Background/Purpose

Rheumatoid arthritis (RA) and other rheumatic diseases (RD) are associated with depression, fatigue, and disturbed sleep, symptoms that often impact behavior. Many smartphone apps allow patients with RDs to regularly report their disease activity for better self-management and clinical followup. Recent advances in reality mining technology combined with the growing use of smartphones have shown measurable changes in phone behavior due to health issues like depression, stress, and influenza. We sought to learn if there are associations of phone behavior with RD patient reported outcomes.

Methods

We invited 700 patients in the National Data Bank for Rheumatic Diseases to participate by installing a custom app on their smartphone and answering questions regularly: a daily pain VAS for 60 days and a weekly Patient Activity Scale-II (PAS-II) for 6 months. Passive data collected included mobility distance, number of unique calls and text messages, call durations, call counts, and number of missed calls. A principal component analysis (PCA) based on the correlation data was performed on the passive data. The scree plot and the Kaiser criterion were used to select the number of components. A hierarchical cluster analysis was also performed using Euclidean dissimilarity metric and the average linkage criterion, with Calinski/Haralasz rule for the optimal choice of number of clusters. In addition, GEE models examined the association of passive data with weekly PAS-II components. Possible confounders included age, sex, disease, education, income, employment, smoking, and marital status in addition to variables such as season, holiday weekend, and time of day. QIC criterion was used to select the best models.

Results

While 150 patients participated, we limited our analysis to the 55 with Android phones due to more extensive passive data. Three components were extracted, explaining a total variation of 81%. The first was an overall measure of the use of the phone (50% explained variance); the second was a contrast between the use of text messages vs. calls (20%); and a third was a function of mobility and radius. The cluster analysis found 2 groups with a larger cluster (N=42) mostly of RA patients, with a lower overall use of the phone, lower time on the phone, and lower radius, and a second cluster having a greater overall use of the phone, higher radius, and having mostly OA and Fibromyalgia.  The use of texting compared to calls was always preferred in both groups. Multivariable models suggested that patients with worse disease activity tended to answer less calls and talk for less time, but compensated with more text messages. They also tended to have greater mobility but less overall travel radius. Holidays and the summer were comparatively better for their diseases. Worse-off patients also left the study sooner.

Conclusion

With our exploratory analysis, we were able to characterize phone behavior in 3 components and 2 profiles that well-distinguished RD diagnosis. Our longitudinal models showed significant association of phone behavior with pain and PAS-II scores over time. This pilot study holds promise for passive behavior to be used in patient self-management and clinical followup.

---

« Back to 2014 ACR/ARHP Annual Meeting

ACR Meeting Abstracts - https://acrabstracts.org/abstract/use-of-smartphones-in-collecting-patient-reported-outcomes-can-passively-collected-behavior-determine-rheumatic-disease-activity-early-results-from-a-nation-wide-pilot-study/