The Determinants of Productivity in Medical Testing: Intensity and Allocation of Care

Abstract

A large body of research has investigated whether physicians overuse care. There is less evidence on whether, for a fixed level of spending, doctors allocate resources to patients with the highest expected returns. We assess both sources of inefficiency exploiting variation in rates of negative imaging tests for pulmonary embolism. We document enormous across-doctor heterogeneity in testing conditional on patient population, which explains the negative relationship between physicians' testing rates and test yields. Furthermore, doctors do not target testing to the highest risk patients, reducing test yields by one third. Our calibration suggests misallocation is more costly than overuse.

Figure 1

Stylized relationship between testing thresholds, testing rates, and test yields Notes: Figure illustrates the theoretic relationship between testing thresholds, test yields and fraction of patients tested for two hypothetical doctors, A and B. Patients are sorted along the x-axis according to their risk of PE, q_id, from highest risk to lowest risk. Each point (x, y) along the plotted curve shows the fraction of patients x for whom q_id ≥ y. For example, at point (T^A = 2/3, τ^A = 1/2) in Panel A, the graph indicates that 2/3 of patients have a risk of PE that equals or exceeds 1/2. τ_A denotes doctor A’s testing threshold, T^A denotes the fraction of patients tested by doctor A, Z^A denotes doctor A’s test yield (among tested patients), and likewise for doctor B. In Panel A, both doctors face patient populations with the same distribution of PE risk. In Panel B, Doctor B’s patients are higher risk, i.e. for any given probability of a positive test q, a greater fraction of doctor B’s patients meet or exceed that threshold compared to doctor A.

Figure 2

Binned scatterplot of physician test yield by fraction of patients tested Notes: Figures displays a binned scatterplot based on our sample of Medicare claims data. Physicians are binned into deciles according to the fraction of patients they test. Panel A reports results across all patients evaluated by each doctor; the x-axis reports the average fraction of patients tested and the y-axis reports the rate of positive test results among tested patients, within each physician decile. The slope coefficient and standard error on the simple bivariate regression of average test yield on fraction of patients tested is reported on the panel. Panels B, C, and D maintain the same definitions of physician groups by deciles of test rate as in Panel A, but splits each doctor’s patients into groups according to whether they have a particular risk characteristic. We report average test rates and test yields by physician’s test decile, for patients with and without the listed characteristic.

Figure 3

Binned scatterplot of physician test yield by testing propensity index: Estimation results and simulations Notes: Figure displays a binned scatterplot based on our estimation and simulation results; physicians are binned into deciles based on the average estimated value of the testing propensity index ${\stackrel{\sim }{I}}_{\mathit{id}}^{\prime }$. The open circle markers plots the relationship between physicians’ actual test yields and physicians’ average ${\stackrel{\sim }{I}}_{\mathit{id}}^{\prime }$. Th solid square markers display the simulated relationship between testing propensities and test yields under a counterfactual with no variation in physician testing thresholds, and instead all physicians assigned the average testing threshold E(τ_d). The X-shaped markers displays the simulated relationship between testing propensities and test yields if there were no variation in physician testing thresholds and there were no misweighting of observable risk factors.