Administrative coding data, compared with CDC/NHSN criteria, are poor indicators of health care–associated infections

Abstract

ICD-9-CM coding alone has been proposed as a method of surveillance for health care-associated infections (HAIs). The accuracy of this method, however, relative to accepted infection control criteria is not known. Retrospective analysis of patients at an academic medical center in 2005 who underwent surgical procedures or who were at risk for catheter-associated bloodstream infections or ventilator-associated pneumonia was performed. Patients previously identified with HAIs by Centers for Disease Control and Prevention's National Healthcare Safety Network surveillance methods were compared with those of the same risk group identified by secondary infection ICD-9-CM codes. Discordant cases identified by only coding were all rereviewed and adjusted prior to final analysis. When coding and surveillance were both negative, a sample of patients was used to estimate the proportion of false negatives in this group. The positive predictive values (PPVs) ranged from 0.14 to 0.51 with an aggregate of 0.23, even after adjustment for additional cases detected on subsequent medical record review. The negative predictive values (NPVs) ranged from 0.91 to 1.00, with an aggregate of 0.96. The estimates of the true variance of PPVs and NPVs across surgical procedures were small (0.0129, standard error, 0.009; 0.000145, standard error, 0.00019, respectively) and could be mostly explained by variation in prevalence of surgical site infections. Administrative coding alone appears to be a poor tool to be used as an infection control surveillance method. Its proposed use for routine HAI surveillance, public reporting of HAIs, interfacility comparisons, and nonpayment for performance should be seriously questioned.