Background: Gram-negative resistance is a growing problem worldwide. It is generally believed that rates of resistant bacteria within a hospital are a function of antibiotic use, resistant organisms brought into the hospital, infection control efforts, and underlying severity of patient illness. The relative contribution of each to a particular resistance phenotype is unclear. P. aeruginosa is responsible for many hospital acquired infections and it may become resistant to carbapenems. In addition, newer threats to the future utility of the carbapenems are carbapenemase-producing K. pneumoniae Purpose: To determine if there is an association between the volume and composition of antibiotic use, geography, severity of illness and rates of carbapenem-resistant P. aeruginosa and K. pneumoniae. Methods: This is a retrospective ecological longitudinal investigation within the University HealthSystem Consortium affiliated academic medical centers. Antibiotic use data between January 1, 2006 and December 31, 2009 were obtained from billing records and reported as days of therapy per 1000 patient days (DOT/1000 PD), in addition to hospital characteristics (e.g. geographical location, bed size, case mix index). “Whole house” antibiograms were obtained to determine rates and proportions of carbapenem-resistant P. aeruginosa (CR-PA) and carbapenem resistant K. pneumoniae (CR-KP). Also, CR-KP isolation was generated as a binary outcome. Generalized estimating equations (GEE) were used to model CR-KP and CR-PA. Results: CR-KP rates (1000PDs) increased from 0.07 in 2006 to 0.15 in 2009 (P= 0.0118) and CR-KP proportions increased from 1.3% in 2006 to 3.1% in 2009 (0.0003) within 40 hospitals over 2006-2009. However, CR-PA rates and proportions were stable over the same period. Geographical location, carbapenems use, and antipseudomonal penicillins use were significantly associated with CR-KP isolation. Thus, for every ten DOT/1000 PDs increase in carbapenem use, the odds of CR-KP isolation increased by 42% (P=0.0149). In contrast, for every ten DOT/1000 PDs increase in antipseudomonal penicillin use, the odds of CR-KP isolation decreased by 14%. However, there was no significant model to explain CR-PA rates and proportions. Conclusion: Carbapenems, antipseudomonal penicillins, and geographical location were identified as risk factors associated with CR-KP isolation. These findings emphasize the challenges associated with the treatment of multidrug- gram-negative bacteria.

This dissertation describes a case study comparing the effectiveness of twoinformation systems that assess the quality of surgical care, the National SurgicalQuality Improvement Program (NSQIP) and the University HealthSystemConsortium Clinical Database (UHCCD). For the comparison, it develops aframework for assessing contextual data quality (CDQ) from the decision maker'sperspective. The differences in quality assessment systems to be studied areposited to be due to the differing contexts in which the data is encoded,transformed and managed impacting data quality for the purpose of surgicalquality assessment.Healthcare spending in the United States has risen faster than the rate of inflationfor over a decade and currently stands at about fifteen percent of the GrossDomestic Product. This has brought enormous pressures on the healthcareindustry to reduce costs while maintaining or improving quality. Numeroussystems to measure healthcare quality have been, and are being, developedincluding the two being studied. A more precise understanding of the differencesbetween these two systems' effectiveness in the assessment of surgical healthcarequality informs decisions nationally regarding hospital accreditation and qualitybasedreimbursements to hospitals.The CDQ framework elaborated is also applicable to executive informationsystems, data warehouses, web portals, and other information systems that drawinformation from disparate systems. Decision makers are more frequently havingdata available from across functional and hierarchical areas within organizationsand data quality issues have been identified in these systems unrelated to thesystem performance from which the data comes.The propositions explored and substantiated here are that workgroup contextinfluences data selection and definition, the data entry and encoding process,managerial control and feedback, and data transformation in information systems.These processes in turn influence contextual data quality relative to a particulardecision model.The study is a cross-sectional retrospective review of archival quality datagathered on 26,322 surgical patients at the University of Kentucky Hospital alongwith interviews of process owners in each system. The quality data includepatient risk/severity factors and outcome data recorded in the National SurgeryQuality Improvement Program (NSQIP) database and the UniversityHealthSystem Consortium Clinical Database (UHCCD).