸ Adherence to ART (as measured within the study, eg, pill counts, recall methods, digital methods) ▸ Acceptability to pharmacy personnel, non-pharmacy personnel and patients ▸ Harm, including error rates.

Search methods for identification of studies
We will perform a comprehensive and exhaustive search of electronic databases and conference proceedings in an attempt to identify all relevant studies available by the search date, regardless of language of publication or publication status ( published, unpublished, in press or in progress).

Databases of peer-reviewed literature
We will search the following electronic databases, from 1 January 1996 to the search date: Along with appropriate Medical Subject Heading (MeSH) terms and relevant keywords, we will use the Cochrane Highly Sensitive Search Strategy for identifying reports of randomised controlled trials in MEDLINE, 19 and the Cochrane validated strategies for identifying references relevant to HIV infection and AIDS. To identify other study designs, the RCT string will be omitted. The search strategy will be iterative in that references of included studies will be searched for additional references. See table 1 for our provisional search strategy for electronic databases.

Conference databases
We will search conference abstract archives on the web sites of the Conference on Retroviruses and Opportunistic Infections (CROI), the International AIDS Conference (IAC) and the International AIDS Society Conference on HIV Pathogenesis, Treatment and Prevention (IAS), for all available abstracts presented at these conferences from 1996 to the search date.

Searching other resources
We will also search the references of relevant articles as well as the WHO International Clinical Trials Registry Platform (ICTRP) and Clinicaltrials.gov. We will contact relevant experts or organisations who may be aware of additional studies in this field.

Data collection and analysis
We will base the methodology for data collection and analysis on the guidance provided in the Cochrane Handbook of Systematic Reviews of Interventions. 19 Selection of studies for inclusion Two authors will read and assess the abstracts of identified publications for potentially eligible studies. We will obtain full text articles for all abstracts judged by at least one of the two authors, to be potentially eligible. Two authors will independently inspect these potentially eligible publications to establish the relevance of the article to the review according to the pre-specified criteria regarding study design, participants, interventions and outcome measures.

Data extraction and management
Two authors will independently extract data into a prepiloted data extraction form. The following characteristics will be extracted from each included study: Study details: Complete citations of publications associated with the study, start and end dates, location, study design characteristics, type of facility involved, investigators, funding sources, recruitment, method of randomisation, sequence generation, method of allocation concealment, blinding of participants and personnel, blinding of outcome assessment, length of follow-up, losses to follow-up, withdrawals or drop-outs and other relevant details.
Details of the intervention: training of the cadre of health workers who were dispensing or distributing ART, what training or other support or supervision they received and other relevant details.
Details of participants: Trial inclusion and exclusion criteria, numbers of participants entering the trial, sex, clinical staging, CD4 count and other pertinent details.
Outcome details: Definitions of outcomes, details of how outcomes were assessed, numerators and denominators associated with each outcome, completeness of outcome data, effect estimates reported and other relevant outcome information.

Assessment of risk of bias in included studies
We will assess the risk of bias in RCTs using the Cochrane risk of bias assessment tool for randomised studies. 19 For non-RCT studies, we will use the Cochrane Risk of Bias Assessment Tool for Non-Randomised Studies of Interventions (ACROBAT-NRSI). 20   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  We will resolve any disagreements between the authors conducting duplicate independent screening of search outputs, assessments of study eligibility, extraction of data, and risk of bias assessment by discussion and consensus. Should this fail to resolve the differences, a third author will arbitrate.

Measures of effect
We will calculate and report risk ratios for dichotomous and time-to-event data and mean differences for continuous data with their 95% CIs.

Unit of analysis issues
The unit of analysis will be the individual study participant. Cluster-randomised trials will be included in meta-analyses only after adjustments are made for design effect. Design effects for cluster-randomised studies will be corrected by using standard procedures, using the formula: design effect=1+(m−1)r, where m is the average cluster size and r is the intra-cluster correlation coefficient.

Dealing with missing data
We will contact study authors if it is necessary to obtain data missing from published reports.

Assessment of heterogeneity
We will examine statistical heterogeneity between study results using the χ 2 test of homogeneity, with a significance α-level of 0.1. In addition, we will use the I 2 statistic to measure the amount of heterogeneity among the trials in each analysis. If we identify significant heterogeneity (ie, p<0.1), we will explore it by prespecified subgroup analysis. If heterogeneity persists, we will perform sensitivity analyses, report results separately and propose reasons for the observed heterogeneity.

Assessment of reporting biases
If any meta-analysis in our review includes 10 or more studies, we will assess the potential for publication bias using a funnel plot. 21 We will attempt to minimise the potential for publication bias through a comprehensive search of published and unpublished literature.

Data synthesis
We will conduct meta-analysis, if appropriate, using the Cochrane Review Manager software (RevMan [Computer program] The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, 2015; Version 5.3). If we find no significant statistical heterogeneity of effects, we will use the fixed effect method of meta-analysis. Otherwise, we will use the random effects model.

Subgroup analysis
In pooled results with significant statistical heterogeneity, we will explore the cause of the heterogeneity through subgroup analyses, with subgroups defined by type of intervention (eg, cadre of health provider), comparison group and region of study (eg, sub-Saharan Africa, Southeast Asia, etc).

Sensitivity analysis
We will conduct a sensitivity analysis to investigate the effect of excluding studies with high risk of bias, with a focus on bias introduced by inadequate allocation concealment, inadequate blinding of outcome assessment and substantial losses to follow-up.

Certainty of evidence
We will assess the certainty (or quality) of evidence using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach, 22 which defines the certainty of evidence for each outcome as "the extent of our confidence that the estimates of effect are correct". 19 The quality rating across studies has four levels: high, moderate, low and very low. Randomised trials are considered to be of high quality but can be downgraded for any of five reasons: risk of bias, indirectness of evidence, unexplained heterogeneity of effects, imprecision of effect estimates and high probability of publication bias. Similarly, observational studies are considered to be of low quality, but can be upgraded for any of three reasons. The quality level of a body of evidence can be increased if there is a large magnitude of effect, if all plausible confounding would reduce a demonstrated effect, and if there is a dose-response gradient.

Reporting of this review
The findings of this review will be presented in a number of ways. The study selection process will be summarised using a flow diagram, and if we identify 10 or more eligible studies, we will assess publication bias using funnel plots. Where appropriate, we will use risk of bias graphs, forest plots and GRADE summary of findings tables. The non-quantitative outcomes will be reported descriptively. We will provide tables of both included and excluded studies. We have prepared this protocol as recommended by the PRISMA-P (Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols) guidelines 23 and will report the findings of the review as recommended by the PRISMA statement. 24

ETHICS AND DISSEMINATION
Since systematic reviews do not directly involve human participants, they do not require ethical clearance. 25 We will provide the findings of this review to the WHO, with the hope that they may guide policy recommendations of this normative agent regarding the shifting of ART dispensing or distribution from pharmacy to nonpharmacy personnel. Although the majority of national programmes in low and middle-income countries have adopted task shifting in ART care at different levels, Mbeye  there has been no global policy to guide the practice for task shifting from pharmacy to non-pharmacy personnel. We will also publish the findings of the systematic review in a peer-reviewed journal.

Twitter Follow Charles Wiysonge at @CharlesShey
Acknowledgements The authors acknowledge the Editor and the referees for critical and constructive comments on earlier versions of this manuscript. The authors would also like to acknowledge the various contributions of Eyerusalem Negussie.
Contributors NMM led the development of the protocol, wrote the first draft, coordinated and integrated comments from co-authors and approved the final version for publication. TK and CSW conceived the study, critically revised successive drafts of the manuscript, and approved the final version for publication. CSW is the guarantor of the manuscript.
Funding Funding for preparing this review was provided by the Department of HIV/AIDS at the WHO Headquarters in Geneva, Switzerland. The authors would like to thank Hacsi Horvath (Global Health Sciences, University of California, San Francisco, USA) for his assistance in developing the search strategy for electronic databases. Neither the authors' institutions nor the funder played a role in preparing the manuscript and the views expressed therein are solely those of the authors.

Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.

