JMIR Publications

ABSTRACT

Background:

Accurate population size estimates (PSE) of key populations, those who are disproportionately affected by HIV, are critical to forecast need and inform HIV prevention and treatment programs, though can be difficult to ascertain due to limited visibility of these groups. In Zambia, where punitive laws against drugs and injection equipment impede harm reduction services for people who inject drugs (PWID), reliable estimates on the number of PWID are limited, inhibiting a data-driven public health response.

Objective:

We sought to estimate the population size of PWID in three large cities in Zambia, document how PSE vary across empiric estimation methods and explore strengths and limitations of each method.

Methods:

We applied 2- and 3-source capture-recapture, successive sampling, and wisdom of the crowd methods to estimate the population size of PWID in Lusaka, Livingstone, and Ndola, Zambia. 3-source capture-capture (3S-CRC) methods included 2-source capture-capture (2S-CRC) in combination with a respondent-driven sampling (RDS) survey. Data were collected from November 2021 to February 2022 and analyzed using a Bayesian nonparametric latent class model. Successive sampling PSE were produced using the RDS recruitment and network sizes. Final estimates for each city were generated using a Bayesian consensus estimator. Kruskal tests and general linear models were used to assess which characteristics measured in the RDS survey were associated with being captured in one or both of the first two captures.

Results:

Consensus PSE ranged between 0.5%-1.8% of the adult male population and below 1% of the total adult population in each city. Consensus estimates were highest in Lusaka (3,700 95% credible interval [CI]:1,500-7,500) followed by Ndola (2,200 95%CI:1,600-2,900), and Livingstone (1,200 95%CI:900-1,900). There was wide variability in the PWID estimates by empiric method, with wisdom of the crowd generally providing the lowest estimates across cities. Across empiric methods, PSE for PWID ranged from 70 (95%CI:50-80) to 2,620 (95%CI:1,510-4,680) in Livingstone, 450 (95%CI:320-580) to 4,350 (95%CI:1,410-18,890) in Lusaka, and 200 (95%CI:130-270) to 4,030 (95%CI:960-5,480) in Ndola. In all cities, fewer recaptures occurred with 3S-CRC than with location sampling via 2S-CRC. RDS participants who had been captured in Capture 1 or 2 had differences in sociodemographic and behavioral risk factors compared to those only captured through the RDS survey though there was heterogeneity by city.

Conclusions:

As demonstrated in this study, RDS methods can facilitate capturing key populations with less social visibility, ensuring a more diverse reflection of the true population, to provide more accurate PSE. This study employed strong empiric methods to produce PSE for PWID in Zambia and is the first study to produce PSE for PWID for major geographies in the country. Results can inform program planning, including target setting and coverage estimation, and future surveillance activities for PWID in the country.