Background/Purpose: The increasing uptake of smartphones and health apps provides opportunities for epidemiological studies to collect regular information from large numbers of people. This offers temporally-rich self-reported data as well as access to novel data such as geolocation. However, the extent to which participants will join and remain engaged with such studies over time is not clear. It is also unclear whether certain subgroups will remain more engaged than others, introducing possible selection bias. Cloudy with a Chance of Pain is a UK smartphone-based study investigating the link between the weather and pain in people with chronic pain. This study aimed to examine patterns of engagement and factors associated with it over the first six months.

Methods: Eligible participants in this study were recruited between 01/20/2016 and 02/29/2016. Follow-up data were available until 06/14/2016. After completing baseline characteristics, participants received daily prompts to complete a motif of 10 symptoms including pain, fatigue, and mood. Participants were categorised as ‘active’ if they reported data within the last 28 days of follow-up (05/17/2016-06/14/2016); we determined their rates of data completion prior to this period. Participants without data in the last 28 days were considered ‘inactive’. The proportion of motifs completed was defined as the number of days with complete motifs divided by the number of possible complete motifs since recruitment. We examined differences between active and inactive participants, and between active users with high ( ≥80% motifs completed) and low (<80% motifs completed) engagement.

Results: We included 7124 participants reporting at least one symptom; 1459 (20%) were considered ‘active’. Active participants were older (median (IQR): 54(45-62) vs 48 (40-58), p<0.001) and more likely to be female (83.4% vs 79.2%, p<0.001) compared to those inactive. Of the self-reported conditions, only headache and non-specific arthritis differed between active and inactive participants. Active participants completed 67% of possible motifs during the study period, compared to inactive participants who completed 8.7%. Of active users, 49% had high engagement throughout follow-up. Those with low engagement completed 43.2% of possible entries; only age was found to significantly differ between these groups (Table). Of the inactive participants, 1038 completed only one motif.

Conclusion: Following high initial recruitment, 20% of participants remained active after more than four months of data collection, half of whom provided data on more than 80% of eligible days. Engagement rates were higher in older participants. Although selection bias due to attrition needs to be considered when analysing results, these results suggest that our method is a viable and sustainable alternative to more traditional data collection.