Estimating cumulative incidence from partially missing time series of respiratory viral infections

Background Seasonal respiratory viruses, including respiratory syncytial virus (RSV) and human metapneumovirus (hMPV), are major contributors to global respiratory infection burden. Estimating viral incidence using real-world data (RWD) is challenging as misalignment of data collection with seasonal circulation can lead to underestimated incidence and hinder inter-study comparisons. Methods We developed a regression-based approach to retrospectively correct partial-season time series with contiguous missing weeks at the start or end of a season. Weekly RSV clinical incidence (per 100,000) was extracted from U.S. RWD sources (Optum ® CDM and PharMetrics Plus, adults ≥65 years; Kaiser Permanente Northwest [KPNW], all ages) for seasons 2018–19 to 2023–24 and aligned to RSV‑Net hospitalization surveillance. Weekly early- and late-season missingness scenarios were created (approximately 10%, 15%, 20%, and 25% of weeks removed). Models were fit within each season using harmonic terms and the surveillance-aligned covariate. Uncertainty was quantified using moving block bootstrap 95% prediction intervals. Weekly accuracy was evaluated using RMSE and normalized RMSE, and seasonal burden reconstruction accuracy using cumulative percent error and signed difference. For hMPV, we evaluated an illustrative extension using Optum ® CDM (adults ≥65 years; 2022–23 and 2023–24) and NREVSS test positivity; because positivity is bounded and depends on testing volume, we used a denominator-aware binomial/logit model (positives out of tests). Results For RSV, weekly prediction error increased with missingness level, while reconstructed seasonal burden generally remained close to observed totals in applicable scenarios. Early-season gaps were associated with larger cumulative errors than late-season gaps in Optum ® CDM and PharMetrics Plus, whereas late-season median percent errors were closer to zero across missingness levels. Applicability was reduced primarily for higher early-season missingness in KPNW due to peak overlap (e.g., early 25% scenarios not applicable). For hMPV test positivity (illustrative; two seasons), weekly prediction RMSE was on the order of ~1–2 percentage points across missingness scenarios. Conclusion Across multiple RSV seasons and data sources, reconstructed seasonal burden generally agreed with observed totals when missingness was confined to contiguous early- or late-season weeks and did not overlap peak activity. This approach provides a transparent method to improve the utility and comparability of partial-season respiratory virus time series. However, applicability is reduced when peak weeks are missing, and hMPV findings are illustrative given limited seasons and the use of test positivity.