Risk factors and prediction model for repeated hand, foot, and mouth disease infections in Songjiang District, Shanghai, China: a retrospective cohort study (2016–2019) with temporal external validation using 2021–2025 surveillance data

Abstract Background Hand, foot, and mouth disease (HFMD) is a major public health concern in China, with repeated infections occurring in a subset of children. We aimed to identify factors independently associated with repeated HFMD infection using a rate-based framework with person-time offsets and develop a fixed-window prediction model with temporal external validation. Methods A retrospective cohort study used surveillance data from Songjiang District, Shanghai (2016–2019), including 14,408 individuals with valid 18-digit identification numbers. Repeated infection was a second episode ≥ 14 days later. The primary analysis estimated repeat-infection rate ratios (RRs) using Poisson regression with the logarithm of each individual’s observation time (index episode to end-2019) as an offset; overdispersion was assessed; negative binomial models served as sensitivity analyses. Sparse categories were merged before fitting. A two-year prediction model (repeated infection within 730 days) used L2-regularized logistic regression in individuals with a 2016–2017 index episode ( n = 7,684, 130 events), validated internally and temporally (2021–2023 index episodes; n = 5,191, 39 events). Results Among 14,408 individuals, 236 repeated episodes occurred over 30,339 person-years (0.78 per 100 person-years). In the multivariable model, male sex was associated with a higher repeat-infection rate (RR = 1.380, 95% CI: 1.053–1.808, P = 0.020), whereas kindergarten attendance (RR = 0.549, 95% CI: 0.393–0.769, P < 0.001) and student status (RR = 0.279, 95% CI: 0.106–0.733, P = 0.010) were associated with lower rates than scattered (home-cared) childcare. Summer (RR = 0.638, P = 0.008) and autumn (RR = 0.570, P = 0.004) index onsets were associated with lower rates than spring onset; these associations were attenuated after adjustment for calendar season during follow-up, indicating that they partly reflect the exposure structure. The prediction model remained exploratory with limited predictive utility (apparent AUC 0.649; temporal external AUC 0.612, 95% CI: 0.532–0.696; calibration slope 0.508). Conclusions Male sex and scattered (home-cared) childcare were independently associated with a higher repeat-infection rate, whereas the index-season associations should be interpreted cautiously because they partly reflect calendar-time exposure structure. The prediction model is exploratory with limited predictive utility. Findings may help characterize higher-risk subgroups pending prospective confirmation.