- Center on Health Equity and Access
- Clinical
- Health Care Cost
- Health Care Delivery
- Insurance
- Policy
- Technology
- Value-Based Care
Can an Algorithm for Appropriate Prescribing Predict Adverse Drug Events?
Objective: To evaluate whether a medication-appropriatenessalgorithm applied to pharmacy claims data can identify ambulatorypatients at risk for experiencing adverse drug events (ADEs) fromthose medications.
Study Design: Cohort study.
Methods: We surveyed a random sample of 211 community-dwellingMedicare managed care enrollees over age 65 years whowere identified by pharmacy claims as taking a potentially contraindicatedmedication (exposed enrollees) and a random sampleof 195 enrollees who were identified as not taking such a medication(unexposed enrollees). The primary outcome of interest wasthe prevalence of self-reported events in previous 6 months.
Results: Ninety-nine (24.4% of total sample) respondentsreported a total of 134 ADEs during the previous 6 months.Exposed enrollees had a significantly higher number of chronicconditions and were taking more prescription and nonprescriptionmedications. However, the higher rate of self-reported ADEsamong exposed enrollees was not statistically significant from thatof unexposed enrollees (prevalence odds ratio = 1.42; 95% confidenceinterval [CI] = 0.90, 2.25). Only 1.5% (2/134) of the selfreportedADEs were attributed to a medication from the potentiallycontraindicated list. Instead, most ADEs were attributed to medicationsthat are commonly used in older patients, including cardiovascularagents (21.6%), anti-inflammatory agents (12.2%), andcholesterol-lowering agents (7.9%).
Conclusions: A medication-appropriateness algorithm usingpharmacy claims data was not able to identify a subgroup ofenrollees at higher risk of experiencing an ADE from those medications.The vast majority of ADEs were attributable to commonlyprescribed medications.
(Am J Manag Care. 2005;11:145-151)
Adverse drug events (ADEs) are an importantcause of morbidity and hospitalizations amongthe elderly.1,2 Among the 1.6 million US residentsof nursing homes, drug-related injuries are estimatedto occur at a rate of 350 000 events per year, andmore than half may be preventable.3 Using a multimodalassessment of medical records, the prevalence of ADEsin community-dwelling elderly persons was recentlyestimated to be 50.1 per 1000 person-years, with 27.6%of events considered to be preventable.4 A survey ofadult primary care patients found a much higher self-reportedADE rate of 27 per 100 patients.5 The elderlyare considered to be at higher risk for ADEs due toprevalent chronic health conditions, age-related physiologicchanges, social isolation, and polypharmacy.6-8Although not all ADEs are the result of medication error,inappropriate prescribing is a preventable cause ofADEs. In 1991, Beers et al published explicit criteria forinappropriate medication use in institutionalized elderlypatients, and the criteria were expanded in 1997.9 Theseand several other criteria have been used extensively indrug utilization review as well as population-basedstudies to measure the prevalence of inappropriate prescribing.10-13 A population survey of older Americansfound that 28.7% of those prescribed any medicationreceived at least 1 potentially inappropriate drug.11When dose and length of therapy were considered, theincidence of inappropriate prescribing rose to 40%.
The objective of the current study is to evaluatewhether a medication algorithm based upon inappropriate-prescribing criteria can be used along with managedcare pharmacy claims to identify a subset of patients athigh risk of suffering an ADE. Administrative-claimalgorithms have the potential to be an inexpensivemethod of identifying patients at risk for ADEs. Thestudy hypothesis is that patients identified as taking apotentially inappropriate medication have a higher likelihoodof subsequently suffering a self-reported ADErelated to those medications.
