Google Searches and Detection of Conjunctivitis Epidemics Worldwide

Abstract

Purpose: Epidemic and seasonal infectious conjunctivitis outbreaks can impact education, workforce, and economy adversely. Yet conjunctivitis typically is not a reportable disease, potentially delaying mitigating intervention. Our study objective was to determine if conjunctivitis epidemics could be identified using Google Trends search data.

Design: Search data for conjunctivitis-related and control search terms from 5 years and countries worldwide were obtained. Country and term were masked. Temporal scan statistics were applied to identify candidate epidemics. Candidates then were assessed for geotemporal concordance with an a priori defined collection of known reported conjunctivitis outbreaks, as a measure of sensitivity.

Participants: Populations by country that searched Google's search engine using our study terms.

Main outcome measures: Percent of known conjunctivitis outbreaks also found in the same country and period by our candidate epidemics, identified from conjunctivitis-related searches.

Results: We identified 135 candidate conjunctivitis epidemic periods from 77 countries. Compared with our a priori defined collection of known reported outbreaks, candidate conjunctivitis epidemics identified 18 of 26 (69% sensitivity) of the reported country-wide or island nationwide outbreaks, or both; 9 of 20 (45% sensitivity) of the reported region or district-wide outbreaks, or both; but far fewer nosocomial and reported smaller outbreaks. Similar overall and individual sensitivity, as well as specificity, were found on a country-level basis. We also found that 83% of our candidate epidemics had start dates before (of those, 20% were more than 12 weeks before) their concurrent reported outbreak's report issuance date. Permutation tests provided evidence that on average, conjunctivitis candidate epidemics occurred geotemporally closer to outbreak reports than chance alone suggests (P < 0.001) unlike control term candidates (P = 0.40).

Conclusions: Conjunctivitis outbreaks can be detected using temporal scan analysis of Google search data alone, with more than 80% detected before an outbreak report's issuance date, some as early as the reported outbreak's start date. Future approaches using data from smaller regions, social media, and more search terms may improve sensitivity further and cross-validate detected candidates, allowing identification of candidate conjunctivitis epidemics from Internet search data potentially to complementarily benefit traditional reporting and detection systems to improve epidemic awareness.

Figure 1.. Five illustrative countries demonstrating daily search data, candidate epidemics identified from that data, and reported outbreaks

For each country (column), the timespan provided is from the earliest to latest occurring candidate conjunctivitis epidemics or reported epidemics, within the full study period (i.e. first four countries shown had only a single continuum period containing any candidates or reports). The center of each point represents the start date for candidate epidemics and the issuance date for reports. Search terms not shown if all five countries had no available daily relative search interest values. (Legend: Y- axis for each time series is normalized (% of max value) daily search values. Daily values are indicated by colored vertical bars; ProMED, PubMed, and Other Online Reports by large gold inverted triangles. Conjunctivitis candidates identified from conjunctivitis-related search terms are shown by red triangles).

Figure 2.. Time series of reported outbreaks compared to detected conjunctivitis candidate epidemic dates.

For each country, conjunctivitis candidate epidemics (red-filled triangles) are plotted based on their start dates, and any reported outbreaks (gold-filled inverted triangles) for that country are plotted based on the reported start date of the report. The center of each point represents the actual dates. Each new continuum period within a country corresponds to a different triangle border color for the outbreak reports and candidate epidemics; triangles with identical border color represent reports and/or candidates occurring within the same continuum period. For all reported outbreaks that had an issuance (publication or first online) date that was one or more weeks after that report’s reported start date, a dotted black line leads to a vertical black line indicating the report’s issuance date. Note: some reported start dates (used to identify continuum ID periods and compare to candidates) were much earlier than when the report was actually issued (e.g. see Réunion, Tonga). Countries with no reported outbreaks, not shown. (Legend: Gold inverted triangles represent issuance date of ProMED, PubMed and Other Online Reports; red triangles represent candidate conjunctivitis epidemics identified in this study from Google search term data. Border colors represent unique 45-day continuums in a country’s time series. Minor breaks: 1 month).