. Author manuscript; available in PMC: 2014 Aug 28.

Published in final edited form as: JAMA. 2013 Aug 28;310(8):812–820. doi: 10.1001/jama.2013.276205

Table 3.

Associations Between Patient Intraoperative Cholangiography Use (No vs Yes) and Odds of Common Duct Injury and ERCP or Common Duct Exploration According to Method of Risk-Adjustment

	Risk-Adjustment Method
	Model 1		Model 2		Model 3		Model 4
	Unadjusted Logistic Regression, OR (95% CI)	P Value	Multivariable Logistic Regression, OR (95% CI)	P Value	Multilevel Logistic Regression, OR (95% CI)^a	P Value	Instrumental Variable Analysis, OR (95% CI)	P Value
Common duct injury^b
Percentage of hospital IOC use^c	1.73 (1.33–2.24)^c	<.001	1.76 (1.34–2.32)^d	<.001	1.79 (1.35–2.36)^d	<.001	1.26 (0.81–1.96)^d	.31
Percentage of surgeon IOC use^e	1.81 (1.38–2.37)^e	<.001	1.74 (1.32–2.29)^f	<.001	1.77 (1.34–2.36)^f	<.001	1.31 (0.91–1.89)^f	.14
ERCP/common duct exploration^g
Percentage of hospital IOC use^c	0.65 (0.62–0.68)^c	<.001	0.65 (0.62–0.69)^d	<.001	0.67 (0.64–0.72)^d	<.001	0.61 (0.56–0.67)^d	<.001
Percentage of surgeon IOC use^e	0.65 (0.62–0.69)^e	<.001	0.66 (0.62–0.69)^f	<.001	0.69 (0.65–0.74)^f	<.001	0.60 (0.56–0.64)^f	<.001

Abbreviations: ERCP, endoscopic retrograde cholangiopancreatography; IOC, intraoperative cholangiography; OR, odds ratio.

The multilevel model is based on patients clustered within hospitals. The model was also repeated with patients clustered within surgeons. The results were similar, so only 1 model is shown. The full multilevel models (outcome = common duct injury) for patients clustered within hospitals are shown in eTable 3 in the Supplement.

Models 2 through 4 for common duct injury are adjusted for age, sex, race, diagnosis, comorbidity, year of surgery, surgeon volume, US medical school training, hospital teaching status, and hospital bed size.

Model is restricted to hospitals with 20 or more cholecystectomies during the study period (n = 92 932).

The adjusted analyses sample size was 90 818 patients; 2.2% of the sample (n = 2114) patients had missing surgeon-level data: 1238 did not have a surgeon identification, 827 patients had a surgeon who was missing American Medical Association data, and 49 were missing hospital-level data. Sensitivity analyses excluding the missing surgeons and hospitals from the models showed similar results with no change in conclusion.

Models are restricted to hospitals performing 20 or more cholecystectomies and surgeons performing 5 or more cholecystectomies during the study period (n = 90 932).

The adjusted analyses sample size was 90 095 patients; 0.92% of the sample (n = 837) patients had missing surgeon-level or hospital-level data.

Models 2 through 4 for ERCP and/or common duct exploration are adjusted for age, sex, race, diagnosis, comorbidity, year of surgery, surgeon age, years in practice, surgeon volume, type of hospital, hospital teaching status, and hospital volume.