Neighbors-based Clinical Prediction Model for Walking Distance in Patients With Intermittent Claudication After Supervised Exercise Therapy

Abstract

The recommended therapy of choice for symptom relief in patients with intermittent claudication is supervised exercise therapy (SET). Although SET is known to be highly effective in increasing maximal and functional walking distance, Results vary greatly between patients. Evidence has mostly originated from research conducted at the population or group level, rather than at the level of the individual patient. Gaining greater insight into individuals’ prognosis may support patients and providers in tailoring care to the needs and priorities of the individual. Previous studies have explored the use of regression to develop prediction models for SET outcomes, but these models have demonstrated limitations in accuracy and external validity. We aimed to develop individual prognostic profiles on the outcome of SET after 6 months for patients with intermittent claudication, using a neighbors-based prediction approach to attempt to overcome the limitations of the previous prediction attempts. By this approach, similar patients, or matches, were selected using an adaptation of predictive mean matching. The following variables were used as matching characteristics: age, sex, smoking status, motivation and baseline walking distance. Historical outcomes data of the matches were then used to create the individual prediction. The final database for analyses included 3799 patients with 15,330 walking distance measurements. The number of matches with optimal prediction performance was m = 150, based on the average bias (–0.04 standard deviations; ideal = 0) coverage (48.7%; ideal = 50%) and precision (311 meters; 20% improvement over population-level prediction). Prediction calibration was good via both the within-sample and out-of-sample testing (Figs 1 and 2). Several features of our prediction approach may have contributed to the relatively good performance compared to previous attempts. First, historical observational data from daily practice of enough quality and quantity was available. Second, by our approach a new prediction model is developed for each new patient, using the realized data of the closest matches. This may have improved external validity. Finally, the resulting prediction is made accessible, user-friendly and supportive to the principles of patient-centered care. In conclusion, we have successfully developed neighbors-based predictions for walking distance in SET, which may provide a method to assist in patient-centered information regarding prognosis and progress during therapy.Fig 2View Large Image Figure ViewerDownload Hi-res image Download (PPT)