Measuring Physical Activity in Older Adults through Data Donation: Consent Rates, Donation Success, & Bias

Physical activity (PA) is a key predictor of many health outcomes, especially for aging populations. The accurate measurement of PA is key to identifying determinants of health and developing appropriate interventions. To measure PA, many studies use self-reports which are usually limited to global measures and can suffer from measurement error. More fine-grained day-reconstruction methods are burdensome for respondents and prone to recall error. As an alternative, researchers are providing study participants with wearable devices that passively track PA, which reduces reactivity and recall error. However, participants’ non-compliance and high device costs are problematic. Many older adults now have smartphones that track physical activity, and individuals can share these passively collected physical activity data with researchers. We test a data donation approach among older adults. Based on legal requirements (e.g., the EU General Data Protection Regulation), data controllers (i.e., social media platforms, health apps, etc.) must provide users access to all their data in a machine-readable format. We study (1) how willing and successful older adults are to donate their PA data from different smartphone apps; (2) what drives donation of PA data at the different stages of participation; and (3) what biases arises from selective data donation? We use a privacy-preserving data donation tool integrated in a probability-based online panel of the Dutch general population to collect PA data from various health apps. 2,086 members of the LISS panel aged 50 years and older completed a web survey in early 2024. All iPhone and Android smartphone owners were asked to download passively collected PA data from their devices (Apple Health, Google Location History, or Samsung Health) and donate them via the Port platform. Out of the 2,086 survey participants, 1,889 (91%) reported owning an iPhone or Android phone compatible for data donation, 606 (29%) reported being willing to donation their PA data, 354 (17%) started the data donation, and 256 (12%) successfully provided a data package. Gender, age, educational attainment, income, and smartphone ownership and usage behavior, privacy related attitudes, and type of health app from which the data were requested correlated with behavior at the different stages of study participation. Compared to the entire sample, older adults who successfully data donors reported higher levels of self-rated health, fewer health-related limitations, fewer difficulties performing everyday activities, and more physical activity. Our study shows that data donation from smartphones as part of a probability-based web survey of older adults is a feasible alternative for the measurement of PA. Nonparticipation correlates strongly with characteristics related to smartphone ownership and comfort with device use. However, substantive bias in health and PA outcomes persist for those who donated in comparison to all survey respondents.