Heroin use cannot be measured adequately with a general population survey

Globally, heroin and other opioids account for more than half of deaths and years-of-life-lost due to drug use and comprise one of the four major markets for illegal drugs. Having sound estimates of the number of problematic heroin users is fundamental to formulating sound health and criminal justice policies. Researchers and policymakers rely heavily upon general population surveys (GPS), such as the US National Survey on Drug Use and Health (NSDUH), to estimate heroin use, without confronting their limitations. GPS-based estimates are also ubiquitous for cocaine and methamphetamine, so insights pertaining to GPS for estimating heroin use are also relevant for those drug markets.

Four sources of potential errors in NSDUH are assessed: selective non-response, small sample size, sampling frame omissions and under-reporting. An alternative estimate drawing on a variety of sources including a survey of adult male arrestees is presented and explained. Other approaches to prevalence estimation are discussed.

Under-reporting and selective non-response in NSDUH are likely to lead to substantial underestimation. Small sample size leads to imprecise estimates and erratic year-to-year fluctuations. The alternative estimate provides credible evidence that NSDUH underestimates the number of frequent heroin users by at least three-quarters and perhaps much more.

GPS, even those as strong as NSDUH, are doomed by their nature to estimate poorly a rare and stigmatized behavior concentrated in a hard-to-track population. Although many European nations avoid reliance upon these surveys, many others follow the US model. Better estimation requires models that draw upon a variety of data sources, including GPS, to provide credible estimates. Recent methodological developments in selected countries can provide guidance. Journals should require researchers to critically assess the soundness of GPS estimates for any stigmatized drug-related behaviors with low prevalence rates.