Personalized Physical Activity Coaching: A Machine Learning Approach

Approaches for the use of Artificial Intelligence in workplace health promotion and prevention: A systematic scoping review

BACKGROUND

Artificial intelligence (AI) is an umbrella term for various algorithms and rapidly emerging technologies with huge potential for workplace health promotion and prevention (WHPP). WHPP interventions aim to improve people's health and well-being through behavioral and organizational measures or by minimizing the burden of workplace-related diseases and associated risk factors. While AI has been the focus of research in other health-related fields, such as Public Health or biomedicine, the transition of AI into WHPP research has yet to be systematically investigated.

OBJECTIVE

The systematic scoping review aims to comprehensively assess an overview of the current use of AI in WHPP. The results will be then used to point to future research directions. The following research questions were derived: (1) what are the study characteristics of studies on AI algorithms and technologies in the context of WHPP, (2) what specific WHPP fields (prevention, behavioral, and organizational approaches) were addressed by the AI algorithms and technologies, and (3) what kind of interventions lead to which outcomes?

METHODS

A systematic scoping literature review (PRISMA-ScR) was conducted in the three academic databases PubMed, IEEE, and ACM in July 2023, searching for articles published between January 2000 and December 2023. Studies needed to be 1) peer-reviewed, 2) written in English, and 3) focused on any AI-based algorithm or technology that (4) were conducted in the context of WHPP or (5) an associated field. Information on study design, AI algorithms and technologies, WHPP fields, and the PICO framework were extracted blindly with Rayyan and summarized.

RESULTS

A total of ten studies were included. Risk prevention and modeling were the most identified WHPP fields (n=6), followed by behavioral health promotion (n=4) and organizational health promotion (n=1). Four studies focused on mental health. Most AI algorithms were machine learning-based, and three studies used combined deep learning algorithms. AI algorithms and technologies were primarily implemented in smartphone applications (eg, in the form of a Chatbot) or used the smartphone as a data source (eg, GPS). Behavioral approaches ranged from 8 to 12 weeks and were compared to control groups. Three studies evaluated the robustness and accuracy of an AI model or framework.

CONCLUSIONS

Although AI has caught increasing attention in health-related research, the review reveals that AI in WHPP is marginally investigated. Our results indicate that AI is promising for individualization and risk prediction in WHPP, but current research does not cover the scope of WHPP. Beyond that, future research will profit from an extended range of research in all fields of WHPP, longitudinal data, and reporting guidelines.

CLINICALTRIAL

Registered on 5th July 2023 at Open Science Framework [1].

A Data-Driven Approach to Predicting Recreational Activity Participation Using Machine Learning

Purpose: With the popularity of recreational activities, the study aimed to develop prediction models for recreational activity participation and explore the key factors affecting participation in recreational activities. Methods: A total of 12,712 participants, excluding individuals under 20, were selected from the National Health and Nutrition Examination Survey (NHANES) from 2011 to 2018. The mean age of the sample was 46.86 years (±16.97), with a gender distribution of 6,721 males and 5,991 females. The variables included demographic, physical-related variables, and lifestyle variables. This study developed 42 prediction models using six machine learning methods, including logistic regression, Support Vector Machine (SVM), decision tree, random forest, eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM). The relative importance of each variable was evaluated by permutation feature importance. Results: The results illustrated that the LightGBM was the most effective algorithm for predicting recreational activity participation (accuracy: .838, precision: .783, recall: .967, F1-score: .865, AUC: .826). In particular, prediction performance increased when the demographic and lifestyle datasets were used together. Next, as the result of the permutation feature importance based on the top models, education level and moderate-vigorous physical activity (MVPA) were found to be essential variables. Conclusion: These findings demonstrated the potential of a data-driven approach utilizing machine learning in a recreational discipline. Furthermore, this study interpreted the prediction model through feature importance analysis to overcome the limitation of machine learning interpretability.

AI and semantic ontology for personalized activity eCoaching in healthy lifestyle recommendations: a meta-heuristic approach

BACKGROUND

Automated coaches (eCoach) can help people lead a healthy lifestyle (e.g., reduction of sedentary bouts) with continuous health status monitoring and personalized recommendation generation with artificial intelligence (AI). Semantic ontology can play a crucial role in knowledge representation, data integration, and information retrieval.

METHODS

This study proposes a semantic ontology model to annotate the AI predictions, forecasting outcomes, and personal preferences to conceptualize a personalized recommendation generation model with a hybrid approach. This study considers a mixed activity projection method that takes individual activity insights from the univariate time-series prediction and ensemble multi-class classification approaches. We have introduced a way to improve the prediction result with a residual error minimization (REM) technique and make it meaningful in recommendation presentation with a Naïve-based interval prediction approach. We have integrated the activity prediction results in an ontology for semantic interpretation. A SPARQL query protocol and RDF Query Language (SPARQL) have generated personalized recommendations in an understandable format. Moreover, we have evaluated the performance of the time-series prediction and classification models against standard metrics on both imbalanced and balanced public PMData and private MOX2-5 activity datasets. We have used Adaptive Synthetic (ADASYN) to generate synthetic data from the minority classes to avoid bias. The activity datasets were collected from healthy adults (n = 16 for public datasets; n = 15 for private datasets). The standard ensemble algorithms have been used to investigate the possibility of classifying daily physical activity levels into the following activity classes: sedentary (0), low active (1), active (2), highly active (3), and rigorous active (4). The daily step count, low physical activity (LPA), medium physical activity (MPA), and vigorous physical activity (VPA) serve as input for the classification models. Subsequently, we re-verify the classifiers on the private MOX2-5 dataset. The performance of the ontology has been assessed with reasoning and SPARQL query execution time. Additionally, we have verified our ontology for effective recommendation generation.

