Objective: To determine the accuracy of computerized medicationhistories.
Study Design: Cross-sectional observational study.
Patients and Methods: The study sample included 493Department of Veterans Affairs primary care patients aged 65 yearsor older who were receiving at least 5 prescriptions. A semistructuredinterview confirmed medication, allergy, and adverse drugreaction (ADR) histories. Accuracy of the computerized medicationlists was assessed, including omissions (medications not on thecomputer record) and commissions (medications on the computerrecord that were no longer being taken). Allergy and ADR recordsalso were assessed.
Results: Patients were taking a mean of 12.4 medications:65% prescription, 23% over-the-counter products, and 12% vitamins/herbals. There was complete agreement between the computermedication list and what the patient was taking for only 5.3%of patients. There were 3.1 drug omissions per patient, and 25% ofthe total number of medications taken by patients were omittedfrom the electronic medical record. There were 1.3 commissionsper patient, and the patients were not taking 12.6% of all activemedications on the computer profile. In addition, 23.2% of allergiesand 63.9% of ADRs were not in the computerized record.
Conclusions: Very few computerized medication histories wereaccurate. Inaccurate medication information may compromisepatient care and limit the utility of medication databases forresearch and for assessment of the quality of prescribing and diseasemanagement.
(Am J Manag Care. 2004;10(part 2):872-877)
Studies have demonstrated that the medication profilein outpatient and inpatient medical chartsoften is inaccurate.1,2 Due to the lack of reliabilityof the medical record as an accurate source of medicationhistory, many hospitals and clinics have begunusing computerized medication profiles, and manygroups and government agencies advocate computerizedmedical records and physician order entry toreduce the incidence of adverse drug events and medicationerrors.3-5 However, little is known about theaccuracy and reliability of computerized medication lists.
In addition, pharmacy benefit management (PBM)databases are increasingly being utilized in clinicalresearch. Information from these large databases hasbeen used to assess compliance and adverse drug eventsin several studies.6-8 In addition to these private insurancedatabases, the Department of Veterans Affairs(VA) has a large pharmacy database used for clinicalresearch. This VA pharmacy database has been utilizedto calculate the Chronic Disease Score (RxRisk-V) toassess the burden of chronic disease on treated populations,9 assess healthcare utilization within the VA system,10 and evaluate prescribing practices.11 To ourknowledge, no studies have evaluated the validity ofclinical data found in computer medical records.
This study was performed to evaluate the agreementbetween information in the VA computerized medicationprofile and information obtained through a structuredmedication history.
The study was conducted at the Iowa City, Iowa VAMedical Center (VAMC) primary care clinics. The IowaCity VAMC is a 100-bed hospital and a primary teachingaffiliate of the University of Iowa Carver College ofMedicine. Sixty internal medical residents, 10 staffphysicians, 4 physician's assistants, and 3 nurse practitionersstaff the primary care clinics.
The patients in this evaluation were aged 65 years andolder, were enrolled in a primary care clinic at the IowaCity VAMC, and had active prescriptions for 5 or moreregularly scheduled nontopical medications. Patients withimpaired cognitive function or enrolled in a pharmacist-basedanticoagulation clinic were excluded. Patients gaveinformed consent, and the institutional review board atthe University of Iowa Carver College of Medicine andIowa City VAMC approved the study protocol.
A clinical pharmacist evaluated patients at their clinicvisit. The evaluation consisted of a focused recordreview of the computerizedmedical record, a semistructuredpatient interview, identificationand classification ofmedication-related problems,and a history of allergies andadverse drug reactions (ADRs).Medication lists compiled bystructured patient interviewwere compared with computerizedmedication profiles atthe time of the interview todetermine overall agreement.Patients were instructed tobring all of their medicationswith them for the interview.
Accuracy of the computerizedmedication list wasassessed by 3 methods. Thefirst was the percentage ofpatients who had between the medicationprofile gathered viastructured interview (ie, theactual number and names of the medications taken)and the number and names of the medications on thecomputerized record. The second method assessed, which were defined as medications thatwere not on the computer record, but that currentlywere taken by the patient. To determine the overall percentageof omissions, the denominator was the actualnumber of medications taken by the patient. Finally, were defined as medications that were onthe computer record, but that were not currently takenby the patient. To determine the overall percentage ofcommissions, the denominator was the total number ofmedications on the computer record. Omissions andcommissions were reported as both the mean numberof medications per patient and a percentage of the totalnumber of medications for all patients in aggregate.
