Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet

Existing accelerometer-based human activity recognition (HAR) benchmark datasets that were recorded during free living suffer from non-fixed sensor placement, the usage of only one sensor, and unreliable annotations. We make two contributions in this work. First, we present the publicly available Human Activity Recognition Trondheim dataset (HARTH). Twenty-two participants were recorded for 90 to 120 min during their regular working hours using two three-axial accelerometers, attached to the thigh and lower back, and a chest-mounted camera. Experts annotated the data independently using the camera’s video signal and achieved high inter-rater agreement (Fleiss’ Kappa $=0.96$). They labeled twelve activities. The second contribution of this paper is the training of seven different baseline machine learning models for HAR on our dataset. We used a support vector machine, k-nearest neighbor, random forest, extreme gradient boost, convolutional neural network, bidirectional long short-term memory, and convolutional neural network with multi-resolution blocks. The support vector machine achieved the best results with an F1-score of $0.81$ (standard deviation: $\pm 0.18$), recall of $0.85\pm 0.13$, and precision of $0.79\pm 0.22$ in a leave-one-subject-out cross-validation. Our highly professional recordings and annotations provide a promising benchmark dataset for researchers to develop innovative machine learning approaches for precise HAR in free living.

Pooling data from thigh-worn accelerometers across multiple studies has great potential to advance evidence on the health benefits of physical activity. This requires harmonization of information on body postures, physical activity types, volumes and time patterns across different brands of devices. The aim of this study is to compare the physical behavior estimates provided by three different brands of thigh-worn accelerometers. Twenty participants volunteered for a 7-day free-living measurement. Three accelerometers - ActiGraph GT3X+, Axivity AX3 and ActivPAL Micro4 - were randomly placed in a vertical line on the midsection of the right thigh. Raw data from each accelerometer was processed and classified into 8 physical activities and postures using the Acti4 software. Absolute differences between estimates and the respective coefficient of variation (CV) were calculated. We observed very minor differences between physical behavior estimates from three different accelerometer brands. When averaged over 24 h (1,440 min), the absolute difference (CV) between accelerometers were: 1.2 mins (0.001) for lying/sitting, 3.4 mins (0.02) for standing, 3.5 mins (0.06) for moving, 1.9 mins (0.03) for walking, 0.1 mins (0.19) for running, 1.2 mins (0.19) for stair climbing, 1.9 mins (0.07) for cycling. Moreover, there was an average absolute difference of 282 steps (0.03) per 24 h. Physical behaviors were classified with negligible difference between the accelerometer brands. These results support harmonization of data from different thigh-worn accelerometers across multiple cohorts when analyzed in an identical manner.

Importance: Self-management is a key element in the care of persistent neck and low back pain. Individually tailored self-management support delivered via a smartphone app in a specialist care setting has not been tested.Objective: To determine the effect of individually tailored self-management support delivered via an artificial intelligence–based app (SELFBACK) adjunct to usual care vs usual care alone or nontailored web-based self-management support (e-Help) on musculoskeletal health.Design, Setting, and Participants: This randomized clinical trial recruited adults 18 years or older with neck and/or low back pain who had been referred to and accepted on a waiting list for specialist care at a multidisciplinary hospital outpatient clinic for back, neck, and shoulder rehabilitation. Participants were enrolled from July 9, 2020, to April 29, 2021. Of 377 patients assessed for eligibility, 76 did not complete the baseline questionnaire, and 7 did not meet the eligibility criteria (ie, did not own a smartphone, were unable to take part in exercise, or had language barriers); the remaining 294 patients were included in the study and randomized to 3 parallel groups, with follow-up of 6 months.Interventions: Participants were randomly assigned to receive app-based individually tailored self-management support in addition to usual care (app group), web-based nontailored self-management support in addition to usual care (e-Help group), or usual care alone (usual care group).Main Outcomes and Measures: The primary outcome was change in musculoskeletal health measured by the Musculoskeletal Health Questionnaire (MSK-HQ) at 3 months. Secondary outcomes included change in musculoskeletal health measured by the MSK-HQ at 6 weeks and 6 months and pain-related disability, pain intensity, pain-related cognition, and health-related quality of life at 6 weeks, 3 months, and 6 months.Results: Among 294 participants (mean [SD] age, 50.6 [14.9] years; 173 women [58.8%]), 99 were randomized to the app group, 98 to the e-Help group, and 97 to the usual care group. At 3 months, 243 participants (82.7%) had complete data on the primary outcome. In the intention-to-treat analysis at 3 months, the adjusted mean difference in MSK-HQ score between the app and usual care groups was 0.62 points (95% CI, −1.66 to 2.90 points; P = .60). The adjusted mean difference between the app and e-Help groups was 1.08 points (95% CI, −1.24 to 3.41 points; P = .36).Conclusions and Relevance: In this randomized clinical trial, individually tailored self-management support delivered via an artificial intelligence–based app adjunct to usual care was not significantly more effective in improving musculoskeletal health than usual care alone or web-based nontailored self-management support in patients with neck and/or low back pain referred to specialist care. Further research is needed to investigate the utility of implementing digitally supported self-management interventions in the specialist care setting and to identify instruments that capture changes in self-management behavior.Trial Registration: ClinicalTrials.gov Identifier: NCT04463043

