Role of Computerized Physician Order Entry Systems in Facilitating Medication Errors
Koppel R, Metlay JP, Cohen A, et al. Role of computerized physician order entry systems in facilitating medication errors. JAMA. 2005;293:1197-1203.
Computerized Physician Order Entry (CPOE) has been touted as an effective means to reduce medical errors, especially medication errors. There have been preliminary studies that showed both potential and actual error reductions with CPOE. More recent data suggested that there may be potential for facilitating errors as well.
Koppel et al. aimed to study CPOE system-related factors that may actually increase risk of medication errors. The authors conducted structured interviews with end users (housestaff, pharmacists, nurses, nurse managers, and attending physicians), real-time observations of end users interfacing with the system, entering orders, charting medications, and reviewing orders, and focus groups with housestaff. These qualitative data were used to help generate a 71-question structured survey subsequently given to the housestaff. These questions pertain to working conditions, sources of stress, and errors. There were 261 responses representing an 88% response rate.
Twenty-two previously unexplored potential medication error sources abstracted from the survey were grouped into the 2 categories: 1) information errors, and 2) human-machine interface flaws. The first category refers to fragmented data and the disparate information systems within hospitals. The latter category includes rigid machine programming that does not correspond to or facilitate workflow. Only 10 survey elements with sufficiently robust results were reported. About 40% of respondents used CPOE to determine dosage of infrequently prescribed medications at least once a week or more. Incorrect doses may be ordered if users follow the dosage information in the system that is based on drug inventory rather than clinical recommendations. Twenty-two percent of respondents noted that more than once a week duplicate or conflicting medications were ordered and not detected for several hours. Disorganized display of patient medications was believed to be partly responsible. More than 80% of respondents noted unintended delay in renewing antibiotics at least once. Such gaps were possible partially because the reminder system occurred in the paper chart while order entry was done with the computer. With respect to the human-machine interface, 55% reported difficulty identifying the correct patient because of poor or fragmented displays, and 23% reported this occurring more than a few times per week. System downtime leading to delay in order entry was reported by 47% to occur more than once a week. System inflexibility also led to difficulties in specifying medications and ordering nonformulary medications. This was reported by 31% to occur at least several times a week, and 24% reported this daily or more frequently.
This was a survey of end users of a CPOE system in a single institution, and the survey elements were mainly estimates of error risks. Nevertheless, it appropriately draws attention to the importance of the unique culture of each institution, efficient workflow, and coherent human-machine interface. The anticipated error reductions may not materialize if these issues are neglected. Hospitalists can serve a critical role in implementation and customization of CPOE systems that allow clinicians to do the right thing more timely and efficiently.
Risk Stratification for In-hospital Mortality in Acutely Decompensated Heart Failure: Classification and Regression Tree Analysis
Fonarow GC, Adams KF, Abraham WT, Yancy CW, Boscardin WJ; ADHERE Scientific Advisory Committee, Study Group, and Investigators. Risk stratification for in-hospital mortality in acutely decompensated heart failure: classification and regression tree analysis. JAMA. 2005;293:572-80.
Heart failure is an important and growing cause of hospitalization in this country, and it is one of the most common clinical entities encountered by hospitalists. While there are some risk assessment tools available for outpatients with heart failure, there has not been a risk stratification tool published for inpatients. In this study by Fonarow et al. in JAMA, the authors describe a simple risk-stratification formula for in-hospital mortality in patients with acutely decompensated heart failure. Data from the ADHERE registry (Acute Decompensated Heart Failure National Registry, which is industry sponsored, as was this study) were used to model the risk of in-hospital death using a classification and regression tree (CART) analysis. This was done in a 2-stage process. First, investigators established a derivation cohort of approximately 33,000 patients (sequential hospital admissions from October 2001 to February 2003) from the ADHERE registry, and used the CART method to analyze 39 clinical variables to determine which were the best predictors of in-hospital mortality. This analysis was used to derive a risk tree to partition patients into low-, intermediate-, and high-risk groups. Second, the validity of this method was tested by applying the prediction tool to a cohort of the subsequent 32,229 patients hospitalized in the ADHERE registry, from March 2003 to July 2003. The results were striking. Baseline characteristics and clinical outcomes between the derivation and validation cohorts were similar across the wide range of parameters examined. The difference in mortality between the low-, intermediate-, and high-risk groups was 23.6% in the highest-risk category and 1.8% in the low-risk category, while the intermediate group was stratified into 3 levels, with 20.0%, 5.0%, and 5.1% mortality risk in intermediate group levels 1, 2, and 3, respectively. Aside from the more than 10-fold range in mortality risk across the various groups, the outstanding feature of the authors’ findings was that 3 simple parameters were the most significant predictors of in-hospital mortality risk: BUN, SBP, and serum creatinine. Specifically, combinations of a serum BUN of 43 or greater, a serum creatinine of 2.75 or greater, and a systolic blood pressure of less than 115 were associated with higher mortality. They note that adding other predictors did not meaningfully increase the model’s accuracy. The authors comment that unlike other predictive models based on multivariate analyses (which are often complex, and therefore difficult to employ at bedside), this simple tool is easy to use. An additional advantage is that the data needed are typically available at time of admission and can therefore be used to make a timely clinical decision in terms of triage into an appropriate level of care. Similar risk assessment tools exist for the risk stratification of patients with the acute coronary syndrome, and given the frequency with which patients are admitted with acutely decompensated heart failure, this new tool should prove a welcome addition to the clinical decision-making abilities of hospitalists.