Article Text
Abstract
Objectives Antimicrobial resistant (AMR) infections are a major public health problem and the burden on population level is not yet clear. We developed a method to calculate the excess burden of resistance which uses country-specific parameter estimates and surveillance data to compare the mortality and morbidity due to resistant infection against a counterfactual (the expected burden if infection was antimicrobial susceptible). We illustrate this approach by estimating the excess burden for AMR (defined as having tested positive for extended-spectrum beta-lactamases) urinary tract infections (UTIs) caused by E. coli in the Netherlands in 2018, which has a relatively low prevalence of AMR E. coli, and in Italy in 2016, which has a relatively high prevalence.
Design Excess burden was estimated using the incidence-based disability-adjusted life-years (DALYs) measure. Incidence of AMR E. coli UTI in the Netherlands was derived from ISIS-AR, a national surveillance system that includes tested healthcare and community isolates, and the incidence in Italy was estimated using data reported in the literature. A systematic literature review was conducted to find country-specific parameter estimates for disability duration, risks of progression to bacteraemia and mortality.
Results The annual excess burden of AMR E. coli UTI was estimated at 3.89 and 99.27 DALY/100 0000 population and 39 and 2786 excess deaths for the Netherlands and Italy, respectively.
Conclusions For the first time, we use country-specific and pathogen-specific parameters to estimate the excess burden of resistant infections. Given the large difference in excess burden due to resistance estimated for Italy and for the Netherlands, we emphasise the importance of using country-specific parameters describing the incidence and disease progression following AMR and susceptible infections that are pathogen specific, and unfortunately currently difficult to locate.
- epidemiology
- infectious diseases
- public health
- qualitative research
- statistics & research methods
- urinary tract infections
Data availability statement
Data are available in a public, open access repository. The code used to calculate the incidence in Italy, the spreadsheet in which the figures were created and the spreadsheets used to calculated the excess burden are available in the Github repository https://github.com/NoorGo/ExcessBurden.
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
- epidemiology
- infectious diseases
- public health
- qualitative research
- statistics & research methods
- urinary tract infections
STRENGTHS AND LIMITATIONS OF THIS STUDY
The strength of this study method is the application of the novel method to estimate the excess burden of an infection in two example countries to demonstrate its use.
We used country-specific and pathogen-specific parameters to estimate the excess burden of disease (BoD).
National-level surveillance data of the Netherlands informed the estimation of the incidence of resistant E. coli urinary tract infections.
The main limitation was that assumptions had to be made for some country-specific parameters for which no suitable studies were found; this might have affected the estimated difference in the burden and excess burden between the Netherlands and Italy.
Most parameter estimates used in the calculation of excess BoD were derived from studies in hospital populations whereas data from studies in the general population could lead to more accurate and better generalisable estimates.
Introduction
Information on incidence and burden of disease (BoD) of infections with antimicrobial-resistant (AMR) bacteria is valuable for setting public health priorities, designing and evaluating interventions.1 However, such information is scarce,2 even though AMR has been identified in the European Union/European Economic Area (EEA) as a major public health problem.3
To gain insight into the AMR-associated BoD, composite health measures, such as the disability-adjusted life-years (DALYs) measure, which can be derived from clinical pathway progression models, and suitable data on mortality and morbidity4 5 are useful. Composite health measures allow diseases and their infectious causes to be ranked in terms of burden,6 and—particularly if based on incidence data—also facilitate measurement of the impact of public health interventions. In the case of AMR, the DALY approach can also be applied to compare the burden across resistant infectious agents, between countries or regions, and across time.
