Comparative performance of multiple-list estimators of key population size

Area of Science:

• Epidemiology
• Biostatistics
• Population Health

Background:

• Accurate estimation of key populations (KPs) sizes is vital for effective HIV epidemic control.
• Discrepancies among different estimation methods hinder targeted interventions.
• An objective basis for selecting appropriate estimation techniques is lacking.

Purpose of the Study:

• To compare the performance of three statistical methods for estimating population sizes from multiple data sources.
• To evaluate loglinear model selection (LLM-AIC), Bayesian model averaging of loglinear models (LLM-BMA), and Bayesian nonparametric latent-class modeling (BLCM).
• To provide guidance on the optimal number of lists for accurate population size estimation.

Main Methods:

• A simulation study was conducted using 36,000 samples generated from populations of varying sizes (1,000 to 20,000) and encounter probabilities.
• Population sizes were estimated using LLM-AIC, LLM-BMA, and BLCM with 2 to 5 cumulative lists.
• Performance was assessed using root mean-squared error, bias, and uncertainty interval coverage.

Main Results:

• LLM-BMA and BLCM demonstrated robustness and comparable performance, outperforming LLM-AIC in accuracy and bias reduction.
• All methods produced uncertainty intervals with inadequate coverage, but LLM-BMA offered the best balance of accuracy and coverage.
• Estimating KP sizes using at least three lists is recommended over two-list estimations, which are more vulnerable.

Conclusions:

• Bayesian approaches (LLM-BMA and BLCM) are superior for estimating key population sizes from multiple lists compared to traditional loglinear model selection.
• The use of at least three data sources enhances the reliability of population size estimates.
• Further research may refine uncertainty interval estimation in these models.