METHODS
Algorithms
Historically, Aetna has provided performance reportsto physicians that assess pharmaceutical use consistentwith nationally accepted guidelines of care. These performancereports were published annually and supported by administrative-data algorithms. The medicationsselected for inclusion in the algorithm were taken frommedications included in 1 of 2 published consensus recommendationsand deemed to be either ineffective ormore toxic than equally effective alternatives (seeAppendix).9,12
Because the algorithm had been adapted from 2 publishedsets of criteria, secondary analyses using those 2sets of criteria in their original forms also were performed.9,12 The McLeod list was developed by using amodified Delphi approach to arrive at consensus recommendationsby an expert panel of clinical pharmacologists,geriatricians, family practitioners, andpharmacists.12 The McLeod criteria include a broadrange of barbiturates and nonsteroidal anti-inflammatoryagents. The Beers list was created by a second groupof nationally recognized experts in geriatric care andpharmacology who also used a modified Delphi approachto update consensus recommendations.9 The Beers criteriainclude a broad range of potentially inappropriatemedications, including muscle relaxants and antispasmodicagents and, in addition, take into account the prescribedstrength. For example, selected benzodiazepinesare considered appropriate for the elderly populationwhen prescribed at a lower daily dose and inappropriateat a higher daily dose.
Participants
Study participants were managed care Medicare planmembers continuously enrolled between July 1, 1999,and June 30, 2000. All participants lived in southeasternPennsylvania and were aged 65 years or older at thetime of enrollment. Pharmacy claims data were used toidentify a subgroup of enrollees over age 65 who had aprescription filled for at least 1 of the potentially contraindicatedmedications during the previous 6 months(exposed enrollees). We identified a second subgroupwho according to pharmacy claims were not taking anyof those medications (unexposed enrollees). These 2subgroups were randomly sampled to assemble thestudy subjects. A power analysis indicated that a samplesize of 184 exposed and 184 unexposed enrollees wouldallow the detection of a difference of 0.15 with a 1-tailedαof .025 and βof .10.14
Potential participants received an introduction lettervia the mail, and a telephone contact was attemptedwithin 2 weeks of the mailing. A single interviewer whowas blinded to participants'study group assignmentattempted contact both on weekdays and weekends.Contact was attempted a minimum of 6 times before anenrollee was designated as "unable to contact."Whencontact was established with the potential participants,the interviewer explained the purpose of the study andobtained oral consent to participate. Enrollees were eligibleto participate in the interview if they spoke Englishor had a translator. Enrollees were excluded from participationif they had a diagnosis of cancer, were hearingimpaired, or had an inaccurate telephone number. Boththe Emory University Human Investigations Committeeand the USQA Center for Health Care Research institutionalreview board approved the study design.
Data Collection and Analysis
The structured interview consisted of 42 questionsregarding general health status; prescription drug use;adherence to medical regimen; source of medicationinformation; ADEs; number of healthcare providers;vitamin, mineral, and herbal supplement use; otherover-the-counter medication use; and demographiccharacteristics. The survey was pilot-tested with 29 participantsand subsequently revised. The revised surveyhad a mean administration time of 16 minutes (range:4-44 minutes). The interviewer recorded answers on aresponse sheet and later entered them into an SPSSdatabase.15 A second member of the research team verifiedthe data entry for accuracy.
The outcome examined for the purpose of this studywas prevalence of self-reported ADEs. The interviewerinitially asked participants to name all of the prescriptionmedications that they currently were taking.Subsequently, ADEs were measured using an adaptedversion of a previously published survey.16,17 Five questionswere asked: (1) whether the participant had experiencedan ADE within the last 6 months, (2) whichdrugs were involved in the ADE, (3) whether a doctorhad been notified about the ADE, (4) what changes weremade to the treatment regimen because of the ADE, and(5) whether hospitalization was required because of theADE. All ADEs reported by the study participants wereincluded. A second member of the research team, whowas blinded to study group assignment, evaluated allmedications reported by the respondents to determinewhether the medications were contraindicated accordingto the algorithm used to identify the cases.
The tested study hypothesis was that the proportionof exposed enrollees reporting an ADE would be higherthan the proportion of unexposed enrollees reporting anADE. This hypothesis was tested by using chi-squaretests of categorical data. The prevalence odds ratio of anADE among exposed enrollees also was calculated. Allanalyses were conducted using SPSS.15
RESULTS
Subject Recruitment
Letters were sent to a total of 719 enrollees (Figure).The telephone interviewer wasunable to reach 138 (19.2%) ofthe participants, 12 (1.7%) participantswere deceased, 23(3.2%) were too infirm to participate,23 (3.2%) were unable tocommunicate by telephone, and117 (16.3%) refused to participate.Thus, the final sample forthis study consisted of 406 outof 719 participants (56.5%), 211exposed and 195 unexposed.Exposed enrollees were lesslikely to participate in the survey,211/397 (53.1%) versus195/322 (60.6%). The characteristicsof participants andnonparticipants are shown inTable 1. Nonparticipants were older and more likely tobe male. Physician and hospital utilization rates, however,were similar for participants and nonparticipants.
