Can an Algorithm for Appropriate Prescribing Predict Adverse Drug Events?

Published Online: March 01, 2005
Kimberly J. Rask, MD, PhD; Kristen J. Wells, MPH; Gregg S. Teitel, PharmD; Jonathan N. Hawley, BS; Calita Richards, PharmD; and Julie A. Gazmararian, MPH, PhD

Objective: To evaluate whether a medication-appropriateness algorithm applied to pharmacy claims data can identify ambulatory patients at risk for experiencing adverse drug events (ADEs) from those medications.

Study Design: Cohort study.

Methods: We surveyed a random sample of 211 community-dwelling Medicare managed care enrollees over age 65 years who were identified by pharmacy claims as taking a potentially contraindicated medication (exposed enrollees) and a random sample of 195 enrollees who were identified as not taking such a medication (unexposed enrollees). The primary outcome of interest was the prevalence of self-reported events in previous 6 months.

Results: Ninety-nine (24.4% of total sample) respondents reported a total of 134 ADEs during the previous 6 months. Exposed enrollees had a significantly higher number of chronic conditions and were taking more prescription and nonprescription medications. However, the higher rate of self-reported ADEs among exposed enrollees was not statistically significant from that of unexposed enrollees (prevalence odds ratio = 1.42; 95% confidence interval [CI] = 0.90, 2.25). Only 1.5% (2/134) of the selfreported ADEs were attributed to a medication from the potentially contraindicated list. Instead, most ADEs were attributed to medications that are commonly used in older patients, including cardiovascular agents (21.6%), anti-inflammatory agents (12.2%), and cholesterol-lowering agents (7.9%).

Conclusions: A medication-appropriateness algorithm using pharmacy claims data was not able to identify a subgroup of enrollees at higher risk of experiencing an ADE from those medications. The vast majority of ADEs were attributable to commonly prescribed medications.

(Am J Manag Care. 2005;11:145-151)

Adverse drug events (ADEs) are an important cause of morbidity and hospitalizations among the elderly.1,2 Among the 1.6 million US residents of nursing homes, drug-related injuries are estimated to occur at a rate of 350 000 events per year, and more than half may be preventable.3 Using a multimodal assessment of medical records, the prevalence of ADEs in community-dwelling elderly persons was recently estimated to be 50.1 per 1000 person-years, with 27.6% of events considered to be preventable.4 A survey of adult primary care patients found a much higher self-reported ADE rate of 27 per 100 patients.5 The elderly are considered to be at higher risk for ADEs due to prevalent chronic health conditions, age-related physiologic changes, social isolation, and polypharmacy.6-8 Although not all ADEs are the result of medication error, inappropriate prescribing is a preventable cause of ADEs. In 1991, Beers et al published explicit criteria for inappropriate medication use in institutionalized elderly patients, and the criteria were expanded in 1997.9 These and several other criteria have been used extensively in drug utilization review as well as population-based studies to measure the prevalence of inappropriate prescribing.10-13 A population survey of older Americans found that 28.7% of those prescribed any medication received at least 1 potentially inappropriate drug.11 When dose and length of therapy were considered, the incidence of inappropriate prescribing rose to 40%.

The objective of the current study is to evaluate whether a medication algorithm based upon inappropriate- prescribing criteria can be used along with managed care pharmacy claims to identify a subset of patients at high risk of suffering an ADE. Administrative-claim algorithms have the potential to be an inexpensive method of identifying patients at risk for ADEs. The study hypothesis is that patients identified as taking a potentially inappropriate medication have a higher likelihood of subsequently suffering a self-reported ADE related to those medications.



Historically, Aetna has provided performance reports to physicians that assess pharmaceutical use consistent with nationally accepted guidelines of care. These performance reports were published annually and supported by administrative-data algorithms. The medications selected for inclusion in the algorithm were taken from medications included in 1 of 2 published consensus recommendations and deemed to be either ineffective or more toxic than equally effective alternatives (see Appendix).9,12

Because the algorithm had been adapted from 2 published sets of criteria, secondary analyses using those 2 sets of criteria in their original forms also were performed.9,12 The McLeod list was developed by using a modified Delphi approach to arrive at consensus recommendations by an expert panel of clinical pharmacologists, geriatricians, family practitioners, and pharmacists.12 The McLeod criteria include a broad range of barbiturates and nonsteroidal anti-inflammatory agents. The Beers list was created by a second group of nationally recognized experts in geriatric care and pharmacology who also used a modified Delphi approach to update consensus recommendations.9 The Beers criteria include a broad range of potentially inappropriate medications, including muscle relaxants and antispasmodic agents and, in addition, take into account the prescribed strength. For example, selected benzodiazepines are considered appropriate for the elderly population when prescribed at a lower daily dose and inappropriate at a higher daily dose.