Attempts to comprehensively estimate the BoD of resistant infection using DALY have only recently been published, and report a large burden of resistance.2 To calculate BoD, parameters for, among others, the chance of progression from acute infection to severe health outcomes, the risk of mortality and duration in each health outcome are needed. These parameter values are needed for AMR and antimicrobial susceptible (AMS) infections separetely because some previous studies observed worse outcomes for AMR infections. On the other hand, a study on complicated P. aeruginosa urinary tract infections (UTIs) and multidrug resistance did not find a difference in 30-day mortality and another study on bacteraemic UTI did also not find an association between 30-day mortality and resistant profiles.3 7 Parameters to calculate the BoD using the DALY measures should be chosen based on study findings of specific pathogens and infection site to provide more insight on whether resistance increases BoD. Moreover, estimating the BoD brings conceptual challenges, such as determining to what health state resistant infections should be compared, as discussed previously by de Kraker and Lipsitch. For instance, AMR infections can be compared with AMS infections or to the situation in which the infections do not occur and the choice of comparison method influences the calculated excess harm caused by resistance.8
The aim of this paper is to introduce a method to calculate the excess BoD. By ‘excess BoD’ we mean the mortality and morbidity (computed as DALY) associated with resistance, over and above the mortality and morbidity associated with infection by the same—but AMS pathogen. In this approach, AMS infections with incidence identical to that for AMR infections serve as a counterfactual to estimate the additional health burden that is attributable to resistance. Our approach is new in that we combine country-specific incidence numbers from surveillance data with country-specific parameter values to calculate the excess BoD for infection caused by a specific resistant pathogen. Methods in previous studies did not include country-specific and pathogen-specific data to estimate the BoD. Subsequently, the method is demonstrated by calculating the excess BoD for a single infection site (UTI) and a single bacterial agent (E. coli) as AMR compared with AMS E. coli, where a resistant E. coli UTI is defined as a tested urine sample containing E. coli which produce extended spectrum beta-lactamases (ESBLs) as confirmed by a laboratory. The excess BoD of these infections was assessed for two countries: Italy, which was previously estimated to have the highest antibiotic-resistant BoD in the EEA, and the Netherlands, which was ranked third from last in the list of highest antibiotic resistant BoD in the EEA.2 Note that our goal is to illustrate how the methodology can be applied to countries with differing AMR E. coli prevalence and with differing surveillance data available, and not to conduct a formal comparison of these countries in terms of excess burden. We selected UTIs because they are among the most frequent infections in both the outpatient and inpatient setting and we choose E. coli UTIs specifically because UTIs are frequently caused by E. coli.9 10 Furthermore, UTI is a common cause of sepsis a life-threatening complication with a very high mortality rate for all ages.11 The excess BoD for AMR E. coli has not been estimated previously for the Netherlands and Italy using national-level data and country-specific parameter values.
Methods
We begin by reviewing the parameter requirements for DALY estimation, then describe the systematic reviews that were carried out to locate country-specific parameter values, and finally detail the calculation of AMR E. coli UTI incidence for both target countries.
Outcome trees
We modified an existing outcome tree (OT) developed by the European Centre of Disease Control (ECDC) describing the clinical progression pathway for UTI,2 shown in figure 1. We describe the separate transition probability parameters, disability durations (DDs), and disability weights (DWs) that are needed to quantify the BoD, in DALYs, due to infection with either the susceptible or resistant strain as shown in figure 1. The method simulates an incidence of AMS E. coli that is equal to resistant E. coli to estimate what the additional burden would be of resistant E.coli compared with the same number of AMS E. coli infections. Our excess BoD approach involves subtracting the estimated annual DALY for AMS UTIs, using the ‘susceptible’ version of the OT, from the annual DALY for AMR E. coli UTIs, using the ‘resistant’ version of the OT, while simulating that incidence is identical. We simulate this identical incidence for calculating the excess burden, because we assume that a person would have had a susceptible infection in case they would not have had a resistant infection. Thus, only the OT parameters for resistant and susceptible E. coli UTIs differ.
The starting health outcome of the OT is a symptomatic UTI, after which patients can recover, or progress to secondary bacteraemia, and following bacteraemia progress to several long-term sequelae or death.
