Article Text
PDF
Abstract
Objectives To systematically review and provide estimates of the minimal important change (MIC) and difference (MID) for outcome tools in people with knee osteoarthritis (OA) after non-surgical interventions.
Design
A systematic review.
Data sources MEDLINE, CINAHL, Web of Science, Scopus and Cochrane databases were searched up to 21 September 2021.
Eligibility criteria We included studies that calculated MIC and MID using any calculation method including anchor, consensus and distribution methods, for any knee OA outcome tool after non-surgical interventions.
Data extraction and synthesis We extracted reported MIC, MID and minimum detectable change (MDC) estimates. We used quality assessment tools appropriate to the studies’ methods to screen out low-quality studies. Values were combined to produce a median and range, for each method.
Results Forty-eight studies were eligible (anchor-k=12, consensus-k=1 and distribution-k=35). MIC values for 13 outcome tools including Knee injury and Osteoarthritis Outcome Score (KOOS)-pain, activities of daily living (ADL), quality of life (QOL) and Western Ontario and McMaster Universities Arthritis Index (WOMAC)-function were estimated using 5 high-quality anchor studies. MID values for 23 tools including KOOS-pain, ADL, QOL and WOMAC-function, stiffness and total were estimated using 6 high-quality anchor studies. One moderate quality consensus study reported MIC for pain, function and global assessment. MDC values from distribution method estimates for 126 tools including KOOS-QOL and WOMAC-total were estimated using 38 good-to-fair-quality studies.
Conclusion Median MIC, MID and MDC estimates were reported for outcome tools in people with knee OA after non-surgical interventions. The results of this review clarify the current understanding of MIC, MID and MDC in the knee OA population. However, some estimates suggest considerable heterogeneity and require careful interpretation.
PROSPERO registration number CRD42020215952.
- STATISTICS & RESEARCH METHODS
- Knee
- Musculoskeletal disorders
Data availability statement
All data relevant to the study are included in the article or uploaded as online supplemental information.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
STRENGTHS AND LIMITATIONS OF THIS STUDY
We estimated minimal important change (MIC) (within-group), minimal important difference (MID) (between-groups) and minimum detectable change values using anchor, consensus or distribution methods papers, respectively.
This systematic review included a defined population of people with knee osteoarthritis, after non-surgical interventions.
High-quality anchor studies were used to contribute to MIC and MID estimates were assessed using a credibility tool specially designed to evaluate anchor method papers.
Consensus and distribution methods papers were evaluated using quality assessment tools suited to each method.
Median estimates were used to reflect the synthesised data due to data skewness.
Introduction
The efficacy of therapeutic interventions is commonly evaluated using statistical significance regardless of patient importance.1 To understand whether differences in outcome measures after treatment are important to patients, it is necessary to know what constitutes a minimum important change or difference for the individual or cohort. These changes and differences are called the minimal important change (MIC) and difference (MID). There are numerous outcome measures for knee osteoarthritis (OA) and many estimates of MIC and MID. However, these estimates can arise from different methodologies leading to variability, confusion and misinterpretation.2–5 Achieving clarity in this space is crucial as these values are used in regulatory and clinical decision-making.6 7 This systematic review aimed to provide estimates of MIC and MID for knee OA outcome measurement tools in people with knee OA after non-surgical interventions.
MIC and MID are defined as the minimum value of an outcome measure that the patient, clinician or relevant others perceive as an important change or difference.4 8 9 The MIC considers the change in a clinical outcome measure within a single group or an individual over time. In contrast, MID considers the difference between independent groups or between individuals.4 10–12 However, the terminology of MIC and MID is used inconsistently.13 The concept was first described by Jaeschke, who studied patients’ perceptions of preintervention and postintervention beneficial change.8 This concept later included both improvement14 and worsening.15
Three methods are used to estimate MIC and MID: anchor, consensus and distribution.6 16 For the anchor method, MIC or MID values are usually estimated by referencing the patients’ responses against an externally validated scale (‘anchor’).17 The ‘global rating of change’ is most commonly used as the anchor but other methods (proxy responses or performance based measures) are also used.18 The receiver operating characteristic (ROC) method for deriving an estimate from anchor questions has been suggested to be more precise for clinical settings than the mean change method.9 15 In the consensus method, values are directly estimated by a group of experienced clinicians or patients until a consensus is achieved.6 In the distribution method, values are estimated statistically, based on the variance of the outcome data using half the SD,19 1 SEM20 or minimum detectable change (MDC) which is based on SEM.6 21 The anchor method is widely considered to be the most valid because it is based on patient perception of what constitutes minimal change or difference.16 22 In this paper, we have included MIC and MID estimates from anchor-based and consensus-based papers as well as MDC estimates. MDC estimates, as a distribution measure, are less meaningful because they do not reflect patient perception, but they do estimate instrument error which is of value to researchers.6 21 For this reason, we included MDC as well as the anchor and consensus estimates.