RESULTS

We have tested several standard AI algorithms and selected the best-performing model with optimized configuration for our use case by empirical testing. We have found that the autoregression model with the REM method outperforms the autoregression model without the REM method for both datasets. Gradient Boost (GB) classifier outperforms other classifiers with a mean accuracy score of 98.00%, and 99.00% for imbalanced PMData and MOX2-5 datasets, respectively, and 98.30%, and 99.80% for balanced PMData and MOX2-5 datasets, respectively. Hermit reasoner performs better than other ontology reasoners under defined settings. Our proposed algorithm shows a direction to combine the AI prediction forecasting results in an ontology to generate personalized activity recommendations in eCoaching.

CONCLUSION

The proposed method combining step-prediction, activity-level classification techniques, and personal preference information with semantic rules is an asset for generating personalized recommendations.

A theory-based and data-driven approach to promoting physical activity through message-based interventions

Objective

We investigated how physical activity can be effectively promoted with a message-based intervention, by combining the explanatory power of theory-based structural equation modeling with the predictive power of data-driven artificial intelligence.

Methods

A sample of 564 participants took part in a two-week message intervention via a mobile app. We measured participants' regulatory focus, attitude, perceived behavioral control, social norm, and intention to engage in physical activity. We then randomly assigned participants to four message conditions (gain, non-loss, non-gain, loss). After the intervention ended, we measured emotions triggered by the messages, involvement, deep processing, and any change in intention to engage in physical activity.

Results

Data analysis confirmed the soundness of our theory-based structural equation model (SEM) and how the emotions triggered by the messages mediated the influence of regulatory focus on involvement, deep processing of the messages, and intention. We then developed a Dynamic Bayesian Network (DBN) that incorporated the SEM model and the message frame intervention as a structural backbone to obtain the best combination of in-sample explanatory power and out-of-sample predictive power. Using a Deep Reinforcement Learning (DRL) approach, we then developed an automated, fast-profiling strategy to quickly select the best message strategy, based on the characteristics of each potential respondent. Finally, the fast-profiling method was integrated into an AI-based chatbot.

Conclusion

Combining the explanatory power of theory-driven structural equation modeling with the predictive power of data-driven artificial intelligence is a promising strategy to effectively promote physical activity with message-based interventions.

An automatic and personalized recommendation modelling in activity eCoaching with deep learning and ontology

Electronic coaching (eCoach) facilitates goal-focused development for individuals to optimize certain human behavior. However, the automatic generation of personalized recommendations in eCoaching remains a challenging task. This research paper introduces a novel approach that combines deep learning and semantic ontologies to generate hybrid and personalized recommendations by considering "Physical Activity" as a case study. To achieve this, we employ three methods: time-series forecasting, time-series physical activity level classification, and statistical metrics for data processing. Additionally, we utilize a naïve-based probabilistic interval prediction technique with the residual standard deviation used to make point predictions meaningful in the recommendation presentation. The processed results are integrated into activity datasets using an ontology called OntoeCoach, which facilitates semantic representation and reasoning. To generate personalized recommendations in an understandable format, we implement the SPARQL Protocol and RDF Query Language (SPARQL). We evaluate the performance of standard time-series forecasting algorithms [such as 1D Convolutional Neural Network Model (CNN1D), autoregression, Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU)] and classifiers [including Multilayer Perceptron (MLP), Rocket, MiniRocket, and MiniRocketVoting] using state-of-the-art metrics. We conduct evaluations on both public datasets (e.g., PMData) and private datasets (e.g., MOX2-5 activity). Our CNN1D model achieves the highest prediction accuracy of 97[Formula: see text], while the MLP model outperforms other classifiers with an accuracy of 74[Formula: see text]. Furthermore, we evaluate the performance of our proposed OntoeCoach ontology model by assessing reasoning and query execution time metrics. The results demonstrate that our approach effectively plans and generates recommendations on both datasets. The rule set of OntoeCoach can also be generalized to enhance interpretability.

Mining Sensor Data to Assess Changes in Physical Activity Behaviors in Health Interventions: Systematic Review

BACKGROUND

Sensors are increasingly used in health interventions to unobtrusively and continuously capture participants' physical activity in free-living conditions. The rich granularity of sensor data offers great potential for analyzing patterns and changes in physical activity behaviors. The use of specialized machine learning and data mining techniques to detect, extract, and analyze these patterns has increased, helping to better understand how participants' physical activity evolves.

OBJECTIVE

The aim of this systematic review was to identify and present the various data mining techniques employed to analyze changes in physical activity behaviors from sensors-derived data in health education and health promotion intervention studies. We addressed two main research questions: (1) What are the current techniques used for mining physical activity sensor data to detect behavior changes in health education or health promotion contexts? (2) What are the challenges and opportunities in mining physical activity sensor data for detecting physical activity behavior changes?

METHODS

The systematic review was performed in May 2021 using the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. We queried the Association for Computing Machinery (ACM), IEEE Xplore, ProQuest, Scopus, Web of Science, Education Resources Information Center (ERIC), and Springer literature databases for peer-reviewed references related to wearable machine learning to detect physical activity changes in health education. A total of 4388 references were initially retrieved from the databases. After removing duplicates and screening titles and abstracts, 285 references were subjected to full-text review, resulting in 19 articles included for analysis.

RESULTS

All studies used accelerometers, sometimes in combination with another sensor (37%). Data were collected over a period ranging from 4 days to 1 year (median 10 weeks) from a cohort size ranging between 10 and 11615 (median 74). Data preprocessing was mainly carried out using proprietary software, generally resulting in step counts and time spent in physical activity aggregated predominantly at the daily or minute level. The main features used as input for the data mining models were descriptive statistics of the preprocessed data. The most common data mining methods were classifiers, clusters, and decision-making algorithms, and these focused on personalization (58%) and analysis of physical activity behaviors (42%).