The effect of copayment status on commission andomission rates also was examined. Veterans who receivemedications dispensed by the VA may be requiredto make a copayment for their medications. The typicalcopayment is $7.00 for a 30-day supply of a medication.
Allergy and ADR agreement between computerizedprofiles and structured patient interview was assessed.An allergy was defined as a known sensitivity or hypersensitivityto a drug, and an ADR was defined as any noxious,unintended, and undesired effect of a drug afterdoses used in humans for prophylaxis, diagnosis, or therapy.The structured interview was compared with thecomputerized allergy/ADR information to assess perfectagreement of the allergy/ADR information. Omissionswere reported as both the number of patients with anallergy or ADR not included on the medication profile,and the number of allergy and ADR omissions for allpatients in aggregate. Commissions included the numberof allergies and ADRs found on the computer medicalrecord, but denied by the patient.
Proportions, means, and standard deviations werereported where appropriate. Differences betweencopayment status were compared with a Student's test. All analyses were conducted using SAS version 8.1for Windows (SAS Institute. Cary, NC).
A total of 493 patients were evaluated. Their mean agewas 74.3 years, 97.8% were male, and 70.0% made acopayment for their medications (Table 1). Patients hada mean of 10.7 medications on their computer medicationprofile and were taking a mean of 12.4 medications.Of all medications, 65% were prescription, 23% wereover-the-counter (OTC) products, and 12% were vitamins/herbals. The percentage of patients with completeagreement between their computerized medication profileand what they were actually taking was 5.3%.
There was a mean of 1.3 commissions per patient;12.6% of all medications on the computer list were notbeing taken by patients (Table 1). There was a mean of 3.1omissions per patient; 25.0% of all medications the patientswere taking were not included on the computerizedmedication profile. Our results indicate that very fewpatients had complete agreement between the structuredmedication history and computerized medication lists.
We also evaluated the effect of copayment status onthe accuracy of the medication profile (Table 2). In oursample, 70% of the patients were required to make acopayment for their medications. Patients with a copaymenthad a mean of 9.6 medications listed on their computerizedmedication profile, compared with 13.0 in thegroup without a copayment ( < .01). The copaymentgroup had a mean of 1.3 commissions per patient, andthe group without a copayment had a mean of 1.6 commissionsper patient ( = .06). That is, in the copaymentgroup, 13.0% of medications were commissions, comparedwith 12.0% in the group that did not have to makea copayment. Copayment status was not significantlyassociated with number of commissions.
To evaluate the omissions, the denominator becomesthe total number of medications the patient was actuallytaking. The copayment group was actually taking a meanof 11.8 medications, compared with 13.9 for the groupwithout a copayment ( < .01). The copayment group had3.4 omissions per patient, and the group without a copaymenthad 2.4 ( < .01) That is, 28.7% of the medicationstaken by patients with a copayment were not included onthe computerized medication record, compared withonly 17.5% for patients who did not have a copayment.Patients with a copaymenthad a significantlyhigher number ofomissions on their computerizedrecord.
Table 3 lists thecommissions andomissions by mutuallyexclusive drugclasses. Cardiovascular(16.2%), topical(13.3%), and gastrointestinal(11.5%) agentsrepresent the drugclasses most frequentlyincluded on thecomputerized profilethat the patients wereno longer taking(commissions). Vitamins/minerals (26%),anticoagulant/antiplateletagents (12.2%),and gastrointestinal agents (11.5%) were the classes ofmedications most frequently omitted from the computerizedmedication list (omissions).
Table 4 lists the top 10 commissions and omissionsby individual drug name. Aspirin (5.0%), docusate(3.5%), and albuterol (3.1%) were the 3 agents most likelyto be found on the computerized list, but whichpatients were no longer taking (commissions). Sixty-sixpercent of all commissions were prescription medications.The individual drugs most frequently omitted onthe computerized profile were aspirin (10.4%), multivitamins(8.2%), and acetaminophen (6.7%). Thirty-fourpercent of all omissions were prescription medications.