Knee OA is a common cause of pain and disability.23 Outcome measurement tools that include the domains of pain, physical function, patient global assessment and imaging are recommended to determine the efficacy of therapeutic interventions in knee OA studies.24 MIC and MID values have been estimated for knee OA outcome measures in these domains using anchor, consensus and distribution methods with variable results. The variability of the methods used makes the selection of an appropriate estimate confusing for clinicians, researchers and regulatory bodies.2–4
The primary objective of this systematic review was to estimate MIC and MID for knee OA outcome measurement tools based on estimates from high-quality anchor studies only. The secondary objectives were to determine MIC and MID estimates based on consensus method and to synthesis MDC values derived from distribution methods.
Methods
This systematic review was designed and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement.25 The protocol was registered on PROSPERO (registration number: CRD42020215952).
Literature search
Five databases (MEDLINE, CINAHL, Web of Science, Scopus and Cochrane) were searched from each database’s respective inception to 21 September 2021. A comprehensive search strategy was developed to capture all relevant articles, and database-specific MESH terms were used. The search strategy was as follows. (*knee OR genu OR tibiofemoral OR patellofemoral) AND (osteoarthr* OR degenerat*) AND ((“MCIC” OR “MCID” OR “MCII” OR “MIC” OR “MII” OR “MPCC” OR “MPCD” OR “MPCI” OR “MDC” OR “SDC” OR “SDD” OR “CIC” OR “CID”) OR (“minim* clinical* important change*” OR “minim* clinical* important difference*” OR “minim* clinical* important improvement*” OR “minim* important change*” OR “minim* important difference*” OR “minim* important improvement*” OR “minim* perceptible clinical* change*” OR “minim* perceptible clinical* difference*” OR “minim* perceptible clinical* improvement*” OR “minim* detectable change*” OR “small* detectable change* " OR “small* detectable difference* " OR “clinical* important change*” OR “clinical* important difference*")). The records were exported to EndNote V.X9.2 for reference management.
Study screening and selection criteria
Covidence software (Covidence systematic review software, Veritas Health Innovation, Melbourne, Australia (www.covidence.org) was used to manage the selection process. Records identified in the search were uploaded and duplicates were removed. Screening of titles and abstracts, then full texts, were performed independently by two reviewers (MDCS and JMC) and conflicts were resolved by a third reviewer (JMS). Included studies incorporated any design that calculated MIC and MID for any knee OA outcome measurement tool considering improvement after non-surgical intervention for adults with knee OA, and using any calculation method: anchor, consensus or distribution methods. We included studies that reported MDC because MDC is considered as an estimate from the distribution method.3 Though distribution-based approaches such as MDC do not reflect the patients perception, MDC values are important for researchers to get some idea about instrument error.6 21 We considered studies with MIC or MID values for improvement only and excluded values for deterioration because improvement values are used to evaluate the efficacy of treatment. Studies were excluded if the data from participants with knee OA could not be separated from other conditions, for example, hip OA or other knee pathologies. Studies of MIC, MID and MDC were included even if they used a different terminology for example, minimal clinically important change for MIC, minimal clinically important difference for MID and smallest detectable change or difference or minimal detectable difference for MDC.
For consistency, we defined the MIC as the pre–post change of one group that is, threshold for those who responded that they had minimally improved on the anchor measure. The MID is defined as the difference (pre–post change) between two groups, that is, ‘minimally improved’ and ‘stayed the same’ groups using the anchor response as defined in previous studies.10 12 26 The MDC is the minimum change above the measurement error based on a given level of confidence.6 21 MDC values for a 90 or 95 CI are labelled as MDC90 or MDC95.
