Estimating hospital catchments from in-patient admission records: a spatial statistical approach applied to malaria

Quantifying the impact of hospital catchment area definitions on hospital admissions forecasts: COVID-19 in England, September 2020-April 2021

BACKGROUND

Defining healthcare facility catchment areas is a key step in predicting future healthcare demand in epidemic settings. Forecasts of hospitalisations can be informed by leading indicators measured at the community level. However, this relies on the definition of so-called catchment areas or the geographies whose populations make up the patients admitted to a given hospital, which are often not well-defined. Little work has been done to quantify the impact of hospital catchment area definitions on healthcare demand forecasting.

METHODS

We made forecasts of local-level hospital admissions using a scaled convolution of local cases (as defined by the hospital catchment area) and delay distribution. Hospital catchment area definitions were derived from either simple heuristics (in which people are admitted to their nearest hospital or any nearby hospital) or historical admissions data (all emergency or elective admissions in 2019, or COVID-19 admissions), plus a marginal baseline definition based on the distribution of all hospital admissions. We evaluated predictive performance using each hospital catchment area definition using the weighted interval score and considered how this changed by the length of the predictive horizon, the date on which the forecast was made, and by location. We also considered the change, if any, in the relative performance of each definition in retrospective vs. real-time settings, or at different spatial scales.

RESULTS

The choice of hospital catchment area definition affected the accuracy of hospital admission forecasts. The definition based on COVID-19 admissions data resulted in the most accurate forecasts at both a 7- and 14-day horizon and was one of the top two best-performing definitions across forecast dates and locations. The "nearby" heuristic also performed well, but less consistently than the COVID-19 data definition. The marginal distribution baseline, which did not include any spatial information, was the lowest-ranked definition. The relative performance of the definitions was larger when using case forecasts compared to future observed cases. All results were consistent across spatial scales of the catchment area definitions.

CONCLUSIONS

Using catchment area definitions derived from context-specific data can improve local-level hospital admission forecasts. Where context-specific data is not available, using catchment areas defined by carefully chosen heuristics is a sufficiently good substitute. There is clear value in understanding what drives local admissions patterns, and further research is needed to understand the impact of different catchment area definitions on forecast performance where case trends are more heterogeneous.

Mathematical modeling of malaria transmission dynamics in humans with mobility and control states

Malaria importation is one of the hypothetical drivers of malaria transmission dynamics across the globe. Several studies on malaria importation focused on the effect of the use of conventional malaria control strategies as approved by the World Health Organization (WHO) on malaria transmission dynamics but did not capture the effect of the use of traditional malaria control strategies by vigilant humans. In order to handle the aforementioned situation, a novel system of Ordinary Differential Equations (ODEs) was developed comprising the human and the malaria vector compartments. Analysis of the system was carried out to assess its quantitative properties. The novel computational algorithm used to solve the developed system of ODEs was implemented and benchmarked with the existing Runge-Kutta numerical solution method. Furthermore, simulations of different vigilant conditions useful to control malaria were carried out. The novel system of malaria models was well-posed and epidemiologically meaningful based on its quantitative properties. The novel algorithm performed relatively better in terms of model simulation accuracy than Runge-Kutta. At the best model-fit condition of 98% vigilance to the use of conventional and traditional malaria control strategies, this study revealed that malaria importation has a persistent impact on malaria transmission dynamics. In lieu of this, this study opined that total vigilance to the use of the WHO-approved and traditional malaria management tools would be the most effective control strategy against malaria importation.

Geographic accessibility and hospital competition for emergency blood transfusion services in Bungoma, Western Kenya

BACKGROUND

Estimating accessibility gaps to essential health interventions helps to allocate and prioritize health resources. Access to blood transfusion represents an important emergency health requirement. Here, we develop geo-spatial models of accessibility and competition to blood transfusion services in Bungoma County, Western Kenya.

METHODS

Hospitals providing blood transfusion services in Bungoma were identified from an up-dated geo-coded facility database. AccessMod was used to define care-seeker's travel times to the nearest blood transfusion service. A spatial accessibility index for each enumeration area (EA) was defined using modelled travel time, population demand, and supply available at the hospital, assuming a uniform risk of emergency occurrence in the county. To identify populations marginalized from transfusion services, the number of people outside 1-h travel time and those residing in EAs with low accessibility indexes were computed at the sub-county level. Competition between the transfusing hospitals was estimated using a spatial competition index which provided a measure of the level of attractiveness of each hospital. To understand whether highly competitive facilities had better capacity for blood transfusion services, a correlation test between the computed competition metric and the blood units received and transfused at the hospital was done.

