Studies of bloodstream infection outcomes: reading between the lines.

Abstract

Understanding the origins, treatment, prevention, and outcomes of nosocomial bloodstream infections (BSIs) is of utmost importance because they are one of the most frequent and severe infectious complications of hospitalization and medical care. The outcomes of BSIs may differ depending on patient factors including underlying conditions and immune status; organism factors including virulence and resistance; and treatment factors such as agent and dose used, delay in appropriate therapy, removal of hardware, and associated supportive therapy. This issue of Infection Control and Hospital Epidemiology includes four studies that examine different factors that affect the outcome of nosocomial BSIs.1-4 These studies present an opportunity to discuss some methodologic principles in measuring the outcomes of hospital-acquired infections and how methods used in such studies may affect study results. The design of studies that measure the effect of an infectious complication during hospitalization on the outcome of a patient is important and challenging because the many other factors that contribute to the outcome have to be neutralized. A patient who develops an infection would optimally be compared with himself or herself if no infection occurred; however, given that this is not possible, various methods of study design and analysis have been developed to adjust for factors that contribute to outcomes. One approach is to look for the “identical twin patient” in whom an infection has not occurred. This is “control by study design” in which a case-patient is matched to a controlpatient who is similar in every aspect, and the difference in outcome is compared between the patients. This approach is easy to understand and therefore has gained wide popularity. However, given that the twin patient never exists, the researcher must decide which factors are the most important and match on them. In our experience, if more than three factors are chosen, in most hospital-based studies it will be almost impossible to find appropriate controlpatients for many of the case-patients. The other approach is “control by analysis” in which patients with and without infection are identified and the effect of each factor on the outcome is adjusted for by stratification or multivariable analysis. Thus, the effect of the infection on the outcome is isolated from other effects. This method requires more reliance on the skills of the statistician and the interpretation of results by the investigator and is less intuitive to the reader, who must assume that the methodology is executed correctly. With the widespread availability of user-friendly statistical packages that allow less experienced investigators to perform complex statistical analysis, this method has gained popularity, perhaps at a cost of reduction in the quality of the analyses. Irrespective of which method is used, several issues remain important in the design of studies that examine the outcomes of patients with infections. Patients’ underlying diseases and severity of illness before the infectious complication occurs are important determinates of outcome; therefore, appropriate adjustment for these factors is essential. Because severely ill patients with multiple comorbidities are more likely to develop nosocomial infections as well as to suffer adverse outcomes, failure to adjust for underlying diseases can lead to an inappropriately high estimate of the impact of an infection on patient outcomes. Investigators have employed several methods to facilitate such adjustment, including use of Acute Physiology and