Volume 36, Issue 10 , Pages S172.e1-S172.e3, December 2008
Cost of bloodstream infections
Article Outline
Background
Hospital-acquired bloodstream infections (BSIs) are relatively rare but do not occur randomly. This suggests that unobserved confounding factors can bias estimates of BSI-associated incremental costs of care. Compared with previous studies, this analysis used a large sample size for greater precision, actual cost-accounting data, and case matching combined with bounding estimates to correct for bias.
Methods
Data from 1,355,647 admissions during 69 months in 55 hospitals were collected from a large population database. BSIs were identified by the Nosocomial Infection Marker, a well-validated, electronic, laboratory-based marker used for automatic infection surveillance. Costs were obtained by matching laboratory data with hospital accounting system calculations and converted to 2006 US dollars.
Results
Of 58,376 presumed nosocomial infections, 12,578 (21.6%) were identified as BSIs. More than 50% of BSIs occurred within the first week of hospitalization and 80% during the first 2 weeks. Various analyses resulted in the following estimates of BSI-associated incremental costs: basic regression analysis, $19,643 (P < .0001; 95% confidence interval [CI]: $9026-$30,260); excluding infections occurring after day 14, $19,427 (P < .001; 95% CI: $8867-$29,986); excluding infections occurring after day 7, $20,600 (P < .001; 95% CI: $10,123-$30,077); controlling for other nosocomial infections, $12,774 (P < .001; 95% CI: $6257-$19,290); and controlling for length of stay, $5534 (P < .012; 95% CI: $1282-$9785).
Conclusion
Even when intentionally underestimated, BSI-associated increased costs are substantial. True costs of BSIs are likely to be between $10,000 and $20,000. More research is needed to explore how controlling BSI costs may affect the cost of inpatient care.
Nosocomial bloodstream infections (BSIs) are among the most serious health care-associated adverse events. Many studies have estimated the effects of BSIs on morbidity, mortality, and costs of care; however, making accurate, unbiased estimates present some challenges. BSIs are relatively rare events, with reported rates in some institutions as low as 0.08 per 1000 catheter-days.1 As a result, researchers must either collect large amounts of data or work with very small sample sizes. BSIs do not occur randomly, which suggests that patients with BSIs may be different from patients without BSIs. This leads to a problem referred to as “endogeneity” or “confounding by indication.” For example, if more acutely ill patients were more likely to acquire BSIs, they could be expected to have increased lengths of stay, costs of care, and mortality rates as a result of their underlying conditions, independent of the infection.
Previous studies have attempted to address the issue of endogeneity through the use of instrumental variables,2, 3 other selection correction methods,4 case matching approaches, or carefully bounding potential biases in parameter estimates.5 This study combines case matching with bounding estimates to correct for bias that might arise from unobserved confounding factors.
Data and methods
Study data were gathered from the records of 1,355,647 admissions during a period from March 2001 through January 2006 in 55 participating hospitals using the Cardinal Health/MedMined (Birmingham, AL) services installed database. Nosocomial infections (NIs) and BSIs were identified by the Nosocomial Infection Marker (NIM; copyright, patent pending, Cardinal Health, Inc), a well-validated, electronic, laboratory-based marker used for automatic infection surveillance.6 The NIM is defined as a nonduplicate isolate collected from a single source (sputum, wound, and others) on or after hospital day 3 or within 14 days of discharge. NIMs are determined from laboratory and patient admit-discharge-transfer data. The sensitivity and specificity of the NIM in detecting NIs have been estimated to be 0.86 and 0.94, respectively. NIs missed by the NIM were empirically treated surgical site infections that were never cultured so that NIM sensitivity for BSIs are likely to be higher. Costs were obtained by matching laboratory data with hospital accounting system calculations of both fixed and variable costs of care. All costs were converted into constant 2006 US dollars using the Consumer Price Index.7
Statistical analyses were performed using the Stata version 9.2 (StataCorp, College Station, TX) absorbing regression (areg) procedure. Only NIMs associated with BSIs were included in this analysis. The absorbing regression was used to allow separate mean costs to be estimated for each diagnostic-related group, and robust (Huber-White) standard errors were calculated to provide for the clustering of observed admission costs within hospitals. First, an unrestricted estimate of costs was calculated followed by restricted cost estimates that excluded BSIs identified 7 to 14 days after admission. This controlled for bias that could arise from a tendency of patients with long inpatient stays to develop infections because of either intrinsic patient factors (eg, compromised immune status) or prolonged exposure to the hospital environment. BSI costs were also regressed as before but controlled for the presence of other types of infection. BSI cost estimates were calculated while controlling for hospital length of stay (LOS), which provided an absolute lower bound estimate for BSI costs.
