Patterns in disease across space and time are important to epidemiologists and health professionals because they may indicate underlying elevated disease risk. In some cases, elevated risk may be driven by environmental exposures, infectious diseases or other factors where timely public health interventions are important. The spatial and spatio-temporal scan statistics identify a single most likely cluster or equivalently select a single correct model. We instead consider an ensemble of single cluster models. We use stacking, a model-averaging technique, to combine relative risk estimates from all of the single cluster models into a sequence of meta-models indexed by the effective number of parameters/clusters. The number of parameters/spatiotemporal clusters is chosen using information criteria. A simulation study is conducted to demonstrate the statistical properties of the stacking method. The method is illustrated using a dataset of female breast cancer incidence data at the municipality level in Japan.

Introduction: There are significant disparities in the rates of maternal and infant morbidity and mortality in the United States – a discrepancy of particular importance in Wisconsin, where Non-Hispanic Black women experience the highest mortality rates in the country. The adverse effects of neighborhood socioeconomic status and geographical distance to obstetrical care outcomes have been demonstrated previously, with poor neighborhood socioeconomic status having been linked to higher rates of preterm births and low birth weight infants, which both increase the risk of neonatal morbidity and mortality. The objective of this study was to investigate the contributions of Area Deprivation Index and geographic location on age-matched birth weight z-scores. Methods: We conducted a retrospective cohort study of all singleton births >22 weeks’ gestation in Dane County, Wisconsin, from January 2016 through June 2018. Generalized additive models were adjusted for race/ethnicity, cigarette use, delivery route, pregnancy-related or chronic hypertension, pregestational and gestational diabetes, number of prenatal visits, maternal age, total weight gain, and pre-pregnancy body mass index. Results: There is evidence of an association between birth weight z-score and spatial location (median P value 0.006). With area deprivation, we found no evidence of an association with birth weight z-score (-0.01; 95% CI, -0.03 to 0.01; P = 0.109). Mean birth weight z-scores were lowest (-0.72) in the urban center of Madison, while mean birth weight z-score was highest (0.18) in rural areas near the northeast, southeast, and southwest county borders. We found an effect of race/ethnicity on birth weight. Conclusions: We identified geographic variations in birth weight at a granular level using census block groups and a holistic measure of deprivation, which can inform targeted public health interventions. The lack of evidence of area deprivation on birth outcomes but significant spatial trends demonstrated continued geographic disparities in our health care systems.

Context: Population health rankings can be a catalyst for the improvement of health by drawing attention to areas in need of relative improvement and summarizing complex information in a manner understood by almost everyone. However, ranks also have unintended consequences, such as being interpreted as “hard truths”, where variations may not be significant. There is a need to improve communication about uncertainty in ranks, with accurate interpretation. The most common solutions discussed in the literature have included modeling approaches to minimize statistical noise or borrow strength from covariates. However, the use of complex models can limit communication and implementation, especially for broad audiences. Objectives: Explore data-informed grouping (cluster analysis) as an easier-to-understand, empirical technique to account for rank imprecision that can be effectively communicated both numerically and visually. Design: Cluster analysis, specifically k-means clustering with Wasserstein (earth mover’s) distance, was explored as an approach to identify natural and meaningful groupings and gaps in the data distribution for the County Health Rankings’ (CHR) health outcomes ranks. Setting: County-level health outcomes from the 2022 CHR Participants: 3,082 counties that were ranked in the 2022 CHR Main Outcome Measure: Data-informed health groups Results: Cluster analysis identified 30 health groupings among counties nationwide, with cluster size ranging from nine to 184 counties. On average, states had 16 identified clusters, ranging from 3 in Delaware and Hawaii to 27 in Virginia. Number of clusters per state was associated with number of counties per state and population of the state. The method helped address many of the issues that arise from providing rank estimates alone. Conclusions: Public health practitioners can use this information to understand uncertainty in ranks, visualize distances between county ranks, have context around which counties are not meaningfully different from one another, and compare county performance to peer counties.

Prior research has shown that cancer risk varies by geography, but scan statistic methods for identifying cancer clusters in case-control studies have been limited in their ability to identify multiple clusters and adjust for participant-level risk factors. We develop a method to identify geographic patterns of breast cancer odds using the Wisconsin Women’s Health Study, a series of 5 population-based case-control studies of female Wisconsin residents aged 20-79 enrolled in 1988-2004 (cases=16,076, controls=16,795). We create sets of potential clusters by overlaying a 1 km grid over each county-neighborhood and enumerating a series of overlapping circles. Using a two-step approach, we fi=t a penalized binomial regression model to the number of cases and trials in each grid cell, penalizing all potential clusters by the least absolute shrinkage and selection operator (Lasso). We use BIC to select the number of clusters, which are included in a participant-level logistic regression model. We identify 15 geographic clusters, resulting in 23 areas of unique geographic odds ratios. After adjustment for known risk factors, confidence intervals narrowed but breast cancer odds ratios did not meaningfully change; one additional hotspot was identified. By considering multiple overlapping spatial clusters simultaneously, we discern gradients of spatial odds across Wisconsin.

Spatial and spatio-temporal cluster detection are important tools in public health and many other areas of application. Cluster detection can be approached as a multiple testing problem, typically using a space and time scan statistic. We recast the spatial and spatio-temporal cluster detection problem in a high-dimensional data analytical framework with Poisson or quasi-Poisson regression with the Lasso penalty. We adopt a fast and computationally-efficient method using a novel sparse matrix representation of the effects of potential clusters. The number of clusters and tuning parameters are selected based on (quasi-)information criteria. We evaluate the performance of our proposed method including the false positive detection rate and power using a simulation study. Application of the method is illustrated using breast cancer incidence data from three prefectures in Japan.