Epidemiology of Patient Safety Events in an Academic Teaching Hospital

Description
Title: Epidemiology of Patient Safety Events in an Academic Teaching Hospital
Authors: Leeder, Ciera
Date: 2016
Abstract: Background: Adverse events are poor health outcomes caused by medical care rather than the underlying disease process. Voluntary reporting is a key component to adverse event reduction; however, incident reporting systems contain many limitations. The Patient Safety Learning System (PSLS) is an electronic incident reporting system with several unique features that were designed to address the weaknesses of previous systems, including a process for physician assessment of reported events to determine their significance. The primary objectives for this study were to determine the positive predictive value of the PSLS for identifying adverse events. Secondary objectives were to identify event, patient, and system-level factors associated with true events, and to assess event rates over time. Methods: I performed a retrospective cohort study using electronic health care data collected data from the Ottawa Hospital, between April 1 2010 and September 30, 2011. We Included all reported patient safety events if they occurred in adults aged 18 and older, admitted to an inpatient ward at the Civic, General, or Heart Institute campus. Events that occurred on Psychiatry, Rehabilitation services, were excluded due to data restrictions. A Clinical Reviewer manually reviewed each event to distinguish true events from non-events. For each hospital program, we used a generalized linear mixed model (GLIMMIX) to predict true events, using the role of the reporter as a random effect. Results: Over the study period, there were 2,569 events reported by hospital staff and physicians. Of these, 660 were rated as adverse events and 1,909 were rated as near misses. This yielded an overall positive predictive value of the PSLS system of 63% (95% CI 62-65%). The variance between reporters was not significant for Critical Care, Heart Institute, Nephrology, Obstetrics and Gynecology, Surgery and Periops, therefore I used a traditional logistic regression model with a common intercept. Number of months the PSLS was available was the only significant covariate found in all programs; the direction of the relationship was the same across all programs, and showed a decrease in true events reported over time. Other common covariates included: time from admission to event, severity of illness, and admission type. All models achieved a good calibration, yet discrimination was poor (c <0.70) in all models except Heart Institute. Discrimination ranged from 65% in Critical Care to 77% in the Heart Institute. Overall, the rate of patient safety events reported for inpatients was 6.39 per 1000 patient days. After an initial learning period, from April 2010-January 2011, in which rates were low, reporting rates increased and stabilized; remaining constant from month to month. The rate of true patient safety event reporting fluctuated greatly from April 2010-January 2011, after which they began to steadily decline. Trends in reporting were similar across hospital campus, reporter, and program. The majority of patient safety events were reported by nurses (44%), and laboratory staff (42%). The remaining 14% of events were reported by the classification ‘Other,’ which included all other hospital staff, such as technicians, physicians, and administrative staff. Only 7 physicians reported events to the PSLS during my study period, therefore, they were categorized under ‘Other’. Conclusions: Despite the many unique advantages of the PSLS, the proportion of true events reported has remained low. The overall utility of statistical models to predict patient safety events is limited. The traditional patient and system-level covariates, which are used to predict risk of adverse outcomes with high accuracy, did not help us discriminate between true patient safety events from non events. It is possible that many different individual and institutional barriers are influencing reporting and perhaps reviewing behavior, which in turn leads to non-clinical variability in what gets reported and classified as a patient safety event.
URL: http://hdl.handle.net/10393/34294
http://dx.doi.org/10.20381/ruor-5227
CollectionThèses, 2011 - // Theses, 2011 -
Files