Demographic Characteristics
P
Demographic characteristics of the study subjectsare shown in Table 2. The mean age, race, marital status,educational attainment, and home ownership weresimilar in both groups. Women were more likely to betaking a potentially contraindicated medication. Thevast majority of respondents (93.6%) were taking at least1 prescription medication. However, the mean numberof prescription medications taken was significantlyhigher for exposed enrollees (4.7 vs 3.3, <.001). Theprevalence of over-the-counter medication use also washigh in both groups, with exposed enrollees taking significantlymore over-the-counter medications. The distributionof chronic medical conditions also wassignificantly different between the 2 groups. Exposedenrollees were significantly more likely to report havinghypertension, seasonal allergies, back pain, depression,anxiety, insomnia, or gastrointestinal symptoms.
Active use of potentially contraindicated medicationsis shown in Table 3. Thirty-three percent (70/211) ofexposed enrollees and 1.0% (2/195) of unexposedenrollees were taking a potentially contraindicated medicationat the time of the telephone interview.
Prevalence of Adverse Drug Events
P
P
Overall, 99 respondents reported having experiencedat least 1 ADE in the previous 6 months. Exposedenrollees reported more ADEs, but the difference wasnot statistically significant (27.5% vs 21.0%, = .13).The prevalence odds ratio of reporting an ADE was 1.42(95% confidence interval [CI] = 0.90, 2.25). Enrolleeswho were still taking a potentially contraindicatedmedication at the time of the telephone interview weremore likely to report having had an ADE, but this differencealso did not reach statistical significance (31.9%vs 22.8%, = .10).
P
Irrespective of study group assignment, enrolleesreporting an ADE were taking an average of 1 moreprescription medication (4.7 vs 3.8, = .05) thanenrollees not reporting an ADE. Sixty-two of 201(30.8%) enrollees taking 4 or more medicationsreported an ADE, compared with only 36 of 205(17.6%) enrollees taking 3 or fewer medications. Thealgorithm did identify a group of enrollees takingmore medications and with more chronic conditions;however, this group still did not have a statisticallyhigher rate of ADEs.
Published Prescribing Criteria and Adverse DrugEvents
P
P
To evaluate the potential effectiveness of the publishedappropriate-prescribing criteria, active prescriptionmedications were compared with the Beers list9 andthe McLeod list.12 Unlike the case for the managed carealgorithm, enrollees currently taking a medication on theBeers list were significantly more likely to report an ADE(35.0% vs 20.9%, = .004). Enrollees taking a medicationon the McLeod list also were significantly more likely toreport an ADE than those not taking a medication on thelist (35.1% vs 22.6%, = .042).
Medications Linked toAdverse Drug Events
The medications most commonlyreported by enrollees tohave resulted in an ADE areshown in Table 4. The mostcommon category was cardiovascularagents, followed byanti-inflammatory agents. Anxiolyticsand antidepressants,medications commonly cited inpublished inappropriate-prescribinglists, were infrequentlyreported by patients as havingcaused an ADE. Only 2 of themedications associated with aself-reported ADE were on thehealth plan's list of potentiallycontraindicated medications.Both were benzodiazepines, andthey accounted for a total of 4reported events. A total of 99enrollees were taking a benzodiazepine.Forty (40%) were takingit on an as-needed basis and59 (60%) were taking it at leastdaily. Only 6 medications fromthe Beers list of contraindicatedmedications were reported tohave caused an ADE, accountingfor only 6.0% (8/134) of thereported ADEs. Three medicationsfrom the McLeod list werereported to have caused an ADE,accounting for only 3.0% (4/134)of the reported ADEs.
