Beyond Single-Pathogen Forecasts: Predicting Shared Respiratory Hospital Burden from Influenza, COVID-19, and RSV

Gaurob Saha; MinJae Woo; Lior Rennert

cs.AIOct 12, 2026

Beyond Single-Pathogen Forecasts: Predicting Shared Respiratory Hospital Burden from Influenza, COVID-19, and RSV

Authors: Gaurob Saha, MinJae Woo, Lior Rennert

Organizations: Department of Public Health Sciences and Center for Public Health Modeling and Response, Clemson University, Clemson, SC 29601, USA

Abstract

Co-circulating influenza, COVID-19, and respiratory syncytial virus (RSV) create a shared hospital-admission burden, because beds, staff, testing capacity, and respiratory support are allocated to total demand rather than to one pathogen at a time. This makes respiratory forecasting a target-construction problem as well as a modeling problem: disease-specific forecasts preserve attribution, whereas hospitals also need a single planning signal for combined burden. We developed a supervised 1- to 4-week-ahead framework for weekly admissions and compared two total-burden strategies: summing independently generated pathogen-specific forecasts and forecasting the combined admission target directly. Evaluation included retrospective held-out testing and prospective operational forecasts issued before the corresponding admission weeks were observed, with performance summarized using MAE and Percent Agreement (PA; the smaller of model-generated and observed counts divided by the larger). In retrospective total-burden evaluation, the combined-burden model had PA values of 0.85, 0.78, 0.74, and 0.76 and MAE values of 80.63, 200.09, 183.97, and 156.13 from week 1 to week 4 ahead, compared with summed-forecast PA values of 0.85, 0.80, 0.69, and 0.69 and MAE values of 100.14, 200.06, 205.65, and 217.57. Prospectively, the combined-burden model had higher PA and lower MAE at all four horizons, with the largest separations at weeks 3 and 4 ahead. These results support layered respiratory forecasting: pathogen-specific forecasts remain necessary for etiologic interpretation, while combined-burden forecasts better align with hospital preparedness and shared-capacity planning.

Figures & tables

Figure 1: Weekly respiratory hospital admissions by pathogen across respiratory seasons. The figure shows the changing relative contributions of COVID-19, influenza, and RSV to the overall respiratory hospitalization burden, motivating a unified multi-pathogen forecasting framework.

Target	Horizon	Selected Features	Engineered Features
Influenza	T+1	20	242
Influenza	T+2	18	230
Influenza	T+3	19	242
Influenza	T+4	17	218
COVID-19	T+1	20	242
COVID-19	T+2	20	242

Table 1: Disease-specific selected and engineered feature counts from the model runs.

Target	Horizon	Selected Features	Engineered Features
Combined	T+1	28	280
Combined	T+2	28	280
Combined	T+3	28	280
Combined	T+4	26	256

Table 2: Combined-target selected and engineered feature counts from the model runs.

Target	Horizon	Trees	Depth	$\eta$	Subsample	Colsample	Child	Gamma	$\alpha$	$\lambda$
Influenza	T+1	414	4	0.0588	0.656	0.921	1	0.987	0.772	0.994
Influenza	T+2	537	7	0.0666	0.828	0.707	1	0.896	0.741	3.755
Influenza	T+3	383	3	0.0716	0.776	0.649	5	0.034	0.909	1.294
Influenza	T+4	256	7	0.0283	0.708	0.851	3	0.564	0.396	0.552
COVID-19	T+1	786	3	0.0432	0.949	0.751	3	0.222	0.010	0.368
COVID-19	T+2	286	3	0.0674	0.779	0.905	5	0.226	0.791	3.333

Table 3: Optuna-selected XGBoost hyperparameters by disease and forecast horizon from the model runs.

Figure 2: Training, gap, retrospective, and prospective data split used for respiratory-admission forecasting.

Disease	Horizon	Baseline XGBoost MAE	SARIMA MAE	Random Forest MAE	Adaptive XGBoost MAE	Baseline XGBoost PA	SARIMA PA	Random Forest PA	Adaptive XGBoost PA
Influenza	T+1	88.84	92.01	57.16	74.64	0.80	0.75	0.83	0.84
Influenza	T+2	190.88	209.70	142.90	189.58	0.77	0.51	0.69	0.75
Influenza	T+3	164.33	286.84	152.86	176.53	0.63	0.40	0.69	0.62
Influenza	T+4	194.53	338.12	158.58	191.29	0.58	0.28	0.67	0.62
COVID-19	T+1	18.66	26.00	18.39	20.70	0.79	0.66	0.78	0.75
COVID-19	T+2	21.53	34.48	29.53	17.92	0.76	0.57	0.71	0.80

Table 5: Retrospective results across disease-specific models.

Disease	Horizon	Baseline XGBoost MAE	SARIMA MAE	Random Forest MAE	Adaptive XGBoost MAE	Baseline XGBoost PA	SARIMA PA	Random Forest PA	Adaptive XGBoost PA
Combined	T+1	114.42	144.52	114.32	80.63	0.83	0.73	0.81	0.85
Combined	T+2	228.97	251.86	216.66	200.09	0.67	0.76	0.78	0.78
Combined	T+3	289.41	257.83	160.32	183.97	0.62	0.72	0.74	0.74
Combined	T+4	291.27	140.57	164.56	156.13	0.61	0.74	0.74	0.76

Table 6: Retrospective results across combined-burden models.