Method 28
We conducted comprehensive searches of peer-reviewed and grey literature. Two 29 authors independently screened search outputs, selected controlled trials, extracted 30 data and resolved discrepancies by consensus. We performed random-effects meta-31 analysis and assessed certainty of evidence using GRADE. 32

34
Three studies with 1993 participants met the inclusion criteria, including two cluster 35 trials conducted in Kenya and Uganda and an individually randomised trial conducted 36 in Brazil. Pooling data we found that there may be no differences between pharmacy 37 and non-pharmacy models of dispensing ART on mortality (RR 1.86, 95% confidence 38 interval 0.44 to 7.95), virological failure (RR 0.92, 0.73 to 1.15) and loss to follow-up 39 (RR 1.13, 0.68 to 1.91). We found some evidence that costs may be reduced for the 40 patient and health system when task-shifting is undertaken. The overall quality of the 41 evidence was low for all outcomes due to serious indirectness (trials compared 42 complex interventions involving, but not limited to, pharmacy to non-pharmacy 43 personnel) and imprecision (effects ranged from appreciable benefit to harm). 44  Health outcomes may be similar when ART dispensing or distribution is shifted from 46 pharmacy to non-pharmacy personnel. However, the low-certainty evidence implies a 47 high likelihood that further research may find the effects of the intervention to be 48 substantially different from our findings. If resource-constrained countries decide to 49 shift ART dispensing and distribution from pharmacy to non-pharmacy personnel, 50 this should be accompanied by robust monitoring and impact evaluation.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  both at patient and population levels 3 4 . Scale up of ART in low and middle-income 79 countries has averted more than 5 million deaths; however, bottlenecks preventing 80 universal access to ART still exist. One challenge is the critical shortage of human 81 resources for health (HRH), including for delivery of essential pharmacy services. 82 As for pharmacists, although the World Health Organization (WHO) recommends a 83 minimum of one pharmacist per 2300 population 5 , most countries in low-resource 84 settings such as sub-Saharan Africa do not meet this target. In addition to the absolute 85 shortage, it is likely that there is an uneven distribution of pharmacists in such 86 settings, as is the case with other specialist healthcare workers who tend to 87 concentrate in urban areas and the private sector, further aggravating the HRH 88 shortage. For instance in South Africa, which is home to the largest number of PLHIV 89 in any country worldwide, in 2010 only 24% of registered pharmacists worked in the 90 public sector where 80% of the population received care 6 . 91

93
Studies and programmes report that involvement of pharmacy personnel in HIV care 94 results in improved patient outcomes 7 . For instance, in the United States of America 95 (USA), the use of a multidisciplinary team approach with pharmacists assuming a 96  treatment outcomes such as viral load, patient retention and medication adherence 8 . 98 The work of pharmacists includes supply management, dispensing and distributing 99 medications, promoting adherence, identifying and preventing potential medication-100 related issues (such as over dosage, sub-therapeutic dosage, adverse drug reactions, 101 medication errors and untreated indications), and monitoring and reporting adverse 102 drug events 9 10 . In some settings, programmes have implemented alternative models 103 of pharmacy services that shift selected tasks from pharmacy to non-pharmacy 104 personnel. Such alternative models could potentially increase the number of health 105 workers involved in ART distribution, adherence counselling and patient education; 106 free more time for pharmacy personnel; support the integration of ART in primary 107 care settings; minimise the number of facility visits for ART collection; and, reduce 108 pharmacy queue waiting times for patients. 109 The specifics of shifting ART-related tasks from pharmacy personnel have not been 110 addressed in a systematic review. Available systematic reviews on task shifting focus 111 on clinical services where nurses and non-physician clinicians provide care 112 comparable to physicians 11 . We therefore synthesised the evidence for task shifting in 113 pharmacy personnel services, where non-pharmacy personnel undertake ART 114 distribution and medication adherence counselling. Table 1 provides definitions for 115 the pharmacy functions distribution and dispensing as they have been used in this 116 review. However, these definitions seem to overlap at facility level. 117

Term Definition
Dispensing • Dispensing is a controlled act that authorises one to select, prepare and provide stock medication that has been prescribed to a patient or client (or his/her representative) for • To dispense is to prepare and supply to a patient a course of therapy on the basis of a prescription 13 ..
• Dispensing is preparation and distribution of a course of therapy to a patient, with appropriate instructions based on a prescription 13 .

Distribution
• At facility level, drug distribution mainly refers to drugs dispensed by licensed practitioners, such as nurses, doctors, pharmacists or pharmacy assistants and collected by a patient.
• At community level, drug distribution refers to trained lay people collecting pre-packed medications from the facility and delivering them to HIV patients in the community 12 .

120
How task shifting using non-pharmacy personnel might 121 work 122 Within the last decade, several high HIV burden countries adopted task shifting 123 strategies where nurses and non-physician clinicians initiate and maintain ART. 124 Although this has undeniably expanded access to ART, it is also increasingly essential 125 that long facility waiting times and frequent facility visits to collect ART are 126 addressed to alleviate the burden of care, both for patients and healthcare providers 9 127 10 . 128 Recent studies in Uganda, Kenya and Mozambique have shown positive outcomes 129 when non-health professionals (lay people) delivered ART at the community level 14 . 130 In Mozambique, the use of PLHIV for distributing ART, monitoring adherence, 131 reporting outcomes and referring sick patients to health facilities yielded a retention 132 rate of 97.5% among stable patients on ART 14 . In a cluster randomised trial in 133 Uganda, the use of community health workers produced comparable results with the 134 facility-based ART programme in terms of patient retention, viral load suppression 135 and mortality rate 15 . Similar findings were also obtained in Kenya and some other 136 parts of Uganda when lay providers were engaged in ART delivery 15 16 . 137  The aim of this review was to evaluate the efficacy and safety of shifting pharmacy 160 related tasks, including ART distribution and adherence assessment, from pharmacy 161 to non-pharmacy personnel. 162        1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59

Sequence generation (checking for selection bias) 252
We considered a study as having a low risk of selection bias due to sequence 253 generation if investigators described a random component in the sequence-generation 254 process, such as the use of random number tables; coin tossing, card or envelope 255 shuffling. We classified a study as having a high risk of bias if investigators described 256 a non-random component in the sequence-generation process, such as the use of date 257  we classified the study as having unclear risk of bias. 260

Allocation concealment (checking for selection bias) 261
We classified the study as having a low risk due to allocation concealment if 262 participants and investigators enrolling participants could not foresee assignment (e.g. We assigned a study as having an unclear risk of selection bias when the study 268 reported insufficient information to permit judgment of the allocation concealment or 269 the methods used to conceal allocation were not described. 270

Blinding (checking for performance and detection bias) 271
We determined that a study was at low risk of performance bias when there was 272 blinding of the participants and key study personnel, and it was unlikely that the 273 blinding could have been broken or there was no blinding in the situation but non-274 blinding was unlikely to introduce bias. A study with no blinding or incomplete 275 blinding when the outcome was likely to be influenced by lack of blinding was 276 classified as having a high risk of performance bias. In the absence of sufficient 277 information to permit judgment of adequacy of the blinding of participants and 278 investigators, the study was classified as having unclear risk of performance bias. 279 Regarding detection bias, we determined that a study was at low risk of bias when 280 there was blinding of outcome assessors. We classified a study as having a high risk 281 of detection bias if there was no blinding or the outcome was likely to be influenced 282  Incomplete outcome data (checking for attrition bias) 286 We classified a study as having a low risk of attrition bias if there was no missing 287 outcome data, reasons for missing outcome data were unlikely to be related to the 288 outcome, or the proportion of missing outcome data was balanced across groups. We 289 classified a study as having a high risk of attrition bias if there was imbalance in 290 proportions lost to follow-up across groups and the reason for missing outcome data 291 was likely to be related to the outcome. When there was insufficient reporting of 292 attrition or exclusions, we classified the study as having an unclear risk of attrition 293 bias. 294

Selective reporting (checking for reporting bias) 295
We determined a study as having a low risk of reporting bias if a protocol was 296 available and primary outcomes in the final trial report corresponded closely to those 297 presented in the protocol. High risk of selective reporting was considered to be 298 present when the primary outcomes differed between the protocol and final trial 299 report. When no trial protocol was available or there was insufficient reporting to 300 determine if selective reporting was present, we classified the study as having an 301 unclear risk of reporting bias. 302

Checking for other forms of bias 303
We classified a study as having a low risk of other forms of bias, when there was no 304 evidence of bias from other sources. A study was classified as having a high risk from 305 other sources when potential bias was present from other sources (e.g. early stopping 306 of trial, fraudulent activity, financial conflicts of interest). We assigned a study as 307 Measures of effect 310 We calculated and presented summary statistics for the risk ratio (RR) for 311 dichotomous outcomes and the weighted-mean difference for continuous outcomes, 312 with their 95% confidence intervals (CI). 313

315
The unit of analysis was the individual study participant for all the trials. 316 Dealing with missing data 317 318 We contacted study authors for one of the included trials to obtain information on the 319 composition of health professionals in the control groups involved in dispensing ART 320 in order to establish whether pharmacy personnel were part of the team. Unfortunately 321 the author could not remember details but assumed that health professionals also 322 included pharmacy personnel. 323

325
We used the I 2 statistic to measure heterogeneity among the trials. We planned to 326 explore substantial heterogeneity (I 2 >50%) by pre-specified subgroup analysis. 327 However, there was no evidence of serious heterogeneity in our review hence we did 328 not perform sub-group analyses. 329

331
We contacted study authors to provide any missing outcome data. However, only one 332 author from one included trial responded but did not provide any additional 333 information. If any meta-analysis in our review was to include 10 or more studies, we 334 planned to assess the potential for publication bias using a funnel plot 20 21 . We 335 minimised the potential for publication bias by using a comprehensive search strategy, 336 as described above. 337

Description of interventions: Pharmacy personnel group:
Patients received this intervention on a monthly basis through the Da´der method which consisted a series of scheduled meetings where the pharmacist and patient addressed, reviewed, and solved drug-related problems for a 12month period. Patients received structured counselling from pharmacists on their prescription regimens at the time of their initial drug dispensing and at monthly refill visits. Key elements included: • Reviewing the prescription with the patient.
• Reviewing a card on which medications were colour-coded to facilitate recognition and reduce confusion that might arise from complicated drug names. • Reviewing the schedule, length and date of next appointment.
• Reviewing patient's understanding of the prescription by asking patient to describe it and giving patients verbal information on the expected side effects of their medications and instructing them to seek medical assistance by calling the pharmacist if side effects occurred.