RESULTS

15 hospitals in Bungoma county provide transfusion services, however these are unevenly distributed across the sub-counties. Average travel time to a blood transfusion centre in the county was 33 min and 5% of the population resided outside 1-h travel time. Based on the accessibility index, 38% of the EAs were classified to have low accessibility, representing 34% of the population, with one sub-county having the highest marginalized population. The computed competition index showed that hospitals in the urban areas had a spatial competitive advantage over those in rural areas.

CONCLUSION

The modelled spatial accessibility has provided an improved understanding of health care gaps essential for health planning. Hospital competition has been illustrated to have some degree of influence in provision of health services hence should be considered as a significant external factor impacting the delivery, and re-design of available services.

Defining catchment boundaries and their populations for Aotearoa New Zealand's rural hospitals

Introduction There is considerable variation in the structure and resources of New Zealand (NZ) rural hospitals; however, these have not been recently quantified and their effects on healthcare outcomes are poorly understood. Importantly, there is no standardised description of each rural hospital's catchment boundary and the characteristics of the population living within this area. Aim To define and describe a catchment population for each of New Zealand's rural hospitals. Methods An exploratory approach to developing catchments was employed. Geographic Information Systems were used to develop drive-time-based geographic catchments, and administrative health data (National Minimum Data Set and Primary Health Organisation Data Set) informed service utilisation-based catchments. Catchments were defined at both the Statistical Area 2 (SA2) and domicile levels, and linked to census-based population data, the Geographic Classification for Health, and the area-level New Zealand Index of Socioeconomic Deprivation (NZDep2018). Results Our results highlight considerable heterogeneity in the size (max: 57 564, min: 5226) and characteristics of populations served by rural hospitals. Substantial differences in the age structure, ethnic composition, socio-economic profile, 'remoteness' and projected future populations, are noted. Discussion In providing a standardised description of each rural hospital's catchment boundary and its population characteristics, the considerable heterogeneity of the communities served by rural hospitals, both in size, rurality and socio-demographic characteristics, is highlighted. The findings provide a platform on which to build further research regarding NZ's rural hospitals and inform the delivery of high-quality, cost-effective and equitable health care for people living in rural NZ.

Application of Bayesian Regression for the Identification of a Catchment Area for Cancer Cases in Dogs and Cats

Research on cancer in dogs and cats, among other diseases, finds an important source of information in registry data collected from hospitals. These sources have proved to be decisive in establishing incidences and identifying temporal patterns and risk factors. However, the attendance of patients is not random, so the correct delimitation of the hospital catchment area (CA) as well as the identification of the factors influencing its shape is relevant to prevent possible biases in posterior inferences. Despite this, there is a lack of data-driven approaches in veterinary epidemiology to establish CA. Therefore, our aim here was to apply a Bayesian method to estimate the CA of a hospital. We obtained cancer (n = 27,390) and visit (n = 232,014) registries of dogs and cats attending the Veterinary Medical Teaching Hospital of the University of California, Davis from 2000 to 2019 with 2,707 census tracts (CTs) of 40 neighboring counties. We ran hierarchical Bayesian models with different likelihood distributions to define CA for cancer cases and visits based on the exceedance probabilities for CT random effects, adjusting for species and period (2000-2004, 2005-2009, 2010-2014, and 2015-2019). The identified CAs of cancer cases and visits represented 75.4 and 83.1% of the records, respectively, including only 34.6 and 39.3% of the CT in the study area. The models detected variation by species (higher number of records in dogs) and period. We also found that distance to hospital and average household income were important predictors of the inclusion of a CT in the CA. Our results show that the application of this methodology is useful for obtaining data-driven CA and evaluating the factors that influence and predict data collection. Therefore, this could be useful to improve the accuracy of analysis and inferences based on registry data.

Delineating natural catchment health districts with routinely collected health data from women's travel to give birth in Ghana

Background

Health service areas are essential for planning, policy and managing public health interventions. In this study, we delineate health service areas from routinely collected health data as a robust geographic basis for presenting access to maternal care indicators.

Methods

A zone design algorithm was adapted to delineate health service areas through a cross-sectional, ecological study design. Health sub-districts were merged into health service areas such that patient flows across boundaries were minimised. Delineated zones and existing administrative boundaries were used to provide estimates of access to maternal health services. We analysed secondary data comprising routinely collected health records from 32,921 women attending 27 hospitals to give birth, spatial demographic data, a service provision assessment on the quality of maternal healthcare and health sub-district boundaries from Eastern Region, Ghana.