DALY parameters and calculation
The principal ‘input’ to the DALY computation is the number of incident cases, in the current example the number of people experiencing an AMR E. coli UTI in 1 year. Transition probabilities between symptomatic UTI and all subsequent health outcomes are required. These estimates are required for AMR and AMS E. coli UTI separately because the probability of transitioning from one health state to another is often not the same for AMR and AMS infections. We use the notation P(Outcome2|Outcome1) to indicate the progression probability from Outcome1 to Outcome2. For instance, P(Bact|UTI) is the probability of progression to bacteraemia given symptomatic UTI. No mortality risk is assumed following a UTI that does not progress to secondary bacteraemia. The OT specifies mortality risk as the parameter P(Death|Bact).
In general, DALYs are calculated as follows: the years of life lost (YLL) are added to the total years lost due to disability (YLD) which is calculated by summing over the YLD for each (non-fatal) health outcome in the OT:
YLL=No. deaths×life expectancy (LE) at age of death.
Ni=the yearly incidence of health outcome i.
DWi=the average disability weight of health outcome i.
DDi=Average duration of disability i.
DALY combines the YLL due to premature mortality and YLD, which captures time lived by an individual in less than full health. A loss of 1 year of full health is equivalent to one DALY.12 For the computation of YLDs, DWs and DDs for each health outcome are required. Given availability of hospital length of stay (LOS) data in the literature, LOS data can serve as a measure of DD if the health state can involve hospital stay. When a patient can transition to more than one, simultaneously experienced, health outcome (so-called ‘internal comorbidity’), such as the long-term sequelae following secondary bacteraemia (figure 1), DWs of the overlapping health outcomes can be adjusted to take this into account.13 We decided a priori to adopt the same DWs as used by ECDC.2 14
The risk of recurrent UTI episodes per patient was incorporated using a simple multiplier approach. Dealing with recurrence is necessary as the incidence data consist of the number of patients with at least one UTI episode in 1 year, and the transition probability from UTI to bacteraemia is defined per patient, but the annual BoD will depend on the total number of episodes in a year. Therefore, given an average annual number of episodes per patient, j, the total duration of time spent in the health outcome symptomatic UTI in a year is defined as j×DD[UTI].
For the computation of YLL, normative LE values by age-group at death are needed. Consistent with previous BoD exercises,2 15 we chose to use the Global Burden of Disease project (GBD-2010)16 values.
All BoD measures were estimated using pre-existing software, the BCoDE toolkit V.1.4.17 In this software, Monte-Carlo simulation with 1000 iterations is employed to compute 95% uncertainty intervals around the BoD. We present the excess BoD and resistant BoD as DALY per 100 000 population (to allow comparison between countries), DALY per 100 cases (for assessing the patient-level burden; also useful for between-country comparison), YLDs and YLL.
Systematic reviews
We performed systematic literature reviews to locate parameter estimates for the risk of progression to bacteraemia, risk of progression to health states following bacteraemia, LOS, other indicators of DDs and mortality risk. The systematic reviews, performed separately for the Netherlands and Italy, are described in detail in online supplemental appendices 1–3, figures S1 and S2.
Supplemental material
AMR E. coli UTI incidence in the Netherlands
Data of 2018 from ISIS-AR, a laboratory based AMR surveillance system in the Netherlands18 were used to estimate AMR E. coli UTI incidence. ISIS-AR contains results of antimicrobial susceptibility testing of bacterial isolates routinely tested in medical microbiology laboratories in the Netherlands. ISIS-AR contains data on all consecutive samples of patients, sampled in hospitals (inpatient and outpatient), general practices and long-term care facilities.19 The coverage of the surveillance system is shown in online supplemental figure S3. ISIS-AR contains data of 46 laboratory which represent around 80% of the Dutch hospitals.20
AMR E. coli UTI incidence was defined as the number of persons having at least one urinary AMR E. coli isolate in 2018 per 1000 population. The incidence was stratified by sex and 5-year age-group. Online supplemental table S1) shows the data used per sex and age-group to calculate the incidence and recurrence rate. Incidence is thus calculated as the total number of resistant E. coli UTI in 2018 per sex and age group divided by the number of inhabitants of the Netherlands per sex and age group in 2018, and subsequently multiplied by 1000. An algorithm was created which calculated the days in between two urinary test samples of the same patient to determine if two consecutive tests had been conducted within 2 weeks in the same patient. If the urinary samples were more than 2 weeks apart, the UTI was labelled as recurrent and then only one isolate was counted. If two tests conducted for the same individual were more than 2 weeks apart, the UTI was defined as recurrent. As a sensitivity analyses, we also show the incidence if we would have defined a recurrent UTI as being longer than 3 months apart. We estimated the average number of recurrent episodes per patient per year. Moreover, we estimate the total incidence of E. coli UTIs regardless of resistance to indicate the percentage of resistant E. coli UTIs in 2018. The analysis to estimate the incidence were performed in R V.4.0.2.
