medRxiv preprint doi: https://doi.org/10.1101/2020.04.02.20051151; this version posted April 6, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license . 1 Socio-demographic, not environmental, risk factors explain fine-scale spatial 2 patterns of diarrheal disease in Ifanadiana, rural Madagascar 3 Michelle V Evans1,2*, Matthew H Bonds3,4, Laura F Cordier4, John M Drake1,2, Felana 4 Ihantamalala3,4, Justin Haruna4, Ann C Miller3, Courtney C Murdock1,2,5, Marius 5 Randriamanambtsoa9, Estelle M Raza-Fanomezanjanahary10, Bénédicte R. Razafinjato 6 4, Andres Garchitorena4,11 7 8 1 Odum School of Ecology, University of Georgia, Athens GA USA 9 2 Center for Ecology of Infectious Diseases, University of Georgia, Athens GA USA 10 3 Department of Global Health and Social Medicine, Harvard Medical School, Boston 11 MA USA 12 4 PIVOT, Ranomafana, Madagascar/ Boston MA USA 13 5 Department of Infectious Diseases, College of Veterinary Medicine, University of 14 Georgia, Athens GA USA 15 6 Center for Tropical and Emerging Global Diseases, University of Georgia, Athens GA 16 USA 17 7 Center for Vaccines and Immunology, University of Georgia, Athens GA USA 18 8 River Basin Center, University of Georgia, Athens GA USA 19 9 National Institute of Statistics, Madagascar 20 10 Ministry of Health, Madagascar 21 11 MIVEGEC, Univ. Montpellier, CNRS, IRD, Montpellier, France 22 23 *Corresponding author: [email protected] 24 25 Keywords: diarrheal disease, precision health mapping 26 NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. 1 medRxiv preprint doi: https://doi.org/10.1101/2020.04.02.20051151; this version posted April 6, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license . 27 Abstract 28 Diarrheal disease (DD) is responsible for over 700,000 child deaths annually, the 29 majority in the tropics. Due to its strong environmental signature, DD is amenable to 30 precision health mapping, a technique that leverages spatial relationships between 31 socio-ecological variables and disease to predict hotspots of disease risk. However, 32 precision health mapping tends to rely heavily on data collected at coarse spatial scales 33 over large spatial extents. There is little evidence that such methods produce 34 operationally-relevant predictions at sufficiently fine enough spatio-temporal scales (e.g. 35 village level) to improve local health outcomes. Here, we use two fine-scale health 36 datasets (<5 km) collected from a health system strengthening initiative in Ifanadiana, 37 Madagascar and identify socio-ecological covariates associated with childhood DD. We 38 constructed generalized linear mixed models including socio-demographic, climatic, and 39 landcover variables and estimated variable importance via multi-model inference. We 40 find that socio-demographic variables, and not environmental variables, are strong 41 predictors of the spatial distribution of disease risk at both an individual and commune- 42 level spatial scale. Specifically, a child’s age, sex, and household wealth were the 43 primary determinants of disease. Climatic variables predicted strong seasonality in DD, 44 with the highest incidence in the colder, drier months of the austral winter, but did not 45 predict spatial patterns in disease. Importantly, our models account for less than half of 46 the total variation in disease incidence, suggesting that the socio-ecological covariates 47 identified as important via global precision health mapping efforts have reduced 48 explanatory power at the local scale. More research is needed to better define the set of 2 medRxiv preprint doi: https://doi.org/10.1101/2020.04.02.20051151; this version posted April 6, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license . 49 conditions under which the application of precision health mapping can be operationally 50 useful to local public health professionals. 51 52 3 medRxiv preprint doi: https://doi.org/10.1101/2020.04.02.20051151; this version posted April 6, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license . 