Medication Errors in United States Hospitals

C. A. Bond, PharmD, FASHP, FCCP, Cynthia L. Raehl, PharmD, FASHP, FCCP, and Todd Franke, PhD


Pharmacotherapy. 2001;21(9) 

In This Article


Hospital medication error information was collected as part of the 1992 National Clinical Pharmacy Services database survey.[16] Pharmacy directors were asked whether their hospital had a medication error reporting system, defined as an ongoing systematic program for reporting, monitoring, and reviewing medication errors. In addition, each pharmacy director was asked to report the total number of medication errors for the previous 12 months and the number of medication errors determined to adversely affect patient outcomes (defined as requiring additional drug therapy, increasing length of stay, or causing permanent harm or patient death). The variables used in this study to compare and contrast medication error data previously were shown to be associated with heath care outcomes and the provision of pharmaceutical services.[16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34]

Sources of Data

Data for pharmacy teaching affiliation, pharmacy directors' degree, pharmacists' location within each hospital, and drug costs were obtained from the 1992 database of the National Clinical Pharmacy Services.[16,33,34] The methods used for data analysis previously were published.[16,33,34] Mortality rate information was obtained from the Health Care Financing Administration.[35] Data on census region information, size, hospital ownership, hospital staffing, admissions data, occupancy rates, teaching affiliation, length of stay, and total cost of care for each hospital were obtained from the American Hospital Association's (AHA) Abridged Guide to the Health Care Field.[36] The survey instrument of the National Clinical Pharmacy Services was updated and pretested by 25 directors of pharmacy.[16,33,34] The questionnaire was mailed to the director of pharmacy in each acute care, general-medical, surgical hospital listed in the AHA database.[36] The methodology, variables, and demographic results of this study were previously published.[16,33,34]

The National Clinical Pharmacy Services database is the largest hospital and clinical pharmacy database in the U.S. This information was integrated into one database, and Stata Version 7, implemented on a personal computer (Pentium 450Mz), was used for all statistical analysis.[37] Only inpatient data were analyzed. Respondents from 1116 of the 1597 hospitals identified in the 1992 National Clinical Pharmacy Services database[16] provided information on the number of medication errors at their institutions. These hospitals constituted the study population.

Definitions and subsequent groupings used in the analysis are provided in the appendix. Data analysis was based on grouping hospitals by seven factors associated with health care outcomes and the provision of clinical pharmacy services.[16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34] Hospitals were assigned to one of nine geographic regions defined by the U.S. Bureau of the Census and to one of three size categories. Hospital pharmacy teaching affiliation was identified (affiliation with a college of pharmacy, no college of pharmacy affiliation but an affiliation with other health education programs, or no affiliation with any health education program). Hospital teaching affiliation was defined as teaching or nonteaching, with teaching determined by membership in the Council of Teaching Hospitals or the American Osteopathic Teaching Hospital Association. The pharmacy directors' educational backgrounds were grouped into four categories. Hospital ownership was grouped into four categories. Pharmacists' predominant location within the hospital was grouped into three categories: centralized, decentralized, centralized with ward visits.

Personnel variables used from the AHA database included the number of administrators, physicians, medical residents, registered nurses, licensed practical/vocational nurses, physician assistants, registered pharmacists, medical technologists, dietitians, occupational therapists, physical therapists, respiratory therapists, social workers, and total hospital personnel. In addition, several ratios were employed: ratio of board-certified physicians to all physicians, ratio of registered nurses to licensed practical/vocational nurses, and ratio of registered pharmacists to pharmacy technicians. Only full-time personnel were included in this analysis, since this was the only personnel measure common to the 14 personnel categories in the AHA database.[36] Each personnel variable was divided by the mean number of occupied beds for that hospital, to provide a hospital-specific staffing level based on a common workload measure (staffing/occupied bed). Simple regression was used to analyze staffing, staffing ratio variables, and medication errors.

Multiple regression analysis between the number of medication errors and mortality rates, drug costs/occupied bed/year, total cost of care/occupied bed/year, and mean length of stay were adjusted for severity of illness, which is known to influence these variables. Severity of illness was controlled by forcing three variables into the multiple regression analysis model: percentage of intensive care unit (ICU) days (calculated as ICU days divided by total inpatient days), annual number of emergency room visits divided by the average daily census, and percentage of Medicaid patients (calculated as Medicaid discharges divided by total discharges). These variables previously were validated as severity of illness measures in similar types of studies.[17,18,19,20,21,22,25,26,38,39] These specific variables were chosen because they are the only variables validated as adjusters for severity of illness using these national databases.[17,18,19,20,21,22,25,26,38,39] Whereas other variables (e.g., APACHE [Acute Physiology and Chronic Health Evaluation] scores, specific patient case mix, patient age, number of surgical patients, physician experience, length of shifts, or patient workloads) have been used to adjust for severity of illness with smaller patient populations, these variables were not available for the study hospitals. Diagnosis-related groups are not reliable severity of illness-adjusters, since many hospitals have inflated these measures.

Patient care outcome measures must adjust for patient characteristics that influence the outcome measure.[39,40,41] If outcome measures (e.g., mortality rate) do not adjust for severity of illness, conclusions for hospitals that treat more severely ill patients will be inaccurate.

Finally, a multiple regression analysis adjusted for severity of illness was employed with the hospital demographic variables, pharmacy and staffing variables, and health care outcome variables (mortality rate, drug costs, total cost of care, and length of stay). This multiple regression analysis is the most important of our analysis models, as it adjusted for severity of illness and considered the collective effects of demographic, pharmacy, staffing, and health care outcome variables on medication errors.

Statistical Analysis

All multiple regression models used a severity of illness-adjusted model. The severity of illness variables were forced into the multiple regression model before any other variables were allowed to enter. Stepwise procedures were used to select the remaining variables for the model.[42,43,44]

The variables selected through this method were confirmed by the use of both forward and backward regression techniques. Both techniques selected the same set of variables. The slope measures the rate of change for the variable and is expressed as either positive (e.g., as the medication error rate increased, the mortality rate increased) or negative (e.g., as the number of medical residents/occupied bed increased, the medication error rate decreased). A high slope indicates that changes in the specified variable were associated with significant changes in the other variable (e.g., decentralized location of pharmacists). The multiple regression analysis allowed us to determine the direct relationships and associations between medication errors and mortality rates, drug costs, total cost of care, and length of stay. We also determined the direct relationships and associations between medication errors and hospital demographic, pharmacy, staffing, and health care outcome variables.

Statistical analysis was employed for both the total number of medication errors and the number of medication errors that adversely affected patient outcomes/hospital. Statistical tests employed were the t test, analysis of variance (ANOVA), simple regression, and multiple regression. The a priori level of significance for all tests was 0.05.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.