Talko Bernhard Dijkhuis

Wearable technology and the Internet of Things have increased access to personal data, enabling applications that deliver individualized treatment and therapy within clinical pathways. To optimize coaching and interventions within such pathways, it is essential to identify all relevant factors in the available data. Feature selection can be a useful tool for achieving this, but with small, high-dimensional datasets, common in healthcare, it can be challenging. The goal of this study is to develop a method for identifying the most relevant features in small, high-dimensional datasets and to introduce a new ordering index that measures the quality of the orderings produced by feature selection methods. This novel index is sensitive to the quality of feature ordering and to the prediction model’s performance metrics when combined with a feature selection method. The index reaches its maximum when the number-of-features-versus-accuracy graph has an ideal concave-downward shape, reflecting increasing accuracy with each informative feature added and decreasing accuracy with each confounding feature added. Using this index, we define six feature orderings derived from the results of four standard feature selection methods. Using the performance metrics and our new ordering index, we show that the resulting orderings can identify more relevant features and improve the overall performance of the classification models, and that the ordering index is a useful contribution to feature selection techniques.

In today’s society, the negative consequences of physical inactivity have become increasingly prevalent, underscoring the sedentary epidemic [1,2]. The World Health Organisation (WHO) identifies in...

Human Digital Twins are an emerging type of Digital Twin used in healthcare to provide personalized support. Following this trend, we intend to elevate our virtual fitness coach, a coaching platform using wearable data on physical activity, to the level of a personalized Human Digital Twin. Preliminary investigations revealed a significant difference in performance, as measured by prediction accuracy and F1-score, between the optimal choice of machine learning algorithms for generalized and personalized processing of the available data. Based on these findings, this survey aims to establish the state of the art in the selection and application of machine learning algorithms in Human Digital Twin applications in healthcare. The survey reveals that, unlike general machine learning applications, there is a limited body of literature on optimization and the application of meta-learning in personalized Human Digital Twin solutions. As a conclusion, we provide direction for further research, formulated in the following research question: how can the optimization of human data feature engineering and personalized model selection be achieved in Human Digital Twins and can techniques such as meta-learning be of use in this context?

Nutrition of patients in ICU settings has been an ongoing matter of controversy for a long time. Under- and overfeeding are common and are strongly as

In recent decades, the number of cases of knee arthroplasty among people of working age has increased. The integrated clinical pathway ‘back at work after surgery’ is an initiative to reduce the possible cost of sick leave. The evaluation of this pathway, like many clinical studies, faces the challenge of small data sets with a relatively high number of features. In this study, we investigate the possibility of identifying features that are important in determining the duration of rehabilitation, expressed in the return-to-work period, by using feature selection tools. Several models are used to classify the patient’s data into two classes, and the results are evaluated based on the accuracy and the quality of the ordering of the features, for which we introduce a ranking score. A selection of estimators are used in an optimization step, reorganizing the feature ranking. The results show that for some models, the proposed optimization results in a better ordering of the features. The ordering of the features is evaluated visually and identified by the ranking score. Furthermore, for all models, higher accuracy, with a maximum of 91%, is achieved by applying the optimization process. The features that are identified as relevant for the duration of the return-to-work period are discussed and provide input for further research.

Although causal inference has shown great value in estimating effect sizes in, for instance, physics, medical studies, and economics, it is rarely used in sports science. Targeted Maximum Likelihood Estimation (TMLE) is a modern method for performing causal inference. TMLE is forgiving in the misspecification of the causal model and improves the estimation of effect sizes using machine-learning methods. We demonstrate the advantage of TMLE in sports science by comparing the calculated effect size with a Generalized Linear Model (GLM). In this study, we introduce TMLE and provide a roadmap for making causal inference and apply the roadmap along with the methods mentioned above in a simulation study and case study investigating the influence of substitutions on the physical performance of the entire soccer team (i.e., the effect size of substitutions on the total physical performance). We construct a causal model, a misspecified causal model, a simulation dataset, and an observed tracking dataset of individual players from 302 elite soccer matches. The simulation dataset results show that TMLE outperforms GLM in estimating the effect size of the substitutions on the total physical performance. Furthermore, TMLE is most robust against model misspecification in both the simulation and the tracking dataset. However, independent of the method used in the tracking dataset, it was found that substitutes increase the physical performance of the entire soccer team.

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.

Injuries of runners reduce the ability to train and hinder competing. Literature shows that the relation between potential risk factors and injuries are not definitive, limited, and inconsistent. In team sports, workload derivatives were identified as risk factors. However, there is an absence of literature in running on workload derivatives. This study used the workload derivatives acute workload, chronic workload, and acute: chronic workload ratios to investigate the relation between workload and injury risk in running. Twenty-three competitive runners kept a daily training log for 24 months. The runners reported training duration, training intensity and injuries. One-week (acute) and 4-week (chronic) workloads were calculated as the average of training duration multiplied by training intensity. The acute:chronic workload ratio was determined dividing the acute and chronic workloads. Results show that a fortnightly low increase of the acute:chronic workload ratio (0.10–0.78) led to an increased risk of sustaining an injury (p<0.001). Besides, a low increase of the acute:chronic workload ratio (0.05–0.62) between the second week and third week before an injury showed an association with increased injury risk (p=0.013). These findings demonstrate that the acute:chronic workload ratio relates to injury risk.