METHODS OF MEASURING MEDICATION USE/COMPLIANCE IN THE COMMUNITY DWELLING POPULATION OF OLDER PEOPLE
The use of pharmacy records to estimate compliance based on pharmacy refills correlates favorably with electronic measurement, but shares some of the same problems that are intrinsic to pill counts. A refill record will provide information about how much medication was dispensed in a given interval, but it can not validate that the medication was actually consumed or consumed at the correct time. In addition, although refill records across pharmacies can be tracked in many cases, these records do not include medications obtained from other sources, such as free samples from physicians and pharmaceutical companies, or medications obtained through sharing with other family members or friends. Free samples given to patients by their physicians and pharmacists (and medications obtained from sources outside of the pharmacy network) would have the consequence of overestimating noncompliance as measured with the pharmacy database.
Some of the problems related to appropriate medication use in the elderly, such as drug-drug interactions and drug-drug duplication can be monitored using electronic pharmacy claims data and drug utilization review computer software (Fillit et al., 1999). However, exclusive use of pharmacy claims data ignores the fact that health plan members may purchase both prescription as well as over-the-counter medications outside of the plan’s benefit structure. This can happen when members who need multiple medications exceed their pharmacy benefit plans and purchase some medications out-of-pocket. This may make a portion of their pharmacy usage unavailable to the health plan. In addition, administrative data does not include nonprescription medications—frequently consumed by older people with significant effects on polypharmacy.
Pharmacy records were used in some of the studies described earlier in the brown bag method, as a means of identifying members at risk for polypharmacy for inclusion in those medication-review studies. Other researchers have used pharmacy databases to determine the relative frequency of various combinations of medications, and to conduct case-control studies of the use of medications and adverse outcomes such as motor vehicle crashes (LeRoy, 2004). As noted by LeRoy, an advantage to the use of claims data is that it is not dependent on patient recall of medication and disease information. However, in interpreting results derived from analyzing administrative claims data, the following sources of error or influence must be considered:
LeRoy used two databases, a nonproprietary database (National Ambulatory Medical Care Survey/NAMCS) and a proprietary database (PharMetrics). The content of these databases is discussed in more detail below.
The NAMCS 1998-2000 data (nonproprietary) were used to obtain information on drug use characteristics and disease prevalence for the U.S. population. NAMCS is a national probability sample survey of visits to office-based physicians, whose detailed prescription drug information is recorded by physicians. Data are obtained on patients’ symptoms, physicians’ diagnoses, and medications ordered or provided. Statistics on demographic characteristics, external causes of injury (e.g., motor vehicle crash) and services provided are also included in the database. The unit of analysis in the database is the number of physician visits, not the number of patients. Quality control of the data is provided by the National Center for Health Statistics, Centers for Disease Control and Prevention (CDC), which has instituted a thorough system of data completeness checks and data edits. All medical and drug coding were subjected to a two-way, 10-percent independent verification procedure. Patients were sorted into groups by age (under 50 and over 50) and whether or not they had a motor vehicle crash coded as a reason for the physician visit. The kinds of descriptive analyses that were conducted with this database, sorted by age group and motor vehicle crash versus the entire cohort included the following:
The proprietary database employed by LeRoy (2004) is owned by PharMetrics, who provides anonymized patient-specific medical and pharmaceutical claims-linked datasets derived from claims paid through health insurance programs. It includes inpatient and outpatient diagnoses and procedures, and both standard and mail-order prescription records. It also allows for the analysis of medical usage and disease treatments in a temporal relationship to a motor vehicle crash. The company uses extensive data quality review procedures that use over 100 quality measures. Queries for individuals with E-codes (“external cause of injury”) for motor vehicle crashes, and three controls for each case, provide information about patient demographics, number of medications dispensed, patterns of medication combinations, and disease prevalence for patients with and without motor vehicle crashes in the enrollment population. Occurrences of drug-drug conflicts and drug-disease conflicts were also examined. The same descriptive queries were conducted as were conducted with the nonproprietary database, in addition to a matched pair case-control study. Cases were defined as all patients with one of more claims with a diagnosis code (ICD9-CM) indicative of a motor vehicle crash and with at least six months of continuous enrollment prior to their first claim(s) with a crash code. Three control patients were randomly matched to each case patient, with matching based on the following criteria: no claims with any of the motor vehicle accident codes; age within 5 years of the case; same gender; at least 6 months of continuous enrollment prior to the Case study subject’s first claim with an accident code.