Study participants were managed care Medicare plan members continuously enrolled between July 1, 1999, and June 30, 2000. All participants lived in southeastern Pennsylvania and were aged 65 years or older at the time of enrollment. Pharmacy claims data were used to identify a subgroup of enrollees over age 65 who had a prescription filled for at least 1 of the potentially contraindicated medications during the previous 6 months (exposed enrollees). We identified a second subgroup who according to pharmacy claims were not taking any of those medications (unexposed enrollees). These 2 subgroups were randomly sampled to assemble the study subjects. A power analysis indicated that a sample size of 184 exposed and 184 unexposed enrollees would allow the detection of a difference of 0.15 with a 1-tailed αof .025 and βof .10.14

Potential participants received an introduction letter via the mail, and a telephone contact was attempted within 2 weeks of the mailing. A single interviewer who was blinded to participants'study group assignment attempted contact both on weekdays and weekends. Contact was attempted a minimum of 6 times before an enrollee was designated as "unable to contact."When contact was established with the potential participants, the interviewer explained the purpose of the study and obtained oral consent to participate. Enrollees were eligible to participate in the interview if they spoke English or had a translator. Enrollees were excluded from participation if they had a diagnosis of cancer, were hearing impaired, or had an inaccurate telephone number. Both the Emory University Human Investigations Committee and the USQA Center for Health Care Research institutional review board approved the study design.

Data Collection and Analysis

The structured interview consisted of 42 questions regarding general health status; prescription drug use; adherence to medical regimen; source of medication information; ADEs; number of healthcare providers; vitamin, mineral, and herbal supplement use; other over-the-counter medication use; and demographic characteristics. The survey was pilot-tested with 29 participants and subsequently revised. The revised survey had a mean administration time of 16 minutes (range: 4-44 minutes). The interviewer recorded answers on a response sheet and later entered them into an SPSS database.15 A second member of the research team verified the data entry for accuracy.

The outcome examined for the purpose of this study was prevalence of self-reported ADEs. The interviewer initially asked participants to name all of the prescription medications that they currently were taking. Subsequently, ADEs were measured using an adapted version of a previously published survey.16,17 Five questions were asked: (1) whether the participant had experienced an ADE within the last 6 months, (2) which drugs were involved in the ADE, (3) whether a doctor had been notified about the ADE, (4) what changes were made to the treatment regimen because of the ADE, and (5) whether hospitalization was required because of the ADE. All ADEs reported by the study participants were included. A second member of the research team, who was blinded to study group assignment, evaluated all medications reported by the respondents to determine whether the medications were contraindicated according to the algorithm used to identify the cases.

The tested study hypothesis was that the proportion of exposed enrollees reporting an ADE would be higher than the proportion of unexposed enrollees reporting an ADE. This hypothesis was tested by using chi-square tests of categorical data. The prevalence odds ratio of an ADE among exposed enrollees also was calculated. All analyses were conducted using SPSS.15


Subject Recruitment

Letters were sent to a total of 719 enrollees (Figure). The telephone interviewer was unable to reach 138 (19.2%) of the participants, 12 (1.7%) participants were deceased, 23 (3.2%) were too infirm to participate, 23 (3.2%) were unable to communicate by telephone, and 117 (16.3%) refused to participate. Thus, the final sample for this study consisted of 406 out of 719 participants (56.5%), 211 exposed and 195 unexposed. Exposed enrollees were less likely to participate in the survey, 211/397 (53.1%) versus 195/322 (60.6%). The characteristics of participants and nonparticipants are shown in Table 1. Nonparticipants were older and more likely to be male. Physician and hospital utilization rates, however, were similar for participants and nonparticipants.



Demographic Characteristics

PDF is available on the last page.


Recommended Reading

No Result Found