Lauren A. White

For infectious disease forecasting challenges, individual model performance typically varies across space and time. This phenomenon raises the question: are there properties of the target time series that contribute to a particular season, location, or disease being more difficult to forecast? Here we characterize a time series’ future predictability using a forecastability metric that calculates the spectral entropy of the time series. Forecastability of syndromic influenza hospital admissions for the state of California varied widely across seasons and was positively correlated with peak burden. Next, using archived U.S. state and national forecasts targeting laboratory-confirmed COVID-19 and influenza hospital admissions, we investigated the relationship between forecastability and: (i) population size of the forecasting target, and (ii) forecast performance as measured by mean absolute error, weighted interval score (WIS), and scaled relative WIS. Forecastability increased with increasing population size of the forecasting target, and forecasting performance generally improved with higher forecastability when mitigating the effects of population size across scales. These preliminary results support the idea that some targets and respiratory virus seasons may be inherently more difficult to forecast and could help explain inter-seasonal variation in model performance.

During the COVID-19 pandemic, US federal hospitalization surveillance systems provided a new data source for lab-confirmed influenza hospital admissions; these data were important for monitoring hospital capacity for combined respiratory virus pathogen burden and served as a new forecasting target for seasonal influenza forecasting efforts. For influenza surveillance in the state of California, from February 2, 2022 through May 31, 2025, we explored: (1) how lab-confirmed hospital admissions correlated with other syndromic and sentinel surveillance data; (2) which signals could serve as potential leading signals relative to hospital admissions; and (3) how these relationships varied across respiratory virus seasons and geography. Despite varying across seasons and regions, influenza surveillance data sources in California were strongly correlated with laboratory-confirmed influenza hospitalizations (Spearman’s P ≥ 0.8). These correlations generally strengthened through time from the 2021-2022 season to the most recent 2024-2025 season, especially for death, influenza-like illness, wastewater, and clinical lab data. Electronic laboratory reporting was the only data source to provide consistent leads of 1–2 weeks relative to the admissions signal across all four seasons (Spearman’s P ≥ 0.89). Understanding the relationship between different data sources will improve situational awareness and help inform future estimates of influenza burden.

Background: During the COVID-19 pandemic, U.S. federal hospitalization surveillance systems provided a new data source for lab-confirmed influenza hospital admissions; these data were important for monitoring hospital capacity for combined respiratory virus pathogen burden and served as a new forecasting target for seasonal influenza forecasting efforts. Methods: For influenza surveillance in the state of California, from February 2, 2022 through May 31, 2025, we explored: (1) how lab-confirmed hospital admissions correlated with other syndromic and sentinel surveillance data; (2) which signals could serve as a potential leading input for predicting hospital admissions; and (3) how these relationships varied across respiratory virus seasons and geography. Results: Despite varying across seasons and regions, influenza surveillance data sources in California were strongly correlated with laboratory-confirmed influenza hospitalizations (Spearman’s ρ ≥ 0.8). These correlations generally strengthened through time from the 2021-2022 season to the most recent 2024-2025 season, especially for death, influenza-like illness, wastewater, and clinical lab data. Most of these data sources neither consistently led nor lagged hospital admissions across all four seasons, but electronic laboratory reporting provided consistent leads of one to two weeks relative to the admissions signal across all four seasons (Spearman’s ρ ≥ 0.89). Principal component analysis suggests that 93% of the variation in data signals can be explained by a single axis. Conclusions: Influenza surveillance data sources have inherent trade-offs in geographic coverage, temporal resolution, and reporting frequency. Understanding the relationship between different data sources will inform future predictions of influenza burden, including forecasting and scenario modeling. Key Messages: Influenza surveillance data sources in California were strongly correlated with new laboratory- confirmed influenza hospitalizations data from the National Hospital Safety Network (NHSN), and these correlations strengthened through time from the 2021-2022 to the most recent 2024- 2025 season. Although most data sources did not provide a clear leading or lagging signal, electronic laboratory reporting results provided consistent leads of one to two weeks relative to laboratory-confirmed influenza hospital admissions. This analysis demonstrates that existing influenza surveillance and sentinel data sources correspond well with the new NHSN data, which will be important for ongoing surveillance and public health action during future respiratory virus seasons.