A potentially relevant data set to provide information about medications dispensed to older people is the Veteran’s Administration Pharmacy Benefits Management (PBM) Database. The PBM Database is a national database of information about all prescriptions dispensed within the VHA System beginning with fiscal year 1999. The database was developed by the Pharmacy Benefits Management Strategic Healthcare Group (PBM/SHG), which is a VA entity responsible for managing the national VA drug formulary process (Smith and Joseph, 2003). PBM data files are created and stored by the PBM/SHG at the Edward Hines Jr. VA Hospital in Hines, IL. Smith and Joseph (2003) state that VA data is an important resource for understanding patterns and costs of pharmacy use by a large, predominantly older population. While there are several large non-VA prescription databases that contain information on privately insured individuals, they contain relatively few people over age 65. Although many VA patients are elderly men, the numbers of younger veterans and women allows for analysis of these groups as well. Vulnerable populations, such as people with low incomes, disabilities, or mental health and substance abuse problems are present in substantial numbers. Smith and Joseph (2003) state that the number of published studies employing the VA pharmacy data sources is small but rapidly growing, as the data sources have a great potential for use in health economics and health services research.
The database contains the following common data for each prescription order dispensed for a patient: dosing instructions, National Drug Code (NDC) where applicable, product name, ordering provider, drug product costs, quantity dispensed, formulary status, and VA drug class. Other data elements are available depending on whether the order was an IV, unit dose, or outpatient prescription order. An unusual feature of PMB is the availability of dosing instructions, which is useful to researchers in studies of patient adherence to physician instructions. Data are made available to researchers as a flat file in Microsoft VisualFoxPro, Microsoft Access, or SAS format. The patient’s SSN will be provided to researchers if there is a need to link the PBM data to other data sources. Researchers cannot directly access the PBM database; instead the PBM/SHG will create a custom extract for the researcher. The database is at the level of individual prescriptions, and therefore, a person can have multiple records on a given day. The PBM database contains no information about patient clinical characteristics. Two other VA databases, VISTA and the Decision Support System (DSS) National Data Extracts can be linked to obtain clinical data such as diagnosis and procedure codes for inpatient admissions and outpatient visits, as well as admission and discharge dates for inpatients. However, these linkages do not allow attribution of a prescription to a particular clinic code or diagnosis, if more than one appeared on the outpatient record for a given day. Patient demographic data may be obtained can be obtained by linking the PMB database to the VISTA database. Smith and Joseph (2003) note that obtaining data from VISTA is significantly more difficult, and therefore, unless additional clinical data is needed, the PBM database will be the better choice. VISTA requires specialized programming, permission by each of the 128 VA facilities across the United States (as opposed to the PMB database that extracts data from all facilities monthly), and careful interpretation across facilities. A planned DSS National Pharmacy Extract will contain rich data about patient characteristics, such as gender, date of birth, low-income status (based on the local threshold for federally subsidized low-income housing) and home ZIP code.
Data access requirements are as follows (Smith and Joseph, 2003): Researchers may have access only to special-use data extracts created by the PMB/SHG field office at the Hines, VA, hospital. Although both inpatient and outpatient data are extracted, only the outpatient PBM datafiles are currently available for research.
Requests for PBM data are fulfilled if the PBM/SHG confirms the following: the proposed data use will not conflict with the PBM/SHG’s primary mission of managing the VA formulary process; IRB approval has been granted; all applicable laws, regulations, and VA policies are being followed, including those pertaining to data confidentiality and human rights; and the requestors have completed a use and nondisclosure agreement. Non-VA researchers are provided PBM data by the PBM/SHG only if they are collaborating with a VA employee or belong to an official oversight body. PBM/SHG will not release data for research if the design appears to favor a particular medication or class of medications or if the study is not scientifically valid. Smith and Joseph (2003) report that the time and effort needed to obtain permission to access the data sources is not trivial, especially for non-VA researchers.
In some cases, there will be a charge for PBM data. There is generally no charge to create an extract for pilot VA research projects, and VA employees may access the data for management purposes at no charge. For funded research, the PBM/SHG staff will request payment in proportion to the staff time needed to consult on protocol design and to compile, analyze, and report the data. For simple data extracts that do not require protocol design assistance, there is a nominal charge to cover programmer costs.
Smith and Joseph (2003) raise an important issue regarding patients’ dual use of VA and non-VA systems. Because of the very low co-pay charged by VA pharmacies ($7 in 2002 for non-service-connected conditions), many enrollees use the pharmacy services to supplement their Medicare coverage. More than 50 percent of VA enrollees have Medicare coverage, including 22 percent of those under age 65. Consequently, researchers cannot rely on VA sources alone to present the whole picture of health care services for VA patients. One option is to obtain Medicare records for VA patients, linking them to VA encounter data through Social Security numbers. This would provide a more complete picture of health services and enable research on those who use alternative systems of care.
