An exploratory study of patient hospitalization patterns and behavioral risk factors using mobile phone location data.

Abstract

This exploratory observational study examined associations between behavioral patterns and hospitalization using geographic analysis of human mobility data. Anonymized GPS-based location data were obtained from users of a mobile location data platform in the Tokyo South Medical Region (Shinagawa and Ota wards) between April 1, 2022, and February 28, 2023. Behavioral patterns, including eating-out frequency and healthcare facility visits, were inferred from movement data. Hospitalization was defined as a hospital stay of at least two consecutive days, identified based on location trajectories. The primary endpoint was the association between eating-out behavior and hospitalization during the observation period. Logistic regression models were used to estimate odds ratios (ORs) and 95% confidence intervals (CIs). Venn diagrams were used to illustrate overlapping behavioral characteristics between groups, complementing regression-based analyses by visualizing behaviors that are difficult to capture through average regression effects and by supporting hypothesis generation. In total, 647 participants were included, of whom 580 were classified as outpatients, whereas 67 experienced hospitalization during the study period. Higher odds of hospitalization were observed only among participants who ate out more than eight times per month, compared with those who ate out less frequently (OR = 1.89, 95% CI: 1.05-3.43). No clear association was observed for the other eating-out frequency categories, with confidence intervals that included the null value. Venn diagram analyses suggested that individuals with both frequent outpatient visits and frequent eating-out behavior were more prevalent in the hospitalized group. These findings demonstrate the potential utility of mobile phone-derived mobility data for identifying behavioral patterns associated with hospitalization. However, given the observational design, reliance on indirect behavioral proxies, and missing demographic information (including age and sex), the results should be interpreted with caution. Further studies integrating detailed clinical and demographic data are needed to clarify causal relationships and evaluate the applicability of mobility-based behavioral indicators in healthcare settings.