Objectives: To compare and contrast the predictive ability of consumer health data from wearable activity monitors with conventional medical claims data for the prediction of influenza vaccinations in a large commercially insured population. We hypothesize that the inclusion of WAM data will provide additional information to predictive models beyond the use of conventional medical claims data alone.

Methods: Data source: Medical claims and enrollment records from commercial health coverage from Humana Inc, a national health and well-being company. Humana wellness app for all members that can connect WAM and acts as a portal for logging wellness behaviors like obtaining an influenza vaccination. Selection Criteria: The predictive analysis period covers a one year period from June 1, 2015 to June 1, 2016. Training data for medical claims and for WAM were obtained from the June 1, 2014 – June 1, 2015 period. Eligible participants were adults (age 18– 64), with evidence of commercial coverage during the analysis period. Influenza vaccination status was based on medical claims data and participant reporting through the wellness app. Analysis: Two predictive models were compared: 1)The first model used medical and pharmacy claims data. 2) The second model used a subset of the first cohort that also had WAM data from the wellness platform. A total of 679 features were computed for analysis. For each model, we split the dataset into training and test sets. On the training set, we used L-1 regularized Lasso regression with influenza vaccination as the outcome to identify a subset of most predictive features. We then trained a random forest classifier on the training set and report predictive power with a Receiver Operating Curve and its Area-Under- Curve (AUC) on the test set as the performance metric for the models.

Results: This study found that WAM data in combination with prior influenza vaccination status performed comparably to conventional medical and pharmacy claims data as is typically used in real world evidence. Advantages to WAM data include: 1) Following user consent, WAM is unobtrusively collected and does not require active user engagement; and 2) information can be summarized over different time granularities of day, hour and even minute levels. WAM can interact with mobile applications and cloud based analytic engines for real time predictions and analysis. This enables the delivery of mobile app based interventions that are personalized and thereby relevant to the individual. Current limitations in WAM data such as inconsistent use, device inaccuracies and a user profile that skews toward younger more affluent people will likely become less relevant as technologies improve and use becomes more commonplace within the consumer market. Predictive models using consumer mobile health data gathered by wearable activity monitors in conjunction with prior year vaccination status can be used to predict influenza vaccination behaviors in a large population. This finding has implications on the development of interactive mobile technologies to positively influence healthy behaviors at a population level

Source: SMDM 2018

Further reading

Aug 08 2019

Chan R, Jankovic F, Marinsek N, Foschini L, Kourtis L, Signorini A, Pugh M, Shen J, Yaari R, Maljkovic V, Sunga M, Hee Song H, Joon Jung H, Tseng B, Trister A

Source: KDD 2019