Background: Mathematical models are important tools for evaluating the effectiveness and cost-effectiveness of STI prevention interventions. These models are usually developed for a single infection, rely on sexual behavior data that is of variable quality, and require fitting (i.e. calibration) to observed STI trends. Incorporating other outcomes of sexual behavior, such as unintended pregnancy, may improve this process.
Methods: We constructed a mathematical compartmental model of chlamydia transmission in adolescents aged 15-19 in Minnesota. The model was primarily parameterized with data from the Minnesota Student Survey (2004-2013) and included temporal trends of chlamydia screening and long-acting reversible birth control use. Using Bayesian calibration methods, we estimated distributions for uncertain parameters by fitting model outputs to state-wide trends from 2005-2013. In the first scenario, we calibrated to chlamydia incidence alone. In the second scenario, we calibrated to the incidence of both chlamydia and pregnancy. We compared parameter estimates and the projected effectiveness of a hypothetical condom promotion intervention across the two models
Results: The two models produced two significantly different distributions for the frequency of sex acts, condom use, and per-act transmission probability (Kolmogorov-Smirnov test, p<0.05). Notably, the average number of sex acts per month increased 46.5% when calibrating to both pregnancy and chlamydia trends versus calibrating to chlamydia incidence alone. A 25% increase in condom use was projected to avert 8.0% of annual chlamydia infections in the model calibrated to chlamydia alone compared to 14.7% when also calibrating to pregnancy.
Conclusions: Including pregnancy as an additional calibration target for a model of chlamydia resulted in different projections of effectiveness of an STI prevention intervention. While pregnancy and STI incidence in adolescents are often considered separately, both are outcomes of unprotected sexual activity and, taken together, may be more informative for calibrating STI models and predicting the outcome of interventions.