Quality assessment
The quality of the included studies was assessed according to their methodology. The quality of the anchor studies was assessed using the credibility instrument developed by Devji et al 27 which was designed to assess the credibility of anchor studies assessing MIC and MID of patient-reported outcome measures. However, we adapted this tool in the following ways. The credibility instrument includes five core criteria, namely: (1) The anchor is rated by the patient, (2) The anchor is interpretable and relevant to the patient, (3) The MIC or MID estimate of patient-reported outcome measure is precise, (4) The correlation between the anchor and the outcome measure reported by the patient is satisfactory and (5) The authors select a threshold on the anchor that reflects a small but important difference). We adapted criteria 1, 3 and 4. For criteria 1 and 4, we included both patient and clinician as relevant anchor respondents. For criteria 3, we included performance measures as well as patient-reported outcome measures .We considered the paper to be ‘high’ quality if at least three of the five criteria were ‘yes’, ‘definitely yes’ or ‘to a great extent’ and of ‘low’ quality if not.28 Consensus studies were assessed using the Critical Appraisal Screening Programme qualitative tool29 which is designed to assess qualitative studies and is well suited to consensus studies. The quality was rated as ‘high’, ‘moderate’ or ‘low’ based on reliability and credibility.30 Distribution studies were evaluated using the National Heart, Lung and Blood Institute, National Institute of Health) quality assessment tool for before-and-after (pre–post) studies with no control group31 and ratings included ‘good’, ‘fair’ or ‘poor’ based on reliability and credibility.31 The quality assessment of included studies was performed by one reviewer (MDCS) and a random sample of 20% had an independent second review (AMF or JMC) to improve the accuracy.32
Data extraction and analysis
We extracted study characteristics including sample size, participant demographics, details of the intervention, follow-up time, outcome measurement tools, calculation method and actual estimate reported based on the method (MIC or MID or MDC). In addition, we extracted the details of the anchor used in each study.
We extracted reported MIC, MID and MDC values from each study. We normalised the values to a 0–100 scale. If a study reported MDC as a percentage of the grand mean (MDC divided by grand mean percentage),33 we converted the data into MDC90 or 95 for the synthesis.
All data were synthesised and described as the median estimates. The median and range (minimum and maximum) of MIC, MID and MDC were calculated using multiple estimates from the included studies arising from different non-surgical interventions, calculation methods, time points and anchors. Mean values were not calculated due to skewness of distributions.34 We excluded low-quality anchor studies from the median MIC and MID synthesis. Furthermore, we conducted a subanalysis to determine median MID based on the ROC method where available because the ROC estimates are considered to be more precise than mean-change estimates and recommended at both individual-level and group-level analyses, and in clinical settings.9 15 Though we planned to conduct a subanalysis to determine the effect of follow-up time on MIC and MID, we were unable to do reliable rate estimates because of a limited number of studies. Therefore, we plotted the values against time including only studies where the outcome measures were assessed at three time points or more.
Patient and public involvement
This is a systematic review. Patients or the public were not involved in this study.
Deviations from the protocol
The protocol registered in PROSPERO lists searches in MEDLINE, Embase, CENTRAL (Cochrane Central Register of Controlled Trial), Web of Science and CINAHL. However, Embase ceased to be available to the research team, so, Scopus was substituted. The data synthesis plan to assess MIC and MID in terms of standardised mean difference was not performed due to the skewness of the distributions. Therefore, we reported median and range for each measure without comparison. The planned meta-analysis was prevented by the skewness and homogeneity of the data and a decision was made to follow a simple descriptive approach using median estimates that was more accessible.28
Results
Study selection
The search yielded 2376 studies and after duplicates were removed, 1059 records were screened. Two hundred and seventeen studies were screened in full-text review resulting in 48 eligible studies (k=48) (figure 1). No further studies were identified after checking the reference lists of included studies.
Flow diagram of study selection (PRISMA).25 PRISMA, Preferred Reporting Items for Systematic Reviews and Meta-Analyses.