Non-pharmacy personnel group:
Patients received Usual Care (UC) for ART drug distribution. The control group received UC for ART drug delivery. This included the following: At the first appointment, when patients received their medications, a nurse first provided information on: • The regimen, • How and when to use the medications, •    383 We judged the three trials to have a low risk of selection bias as a result of adequate 384 randomisation and allocation concealment. The studies by Selke et al 16  However, the Selke trial 16 did not give sufficient information to permit judgment of 388 whether or not detection bias was present. The Silveira trial 26 was judged to have a 389 high risk of performance and detection bias due to lack of blinding for participants, 390 personnel and outcome assessment respectively for all study outcomes. The three 391 studies were judged to have a low risk of attrition bias as they did not have differential 392 or large numbers of losses to follow-up across the intervention arms. The risk of 393 selective reporting bias was judged to be low for the study by Jaffar et al 25  information and as having a high risk of other biases due to inadequate sample size. 397 Figure 2 provides a graphical summary of the risk of bias assessments. 398

399
There was one reported death in the non-pharmacy personnel group in the Selke trial 400 16 during the study period. However, the authors reported the death as unrelated to 401 HIV but rather due to use of herbal abortifacients. The trial by Jaffar 25 in Uganda did 402 not find differences in mortality rates between the non-pharmacy and pharmacy 403 personnel groups: adjusted rate ratio RR 0.95 (95% CI: 0.71 to 1.28). Five deaths 404 were reported in the non-pharmacy personnel group compared to none in the 405 pharmacy personnel group in the Silveira trial 26 . This difference was not statistically 406 significant (RR 11.00, 95% CI: 0.61 to 197.35). When these data were pooled as 407 shown in Figure 3, there was no difference in mortality between the groups cared for 408 by non-pharmacy personnel and those cared for by pharmacy personnel (RR 1.86; 409 95% CI: 0.44 to 7.95), with no significant heterogeneity (Chi 2 = 3.34; df = 2; p = 410 0.19; I 2 = 40%). The certainty of the evidence for this outcome was low due to serious 411 indirectness and imprecision as shown in Table 5. We downgraded for indirectness 412 because two of the three trials (Selke 16 and Jaffar 25 ) compared complex interventions 413 that included, but were not limited to, pharmacy and non-pharmacy personnel as 414 shown in Table 5. There were few deaths in the trials and the effect estimate was 415 imprecise, with wide confidence intervals ranging from appreciable benefit to 416 substantial harm as reported in Table 5.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59 1 We downgraded by one for serious indirectness: Two trials compared complex interventions that included, but were not merely limited to, pharmacy and non-pharmacy personnel. 2 We downgraded by one for serious imprecision: There are few events and the effect estimates have wide confidence intervals, ranging from appreciable benefit to harm.   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  In a meta-analysis, we detected no differences in virological failure between the group 447 cared for by non-pharmacy personnel and that cared for by pharmacy personnel (RR 448 0.92; 95% CI: 0.73 to 1.15), with no significant heterogeneity detected between the 449 trials (Chi 2 = 0.24; df = 2; p=0.89; I 2 = 0%) as shown in Figure 4. The quality of the 450 evidence for this outcome was low due to serious indirectness and imprecision as 451 shown in Table 5. In our meta-analysis, we detected no difference in loss to follow up between the 467 groups (RR 1.13; 95% CI: 0.68 to 1.91), with no significant heterogeneity (Chi 2 = 468 0.54; df = 2; p=0.76; I 2 = 0%) as shown in Figure 4. The quality of the evidence for 469 this outcome was low due to serious indirectness and imprecision as shown in Table  470 5. 471

473
Selke 16 reported significantly more clinic visits in the pharmacy compared to the 474 non-pharmacy personnel group (mean visits 12.6 versus 6.4), p<0.001. Despite fewer 475 clinic visits for the non-pharmacy personnel group, clinical outcomes between the two 476 groups were comparable with an incidence of 13.6 HIV opportunistic infections per 477 100 person-years in the non-pharmacy personnel group and 19.8 HIV opportunistic 478 infections per 100 person-years in the pharmacy personnel group, p=0.42. Although 479 the non-pharmacy personnel group was found to have fewer clinic visits, the authors 480 observed that this group undertook 64% more clinic visits than originally scheduled. 481 The study by Jaffar 25  year were higher in the pharmacy personnel group ($54) compared to non-pharmacy 502 personnel group ($18). Costs to access care included transport, lunch, childcare and 503 lost work time. The outcome cost was not pre-specified in the protocol, however, this 504 provides additional data that may be relevant for decision makers and was therefore 505 included for consideration. 506 Acceptability to pharmacy personnel, non-pharmacy personnel and 507 patients and harm, including error rates 508 509 These outcomes were not reported in the included trials. 510

513
Two cluster randomised clinical trials and one non-blinded randomised controlled 514 trial were included in this review of effects of shifting responsibility from pharmacy 515 to non-pharmacy personnel for adherence assessment and dispensing antiretroviral 516 drugs to HIV-infected patients. 517 In our meta-analysis, we found low quality evidence that there may probably be no 518 difference in mortality, virological response, and loss to follow up between non-519 pharmacy personnel and pharmacy personnel distributing ART to patients. In 520 addition, we did not observe significant differences in adherence to treatment. Selke et 521 al 16 reports that the non-pharmacy personnel group had significantly fewer all-cause 522 sick visits to the clinic compared to the pharmacy personnel group. Although the main 523  16 and Jaffar 25 trials was the level of education of the 524 non-pharmacy personnel, it did not seem to affect their overall performance 16 25 . 525 Additionally, the Jaffar trial 25 found the non-pharmacy dispensing strategy to be 526 cost-effective and cheaper to run than the pharmacy dispensing strategy by almost 45 527 US dollars per patient per annum and this accounted for only 6% of the total cost of 528 healthcare service expenditure for the intervention group 25 . The patients who were 529 accessing their medications and other healthcare services from pharmacy personnel 530 incurred more costs in terms of transportation, lunch, childcare costs and lost work 531 time. In the first year of the study, the pharmacy personnel group incurred double the 532 cost of healthcare services per patient per annum compared to the non-pharmacy 533 personnel group. This is a challenge in most poor settings in Africa where many 534 cannot afford basic necessities and this further impedes their access to treatment 25 . 535 Lack of or inadequate financial resources are major factors in late presentation to 536 health facilities, poor access to HIV care and support, and low retention in care after 537 initiation of ART 32 33 . Although health services are provided for free in most public 538 institutions, the cost of accessing the services may serve as a hindrance to achieving 539 optimal care and support for HIV-infected patients 32 . The non-pharmacy personnel 540 dispensing ART could therefore considerably reduce the cost of services for both 541 patients and governments with potentially favourable clinical outcomes. 542 Although issues related to drug-drug interaction, medication errors, pre-screening for 543 ART and some special cases of adherence to therapy may be beyond the capacity of 544 non-pharmacy personnel, targeted pharmacy support to non-pharmacy personnel 545 could offer benefits to patients having challenges with their medications especially 546 those who are taking concomitant medications with ART 34 . 547  Although it may not be possible to rule out publication and language biases in 564 systematic reviews and that this could potentially threaten the conduct of this review, 565 our trial search was not restricted to any language, setting or publication status. We 566 are of the opinion that we identified all the existing randomised controlled trials 567 relevant to our research question. However, there is a possibility that some 568 unpublished or published trials were not accessible by our search.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  We assessed the certainty or quality of evidence using the GRADE approach 35 . . 579 Evidence from this review should be applied with caution considering its low quality. 580 Three outcomes in this review were found to be of low quality evidence. We are 581 therefore not confident enough to state that the estimates lie close to the true value 582 and it is possible that different and further studies are needed to contribute to the body 583 of knowledge to increase our certainty in these review findings. 584

8-9
Information sources 7 Describe all information sources (e.g., databases with dates of coverage, contact with study authors to identify additional studies) in the search and date last searched. 9 Search 12 Data collection process 10 Describe method of data extraction from reports (e.g., piloted forms, independently, in duplicate) and any processes for obtaining and confirming data from investigators.