Two thirds of patients (n = 318) had complete agreementbetween the computer record and patient reportfor allergies and ADRs (Table 5). Thirty-eight patients(7.7%) reported at least 1 allergy that was not on thecomputerized medication record. Of the 215 allergiesreported by these patients (some patients had morethan 1 allergy), 50 allergy omissions (23.2%) were not inthe computerized medication record. There were only 2(0.9%) allergy commissions, in which the patient deniedan allergy that was in the computer medical record.
A total of 140 patients (28.3%) reported at least 1ADR that was not on the computer medication record.Of the 360 total ADRs confirmed by patients, 230(63.9%) were not included in the computerized medicationrecord. There were 6 ADR commissions (1.6%), inwhich the patient denied an ADR that was recorded inthe computer medical record.
Our findings have significant implications for patientcare as well as research using PBM databases. Becauseonly approximately 1 in 20 patients had perfect agreementbetween their computerized medication profileand what they were actually taking, systems need to bein place to review medication use at every clinic visitand hospital discharge. This lack of agreement is a functionof multiple factors. The first factor influencing theaccuracy of the computerized medication profile is theinability to add medications to the VA computer profilethat the patient purchased over the counter or thatwere prescribed by non-VA providers. The VA pharmacycomputer system was designed only to include medicationsdispensed by the VA. This would explain whythe top 6 most frequent omissions were OTC medicationsmost likely purchased by the patients outside theVA system. However, the fact that 34% of all omissionswere prescription medications purchased outside theVA system is cause for concern. This limitation of the VAcomputer system significantly hampers the ability toaccurately record all the medications a patient is taking.
The VA has recognized this problem, and in the fallof 2004, it is revising the computerized medication profileacross the VA to allow non-VA prescribed medicationsto be recorded. Although this change will allowcare providers to input outside medications, it stillrequires someone to take an accurate medication historyand keep the computer profile up to date. This limitationin medication profiles also has been identified bythe Joint Commission on Accreditation of HealthcareOrganization and has been targeted in their 2005 ambulatory-care national patient safety goals, which statethat organizations will "accurately and completely reconcilemedications across the continuum of care."12
A second reason for the inaccuracy of the computerizedprofile is failure to update the computerized medicalrecord. Medications may be discontinued by thepatient or non-VA providers without informing the primarycare provider. Non—primary care providers in theVA also may discontinue or start a medication, especiallyover the telephone, and fail to update the computerizedprofile. Entire medication profiles often arerenewed at each visit or hospitalization without closelyexamining the list. This practice may lead to carry overof medications that were previously discontinued by theprimary care provider or other providers. The proposedVA computer modifications will not correct the 12.6%commission rate; fixing this problem will require careproviders to remove medications that a patient is nolonger taking from the active list.
Copayment status also played a significant role in theaccuracy of the computerized medication record.Although there was no significant difference in thecommission rate by copayment status, patients with acopayment were much more likely to have medicationomissions than those without a copayment. This probablyis because many medications, both prescription andnonprescription, can be obtained outside the VA for lessthan the $7.00 copayment for a 30-day supply. Forexample, aspirin, multivitamins, and some analgesicsand cardiovascular agents can be obtained for less costto the patient outside the VA. In addition, for the 30% ofpatients who did not have a copayment, there was likelyno financial incentive to purchase any medicationsoutside the VA. Interestingly, patients with no copaymentalso were taking, on average, more medicationsthan patients who had a copayment. This may bebecause the medicationswere free, or because thisgroup of patients hadgreater illness burdenrequiring more chronicmedication use.
Allergy and ADR informationalso was frequentlyinaccurate, with 23.2% ofallergies and 63.9% ofADRs not documented inthe computerized medicationrecord. This lack ofdocumentation can pose arisk to patients if they areprescribed a medication towhich they have a seriousallergy such as anaphylaxis.In addition, adverseoutcomes or unnecessarymedical care could resultif patients are prescribedmedications to which theyalready have a knownintolerance (ADR).
