Patterns and predictors of variability in patient-generated daily pain severity collected via a mobile health...

Digital-health technologies support the collection of patient-generated health data that is frequent, longitudinal, and collected in participant's own environments. Such high-frequency data could detect patterns of variation in disease and associated symptoms, but characterizing, interpreting, and understanding the reasons for this variability remain open questions. Here, we examine 2070 people living with chronic pain to quantify daily variability in pain severity across seven-day periods and identify factors associated with that variability. Data were collected via a smartphone application from a population-based mobile-health study, Cloudy with a Chance of Pain. Summary statistics and distributions of pain changes on consecutive days were calculated within 13,052 complete weeks of data, which had been assigned to one of four clusters via a previously published k-mediods clustering algorithm: no/low pain, mild pain, moderate pain, and severe pain. Cumulative-probit models were used to identify associations between changes in pain severity and changes in exposure data. Across the four clusters, the no/low-pain cluster had the highest proportion of weeks with no within-week changes (59%) in pain severity compared to the other clusters (48-53%). When pain did change, it changed one unit (out of five) about 20% of the time, but larger changes of two to four units also occurred. Changes in pain severity were associated most strongly with changes in pain interference (i.e., how pain impacts daily activities) and were also associated with changes in fatigue, morning stiffness, mood, and participant well-being. Thus, this study showed that data collected frequently through digital-health technologies can be used to explore variability in symptoms and their associations with other variables. That pain severity was associated with changes in modifiable variables (e.g., fatigue, mood) suggests opportunities for different treatment and self-management regimes for different patient subtypes within the four clusters.