Does My High Blood Pressure Improve Your Survival? Overall and Subgroup Learning Curves in Health

Learning curves in health are of interest for a wide range of medical disciplines, for multiple types of healthcare providers and policy makers. In this paper, we distinguish between three types of learning when identifying overall learning curves: static learning, learning from cumulative experience and human capital depreciation. In addition, we approach the question of how treating more patients with specific characteristics improves provider performance. Information on the role of subgroups has the potential to better inform new or low outcome providers on how to improve. Statistically however, capturing all subgroup experiences in one analysis introduces strong collinearities among regressors. To soften collinearity problems, we explore the use of Lasso regression as a variable selection method and Theil-Goldberger mixed estimation to augment the available information. We use data from the Belgian Transcatheter Aorta Valve Implantation (TAVI) registry, containing information on the first 860 TAVI procedures in Belgium. Ultimately, we find evidence for both overall and subgroup learning effects: for 2-year survival, we find that the probability of survival is increased by about 0.16%-points for each additional patient treated. For adverse events like renal failure and stroke, we find that an extra day between procedures increases the probability for these events by 0.12%-points and 0.07%-points respectively. These overall effects are then split into subgroup effects where we find evidence for positive learning effects from physicians' experience with defibrillation, treating patients with hypertension and the use of certain types of replacement valves during the TAVI procedure.