Influenza surveillance: A Review of Data Sources
Objective
In recent years, influenza surveillance systems have started to include alternative data sources such as hospital emergency department data, school absenteeism reports, pharmaceutical sales, website access and health/medical advice calls. This scoping review aims to map the published papers on the use of non-traditional/alternative sources of syndromic surveillance data for influenza.
Study Background
The global history of pandemics in the 20th and 21st century indicates the potential of influenza viruses to undergo antigenic evolution and cause outbreaks. Globally, influenza is estimated to claim 290,000 to 650,000 lives every year. In 1947, WHO Interim Committee of the United Nations established the Global Influenza Program (GIP), which was renamed global influenza surveillance and response system (GISRS) later on. The H1N1 pandemic of 2009 tested the capacities of countries in early detection and investigation of the disease. This strengthened the need for a standardized global surveillance approach. And thus, the first Global Epidemiological Surveillance Standards for Influenza was developed. The goal of the influenza surveillance program is to monitor the associated disease burden, identify the ideal timing of the vaccination campaign and evaluate its effectiveness with immunization strategies.
Sentinel surveillance for influenza is a resource-effective approach to collecting critical epidemiological and virological data. However, it has several limitations. For example, it captures only the data of patients who seek medical attention. The delay between the onset of symptoms and seeking medical care results in a lag of 1 to 2 weeks between data collection and reporting. Third, the sentinel system depends on good laboratory settings and trained staff, which are factors that are not always feasible, especially in low-resource countries. Lastly, the sentinel approach introduces sampling bias.
To address these limitations, there has been a global quest to develop influenza surveillance systems that use alternative sources of data. In this study, the authors aimed to compile the published literature on the use of alternative sources of data and compare their timeliness and relevance with traditional data.
Methods
Study concept
The core concepts examined in this review were
- Traditional influenza surveillance systems (the ones using sentinel influenza-like illness (ILI) and severe acute respiratory infection (SARI) data).
- Non-traditional influenza surveillance systems (using alternative data such as internet search queries and electronic health records (EHR).
Literature search
The authors searched PubMed, Web of Science, and Scopus for studies in English, French, and Spanish, published between January 1, 2007, and January 28, 2022. They used the terms “influenza” AND “surveillance methods”, “correlation”, OR “timeliness.”.
Geographic setting
There was no geographic restriction. The review included studies done in both healthcare (pharmacies, hospitals, and long-term care facilities) and non-healthcare (educational institutions) settings.
Study selection
Before the systematic literature search, a list of inclusion and exclusion criteria was created. All literature pulled was imported into Covidence to remove any duplicate entries. Next, three reviewers screened each article’s title and abstract, using the inclusion/exclusion criteria to determine relevance. Any conflicts were resolved by discussion. Last, all relevant articles were moved into the Full-Text Review, where reviewers read through the full text of the articles and decided which to exclude based on the eligibility criteria. All included articles were then moved forward for data extraction.
Eligibility criteria
This review included an article published in the selected databases that i) considered percent positivity for influenza or influenza cases ii) had ILI or SARI as syndromic indicators and iii) included correlation/concurrence and timeliness as primary outcomes.
Exclusion criteria
All papers published in or before 2006 were excluded. Reviews, animal studies, nowcasting or forecasting studies, studies that considered only Google flu trends as the non-traditional system, as well as articles published in languages other than English, French, and Spanish were excluded.
Data extraction
A survey built by the review team was used to extract data. A second reviewer decided on the data accuracy and comprehensiveness. Any discrepancies in the data extraction process were discussed and resolved by a third reviewer.
RESULTS
Study inclusion
A total of 823 articles were retrieved from the database search. After removing duplicates, 785 articles were assessed for relevance. Of those, 627 were excluded due to irrelevancy, leaving 158 full texts for review. After the full-text review, only 57 were included for data extraction. However, many articles studied multiple surveillance systems, were conducted in multiple geographical regions, or performed over multiple seasons. So, this review has more than 57 articles.
