An evaluation of the edge effects in disease map modelling

Model-based disease mapping using primary care registry data

BACKGROUND

Spatial modeling of disease risk using primary care registry data is promising for public health surveillance. However, it remains unclear to which extent challenges such as spatially disproportionate sampling and practice-specific reporting variation affect statistical inference.

METHODS

Using lower respiratory tract infection data from the INTEGO registry, modeled with a logistic model incorporating patient characteristics, a spatially structured random effect at municipality level, and an unstructured random effect at practice level, we conducted a case and simulation study to assess the impact of these challenges on spatial trend estimation.

RESULTS

Even with spatial imbalance and practice-specific reporting variation, the model performed well. Performance improved with increasing spatial sample balance and decreasing practice-specific variation.

CONCLUSION

Our findings indicate that, with correction for reporting efforts, primary care registries are valuable for spatial trend estimation. The diversity of patient locations within practice populations plays an important role.

Evaluation and comparison of spatial cluster detection methods for improved decision making of disease surveillance: a case study of national dengue surveillance in Thailand

BACKGROUND

Dengue is a mosquito-borne disease that causes over 300 million infections worldwide each year with no specific treatment available. Effective surveillance systems are needed for outbreak detection and resource allocation. Spatial cluster detection methods are commonly used, but no general guidance exists on the most appropriate method for dengue surveillance. Therefore, a comprehensive study is needed to assess different methods and provide guidance for dengue surveillance programs.

METHODS

To evaluate the effectiveness of different cluster detection methods for dengue surveillance, we selected and assessed commonly used methods: Getis Ord [Formula: see text], Local Moran, SaTScan, and Bayesian modeling. We conducted a simulation study to compare their performance in detecting clusters, and applied all methods to a case study of dengue surveillance in Thailand in 2019 to further evaluate their practical utility.

RESULTS

In the simulation study, Getis Ord [Formula: see text] and Local Moran had similar performance, with most misdetections occurring at cluster boundaries and isolated hotspots. SaTScan showed better precision but was less effective at detecting inner outliers, although it performed well on large outbreaks. Bayesian convolution modeling had the highest overall precision in the simulation study. In the dengue case study in Thailand, Getis Ord [Formula: see text] and Local Moran missed most disease clusters, while SaTScan was mostly able to detect a large cluster. Bayesian disease mapping seemed to be the most effective, with adaptive detection of irregularly shaped disease anomalies.

CONCLUSIONS

Bayesian modeling showed to be the most effective method, demonstrating the best accuracy in adaptively identifying irregularly shaped disease anomalies. In contrast, SaTScan excelled in detecting large outbreaks and regular forms. This study provides empirical evidence for the selection of appropriate tools for dengue surveillance in Thailand, with potential applicability to other disease control programs in similar settings.

Unmatched spatially stratified controls: A simulation study examining efficiency and precision using spatially-diverse controls and generalized additive models

Unmatched spatially stratified random sampling (SSRS) of non-cases selects geographically balanced controls by dividing the study area into spatial strata and randomly selecting controls from all non-cases within each stratum. The performance of SSRS control selection was evaluated in a case study spatial analysis of preterm birth in Massachusetts. In a simulation study, we fit generalized additive models using controls selected by SSRS or simple random sample (SRS) designs. We compared mean squared error (MSE), bias, relative efficiency (RE), and statistically significant map results to the model results with all non-cases. SSRS designs had lower average MSE (0.0042-0.0044) and higher RE (77-80%) compared to SRS designs (MSE: 0.0072-0.0073; RE across designs: 71%). SSRS map results were more consistent across simulations, reliably identifying statistically significant areas. SSRS designs improved efficiency by selecting controls that are geographically distributed, particularly from low population density areas, and may be more appropriate for spatial analyses.

A comparison of multiple neighborhood matrix specifications for spatio-temporal model fitting: a case study on COVID-19 data

Establishing proper neighbor relations between a set of spatial units under analysis is essential when carrying out a spatial or spatio-temporal analysis. However, it is usual that researchers choose some of the most typical (and simple) neighborhood structures, such as the first-order contiguity matrix, without exploring other options. In this paper, we compare the performance of different neighborhood matrices in the context of modeling the weekly relative risk of COVID-19 over small areas located in or near Valencia, Spain. Specifically, we construct contiguity-based, distance-based, covariate-based (considering mobility flows and sociodemographic characteristics), and hybrid neighborhood matrices. We evaluate the goodness of fit, the overall predictive quality, the ability to detect high-risk spatio-temporal units, the capability to capture the spatio-temporal autocorrelation in the data, and the goodness of smoothing for a set of spatio-temporal models based on each of the neighborhood matrices. The results show that contiguity-based matrices, some of the distance-based matrices, and those based on sociodemographic characteristics perform better than the matrices based on k-nearest neighbors and those involving mobility flows. In addition, we test the linear combination of some of the constructed neighborhood matrices and the reweighting of these matrices after eliminating weak neighbor relations, without any model improvement.

Does the edge effect impact on the measure of spatial accessibility to healthcare providers?

