Abstract

Training, especially in healthcare, is resource intensive; particularly, as new technologies and techniques emerge, new training paradigms are needed. Robotic-assisted surgery (RAS) is one such technology that has become prevalent in various surgical specialties. Additionally, there are additional training demands on surgical trainees to learn skills for RAS. However, due to the limited resources of equipment availability, training space, and busy surgeon and trainee schedules, training operators to use the system can be difficult to schedule. This is critical to consider as waiting too long between training sessions can lead to forgetting and skill decay, while scheduling sessions too close together can lead to limited information gain. Furthermore, the mental workload of a training task contributes to how well the skill is processed during the training. All these factors culminate in a contradictory time-dependent issue: trainees need additional practice to improve their skills but also may not need as much time as they become more proficient. With these challenges, there is a need for optimal training strategies that incorporate the mental workload of trainees and training scheduling to limit forgetting in trainees. The purpose of this study was to model RAS training as a decision-making process that recommends the optimal schedules for training sessions. This work utilized a previously published dataset of 15 trainees performing skills tasks on the da Vinci Skill Simulator. Performance metrics, eye-tracking metrics of workload (i.e., gaze entropy), and training characteristics (e.g., tasks and training session intervals) were obtained. The RAS training was modelled as an adaptive Bayesian training model that considered the trainee’s forgetfulness, based on the Ebbinghaus forgetting curve, and workload during the training. The model was used to optimize the expected rewards (i.e., performance score improvement) for future training sessions. The model’s objective was to maximize the total performance score based on the trainees’ previous performance and the individual’s knowledge retention. It was found that a nonlinear scheduling and task assortment model considering the Ebbinghaus forgetting curve jointly with the learning curve was more accurate than a linear model that only considered immediate learning. These findings can help shape future adaptive and personalized training curriculums that most benefit the individual based on their task performance, information retention, and to identify the ideal time to schedule training sessions. By generating a structure that maximizes the trainees’ skills learned, it will help medical educators plan training sessions to optimize both institutional resources and trainees’ time to improve not only the skills of RAS trainees, but learners across healthcare and various domains.