The Comparative Effectiveness of 2 Electronic Prescribing Systems | Page 2
Published Online: December 16, 2011
Rainu Kaushal, MD, MPH; Yolanda Barron, MS; and Erika L. Abramson, MD, MS
Prescription Collection. We collected carbon copies at baseline and electronic downloads of all prescriptions written by providers during a 2-week period at 1 year. During both time periods, we obtained a minimum of 75 prescriptions on at least 25 patients per provider, extending data collection if necessary, and limited review to 3 prescriptions per patient to minimize clustering of errors. Non-duplicate prescription pads were removed at baseline to ensure compliance with use of the duplicate prescription pads.
Prescription Review. A research nurse reviewed each prescription in an identical manner guided by extensively utilized, standardized methodology.9-11 Training included review of error definitions, legibility assessments, and review of test and actual cases. Two data collectors jointly reviewed cases, after which reviewers worked independently. Reviewers classified prescribing errors, and rule violations, and evaluated the use of ADE-trigger drugs. Inappropriate abbreviation errors were from the Joint Commission on Accreditation of Healthcare Organizations’ “Do Not Use” list, established to denote abbreviations with great potential to cause medical errors.12
We determined interrater reliability by having 2 reviewers examine the same random sample of 2% of the data. Interrater agreement for overall and type of error was 1.0, indicating excellent agreement.
Chart Review. An ambulatory chart review was performed for suspected near misses or when a drug often used to treat an ADE was prescribed.
Physician Review and Classification. Two physicians blinded to the providers’ prescribing method independently reviewed all suspected near misses and ADEs. Interrater agreement for the presence of prescribing errors and near misses was 0.96 and 0.93, indicating excellent agreement.
Statistical Analysis. We compared error rates per 100 prescriptions of (1) stand-alone and integrated adopters at baseline, (2) stand-alone adopters at baseline and 1 year, (3) integrated adopters at baseline and 1 year, and (4) between stand-alone and integrated adopters at 1 year using mixedeffects Poisson regression models that included e-prescribing system type, study time, and an interaction term system type and study time. We adjusted for clustering at the provider level and assumed an independent correlation structure for these Poisson models. We calculated 95% Poisson confidence intervals (CIs) with cluster robust standard errors for the rates. We used SAS for PC version 9.2 (SAS Institute Inc, Cary, North Carolina) to estimate Kappa statistics, χ2, and t tests, and Stata 11 (StataCorp, College Station, Texas) to
estimate mixed-effects Poisson regression models and to calculate 95% Poisson CIs with clustered robust standard errors.
There was no significant difference between provider groups (Table 1). All physicians were board certified in internal or family medicine. The mean number of prescriptions written per provider and patients seen per provider were similar at baseline (0.65 and 0.26, respectively). At 1 year, integrated adopters saw significantly more patients and wrote significantly more prescriptions during the study period (0.002 and 0.002, respectively).
For stand-alone adopters, 1273 unique patients seen at baseline and 1598 unique patients seen at 1 year received prescriptions (Table 2). For integrated adopters, 481 unique patients seen at baseline and 368 unique patients seen at 1 year received prescriptions. At baseline, patients of stand-alone adopters were significantly older (56 vs 53 years, P <.001) and at 1 year more were female (60% vs 54%, P = .03).
Baseline Error Rates
We reviewed 2550 prescriptions at baseline, 1901 of which were written by stand-alone adopters and 649 of which were from integrated adopters (Tables 3A and 3B).
Rates of baseline prescribing errors for both groups were high, with planned adopters of integrated e-prescribing 1.6- fold less likely to have prescribing errors (P = .03). Rule violations were also high at baseline for both groups, but with no statistical differences. Near misses were detected infrequently and without statistical differences.
Rates of Errors for Stand-alone E-prescribing Adopters at Baseline and 1 Year
We reviewed 2305 prescriptions at 1 year for stand-alone adopters. Stand-alone e-prescribing reduced prescribing error rates by 6.7-fold (P <.001). Rule violations also decreased significantly (P <.001). Near misses remained low at 1 year and unchanged from baseline.
Rates of Errors for Integrated E-prescribing Adopters at Baseline and 1 Year
We reviewed 536 prescriptions at 1 year for integrated e-prescribing adopters. Use of an integrated system reduced prescribing error rates by 1.6-fold, suggesting a trend for decreasing prescribing errors but without achieving statistical significance. Rule violations significantly decreased (P <.001). Rates of near misses remained low and unchanged.
Rates of Errors for Stand-alone Versus Integrated E-prescribing Adopters at 1 Year
Stand-alone adopters were 2.45 times less likely to have prescribing errors than integrated adopters at 1 year (P <.001). After accounting for differences in baseline error rates, standalone users had a 4-fold lower rate of errors at 1 year (P <.001). There was no significant difference in rates of rule violations (P = .16) or near misses (P = .94) at 1 year.
Types of Prescribing Errors at Baseline and 1 Year
Use of the stand-alone system reduced all types of prescribing errors, while use of the integrated system led to a decrease in only some error types (Tables 4A and 4B).
PDF is available on the last page.