To study the extent to which selection bias poses problems for estimating the attributable cost of deep chest surgical-site infection (SSI) following coronary artery bypass graft (CABG) surgery.

Reanalysis of a prospective case–control study.

A large, Midwestern community medical center.

Cases were all patients who had an SSI (N = 41) following CABG and CABG and valve surgery between April 1996 and March 1998. Controls were every tenth uninfected patient (N = 160).

Estimates of the attributable cost of deep chest SSI were computed using unmatched comparison, matched comparison, linear regression, and Heckman's two-stage approach.

The attributable cost of deep chest SSI was estimated to be $20,012 by unmatched comparison, $19,579 by matched comparison, $20,103 by linear regression, and $14,211 by Heckman's two-stage method. Controlling for selection bias substantially reduced the cost estimate, but the coefficient capturing selection bias was not statistically significant.

Deep chest SSI significantly increases the cost of care for patients who undergo CABG surgery. Unmatched comparison, matched comparison, and linear regression estimated the attributable cost to be approximately $20,000. Although controlling for selection bias with Heckman's two-stage method resulted in a substantially smaller estimate, the coefficient for selection bias was not statistically significant, suggesting that the estimates derived from the other models should be acceptable. However, the magnitude of the difference between the models shows that the effect of selection bias can be substantial. Some exploration for selection bias is recommended when estimating the attributable cost of SSIs.

Bloodstream infections are a major cause of morbidity and mortality in patients receiving long-term hemodialysis. We wanted to determine the incidence of hemodialysis-related bloodstream infections in Canadian centers participating in the Canadian Nosocomial Infection Surveillance Program.

Prospective surveillance for hemodialysis-related bloodstream infections was performed in 11 centers during a 6-month period. Bloodstream infections were defined by published criteria. Hemodialysis denominators included the number of dialysis procedures, the number of patient-days on dialysis, and the frequencies of different types of vascular access.

There were 184 bloodstream infections in 133,158 dialysis procedures (1.4 per 1,000) and 316,953 patient-days (0.6 per 1,000). Hemodialysis access through arteriovenous (AV) fistulae was associated with the lowest risk for bloodstream infection (0.2 per 1,000 dialysis procedures). The relative risk for infection was 2.5 with AV graft access, 15.5 with cuffed and tunneled central venous catheter (CVC) access, and 22.5 with uncuffed CVC access (P < .001). There was marked variation among the 11 centers in the means of vascular access used for hemodialysis. Significant variation in infection rates was observed among the centers when controlling for types of access.

There was a hierarchy of risk of hemodialysis-related bloodstream infection according to type of vascular access. There was significant variation in the type of vascular access being used among the Canadian hemodialysis centers, and also variation in access-specific infection rates between centers.