Estimation of AMR E. coli UTI incidence in Italy
No Italian source comparable to ISIS-AR was found. Therefore, we took seven steps to calculate the incidence.
Step 1. We took the number of UTIs (n=57 271) reported in a study that retrospectively used primary care electronic medical records of around 1.1 million Italian general practitioner (GP) patients from 1 January 2016 through 31 December 2016.21 The coverage of this study around 2%22 and the Italian population size in 2016 reported on ISTAT was used to estimate the total number of patients with a UTI in the entire population in 2016.22
Step 2. The sex and age-group distribution from a study on UTIs in 2015–2019 in an academic Italian high-volume centre, namely the University Hospital ‘San Giovann di Dio e Ruggi d’Aragona’ in Salerno, was used to distribute the total estimated UTIs among women (62.33%), men (37.77%) and age-groups.23
Step 3. The number of E. coli UTIs was calculated assuming that 59.9% of UTIs were caused by E. coli as reported in Cardone et al24 which we identified in the systematic review (online supplemental appendix 1).24 From January 2013 to June 2017, Cardone et al24 included urine samples collected in the emergency department and used two inclusion criteria. The urine samples had to be collected in (1) patients with UTI symptoms and (2) it had to be their first positive culture urine culture in a given year.
Step 4. A large study from April 2007 to April 2008 in 20 microbiology laboratories found that 15.1% of E. coli bacteraemia produced ESBL25 and this percentage was then applied to the results of Step 3 to estimate the AMR E. coli UTI incidence.
Step 5. To estimate the incident number of AMR E. coli UTIs per 5-year age category as needed for the BCoDE toolkit V.1.417 (eg, 10–14, 15–19), we distributed UTIs within the age-categories used in Serretiello et al23 proportionally according to the age-category and sex-specific population size.
Step 6. To calculate the incident number of AMR E. coli UTIs including clinical and outpatient cases, we used the same ratio of hospital to GP cases and outpatient to GP cases, sex and age-stratified, as in the Netherlands. We used the same recurrence rate as we found in the Netherlands, as we were unable to identify a better estimate.
All calculations for the Italian incidence can be found online (https://github.com/NoorGo/ExcessBurden).
Patient and public involvement
There was no direct patient or public involvement in the design of this study.
Results
The results of the systematic review are discussed in online supplemental appendix 4, and the identified parameter values are described in table 1.
Parameters
The Netherlands
P(Death|Bact) for AMS E. coli was 11.3% and for AMR E. coli 27.5%. We estimated the DD(UTI) for AMS E. coli at 5.1 days (95% CI 4.3 to 5.9) and for AMR E. coli at 8.7 days (95% CI 7.0 to 10.8). DD(Bact) for AMS E. coli is 2.9 days (95% CI 1.7 to 4) and for AMR E. coli 7.9 days (95% CI 3.5 to 13.0). All parameters and their sources can be found in table 1.
Italy
P(Death|Bact) for AMS E. coli was 5.47% and for AMR E. coli this was estimated to be 26.5%.5 We were only able to find a single Italian parameter value for DD(UTI), which did not distinguish between AMS E. coli and AMR E. coli (10.7 days, IQR (7–17)). DD(Bact) for AMS E. coli was estimated at 13 days (SD=9) and for AMR E. coli at 20 days (SD=17).