53 Introduction 54 Over 700,000 child deaths are attributed to diarrheal disease (DD) annually1. The 55 burden of DD is unequally distributed across the globe: 73% of deaths occur in just 56 fifteen low income countries, driven by inequalities in water and sanitation infrastructure 57 and environmental conditions2. Understanding risk factors of spatio-temporal patterns of 58 DD is instrumental in designing public health interventions that target populations most 59 at risk. Precision health mapping is an approach that incorporates increasingly available 60 fine-scale social and environmental information into spatial models to explain and 61 predict spatial disease patterns at resolutions finer than those previously possible3. This 62 approach has recently emerged as a method to identify areas and populations at risk of 63 disease, and has been successfully used to map the global distribution of diseases with 64 strong environmental signatures, such as mosquito-borne (e.g. malaria4, dengue5, 65 lymphatic filariasis6) and water-borne diseases (e.g. schistosomiasis7). There are few 66 examples, however, of efforts that are precise enough and appropriately integrated with 67 implementation to be able to improve local public health strategies. 68 Diarrheal disease is ideal for precision health mapping because the determinants 69 of environmental suitability for a diarrheal pathogen, such as climate8,9, land cover10, 70 and soil characteristics11, are well known. Hydrological networks, as well as 71 infrastructure and WASH (water, sanitation, and hygiene) practices influence 72 transmission dynamics and individual risk of DD1012,13. Upstream land cover has been 73 shown to predict DD prevalence in rural areas of the tropics10, with cumulative effects 74 for populations that are downstream of sources of water contamination (e.g. livestock or 75 agricultural run-off14). Global analyses of coastal Vibrio cholera, the causative agent of 4 medRxiv preprint doi: https://doi.org/10.1101/2020.04.02.20051151; this version posted April 6, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license . 76 cholera, for example, have found that chlorophyll levels, a common consequence of 77 runoff-induced algal blooms, are the strongest predictor of the pathogen’s presence15. 78 Such studies tend to rely on data extracted from large national surveys, such as 79 Demographic and Health Surveys, that are powered to estimate indicators at broad 80 spatial and temporal scales, such as a country or region every five years. When 81 projected to more granular geospatial data, they produce fitted values on the 82 assumption that relationships found at broad spatial scales exist at fine spatial scales. 83 However, because these localized predictions are not typically fitted with localized 84 health data at fine spatio-temporal scales, they may fail to explain patterns of disease at 85 local scales, limiting their ability to inform priorities set by local health actors. 86 Socio-ecological determinants of DD may be region- and pathogen- specific, so 87 that relationships identified at the global level may not hold locally. For example, despite 88 a proven biological pathway of fecal-oral transmission, the effects of WASH 89 interventions on DD prevalence are ambiguous due to differences in socio-ecological 90 context and specific etiological agents16. Pathogens’ temperature responses differ 91 between viral (higher survival in colder environments) and bacterial agents (lower 92 growth in colder environments)17, with net effects on DD dynamics dependent on the 93 prevailing set of pathogens. Additionally, precipitation leads to the highest disease 94 prevalence at extremes (i.e. droughts and flooding)18, a non-linearity that may be 95 difficult to identify at the local level due to lower variability. At the national and global 96 scale, increased forest cover is consistently associated with lower incidence of 97 DD10,19,20, but these associations have not been tested at the fine spatial scale relevant 5 medRxiv preprint doi: https://doi.org/10.1101/2020.04.02.20051151; this version posted April 6, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. It is made available under a CC-BY-NC-ND 4.0 International license . 98 for local action. Addressing this difficulty in downscaling precision health mapping tools 99 is central to evaluating their potential utility. 100 Here, we investigated the potential for precision health mapping of DD at a fine 101 scale (village-level) in the rural health district of Ifanadiana, in southeastern 102 Madagascar. We leveraged multiple spatio-temporal datasets, including a district- 103 representative longitudinal cohort study and health center case reports in Ifanadiana, to 104 identify the socio-ecological risk factors of DD. We then assessed our ability to predict 105 disease risk at a scale relevant to public health managers.