Public health in genetic spaces: a statistical framework to optimize cluster-based outbreak detection

Area of Science:

• Epidemiology and Public Health
• Viral Evolution and Phylogenetics
• Biostatistics and Mathematical Modeling

Background:

• Genetic clustering is used to understand viral transmission rates and detect outbreaks.
• Current methods lack objective criteria for setting clustering parameters, hindering real-time application.
• Understanding transmission variation is crucial for effective public health interventions.

Purpose of the Study:

• To develop a statistical framework for optimizing genetic clustering methods.
• To improve the accuracy of forecasting new cases and detecting outbreaks.
• To establish objective guidelines for setting genetic clustering criteria in public health.

Main Methods:

• Analyzed pairwise Tamura-Nei (TN93) genetic distances for HIV-1 subtype B pol sequences from diverse populations.
• Developed and compared two models: a null model (proportional growth) and a weighted model (including covariates like diagnosis recency).
• Optimized TN93 thresholds by maximizing the difference in information loss between models, indicating effective covariate use.

Main Results:

• Optimal TN93 thresholds varied significantly across datasets (e.g., 0.0104 in Alberta vs. 0.016 in Seattle/Tennessee).
• The weighted model showed greater predictive accuracy within a narrow range of thresholds (±0.005 units) for each population.
• Recency of HIV diagnosis was a stronger predictor of new cases than sample collection dates.

Conclusions:

• Genetic clustering methods must be calibrated to specific public health settings and epidemic contexts.
• Relying on historical or conventional thresholds can misdirect prevention efforts.
• The developed framework allows for method calibration and evaluation of predictive models for cluster growth.