Another potential source of pharmacy data is the Medicare Current Beneficiary Survey (MCBS), which is a continuous, multipurpose survey of a nationally representative sample of aged, disabled, and institutionalized Medicare beneficiaries. MCBS, which is sponsored by the Centers for Medicare & Medicaid Services (CMS), is the only comprehensive source of information on the health status, health care use and expenditures, health insurance coverage, and socioeconomic and demographic characteristics of the entire spectrum of Medicare beneficiaries (http://www.cms.hhs.gov/mcbs/default.asp.) An e-mail inquiry to the Centers for Medicare and Medicaid Services regarding the use of this database in projects sponsored by NHTSA to assess the effects of polypharmacy on driving yielded the following response:40
Information obtained from the CMS Web site for researchers’ use of MCBS data is as follows. The Research Data Assistance Center (ResDAC) provides free assistance to academic and non-profit researchers interested in using Medicare, Medicaid, SCHIP, and Medicare Current Beneficiary Survey (MCBS) data for research. Primary funding for ResDAC comes from a CMS research contract. ResDAC is a consortium of faculty and staff from the University of Minnesota, Boston University, Dartmouth Medical School, and the Morehouse School of Medicine. ResDAC offers a number of services for researchers with all levels of experience using or planning to use CMS data. Services include technical data assistance, information on available data resources, and training. CMS releases MCBS data only under a data use agreement. CMS will release some billing and administrative data with the MCBS survey data, commensurate with demonstrated need. Researchers who have specific needs for more detailed geographic information or for Medicare claims data may request Limited Data Set (LDS) Files from CMS. Requests for these files must include a study protocol with specific justification for the additional data required, along with an Identifiable Data Use Agreement. The MCBS Access to Care files (1991 through 2002) and the MCBS Cost and Use files (1992 through 2001) and accompanying documentation are available. Data files are supplied in EBCDIC format on tapes with IBM standard labels. Each data file costs $480.
The Cost and Use file is a more complete file with regard to health expenditures and medical events. The MCBS Cost and Use files link Medicare claims to survey-reported events and provides complete expenditure and source of payment data on all health care services, including those not covered by Medicare. Expenditure data were developed through a reconciliation process that combines information from survey respondents and Medicare administrative files. The process produces a comprehensive picture of health services received, amounts paid, and sources of payment. The file can support a broader range of research and policy analyses on the Medicare population than would be possible using either survey data or administrative claims data alone. The strength of Cost and Use files stem from the integration of information that can be obtained only from a beneficiary and from Medicare claims data on provider services and covered charges. Survey-reported data include information on the use and cost of all types of medical services, as well as information on supplementary health insurance, living arrangements, income, health status, and physical functioning. Medicare claims data includes use and cost information on inpatient hospitalizations, outpatient hospital care, physician services, home health care, durable medical equipment, skilled nursing home services, hospice care, and other medical services.
Chrischilles et al. (2004) used claims data from 117 pharmacies participating in the Iowa Medicaid Pharmaceutical Case Management (PCM) program to identify eligible patients for an evaluation of the PCM program. Eligible patients consisted of noninstitutionalized Iowa Medicaid patients taking four or more long-term medications, including at least one medication representing 1 of 12 specified diseases (congestive heart disease, ischemic heart disease, diabetes mellitus, hypertension, hyperlipidemia, asthma, depression, atrial fibrillation, osteoarthritis, gastroesophageal reflux, peptic ulcer disease, and chronic obstructive pulmonary disease). In order to have complete claims data, patients who were not continuously eligible for Medicaid from 6 months before through 12 months after the date on which they became eligible for PCM were excluded from analysis. The prevalence of adverse drug reactions was assessed in patient questionnaires. Of the total of 3,037 patients eligible for PCM services during the study enrollment year, approximately 28 percent were 65 or older. Of the 3,037 patients identified, 2,211 were continuously eligible and constituted the analysis data set. Surveys were sent to the 2,211 subjects, and the response rate was 39 percent (659 of 2,211 were completed and returned).
The Medical Expenditure Panel Survey (MEPS), used by Aparasu and Mort (2004) to evaluate the prevalence and correlates of potentially inappropriate medications among older community-dwelling people, is a representative national sample survey of the U.S. noninstitutionalized population. The survey is conducted by the Agency for Healthcare Research and Quality in collaboration with the National Center for Health Statistics. It uses a household component (HC) and a medical provider component (MPC). The HC consisted of three rounds of computer-assisted in-person interviews with household respondents in a year, to collect information on sociodemographic characteristics, health conditions, health status, and healthcare use. The MPC validated and supplemented the HC information through a mail survey of medical providers and pharmacies. Pharmacies provided data on prescriptions dispensed to respondents. Data provided by the 21,571 respondents to the 1996 MEPS were employed to develop national estimates of health care use. For Aparasu and Mort’s (2004) study, data on the use of psychotropic medications by individuals 65 and older were extracted from the MEPS prescription file. Based on classifications described by Mort and Aparasu (2000), psychotropic medications were classified as antidepressants, antianxiety agents, sedative/hypnotics, antipsychotic agents, and stimulants. Generic product identifiers in the Price-Check PC software (First DataBank, Inc., Indianapolis, IN) were used to classify products according to their generic ingredients based on the products’ prescription National Drug Codes. The data extraction procedure resulted in an unweighted total of 471 records for older patients using psychotropic medications of the 2,455 records for elderly patients (19%). This translated to an estimate of 6.09 million (19%) of the 32.29 million elderly patients.