CONCLUSIONS

Mining sensor data offers great opportunities to analyze physical activity behavior changes, build models to better detect and interpret behavior changes, and allow for personalized feedback and support for participants, especially where larger sample sizes and longer recording times are available. Exploring different data aggregation levels can help detect subtle and sustained behavior changes. However, the literature suggests that there is still work remaining to improve the transparency, explicitness, and standardization of the data preprocessing and mining processes to establish best practices and make the detection methods easier to understand, scrutinize, and reproduce.

Machine learning and ontology in eCoaching for personalized activity level monitoring and recommendation generation

Leading a sedentary lifestyle may cause numerous health problems. Therefore, passive lifestyle changes should be given priority to avoid severe long-term damage. Automatic health coaching system may help people manage a healthy lifestyle with continuous health state monitoring and personalized recommendation generation with machine learning (ML). This study proposes a semantic ontology model to annotate the ML-prediction outcomes and personal preferences to conceptualize personalized recommendation generation with a hybrid approach. We use a transfer learning approach to improve ML model training and its performance, and an incremental learning approach to handle daily growing data and fit them into the ML models. Furthermore, we propose a personalized activity recommendation algorithm for a healthy lifestyle by combining transfer learning, incremental learning, the proposed semantic ontology model, and personal preference data. For the overall experiment, we use public and private activity datasets collected from healthy adults (n = 33 for public datasets; n = 16 for private datasets). The standard ML algorithms have been used to investigate the possibility of classifying daily physical activity levels into the following activity classes: sedentary (0), low active (1), active (2), highly active (3), and rigorous active (4). The daily step count, low physical activity, medium physical activity, and vigorous physical activity serve as input for the classification models. We first use publicly available Fitbit datasets to build the initial classification models. Subsequently, we re-use the pre-trained ML classifiers on the real-time MOX2-5 dataset using transfer learning. We test several standard algorithms and select the best-performing model with optimized configuration for our use case by empirical testing. We find that DecisionTreeClassifier with a criterion "entropy" outperforms other ML classifiers with a mean accuracy score of 97.50% (F1 = 97.00, precision = 97.00, recall = 98.00, MCC = 96.78) and 96.10% (F1 = 96.00, precision = 96.00, recall = 96.00, MCC = 96.10) in Fitbit and MOX2-5 datasets, respectively. Using transfer learning, the DecisionTreeClassifier with a criterion "entropy" outperforms other classifiers with a mean accuracy score of 97.99% (F1 = 98.00, precision = 98.00, recall = 98.00, MCC = 96.79). Therefore, the transfer learning approach improves the machine learning model performance by ≈ 1.98% for defined datasets and settings on MOX2-5 datasets. The Hermit reasoner outperforms other reasoners with an average reasoning time of 1.1-2.1 s, under defined settings in our proposed ontology model. Our proposed algorithm for personalized recommendations conceptualizes a direction to combine the classification results and personal preferences in an ontology for activity eCoaching. The proposed method of combining machine learning technology with semantic rules is an invaluable asset in personalized recommendation generation. Moreover, the semantic rules in the knowledge base and SPARQL (SPARQL Protocol and RDF Query Language) query processing in the query engine helps to understand the logic behind the personalized recommendation generation.

Unsupervised Early Detection of Physical Activity Behaviour Changes from Wearable Accelerometer Data

Wearable accelerometers record physical activity with high resolution, potentially capturing the rich details of behaviour changes and habits. Detecting these changes as they emerge is valuable information for any strategy that promotes physical activity and teaches healthy behaviours or habits. Indeed, this offers the opportunity to provide timely feedback and to tailor programmes to each participant's needs, thus helping to promote the adherence to and the effectiveness of the intervention. This article presents and illustrates U-BEHAVED, an unsupervised algorithm that periodically scans step data streamed from activity trackers to detect physical activity behaviour changes to assess whether they may become habitual patterns. Using rolling time windows, current behaviours are compared with recent previous ones, identifying any significant change. If sustained over time, these new behaviours are classified as potentially new habits. We validated this detection algorithm using a physical activity tracker step dataset (N = 12,798) from 79 users. The algorithm detected 80% of behaviour changes of at least 400 steps within the same hour in users with low variability in physical activity, and of 1600 steps in those with high variability. Based on a threshold cadence of approximately 100 steps per minute for standard walking pace, this number of steps would suggest approximately 4 and 16 min of physical activity at moderate-to-vigorous intensity, respectively. The detection rate for new habits was 80% with a minimum threshold of 500 or 1600 steps within the same hour in users with low or high variability, respectively.

