Publication bias in surgery: implications for informed consent.

BACKGROUND: Patients consenting for pancreas surgery are often quoted an operative risk of 1% to 3% based on the literature. However, these results are often from centers of excellence, and as a result the literature mortality rates may not be representative or generalizable.
METHODS: A MEDLINE search was performed to identify the major studies of pancreaticoduodenectomy (PD) and total pancreatectomy (TP) over a 6-y period (January 1998-December 2003). To obtain a literature-based mortality rate, we performed a meta-analysis of these published series and compared them with actual in-hospital mortality rates based on a representative 20% sample of hospital data in 37 states (the Nationwide Inpatient Sample). The sample included approximately 8 million patient records per year. Literature versus actual mortality rates were compared for the same 6-y period and stratified by academic versus nonacademic medical centers.
RESULTS: We identified 16 major studies on PD and TP totaling 3641 patients with an overall mortality rate of 3.2% (range 0%-9.3%). The actual mortality rate based on the Nationwide Inpatient Sample (n = 7604) was 2.4-fold higher than the literature rate (adjusted rate of 7.6% versus 3.2%, P < 0.0001). All literature-based series were published from academic medical centers. By contrast, in the national database, 26.3% of PDs (2003/7604) were performed at nonacademic medical centers with a mortality rate of 11.4% (229/2003). The actual mortality rate at academic medical centers was lower than nonacademic medical centers (6.4% (360/5601), P < 0.0001), but still higher than the literature-based rate of 3.2% (P < 0.0001).
CONCLUSIONS: Mortality rates for pancreatic resections in actual practice are 2.4-fold higher than those reported in the literature. Proper informed consent for surgical procedures should include an accurate description of the risks, using actual local and national mortality rates.