Spatio-temporal and statistical prediction models of human infectious diseases in Rwanda

Abstract

Due to limited resources and inadequate reporting systems, routine data on infectious diseases in countries with resource constraints often lack complete and accurate information. As a result, analyzing disease patterns and trends can be difficult, hindering effective public health interventions. To address this, more advanced statistical methods can be used to analyze and predict disease patterns based on spatial and temporal factors using health facility routine data. This provides a more accurate understanding of disease patterns and trends, which can guide the development and implementation of effective public health interventions. By integrating statistical prediction models into existing electronic health systems in resource-constrained countries, the overall capacity of the health systems can be enhanced, leading to improved health outcomes and reduced disease burden. The objective of this PhD thesis was to bridge the gap in statistical analysis of routine data using spatio-temporal and statistical prediction models that could be integrated into existing electronic health systems in resource-constrained countries. The study used Malaria and COVID-19 data from Rwanda's integrated health management information system, census data from the National Institute of Statistics, and Demographic Health Surveys from ICF Macro. These multiple sources of data were used to build statistical models that dealt with the spatial and temporal structure of the data. The Bayesian approach was used due to its flexibility in dealing with uncertainty and missing data, and integrated nested Laplace approximations (INLA) were used to optimize computation capability on larger datasets structured on space and time. The findings of the thesis demonstrated that routinely collected data are equally important as survey data in both evaluating and making decisions, provided that the data quality is high. Routine data is collected more frequently and provides more timely assessments of health burdens. The thesis also showed a novel approach of using the excess probability method to evaluate attainment of targets or identifying thresholds, which was recommended as a relevant statistical method that enables the monitoring and control of the effectiveness of targeted interventions. The thesis recommends estimating malaria burden using multiple data sources in a two-step modeling approach. Survey data may not provide an accurate picture of the malaria burden in countries with low malaria prevalence, which can affect malaria control, prevention measures, and resource allocation. The thesis also provided insights on using spatio-temporal models in routine monitoring of COVID-19 outbreaks in Rwanda. All scientific outcomes were published in peer-reviewed journals.