Gurwitz et al. (2003) used administrative data regarding outpatient health service utilization and prescription medication for their sample of 27,617 Medicare+Choice Plan enrollees who were followed by a large multispecialty group practice. The group practice provides health care to more than 30,000 people 65 and older living in a single geographic area, approximately 90 percent of whom are enrolled in the Medicare+Choice Plan. The practice provides care to members of a New England-based health maintenance organization. All Medicare+Choice plan enrollees had a drug benefit plan during the study. Comparison of this population to the overall U.S. population 65 and older demonstrated very similar age and gender characteristics. The authors state that this particular setting (a large multispecialty group practice providing care to patients 65 and older who live in a single geographic area, and who have prescription drug medical coverage) is ideal for research on prescription drug use and the incidence of adverse effects, because automated data on prescription medications, laboratory results, and electronic clinic notes were readily available. At the time of the study, while only 17 percent of all Medicare beneficiaries nationally were Medicare+Choice plan enrollees, the age and sex characteristics of the study population closely mirrored the overall U.S. population 65 and older.
Hennessy, Bilker, Weber, and Strom (2002) recommend that whenever possible, researchers using administrative data should carry out macro-level descriptive analyses on the parent data set. They reached this conclusion after examining the integrity of six Medicaid databases for use in pharmacoepidemiology research. They examined four categories of potential data errors: incomplete claims for certain time periods; absence of an accurate indicator of inpatient hospitalizations; missing hospitalizations for those 65 and older; and diagnostic codes in demographic groups in which those conditions should be rare. They found that prescription claims were missing intermittently in some States. For three of the six States, no valid marker was found for inpatient hospitalizations. Hospitalizations were missing to varying degrees for those 65 and older. They mention that studies using Medicaid data linked to Medicare data would hopefully eliminate the problem of incomplete hospitalization data, since in the United States, Medicare is almost always the primary payer for hospital claims for those 65 and older. Medicaid studies without access to Medicare data may have limited ability to study hospitalization outcomes in the elderly. There were no widespread gross errors in diagnostic codes or demographic data. In particular, Hennessy et al. recommended that researchers examine the number of claims of different types (e.g., prescription, inpatient medical, outpatient medical) over time, looking for apparent gaps. Validity of markers of hospitalization should be assessed, with comparison with an external standard undertaken whenever possible. The accuracy of diagnoses and demographic data should be evaluated by examining the frequency of select diagnoses stratified by demographic group. Further, they state that the practice of obtaining data only on the set of enrollees who will be included in the results of the study precludes the macro-level quality assurance checks that should be conducted. With regard to prescription claims, it was recommended that investigators consider the possibility of incomplete prescription claims in the execution and interpretation of studies. They cite, for example, longitudinal studies examining prescription refill patterns might incorrectly interpret incomplete claims as the failure of subjects to obtain prescription refills. Examining whether a greater number than expected of such gaps occur simultaneously in calendar time within the study cohort may help identify such study gaps.
A particular item of importance in the use of pharmaceutical databases is the time window chosen to assess drug use. Kennerfalk et al. (2002) found that use of a random date compared with a one-month period resulted in a significant underestimation of the amount of drugs used for acute conditions and, consequently, the risk of polypharmacy. Polypharmacy was defined in this study as the concomitant use of 5 or more drugs. They analyzed the number of different prescribed drugs used at the index (random) date and during the month following the index date in a sample of 5,000 patients 65 to 90 years old from the General Practice Research Database in the United Kingdom. The prevalence of polypharmacy in men at the index date was 9.7 percent compared to 15.9 percent in the month after the index date. For women, the prevalence of polypharmacy at the index date was 10.9 percent compared to 18.7 percent in the month after the index date. The difference between the two time windows was statistically significant for both men and women. The authors state that the use of a short time window will accurately include drugs for continuous use (e.g., for treatment of chronic conditions), while agents used for the treatment of acute conditions will be underestimated or even excluded. They further state that one month may be too short a time interval to reflect the risk of multiple drug use.