Problem drug use prevalence estimation revisited: heterogeneity in capture–recapture and the role of external evidence

Abstract

Background and AimsCapture–recapture (CRC) analysis is recommended for estimating the prevalence of problem drug use or people who inject drugs (PWID). We aim to demonstrate how naive application of CRC can lead to highly misleading results, and to suggest how the problems might be overcome.MethodsWe present a case study of estimating the prevalence of PWID in Bristol, UK, applying CRC to lists in contact with three services. We assess: (i) sensitivity of results to different versions of the dominant (treatment) list: specifically, to inclusion of non‐incident cases and of those who were referred directly from one of the other services; (ii) the impact of accounting for a novel covariate, housing instability; and (iii) consistency of CRC estimates with drug‐related mortality data. We then incorporate formally the drug‐related mortality data and lower bounds for prevalence alongside the CRC into a single coherent model.ResultsFive of 11 models fitted the full data equally well but generated widely varying prevalence estimates, from 2740 [95% confidence interval (CI) = 2670, 2840] to 6890 (95% CI = 3740, 17680). Results were highly sensitive to inclusion of non‐incident cases, demonstrating the presence of considerable heterogeneity, and were sensitive to a lesser extent to inclusion of direct referrals. A reduced data set including only incident cases and excluding referrals could be fitted by simpler models, and led to much greater consistency in estimates. Accounting for housing stability improved model fit considerably more than did the standard covariates of age and gender. External data provided validation of results and aided model selection, generating a final estimate of the number of PWID in Bristol in 2011 of 2770 [95% credible interval (Cr‐I) = 2570, 3110] or 0.9% (95% Cr‐I = 0.9, 1.0%) of the population aged 15–64 years.ConclusionsSteps can be taken to reduce bias in capture–recapture analysis, including: careful consideration of data sources, reduction of lists to less heterogeneous subsamples, use of covariates and formal incorporation of external data.

Impact of current and scaled‐up levels of hepatitis C prevention and treatment interventions for people who inject drugs in three UK settings—what is required to achieve the WHO's HCV elimination targets?

“…The model was calibrated to temporal data on HCV prevalence, recent estimates of coverage of OST and HCNSP, proportion of PWID with high‐risk attributes, population‐size estimates of PWID and their distribution by injecting duration (Table and Supporting information, Table S2) . Data on HCV incidence for Dundee and Bristol and HCV prevalence after 2006 for Bristol and Walsall were used for model validation, with both extracted from routine surveys of PWID [needle exchange surveillance initiative in Scotland (NESI) , unlinked anonymous monitoring survey (UAM) in England and Wales and two additional community surveys using respondent‐driven sampling from Bristol ; see Supporting information for details of the surveys].…”

Assessing the impact and cost-effectiveness of needle and syringe provision and opioid substitution therapy on hepatitis C transmission among people who inject drugs in the UK: an analysis of pooled data sets and economic modelling

“…In Bristol and Walsall, size estimation data suggest that the PWID population has decreased by between 10% and 30% between 2009 and 2011. [104][105][106]108 Concurrently, survey data 44,70,73,88,89 suggest that the proportion of PWID injecting for > 10 years has increased, whereas the proportion injecting for between 3 and 10 years has decreased ( Figure 18). There has been little change in the proportion injecting for < 3 years.…”

Estimating high‐risk cannabis and opiate use in Ankara, Istanbul and Izmir

“…Data were analysed using the CRC [15]. For all analyses, data were stratified by age group (15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32)(33)(34) and 35-64 years) and gender, and log-linear Poisson regression models were fitted separately to the stratified pattern of overlaps between the sources of data. It is not recommended to use CRC with only two sources, because it is then necessary to assume independence of the two sources [16]; this is an untestable and usually improbable assumption.…”

“…This risk was reduced by stratifying samples according to age and gender. Nevertheless, recent research has examined the impact of possible breaches of the traditional capture-recapture assumptions, such as heterogeneity of capture and independence of data sources [29,30]. As all of these mechanisms may simultaneously impact on the overlap patterns of the data used for CRC, it can be said that influences lowering the overlap will generally result in overestimating the true prevalence, whereas increasing the overlap will result in underestimating the true prevalence.…”