Respiratory syncytial virus tracking using internet search engine data

Eyal Oren, Justin Frere, Eran Yom-Tov, Elad Yom-TovBMC Public Health 2018 · OpenReview

Abstract: Respiratory Syncytial Virus (RSV) is the leading cause of hospitalization in children less than 1 year of age in the United States. Internet search engine queries may provide high resolution temporal and spatial data to estimate and predict disease activity.

Introduction

Respiratory Syncytial Virus (RSV) represents a significant public health challenge, serving as the primary cause of infant hospitalizations in the U.S. Despite its impact, RSV is not a nationally notifiable disease, unlike influenza. This lack of mandatory, centralized reporting leaves health officials dependent on passive surveillance data that often lacks the spatial and temporal resolution required for timely interventions. This study investigates whether "digital epidemiology"—leveraging search engine query data—can provide a robust, real-time proxy for RSV activity.

Methods

The researchers developed a multi-stage pipeline to transform noisy search data into a predictive tool. The process began with Feature Selection, narrowing a list of 613 potential search terms (covering symptoms and related diseases) down to the 50 most highly correlated with clinical hospital discharge data from Arizona.

Using Google Trends data from 2004 to 2016, the team built linear models for five "source" states (California, Michigan, Ohio, Pennsylvania, and Texas) where reliable laboratory surveillance data from the National Respiratory and Enteric Virus Surveillance System (NREVSS) was available. To address the "cold start" problem for states lacking historical ground truth data, the study introduced a Domain Adaptation technique. They identified "reference states" (frequently Ohio) with similar search behaviors to target states and mathematically adjusted the model parameters to project RSV incidence across the remaining 45 states.

Correlation between laboratory-confirmed RSV cases (blue) and Google Trends estimates (orange) across five source states.

Key Results and Insights

The study demonstrated that internet search queries are highly effective at tracking RSV. In the five source states with clinical ground truth, the models achieved a high correlation (rho) of 0.81 to 0.84. Even in states without historical clinical data, the domain-adapted models maintained a respectable correlation of 0.62, proving the scalability of the approach.

Comparison of CDC reporting (blue) versus search-based predictions (orange) for New York and Minnesota, demonstrating the effectiveness of domain adaptation in target states.

Several key insights emerged from the data:

  • Symptom Overlap: The most predictive search terms were not limited to "RSV" but included related conditions like "flu," "pneumonia," and "bronchiolitis." This suggests that collective search behavior for general respiratory symptoms creates a robust proxy for specific viral activity.
  • Viral Migration: The model successfully captured a distinct geographical trend where RSV peaks typically begin in the Southeast (Florida) and migrate toward the Northwest. This "wave" pattern allows for regional forecasting.
  • Model Robustness: Utilizing a broad set of 50 terms proved superior to tracking the single term "RSV," as the multi-term approach is less susceptible to artificial spikes caused by sudden media interest.

Discussion and Limitations

While search data provides high-resolution insights, the authors note several limitations. The clinical "ground truth" (NREVSS) is itself a passive system that may underrepresent the true disease burden. Furthermore, search data is anonymized, preventing demographic-specific analysis (e.g., distinguishing between a parent searching for a child versus an elderly individual). Finally, as search engine algorithms and public health literacy evolve, models must be continuously updated to reflect changing online behaviors.

Future research may involve integrating genetic strain data to identify if different RSV subtypes produce unique search signatures, further refining the precision of digital surveillance.

Summary updated

Abstract

Respiratory Syncytial Virus (RSV) is the leading cause of hospitalization in children less than 1 year of age in the United States. Internet search engine queries may provide high resolution temporal and spatial data to estimate and predict disease activity.
BibTeX
@inproceedings{oren2018respiratory,
  author = {Eyal Oren and Justin Frere and Eran Yom-Tov and Elad Yom-Tov},
  title = {{Respiratory syncytial virus tracking using internet search engine data}},
  booktitle = {BMC Public Health},
  year = {2018},
  url = {https://openreview.net/forum?id=V0Hn1oymDk},
}
Figure