Evaluating City-Cut Geographic Risk Classification for a Pandemic Travel-Record System in Shanghai, China: Retrospective Observational Analysis
JMIR mHealth and uHealth · 2026-11-09 · 14 · e64941-e64941
DOI: 10.2196/64941 ↗Astratt
Abstract Background Smartphone-delivered travel-record and health-code systems were deployed across China during the COVID-19 pandemic to classify risk based on travel history and regulate access to venues and travel. The geographic classification rule is a key design choice for these systems. A city-boundary rule (city-cut) is administratively convenient but may group areas with different mobility patterns under one risk label, whereas a fixed-radius rule (radius-cut) offers greater spatial precision but may exclude more distant areas connected by routine travel. Whether a classification rule aligns with observed mobility is a design question distinct from evaluating a deployed system’s epidemiological effects. Objective This study examined whether observed mobility in Shanghai supported a city-cut or a radius-cut geographic design for a mobile health travel-record system and whether mobility-derived communities corresponded to administrative districts. Methods We conducted a retrospective observational analysis of 1.44 billion deidentified China Unicom mobility records collected for two 46-day periods in 2019 and 2020. The analytical unit was the daily origin-area movement proportion. We calculated within-Shanghai, confirmed external-city, and nonadjacent-district movement proportions; for each origin area, we paired them by calendar day. We constructed grid-level transition matrices, used spectral clustering followed by k-means++, and checked the groupings across repeated initializations and selected dates. Results Across 46 daily city-average observations in 2019, the intracity proportion was 97.2% (95% t CI 97.1%‐97.3%). Nonadjacent-district movement was 13.7% (95% t CI 13.5%‐13.9%), compared with 2.8% (95% t CI 2.7%‐2.9%) for confirmed external-city movement. A paired, 2-tailed Newey-West heteroskedasticity- and autocorrelation-consistent (HAC) t test (α=.05) gave a mean difference of 10.9 (95% HAC CI 10.7‐11.1; P
EUVIMED