Severity of Adverse DrugEvents
The most common enrollee-reportedsymptoms are listed inTable 5. A total of 171 symptomswere reported from the 134 ADEs.Gastrointestinal symptoms, fatigue,and dizziness were most common. As a gauge ofsymptom severity, most episodes (78.9%) werereported to the enrollee's physician, and thephysician made a change in prescription medicationsfor the majority (65.7%) of events. FourADEs (3.0%) resulted in a hospitalization. The 4patients who reported requiring hospitalizationattributed their reaction to 1 of the followingmedications: verapamil, citalopram, diazepam,or metoprolol. One of the medications, diazepam,is a contraindicated medication according to all 3lists. The other 3 medications are not on any ofthe criteria lists.
DISCUSSION
Adverse drug events are common, with onefourth of the study population reporting an ADEin the previous 6 months. An algorithm usingmanaged care pharmacy claims data to identifypotentially inappropriate prescribing practiceswas not able to identify a subgroup with a higherprevalence of self-reported ADEs from thosemedications. In fact, the vast majority of ADEswere attributed to commonly prescribed medications.Patients taking a medication from theBeers9 or McLeod12 lists, which include a broaderrange of medications, were significantly morelikely to report having suffered an ADE.However, there was still very little overlapbetween the medications reported by enrolleesto have caused an ADE and the medications listedas potentially inappropriate according to anyof the published lists.
Although the appropriate-prescribing algorithmdid not identify a significantly highernumber of ADEs, the algorithm did identify asubset of patients who were taking significantly moremedications, both prescription and nonprescription.This subgroup also had a higher prevalence of chronicmedical conditions, including anxiety and depression.Given the substantial number of enrollees taking benzodiazepines,it is surprising that so few ADEs wereattributed to that class of medication. Polypharmacy isa risk factor for poor adherence as well as medicationinteractions.18 The higher rate of ADEs among theexposed enrollees, although not statistically significant,may have been due to polypharmacy rather than thespecific use of contraindicated medications.Interventions to reduce polypharmacy, however, havehad variable long-term success.19-21
Our findings suggest that there are limitations to theuse of inappropriate-prescribing algorithms to reducethe prevalence of ADEs. First, many of the medicationson the Beers and McLeod lists are no longer widelyused.10,13 Second, medications that are frequently implicatedin preventable events (eg, warfarin) are not onthese "bad drug"lists.3 Few medications are absolutelycontraindicated in older patients, making these lists extremelynarrow. Algorithms have content validity andcan identify questionable prescribing, but their predictivevalidity remains uncertain.22 The Beers criteria haverecently been updated to include medications thatshould generally be avoided in persons aged 65 years orolder and medications that should not be used in olderpersons known to have specific medical conditions.23Further research is needed to test this revised tool inolder ambulatory populations.
Medications commonly and appropriately prescribedin the ambulatory setting were blamed for the vastmajority of ADEs. This confirms other recentfindings.4,5,24 Cardiovascular agents are commonlylinked to ADEs, in this study representingalmost one third of the events. Renal orelectrolyte drug events were less frequent thanpreviously reported, probably because our studyrelied on patient self-report while the previousstudy used laboratory and medical recordreview.4 A study of recently hospitalizedpatients reported that antibiotics were the mostcommon drug type linked to an ADE.24 In ourcommunity-dwelling elderly, antibiotics werelinked to only 5% of ADEs. The overall rate ofADEs identified in this study was very similar tothe rate found in an adult ambulatory populationthat also included self-reported data5 andmuch higher than the rate found in a study thatreviewed medical records but gathered no informationdirectly from patients.4