Excess burden
The Netherlands
Per 100 0000 inhabitants we found an excess burden of 3.9 DALY/100 000. The YLL component accounted for 98% of the excess BoD. We found 39 (59%) excess deaths compared with the AMS model. Figure 2 shows the YLL and YLD for the Netherlands, while assuming equal incidence of susceptible and AMR E. coli. Per 100 cases the excess burden was estimated at 8.8 DALY/100 cases. The greatest excess burden was observed for bacteraemia (658 DALY) as can be seen in figure 3 which shows the excess burden for each of the six specified health outcomes in the clinical pathway progression model for UTI. Sex-group and age-group differences in both BoD and excess burden were apparent (figure 4); the latter was two times greater for females (527 compared with 257 DALY per year in the population of males).
Italy
Per 100 000 inhabitants In Italy, we estimated an excess burden of 99 DALY/100 000. The YLL component accounted for 99.7% of the excess burden and 2786 (77.0%) excess deaths were estimated. Per 100 cases the excess BoD was estimated at 12.3 DALY/100 cases. Figure 5 shows the YLL and YLD for Italy for AMR E. coli UTI and when simulating equal incidence of the counterfactual AMS E. coli UTI. Figure 6 which shows the excess burden for each of the six specified health outcomes in the clinical pathway progression model for UTI. Sex-group and age-group differences in both BoD and excess burden were apparent (figure 7); the excess burden was 1.3 times greater for females (34 036 compared with 26 184 DALY). The 5-year age-group contributing the largest estimated excess BoD was 55 to 59-year-old women and 65 to 69-year-old men (5990 and 6041 DALY, respectively).
Resistant burden
The Netherlands
In the Netherlands a total of 9623 AMR E. coli UTIs occurred in 2018 based on the tested isolates in ISIS-AR, corresponding to an annual incidence of 0.56 AMR E. coli UTIs/1000 inhabitants. This incidence includes recurrent UTIs. These UTIs occurred in 7586 unique patients, resulting in an annual incidence of 0.44 AMR E. coli UTIs/1000 inhabitants, excluding recurrent UTIs. Online supplemental table S1 was used to calculate the AMR E. coli UTI incidence and recurrence rate per age and sex group. Of the unique AMR E. coli UTIs, 64.2% occurred in women and 62.3% in people aged 65 years or older. The total number of E. coli UTI in 2018 was 199 441 and excluding recurrent UTI 165 258. The incidence including recurrent UTIs was 11.61/1000 inhabitants and 9.62/1000 inhabitants excluding recurrent E. coli UTI. The percentage resistant E. coli UTIs was 4.8% including recurrent UTIs and 4.6% excluding recurrent UTIs of the total number of E. coli UTIs in 2018. Online supplemental table S2 was used to calculate the E. coli UTI incidence and recurrence rate per age and sex group. In the sensitivity analysis in which we assumed a recurrent UTI to be more than 3 months apart we found an overall incidence of 0.47 AMR E. coli UTIs/1000 inhabitants and an incidence of 0.44 AMR E. coli UTI/1000 inhabitants excluding recurrent UTIs. Online supplemental table S3 shows the data of the incidence calculation for the sensitivity analysis.
Per 100 000 inhabitants in the Netherlands, we estimated an AMR E. coli UTI incidence of 9.2 DALY/100 000 inhabitants (95% UI 8.5 to 9.9). The YLL component accounted for 71.0% of the resistant BoD and 66 deaths were estimated. The sex-aggregated and age-aggregated BoD for AMR E. coli UTI in the Dutch population in 2018 was estimated at 1581 DALY (95% UI 1467 to 1701), or per 100 cases 20.8 DALY (95% UI 19.3 to 22.3) DALY (table 2). The resistant BoD for females was approximately two times that for males (1011 compared with 570 DALY) as shown in figure 4. Figure 3 shows the BoD for the specified health outcomes in the UTI clinical pathway progression model. The health outcome with the highest BoD for UTIs caused by AMR E. coli was bacteraemia (1127 DALY, 95% UI 1020 to 1238).
Italy
In Italy in 2016, we estimated 490 332 AMR E. coli UTI and an incidence of 8.1 UTIs/1000 inhabitants excluding recurrent UTI. In women, 56% of infections occurred and 44% occurred in people aged ≥65 years. Incidences per age and sex group can be found in tables 3 and 4.