Geographical regions
In total, 22 geographical areas were mentioned. The most studied country was the United States (n = 21), followed by China (n = 8), Canada (n = 5), and Spain (n=5).
Temporal distribution
The publications showed an upward trend from 2007 to 2014, with a peak in 2014, after which their numbers remained stable.
Surveillance systems by data source
Of the 57 articles included, 36 and 46 studies considered, respectively, compared the timeliness and correlative value of non-traditional surveillance systems with traditional ones. Below are the non-traditional influenza surveillance systems studied.
- EHR-based surveillance system
EHR-based studies were the most common (n = 15, 42.4%), suggesting that EHR-based surveillance systems may be both timely and correlate well with sentinel surveillance data.
- Participatory survey-based surveillance system
About 31.1% of the studies (n = 11) focused on participatory survey-based surveillance systems. Most studies identified participatory survey-based systems as being a reliable complement to the sentinel surveillance system. However, Lwin et al. reported a rather weaker correlation than in other studies between ILI symptoms reported by healthcare workers in Singapore.
Rehn et al. reported that ILI reports from a Swedish internet-based monitoring system had the strongest correlation with lab data when no time lag was used for the influenza seasons of 2011–12 and 2012–13.
- Online searches and internet traffic
10 articles addressed online searches and webpage traffic (28.3%). Yuan et al. and Dong et al. found strong positive correlations between influenza-related search terms and influenza case data. Some of these correlations were stronger even with a lag of 1, 2, or 3 months. Hulth et al. showed that influenza-related searches peaked 1 or 2 weeks ahead of the sentinel ILI. Chang et al. found strong correlations between the Google trends data and ILI data when a week’s time lag was modeled.
- Twitter-based surveillance system
Out of the 57, only 7 studies (19.8%) compared traditional sentinel surveillance to Twitter-based systems. And, the evidence was conflicting. Four studies from the USA and one from Korea found strong correlations between Twitter search and influenza data, while two other studies reported weaker correlations.
- Absenteeism-based surveillance system
Four studies (14.1%) used absenteeism data from educational settings. However, there was no concurrence in results between those studies. Fan et al. reported a peak in absenteeism 2-4 days before the ILI incidence in rural China. Duchemin et al. found that more workplace sick leave occurred several weeks before ILI outbreaks in France.
- Miscellaneous surveillance system
The miscellaneous surveillance systems (14.1%) identified include systems that use body temperature, restaurant reservations, rapid influenza detection tests (RIDT), and online health tools. They all provided a stronger signal than influenza markers and ILI data.
- Telephonic health-line-based surveillance system
Only four studies (11.3%) considered telephonic helplines as a syndromic surveillance system. And all of them lacked a clear consensus regarding timeliness and correlation.
- Medication sales-based surveillance system
Only three studies (8.5%) looked at prescription medication sales for colds and acute respiratory tract infections as a surveillance system. Of which 2 showed a moderate correlation and one a strong correlation between ILI-related drug sales and ILI outbreaks.
- Media reports
Two articles (5.7%) considered media reports and the correlation was moderate.
Discussion
This scoping review identified eight non-traditional and a miscellaneous group of non-traditional influenza surveillance systems with varying timeliness and correlation to traditional surveillance systems. Though EHR and participatory survey-based surveillance systems were found to be correlated optimally and complement the current sentinel surveillance system, there were disagreements between studies for each of the data sources studied.
The authors found that some of the surveillance systems were geography-specific. For example, 6 out of 7 Twitter-based studies were from the USA, the country with the most users. Similarly, China had the highest number of online search-based studies. Since internet usage is relatively lower in low-resource countries, the question remains whether such online-based complementary systems can provide comprehensive and minimally biased data.
Limitations
While the authors used a broad search strategy using three languages, a limitation of this review is that it is limited to studies published after 2007. Some search terms may not have been globally applicable, so the authors may have missed some relevant articles. Due to the scoping nature of the review, this study did not assess literature bias or other methodological limitations.
Oncology Related Tools
Other
Latest Research
