Arrieta, Alejandro. Measuring overtreatment: a structural model to estimate the impact of non-clinical factors on healthcare utilization. Retrieved from https://doi.org/doi:10.7282/T3S46S9F
DescriptionIt is well acknowledged that, in the agency relationship between physicians and patients, the informational advantage gives doctors an incentive to deviate from the appropriate treatment, thus incurring over- or under- utilization. However, the empirical consequence of this problem has not been adequately considered. In particular, physician agency creates a gap between appropriate treatment and actual treatment whose characteristics and effects on estimation are analogous to a classification error.
This thesis proposes a structural model based on misclassification in which the physician behavior characterizes the structure of the measurement error. The model produces consistent estimators and is able to measure the degree of over- and under-utilization by separating out the effect of clinical and non-clinical variables on treatment decision. The model is applied to cesarean section deliveries performed in New Jersey in 1999-2002. The results show a moderate but growing rate of non-clinically required c-sections of around 3.2%, implying that the rapid growth of c-section rates over these years is explained mainly by non-clinical factors.
In the second chapter, the model is used to study how reform in the Peruvian health system has increased physician incentives to overuse c-sections in private hospitals. C-section rates in the private sector grew from 27% to 48% after the health reform of 1997, while the rates remained constant at 19% in the public sector. Using a national survey, it is estimated that each year more than 13 thousand women are over-treated, having a c-section without medical reasons. This document highlights the consequences of unnecessary c-sections on women's reproductive rights, and establishes important implications and recommendations for other health reforms in Latin America.
The third chapter extends the parametric estimation of the structural misclassification model to a semi-parametric estimation based on a double-index semi-parametric maximum likelihood with bias correction (Klein and Vella, 2008). I show that misspecification error due to a wrong assumption in error distribution may lead to an important inconsistency in parametric estimates, thus justifying the use of a semi-parametric technique to support results. The parametric and semi-parametric models are compared using a Monte Carlo study and an application for c-section deliveries.