BACKGROUND

Spatial accessibility indices are increasingly applied when investigating inequalities in health. Although most studies are making mentions of potential errors caused by the edge effect, many acknowledge having neglected to consider this concern by establishing spatial analyses within a finite region, settling for hypothesizing that accessibility to facilities will be under-reported. Our study seeks to assess the effect of edge on the accuracy of defining healthcare provider access by comparing healthcare provider accessibility accounting or not for the edge effect, in a real-world application.

METHODS

This study was carried out in the department of Nord, France. The statistical unit we use is the French census block known as 'IRIS' (Ilot Regroupé pour l'Information Statistique), defined by the National Institute of Statistics and Economic Studies. The geographical accessibility indicator used is the "Index of Spatial Accessibility" (ISA), based on the E2SFCA algorithm. We calculated ISA for the pregnant women population by selecting three types of healthcare providers: general practitioners, gynecologists and midwives. We compared ISA variation when accounting or not edge effect in urban and rural zones. The GIS method was then employed to determine global and local autocorrelation. Lastly, we compared the relationship between socioeconomic distress index and ISA, when accounting or not for the edge effect, to fully evaluate its impact.

RESULTS

The results revealed that on average ISA when offer and demand beyond the boundary were included is slightly below ISA when not accounting for the edge effect, and we found that the IRIS value was more likely to deteriorate than improve. Moreover, edge effect impact can vary widely by health provider type. There is greater variability within the rural IRIS group than within the urban IRIS group. We found a positive correlation between socioeconomic distress variables and composite ISA. Spatial analysis results (such as Moran's spatial autocorrelation index and local indicators of spatial autocorrelation) are not really impacted.

CONCLUSION

Our research has revealed minor accessibility variation when edge effect has been considered in a French context. No general statement can be set up because intensity of impact varies according to healthcare provider type, territorial organization and methodology used to measure the accessibility to healthcare. Additional researches are required in order to distinguish what findings are specific to a territory and others common to different countries. It constitute a promising direction to determine more precisely healthcare shortage areas and then to fight against social health inequalities.

An approach for de-identification of point locations of livestock premises for further use in disease spread modeling

We describe a method for de-identifying point location data used for disease spread modeling to allow data custodians to share data with modeling experts without disclosing individual farm identities. The approach is implemented in an open-source software program that is described and evaluated here. The program allows a data custodian to select a level of de-identification based on the K-anonymity statistic. The program converts a file of true farm locations and attributes into a file appropriate for use in disease spread modeling with the locations randomly modified to prevent re-identification based on location. Important epidemiological relationships such as clustering are preserved to as much as possible to allow modeling similar to those using true identifiable data. The software implementation was verified by visual inspection and basic descriptive spatial analysis of the output. Performance is sufficient to allow de-identification of even large data sets on desktop computers available to any data custodian.

Distance-based mapping of disease risk

In this article, we consider the problem of comparing the distribution of observations in a planar region to a pre-specified null distribution. Our motivation is a surveillance setting where we map locations of incident disease, aiming to monitor these data over time, to locate potential areas of high/low incidence so as to direct public health actions. We propose a non-parametric approach to distance-based disease risk mapping inspired by tomographic imaging. We consider several one-dimensional projections via the observed distribution of distances to a chosen fixed point; we then compare this distribution to that expected under the null and average these comparisons across projections to compute a relative-risk-like score at each point in the region. The null distribution can be established from historical data. Scores are displayed on the map using a color scale. In addition, we give a detailed description of the method along with some desirable theoretical properties. To further assess the performance of this method, we compare it to the widely used log ratio of kernel density estimates. As a performance metric, we evaluate the accuracy to locate simulated spatial clusters superimposed on a uniform distribution in the unit disk. Results suggest that both methods can adequately locate this increased risk but each relies on an appropriate choice of parameters. Our proposed method, distance-based mapping (DBM), can also generalize to arbitrary metric spaces and/or high-dimensional data.

A generalized Poisson-gamma model for spatially overdispersed data

Modern disease mapping commonly uses hierarchical Bayesian methods to model overdispersion and spatial correlation. Classical random-effects based solutions include the Poisson-gamma model, which uses the conjugacy between the Poisson and gamma distributions, but which does not model spatial correlation, on the one hand, and the more advanced CAR model, which also introduces a spatial autocorrelation term but without a closed-form posterior distribution on the other. In this paper, a combined model is proposed: an alternative convolution model accounting for both overdispersion and spatial correlation in the data by combining the Poisson-gamma model with a spatially-structured normal CAR random effect. The Limburg Cancer Registry data on kidney and prostate cancer in Limburg were used to compare the conventional and new models. A simulation study confirmed results and interpretations coming from the real datasets. Relative risk maps showed that the combined model provides an intermediate between the non-patterned negative binomial and the sometimes oversmoothed CAR convolution model.

Community-level risk factors for depression hospitalizations

This study measured geographic variation in depression hospitalizations and identified community-level risk factors. Depression hospitalizations were identified from the Statewide Inpatient Database. The dependent variable was specified as the indirectly standardized hospitalization rate. County-level data for 14 states were collected from federal agencies. The Bayesian spatial regression model included socio-demographic, economic, and health system characteristics as independent variables. There were 8.5 depression hospitalizations per 1,000 residents. 8.8% of counties had hospitalization rates 33% greater than the standardized rate. Significant risk factors included unemployment, poverty, physician supply, and hospital bed supply. Significant protective factors included rurality, economic dependence, and housing stress.