Results

Clear patterns of cross border movement to give birth emerged from the analysis, but more women originated closer to the hospitals. After merging the 250 sub-districts in 33 districts, 11 health service areas were created. The minimum percent of internal flows of women giving birth within any health service area was 97.4%. Because the newly delineated boundaries are more “natural” and sensitive to observed flow patterns, when we calculated areal indicator estimates, they showed a marked improvement over the existing administrative boundaries, with the inclusion of a hospital in every health service area.

Conclusion

Health planning can be improved by using routine health data to delineate natural catchment health districts. In addition, data-driven geographic boundaries derived from public health events will improve areal health indicator estimates, planning and interventions.

Geographic barriers to establishing a successful hospital referral system in rural Madagascar

Background

The provision of emergency and hospital care has become an integral part of the global vision for universal health coverage. To strengthen secondary care systems, we need to accurately understand the time necessary for populations to reach a hospital. The goal of this study was to develop methods that accurately estimate referral and prehospital time for rural districts in low and middle-income countries. We used these estimates to assess how local geography can limit the impact of a strengthened referral programme in a rural district of Madagascar.

Methods

We developed a database containing: travel speed by foot and motorised vehicles in Ifanadiana district; a full mapping of all roads, footpaths and households; and remotely sensed data on terrain, land cover and climatic characteristics. We used this information to calibrate estimates of referral and prehospital time based on the shortest route algorithms and statistical models of local travel speed. We predict the impact on referral numbers of strategies aimed at reducing referral time for underserved populations via generalised linear mixed models.

Results

About 10% of the population lived less than 2 hours from the hospital, and more than half lived over 4 hours away, with variable access depending on climatic conditions. Only the four health centres located near the paved road had referral times to the hospital within 1 hour. Referral time remained the main barrier limiting the number of referrals despite health system strengthening efforts. The addition of two new referral centres is estimated to triple the population living within 2 hours from a centre with better emergency care capacity and nearly double the number of expected referrals.

Conclusion

This study demonstrates how adapting geographic accessibility modelling methods to local scales can occur through improving the precision of travel time estimates and pairing them with data on health facility use.

Using Catchment Population to Estimate Sporadic Creutzfeldt-Jakob Disease Incidence

INTRODUCTION

Catchment populations have several uses. A method using catchment population to estimate the incidence of sporadic Creutzfeldt-Jakob disease (sCJD) is described.

MATERIALS AND METHODS

A cohort of nine consecutive patients diagnosed with sCJD, symptom onset spanning 26 months, were observed at a rural tertiary university medical center that has approximately 40,000 hospital discharges annually. An effective catchment population was determined using surrounding county utilization frequency that captured all nine sCJD patients and accounted for over 87% of discharges.

RESULTS

The effective sCJD hospital catchment population was 1.266 million, implying an annual sCJD incidence rate of 3.39 per million (95% CIs, 1.55-6.43), assuming a Poisson distribution for sCJD occurrence.

CONCLUSIONS

This annual incidence rate suggests that many sCJD patients are unrecognized and unreported. An advantage of this catchment population method is independence from death certificate accuracy, important in rare diseases that are both rapidly and invariably fatal. The relative absence of significant healthcare systems competition in this rural population enhances the reliability of this finding. The most likely explanation for the high sCJD incidence rate suggested by this study is enhanced clinical suspicion and improved diagnostic accuracy.

Estimating malaria incidence from routine health facility-based surveillance data in Uganda

Background

Accurate measures of malaria incidence are essential to track progress and target high-risk populations. While health management information system (HMIS) data provide counts of malaria cases, quantifying the denominator for incidence using these data is challenging because catchment areas and care-seeking behaviours are not well defined. This study’s aim was to estimate malaria incidence using HMIS data by adjusting the population denominator accounting for travel time to the health facility.

Methods

Outpatient data from two public health facilities in Uganda (Kihihi and Nagongera) over a 3-year period (2011–2014) were used to model the relationship between travel time from patient village of residence (available for each individual) to the facility and the relative probability of attendance using Poisson generalized additive models. Outputs from the model were used to generate a weighted population denominator for each health facility and estimate malaria incidence. Among children aged 6 months to 11 years, monthly HMIS-derived incidence estimates, with and without population denominators weighted by probability of attendance, were compared with gold standard measures of malaria incidence measured in prospective cohorts.