In Italy, we estimated 192 DALY/100 000 (95% UI 181 to 203). The YLL component accounted for 66.9% of the resistant UTI BoD. For the AMR model 3617 (95% UI 3352 to 3884) deaths were estimated. The sex-aggregated and age-aggregated BoD for resistant AMR E. coli UTI in the Italian population in 2016 was estimated at 166 488 (95% UI 1 09 744 to 123 106) DALY, or 23.8 DALY per 100 cases (table 5). Just as for the Netherlands, the health outcome with the highest BoD for UTIs caused by AMR E. coli was bacteraemia (78 686 DALY, 95% UI 72 736 to 84 493), which also caused the larger excess burden (69 885 DALY) as can be seen in figures 3 and 6). The resistant BoD for females was approximately 1.3 times that for males (64 878 compared with 51 610 DALY). The 55 to 59-year-old women (9688 DALY) and 65 to 69-year-old men contributed the most (9765 DALY).
Discussion
We developed a method for estimating the excess BoD due to antimicrobial resistance, and applied the method to AMR E. coli UTI for two countries using country-specific parameters and incidence data. Using country-specific parameters for BoD estimates is crucial, as outcome measures (eg, mortality) are not only influenced by resistance itself, but can also be influenced by inappropriate treatment,8 and BoD depends on the prevalence of comorbidities, as well as country-specific differences in hospital and prevention policies.26 Previous large BoD studies such as Cassini et al2 did not use country-specific parameter estimates,2 whereas our results indicate that this is important. Two examples, among others that we found in our study, of why the use of country-specific parameters is important are that parameters such as the risk of death following bacteraemia and the disease duration of bacteraemia we found in the literature differed between Italy and the Netherlands. Subsequently these parameter differences between Italy and the Netherlands contribute to the differences in the excess burden between Italy and the Netherlands.
YLL accounted for most of the estimated AMR BoD in the Netherlands and in Italy (71% and 66.3%, respectively). A previous study on healthcare-associated (HA) infections, including bloodstream infections and UTI, based on data of Italy in 2016, also found that the majority of the BoD of AMR was attributable to YLL (79.7%).27 Regarding the burden of AMR in DALYs per 100 000 population, HA UTIs were estimated at 81.2 (69.0–94.4) DALYs/100 000 population. Both studies noted that UTIs were the second14 or most frequent27 HA in terms of incidence. The difference in excess BoD and in the AMR disease burden between the Netherlands and Italy that we found might be partly due to differences in treatment and resistance testing policies. Since our literature search, a Dutch study in eight hospitals was published suggesting a different mortality when comparing highly resistant to non-highly resistant bacteraemia, namely an RR of 1.08 (95% CI 0.48 to 2.41).28 This estimated mortality would imply that our estimates of the excess burden for NL may be over-estimated as the mortality risk difference of Rottier et al28 is smaller than that of van Hout et al.29 However, the CI of Rottier et al28 is relatively large and of the bacteraemia that were included, only 52% (n=1001) had the urinary tract as source and 62% (n=1190) was caused by E. coli.
Previous incidence estimates of resistant E. coli UTI based on data from 2015 indicate a third generation cephalosporin resistant E. coli UTI incidence in Italy that is 7.3 times higher than in the Netherlands, and a carbapenem resistant E. coli UTI incidence that is 12.3 times higher.2 In the current study, we estimated AMR E. coli incidence to be 18.3 times higher in Italy in 2016 than in the Netherlands in 2018. However, these previous estimates from Cassini et al2 were derived using a different approach2; namely, the incidence of blood-stream infection served as primary data, which was then extrapolated to specific infection sites and to each EU/EAA country. Also, in contrast to the study of Cassini et al,2 we use country-specific parameters which might be more suitable to indicate differences between countries in contributors to BoD. In a recent burden study DALYs attributable to and associated with bacterial AMR for 23 pathogens and 88 pathogen–drug combinations in 204 countries and territories in 2019 are provided. The authors mention the difficulty of understanding the burden of AMR when data are sparse and mention that because of data sparsity, they assumed the relative risk of death was the same for every syndrome, location and age group.30 We also found it difficult to locate country-specific mortality risks and other parameter values, and have argued that such data is important for accurate excess burden estimation at country level because country-specific parameters of for example mortality differ between Italy and the Netherlands.
