Current crisis or artifact of surveillance: insights into rebound chlamydia rates from dynamic modelling

Abstract

Background: After initially falling in the face of intensified control efforts, reported rates of sexually transmitted chlamydia in many developed countries are rising. Recent hypotheses for this phenomenon have broadly focused on improved case finding or an increase in the prevalence. Because of many complex interactions behind the spread of infectious diseases, dynamic models of infection transmission are an effective means to guide learning, and assess quantitative conjectures of epidemiological processes. The objective of this paper is to bring a unique and robust perspective to observed chlamydial patterns through analyzing surveillance data with mathematical models of infection transmission.

Methods: This study integrated 25-year testing volume data from the Canadian province of Saskatchewan with one susceptible-infected-treated-susceptible and three susceptible-infected-treated-removed compartmental models. Calibration of model parameters to fit observed 25-year case notification data, after being combined with testing records, placed constraints on model behaviour and allowed for an approximation of chlamydia prevalence to be estimated. Model predictions were compared to observed case notification trends, and extensive sensitivity analyses were performed to confirm the robustness of model results.

Results: Model predictions accurately mirrored historic chlamydial trends including an observed rebound in the mid 1990s. For all models examined, the results repeatedly highlighted that increased testing volumes, rather than changes in the sensitivity and specificity of testing technologies, sexual behaviour, or truncated immunological responses brought about by treatment can, explain the increase in observed chlamydia case notifications.

Conclusions: Our results highlight the significant impact testing volume can have on observed incidence rates, and that simple explanations for these observed increases appear to have been dismissed in favor of changes to the underlying prevalence. These simple methods not only demonstrate geographic portability, but the results reassure the public health effort towards monitoring and controlling chlamydia.

Figure 1

Historic trends of Chlamydia trachomatis infections in Saskatchewan and Canada between 1983 and 2007. Asterisk indicates when chlamydia infections became reportable in Saskatchewan (1984). Saskatchewan Incidence rates (per 100,000 population per year) are crude rates and are calculated as the ratio of number of reported cases and the population of Saskatchewan.

Figure 2

Schematic stock and flow diagram of the susceptible-infected-treated-removed model. We adopted a deterministic, compartmental, susceptible-infected-treated-removed-susceptible (or SITRS) framework. The susceptible class, S, contained those who were sexually active and could become infected; the infected class, I, contained those who were infectious; the treated class, T, containing both true and false positives; these people were assumed to have been tested, diagnosed as cases, and treated appropriately; the removed class, R, contained those who had naturally recovered from infection, and those who were given a positive diagnosis and temporarily reduced their sexual risk-taking behaviour after being treated. People in the removed class were assumed to eventually exit the sexually active population, or return to the susceptible class as a result of waned immunity or relapse into previous risky sexual behaviour. Additional model structures that were also investigated are displayed in Additional file 1.

Figure 3

Comparison of model calibrations to observed trends. Calibrated numbers of (A) cases from the models compared to reported numbers of cases in Saskatchewan. Parts (B) and (C) cross-check the model to the observed proportion positive among those tested (B), and the reported incidence (per 100,000 population) (C). Model-generated curves in parts (A), (B), and (C) were arbitrarily chosen from 50 million optimization simulations. Part (D) is a visual comparison of testing volume between 1983 and 2007 to the "actual" prevalence in Saskatchewan generated by the model.

Figure 4

Uncertainty and sensitivity analysis of model results. Parts (A) and (B) are the results of a sensitivity analysis on the "actual" prevalence generated by the model and the fraction of infectives that have recovered via treatment, respectively. In both parts (A) and (B), the black line represents the mean value, and the coloured bands represent the 50 (red), 75 (yellow), 95 (green), and 99 per cent (blue) confidence intervals.