Target	Horizon	Adaptive XGBoost MAE	Adaptive XGBoost PA
Summed	T+1	100.14	0.85
Summed	T+2	200.06	0.80
Summed	T+3	205.65	0.69
Summed	T+4	217.57	0.69

Table 7: Retrospective results for summed disease-specific forecasts.

Target	Horizon	Adaptive XGBoost MAE	Adaptive XGBoost PA
Combined	T+1	80.63	0.85
Combined	T+2	200.09	0.78
Combined	T+3	183.97	0.74
Combined	T+4	156.13	0.76

Table 8: Retrospective results for the combined-burden model.

Target	Horizon	Adaptive XGBoost MAE	Adaptive XGBoost PA
Summed	T+1	40.21	0.70
Summed	T+2	42.46	0.69
Summed	T+3	64.43	0.62
Summed	T+4	43.47	0.66

Table 9: Prospective results for summed disease-specific forecasts.

Target	Horizon	Adaptive XGBoost MAE	Adaptive XGBoost PA
Combined	T+1	36.33	0.71
Combined	T+2	37.57	0.70
Combined	T+3	33.19	0.70
Combined	T+4	21.97	0.76

Table 10: Prospective results for the combined-burden model.

Epidemic Forecasting

Epidemic forecasting predicts disease incidence, outbreaks, and healthcare burden across time and regions to guide surveillance and public-health decisions.

See all 36 papers

EpiFlow: A framework for improving the utility of wastewater signals for disease forecasting
Aug 7, 2026Aniruddha Adiga, Jingyuan Chou, Gursharn Kaur +8Epidemic Forecasting Probabilistic Forecasting
Biocomplexity Institute, University of Virginia · Department of Computer Science, University of Virginia · Virginia Department of Health +3
Understanding Key Features of Time Series Foundation Models from Epidemic Forecasting
Jun 17, 2026Alireza Jafari, Judy Fox, Geoffrey C. Fox +2Epidemic Forecasting LLM-Based Time Series Forecasting
Department of Computer Science, School of Engineering and Applied Science, University of Virginia, Charlottesville, VA 22904 USA · School of Data Science, University of Virginia, Charlottesville, VA 22904 USA · Biocomplexity Institute and the Department of Computer Science, School of Engineering and Applied Science, University of Virginia, Charlottesville, VA 22904 USA +1
Generative diffusion models for spatiotemporal influenza forecasting
Apr 27, 2026Joseph Lemaitre, Justin LesslerEpidemic Forecasting Spatiotemporal Forecasting
Department of Epidemiology, Gillings School of Global Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA · Department of Epidemiology, Johns Hopkins Bloomberg School of Public Health, Baltimore, MD 21205, USA · Carolina Population Center, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599, USA

Aug 7, 2026J/K move · Enter open · S save

EpiFlow: A framework for improving the utility of wastewater signals for disease forecasting

Aniruddha Adiga, Jingyuan Chou, Gursharn Kaur, Andrew Warren+7

Biocomplexity Institute, University of Virginia · Department of Computer Science, University of Virginia · Virginia Department of Health+3

Epidemic Forecasting Probabilistic Forecasting Non-Stationary Time Series Forecasting

Open paper AskPDF arXiv

Wastewater-based surveillance is an effective tool for disease monitoring and can provide early warning of outbreaks. Although wastewater viral loads (WVL) correlate with disease burden, their utility for improving real-time forecasting remains under investigation. During the early phases of an epidemic, many indicators can effectively monitor disease spread, but their reliability may decline because of reporting fatigue and low prevalence. Hospital burden can vary substantially even during low-prevalence periods, making accurate forecasting of burden indicators essential for minimizing disease impacts. In this paper, we present principled approaches for processing wastewater data, characterizing its relationship with burden indicators, and generating real-time forecasts. We assess the predictability of WVL using entropy measures. We analyze the relationship between WVL and burden indicators using causality tests that capture temporal dynamics and the leading-indicator behavior of WVL. We incorporate these insights into a time-varying forecasting model that accounts for the evolving relationship between the signals. We also evaluate the effects of delays in WVL reporting through simulations. We test the utility of our methods by forecasting COVID-19 hospital admissions across Virginia and its health regions during periods of varying disease prevalence. Incorporating WVL improves forecast accuracy relative to baseline models, particularly during critical epidemic phases, and results in a 20 percentage point improvement in forecast coverage. Our results demonstrate that WVL signals can improve infectious disease forecasting even under conditions of low prevalence or delayed reporting.

Abstract

Epidemic Forecasting

Explore similar work

EpiFlow: A framework for improving the utility of wastewater signals for disease forecasting

Understanding Key Features of Time Series Foundation Models from Epidemic Forecasting

Generative diffusion models for spatiotemporal influenza forecasting