Introduction
 Accurate measurement of physical activity (PA) in older adults is important, both in health research and personalized prevention. Accelerometers, used to overcome the limitations of self-reporting, were initially worn on the hips, but are increasingly worn on the non-dominant wrist. While this can improve wear compliance, the accuracy of PA intensity classification can be compromised. Given the high prevalence of mild to severe hearing loss in the older population, this study explores a novel approach: integrating an accelerometer into a hearing aid (ear sensor). We aimed to assess its accuracy and compare it to research-grade sensors worn at different locations.
 Methods 60 middle-aged to older adults (64.0 ± 8.0 years, 48% women) were included in this study. Each subject performed 12-13 different activities, which were pseudo-randomly selected from a list of 33 activities of daily living. Each activity lasted 8 min and included sedentary activities (e.g., lying, playing cards) low-intensity activities (e.g., hanging laundry), activities of changing intensity or without physical displacement (e.g., yoga, squats), indoor activities related to locomotion (e.g., walking, running), outdoor activities (e.g., walking uphill, cycling), and activities with aids (e.g., walking with a stroller). Oxygen consumption was measured via indirect calorimetry and used to classify activity intensity into sedentary behavior (SB, metabolic equivalent of task [MET] < 1.5), light intensity PA (LPA, 1.5 ≤ MET < 3.0), or moderate to vigorous intensity PA (MVPA, MET ≥ 3.0). The ear sensor was placed behind the left ear, while the research-grade sensors were placed on both wrists and ankles, on the hip, chest, and forehead. Estimation of PA intensity classes was done using mean amplitude deviations and ROC analyses. Contingency tables were used to determine classification accuracy.
 Results
 Overall accuracy of the ear sensor was 82.6%, performing better than both wrists (left 81.1%, right 76.0%) and both ankles (left 81.1%, right 81.9%), but worse than the forehead (83.6%), hip (85.6%) and the chest (85.9%). ROC analyses show that all sensors can effectively discriminate between sedentary vs. non-sedentary activities (AUC 0.97-0.98, exception ankles: AUC 0.95-0.96) and between MVPA vs. other (AUC 0.96-0.97, exception wrists: AUC 0.89-0.92).
 Discussion/Conclusion
 This study is the first to show that an accelerometer integrated into a hearing aid can accurately classify PA intensity and differentiate MVPA and sedentary behavior in older adults. It also confirms previous investigations showing that wrist-worn sensors – although increasingly being used to monitor PA – are less effective in capturing MVPA compared to sensors worn closer to the center of mass (including the head/ear in our study). Although the optimal wear site in older adults is a subject of ongoing debate, our data shows that a sensor integrated into a hearing aid offers a promising balance of classification accuracy and (possibly) user compliance. Further studies should explore integrating in-ear heart rate monitoring to enhance accuracy even further.