In the paper of de Kraker and Lipsitch8 it is proposed to let the counterfactual in the BoD calculation depend on the type of intervention.8 The excess BoD method proposed in the current study defines the susceptible counterfactual to have identical incidence as resistant infection. This method could accordingly be useful for estimating the effect of reduction of broad spectrum antimicrobial use, vaccination against pathogens that are associated with antimicrobial use, introduction of new antibiotics, reduction of environmental or agricultural antibiotic use, and a combination of interventions targeted at the resistant strain. For these estimations, the model parameters could for example be adjusted and made specific for another pathogen and for a new intervention. The susceptible counterfactual is relevant under the assumption that resistant and susceptible strain compete as previously indicated to be the case by Godijk et al.31 Under the assumption that the replacement scenario is (mostly) occurring, the comparison group should be the same group of patients with infections caused by AMS pathogens to calculate excess mortality and BoD.32
A strength of this study is that we used national-level surveillance data of the Netherlands to calculate the incidence of resistant E. coli UTI. The use of these data enabled us to estimate the incidence of AMR E. coli as a basis for the BoD estimate. However, the use of these data harbour some limitations. First, the national coverage is less than 100%; therefore AMR E. coli UTI incidence is underestimated. Also, in Italy the study on which we based our estimation of the proportion of resistant E. coli is dependent on samples being taken, which is also sensitive to testing practice and does not have a complete national coverage. However, the BoD experienced by these ‘missed’ patients is expected to be small because their UTI resolved on first line treatment and therefore, they experienced little BoD. Their chance of progressing to bacteraemia would be minimal. Our DALY estimate is mostly determined by those patients that develop bacteraemia, which has an accompanying high risk of mortality. Second, the surveillance date are routine data from medical microbiological laboratories. The ISIS-AR data only contains UTIs that have been sampled and tested for resistance. In general practices in the Netherlands, UTIs are often sampled only when infection is not eliminated after initial treatment. A part of the UTIs, therefore, may have been missed in our study. However, since we based our calculations on AMR infections only, we do not expect that this has largely influenced our estimates.
Another strength of this study is that we not only propose a new method to calculate the excess BoD, but that we also apply our method to two countries to demonstrate its use and explore the methods drawbacks. A drawback of this method, as mentioned previously,33 is that it often is difficult to locate high-quality AMR surveillance data and country-specific AMR attributable mortality and morbidity parameters, as we experienced in the current study. Even though we performed a systematic review, we were not able to locate relevant studies and/or recent estimates for all parameters. In low-income and middle-income countries data scarcity is an even larger problem, which makes using country-specific parameter estimates and incidence data as we advise for out method harder, even though the use of country-specific parameters is probably even more important when comparing developing to developed countries. Apart from the higher percentage of resistance in Italy, the difference in parameter estimates between Italy and the Netherlands explain the larger BoD and excess BoD for Italy. For the Netherlands, available studies showed a smaller difference in the bacteraemia mortality rate for AMR E. coli and AMS E. coli (27.5% vs 11.3%, respectively) than for Italy (26.2% vs 5.5%, respectively). Moreover, for the Netherlands DDs for the UTI and bacteraemia health outcomes were shorter. However, we had to make multiple assumptions of the model parameters, especially for Italy, as country-specific data were not available for all estimates. These assumptions may also affect the estimated difference in the burden and excess burden between the Netherlands and Italy. For example, we used the same ratio of hospital to GP cases and outpatient to GP cases for Italy as for the Netherlands because we could not find specific data for Italy. However, in both the Netherlands and Italy antibiotics are not sold over the counter (in Italy there are some exceptions, eg, when the drug is necessary in order not to interrupt the treatment of a chronic disease34); thus prescriptions are required,34 35 and it is most common in both countries to first visit the GP, get treatment if necessary, and thereafter get additional care if needed. For this reason, we choose to use