Background: Infectious disease surveillance is an essential component of public health for preventing and mitigating outbreaks. Systematically applying statistical methods for anomaly detection to surveillance data can expedite outbreak response through early warning. A commonly used approach is the usage of spatiotemporal scan statistics as implemented in SaTScan, a software that analyzes spatiotemporal data to identify clusters of events over space and time that deviate from expected values. Some health departments identify outbreaks and prioritize resources using SaTScan for early cluster detection for diseases such as salmonellosis, legionellosis, and COVID-19. However, as a standalone software, SaTScan v10.2.1 does not provide functionality to easily disseminate visual cluster results over time in a way that is tailored to epidemiologists’ needs for real-time disease surveillance. Results: We developed an open source dashboard that provides a customizable framework for displaying results and facilitating the use of SaTScan for public health surveillance. The California Clustering for Operational Real-time Visualization of Infectious Diseases (CalCORVID) dashboard is built using RShiny, is specifically designed for SaTScan outputs, and can be easily adapted to display any jurisdiction’s results. This dashboard features a map and corresponding results table, the option to view historical results, integration of the Social Vulnerability Index (SVI) to contextualize clusters, and interactive elements to enhance usability for epidemiologists. Conclusions: We present CalCORVID as a complementary tool to native outputs of SaTScan v10.2.1, allowing users to visualize, customize, and distribute their results for specific public health use cases. Epidemiologists currently using SaTScan can adapt the provided code repository and dashboard template to display their own jurisdictions’ results, facilitating dissemination of cluster results for real-time, ongoing disease surveillance.

For infectious disease forecasting challenges, individual model performance typically varies across space and time. This phenomenon raises the question: are there properties of the target time series that contribute to a particular season, location, or disease being more difficult to forecast? Here we characterize a time series’ future predictability using a forecastability metric that calculates the spectral density of the time series. Forecastability of syndromic influenza hospital admissions for the state of California varied widely across seasons and was positively correlated with peak burden. Next, using archived U.S. state and national forecasts targeting laboratory-confirmed COVID-19 and influenza hospital admissions, we investigated the relationship between forecastability and: (i) population size of the forecasting target, and (ii) forecast performance as measured by mean absolute error, weighted interval score (WIS), and scaled relative WIS. Forecastability increased with increasing population size of the forecasting target, and forecasting performance generally improved with higher forecastability when controlling for population size across scales. These preliminary results support the idea that some targets and respiratory virus seasons may be inherently more difficult to forecast and could help explain inter-seasonal variation in model performance. Author summary: Could intrinsic properties of an epidemiological time series help explain why a particular season, location, or disease is more difficult to predict in the future? To answer this question, this analysis uses a measure of a time series’ future predictability called “forecastability,” which describes the inherent uncertainty or surprise in the signal. Influenza and COVID-19 hospital admissions had higher forecastability scores for locations with larger population sizes, possibly due to larger counts leading to smoother time series. At the same time, forecasting performance generally improved for time series with higher forecastability scores when controlling for population size, suggesting that this metric is helpful for understanding ease of forecasting. These preliminary results support the idea that some epidemiological targets and respiratory virus seasons may be inherently more difficult to forecast and could help explain why forecasting model performance changes across different respiratory virus seasons.

Individuals infected with SARS-CoV-2 may develop post-acute sequelae of COVID-19 (“long COVID”) even after asymptomatic or mild acute illness. Including time varying COVID symptom severity can provide more informative burden estimates for public health response. Using a compartmental model driven by confirmed cases, this study estimated long COVID burden by age group (0–4, 5–17, 18–49, 50–64, 65+) in California as measured by the cumulative and severity-specific proportion of quality-adjusted life years (QALYs) lost. Long COVID symptoms were grouped into severe, moderate, and mild categories based on estimates from the Global Burden of Disease study, and symptoms were assumed to decrease in severity in the model before full recovery. All 10,945,079 confirmed COVID-19 cases reported to the California Department of Public Health between March 1, 2020, and December 31, 2022, were included in the analysis. Most estimated long COVID-specific QALYs [59,514 (range: 10,372–180,257)] lost in California were concentrated in adults 18–49 (31,592; 53.1%). Relative to other age groups, older adults (65+) lost proportionally more QALYs from severe long COVID (1,366/6,984; 20%). Due to changing case ascertainment over time, this analysis might underestimate the actual total burden. In global sensitivity analysis, estimates of QALYs lost were most sensitive to the proportion of individuals that developed long COVID and proportion of cases with each initial level of long COVID symptom severity (mild/moderate/severe). Models like this analysis can help translate observable metrics such as cases and hospitalizations into quantitative estimates of long COVID burden that are currently difficult to directly measure. Unlike the observed relationship between age and incident severe outcomes for COVID-19, this study points to the potential cumulative impact of mild long COVID symptoms in younger individuals.