12-13
Data items 11 List and define all variables for which data were sought (e.g., PICOS, funding sources) and any assumptions and simplifications made.

12-13
Risk of bias in individual studies 12 Describe methods used for assessing risk of bias of individual studies (including specification of whether this was done at the study or outcome level), and how this information is to be used in any data synthesis.

16
Synthesis of results 14 Describe the methods of handling data and combining results of studies, if done, including measures of consistency (e.g., I 2 ) for each meta-analysis.

Study selection
17 Give numbers of studies screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally with a flow diagram.

18
Study characteristics 18 For each study, present characteristics for which data were extracted (e.g., study size, PICOS, follow-up period) and provide the citations.

19-23
Risk of bias within studies 19 Present data on risk of bias of each study and, if available, any outcome level assessment (see item 12).

23-24
Results of individual studies 20 For all outcomes considered (benefits or harms), present, for each study: (a) simple summary data for each intervention group (b) effect estimates and confidence intervals, ideally with a forest plot.

24-29
Synthesis of results 21 Present results of each meta-analysis done, including confidence intervals and measures of consistency.

24-29
Risk of bias across studies 22 Present results of any assessment of risk of bias across studies (see Item 15).
n/a DISCUSSION Summary of evidence 24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers).

29-31
Limitations 25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias).

31-32
Conclusions 26 Provide a general interpretation of the results in the context of other evidence, and implications for future research.   (ART) tasks from specialised healthcare workers to those with shorter or less formal training has been implemented in resource-limited settings to alleviate critical shortages of human resources for health. However, the specifics of shifting ART dispensing from pharmacy to non-pharmacy personnel have not been addressed in a systematic review, although this can potentially increase access to ART. We will assess the effects of shifting dispensing and distribution of ART and adherence assessment from pharmacy to nonpharmacy personnel in low and middle-income countries.

31-32
Methods and analysis: We will search PubMed, CENTRAL, EMBASE, WHO Global Health Library and relevant grey literature for eligible controlled trials. Two authors will screen the search output, select eligible studies, assess risk of bias and extract data from included studies, resolving discrepancies by discussion and consensus. We will perform meta-analysis using both fixed and random effects models, investigate clinical and statistical heterogeneity, and assess our confidence in the overall evidence using standard Cochrane methods, including GRADE.

Ethics and dissemination:
Only secondary data will be included in this review and ethics approval is not required. We will disseminate the review findings in various scientific fora, including peer-reviewed journals. The findings may help to inform policy makers in defining the scope of work of healthcare workers, and global recommendations for shifting the dispensing and distribution of ART from pharmacy to non-pharmacy personnel.

INTRODUCTION
Description of the condition By March 2015, 15 million (40.7%) of the estimated 36.9 million people living with HIV (PLHIV) globally were receiving antiretroviral therapy (ART). 1 Combination ART is effective for reducing HIV related morbidity and mortality as well as preventing HIV transmission. 2 Initiating ART early in the course of HIV infection has been associated with better health outcomes, both at patient and population levels. 3 4 Scale-up of ART in low and middle income countries has averted more than 5 million deaths; however, bottlenecks preventing universal access to ART still exist. One challenge is the critical shortage of human resources for health (HRH), including for the delivery of essential HIV related pharmacy services. The WHO recommends a minimum of one pharmacist per 2300 population, but most countries in low-resource settings such as sub-Saharan Africa have not yet met this target. 5 In addition to the absolute shortage, it is likely that there is an uneven distribution of pharmacists in such settings, as is the case with other specialist healthcare workers who Strengths and limitations of this study ▪ To our knowledge, this is the first published protocol of a systematic review that will investigate the effects of task shifting from pharmacy to non-pharmacy personnel for dispensing or distributing antiretroviral therapy to patients living with HIV.  tend to concentrate in urban areas and the private sector, further aggravating the HRH shortage. 6 For instance in South Africa, which is home to the largest number of PLHIV in any country in the world, in 2010 only 24% of registered pharmacists worked in the public sector where 80% of the population received care. 7

Description of the intervention
Studies and programme reports indicate that involvement of pharmacy personnel in HIV care results in improved patient outcomes. For instance, in the USA, the use of a multidisciplinary team approach with pharmacists assuming a central role in ART initiation, dispensing and adherence counselling improved treatment outcomes such as viral load, patient retention and medication adherence. 8 The work of pharmacists includes supply management, dispensing and distributing medications, promoting adherence, identifying and preventing potential medication-related issues, and monitoring and reporting adverse events. In some settings, programmes have implemented alternative models of pharmacy services that shift selected tasks from pharmacy to non-pharmacy personnel. Such alternative models could potentially increase the number of health workers involved in ART distribution, adherence counselling and patient education, free more time for pharmacy personnel, support the integration of ART in primary care settings, minimise the number of facility visits for ART collection, and reduce pharmacy queue waiting times for patients. 9 However, the specifics of shifting ART related tasks from pharmacy to non-pharmacy personnel have not yet been addressed in a systematic review. We therefore plan to synthesise the evidence for task shifting in pharmacy services, where non-pharmacy personnel undertake ART dispensing and distribution and medication adherence counselling. For this systematic review, pharmacy personnel will include both pharmacists and pharmacy technicians. Pharmacy technicians constitute an important part of the pharmacy workforce in low and middleincome countries: a survey of 26 low and middle-income countries in 2011 revealed that pharmacy technicians constitute 10% (Nigeria) to 70% (Pakistan) of the pharmacy workforce. 10 How the intervention might work Within the last decade, several high HIV burden countries adopted task shifting strategies where nurses and non-physician clinicians initiate and maintain ART. 11 Although this has undeniably expanded access to ART, it is also increasingly essential that long facility waiting times and frequent facility visits to collect ART are addressed to alleviate the burden of care, both for patients and healthcare providers. 12 13 Recent studies in Uganda, Kenya and Mozambique have shown positive outcomes when non-health professionals (lay people) delivered ART at the community level. 14 In Mozambique the use of PLHIV for distributing ART, monitoring adherence, reporting outcomes and referring sick patients to health facilities yielded a retention rate of 97.5% among stable patients on ART. 14 In a cluster randomised trial in Uganda, the use of trained community health workers produced comparable results with facility-based ART programme in terms of patient retention, viral load suppression and mortality rate. 15 Similar findings were also obtained in Kenya and Uganda when lay providers were engaged in ART delivery. 15 16 Task shifting has therefore been seen as an achievable solution to the critical human resource shortages for scale-up of ART. 17 While it is imperative to increase the rate of recruitment and training of health workers as well as improve working conditions to reduce attrition and emigration, the HIV pandemic requires a more urgent measure to address the critical skills shortage. 18 Such measures may include shifting selected tasks (including dispensing and distributing ART and adherence counselling) from pharmacy to non-pharmacy personnel. The task shifting could free time for pharmacy personnel to focus on more technical functions such as supply management and pharmacovigilance.
Why this review is important Previous systematic reviews of task shifting for increasing ART access focused on clinical services where nurses and non-clinician physicians provide care. 11 Dependence on and shortages of pharmacists are also key constraints on ART expansion, but the specifics of task shifting for ART dispensing or distribution from pharmacy to non-pharmacy personnel have not been reviewed systematically. We will systematically review the scientific literature and assess the efficacy and safety of task shifting models that use non-pharmacy personnel in dispensing or distributing ART and assessing adherence to treatment of HIV infection.

OBJECTIVE
The aim of this review is to evaluate the efficacy and safety of shifting dispensing and distribution of ART as well as assessment of adherence from pharmacy to nonpharmacy personnel.

METHODS
This review protocol has been registered in the PROSPERO International Prospective Register of Systematic Reviews (http://www.crd.york.ac.uk/PROSPERO), registration number CRD42015017034.
Criteria for considering studies for this review Types of studies We will include randomised controlled trials (RCTs) and non-RCTs, irrespective of whether allocation to interventions occurred at the individual or cluster level.

Types of participants
Participants will be PLHIV receiving ART.

