Background: In many studies, it is of interest to identify population subgroups that are relatively homogeneous with respect to an outcome. The nature of these subgroups can provide insight into effect mechanisms and suggest targets for tailored interventions. However, identifying relevant subgroups can be challenging with standard statistical methods. Main text: We review the literature on decision trees, a family of techniques for partitioning the population, on the basis of covariates, into distinct subgroups who share similar values of an outcome variable. We compare two decision tree methods, the popular Classification and Regression tree (CART) technique and the newer Conditional Inference tree (CTree) technique, assessing their performance in a simulation study and using data from the Box Lunch Study, a randomized controlled trial of a portion size intervention. Both CART and CTree identify homogeneous population subgroups and offer improved prediction accuracy relative to regression-based approaches when subgroups are truly present in the data. An important distinction between CART and CTree is that the latter uses a formal statistical hypothesis testing framework in building decision trees, which simplifies the process of identifying and interpreting the final tree model. We also introduce a novel way to visualize the subgroups defined by decision trees. Our novel graphical visualization provides a more scientifically meaningful characterization of the subgroups identified by decision trees. Conclusions: Decision trees are a useful tool for identifying homogeneous subgroups defined by combinations of individual characteristics. While all decision tree techniques generate subgroups, we advocate the use of the newer CTree technique due to its simplicity and ease of interpretation.
Background: Several factors are associated with an increased risk of anterior cruciate ligament (ACL) reconstruction revision. However, the ability to accurately translate these factors into a quantifiable risk of revision at a patient-specific level has remained elusive. We sought to determine if machine learning analysis of the Norwegian Knee Ligament Register (NKLR) can identify the most important risk factors associated with subsequent revision of primary ACL reconstruction and develop a clinically meaningful calculator for predicting revision of primary ACL reconstruction. Methods: Machine learning analysis was performed on the NKLR data set. The primary outcome was the probability of revision ACL reconstruction within 1, 2, and/or 5 years. Data were split randomly into training sets (75%) and test sets (25%). Four machine learning models were tested: Cox Lasso, survival random forest, generalized additive model, and gradient boosted regression. Concordance and calibration were calculated for all 4 models. Results: The data set included 24,935 patients, and 4.9% underwent a revision surgical procedure during a mean follow-up (and standard deviation) of 8.1 ± 4.1 years. All 4 models were well-calibrated, with moderate concordance (0.67 to 0.69). The Cox Lasso model required only 5 variables for outcome prediction. The other models either used more variables without an appreciable improvement in accuracy or had slightly lower accuracy overall. An in-clinic calculator was developed that can estimate the risk of ACL revision (Revision Risk Calculator). This calculator can quantify risk at a patient-specific level, with a plausible range from near 0% for low-risk patients to 20% for high-risk patients at 5 years. Conclusions: Machine learning analysis of a national knee ligament registry can predict the risk of ACL reconstruction revision with moderate accuracy. This algorithm supports the creation of an in-clinic calculator for point-of-care risk stratification based on the input of only 5 variables. Similar analysis using a larger or more comprehensive data set may improve the accuracy of risk prediction, and future studies incorporating patients who have experienced failure of ACL reconstruction but have not undergone subsequent revision may better predict the true risk of ACL reconstruction failure. Level of Evidence: Prognostic Level III. See Instructions for Authors for a complete description of levels of evidence.
We propose and compare combinations of several methods for classifying transportation activity data from smartphone GPS and accelerometer sensors. We have two main objectives. First, we aim to classify our data as accurately as possible. Second, we aim to reduce the dimensionality of the data as much as possible in order to reduce the computational burden of the classification. We combine dimension reduction and classification algorithms and compare them with a metric that balances accuracy and dimensionality. In doing so, we develop a classification algorithm that accurately classifies five different modes of transportation (i.e., walking, biking, car, bus and rail) while being computationally simple enough to run on a typical smartphone. Further, we use data that required no behavioral changes from the smartphone users to collect. Our best classification model uses the random forest algorithm to achieve 96.8% accuracy.