Classification of Parkinson's disease and its stages using machine learning

As digital health technology becomes more pervasive, machine learning (ML) provides a robust way to analyze and interpret the myriad of collected features. The purpose of this preliminary work was to use ML classification to assess the benefits and relevance of neurocognitive features both tablet-based assessments and self-reported metrics, as they relate to Parkinson’s Disease (PD) and its stages [Hoehn and Yahr (H&Y) Stages 1–5]. Further, this work aims to compare perceived versus sensor-based neurocognitive abilities. In this study, 75 participants (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = 50$$\end{document}n=50 PD; \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n = 25$$\end{document}n=25 control) completed 14 tablet-based neurocognitive functional tests (e.g., motor, memory, speech, executive, and multifunction), functional movement assessments (e.g., Berg Balance Scale), and standardized health questionnaires (e.g., PDQ-39). Decision tree classification of sensor-based features allowed for the discrimination of PD from healthy controls with an accuracy of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$92.6\%$$\end{document}92.6%, and early and advanced stages of PD with an accuracy of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$73.7\%$$\end{document}73.7%; compared to the current gold standard tools [e.g., standardized health questionnaires (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$78.3\%$$\end{document}78.3% accuracy) and functional movement assessments (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$70\%$$\end{document}70% accuracy)]. Significant features were also identified using decision tree classification. Device magnitude of acceleration was significant in 12 of 14 tests (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$85.7\%$$\end{document}85.7%), regardless of test type. For classification between diagnosed and control populations, 17 motor (e.g., device magnitude of acceleration), 9 accuracy (e.g., number of correct/incorrect interactions), and 8 timing features (e.g., time to between interactions) were significant. For classification between early (H&Y Stages 1 and 2) and advanced (H&Y Stages 3, 4, and 5) stages of PD, 7 motor, 12 accuracy, and 14 timing features were significant. Finally, this work depicts that perceived functionality of individuals with PD differed from sensor-based functionalities. In early-stage PD was shown to be \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$21.6\%$$\end{document}21.6% lower than sensor-based scores with notable perceived deficits in memory and executive function. However, individuals in advanced stages had elevated perceptions (1.57x) for executive and behavioral functions compared to early-stage populations. Machine learning in digital health systems allows for a more comprehensive understanding of neurodegenerative diseases and their stages and may also depict new features that influence the ways digital health technology should be configured.

Mitigating Issues With/of/for True Personalization

A common but false perception persists about the level and type of personalization in the offerings of contemporary software, information systems, and services, known as Personalization Myopia: this involves a tendency for researchers to think that there are many more personalized services than there genuinely are, for the general audience to think that they are offered personalized services when they really are not, and for practitioners to have a mistaken idea of what makes a service personalized. And yet in an era, which mashes up large amounts of data, business analytics, deep learning, and persuasive systems, true personalization is a most promising approach for innovating and developing new types of systems and services—including support for behavior change. The potential of true personalization is elaborated in this article, especially with regards to persuasive software features and the oft-neglected fact that users change over time.

Are Machine Learning Models on Wrist Accelerometry Robust against Differences in Physical Performance among Older Adults?

Sufficient physical activity (PA) reduces the risk of a myriad of diseases and preserves physical capabilities in later life. While there have been significant achievements in mapping accelerations to real-life movements using machine learning (ML), errors continue to be common, particularly for wrist-worn devices. It remains unknown whether ML models are robust for estimating age-related loss of physical function. In this study, we evaluated the performance of ML models (XGBoost and LASSO) to estimate the hallmark measures of PA in low physical performance (LPP) and high physical performance (HPP) groups. Our models were built to recognize PA types and intensities, identify each individual activity, and estimate energy expenditure (EE) using wrist-worn accelerometer data (33 activities per participant) from a large sample of participants (n = 247, 57% females, aged 60+ years). Results indicated that the ML models were accurate in recognizing PA by type and intensity while also estimating EE accurately. However, the models built to recognize individual activities were less robust. Across all tasks, XGBoost outperformed LASSO. XGBoost obtained F1-Scores for sedentary (0.932 ± 0.005), locomotion (0.946 ± 0.003), lifestyle (0.927 ± 0.006), and strength flexibility exercise (0.915 ± 0.017) activity type recognition tasks. The F1-Scores for recognizing low, light, and moderate activity intensity were (0.932 ± 0.005), (0.840 ± 0.004), and (0.869 ± 0.005), respectively. The root mean square error for EE estimation was 0.836 ± 0.059 METs. There was no evidence showing that splitting the participants into the LPP and HPP groups improved the models’ performance on estimating the hallmark measures of physical activities. In conclusion, using features derived from wrist-worn accelerometer data, machine learning models can accurately recognize PA types and intensities and estimate EE for older adults with high and low physical function.

The effect of two lockdowns on physical activity of school-age children

The introduction of strict quarantine restrictions in many countries initiated a direction in science to study the behavioral characteristics of children and adolescents during the social isolation at the population level. We present our observations during the two lockdowns in Ukraine. The objective of this study was to determine: a) the level of light (LPA) and moderate-to-vigorous (MVPA) physical activity among school-age children, and b) the impact of the external and internal factors on their physical activity during the lockdown. Global Physical Activity Questionnaire (GPAQ) as part of our questionnaire Q-RAPH was used. Parents of 1091 children 6–18 years old (54% boys) filled Q-RAPH at two measurement points in 2020 and 2021. After performing ANCOVA and logistic regression, we found a significant decrease in MVPA by 12.7% in 2021 compared to 2020 (p ​< ​0.001) while LPA was about 1.5 ​h a day during both periods. The proportion of children who reach the recommended levels of MVPA also decreased by 13.7% in 2021 (p ​< ​0.001). Factors negatively affecting the achievement of 60 ​min a day of MVPA were female gender, chronic diseases, overweight/obesity, non-participation in organized sports, and a decrease in the average air temperature. This study evidences the insufficient level of preventive measures and requires an intensification of health education among the Ukrainian population. When developing preventive measures, special attention should be paid to groups vulnerable to MVPA reduction as children who have chronic diseases and/or overweight/obesity as well as non-participation in sports.

Towards Symptom-Specific Intervention Recommendation Systems

BACKGROUND

Mobile devices and their capabilities (e.g., device sensors and human-device interactions) are increasingly being considered for use in clinical assessments and disease monitoring due to their ability to provide objective, repeatable, and more accurate measures of neurocognitive performance. These mobile-based assessments also provide a foundation for the design of intervention recommendations.

