Background

Three-Sample Capture-Recapture (3S-CRC) is considered the gold-standard method for estimating the size of hidden populations. The approach relies on 3 or more lists of the target population (e.g., receiving services, outreach intercepts) and models the total number based on the probabilities of appearing in 1, 2, or 3 “captures”. However, 3S-CRC presents theoretical and practical challenges, including: 1) misaligned definitions of the population; 2) non-independence between captures, 3) missing important sub-populations; 4) duplicates within capture rounds; and 5) matching individuals across rounds. In updating estimates for the number of people who inject drugs (PWID) in San Francisco, we developed innovations to address these issues.

Approach

Captures using service databases often had misaligned definitions of PWID (i.e., injection not recorded), place (i.e., geographic area), and time (i.e., past-year injection), and had incomplete identifiers. We therefore recommend 3 field captures. To improve independence between captures, sampling methods were diversified using street intercepts (rounds 1 and 3) and respondent-driven sampling (RDS, round 2). RDS, a peer-referral recruitment method, reaches less visible subgroups and allows the collection of biological specimens and in-depth behavioral data. Formative mapping ensures geographic alignment and inclusion of subgroups. A $10 incentive for all screened (regardless of eligibility) discourages falsifying criteria and reduces within-round duplication. Distinct-colored popcorn facilitates recall of participation in the current and prior rounds, both needed to measure capture-recapture probabilities. Matching between rounds uses a unique alphanumeric code from participant-known elements (e.g., initials, dates) while maintaining anonymity. Finally, we advocate changing the “capture-recapture” terminology to the human-centered “Multiple Engagement Size Estimation” when referring to this gold standard method.