One of the main concernsabout not having anaccurate computerizedmedication profile ispatient safety. Adverse drugevents and drug—drug ordrug—disease interactionsare common in outpatientsand may lead to excess clinicvisits, hospitalizations,side effects, and costs. Drugclasses with high risk fordrug—drug or drug—diseaseinteractions include cardiovasculardrugs, agentsthat affect the central nervous system, and anticoagulant/antiplatelet agents.13 These classes account for 26% ofthe commissions and 23% of the omissions in our study.In a study of outpatients by Gandhi et al, 25% experiencedan adverse drug events over a 3-month period.14 Themean age of their study population was 52 years, and themean number of medications taken by these patients was1.53. Our study population was older (mean age 74.3years) and taking more medications (mean of 12.4), puttingthem at much higher risk for adverse drug events. Ina recent ambulatory VA study of potential drug—dietarysupplement interactions, 45% had a potential interactionof any severity, and 6% had the potential for a severe interaction.15A large meta-analysis found that at least 5% of allhospital admissions are due to ADRs, and about 4% ofpatients admitted due to drug reactions die.16 Anotherstudy found 5% to 9% of hospital costs were due to ADRs.17These studies underscore the potential risk patients incurwhen taking medications, and although no studies haveshown that having more accurate medication historiesprevents these adverse outcomes, working with moreaccurate information is a vital component to safe prescribingand monitoring of medication use.
Inaccurate medication lists also can impact assessmentsof quality of care. For example, aspirin use in postmyocardialinfarction patients is a frequent qualityindicator. However in our study sample, 5% of commissionand 10% of omission medications were aspirin. Somepatients who appear to be taking aspirin are not, andmany who do not appear to be receiving aspirin are actuallytaking it. In addition, cardiovascular drugs, includingangiotensin-converting enzyme inhibitors and betablockers,represent 8.2% of all omissions. These classesof drugs also are critical in the management of conditionssuch as congestive heart failure and postmyocardialinfarction. These omissions can lead to prescribingof medications that patients are already taking or ofsimilar medications (ie, therapeutic duplication).Similarly, clinicians may fail to realize a patient is notactually taking 1 of the commission drugs and thereforenot prescribe medically necessary medications.
Further, our findings identify some of the limitationsof using PBM data for quality assurance and researchpurposes. Although no computerized medication recordsystem will be perfectly accurate all of the time, ourresults suggest areas for potential improvement. Specifically,a system should be designed that allows for theinclusion of OTC products, vitamins/herbals, and prescriptionmedications from both within and outside thehealthcare system. Another area for improvementwould be to develop a system that allows for systematicevaluation of medication lists by nurses, providers, andpharmacists. This system should reinforce the need toupdate medication lists when providers discontinuemedications and to use caution when renewing entireprofiles. Involving patients by giving them medicationlists to review prior to clinic visits also may be a methodto continuously update medication profiles.
There are some limitations to our study. The first isthat it was conducted in a single VA primary care clinic.Although all VA clinics use the same computerized medicalrecord, there may be systematic differences in theway each facility updates its computerized medicalrecord. Another limitation is the accuracy of thepatients' reports of what they were taking. However, thestudy had a systematic method for reviewing medications,and patients were instructed to bring all of theirmedications to the clinic for review. Therefore, we donot believe that accuracy of patient reports was an areaof significant bias in our study. However, one prior studydid show that patient reports during a clinic visit had atleast 1 omission 48% of the time, so our findings may infact be underrepresenting the problem of omissions.18
Our findings outline potential limitations of the VAcomputerized medication profile. Inaccurate medicationlists could result in risks to patient safety andimpact the assessment of quality of care. In addition,when using pharmacy records for research, it is importantto understand the limitations of the database.When existing computerized medication records aremodified (or new systems are developed), the medicationprofile needs to be systematically assessed and theaccuracy maximized, and users should be informed ofthe inherent limitations of such systems.
From the Center for Research in the Implementation of Innovative Strategies in Practice(CRIISP) at the Iowa City VA Medical Center, Iowa City, Iowa (PJK, ABH, MJB); and theDivision of General Internal Medicine, Department of Medicine, University of Iowa CarverCollege of Medicine, Iowa City (PJK, BJM).