OBJECTIVE

The purpose of this work was to assess the benefits of various physical intervention programs as they relate to Parkinson's disease (PD), its symptoms, and stages (Hoehn and Yahr (H&Y) Stages 1-5).

METHODS

Ninety-five participants (n = 70 PD; n = 25 control) completed 14 tablet-based neurocognitive functional tests (e.g., motor, memory, speech, executive, and multi-function) and standardized health questionnaires. 208 symptom-specific digital features were normalized to assess the benefits of various physical intervention programs (e.g., aerobic activity, non-contact boxing, functional strength, and yoga) for individuals with PD. While previous studies have shown that physical interventions improve both motor and non-motor PD symptoms, this paper expands on previous works by mapping symptom-specific neurocognitive functionalities to specific physical intervention programs across stages of PD.

RESULTS

For early-stage PD (e.g., H&Y Stages 1 & 2), functional strength activities provided the largest overall significant delta improvement (Δ= 0.1883; p = 0.0265), whereas aerobic activity provided the largest overall significant delta improvement (Δ= 0.2700; p = 0.0364) for advanced stages of PD (e.g., H&Y Stages 3-5).

CONCLUSIONS

As mobile-based digital health technology allows for the collection of larger, labeled, objective datasets, new ways to analyze and interpret patterns in this data emerge which can ultimately lead to new personalized medicine programs.

Digital Interventions on Healthy Lifestyle Management: Systematic Review

BACKGROUND

Digital interventions have tremendous potential to improve well-being and health care conveyance by improving adequacy, proficiency, availability, and personalization. They have gained acknowledgment in interventions for the management of a healthy lifestyle. Therefore, we are reviewing existing conceptual frameworks, digital intervention approaches, and associated methods to identify the impact of digital intervention on adopting a healthier lifestyle.

OBJECTIVE

This study aims to evaluate the impact of digital interventions on weight management in maintaining a healthy lifestyle (eg, regular physical activity, healthy habits, and proper dietary patterns).

METHODS

We conducted a systematic literature review to search the scientific databases (Nature, SpringerLink, Elsevier, IEEE Xplore, and PubMed) that included digital interventions on healthy lifestyle, focusing on preventing obesity and being overweight as a prime objective. Peer-reviewed articles published between 2015 and 2020 were included. We used the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines and a framework for an evidence-based systematic review. Furthermore, we improved the review process by adopting the Rayyan tool and the Scale for the Assessment of Narrative Review Articles.

RESULTS

Our initial searches identified 780 potential studies through electronic and manual searches; however, 107 articles in the final stage were cited following the specified inclusion and exclusion criteria. The identified methods for a successful digital intervention to promote a healthy lifestyle are self-monitoring, self-motivation, goal setting, personalized feedback, participant engagement, psychological empowerment, persuasion, digital literacy, efficacy, and credibility. In this study, we identified existing conceptual frameworks for digital interventions, different approaches to provide digital interventions, associated methods, and execution challenges and their impact on the promotion of healthy lifestyle management.

CONCLUSIONS

This systematic literature review selected intervention principles (rules), theories, design features, ways to determine efficient interventions, and weaknesses in healthy lifestyle management from established digital intervention approaches. The results help us understand how digital interventions influence lifestyle management and overcome the existing shortcomings. It serves as a basis for further research with a focus on designing, developing, testing, and evaluating the generation of personalized lifestyle recommendations as a part of digital health interventions.

Early Prediction of Physical Performance in Elite Soccer Matches-A Machine Learning Approach to Support Substitutions

Substitution is an essential tool for a coach to influence the match. Factors like the injury of a player, required tactical changes, or underperformance of a player initiates substitutions. This study aims to predict the physical performance of individual players in an early phase of the match to provide additional information to the coach for his decision on substitutions. Tracking data of individual players, except for goalkeepers, from 302 elite soccer matches of the Dutch 'Eredivisie' 2018-2019 season were used to enable the prediction of the individual physical performance. The players' physical performance is expressed in the variables distance covered, distance in speed category, and energy expenditure in power category. The individualized normalized variables were used to build machine learning models that predict whether players will achieve 100%, 95%, or 90% of their average physical performance in a match. The tree-based algorithms Random Forest and Decision Tree were applied to build the models. A simple Naïve Bayes algorithm was used as the baseline model to support the superiority of the tree-based algorithms. The machine learning technique Random Forest combined with the variable energy expenditure in the power category was the most precise. The combination of Random Forest and energy expenditure in the power category resulted in precision in predicting performance and underperformance after 15 min in a match, and the values were 0.91, 0.88, and 0.92 for the thresholds 100%, 95%, and 90%, respectively. To conclude, it is possible to predict the physical performance of individual players in an early phase of the match. These findings offer opportunities to support coaches in making more informed decisions on player substitutions in elite soccer.

Connecting Social Psychology and Deep Reinforcement Learning: A Probabilistic Predictor on the Intention to Do Home-Based Physical Activity After Message Exposure

Previous research has shown that sending personalized messages consistent with the recipient's psychological profile is essential to activate the change toward a healthy lifestyle. In this paper we present an example of how artificial intelligence can support psychology in this process, illustrating the development of a probabilistic predictor in the form of a Dynamic Bayesian Network (DBN). The predictor regards the change in the intention to do home-based physical activity after message exposure. The data used to construct the predictor are those of a study on the effects of framing in communication to promote physical activity at home during the Covid-19 lockdown. The theoretical reference is that of psychosocial research on the effects of framing, according to which similar communicative contents formulated in different ways can be differently effective depending on the characteristics of the recipient. Study participants completed a first questionnaire aimed at measuring the psychosocial dimensions involved in doing physical activity at home. Next, they read recommendation messages formulated with one of four different frames (gain, non-loss, non-gain, and loss). Finally, they completed a second questionnaire measuring their perception of the messages and again the intention to exercise at home. The collected data were analyzed to elicit a DBN, i.e., a probabilistic structure representing the interrelationships between all the dimensions considered in the study. The adopted procedure was aimed to achieve a good balance between explainability and predictivity. The elicited DBN was found to be consistent with the psychosocial theories assumed as reference and able to predict the effectiveness of the different messages starting from the relevant psychosocial dimensions of the recipients. In the next steps of our project, the DBN will form the basis for the training of a Deep Reinforcement Learning (DRL) system for the synthesis of automatic interaction strategies. In turn, the DRL system will train a Deep Neural Network (DNN) that will guide the online interaction process. The discussion focuses on the advantages of the proposed procedure in terms of interpretability and effectiveness.