Results
Data from 1,355,647 admissions included information on 58,376 presumed NIs; of these, 12,578 (21.6%) were identified as BSIs. Basic regression analysis yielded an estimated BSI-associated incremental cost of $19,643 (P < .0001; 95% confidence interval [CI]: $9026-$30,260). The distribution of BSIs as a function of LOS showed that more than half of all BSIs occurred within the first week of hospitalization and that 80% occurred within the first 2 weeks (Fig 1).
Fig 1.
Bloodstream infections: cumulative frequency.
When infections that occurred after hospital day 14 were excluded, the estimated incremental costs of BSIs were $19,427 (P < .001; 95% CI: $8867-$29,986); when infections that occurred after hospital day 7 were excluded, the estimated incremental costs of BSIs were $20,600 (P < .001; 95% CI: $10,123-$30,077).
There was a moderate correlation between the presence of BSIs and other NIs (Γ = .28). When the analysis controlled for the presence of other NIs, then BSIs were associated with incremental costs of $12,774 (P < .001; 95% CI: $6257-$19,290). A regression analysis that controlled for LOS yielded an estimated incremental BSI cost of $5534 (P < .012; 95% CI: $1282-$9785).
Discussion
This analysis of infections that occurred before hospital day 14 showed that BSIs were associated with increased costs of care of $19,643, which is consistent with previous estimates. The significance of these findings relate to the large sample size, which provided greater precision compared with previous studies and the use of actual cost-accounting data rather than charges or reimbursements.
How cost estimates may be biased by unobserved factors was also explored. When infections that occurred before either hospital day 7 or 14 are excluded, the results do not differ significantly from the basic regression results, suggesting that endogeneity may not be an important influence.
It does not appear that LOS, by itself, is an important source of bias: more than half of BSIs occurred relatively early in the course of hospitalization. This is consistent with studies that found that the most effective prevention efforts focus on catheter placement. Controlling for other infections may help correct for bias that could arise from patients with compromised immune systems. The incremental costs of BSIs decreased to $12,774 when the effects of other infections were taken into account. Finally, an extreme lower bound on BSI costs was estimated by controlling for LOS. The resulting estimate of $5534, an intentional underestimate, still represents substantial additional costs of care associated with BSIs.
This study has several limitations. First, the participating hospitals were not a representative sample of all United States hospitals, so the results may not be widely applicable. Whereas costs of care differed among hospitals, there were not significant differences in the incremental costs of BSIs. Second, the data on patient and hospital characteristics were limited because of their proprietary nature. Obtaining the release of protected health information would have required extensive review at each institution. More precise estimates could be obtained if information on hospital characteristics and patient comorbidities were available. Finally, the only outcome investigated in this study was direct health care costs. BSIs are also associated with increased risk of morbidity and mortality. A complete estimate of the burden of disease associated with BSIs would take these outcomes into account as well.
The results of this study provide additional evidence that BSIs involve substantial costs and suggest that preventing BSIs could produce significant returns on investment. This study also provides an example of the value of automated infection surveillance, which enabled the study to be conducted at very low cost. Future studies that use this approach and augment data on patient characteristics and outcomes are highly desirable and will likely produce more insight about the costs of BSIs and the value of infection control efforts.
Conclusion
Even when intentionally underestimated, BSI-associated increased costs are substantial. The true incremental cost of care associated with BSIs is likely to lie between $10,000 and $20,000. Further research is needed to explore how controlling BSI costs may affect costs of inpatient care.
References
- . New technologies to prevent intravascular catheter-related bloodstream infections. Emerg Infect Dis. 2001;7:197–199
- . Identification of causal effects using instrumental variables. J Am Stat Assoc. 1996;17:268–275
- . Correcting for bias when estimating the cost of hospital-acquired infection: an analysis of lower respiratory tract infections in non-surgical patients. Health Econ. 2005;14:755–761
- . Nonrandom selection and the attributable cost of surgical-site infections. Infect Control Hosp Epidemiol. 2002;23:177–182
- Adjustments for unmeasured confounders in pharmacoepidemiologic database studies using external information. Med Care. 2007;45:S158–S165
- A laboratory-based, hospital-wide, electronic marker for nosocomial infection. Am J Clin Pathol. 2006;125:34–39
- Bureau of Labor Statistics. Consumer Price Index home page. Available at: http://www.bls.gov/cpi/. Accessed May 29, 2007.
Disclosures: Dr. Kilgore received an honorarium for participating in the symposium and writing this article.
PII: S0196-6553(08)00789-X
doi:10.1016/j.ajic.2008.10.004
© 2008 Association for Professionals in Infection Control and Epidemiology, Inc. Published by Elsevier Inc All rights reserved.
Volume 36, Issue 10 , Pages S172.e1-S172.e3, December 2008
