Self-referral in medical imaging: a meta-analysis of the literature.

PURPOSE: In the current political and economic climate, there is a desire to reduce health care costs; diagnostic imaging expenditure is one area of particular interest. The authors present a meta-analysis of the relative frequency of imaging utilization in the setting of self-referral compared with that of non-self-referral and a simulation of increased cost to Medicare Part B on the basis of this relative frequency. METHODS: The MEDLINE database was searched systematically. Specific inclusion criteria for relative frequency calculations were a numerator (number of patients imaged) and denominator (number of total patients seen) in each group (self-referrers and radiologist referrers). The relative risk of self-referral was determined for each group and is defined by the "relative frequency" of imaging utilization for the self-referrers divided by the frequency for the radiologist referrers. Relative frequency represents the increased (if >1) or decreased (if <1) chance of imaging by self-referrers over radiologist referrers. The meta-analysis was used to combine imaging frequencies for each referral condition of the individual studies that met inclusion criteria for an overall estimate of relative frequency, using a random-effects model to account for the variations among the studies. Relative frequency data were then used to perform a cost simulation to Medicare Part B using 2006 data. RESULTS: The initial search yielded 334 articles, 5 of which met the threshold for inclusion. In these 5 studies, 76,905,162 total episodes of care were analyzed. The individual relative frequency of imaging in the setting of self-referral ranged from 1.60 to 4.50. The combined relative frequency was 2.16 (95% confidence interval, 2.15-2.16) using the fixed-effects model and 2.48 (95% confidence interval, 1.90-3.24) using the random-effects model. For 2006 Government Accountability Office (GAO) data, the estimated cost of increased imaging in the setting of self-referral was $3.6 billion, but a range of costs was also provided to account for potential inaccuracies in the GAO data. CONCLUSIONS: The existing literature yields a combined relative frequency of imaging of 2.48 (95% confidence interval, 1.90-3.24) for self-referrers compared with non-self-referrers. Precise extrapolation of Medicare Part B costs attributable to self-referral would require changes in reporting requirements for imaging equipment ownership. Cost simulation results total billions of dollars annually and may be irrespective of potential inaccuracies in the GAO data as a result of Current Procedural Terminology(®) coding ambiguity and nontransparent reporting of equipment ownership.

Duke Scholars

Author Huiman Xie Barnhart Biostatistics & Bioinformatics, Division of Biostatistics