The potential of artificial intelligence in enhancing adult weight loss: a scoping review

OBJECTIVE

To present an overview of how artificial intelligence (AI) could be used to regulate eating and dietary behaviours, exercise behaviours and weight loss.

DESIGN

A scoping review of global literature published from inception to 15 December 2020 was conducted according to Arksey and O'Malley's five-step framework. Eight databases (CINAHL, Cochrane-Central, Embase, IEEE Xplore, PsycINFO, PubMed, Scopus and Web of Science) were searched. Included studies were independently screened for eligibility by two reviewers with good interrater reliability (k = 0·96).

RESULTS

Sixty-six out of 5573 potential studies were included, representing more than 2031 participants. Three tenets of self-regulation were identified - self-monitoring (n 66, 100 %), optimisation of goal setting (n 10, 15·2 %) and self-control (n 10, 15·2 %). Articles were also categorised into three AI applications, namely machine perception (n 50), predictive analytics only (n 6) and real-time analytics with personalised micro-interventions (n 10). Machine perception focused on recognising food items, eating behaviours, physical activities and estimating energy balance. Predictive analytics focused on predicting weight loss, intervention adherence, dietary lapses and emotional eating. Studies on the last theme focused on evaluating AI-assisted weight management interventions that instantaneously collected behavioural data, optimised prediction models for behavioural lapse events and enhance behavioural self-control through adaptive and personalised nudges/prompts. Only six studies reported average weight losses (2·4-4·7 %) of which two were statistically significant.

CONCLUSION

The use of AI for weight loss is still undeveloped. Based on the current study findings, we proposed a framework on the applicability of AI for weight loss but cautioned its contingency upon engagement and contextualisation.

Age Differences in Estimating Physical Activity by Wrist Accelerometry Using Machine Learning

Accelerometer-based fitness trackers and smartwatches are proliferating with incessant attention towards health tracking. Despite their growing popularity, accurately measuring hallmark measures of physical activities has yet to be accomplished in adults of all ages. In this work, we evaluated the performance of four machine learning models: decision tree, random forest, extreme gradient boosting (XGBoost) and least absolute shrinkage and selection operator (LASSO), to estimate the hallmark measures of physical activities in young (20-50 years), middle-aged (50-70 years], and older adults (70-89 years]. Our models were built to recognize physical activity types, recognize physical activity intensities, estimate energy expenditure (EE) and recognize individual physical activities using wrist-worn tri-axial accelerometer data (33 activities per participant) from a large sample of participants (n = 253, 62% women, aged 20-89 years old). Results showed that the machine learning models were quite accurate at recognizing physical activity type and intensity and estimating energy expenditure. However, models performed less optimally when recognizing individual physical activities. F1-Scores derived from XGBoost's models were high for sedentary (0.955-0.973), locomotion (0.942-0.964) and lifestyle (0.913-0.949) activity types with no apparent difference across age groups. Low (0.919-0.947), light (0.813-0.828) and moderate (0.846-0.875) physical activity intensities were also recognized accurately. The root mean square error range for EE was approximately 1 equivalent of resting EE [0.835-1.009 METs]. Generally, random forest and XGBoost models outperformed other models. In conclusion, machine learning models to label physical activity types, activity intensity and energy expenditure are accurate and there are minimal differences in their performance across young, middle-aged and older adults.

An Automatic Ontology-Based Approach to Support Logical Representation of Observable and Measurable Data for Healthy Lifestyle Management: Proof-of-Concept Study

Background

Lifestyle diseases, because of adverse health behavior, are the foremost cause of death worldwide. An eCoach system may encourage individuals to lead a healthy lifestyle with early health risk prediction, personalized recommendation generation, and goal evaluation. Such an eCoach system needs to collect and transform distributed heterogenous health and wellness data into meaningful information to train an artificially intelligent health risk prediction model. However, it may produce a data compatibility dilemma. Our proposed eHealth ontology can increase interoperability between different heterogeneous networks, provide situation awareness, help in data integration, and discover inferred knowledge. This “proof-of-concept” study will help sensor, questionnaire, and interview data to be more organized for health risk prediction and personalized recommendation generation targeting obesity as a study case.

Objective

The aim of this study is to develop an OWL-based ontology (UiA eHealth Ontology/UiAeHo) model to annotate personal, physiological, behavioral, and contextual data from heterogeneous sources (sensor, questionnaire, and interview), followed by structuring and standardizing of diverse descriptions to generate meaningful, practical, personalized, and contextual lifestyle recommendations based on the defined rules.

Methods

We have developed a simulator to collect dummy personal, physiological, behavioral, and contextual data related to artificial participants involved in health monitoring. We have integrated the concepts of “Semantic Sensor Network Ontology” and “Systematized Nomenclature of Medicine—Clinical Terms” to develop our proposed eHealth ontology. The ontology has been created using Protégé (version 5.x). We have used the Java-based “Jena Framework” (version 3.16) for building a semantic web application that includes resource description framework (RDF) application programming interface (API), OWL API, native tuple store (tuple database), and the SPARQL (Simple Protocol and RDF Query Language) query engine. The logical and structural consistency of the proposed ontology has been evaluated with the “HermiT 1.4.3.x” ontology reasoner available in Protégé 5.x.