Included studies calculated MIC and MID by anchor method (k=12), by consensus method (k=1) and MDC by distribution method (k=35).
The methodological quality of included studies
Most anchor studies (k=10) were of high quality and two were of low quality35 36 (online supplemental file 1). The quality of the consensus study (k=1) was moderate (online supplemental file 2). The quality of distribution studies ranged from good (k=11) to fair (k=24) (online supplemental file 3).
Supplemental material
Study characteristics of included studies
All the anchor studies were observational prospective cohort studies14 15 35 37–43 and two of them were nested within randomised controlled trials36 44(table 1). The number of participants in each study ranged from 41 to 1606. The mean age and body mass index ranged from 57.1 to 67.9 years and from 28.1 to 33 kg/m2, respectively. The interventions used in these studies were rehabilitation, exercise, physiotherapy and non-steroidal anti-inflammatory drugs. The follow-up time ranged from 7 days to 1 year. Eleven studies used global rating of change as the external anchor while one study44 used multiple anchors. Most anchor studies (k=6)36–39 42 44 reported MID values only, four studies14 40 41 43 reported MIC values only and two studies,15 35 reported both MIC and MID. Four of these studies35 37 41 42 also reported MDC values. Moreover, studies used different anchor questions and group classifications when calculating MIC and MID. For example, one MIC study35 considered the minimal improved response group as ‘my knee has got better’ and another study40 considered the response group as both ‘good’ and ‘excellent’ improvement groups (online supplemental file 4).
Characteristics of included studies
The consensus study (k=1)45 (table 1) used a questionnaire to survey 27 clinicians from a range of specialities (orthopaedic (38%), rheumatology (33%), internal medicine (19%) and other (9%)). The clinicians were asked about MIC values for pain, function and global assessment for severe knee OA. However, participants were not asked to consider time duration nor the interventions. MIC was termed ‘minimal clinically important improvement’ in this study.
Most distribution studies (k=30) that reported MDC, were test–retest observational studies46–75 assessing the reliability of the outcome tool, and five used datasets from interventional cohort studies76 77 and randomised controlled trials33 78 79 (table 1). In the distribution studies (k=35), the number of participants in each study ranged from 8 to 135. The mean age and body mass index ranged from 50.5 to 70.3 years and from 22.7 to 35 kg/m2, respectively. Studies estimated MDC90 and MDC95. The follow-up time ranged from the same day to 1 year.
The MIC estimates derived using the anchor method
The median MIC for 13 tools (with subscales) were calculated based on 5 high-quality anchor studies14 15 40 41 43 using 23 estimates (table 2). These estimates were based on different underlying calculations, follow-up time and anchor questions. Methods for calculating MIC included: mean change (pre and post mean change of the minimally improved group),8 15 41 75th centile value of the mean change of the group40 43 and 75th centile value adjusted with the baseline score, age and disease duration.14 Most studies included one follow-up time (range: 7 days to 4 weeks), but one study15 reported MIC at two time-points (26 and 52 weeks). One anchor question was used in most studies, however, two studies15 40 reported different MIC values based on two different anchor questions (general health status and functional state).
MIC values of knee osteoarthritis outcome tools derived using the anchor method
The MID estimates derived using the anchor method
The median MID for 23 tools were calculated based on 6 high-quality anchor studies15 37–39 42 44 46 using 83 estimates (table 3). These estimates were based on different underlying calculations, follow-up time and anchor questions. Methods for calculating MID included: ROC method80 only (k=3),38 39 42 mean change (Redilmier and Lorig) method only (pre–post mean change difference between two groups)81 (k=1),37 both ROC and mean change methods (k=1),15 and mean change in T-scores with multiple anchors (k=1).44 Most studies (k=4) included one follow-up (range: 4–12 weeks), but two studies included MID at multiple time points for example, 2, 6 and 12 months.15 42 One anchor question was used in most studies, however, one study used multiple anchors.44
MID values of knee osteoarthritis outcome tools derived using the anchor method
MID estimates based on the ROC method were reported and compared with MID estimates for all methods. Overall, 4 of 6 studies15 38 39 42 (67%) used the ROC method. The ROC estimates were the same as the overall estimates in most cases (table 4).
