Estimating prevalence of injecting drug use: a comparison of multiplier and capture-recapture methods in cities in England and Russia.
We consider the question of what method should be recommended to estimate the prevalence of injecting drug use (IDU); and compare multiplier and capture-recapture (CRC) methods of estimating prevalence of injecting drug use (IDU). The prevalence of injecting drug use in four cities (Brighton, Liverpool, London and Togliatti) was estimated using similar methods: covariate capture-recapture (CRC) and multipliers. The multipliers, generated either from a community recruited survey or historical/literature-based, were applied to a range of 'benchmarks': specialist drug treatment, arrests, accident and emergency department (A&E), syringe exchange, HIV tests and opiate overdose deaths. The CRC estimates were assumed to be 'preferred/gold standard' [2,304 (95% confidence interval 1,514 - 3,737) in Brighton, 2,910 (2,546 - 4,977) in Liverpool, 16,782 (13,793 - 21,620) in 12 London boroughs and 15,039 (12,696 - 18,515) male IDU in Togliatti]. The ranges given by the multiplier estimates obtained through the community survey varied from 200 to 770 in Brighton, 530 to 1,300 in Liverpool, 2,900 to 10,600 in London and 12,400 to 91,000 in Togliatti. Several multipliers gave implausible results, lower than the observed data collected for another benchmark, and in the three English cities all these multiplier estimates were below the lower 95% confidence interval of the CRC estimate. In Togliatti, only one multiplier estimate was close to the preferred CRC estimates, with the rest implausibly high. The multiplier estimates based on historical/literature multipliers also ranged widely from 390 to 4,800 for Brighton, from 1,645 to 2,800 in Liverpool, from 4,650 to 12,600 in the 12 London boroughs and 12,800 to 32,000 in Togliatti. In the three UK cities the mortality multiplier estimates were closest to the capture-recapture estimates. The study was a practical demonstration comparing a range of multiplier estimates with a single CRC study. In almost all the individual comparisons the multiplier estimates performed poorly. CRC methods should be preferred as the means of estimating numbers of drug users with multiplier methods being used with caution and only where CRC is not possible.