Results

The proposed ontology has been implemented for the study case “obesity.” However, it can be extended further to other lifestyle diseases. “UiA eHealth Ontology” has been constructed using logical axioms, declaration axioms, classes, object properties, and data properties. The ontology can be visualized with “Owl Viz,” and the formal representation has been used to infer a participant’s health status using the “HermiT” reasoner. We have also developed a module for ontology verification that behaves like a rule-based decision support system to predict the probability for health risk, based on the evaluation of the results obtained from SPARQL queries. Furthermore, we discussed the potential lifestyle recommendation generation plan against adverse behavioral risks.

Conclusions

This study has led to the creation of a meaningful, context-specific ontology to model massive, unintuitive, raw, unstructured observations for health and wellness data (eg, sensors, interviews, questionnaires) and to annotate them with semantic metadata to create a compact, intelligible abstraction for health risk predictions for individualized recommendation generation.

Fat mass prediction equations and reference ranges for Saudi Arabian Children aged 8-12 years using machine technique method

Background

The number of children with obesity has increased in Saudi Arabia, which is a significant public health concern. Early diagnosis of childhood obesity and screening of the prevalence is needed using a simple in situ method. This study aims to generate statistical equations to predict body fat percentage (BF%) for Saudi children by employing machine learning technology and to establish gender and age-specific body fat reference range.

Methods

Data was combined from two cross-sectional studies conducted in Saudi Arabia for 1,292 boys and girls aged 8-12 years. Body fat was measured in both studies using bio-electrical impedance analysis devices. Height and weight were measured and body mass index was calculated and classified according to CDC 2,000 charts. A total of 603 girls and 374 boys were randomly selected for the learning phase, and 153 girls and 93 boys were employed in the validation set. Analyses of different machine learning methods showed that an accurate, sensitive model could be created. Two regression models were trained and fitted with the construction samples and validated. Gradient boosting algorithm was employed to achieve a better estimation and produce the equations, then the root means squared error (RMSE) equation was performed to decrease the error. Body fat reference ranges were derived for children aged 8-12 years.

Results

For the gradient boosting models, the predicted fat percentage values were more aligned with the true value than those in regression models. Gradient boosting achieved better performance than the regression equation as it combined multiple simple models into a single composite model to take advantage of that weak classifier. The developed predictive model archived RMSE of 3.12 for girls and 2.48 boys. BF% and Fat mass index charts were presented in which cut-offs for 5th, 75th and 95th centiles are used to define 'under-fat', 'normal', 'overfat' and 'subject with obesity'.

Conclusion

Machine learning models could represent a significant advancement for investigators studying adiposity-related issues in children. These models and newly developed centile charts could be useful tools for the estimation and classification of BF%.

The relationship between machine-learning-derived sleep parameters and behavior problems in 3- and 5-year-old children: results from the CHILD Cohort study

STUDY OBJECTIVES

Machine learning (ML) may provide insights into the underlying sleep stages of accelerometer-assessed sleep duration. We examined associations between ML-sleep patterns and behavior problems among preschool children.

METHODS

Children from the CHILD Cohort Edmonton site with actigraphy and behavior data at 3-years (n = 330) and 5-years (n = 304) were included. Parent-reported behavior problems were assessed by the Child Behavior Checklist. The Hidden Markov Model (HMM) classification method was used for ML analysis of the accelerometer sleep period. The average time each participant spent in each HMM-derived sleep state was expressed in hours per day. We analyzed associations between sleep and behavior problems stratified by children with and without sleep-disordered breathing (SDB).

RESULTS

Four hidden sleep states were identified at 3 years and six hidden sleep states at 5 years using HMM. The first sleep state identified for both ages (HMM-0) had zero counts (no movement). The remaining hidden states were merged together (HMM-mov). Children spent an average of 8.2 ± 1.2 h/day in HMM-0 and 2.6 ± 0.8 h/day in HMM-mov at 3 years. At age 5, children spent an average of 8.2 ± 0.9 h/day in HMM-0 and 1.9 ± 0.7 h/day in HMM-mov. Among SDB children, each hour in HMM-0 was associated with 0.79-point reduced externalizing behavior problems (95% CI -1.4, -0.12; p < 0.05), and a 1.27-point lower internalizing behavior problems (95% CI -2.02, -0.53; p < 0.01).

CONCLUSIONS

ML-sleep states were not associated with behavior problems in the general population of children. Children with SDB who had greater sleep duration without movement had lower behavioral problems. The ML-sleep states require validation with polysomnography.

Accuracy of Mobile Applications versus Wearable Devices in Long-Term Step Measurements

Fitness sensors and health systems are paving the way toward improving the quality of medical care by exploiting the benefits of new technology. For example, the great amount of patient-generated health data available today gives new opportunities to measure life parameters in real time and create a revolution in communication for professionals and patients. In this work, we concentrated on the basic parameter typically measured by fitness applications and devices-the number of steps taken daily. In particular, the main goal of this study was to compare the accuracy and precision of smartphone applications versus those of wearable devices to give users an idea about what can be expected regarding the relative difference in measurements achieved using different system typologies. In particular, the data obtained showed a difference of approximately 30%, proving that smartphone applications provide inaccurate measurements in long-term analysis, while wearable devices are precise and accurate. Accordingly, we challenge the reliability of previous studies reporting data collected with phone-based applications, and besides discussing the current limitations, we support the use of wearable devices for mHealth.