Types of interventions
We will include studies that evaluate the shifting of selected tasks from pharmacy personnel to non-pharmacy personnel. The selected tasks include dispensing and distribution of ART and adherence assessment. Pharmacy personnel will include both pharmacists and pharmacy technicians. Non-pharmacy personnel may include (but are not limited to) nurses, non-physician clinicians, and lay providers such as patient peer groups, community volunteers, PLHIV and community health committees.

Types of outcome measures Primary outcomes
The primary outcome for this review is risk of death.

Secondary outcomes
Our secondary outcome measures include: ▸ Virological suppression ▸ Number of all-cause sick visits made to the health facility, including for adverse events ▸ Loss to follow-up ▸ Adherence to ART (as measured within the study, eg, pill counts, recall methods, digital methods) ▸ Acceptability to pharmacy personnel, non-pharmacy personnel and patients ▸ Harm, including error rates.

Search methods for identification of studies
We will perform a comprehensive and exhaustive search of electronic databases and conference proceedings in an attempt to identify all relevant studies available by the search date, regardless of language of publication or publication status ( published, unpublished, in press or in progress).

Databases of peer-reviewed literature
We will search the following electronic databases, from 1 January 1996 to the search date: Along with appropriate Medical Subject Heading (MeSH) terms and relevant keywords, we will use the Cochrane Highly Sensitive Search Strategy for identifying reports of randomised controlled trials in MEDLINE, 19 and the Cochrane validated strategies for identifying references relevant to HIV infection and AIDS. To identify other study designs, the RCT string will be omitted. The search strategy will be iterative in that references of included studies will be searched for additional references. See table 1 for our provisional search strategy for electronic databases.

Conference databases
We will search conference abstract archives on the web sites of the Conference on Retroviruses and Opportunistic Infections (CROI), the International AIDS Conference (IAC) and the International AIDS Society Conference on HIV Pathogenesis, Treatment and Prevention (IAS), for all available abstracts presented at these conferences from 1996 to the search date.

Searching other resources
We will also search the references of relevant articles as well as the WHO International Clinical Trials Registry Platform (ICTRP) and Clinicaltrials.gov. We will contact relevant experts or organisations who may be aware of additional studies in this field.

Data collection and analysis
We will base the methodology for data collection and analysis on the guidance provided in the Cochrane Handbook of Systematic Reviews of Interventions. 19 Selection of studies for inclusion Two authors will read and assess the abstracts of identified publications for potentially eligible studies. We will obtain full text articles for all abstracts judged by at least one of the two authors, to be potentially eligible. Two authors will independently inspect these potentially eligible publications to establish the relevance of the article to the review according to the pre-specified criteria regarding study design, participants, interventions and outcome measures.

Data extraction and management
Two authors will independently extract data into a prepiloted data extraction form. The following characteristics will be extracted from each included study: Study details: Complete citations of publications associated with the study, start and end dates, location, study design characteristics, type of facility involved, investigators, funding sources, recruitment, method of randomisation, sequence generation, method of allocation concealment, blinding of participants and personnel, blinding of outcome assessment, length of follow-up, losses to follow-up, withdrawals or drop-outs and other relevant details.
Details of the intervention: training of the cadre of health workers who were dispensing or distributing ART, what training or other support or supervision they received and other relevant details.
Details of participants: Trial inclusion and exclusion criteria, numbers of participants entering the trial, sex, clinical staging, CD4 count and other pertinent details.
Outcome details: Definitions of outcomes, details of how outcomes were assessed, numerators and denominators associated with each outcome, completeness of outcome data, effect estimates reported and other relevant outcome information.

Assessment of risk of bias in included studies
We will assess the risk of bias in RCTs using the Cochrane risk of bias assessment tool for randomised studies. 19 For non-RCT studies, we will use the Cochrane Risk of Bias Assessment Tool for Non-Randomised Studies of Interventions (ACROBAT-NRSI). 20  We will resolve any disagreements between the authors conducting duplicate independent screening of search outputs, assessments of study eligibility, extraction of data, and risk of bias assessment by discussion and consensus. Should this fail to resolve the differences, a third author will arbitrate.

Measures of effect
We will calculate and report risk ratios for dichotomous and time-to-event data and mean differences for continuous data with their 95% CIs.

Unit of analysis issues
The unit of analysis will be the individual study participant. Cluster-randomised trials will be included in meta-analyses only after adjustments are made for design effect. Design effects for cluster-randomised studies will be corrected by using standard procedures, using the formula: design effect=1+(m−1)r, where m is the average cluster size and r is the intra-cluster correlation coefficient.

Dealing with missing data
We will contact study authors if it is necessary to obtain data missing from published reports.

Assessment of heterogeneity
We will examine statistical heterogeneity between study results using the χ 2 test of homogeneity, with a significance α-level of 0.1. In addition, we will use the I 2 statistic to measure the amount of heterogeneity among the trials in each analysis. If we identify significant heterogeneity (ie, p<0.1), we will explore it by prespecified subgroup analysis. If heterogeneity persists, we will perform sensitivity analyses, report results separately and propose reasons for the observed heterogeneity.

Assessment of reporting biases
If any meta-analysis in our review includes 10 or more studies, we will assess the potential for publication bias using a funnel plot. 21 We will attempt to minimise the potential for publication bias through a comprehensive search of published and unpublished literature.

Data synthesis
We will conduct meta-analysis, if appropriate, using the Cochrane Review Manager software (RevMan [Computer program] The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, 2015; Version 5.3). If we find no significant statistical heterogeneity of effects, we will use the fixed effect method of meta-analysis. Otherwise, we will use the random effects model.

Subgroup analysis
In pooled results with significant statistical heterogeneity, we will explore the cause of the heterogeneity through subgroup analyses, with subgroups defined by type of intervention (eg, cadre of health provider), comparison group and region of study (eg, sub-Saharan Africa, Southeast Asia, etc).

Sensitivity analysis
We will conduct a sensitivity analysis to investigate the effect of excluding studies with high risk of bias, with a focus on bias introduced by inadequate allocation concealment, inadequate blinding of outcome assessment and substantial losses to follow-up.

Certainty of evidence
We will assess the certainty (or quality) of evidence using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach, 22 which defines the certainty of evidence for each outcome as "the extent of our confidence that the estimates of effect are correct". 19 The quality rating across studies has four levels: high, moderate, low and very low. Randomised trials are considered to be of high quality but can be downgraded for any of five reasons: risk of bias, indirectness of evidence, unexplained heterogeneity of effects, imprecision of effect estimates and high probability of publication bias. Similarly, observational studies are considered to be of low quality, but can be upgraded for any of three reasons. The quality level of a body of evidence can be increased if there is a large magnitude of effect, if all plausible confounding would reduce a demonstrated effect, and if there is a dose-response gradient.

Reporting of this review
The findings of this review will be presented in a number of ways. The study selection process will be summarised using a flow diagram, and if we identify 10 or more eligible studies, we will assess publication bias using funnel plots. Where appropriate, we will use risk of bias graphs, forest plots and GRADE summary of findings tables. The non-quantitative outcomes will be reported descriptively. We will provide tables of both included and excluded studies. We have prepared this protocol as recommended by the PRISMA-P (Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols) guidelines 23 and will report the findings of the review as recommended by the PRISMA statement. 24

ETHICS AND DISSEMINATION
Since systematic reviews do not directly involve human participants, they do not require ethical clearance. 25 We will provide the findings of this review to the WHO, with the hope that they may guide policy recommendations of this normative agent regarding the shifting of ART dispensing or distribution from pharmacy to nonpharmacy personnel. Although the majority of national programmes in low and middle-income countries have adopted task shifting in ART care at different levels, there has been no global policy to guide the practice for task shifting from pharmacy to non-pharmacy personnel. We will also publish the findings of the systematic review in a peer-reviewed journal.