Low back pain (LBP) is a leading cause of disability worldwide. Most patients with LBP encountered in primary care settings have nonspecific LBP, that is, pain with an unknown pathoanatomical cause. Self-management in the form of physical activity and strength and flexibility exercises along with patient education constitute the core components of the management of nonspecific LBP. However, the adherence to a self-management program is challenging for most patients, especially without feedback and reinforcement. Here we outline a protocol for the design and implementation of a decision support system (DSS), selfBACK, to be used by patients themselves to promote self-management of LBP. The main objective of the selfBACK project is to improve self-management of nonspecific LBP to prevent chronicity, recurrence and pain-related disability. This is achieved by utilizing computer technology to develop personalized self-management plans based on individual patient data. The decision support is conveyed to patients via a mobile phone app in the form of advice for self-management. Case-based reasoning (CBR), a technology that utilizes knowledge about previous cases along with data about the current patient case, is used to tailor the advice to the current patient, enabling a patient-centered intervention based on what has and has not been successful in previous patient cases. The data source for the CBR system comprises initial patient data collected by a Web-based questionnaire, weekly patient reports (eg, symptom progression), and a physical activity-detecting wristband. The effectiveness of the selfBACK DSS will be evaluated in a multinational, randomized controlled trial (RCT), targeting care-seeking patients with nonspecific LBP. A process evaluation will be carried out as an integral part of the RCT to document the implementation and patient experiences with selfBACK. The selfBACK project was launched in January 2016 and will run until the end of 2020. The final version of the selfBACK DSS will be completed in 2018. The RCT will commence in February 2019 with pain-related disability at 3 months as the primary outcome. The trial results will be reported according to the CONSORT statement and the extended CONSORT-EHEALTH checklist. Exploitation of the results will be ongoing throughout the project period based on a business plan developed by the selfBACK consortium. Tailored digital support has been proposed as a promising approach to improve self-management of chronic disease. However, tailoring self-management advice according to the needs, motivation, symptoms, and progress of individual patients is a challenging task. Here we outline a protocol for the design and implementation of a stand-alone DSS based on the CBR technology with the potential to improve self-management of nonspecific LBP. The selfBACK project will provide learning regarding the implementation and effectiveness of an app-based DSS for patients with nonspecific LBP. RR1-10.2196/9379.

Introduction:Accelerometer-based measurements of physical activity types are commonly used to replace self-reports. To advance the field, it is desirable that such measurements allow accurate detection of key daily physical activity types. This study aimed to evaluate the performance of a machine learning classifier for detecting sitting, standing, lying, walking, running, and cycling based on a dual versus single accelerometer setups during free-living.Methods:Twenty-two adults (mean age [SD, range] 38.7 [14.4, 25–68] years) were wearing two Axivity AX3 accelerometers positioned on the low back and thigh along with a GoPro camera positioned on the chest to record lower body movements during free-living. The labeled videos were used as ground truth for training an eXtreme Gradient Boosting classifier using window lengths of 1, 3, and 5 s. Performance of the classifier was evaluated using leave-one-out cross-validation.Results:Total recording time was ∼38 hr. Based on 5-s windowing, the overall accuracy was 96% for the dual accelerometer setup and 93% and 84% for the single thigh and back accelerometer setups, respectively. The decreased accuracy for the single accelerometer setup was due to a poor precision in detecting lying based on the thigh accelerometer recording (77%) and standing based on the back accelerometer recording (64%).Conclusion:Key daily physical activity types can be accurately detected during free-living based on dual accelerometer recording, using an eXtreme Gradient Boosting classifier. The overall accuracy decreases marginally when predictions are based on single thigh accelerometer recording, but detection of lying is poor.