Accurate forecasts can enable more effective public health responses during seasonal influenza epidemics. For the 2021–22 and 2022–23 influenza seasons, 26 forecasting teams provided national and jurisdiction-specific probabilistic predictions of weekly confirmed influenza hospital admissions for one-to-four weeks ahead. Forecast skill is evaluated using the Weighted Interval Score (WIS), relative WIS, and coverage. Six out of 23 models outperform the baseline model across forecast weeks and locations in 2021–22 and 12 out of 18 models in 2022–23. Averaging across all forecast targets, the FluSight ensemble is the 2^nd most accurate model measured by WIS in 2021–22 and the 5^th most accurate in the 2022–23 season. Forecast skill and 95% coverage for the FluSight ensemble and most component models degrade over longer forecast horizons. In this work we demonstrate that while the FluSight ensemble was a robust predictor, even ensembles face challenges during periods of rapid change.

Background: The effective reproduction number ( R _e ) serves as a metric of population-wide, time-varying disease spread. During the COVID-19 pandemic, R _e was primarily estimated from clinical surveillance data streams ( R _cc ), which have varied in quality and representativeness due to changes in testing volume, test-seeking behavior, and resource constraints. Deriving R _e from alternative data sources such as wastewater could inform future public health responses. Objectives: We estimated county-aggregated, sewershed-restricted wastewater-based SARS-CoV-2 R _e ( R _ww ) from May 1, 2022 to April 30, 2023 for five counties in California of varying population sizes, clinical testing rates, demographics, proportions surveilled by wastewater, and sampling frequencies to validate the reliability of R _ww as a real-time disease surveillance metric. Methods: We produced both instantaneous and cohort sewershed-restricted R _e using smoothed and deconvolved wastewater concentrations. We then population-weighted and aggregated these sewershed-level estimates to arrive at county-level R _e . Using mean absolute error (MAE), Spearman’s rank correlation (ρ), confusion matrix classification, and cross-correlation analyses, we compared the timing and trajectory of two R _ww models to: (1) a publicly available, county-level ensemble of R _cc estimates, and (2) a county-aggregated, sewershed-restricted R _cc . Results: Both R _ww models demonstrated high concordance with traditional R _cc estimates, as indicated by low mean absolute errors (MAE ≤ 0.09), significant positive Spearman correlation (Spearman ρ ≥ 0.66, p < 0.001), and high confusion matrix classification accuracy (≥ 0.81). The relative timings of R _ww and R _cc were less clear, with cross-correlation analyses suggesting strong associations for a wide range of temporal lags that varied by county and R _ww model type. Discussion: This R _e estimation methodology provides a generalizable, robust, and operationalizable framework for estimating county-level R _ww . Our results support the additional use of R _ww as an epidemiological tool for surveillance. Based on this research, we produced publicly available R _ww nowcasts for the California Communicable diseases Assessment Tool ( https://calcat.covid19.ca.gov/cacovidmodels/ ).

Accurate forecasts can enable more effective public health responses during seasonal influenza epidemics. Forecasting teams were asked to provide national and jurisdiction-specific probabilistic predictions of weekly confirmed influenza hospital admissions for one through four weeks ahead for the 2021-22 and 2022-23 influenza seasons. Across both seasons, 26 teams submitted forecasts, with the submitting teams varying between seasons. Forecast skill was evaluated using the Weighted Interval Score (WIS), relative WIS, and coverage. Six out of 23 models outperformed the baseline model across forecast weeks and locations in 2021-22 and 12 out of 18 models in 2022-23. Averaging across all forecast targets, the FluSight ensemble was the 2 ^nd most accurate model measured by WIS in 2021-22 and the 5 ^th most accurate in the 2022-23 season. Forecast skill and 95% coverage for the FluSight ensemble and most component models degraded over longer forecast horizons and during periods of rapid change. Current influenza forecasting efforts help inform situational awareness, but research is needed to address limitations, including decreased performance during periods of changing epidemic dynamics.