Unequal recovery of health facility visits in hospitals and township health centres after COVID-19 disruption in Henan Province, China, 2014–2024: a time-series study

Background The COVID-19 pandemic disrupted health service utilisation worldwide. Its long-term effect remains uncertain, especially after the major adjustment of China’s COVID-19 control policies at the end of 2022. Evidence is also limited on whether recovery differed between hospitals and township health centres (THCs). The research aimed to evaluate the impact of the COVID-19 pandemic and its ending on health facility visits in hospitals and THCs in Henan Province from 2014 to 2024. Methods Monthly aggregate data on total health facility visits in hospitals and THCs in Henan Province from January, 2014, to December, 2024, were obtained from the health statistics system. The observation period was divided into the pre-pandemic period (2014–2019), the pandemic period (2020–2022), and the post-pandemic period (2023–2024). A Quasi-Poisson regression model was trained with data from 2014 to 2018. The residual series from the Quasi-Poisson model was then modelled using an LSTM neural network. Data from 2019 were used as the validation set to compare the predictive performance of the Quasi-Poisson model and the Quasi-Poisson-LSTM hybrid model. The final model was used to generate counterfactual predictions for 2020–2024, and relative differences between observed and predicted values were calculated with 95% confidence intervals (CIs). Key results In 2024, hospitals and THCs in Henan Province recorded 423 million health facility visits. During 2014–2019, the EAPC in per-person visits was 5.0% (95% CI 3.8–6.1) overall, with a faster increase in hospitals (6.5%, 5.8–7.3) than in THCs (2.2%, −0.5 to 5.0). Over 2014–2024, the EAPC increased to 4.4% (3.0–5.8) in THCs, whereas it decreased to 4.8% (3.8–5.9) in hospitals. The Quasi-Poisson-LSTM model had lower prediction error than the Quasi-Poisson model alone for total visits (RMSE 42.7081 vs. 51.7871; MAE 34.2020 vs. 39.8669). Counterfactual predictions showed that total visits were 14.7% (95% CI 13.2–16.2%) lower than expected in 2020, 4.8% (2.8–7.0%) lower in 2021, and 6.8% (4.3–9.3%) lower in 2022. In 2023 and 2024, total visits returned to near-expected levels, with non-significant relative differences of −1.4% (−4.4% to 1.7%) and 2.3% (−1.2% to 6.0%), respectively. However, facility-specific patterns diverged. In 2024, hospital visits remained 6.0% (3.9–9.5%) below expected levels, whereas THCs visits were 39.7% (20.8–45.0) above expected levels. Conclusion The COVID-19 pandemic had divergent effects on health facility visits in hospitals and THCs in Henan Province. Although total visit volume had returned to the expected level in the post-pandemic era, hospital-based visits remained persistently below expected levels, while THCs showed substantial excess use. These findings suggest that care-seeking patterns might have shifted structurally in the post-pandemic era.