Multiple-cluster detection test for purely temporal disease clustering: Integration of scan statistics and generalized linear models

Abstract

The spatial scan statistic is commonly used to detect spatial and/or temporal disease clusters in epidemiological studies. Although multiple clusters in the study space can be thus identified, current theoretical developments are mainly based on detecting a ‘single’ cluster. The standard scan statistic procedure enables the detection of multiple clusters, recursively identifying additional ‘secondary’ clusters. However, their p-values are calculated one at a time, as if each cluster is a primary one. Therefore, a new procedure that can accurately evaluate multiple clusters as a whole is needed. The present study focuses on purely temporal cases and then proposes a new test procedure that evaluates the p-value for multiple clusters, combining generalized linear models with an information criterion approach. This framework encompasses the conventional, currently widely used detection procedure as a special case. An application study adopting the new framework is presented, analysing the Japanese daily incidence of out-of-hospital cardiac arrest cases. The analysis reveals that the number of the incident increases around New Year’s Day in Japan. Further, simulation studies undertaken confirm that the proposed method possesses a consistency property that tends to select the correct number of clusters when the truth is known.

Detecting multiple spatial disease clusters: information criterion and scan statistic approach

“…Thus, the former identifies fewer significant secondary clusters relative to true clusters. This was observed in our simulation study, while the proposed procedure tends to detect more clusters, contrasting the reported result in the purely temporal setting [26], although this may largely depend on the scenario assumed.…”

“…The Monte Carlo hypothesis testing procedure evaluates the statistical significance of appropriate models in the same manner as the standard scan statistic. Under the null hypothesis, a large number of random datasets are generated; however, for each of these, is instead calculated as a test statistic (see details [ 26 ]).…”

“…3 Detected clustered-areas. The blue shaded areas were detected by both the conventional SCP and proposed approaches, whereas the orange shaded areas were selected only by the proposed approach Takahashi and Shimadzu Int J Health Geogr (2020) 19:33 criterion approach is based on the framework proposed by Takahashi and Shimadzu [26] for detecting multiple temporal-clusters. The idea of model selection has been used in more general statistical modeling contexts; for instance, Akaike information criterion (AIC) and Bayesian information criterion (BIC) are used to estimate the number of multiple clusters [31,32] and finite mixtures [33].…”

“…Therefore, a comprehensive approach is needed. A recent study suggests that a combined approach of statistical modeling and model selection can offer a potential solution by illustrating a case study that detects purely temporal clusters with a time series model in Takahashi and Shimadzu [26]. However, it is not always straightforward if the time series framework is directly applicable to a spatial context, which involves an extra dimension.…”

“…Here, we propose a unified framework that enables simultaneous detection and evaluation of multiple spatial-clusters by combining generalized linear models (GLMs) and information criterion approaches. The framework encompasses the procedure proposed for detecting purely temporal clusters in Takahashi and Shimadzu [ 26 ] as a special case. We present an illustrative example, the hospital admission for chronic obstructive pulmonary disease (COPD) in England, available from a textbook [ 27 ], for evaluating the performance of the proposed method.…”