This research was supported by a grant from the Health Services Research andDevelopment Service, Department of Veterans Affairs (SAF 98-152). Dr Kaboli is supportedby a Research Career Development Award from the Health Services Research andDevelopment Service, Department of Veterans Affairs (RCD 03-033-1).
Presented at the 27th Annual Meeting of the Society of General Internal Medicine,Chicago, Ill, May 13, 2004.
Address correspondence to: Peter J. Kaboli, MD, MS, Division of General InternalMedicine, University of Iowa Hospitals and Clinics SE615GH, 200 Hawkins Drive, IowaCity, IA 52242. E-mail: email@example.com.
Br Med J Clin Res Ed.
1. Price D, Cooke J, Singleton S, Feely M. Doctors' unawareness of the drugs theirpatients are taking: a major cause of overprescribing? 1986;292(6513):99-100.
J Am Geriatr Soc.
2. Beers MH, Munekata M, Storrie M. The accuracy of medication histories in thehospital medical records of elderly persons. 1990;38:1183-1187.
Arch Intern Med.
3. Kaushal R, Shojania KG, Bates DW. Effects of computerized physician orderentry and clinical decision support systems on medication safety: a systematicreview. 2003;163:1409-1416.
To Err is Human: Building a Safer Health System.
4. Kohn LT, Corrigan JM, Donaldson MS, eds. The Committee on Quality ofHealth Care in America. Washington, DC: National Academy Press; 2000.
5. The Leapfrog Group for Patient Safety. Rewarding higher standards. 2003. Availableat: http://www.leapfroggroup.org/for_hospitals/leapfrog_safety_practices/cpoe.
South Med J.
6. Dezii CM, Kawabata H, Tran M. Effects of once-daily and twice-daily dosing onadherence with prescribed glipizide oral therapy for type 2 diabetes. 2002;95:68-71.
J Clin Psychopharmacol.
7. Roe CM, Odell KW, Henderson RR. Concomitant use of antipsychotics anddrugs that may prolong the QT interval. 2003;23:197-200.
J ManagCare Pharm.
8. Wogen J, Kreilick CA, Livornese RC, Yokoyama K, Frech F. Patient adherencewith amlodipine, lisinopril, or valsartan therapy in a usual-care setting. 2003;9:424-429.
9. Sloan KL, Sales AE, Liu CF, et al. Construction and characteristics of the RxRisk-V:a VA-adapted pharmacy-based case-mix instrument. 2003;41:761-774.
10. Liu CF, Sales AE, Sharp ND, et al. Case-mix adjusting performance measuresin a veteran population: pharmacy- and diagnosis-based approaches. 2003;38:1319-1337.
J Am Geriatr Soc.
11. Pugh MV, Cramer J, Knoefel J, et al. Potentially inappropriate antiepilepticdrugs for elderly patients with epilepsy. 2004;52:417-422.
12. Joint Commission on Accreditation of Healthcare Organizations. Nationalpatient safety goals for 2005 and 2004. Available at: http://www.jcaho.org/accredited+organizations/patient+safety/npsg.htm. Accessed July 12, 2004.
J Am Geriatr Soc.
13. Hanlon JT, Schmader KE, Koronkowski MJ, et al. Adverse drug events in highrisk older outpatients. 1997;45:945-948.
N Engl J Med.
14. Gandhi TK, Weingart SN, Borus J, et al. Adverse drug events in ambulatorycare. 2003;348:1556-1564.
Arch Intern Med.
15. Peng CC, Glassman PA, Trilli LE, Hayes-Hunter J, Good CB. Incidence andseverity of potential drug-dietary supplement interactions in primary care patients. 2004;164:630-636.
16. Einarson TR. Drug-related hospital admissions. 1993;27:832-840.
Br J Clin Pharmacol.
17. Moore N, Lecointre D, Noblet C, Mabille M. Frequency and cost of seriousadverse drug reactions in a department of general medicine. 1998;45:301-308.
J Gen Intern Med.
18. Yang JC, Tomlinson G, Naglie G. Medication lists for elderly patients: clinic-derivedversus in-home inspection and interview. 2001;16(2):112-115.