Method 28
We conducted comprehensive searches of peer-reviewed and grey literature. Two 29 authors independently screened search outputs, selected controlled trials, extracted 30 data and resolved discrepancies by consensus. We performed random-effects meta-31 analysis and assessed certainty of evidence using GRADE. 32

34
Three studies with 1993 participants met the inclusion criteria, including two cluster 35 trials conducted in Kenya and Uganda and an individually randomised trial conducted 36 in Brazil. We found very low certainty evidence regarding mortality due to the low 37 number of events. Therefore, we are uncertain whether there is a true increase in 38 mortality as the effect size suggests, or a reduction in mortality between pharmacy 39 and non-pharmacy models of dispensing ART (RR 1.86, 95% confidence interval (CI) 40 0.44 to 7.95, n = 1993, 3 trials, very low certainty evidence). There may be no 3 certainty evidence). We found some evidence that costs may be reduced for the 45 patient and health system when task-shifting is undertaken. 46

Conclusions 47
The low certainty regarding the evidence implies a high likelihood that further 48 research may find the effects of the intervention to be substantially different from our 49 findings. If resource-constrained countries decide to shift ART dispensing and 50 distribution from pharmacy to non-pharmacy personnel, this should be accompanied 51 by robust monitoring and impact evaluation. both at patient and population levels 3 4 . Scale up of ART in low and middle-income 79 countries has averted more than 5 million deaths; however, bottlenecks preventing 80 universal access to ART still exist. One challenge is the critical shortage of human 81 resources for health (HRH), including for delivery of essential pharmacy services.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  treatment outcomes such as viral load, patient retention and medication adherence 8 . 98 The work of pharmacists includes supply management, dispensing and distributing 99 medications, promoting adherence, identifying and preventing potential medication-100 related issues (such as over dosage, sub-therapeutic dosage, adverse drug reactions, 101 medication errors and untreated indications), and monitoring and reporting adverse 102 drug events 9 10 . In some settings, programmes have implemented alternative models 103 of pharmacy services that shift selected tasks from pharmacy to non-pharmacy 104 personnel. Such alternative models could potentially increase the number of health 105 workers involved in ART distribution, adherence counselling and patient education; 106 free more time for pharmacy personnel; support the integration of ART in primary 107 care settings; minimise the number of facility visits for ART collection; and, reduce 108 pharmacy queue waiting times for patients. 109 The specifics of shifting ART-related tasks from pharmacy personnel have not been 110 addressed in a systematic review. Available systematic reviews on task shifting focus 111 on clinical services where nurses and non-physician clinicians provide care 112 comparable to physicians 11 . We therefore synthesised the evidence for task shifting in 113 pharmacy personnel services, where non-pharmacy personnel undertake ART 114 distribution and medication adherence counselling. Table 1 provides definitions for 115 the pharmacy functions distribution and dispensing as they have been used in this 116 review. However, these definitions seem to overlap at facility level. 117

Term Definition
Dispensing • Dispensing is a controlled act that authorises one to select, prepare and provide stock medication that has been prescribed to a patient or client (or his/her representative) for • To dispense is to prepare and supply to a patient a course of therapy on the basis of a prescription 13 .
• Dispensing is preparation and distribution of a course of therapy to a patient, with appropriate instructions based on a prescription 13 .

Distribution
• At facility level, drug distribution mainly refers to drugs dispensed by licensed practitioners, such as nurses, doctors, pharmacists or pharmacy assistants and collected by a patient.
• At community level, drug distribution refers to trained lay people collecting pre-packed medications from the facility and delivering them to HIV patients in the community 12 .

248
The unit of analysis was the individual study participant for all the trials. 249 Dealing with missing data 250 251 We contacted study authors for one of the included trials to obtain information on the 252 composition of health professionals in the control groups involved in dispensing ART 253 in order to establish whether pharmacy personnel were part of the team. However, the 254 We used the I 2 statistic to measure heterogeneity among the trials. We planned to 259 explore substantial heterogeneity (I 2 >50%) by pre-specified subgroup analysis. 260 However, there was no evidence of serious heterogeneity. 261

263
We minimised the potential for publication bias by using a comprehensive search 264 strategy. As none of the meta-analysis include 10 or more studies, we did not assess 265 publication bias using a funnel plot 20 21 . 266

Description of interventions: Pharmacy personnel group:
Patients received this intervention on a monthly basis through the Da´der method which consisted a series of scheduled meetings where the pharmacist and patient addressed, reviewed, and solved drug-related problems for a 12month period. Patients received structured counselling from pharmacists on their prescription regimens at the time of their initial drug dispensing and at monthly refill visits. Key elements included: • Reviewing the prescription with the patient.
• Reviewing a card on which medications were colour-coded to facilitate recognition and reduce confusion that might arise from complicated drug names. • Reviewing the schedule, length and date of next appointment.
• Reviewing patient's understanding of the prescription by asking patient to describe it and giving patients verbal information on the expected side effects of their medications and instructing them to seek medical assistance by calling the pharmacist if side effects occurred.

Non-pharmacy personnel group:
Patients received Usual Care (UC) for ART drug distribution. The control group received UC for ART drug delivery. This included the following: At the first appointment, when patients received their medications, a nurse first provided information on: • The regimen, • How and when to use the medications, •

305
The included trials all had low risk of selection bias as a result of adequate 306 randomisation and allocation concealment. Selke et al 16 and Jaffar et al 25 were 307 judged to have low risk of performance bias for blinding because the reported 308 outcomes were objective and unlikely to be influenced by lack of blinding. However, 309 the Selke trial 16 did not give sufficient information to permit judgment of whether or 310 not detection bias was present. The Silveira trial 26 was judged to have a high risk of 311 performance and detection bias due to lack of blinding for participants, personnel and 312 outcome assessment respectively for all study outcomes. All trials have a low risk of 313 attrition bias as they did not have differential or large numbers of losses to follow-up 314 across the intervention arms. Selective reporting bias was judged to be low for the 315 study by Jaffar et al 25   and Jaffar 25 ) compared complex interventions that included, but were not limited to, 327 pharmacy and non-pharmacy personnel as shown in Table 5. There were few deaths 328 and the effect estimate was imprecise, with very wide confidence intervals ranging 329 from appreciable benefit to substantial harm. 330 331  1 We downgraded by one for serious indirectness: Two trials compared complex interventions that included, but were not merely limited to, pharmacy and non-pharmacy personnel. 2 We downgraded by two for serious imprecision: There are few events and the effect estimates have wide confidence intervals, ranging from appreciable benefit to harm. 3 We downgraded by one for serious imprecision: There are few events and the effect estimates have wide confidence intervals, ranging from appreciable benefit to harm.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Virological response 334 We found low certainty evidence that there may be no differences in virological 335 failure between the group cared for by non-pharmacy personnel and that cared for by 336 pharmacy personnel (RR 0.92; 95% CI: 0.73 to 1.15, figure 4), with no heterogeneity 337 (Chi 2 = 0.24; df = 2; p=0.89; I 2 = 0%). The quality of the evidence for this outcome 338 was low due to serious indirectness and imprecision as shown in Table 5. 339

341
We found low certainty evidence that there may be no difference in loss to follow up 342 between the groups (RR 1.13; 95% CI: 0.68 to 1.91), with no heterogeneity (Chi 2 = 343 0.54; df = 2; p=0.76; I 2 = 0%), figure 5). The quality of the evidence was low due to 344 serious indirectness and imprecision as shown in Table 5. 345

347
Selke 16 reported significantly more clinic visits in the pharmacy compared to the non-348 pharmacy personnel group (mean visits 12.6 versus 6.4), p<0.001. Although the non-349 pharmacy personnel group was found to have fewer clinic visits, the authors observed 350 that this group attended 64% more clinic visits than originally scheduled. The study 351 by Jaffar 25 found a high frequency of outpatient attendance (15, 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  year were higher in the pharmacy personnel group ($54) compared to non-pharmacy 371 personnel group ($18). Costs to access care included transport, lunch, childcare and 372 lost work time. The outcome cost was not pre-specified in the protocol, however, this 373 provides additional data that may be relevant for decision makers and was therefore 374 included for consideration. 375 Acceptability to pharmacy personnel, non-pharmacy personnel and 376 patients and harm, including error rates 377 378 These outcomes were not reported in the included trials. 379

382
Two cluster randomised clinical trials and one non-blinded randomised controlled 383 trial were included in this review of effects of shifting responsibility from pharmacy 384  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  In our meta-analysis, we found very low certainty evidence for the outcome mortality. 387 We are therefore not certain whether the intervention may impact on mortality or not. 388 We found low certainty evidence that there may be no difference in virological 389 response, and loss to follow up between non-pharmacy personnel and pharmacy 390 personnel distributing ART to patients. In addition, we did not find significant 391 differences in adherence to treatment. Selke et al 16 reports that the non-pharmacy 392 personnel group had significantly fewer all-cause sick visits to the clinic compared to 393 the pharmacy personnel group. Although the main difference between the Selke 16 and 394 Jaffar 25 trials was the level of education of the non-pharmacy personnel, it did not 395 seem to affect their overall performance 16 25 . 396 Additionally, the Jaffar trial 25 found the non-pharmacy dispensing strategy to be 397 cost-effective and cheaper to run than the pharmacy dispensing strategy by almost 45 398 US dollars per patient per annum and this accounted for only 6% of the total cost of 399 healthcare service expenditure for the intervention group 25 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Although issues related to drug-drug interaction, medication errors, pre-screening for 414 ART and some special cases of adherence to therapy may be beyond the capacity of 415 non-pharmacy personnel, targeted pharmacy support to non-pharmacy personnel 416 could offer benefits to patients having challenges with their medications especially 417 those who are taking concomitant medications with ART 34 . 418 This review and meta-analysis has shown that the use of non-pharmacy personnel 419 comprising either lay people or other health professionals such as nurses who are 420 given well-tailored and comprehensive short trainings with refreshers may be able to 421 support ART dispensing systems for HIV-infected patients, however, the low 422 certainty evidence implies that additional trial evidence may change our results. The 423 use of non-pharmacy personnel such as lay people for dispensing ART and 424 monitoring patients' health would enable identification of psychosocial features that 425 might be overlooked by physicians, nurses and pharmacists at the facility. Some of 426 these psychosocial problems such as gender-based violence, food insecurity and 427 alcohol abuse negatively impact on adherence and retention in care 16 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  Although it may not be possible to rule out publication and language biases in 436 systematic reviews, our electronic search was not restricted to language, setting or 437 publication status. 438 We identified three studies for inclusion with small sample sizes and event rates. The 439 trials did not specifically pose the question of task shifting to non-pharmacy personnel

