Seasonal malaria chemoprevention (SMC) was first recommended by the World Health Organization (WHO) in 2012 to prevent uncomplicated malaria in children and began implementation in Burkina Faso in 2014 under programmatic campaigns. Systematic assessment of the impact of national SMC campaigns requires data with weekly or monthly temporal resolution over several years and broad spatial coverage. Despite the intensive deployment of malaria control interventions, such as SMC, health facilities in Burkina Faso reported increased malaria incidence between 2015 and 2018 among children under the age of 5. This analysis aims to adjust malaria routine case data in Burkina Faso for factors affecting the reported disease burden and describe the changes in the trend of malaria transmission between 2015 and 2018. The goal is to refine the subnational analysis of trends in the malaria burden in Burkina Faso and improve the assessment of SMC impact through quasi-experimental modeling. Monthly data on confirmed malaria cases from 2015 to 2018 was obtained from Burkina Faso’s Health Management Information System (HMIS). The data was aggregated to the health district level and was adjusted for changes in the health facility reporting rate and changes in treatment-seeking behavior. Seasonal trend decomposition was applied to crude malaria incidence, adjusted malaria incidence, all-cause outpatient visits, non-malarial outpatient visits, and the malaria proportion of outpatient visits to determine the trend in each time series for children under 5 years old and individuals over 5 years old. The Sen's slope coefficient was used to quantify the trend components and map them for each health district in the country. Using adjusted malaria outcomes, we quantify the effectiveness of SMC using district-level data against symptomatic malaria with generalized linear mixed-effects models in a difference-in-differences framework. We quantify the effectiveness of SMC in reducing malaria incidence, the malaria proportion of outpatient visits, and the incidence of malaria hospitalizations for each group of health districts treated in the same year. Given the impact of changes in treatment-seeking behavior spurred on by a national change in healthcare policy that removed healthcare fees for children under 5 years old and pregnant women, we modeled the treatment-seeking rate at a 1 x 1 km spatial resolution as a latent process. This latent process was modeled using data from the Malaria Indicator Survey from 2014 and 2017/18 with an integrated nested Laplacian approximation (INLA) using spatial random effects with a Matérn correlation structure. For analysis focused on the malaria burden at the health district level, fine-scale estimates were aggregated for each district using a population-weighted average. For children under 5 years old, trends increased for crude malaria incidence (43/70 health districts). After adjusting for reporting rates and the changes in modeled treatment-seeking rate, adjusted malaria incidence trends increased in 26 districts, decreased in 4 districts, and remained unchanged in 40 districts. The proportion of outpatient visits that were confirmed malaria cases had a decreasing trend in 64/70 districts. For individuals over 5 years old, trends in malaria incidence and outpatient visits increased more modestly or remained unchanged in most districts. SMC reduced the adjusted incidence in children under 5 years old by 27.1% (95% CI: [19.8%, 33.7%]), 43.4% (95% CI: [36.2%, 49.9%]), and 29.7% (95% CI: [20.2%, 38.1%]) in the 2016-2018 SMC groups, respectively. SMC showed a weak protective effect on reducing the malaria proportion of outpatient visits in children under 5 years old: 7.5% (95% CI: [2.3%, 12.5%]) reduction for the 2016 SMC rollout group, 13.7% (95% CI: [7.4%, 19.5%]) reduction for the 2017 group, and no significant effect for the 2018 group. SMC was also associated with a reduction in malaria hospitalizations among children under 5 years old by 39.8% (95% CI: [31.4%, 47.2%]) in the 2016 SMC rollout group and 35.2% (95% CI: [20.5%, 47.2%]) in the 2017 group. SMC was not associated with a significant reduction in any of the three indicators for individuals over 5 years old, except in 2017 where a reduction of 12.2% (95% CI: [1.6%, 21.6%]) was observed for adjusted malaria incidence and a reduction of 15.2% (95% CI: [1.2%, 27.2%]) was observed for malaria hospitalizations. The gaps in reporting quality among healthcare facilities limit our ability to leverage the full HMIS data for the evaluation of the malaria burden and the estimation of the effectiveness of SMC. Using the HMIS data at the healthcare facility level, we address the gaps in reporting by expanding the analysis of the malaria burden among children under 5 years old and the effectiveness of SMC using a spatio-temporal model in an INLA framework. We model the malaria proportion of outpatient visits among children in a generalized linear mixed-effects model and use this to capture the effectiveness of SMC in a difference-in-differences framework. Time series are modeled from health facilities with known GPS locations and the model includes a temporal random effect with a seasonal correlation structure and a spatio-temporal random effect with a Matérn correlation structure. The modeled latent process for the treatment-seeking rate is included in the model to help adjust for changes in treatment-seeking behavior to each healthcare facility. We introduce a novel approach to difference-in-differences modeling frameworks, that helps capture the impact of SMC and changes in treatment seeking on the malaria burden. We map the malaria proportion of outpatient visits at a 1 x 1 km resolution for the 4 months of the high-transmission season for malaria each year in the study period. We derive the effectiveness of SMC using healthcare facility-level data, and compare the results to the model using district-aggregated data. Using healthcare facility-level data in a spatio-temporal model, SMC reduced the malaria proportion of outpatient visits among children under 5 years old by 10.9% (95% CI: [8.3%, 13.5%]) in the 2016 SMC group, 16.2% (95% CI: [12.9%, 19.3%]) in the 2017 SMC group, and 11.0% (95% CI: [5.8%, 15.9%]) in the 2018 SMC group. The results of SMC effectiveness from the 2016 and 2017 groups are consistent with those found in the health district level model. This model was able to capture a significant reduction associated with the introduction of SMC in the 2018 SMC group, whereas the district-level model found no significant reduction. SMC showed a mild to moderate protective effect against symptomatic malaria among children under 5 years old. Changes in treatment-seeking behavior and health facility reporting contributed to the observed increase in malaria incidence in children under 5 years old between 2015 and 2018, but adjusting for these factors did not completely remove the increasing trend across districts. The choice of which malaria outcome to analyze shows strong differences in the protective effectiveness of SMC, as the 2018 rollout group saw a significant reduction in malaria among children under 5 associated with SMC but no such significant reduction in the malaria proportion of outpatient visits. The choice of using the malaria proportion as the outcome variable helps us control for monthly changes in treatment seeking, but does not directly reflect changes in malaria incidence. Using adjusted malaria incidence as the response variable, we cannot reliably control for intra-regional or temporal changes in treatment-seeking behavior. Using spatio-temporal modeling frameworks helps address some of the limitations in the routine case data caused by gaps in reporting quality. These frameworks helped in capturing the effectiveness of SMC on the 2018 SMC group, which was not observed to be significant using a district-aggregated model. Adjusting for sources of bias is a necessary step before analyzing routine data for SMC effectiveness. Our novel approach to differences-in-differences frameworks which include latent processes may be extended to other disciplines which utilize differences-in-differences modeling frameworks while considering latent processes.

Creator

• Rodriguez, Sebastian

DOI

• https://doi.org/10.21985/n2-cg1c-nv66

Subject

• Epidemiology
• Statistics

Language

• en

Alternate Identifier

• http://dissertations.umi.com/northwestern:16220
• etdadmin_upload_927176

Keyword

• Spatio-temporal modeling
• Integrated Nested Laplacian Approximation
• Malaria
• Time-series analysis
• Difference-in-differences
• Seasonal Malaria Chemoprevention

Date created

• 2022-01-01

Resource type

• Dissertation

Rights statement

• In Copyright