Comparison of ROC method-based MID estimates with overall estimates
The effect of follow-up time on MIC and MID
There were insufficient data to establish reliable rate estimates for the effect of time. The MIC of Knee injury and Osteoarthritis Outcome Score (KOOS)-pain and KOOS-quality of life (QOL) and, the MID of KOOS-pain, KOOS-QOL, Knee Outcome Score (KOS)-activity of daily living (ADL), Lower Extremity Functional Scale (LEFS) and Western Ontario and McMaster Universities Arthritis Index (WOMAC)-total were assessed at more than three different time points. The MIC of KOOS-QOL and, MID of KOOS-QOL, LEFS and KOS-ADL appeared to increase with increasing follow-up time. However, MIC of KOOS-pain and MID of WOMAC-total appeared to reduce with follow-up time and KOOS-pain remained constant (online supplemental file 5).
MIC values derived using the consensus method
One consensus study45 reported that MIC for pain, function and global assessment were 20% of the maximum score.
The MDC estimates derived using the distribution method
The median MDC was calculated for 126 tools based on 38 studies (35 good-to-fair distribution and three high-quality anchor studies) using 308 estimates (online supplemental file 6). These estimates were based on different calculation methods and follow-up times. Four included studies reported MDC90 values only41 51 59 77 and 29 studies reported MDC95 values only. Five studies37 42 57 64 74 reported both the MDC90 and MDC95 values. Most studies (k=37) reported unadjusted MDC, while one study reported both the adjusted and unadjusted estimates.37 Six studies separately reported inter-rater/intrarater MDC values.46 52 56 60 67 73 Furthermore, three studies reported distinct values for two patient groups in each study, for example, the placebo group and the treatment group,78 the most painful and the least painful groups,70 and the groups that reported moderate improvement (‘great deal better’) and MCID improvement (‘somewhat better’).41 Most studies assessed the index (worst) knee, but one study based the estimate on all diseased knees (12 knees in 8 patients).52 Regarding the time point of MDC estimation, most studies (k=39) reported MDC estimates at one time point only, but one study reported MDC estimates at three time points (2, 6 and 12 months).42
Discussion
This systematic review provided estimates for MIC and MID of knee OA outcome tools after non-surgical interventions derived using anchor, consensus and distribution methods respectively. This review is unique in that it provides estimates for MIC, MID (based on high-quality studies) and MDC (from good-to-fair-quality studies) of knee OA outcome tools after non-surgical interventions. MDC was reported for a greater number of outcome measures (126) than for MIC (13) or MID (23). MID estimates based on the ROC method were similar to the overall median estimates, however, the majority of MID studies used the ROC method. Although we found that some MIC and MID appear to increase with follow-up time, this was not consistent.
The estimates for MIC and MID reported in this review are lower than those reported previously.82–84 Previous reviews which included knee replacement interventions82–84 produced higher estimates suggesting that knee replacement cohorts need more improvement to be satisfied. The MID values for WOMAC-pain and function in this review ranged from 7.1 to 21 and 11.3 to 14.9 (out of 100), respectively; compared with reviews of total knee replacements which reported values ranging from 4.0 to 47.9 and 1.8 to 33.0 (out of 100).82–84 This disparity may be due to differences in disease severity which has been previously reported based on baseline pain score.6 15 85 Therefore, our data are more applicable to patients and cohorts receiving non-surgical interventions. Furthermore, previous knee OA intervention studies have used MID estimates from studies with combined hip and knee OA.86 87 Given that MID is sensitive to disease type,6 our median estimates are likely to be more applicable to the knee OA population.
Some of the median estimates presented in this study suggest considerable heterogeneity. For example, the MID for WOMAC-pain was 8.7, but, the range extended from 7.1 to 21. These wide ranges are seen for other estimates including MIC for KOOS-pain, KOOS-QOL and LEFS. The median estimate was used because it is robust when data is skewed. However, the uncertainty which accompanies the wide ranges reported must be acknowledged.
MDC was reported for more outcome measures than MIC or MID. MDC is derived from data distribution only, unlike MIC and MID which are related to patients’ perception.17 22 Researchers may use MDC estimates as an option if MIC or MID are not reported. Yet, according to the results of this study and others, MDC can be larger or smaller than MID.4 10 Hence, researchers using MDC estimates from single studies to establish a sample size may overestimate or underestimate the number of participants required for a given power.