448
We assessed the certainty of evidence using the GRADE approach 35 . Evidence from 449 this review should be applied with caution considering its very low and low quality. 450 We are therefore not confident enough to state that the estimates lie close to the true 451 value and it is likely that further studies would contribute to the body of knowledge to 452 inform these results. 453
Methods and analysis: We will search PubMed, CENTRAL, EMBASE, WHO Global Health Library and relevant grey literature for eligible controlled trials. Two authors will screen the search output, select eligible studies, assess risk of bias and extract data from included studies, resolving discrepancies by discussion and consensus. We will perform meta-analysis using both fixed and random effects models, investigate clinical and statistical heterogeneity, and assess our confidence in the overall evidence using standard Cochrane methods, including GRADE.
Ethics and dissemination: Only secondary data will be included in this review and ethics approval is not required. We will disseminate the review findings in various scientific fora, including peer-reviewed journals. The findings may help to inform policy makers in defining the scope of work of healthcare workers, and global recommendations for shifting the dispensing and distribution of ART from pharmacy to non-pharmacy personnel.

INTRODUCTION
Description of the condition By March 2015, 15 million (40.7%) of the estimated 36.9 million people living with HIV (PLHIV) globally were receiving antiretroviral therapy (ART). 1 Combination ART is effective for reducing HIV related morbidity and mortality as well as preventing HIV transmission. 2 Initiating ART early in the course of HIV infection has been associated with better health outcomes, both at patient and population levels. 3 4 Scale-up of ART in low and middle income countries has averted more than 5 million deaths; however, bottlenecks preventing universal access to ART still exist. One challenge is the critical shortage of human resources for health (HRH), including for the delivery of essential HIV related pharmacy services.
The WHO recommends a minimum of one pharmacist per 2300 population, but most countries in low-resource settings such as sub-Saharan Africa have not yet met this target. 5 In addition to the absolute shortage, it is likely that there is an uneven distribution of pharmacists in such settings, as is the case with other specialist healthcare workers who Strengths and limitations of this study ▪ To our knowledge, this is the first published protocol of a systematic review that will investigate the effects of task shifting from pharmacy to non-pharmacy personnel for dispensing or distributing antiretroviral therapy to patients living with HIV. ▪ The protocol was written according to the PRISMA-P (Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols) recommendations. ▪ The review findings may help to inform antiretroviral therapy guidelines by the WHO. ▪ The possible weakness of the planned review would be the limitations of included studies, for example, high risk of bias and heterogeneity of settings, designs and effects. Mbeye 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  tend to concentrate in urban areas and the private sector, further aggravating the HRH shortage. 6 For instance in South Africa, which is home to the largest number of PLHIV in any country in the world, in 2010 only 24% of registered pharmacists worked in the public sector where 80% of the population received care. 7 Description of the intervention Studies and programme reports indicate that involvement of pharmacy personnel in HIV care results in improved patient outcomes. For instance, in the USA, the use of a multidisciplinary team approach with pharmacists assuming a central role in ART initiation, dispensing and adherence counselling improved treatment outcomes such as viral load, patient retention and medication adherence. 8 The work of pharmacists includes supply management, dispensing and distributing medications, promoting adherence, identifying and preventing potential medication-related issues, and monitoring and reporting adverse events. In some settings, programmes have implemented alternative models of pharmacy services that shift selected tasks from pharmacy to non-pharmacy personnel. Such alternative models could potentially increase the number of health workers involved in ART distribution, adherence counselling and patient education, free more time for pharmacy personnel, support the integration of ART in primary care settings, minimise the number of facility visits for ART collection, and reduce pharmacy queue waiting times for patients. 9 However, the specifics of shifting ART related tasks from pharmacy to non-pharmacy personnel have not yet been addressed in a systematic review. We therefore plan to synthesise the evidence for task shifting in pharmacy services, where non-pharmacy personnel undertake ART dispensing and distribution and medication adherence counselling. For this systematic review, pharmacy personnel will include both pharmacists and pharmacy technicians. Pharmacy technicians constitute an important part of the pharmacy workforce in low and middleincome countries: a survey of 26 low and middle-income countries in 2011 revealed that pharmacy technicians constitute 10% (Nigeria) to 70% (Pakistan) of the pharmacy workforce. 10 How the intervention might work Within the last decade, several high HIV burden countries adopted task shifting strategies where nurses and non-physician clinicians initiate and maintain ART. 11 Although this has undeniably expanded access to ART, it is also increasingly essential that long facility waiting times and frequent facility visits to collect ART are addressed to alleviate the burden of care, both for patients and healthcare providers. 12 13 Recent studies in Uganda, Kenya and Mozambique have shown positive outcomes when non-health professionals (lay people) delivered ART at the community level. 14 In Mozambique the use of PLHIV for distributing ART, monitoring adherence, reporting outcomes and referring sick patients to health facilities yielded a retention rate of 97.5% among stable patients on ART. 14 In a cluster randomised trial in Uganda, the use of trained community health workers produced comparable results with facility-based ART programme in terms of patient retention, viral load suppression and mortality rate. 15 Similar findings were also obtained in Kenya and Uganda when lay providers were engaged in ART delivery. 15 16 Task shifting has therefore been seen as an achievable solution to the critical human resource shortages for scale-up of ART. 17 While it is imperative to increase the rate of recruitment and training of health workers as well as improve working conditions to reduce attrition and emigration, the HIV pandemic requires a more urgent measure to address the critical skills shortage. 18 Such measures may include shifting selected tasks (including dispensing and distributing ART and adherence counselling) from pharmacy to non-pharmacy personnel. The task shifting could free time for pharmacy personnel to focus on more technical functions such as supply management and pharmacovigilance.
Why this review is important Previous systematic reviews of task shifting for increasing ART access focused on clinical services where nurses and non-clinician physicians provide care. 11 Dependence on and shortages of pharmacists are also key constraints on ART expansion, but the specifics of task shifting for ART dispensing or distribution from pharmacy to non-pharmacy personnel have not been reviewed systematically. We will systematically review the scientific literature and assess the efficacy and safety of task shifting models that use non-pharmacy personnel in dispensing or distributing ART and assessing adherence to treatment of HIV infection.

OBJECTIVE
The aim of this review is to evaluate the efficacy and safety of shifting dispensing and distribution of ART as well as assessment of adherence from pharmacy to nonpharmacy personnel.

METHODS
This review protocol has been registered in the PROSPERO International Prospective Register of Systematic Reviews (http://www.crd.york.ac.uk/PROSPERO), registration number CRD42015017034.
Criteria for considering studies for this review Types of studies We will include randomised controlled trials (RCTs) and non-RCTs, irrespective of whether allocation to interventions occurred at the individual or cluster level.

Types of participants
Participants will be PLHIV receiving ART.

Types of interventions
We will include studies that evaluate the shifting of selected tasks from pharmacy personnel to non-pharmacy personnel. The selected tasks include dispensing and distribution of ART and adherence assessment. Pharmacy personnel will include both pharmacists and pharmacy technicians. Non-pharmacy personnel may include (but are not limited to) nurses, non-physician clinicians, and lay providers such as patient peer groups, community volunteers, PLHIV and community health committees.

Types of outcome measures Primary outcomes
The primary outcome for this review is risk of death.

Secondary outcomes
Our secondary outcome measures include: ▸ Virological suppression ▸ Number of all-cause sick visits made to the health facility, including for adverse events ▸ Loss to follow-up ▸ Adherence to ART (as measured within the study, eg, pill counts, recall methods, digital methods) ▸ Acceptability to pharmacy personnel, non-pharmacy personnel and patients ▸ Harm, including error rates.

Search methods for identification of studies
We will perform a comprehensive and exhaustive search of electronic databases and conference proceedings in an attempt to identify all relevant studies available by the search date, regardless of language of publication or publication status ( published, unpublished, in press or in progress).

