Is small-area variation in healthcare utilization explained by physician financial self-interest?

Objective:

This study brings together two major, prominent strands of research on variation in health care utilization that have not previously been combined -- small-area variation (SAV), and physician financial self-interest (FSI). SAV research finds that moving the entire U.S. to the health care patterns of low-utilization areas would save up to $600 billion annually without adverse effects on outcomes, quality, or access. However, little is known regarding the causes of utilization variation. The FSI research, which is most developed for imaging, has shown that physicians with FSI order images 2-4 times as much as physicians without FSI, but this research has not quantified the aggregate effect of FSI.

We will investigate whether SAV in the proportion of care provided by physicians with FSI accounts for much of SAV in utilization.

Methods:

We investigate the hypothesis (i) for total imaging, (ii) more narrowly, for each imaging modality (CT, MR, etc.) separately, and (iii) more broadly, by studying the relation of FSI in imaging to total healthcare costs.

Data from Medicare's 5% Research Identifiable Files for 2005 will be used to analyze variations across 3,436 hospital service areas (HSAs) from the Dartmouth Atlas of Healthcare. Multivariable regression will used to measure the contribution of FSI to SAV in per beneficiary utilization or costs of various types of care in HSAs, with controls for beneficiary characteristics (age, sex, etc.). FSI is defined in four alternative ways.

Preliminary findings:

Preliminary findings are based on 2004 Medicare data, and explore variations across counties. We find substantial variability among counties in imaging utilization, overall and for individual modalities (systematic component of variation (SCV) of 330 to 489). Similar variability was found at the county lever for all four definitions of FSI (CVs ranging from 31.68 to 91.91).

The percent of treating physicians with FSI in a county is positively and statistically significantly correlated with images per 1000 beneficiaries in the county (correlation coefficients of 0.13-0.19 across definitions). In multivariable regression (in a simple form at the state level), we find a statistically significant effect of percent of imaging referred by treating physicians with FSI in imaging but no significant effect of radiologists per 1,000 Medicare enrollees on total imaging utilization.

HSA-level analysis is still to be performed.

Conclusions:

If our findings support our hypothesis, we will have a potential strategy and achieving the savings suggested by the SAV research, if deteriorating finances require such action.