These results are subject to several limitations.First, the validity and reproducibility ofpatient self-report of ADEs have been poorlystudied. Drugs can lead to asymptomatic laboratoryabnormalities, for example, that a patientmay never be aware of. At the same time,attributing symptoms to specific drugs is problematic.Nonspecific symptoms are common among outpatientsand may not be related to the specific drugidentified by the patient. Another difficulty is distinguishingthe expected effects of a drug from an adverseevent. To the extent that patients overidentify drug-relatedsymptoms and expected drug events, this studyoverestimates the ADE rate. Patients and the interviewerwere blinded as to study group assignment, so thereshould not have been a differential effect. Both providerreport and medical chart review have been shown tosubstantially underestimate adverse event rates, and thelimitations of implicit chart reviews have been welldescribed.25-28 Most episodes were considered seriousenough by the patient to be reported to a physician, andin two thirds of cases, the event led to a change in treatment.The study population was Medicare managed careenrollees who were disproportionately well educated andCaucasian and may not be typical of older Americans. Asnoted above, though, the rate of ADEs was very similarto that found in another study of adult primary carepatients.5
The initial prescribing event used to identify enrolleesin this study was ascertained from pharmacy claims submittedto the health plan for reimbursement. Only onethird of the enrollees identified as taking a potentiallycontraindicated medication by pharmacy claims werestill taking that medication at the time of the telephonesurvey. The time delay between initial prescription anddata evaluation limits the use of claims data as anadverse event-screening tool. More focused algorithmsof appropriate prescribing for patients with selectedchronic conditions could address some of these limitations.Claims-based algorithms can identify optimal prescribingstandards for patients with specific chronicconditions (eg, use of steroid inhalers in asthmapatients). Pharmacy claims also can be used to screenfor polypharmacy or to identify potentially harmfuldrug-drug interactions.20 Algorithms also can be used inreal time to alert pharmacists to potential drug-druginteractions for new prescriptions. These tools have thepotential to fundamentally change the prescribingprocess for both inpatient and ambulatory medical care,but careful evaluation is needed to ensure that theseinterventions do in fact improve health outcomes.
CONCLUSION
In this study, an appropriate-prescribing algorithmapplied to pharmacy claims data was not helpful foridentifying a high-risk group of elderly patients. Nearlyone fourth of study participants reported an ADE in theprevious 6 months, and the vast majority of reportedevents were adverse reactions to commonly used medications.More work is needed to develop clinically usefulalgorithms that can be applied to older ambulatorypopulations.
Acknowledgments
The authors gratefully acknowledge the contributions of Richard W.Kobylinski and Natalia Oster, RN, MPH, toward the completion of this study.
From the Department of Health Policy and Management, Rollins School of PublicHealth (KJR, JNH, JAG), the Division of General Medicine, Emory University School ofMedicine (KJR), and the Emory Center on Health Outcomes and Quality, formerly theUSQA Center for Health Care Research (KJR, JNH, JAG), Emory University, Atlanta, Ga; theDepartment of Psychology, University of South Florida, Tampa, Fla (KJW); U.S. QualityAlgorithms, Inc, Blue Bell, Pa (GST); and the Georgia Division of Public Health, GeorgiaDepartment of Human Resources, Atlanta, Ga (CR). Dr Teitel now is with Merck andCompany, Inc, Rahway, NJ.
This research was supported by a grant from the Aetna Foundation and the QualityCare Research Fund. The project contents are solely the responsibility of the authors and donot necessarily represent the views of Aetna Inc or its affiliates.
Address correspondence to: Kimberly J. Rask, PhD, MPH, Department of Health Policyand Management, Rollins School of Public Health, Emory University, 1518 Clifton Road,NE, Room 636, Atlanta, GA 30322. E-mail: krask@sph.emory.edu.
Postgrad Med
J.
1. Malhotra S, Karan RS, Pandhi P, Jain S. Drug related medical emergencies inthe elderly: role of adverse drug reactions and non-compliance. 2001;77:703-707.
J Chron Dis.
2. Hutchinson TA, Flegel KM, Kramer MS, Leduc DG, Kong HHP. Frequency,severity and risk factors for adverse drug reactions in adult out-patients: a prospectivestudy. 1986;39:533-542.
Arch Intern Med.
3. Gurwitz JH. Improving the quality of medication use in elderly patients: a not-so-simple prescription. 2002;162;1670-1672.
JAMA.
4. Gurwitz JH, Field TS, Harrold LR, et al. Incidence and preventability ofadverse drug events among older persons in the ambulatory setting. 2003;289:1154-1156.
N Engl J Med.
5. Gandhi TK, Weingart SN, Borus J, et al. Adverse drug events in ambulatorycare. 2003;348:1556-1564.
Gerontologist.
6. Lipton HL, Bird JA. The impact of clinical pharmacists' consultations on geriatricpatients' compliance and medical care use: a randomized controlled trial.1994;34:307-315.
J Am Geriatr Soc.
7. Hanlon JT, Schmader KE, Koronkowski MJ, et al. Adverse drug events in high-riskolder outpatients. 1997;45:945-948.
JAMA.
