Integrating Individual-Level Syndromic Surveillance Data Into Infectious Disease Modeling for COVID-19 Forecasting in Sweden: Comparative Modeling Study

The COVID-19 pandemic has triggered major global challenges, highlighting the need for timely surveillance and accurate forecasting to curb disease transmission and reduce the health burden. Traditional modeling approaches usually rely on aggregated data on confirmed cases. However, testing data can be unreliable due to low availability and uptake, as well as registration delays. In Sweden, the 1177 health care helpline provides valuable syndromic surveillance data that can capture early signals of infection prior to confirmation through testing.

This study aims to evaluate the effect of integrating syndromic surveillance data on short-term COVID-19 forecast accuracy.

We developed a novel epidemiological model named SI_sI_cHRD (susceptible-infected_suspected-infected_confirmed -hospitalized-recovered-deceased), incorporating data from the 1177 health care helpline cross-linked with national registers on COVID-19 testing, hospitalizations, and deaths. In the novel model, the population with infection is divided into suspected infections (I_s; those reporting symptoms to 1177) and confirmed infections (I_c; those who tested positive). Accuracy metrics of this model are compared to a model without data from the 1177 health care helpline. To evaluate model performance across different pandemic phases between March 2 and December 31, 2020, we used a daily rolling-window approach with a 60-day fitting period to predict confirmed infections, hospitalizations, and deaths over the subsequent 30 days. Prediction performance was evaluated using mean absolute error and mean absolute percentage error (MAPE).

The proposed SI_sI_cHRD model successfully captured the temporal dynamics of confirmed, hospitalized, and death cases. The model achieved a median mean MAPE of 61.6% (IQR 37.6%-79.6%) for confirmed cases, 37.5% (IQR 18.9%-57.5%) for hospitalizations and for deaths, the model achieved a median MAPE of 0.65% (IQR 0.3%-1.9%), indicating substantially higher predictive accuracy for severe outcomes compared to confirmed infections and hospital admissions. Compared to the model without syndromic surveillance data, short-term forecasting performance of the SI_sI_cHRD model reduced MAPE by an average of 15.7 percentage points for confirmed cases, 8.17 percentage points for hospitalizations, and 1.99 percentage points for deaths. Overall, the results demonstrate predictive stability for severe outcomes during the early phase of the pandemic.

The proposed model demonstrated predictive performance by integrating individual-level data with health care helpline syndromic data on testing and severe disease outcomes, capturing early signals from symptomatic individuals and distinguishing suspected cases from subsequently confirmed infections, leading to better forecasts of severe outcomes such as hospitalizations and deaths. This framework can support public health decision-making in future outbreaks.