The ROC estimates were similar to the synthesised MID estimates which used all calculation methods. Our synthesised estimates were based on a combination of both the mean change81 and ROC methods.80 However, the ROC estimates are reported to be more precise and can be applied to both individuals and groups and are recommended in clinical settings.9 15 Moreover, the area under the curve of the ROC has the advantage of being able to interpret the level of confidence for the MID estimate from acceptable to outstanding discrimination between responders and non-responders.17 Therefore, we recommend using our median ROC based MID estimates where possible.
Although we found that some MIC and MID appear to increase with follow-up time, this was not consistent. Two previous studies15 42 suggested that there may be an effect of time, due to changing of perceptions over time (response-shift), especially in patients with chronic conditions.15 88 In addition, recall bias is affected by increased follow-up time and may also affect estimates.6 27 89 Therefore, although the consistency of follow-up time must be considered, more data are required to determine the effect of follow-up time on MIC and MID.
One of the studies included in this review used the consensus method. They reported that MIC was 20% of the maximum score for pain, function and global assessment,45 but, our anchor studies data suggest that MIC is highly variable (2.2–27.6 out of 100) depending on the outcome measurement. Therefore, the blanket application of 20% may not be suggested regardless of the tool used.
In this review, we considered only MIC, MID and MDC but there are other measures of clinical improvement. While the MIC and MID are used to assess meaningful clinical effects, recent reports have questioned the applicability of these concepts as they do not consider the costs, risks, benefits and inconvenience of the treatment. The smallest worthwhile effect (SWE) was developed using the benefit-harm trade-off method, described by Barrett et al.90 The SWE is defined as the smallest amount of improvement which is identified by the patient as worthwhile when considering the improvement outweighing risks and inconvenience91 and the estimates are always compared with natural recovery.92 However, only one study has reported SWE for people undergoing total knee replacement.93 Other studies have evaluated ‘patients acceptable symptom state’ (PASS) which is the symptom state that patients consider acceptable or when they feel ‘well’ after treatment.84 94 95 PASS estimates for WOMAC function are reported to be between 31 and 34.4.96 These values are much higher than our MIC median estimate of 17 (9.1–17.1). Although MIC and MID are still commonly used, the development of this field of research will enable value judgements as well as clinical judgements to be considered in the interpretation of clinical trials of interventions.
This systematic review should be considered in light of its limitations. The results of this review have been affected by heterogeneity of the included studies including: sample size, participant demographics, severity of knee OA, varied interventions, follow-up time and calculation methods. Median estimates were used because of the data skewness, but some of the ranges were wide, challenging the certainty of some of these estimates. The reader is encouraged to take the range of the data into account when interpreting the results. Previous evidence suggests that data from follow-up times of less than 1 month are more reliable.27 97 98 However, we included all estimates regardless of follow-up time. Statistical analysis was not conducted to determine the effect of follow-up time due to limited available data, but this is an interesting area for further study. The grey literature was not searched for this review.
This review presents median estimates for MIC, MID and MDC of people with knee OA following non-surgical interventions. A subset of MID estimates based on the ROC method is reported and, where available, this estimate is recommended as the most precise for both individual and group analyses and clinical settings. MDC estimates are available for more outcome measures but are purely statistical and arguably less applicable. This review clarifies the current understanding of MIC, MID and MDC in the knee OA population. However, some estimates suggest considerable heterogeneity and require careful interpretation.
Data availability statement
All data relevant to the study are included in the article or uploaded as online supplemental information.
Ethics statements
Patient consent for publication
References
Footnotes
Twitter @denika.silva, @KneeMotion
Contributors MDCS, DMP, AMF and JMS conceptualised and designed the study. MDCS ran the database search. MDCS and JMC did the screening, full-text review and data acquisition. MDCS, DMP, AMF and JMS were involved in data analysis and Interpretation of the data. MDCS wrote the first draft of the manuscript. MDCS is responsible for the overall content as the guarantor. All authors reviewed the draft, read and approved the final manuscript.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.