Results

A total of 48,898 outpatient visits were recorded across the two sites over the study period. HMIS incidence correlated with cohort incidence over time at both study sites (correlation in Kihihi = 0.64, p < 0.001; correlation in Nagongera = 0.34, p = 0.045). HMIS incidence measures with denominators unweighted by probability of attendance underestimated cohort incidence aggregated over the 3 years in Kihihi (0.5 cases per person-year (PPY) vs 1.7 cases PPY) and Nagongera (0.3 cases PPY vs 3.0 cases PPY). HMIS incidence measures with denominators weighted by probability of attendance were closer to cohort incidence, but remained underestimates (1.1 cases PPY in Kihihi and 1.4 cases PPY in Nagongera).

Conclusions

Although malaria incidence measured using HMIS underestimated incidence measured in cohorts, even when adjusting for probability of attendance, HMIS surveillance data are a promising and scalable source for tracking relative changes in malaria incidence over time, particularly when the population denominator can be estimated by incorporating information on village of residence.

Determinants of subnational disparities in antenatal care utilisation: a spatial analysis of demographic and health survey data in Kenya

BACKGROUND

The spatial variation in antenatal care (ANC) utilisation is likely associated with disparities observed in maternal and neonatal deaths. Most maternal deaths are preventable through services offered during ANC; however, estimates of ANC coverage at lower decision-making units (sub-county) is mostly lacking. In this study, we aimed to estimate the coverage of at least four ANC (ANC4) visits at the sub-county level using the 2014 Kenya Demographic and Health Survey (KDHS 2014) and identify factors associated with ANC utilisation in Kenya.

METHODS

Data from the KDHS 2014 was used to compute sub-county estimates of ANC4 using small area estimation (SAE) techniques which relied on spatial relatedness to yield precise and reliable estimates at each of the 295 sub-counties. Hierarchical mixed-effect logistic regression was used to identify factors influencing ANC4 utilisation. Sub-county estimates of factors significantly associated with ANC utilisation were produced using SAE techniques and mapped to visualise disparities.

RESULTS

The coverage of ANC4 across sub-counties was heterogeneous, ranging from a low of 17% in Mandera West sub-county to over 77% in Nakuru Town West and Ruiru sub-counties. Thirty-one per cent of the 295 sub-counties had coverage of less than 50%. Maternal education, household wealth, place of delivery, marital status, age at first marriage, and birth order were all associated with ANC utilisation. The areas with low ANC4 utilisation rates corresponded to areas of low socioeconomic status, fewer educated women and a small number of health facility deliveries.

CONCLUSION

Suboptimal coverage of ANC4 and its heterogeneity at sub-county level calls for urgent, focused and localised approaches to improve access to antenatal care services. Policy formulation and resources allocation should rely on data-driven strategies to guide national and county governments achieve equity in access and utilisation of health interventions.

The age-specific incidence of hospitalized paediatric malaria in Uganda

BACKGROUND

Understanding the relationship between malaria infection risk and disease outcomes represents a fundamental component of morbidity and mortality burden estimations. Contemporary data on severe malaria risks among populations of different parasite exposures are scarce. Using surveillance data, we compared rates of paediatric malaria hospitalisation in areas of varying parasite exposure levels.

METHODS

Surveillance data at five public hospitals; Jinja, Mubende, Kabale, Tororo, and Apac were assembled among admissions aged 1 month to 14 years between 2017 and 2018. The address of each admission was used to define a local catchment population where national census data was used to define person-year-exposure to risk. Within each catchment, historical infection prevalence was assembled from previously published data and current infection prevalence defined using 33 population-based school surveys among 3400 children. Poisson regression was used to compute the overall and site-specific incidences with 95% confidence intervals.

RESULTS

Both current and historical Plasmodium falciparum prevalence varied across the five sites. Current prevalence ranged from < 1% in Kabale to 54% in Apac. Overall, the malaria admission incidence rate (IR) was 7.3 per 1000 person years among children aged 1 month to 14 years of age (95% CI: 7.0, 7.7). The lowest rate was described at Kabale (IR = 0.3; 95 CI: 0.1, 0.6) and highest at Apac (IR = 20.3; 95 CI: 18.9, 21.8). There was a correlation between IR across the five sites and the current parasite prevalence in school children, though findings were not statistically significant. Across all sites, except Kabale, malaria admissions were concentrated among young children, 74% were under 5 years. The median age of malaria admissions at Kabale hospital was 40 months (IQR 20, 72), and at Apac hospital was 36 months (IQR 18, 69). Overall, severe anaemia (7.6%) was the most common presentation and unconsciousness (1.8%) the least common.

CONCLUSION

Malaria hospitalisation rates remain high in Uganda particularly among young children. The incidence of hospitalized malaria in different locations in Uganda appears to be influenced by past parasite exposure, immune acquisition, and current risks of infection. Interruption of transmission through vector control could influence age-specific severe malaria risk.