Experiences of mobile health in promoting physical activity: A qualitative systematic review and meta-ethnography

OBJECTIVE

Despite evidence supporting physical activity in primary and secondary prevention, many individuals do not meet recommended levels. Mobile health is a field with a growing evidence base and is proposed as a convenient method for delivering health interventions. Despite qualitative exploration of stakeholder perspectives, there is a lack of synthesis to inform evidence-based design. This study aims to resolve this by identifying and synthesising qualitative research on the experience of using mobile health applications to promote physical activity.

METHOD

A systematic review focused on qualitative research, mobile health and physical activity was conducted in October 2017 using CINAHL, ERIC, EMBASE, MEDLINE and PsycINFO databases. The protocol was registered with the Prospero database (Registration: CRD42018080610). Results were synthesised as a meta-ethnography.

RESULTS

Fifteen studies were included, covering a variety of populations, including people with diabetes, obesity, and serious mental illness. Five themes emerged: (a) personal factors and the experience of using mobile health, (b) mobile health and changes in thinking that support physical activity, (c) the experience of mobile health features, including prompts, goal setting and gamification, (d) the experience of personalised mobile health and physical activity, (e) technical and user issues in mobile health and their effect on experience.

CONCLUSION

Personal factors and features of the device influenced the experience of using mobile health to support physical activity. The two mechanisms through which mobile health use facilitated physical activity were strengthening of motivation and changes in self-awareness and strategising. Experiences were not entirely unproblematic as technical issues and adverse effects related to self-monitoring were noted. This synthesis provides insight into the experience of mobile health and is useful for researchers and healthcare practitioners interested in designing user-informed mobile health interventions for promoting physical activity.

An Adaptive, Data-Driven Personalized Advisor for Increasing Physical Activity

In recent years, there has been growing interest in the use of fitness trackers and smartphone applications for promoting physical activity. Many of these applications use accelerometers to estimate the level of activity that users engage in and provide visual reports of a user's step counts. When provided, most recommendations are limited to popular general health advice. In our study, we develop an approach for providing data-driven and personalized recommendations for intraday activity planning. We generate an hour-by-hour activity plan that is based on the user's probability of adhering to the plan. The user's probability of adherence to the plan is personalized, based on his/her past activity patterns and current activity target. Using this approach, we can tailor notifications (e.g., reminders, encouragement) to each user. We can also dynamically update the user's activity plan at mid-day, if his/her actual activity deviates sufficiently from the original plan. In this paper, we describe an implementation of our approach and report our technical findings with respect to identifying typical activity patterns from historical data, predicting whether an activity target will be achieved, and adapting an activity plan based on a user's actual performance throughout the day.

Quantile Coarsening Analysis of High-Volume Wearable Activity Data in a Longitudinal Observational Study

Owing to advances in sensor technologies on wearable devices, it is feasible to measure physical activity of an individual continuously over a long period. These devices afford opportunities to understand individual behaviors, which may then provide a basis for tailored behavior interventions. The large volume of data however poses challenges in data management and analysis. We propose a novel quantile coarsening analysis (QCA) of daily physical activity data, with a goal to reduce the volume of data while preserving key information. We applied QCA to a longitudinal study of 79 healthy participants whose step counts were monitored for up to 1 year by a Fitbit device, performed cluster analysis of daily activity, and identified individual activity signature or pattern in terms of the clusters identified. Using 21,393 time series of daily physical activity, we identified eight clusters. Employment and partner status were each associated with 5 of the 8 clusters. Using less than 2% of the original data, QCA provides accurate approximation of the mean physical activity, forms meaningful activity patterns associated with individual characteristics, and is a versatile tool for dimension reduction of densely sampled data.

Smart Sensing Technologies for Personalised e-Coaching

People living in both developed and developing countries face serious health challenges related to sedentary lifestyles. It is therefore essential to find new ways to improve health so that people can live longer and age well. With an ever-growing number of smart sensing systems developed and deployed across the globe, experts are primed to help coach people to have healthier behaviors. The increasing accountability associated with app- and device-based behavior tracking not only provides timely and personalized information and support, but also gives us an incentive to set goals and do more. This paper outlines some of the recent efforts made towards automatic and autonomous identification and coaching of troublesome behaviors to procure lasting, beneficial behavioral changes.

The influence of a consumer-wearable activity tracker on sedentary time and prolonged sedentary bouts: secondary analysis of a randomized controlled trial

Objective

A recent meta-analysis surmised pedometers were a useful panacea to independently reduce sedentary time (ST). To further test and expand on this deduction, we analyzed the ability of a consumer-wearable activity tracker to reduce ST and prolonged sedentary bouts (PSB). We originally conducted a 12-month randomized control trial where 800 employees from 13 organizations were assigned to control, activity tracker, or one of two activity tracker plus incentive groups designed to increase step count. The primary outcome was accelerometer measured moderate-to-vigorous physical activity.

Results

We conducted a secondary analysis on accelerometer measured daily ST and PSB bouts. A general linear mixed model was used to examine changes in ST and prolonged sedentary bouts, followed by between-group pairwise comparisons. Regression analyses were conducted to examine the association of changes in step counts with ST and PSB. The changes in ST and PSB were not statistically significant and not different between the groups (P < 0.05). Increases in step counts were concomitantly associated with decreases in ST and PSB, regardless of intervention (P < 0.05). Caution should be taken when considering consumer-wearable activity trackers as a means to reduce sedentary behavior.

Trial registration NCT01855776 Registered: August 8, 2012

Electronic supplementary material

The online version of this article (10.1186/s13104-018-3306-9) contains supplementary material, which is available to authorized users.