Data mining and infection control.

Patterns embedded in large volumes of clinical data may provide important insights into the characteristics of patients or care delivery processes, but may be difficult to identify by traditional means. Data mining offers methods that can recognize patterns in these large data sets and make them actionable. We present an example of this capability in which we successfully applied data mining to hospital infection control. The Data Mining Surveillance System (DMSS) uses data from the clinical laboratory and hospital information systems to create association rules linking patients, sample types, locations, organisms, and antibiotic susceptibilities. Changes in association strength over time signal epidemiologic patterns potentially appropriate for follow-up, and additional heuristic methods identify the most informative of these patterns for alerting.

The costs of nosocomial infections.

BACKGROUND: Although nosocomial infections (NIs) are widely regarded as expensive complications of healthcare delivery, their costs have not been rigorously quantified in large-scale studies. Additionally, problems that can bias cost estimates have often gone unaddressed. For example, are NIs more likely to cause significant extra length of stay (LOS) and costs, or are they more likely to be relatively inexpensive and inevitable consequences of long and expensive hospitalizations? This study is the largest of its kind to provide a rigorous analysis of the costs of NIs. OBJECTIVE: To precisely bound the attributable costs of a NI using large-scale data and to determine the effects of endogeneity between NIs and LOS on cost estimates. DESIGN, SETTING AND PATIENTS: Discharge diagnoses, cost, LOS, and NI data were collected for 1,355,347 admissions from March 30, 2001 to January 31, 2006 in 55 hospitals. MAIN OUTCOME MEASURES: The cost effects of NIs (in 2007 $) were estimated using multivariable regression models. Restricted models were applied to determine how cost estimates are confounded by disease severity and LOS. RESULTS: NIs are associated with $12,197 (95% CI, $4862-$19,533, P < 0.001) in incremental cost. A lower bound estimate of infection cost, controlling for LOS, is $4644 (95% CI, $1266-$7391). CONCLUSIONS: NIs are associated with substantial increases in the costs of inpatient care, even when estimates are corrected for potential endogenous confounding.

Clinical correlation of the CLSI susceptibility breakpoint for piperacillin- tazobactam against extended-spectrum-beta-lactamase-producing Escherichia coli and Klebsiella species.

We assessed infections caused by extended-spectrum-beta-lactamase-producing Escherichia coli or Klebsiella spp. treated with piperacillin-tazobactam to determine if the susceptibility breakpoint predicts outcome. Treatment was successful in 10 of 11 nonurinary infections from susceptible strains and in 2 of 6 infections with MICs of >16/4 mug/ml. All six urinary infections responded to treatment regardless of susceptibility.

A laboratory-based, hospital-wide, electronic marker for nosocomial infection: the future of infection control surveillance?

Faced with expectations to improve patient safety and contain costs, the US health care system is under increasing pressure to comprehensively and objectively account for nosocomial infections. Widely accepted nosocomial infection surveillance methods, however, are limited in scope, not sensitive, and applied inconsistently. In 907 inpatient admissions to Evanston Northwestern Healthcare hospitals (Evanston, IL), nosocomial infection identification by the Nosocomial Infection Marker (MedMined, Birmingham, AL), an electronic, laboratory-based marker, was compared with hospital-wide nosocomial infection detection by medical records review and established nosocomial infection detection methods. The sensitivity and specificity of marker analysis were 0.86 (95% confidence interval [CI 95], 0.76-0.96) and 0.984 (CI 95, 0.976, 0.992). Marker analysis also identified 11 intensive care unit-associated nosocomial infections (sensitivity, 1.0; specificity, 0.986). Nosocomial Infection Marker analysis had a comparable sensitivity (P > .3) to and lower specificity (P < .001) than medical records review. It is important to note that marker analysis statistically outperformed widely accepted surveillance methods, including hospital-wide detection by Study on the Efficacy of Nosocomial Infection Control chart review and intensive care unit detection by National Nosocomial Infections Surveillance techniques.