A predictive model for healthcare coverage in Yemen

Abstract

Introduction: The ongoing war in Yemen continues to pose challenges for healthcare coverage in the country especially with regards to critical gaps in information systems needed for planning and delivering health services. Restricted access to social services including safe drinking water and sanitation systems have likely led to an increase in the spread of diarrheal diseases which remains one of greatest sources of mortality in children under 5 years old. To overcome morbidity and mortality from diarrheal diseases among children in the context of severe information shortages, a predictive model is needed to determine the burden of diarrheal disease on Yemeni children and their ability to reach curative health services through an estimate of healthcare coverage. This will allow for national and local health authorities and humanitarian partners to make better informed decisions for planning and providing health care services.

Methods: A probabilistic Markov model was developed based on an analysis of Yemen's health facilities' clinical register data provided by UNICEF. The model combines this health system data with environmental and conflict-related factors such as the destruction of infrastructure (roads and health facilities) to fill in gaps in population-level data on the burden of diarrheal diseases on children under five, and the coverage rate of the under-five sick population with treatment services at primary care facilities. The model also provides estimates of the incidence rate, and treatment outcomes including treatment efficacy and mortality rate.

Results: By using alternatives to traditional healthcare data, the model was able to recreate the observed trends in treatment with no significant difference compared to provided validation data. Once validated, the model was used to predict the percent of sick children with diarrhea who were able to reach, and thus receive, treatment services (coverage rate) for 2019 which ranged between an average weekly minimum of 1.73% around the 28th week of the year to a weekly maximum coverage of just over 5% around the new year. These predictions can be translated into policy decisions such as when increased efforts are needed to reach children and what type of service delivery modalities may be the most effective.

Conclusion: The model developed and presented in this manuscript shows a seasonal trend in the spread of diarrheal disease in children under five living in Yemen through a novel incorporation of weather, infrastructure and conflict parameters in the model. Our model also provides new information on the number of children seeking treatment and how this is influenced by the ongoing conflict. Despite the work of the national and local health authorities with the support of aid organizations, during the mid-year rains up to 98% of children with diarrhea are unable to receive treatment services. Thus, it is recommended that community outreach or other delivery modalities through which services are delivered in closer proximity to those in need should be scaled up prior to and during these periods. This would serve to increase number of children able to receive treatment by lessening the prohibitive travel burden, or access constraint, on families during these times.

Keywords: Childhood illnesses; Computational modeling; Diarrheal disease; Healthcare coverage; Markov model; Yemen.

Fig. 1

Model schematic. Overview of the possible interactions between the four different health status in the model

Fig. 2

Incidence rate comparison. The comparison between data published by El Bcheraoui et al. and the average output of the model after 7 iterations for 2016. Error bars represent the upper and lower bounds of the 95% confidence interval analyzed in GraphPad. Although numerically close, the model’s output for the incidence rate is deemed to be statically different as it falls outside the confidence interval of the target estimation data

Fig. 3

Comparison to Number of Children Treated (a), Treatment Mortality (b) and Treatment Efficacy (c). Comparison of the model against validation data showing no significant difference between the a total number of children under 5 treated (P = .8864) and b the treatment mortality rates from 2018 health facility monitoring data collected through a third party (P = .4479). c The efficacy of Zn/ORS intervention from the same data set (P = .8450) was also not deemed to be significantly different than the model. Significance was assessed with a one sample T test in GraphPad. Error bars shown as standard deviation for n = 7 code iterations

Fig. 4

Health trends for children under five in Yemen during May–September of 2018. a-e. The week number is the number of weeks since the start of the year. a Overall weekly distribution of the health-states for the estimated child population and b the zoomed in to see the bottom half of the graph. c Plot of the total Sick health status population vs time compared against the In-treatment health status vs time, with the In-Treatment health status enlarged in (d). e The coverage rate was listed week by week

Fig. 5

Deterministic Sensitivity Analysis for the maximum estimated coverage (a) and the minimum estimated coverage (b). Parameters were swept from 10 to 200% of their baseline values except for the Weather Amplitude, which was varied from 10 to 190% due to larger values producing unrealistic probabilities (greater than 1 or less than 0). The three most sensitive parameters were the maximum road value, weather amplitude and the minimum road value in all cases. a The maximum coverage was more sensitive to decreasing values compared to the minimum coverage (b) which was equally sensitive to parameter changes in both directions

Fig. 6

Predictions for each state (a), no Deceased and no Healthy (b), or Sick states (c) and weekly coverage rate (d). a Tracked changes over all health status. b Comparison between Sick and In-treatment health status throughout 2019 and c a magnified view of the Sick health status. d The weekly percent treatment coverage during the year