In this study of 4 adult primary care practices, we found prescribing errors in about 1 of 13 prescriptions and nearly half of these errors had potential for harm. Certain classes of medications, such as antibiotics and nonsteroidal anti-inflammatory agents, were commonly implicated in prescribing errors. Basic computerized prescribing was not associated with a reduced rate of prescribing errors or potential ADEs. However, these rates could have been substantially reduced with more advanced decision support such as dose and frequency checking.
Our medication prescribing error rate of 7.6 per 100 prescriptions is substantially higher than inpatient rates of 0.4 to 5 per 100 orders found in studies using similar detection methods.1
This difference may arise in part because outpatient prescriptions typically include more parameters (e.g., number to be dispensed, number of refills) and outpatient providers may work under more intense time pressure. Almost half of prescribing errors in this study were judged to have the potential for injury. This proportion is substantially higher than in the inpatient setting where estimates of medication errors classified as potential ADEs have ranged from 7% to 33%.1,15
The potential for injury may be greater because this study focuses on prescribing errors whereas inpatient studies typically include transcribing, dispensing and administration errors that may be common but less serious (e.g., missing dose errors). The difference also may reflect the lack of resources (feedback from pharmacists) and lack of easy patient monitoring in the outpatient setting. Moreover, since more classes and brands of drugs are available in ambulatory care, providers may be less familiar with correct prescribing parameters.
Fewer outpatient potential ADEs were rated life-threatening or serious compared with an inpatient study (26% vs 55%).14
However, because of the high rate of potential ADEs in the outpatient setting (3% of all prescriptions), the percentage of prescriptions with serious potential ADEs is higher (8 per 1000 prescription). Since 3 billion prescriptions are written annually in the U.S.,16
this extrapolates to 24 million serious potential events in the U.S., if this rate is generalizable. Finally, 4% of patients with prescribing errors experienced a preventable ADE. Although a small number of prescriptions led to actual harm, the large number of prescriptions that have potential to cause serious harm clearly indicates that new error prevention strategies must be introduced into the current prescribing process.
Dose and frequency errors occurred often, paralleling the inpatient experience.1
Certain medication classes were more likely to be involved, in particular antibiotics, narcotics, and nonsteroidal anti-inflammatory medications, likely because they are also commonly prescribed. Thus, these medications are good targets for improved prescribing decision support or physician education about appropriate prescribing. Although half of prescribing errors had no potential for harm, these too reveal systems failures. The same defect that leads to an error without injury because the drug has a broad therapeutic window (e.g., a 3-fold overdose of a stool softener) can lead to a serious injury if the drug is potentially toxic (e.g., a 3-fold overdose of a β blocker). Therefore, we believe it is important to document these errors, as they may indicate a more systemic problem.17
In 2 study sites, computerization of prescribing included printed prescriptions and required fields (drug, dose, quantity, and duration), some default doses and no or only optional checks for allergies and drug interactions. This basic type of electronic prescribing system did not reduce rates of medication prescribing errors. However, our data suggest that computerized prescribing with more advanced decision-support will be vital to reduce the rate of potentially harmful errors. The vast majority of potential ADEs found in the study were judged to be preventable with more advanced decision support. This was true in the inpatient experience, in which basic computerized physician order entry had a smaller effect on serious errors than on medication errors overall.10,11
Advanced decision support that requires complete prescriptions and provides default dosing and frequencies appeared particularly important. In addition, 2 of the 3 preventable ADEs in our study could have been prevented with default doses and frequencies. Because default doses and frequencies can be readily incorporated into most commercial systems, this finding is encouraging.
Computerization of prescribing has been widely touted, but is not yet widely used.18
Only 2% of all prescriptions written by 650,000 U.S. physicians in 1999 were written electronically with any device19
; in contrast two-thirds of general practitioners in Britain in 1993 used electronic prescribing.20
However, there is momentum for change. A 2002 poll by Harris Interactive showed that 82% of patients preferred an electronic prescribing system.21
In addition, legislation currently pending in Congress (Senate bill S.1 and House bill H.R.1) would mandate computerized prescribing. However, the question remains of what features should be included as part of this mandate.
The level of decision support among commercially available electronic prescribing systems varies substantially, and the use of decision support is often optional.12,22
Our 2 practices with basic computerized prescribing had many prescribing errors that could have been prevented with more mandatory safeguards in place. Consequently, as practices implement computerized prescribing, defaults and at least some checks and alerts should be mandatory rather than optional. Furthermore, physicians should be informed about the purposes of such safeguards and the potential dangers of overriding them.
Our study has several limitations. It was conducted for a limited duration in only 4 urban primary care practices including many physicians with part-time practices, so the results may not be generalizable. Our study was not powered to detect modest differences in potential ADE rates between computerized and handwritten sites. In addition, physicians were not blinded to the purpose of the study and might have been particularly careful when prescribing or may have excluded patients they knew to be at high risk. If so, our findings may represent only conservative estimates of the true rates. This study focuses on prescribing errors and was not designed to detect dispensing or patient adherence errors. Future studies should investigate these subsequent steps in the outpatient medication use process. Finally, we used implicit review for our classifications of prescribing errors; however, our reliability for these classifications was good.
Prescribing errors are common in ambulatory care, and often have potential to harm patients. Office practices and health systems are beginning to develop or purchase computerized prescribing systems. Basic computerized prescribing did not result in fewer errors compared with handwritten prescribing. Therefore, to achieve a major safety benefit, computerized prescribing with advanced decision support will likely be needed.