the same ratio of hospital to GP cases and outpatient to GP cases, even though there are some antibiotic prescription and treatment differences between the two countries. Furthermore, the estimated mortality following bacteraemia as a consequence of UTI was estimated to be 11.3% for AMS E. coli and for AMR E. coli 27.5% in the Netherlands,29 whereas a previous study in Finland, Sweden and Canada found a mortality rate of 9.2% of E. coli BSI with third-generation cephalosporin susceptibility and a mortality of 14.1% of E. coli BSIs with third-generation cephalosporin resistance.36 As we found few parameter estimates that were country-specific, we were unable to, for example, do a small meta-analysis, and get more valid estimates. Thus, our results should be interpreted with caution. The codes used to calculate the incidence in Italy, the excel in which the figures were created and the excel sheets used to calculated the excess burden are available on the Github repository (https://github.com/NoorGo/ExcessBurden).37
Moreover, the assumed 15.1% resistance prevalence E. coli UTIs in Italy is likely to be an underestimate, as other data from 2017 suggested around 75% of the E. coli isolates in Italy to be resistant to at least one antibiotic group and around 45% to be resistant to three or more antibiotic groups,38 however the 2017 prevalence was not specific for UTIs and we preferred to use UTI-specific AMR E. coli estimates. Future research would benefit from using more recent country-specific surveillance data, when it becomes available, to more accurately estimate AMR E. coli incidence.
In addition, parameter estimates were limited by restricted analysis of confounders.26 We did, however, stratify our results for age and sex. Moreover, we adjusted the risk of mortality following bacteraemia for age. Future research could use parameter estimates derived from the general population. Most estimates used in this study were derived from studies in hospital populations. Parameter estimates based on studies in the general population could lead to more accurate estimates that are better generalisable to the Dutch and Italian populations. For example, hospital patients presenting with a UTI may more likely progress to bacteraemia, due to an already weakened immune system, than individuals who present with a UTI at the GP. As we were unable to locate parameter estimates in the general population, we also recommend future research to focus on estimating these parameters. An example of such a study could be following GP patients who have a confirmed AMR or AMS E. coli UTI to estimate the probability of progression to bacteraemia and subsequent mortality.
To conclude, for the first time, we use country-specific and pathogen-specific parameters to estimate the excess burden of resistant infections. Given the large excess burden difference between AMR E. coli and AMS E. coli UTI, we emphasise the importance of using country-specific parameters describing the incidence and disease progression following resistant and susceptible infections that are pathogen-specific. Unfortunately, these parameters are currently difficult to locate.
Data availability statement
Data are available in a public, open access repository. The code used to calculate the incidence in Italy, the spreadsheet in which the figures were created and the spreadsheets used to calculated the excess burden are available in the Github repository https://github.com/NoorGo/ExcessBurden.
Ethics statements
Patient consent for publication
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors NGG, SAM and MCJB conceptualised the study. NGG conducted the literature review and performed the data analyses with the help of SAM. NGG generated the figures and drafted the manuscript. WA-vdK and AFS had access to the ISIS-AR data and supplied the required data for the incidence calculations. WA-vdK created online supplemental figure S3. SAM, MCJB, WA-vdK, AFS and EF reviewed the manuscript and performed a critical revision of the manuscript text to clarify the methodology. NGG is guarantor and is responsible for the overall content.
Funding This study was supported by the research project RADAR (Risk Assessment and Disease burden of Antimicrobial Resistance) funded through the One Health European Joint Programme by the EU's Horizon-2020 Research and Innovation Programme (grant 773830).
Map disclaimer The inclusion of any map (including the depiction of any boundaries therein), or of any geographic or locational reference, does not imply the expression of any opinion whatsoever on the part of BMJ concerning the legal status of any country, territory, jurisdiction or area or of its authorities. Any such expression remains solely that of the relevant source and is not endorsed by BMJ. Maps are provided without any warranty of any kind, either express or implied.
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.