Shoulder injuries are a common problem in handball. One likely cause of such injuries is excessive throwing. However, it is difficult to measure the number of player throws in large cohort studies using existing methods accurately. Therefore, the purpose of this study is to develop and validate a method for identifying overhead throws using a low-cost inertial measurement unit (IMU) worn on the wrist. In a two-stage approach, we developed a threshold-based automatic identification method of overhead throws in a laboratory study using the IMU. Subsequently, we validated the suggested thresholds in a field setting by comparing throws identified by the threshold-method to throws identified by video recordings of handball practices. The best set of threshold values resulted in a per-player median sensitivity of 100% (range: 84-100%) and a median positive predictive value (PPV) of 96% (range: 86-100%) in the development study. In the validation study, the per-player median sensitivity dropped to 78% sensitivity (range: 52-91%), while the per-player median PPV dropped to 79% (range: 47-90%). The proposed method is a promising method for automatically identifying handball throws in a cheap and feasible way.

Shoulder injuries are a common problem in handball. One likely cause of such injuries is excessive throwing. However, it is difficult to measure the number of player throws in large cohort studies using existing methods accurately. Therefore, the purpose of this study is to develop and validate a method for identifying overhead throws using a lowcost inertial measurement unit (IMU) worn on the wrist. In a two-stage approach, we developed a threshold-based automatic identification method of overhead throws in a laboratory study using the IMU. Subsequently, we validated the suggested thresholds in a field setting by comparing throws identified by the threshold-method to throws identified by video recordings of handball practices. The best set of threshold values resulted in a per-player median sensitivity of 100% (range: 84-100%) and a median positive predictive value (PPV) of 96% (range: 86-100%) in the development study. In the validation study, the per-player median sensitivity dropped to 78% sensitivity (range: 52-91%), while the perplayer median PPV dropped to 79% (range: 47-90%). The proposed method is a promising method for automatically identifying handball throws in a cheap and feasible way.