Background: Influenza is a major public health and medical concern. Particularly among persons with Type 2 Diabetes it imposes a high burden of morbidity, mortality and economic costs. Surveillance of influenza offers valuable information for understanding the burden of influenza in a given population and also to evaluate the impact of vaccination. Conventional surveillance methods rely on data derived from patient interactions with the health care system – i.e. clinics, pharmacies or hospitals. Recent efforts to use novel digital markers have shown promise as a surveillance method in the general population. For example, the use of Google searches, Twitter feeds, or over-the-counter pharmacy sales may offer earlier detection of outbreaks as well as insights on the symptoms from text analysis of the searches or feeds. The advent of mobile health technologies (mHealth) such as exercise trackers, digital scales or app-enabled smartphones can potentially add more individualized insights by providing person-level data at granularities up to minute level for a number of behavioral types such as walking, exercise, sleeping and diet. In this study, we explore the first effort integrating conventional medical claims with mHealth data to characterize influenza and its impact on persons with Type 2 Diabetes Mellitus.
Methods: Study design is a retrospective population based cohort of persons with Type 2 Diabetes compared to age and gender matched non-diabetic controls and a self controlled study centered around incident influenza diagnosis among persons with Type 2 Diabetes. Study setting is a large national insurance payer in the United States. Potential participant data is drawn from a nearly three year period during which participants were enrolled in the insurance plan. The analysis covers a one year period from June 1, 2016 to June 1, 2017, thereby focusing on incident outcomes during the 2016-2017 influenza season in the US. Eligible participants were adults over the age of 18, with evidence of commercial coverage during the analysis period. Participants included in the retrospective cohort and case-control pre-post analysis were required to have at least one year of continuous commercial coverage that included the 2016-2017 analysis period.
The insurance payer provides a digital wellness app that connects a variety of consumer wearable fitness devices. The wellness app works on mobile phones using the Apple iOS or the Google Android operating system.
The wearable fitness devices provide passively collected behavioral activity data in the form of steps taken and sleep, as well as heart rate data. Bivariate analysis compared influenza related medical outcomes among persons with Type 2 Diabetes to age and gender matched non-diabetic controls. P values were computed with Student’s t-test, chi-square or Mann-Whitney U tests where appropriate. False discovery rate threshold of 10% used to select significant p values.
Activity behaviors and medical outcomes were examined with a pre-post study design comparing the peri-influenza period (the 2 weeks before and 4 weeks after an incident influenza diagnosis) to the 6 week period preceding the peri-influenza period (baseline) at the individual level. We also looked at a sub-cohort of patients with dense data in the 14 days surrounding an influenza event.
Results: Medically-attended influenza events significantly affect physical activity for up to 4 days before and 7 after the event.
Shown is step activity data aggregated per day for diabetics (N=67) and controls (N=243) with medically attended influenza in the 2 weeks prior and after diagnosis. During this period, people with diabetes and medically attended influenza also missed more days of tracking (up to 20% missed days from a baseline of 5%) and this was reflected in reduced time being active, daily step counts and a slower maximum daily walking pace.
Diabetics also change sleep behavior during medically-attended influenza events: earlier bed times precede influenza, followed by more interrupted sleep and more naps. The three key results included: Sleep start-times shift by 30 minutes in advance of the medical visit; Restless sleep peaks after the medical visit, with about 2% more of the night spent restless on average; Naps also peak after the medical visit, with about 0.15 more daily naps per person.
Discussion: Influenza occurs more frequently among persons with Type 2 Diabetes Mellitus compared with controls. This suggests that diabetes increases the risk of influenza infection, and highlights the importance of vaccination in this population. No adverse effects of influenza vaccination on activity behaviors or physiologic signals were observed, confirming a lack of evidence for a common misperception that vaccination can lead to flu effects. Among mHealth users, persons with Type 2 Diabetes differ from non-diabetic trackers in a number of behaviors related to sleep, walking and exercise. Diabetics with influenza also saw more acute events in the peri-influenza period compared to baseline, such as a 75% increase in abnormal glucose events. Among persons with Type 2 Diabetes, mHealth devices detected substantial changes in walking, exercise and sleep in the immediate period before and after an influenza episode. These changes likely lead to further adverse effects on daily living and activities both in work and in personal life. Taken together, these results are the first describing the immediate effects of influenza on the daily life of people with diabetes at a population level. These findings confirm the potential of mHealth for quantifying the impact of influenza on people with diabetes. As mHealth devices become more ubiquitous, their applications for individual and population surveillance and for prevention and management of acute and chronic disease will show greater value. This needs to be assessed with further prospective studies.