Databases of peer-reviewed literature
We will search the following electronic databases, from 1 January 1996 to the search date: ▸ Cochrane Central Register of Controlled Trials (CENTRAL) ▸ Excerpta Medica Database (EMBASE) ▸ PubMed ▸ ISI Web of Science (Science Citation index) ▸ WHO Global Health Library, which includes references from AIM (AFRO), LILACS (AMRO/PAHO), IMEMR (EMRO), IMSEAR (SEARO) and WPRIM (WPRO). Along with appropriate Medical Subject Heading (MeSH) terms and relevant keywords, we will use the Cochrane Highly Sensitive Search Strategy for identifying reports of randomised controlled trials in MEDLINE, 19 and the Cochrane validated strategies for identifying references relevant to HIV infection and AIDS. To identify other study designs, the RCT string will be omitted. The search strategy will be iterative in that references of included studies will be searched for additional references. See table 1 for our provisional search strategy for electronic databases.

Conference databases
We will search conference abstract archives on the web sites of the Conference on Retroviruses and Opportunistic Infections (CROI), the International AIDS Conference (IAC) and the International AIDS Society Conference on HIV Pathogenesis, Treatment and Prevention (IAS), for all available abstracts presented at these conferences from 1996 to the search date.

Searching other resources
We will also search the references of relevant articles as well as the WHO International Clinical Trials Registry Platform (ICTRP) and Clinicaltrials.gov. We will contact relevant experts or organisations who may be aware of additional studies in this field.

Data collection and analysis
We will base the methodology for data collection and analysis on the guidance provided in the Cochrane Handbook of Systematic Reviews of Interventions. 19 Selection of studies for inclusion Two authors will read and assess the abstracts of identified publications for potentially eligible studies. We will obtain full text articles for all abstracts judged by at least one of the two authors, to be potentially eligible. Two authors will independently inspect these potentially eligible publications to establish the relevance of the article to the review according to the pre-specified criteria regarding study design, participants, interventions and outcome measures.

Data extraction and management
Two authors will independently extract data into a prepiloted data extraction form. The following characteristics will be extracted from each included study: Study details: Complete citations of publications associated with the study, start and end dates, location, study design characteristics, type of facility involved, investigators, funding sources, recruitment, method of randomisation, sequence generation, method of allocation concealment, blinding of participants and personnel, blinding of outcome assessment, length of follow-up, losses to follow-up, withdrawals or drop-outs and other relevant details.
Details of the intervention: training of the cadre of health workers who were dispensing or distributing ART, what training or other support or supervision they received and other relevant details.
Details of participants: Trial inclusion and exclusion criteria, numbers of participants entering the trial, sex, clinical staging, CD4 count and other pertinent details.

Assessment of risk of bias in included studies
We will assess the risk of bias in RCTs using the Cochrane risk of bias assessment tool for randomised studies. 19 For non-RCT studies, we will use the Cochrane Risk of Bias Assessment Tool for Non-Randomised Studies of Interventions (ACROBAT-NRSI). 20 We will resolve any disagreements between the authors conducting duplicate independent screening of search outputs, assessments of study eligibility, extraction of data, and risk of bias assessment by discussion and consensus. Should this fail to resolve the differences, a third author will arbitrate.

Measures of effect
We will calculate and report risk ratios for dichotomous and time-to-event data and mean differences for continuous data with their 95% CIs.

Unit of analysis issues
The unit of analysis will be the individual study participant. Cluster-randomised trials will be included in meta-analyses only after adjustments are made for design effect. Design effects for cluster-randomised studies will be corrected by using standard procedures, using the formula: design effect=1+(m−1)r, where m is the average cluster size and r is the intra-cluster correlation coefficient.
Dealing with missing data We will contact study authors if it is necessary to obtain data missing from published reports.

Assessment of heterogeneity
We will examine statistical heterogeneity between study results using the χ 2 test of homogeneity, with a significance α-level of 0.1. In addition, we will use the I 2 statistic to measure the amount of heterogeneity among the trials in each analysis. If we identify significant heterogeneity (ie, p<0.1), we will explore it by prespecified subgroup analysis. If heterogeneity persists, we will perform sensitivity analyses, report results separately and propose reasons for the observed heterogeneity.

Assessment of reporting biases
If any meta-analysis in our review includes 10 or more studies, we will assess the potential for publication bias using a funnel plot. 21 We will attempt to minimise the potential for publication bias through a comprehensive search of published and unpublished literature.

Data synthesis
We will conduct meta-analysis, if appropriate, using the Cochrane Review Manager software (RevMan [Computer program] The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, 2015; Version 5.3). If we find no significant statistical heterogeneity of effects, we will use the fixed effect method of meta-analysis. Otherwise, we will use the random effects model.

Subgroup analysis
In pooled results with significant statistical heterogeneity, we will explore the cause of the heterogeneity through subgroup analyses, with subgroups defined by type of intervention (eg, cadre of health provider), comparison group and region of study (eg, sub-Saharan Africa, Southeast Asia, etc).

Sensitivity analysis
We will conduct a sensitivity analysis to investigate the effect of excluding studies with high risk of bias, with a focus on bias introduced by inadequate allocation concealment, inadequate blinding of outcome assessment and substantial losses to follow-up.

Certainty of evidence
We will assess the certainty (or quality) of evidence using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach, 22 which defines the certainty of evidence for each outcome as "the extent of our confidence that the estimates of effect are correct". 19 The quality rating across studies has four levels: high, moderate, low and very low. Randomised trials are considered to be of high quality but can be downgraded for any of five reasons: risk of bias, indirectness of evidence, unexplained heterogeneity of effects, imprecision of effect estimates and high probability of publication bias. Similarly, observational studies are considered to be of low quality, but can be upgraded for any of three reasons. The quality level of a body of evidence can be increased if there is a large magnitude of effect, if all plausible confounding would reduce a demonstrated effect, and if there is a dose-response gradient.

Reporting of this review
The findings of this review will be presented in a number of ways. The study selection process will be summarised using a flow diagram, and if we identify 10 or more eligible studies, we will assess publication bias using funnel plots. Where appropriate, we will use risk of bias graphs, forest plots and GRADE summary of findings tables. The non-quantitative outcomes will be reported descriptively. We will provide tables of both included and excluded studies. We have prepared this protocol as recommended by the PRISMA-P (Preferred Reporting Items for Systematic Review and Meta-Analysis Protocols) guidelines 23 and will report the findings of the review as recommended by the PRISMA statement. 24

ETHICS AND DISSEMINATION
Since systematic reviews do not directly involve human participants, they do not require ethical clearance. 25 We will provide the findings of this review to the WHO, with the hope that they may guide policy recommendations of this normative agent regarding the shifting of ART dispensing or distribution from pharmacy to nonpharmacy personnel. Although the majority of national programmes in low and middle-income countries have adopted task shifting in ART care at different levels, there has been no global policy to guide the practice for task shifting from pharmacy to non-pharmacy personnel. We will also publish the findings of the systematic review in a peer-reviewed journal.

12-13
Data items 11 List and define all variables for which data were sought (e.g., PICOS, funding sources) and any assumptions and simplifications made.

12-13
Risk of bias in individual studies 12 Describe methods used for assessing risk of bias of individual studies (including specification of whether this was done at the study or outcome level), and how this information is to be used in any data synthesis.

13
Summary measures 13 State the principal summary measures (e.g., risk ratio, difference in means).

Study selection
17 Give numbers of studies screened, assessed for eligibility, and included in the review, with reasons for exclusions at each stage, ideally with a flow diagram.

15
Study characteristics 18 For each study, present characteristics for which data were extracted (e.g., study size, PICOS, follow-up period) and provide the citations.

16-18
Risk of bias within studies 19 Present data on risk of bias of each study and, if available, any outcome level assessment (see item 12).

20-21
Results of individual studies 20 For all outcomes considered (benefits or harms), present, for each study: (a) simple summary data for each intervention group (b) effect estimates and confidence intervals, ideally with a forest plot.

21-24
Synthesis of results 21 Present results of each meta-analysis done, including confidence intervals and measures of consistency.

21-24
Risk of bias across studies 22 Present results of any assessment of risk of bias across studies (see Item 15).
n/a DISCUSSION Summary of evidence 24 Summarize the main findings including the strength of evidence for each main outcome; consider their relevance to key groups (e.g., healthcare providers, users, and policy makers).

24-26
Limitations 25 Discuss limitations at study and outcome level (e.g., risk of bias), and at review-level (e.g., incomplete retrieval of identified research, reporting bias).

27-28
Conclusions 26 Provide a general interpretation of the results in the context of other evidence, and implications for future research.