8. Kaufman DW, Kelly JP, Rosenberg L, Anderson TE, Mitchell AA. Recent patternsof medication use in the ambulatory adult population of the United States:the Sloan Survey. 2002;287:337-344.
Arch Intern Med.
9. Beers MH. Explicit criteria for determining potentially inappropriate medicationuse by the elderly: an update. 1997;157:1531-1536.
JAMA.
10. Zhan C, Sangl J, Bierman AS, et al. Potentially inappropriate medication usein the community-dwelling elderly: findings from the National 1996 MedicalExpenditure Panel Survey. 1991;286:2823-2829.
JAMA.
11. Willcox SM, Himmelstein DU, Woolhandler S. Inappropriate drug prescribingfor the community-dwelling elderly. 1994;272:292-296.
Can Med
Assoc J.
12. McLeod PJ, Huang AR, Tamblyn RM, Gayton DC. Defining inappropriatepractices in prescribing for elderly people: a national consensus panel. 1997;156:385-391.
Arch Intern Med.
13. Pitkala KH, Strandberg TE, Tilvis RS. Inappropriate drug prescribing in home-dwelling,elderly patients: a population-based survey. 2002;162:1707-1712.
Designing Clinical Research.
14. Hulley SB, Cummings SR. Baltimore Md:Williams & Wilkins; 1988.
15. SPSS Inc. SPSS for Windows. Release 10.0.5. Chicago: SPSS Inc; 1999.
Ann Intern Med.
16. Chrischilles EA, Segar ET, Wallace RB. Self-reported adverse drug reactionsand related resource use: a study of community-dwelling persons 65 years of ageand older. 1992;117:634-640.
J Gen
Intern Med.
17. Gandhi TK, Burstin HR, Cook EF. Drug complications in outpatients. 2000;15:149-154.
J Women Aging.
18. Perry BA, Turner LW. A prediction model for polypharmacy: are older, educatedwomen more susceptible to an adverse drug event? 2001;13:39-51.
Drugs Aging.
19. Pitkala KH, Strandberg TE, Tilvis RS. Is it possible to reduce polypharmacy inthe elderly? A randomised, controlled trial. 2001;18:143-149.
Am J Manag Care.
20. Fillit HM, Futterman R, Orland BI, et al. Polypharmacy management inMedicare managed care: changes in prescribing by primary care physicians resultingfrom a program promoting medication reviews. 1999;5:587-594.
Am J Med.
21. Hanlon JT, Weinberger M, Samsa GP, et al. A randomized, controlled trial ofa clinical pharmacist intervention to improve inappropriate prescribing in elderlyoutpatients with polypharmacy. 1996;100:428-436.
Med Care.
22. Hanlon JT, Fillenbaum GG, Kuchibhatla M, et al. Impact of inappropriatedrug use on mortality and functional status in representative community dwellingelders. 2002;40:166-176.
Arch Intern Med.
23. Fick DM, Cooper JW, Wade WE, Waller JL, Maclean JR, Beers MH.Updating the Beers criteria for potentially inappropriate medication use in olderadults. 2003;163:2716-2724.
Ann
Intern Med.
24. Forster AJ, Murff HJ, Peterson JF, Gandhi TK, Bates DW. The incidence andseverity of adverse events affecting patients after discharge from the hospital. 2003;138:161-167.
Jt Comm J Qual Improv.
25. Cullen DJ, Bates DW, Small SD, Cooper JB, Nemeskal AR, Leape LL. Theincident reporting system does not detect adverse drug events: a problem for qualityimprovement. 1995;21:541-548.
Ann Intern Med.
26. O'Neil AC, Petersen LA, Cook EF, Bates DW, Lee TH, Brennan TA. Physicianreporting compared with medical-record review to identify adverse medicalevents. 1993;119;370-376.
JAMA.
27. McDonald CJ, Weiner M, Hui SL. Deaths due to medical errors are exaggeratedin Institute of Medicine report. 2000;284:93-95.
JAMA.
28. Hayward RA, Hofer TP. Estimating hospital deaths due to medical errors: preventabilityis in the eye of the reviewer. 2001;286:415-420.
