• Journal Article

Development and evaluation of a Markov model to predict changes in schistosomiasis prevalence in response to praziquantel treatment: a case study of Schistosoma mansoni in Uganda and Mali

Citation

Deol, A., Webster, J., Walker, M., Basanez, M-G., Hollingsworth, D., Fleming, F., ... French, M. D. (2016). Development and evaluation of a Markov model to predict changes in schistosomiasis prevalence in response to praziquantel treatment: a case study of Schistosoma mansoni in Uganda and Mali. Parasites & Vectors, 9(543).

Abstract

Background
Understanding whether schistosomiasis control programmes are on course to control morbidity and potentially switch towards elimination interventions would benefit from user-friendly quantitative tools that facilitate analysis of progress and highlight areas not responding to treatment. This study aimed to develop and evaluate such a tool using large datasets collected during Schistosomiasis Control Initiative-supported control programmes.

Methods
A discrete-time Markov model was developed using transition probability matrices parameterized with control programme longitudinal data on Schistosoma mansoni obtained from Uganda and Mali. Four matrix variants (A-D) were used to compare different data types for parameterization: A-C from Uganda and D from Mali. Matrix A used data at baseline and year 1 of the control programme; B used year 1 and year 2; C used baseline and year 1 from selected districts, and D used baseline and year 1 Mali data. Model predictions were tested against 3 subsets of the Uganda dataset: dataset 1, the full 4-year longitudinal cohort; dataset 2, from districts not used to parameterize matrix C; dataset 3, cross-sectional data, and dataset 4, from Mali as an independent dataset.

Results
The model parameterized using matrices A, B and D predicted similar infection dynamics (overall and when stratified by infection intensity). Matrices A-D successfully predicted prevalence in each follow-up year for low and high intensity categories in dataset 1 followed by dataset 2. Matrices A, B and D yielded similar and close matches to dataset 1 with marginal discrepancies when comparing model outputs against datasets 2 and 3. Matrix C produced more variable results, correctly estimating fewer data points.

Conclusion
Model outputs closely matched observed values and were a useful predictor of the infection dynamics of S. mansoni when using longitudinal and cross-sectional data from Uganda. This also held when the model was tested with data from Mali. This was most apparent when modelling overall infection and in low and high infection intensity areas. Our results indicate the applicability of this Markov model approach as countries aim at reaching their control targets and potentially move towards the elimination of schistosomiasis.