Weighing In

WEIGHING IN

Selecting an appropriate database for the investi-gation of drug effects warrants consideration of multiple factors. Once it has been determined that a specific drug or set of drugs under investiga-tion is on the formulary, the relative size of the prospective databases may be an important consid-eration, as the process of evaluating the occurrence of rare effects requires large numbers of users of the drug(s) in question. UnitedHealth and Medicaid offer the largest databases, although UnitedHealth is not population-based. GHC is the smallest of these North American databases, but it contains informa-tion on inpatient drug exposures. The combined HMO Research Network is an important new option for large-scale post-marketing drug studies.

Saskatchewan contains a stable, representative, population-based database, in which loss to follow-up is minimal, making it more desirable for study-ing outcomes that have a delayed effect. Among the US databases, Kaiser is the most stable, with 3% loss a year after the first 2 years of enrollment. Compared to the total populations in the areas they serve, the members of GHC, Kaiser and UnitedHealth are disproportionately employed. Medicaid recipients over-represent the poor and disabled.

Drug data vary in their completeness across the databases. Medicaid data would be the most complete, as the formularies are the least restrictive, and Medi-caid patients are unlikely to purchase drugs outside of the insurance plan, as they are economically disad-vantaged individuals who can obtain them without charge through Medicaid. Saskatchewan drug data are likely to be complete, if the drug is on the formu-lary. GHC is missing drug data on Medicare patients, that is, the elderly. Kaiser and UnitedHealth lack phar-macy benefits for 6%–7% of their populations, and Medicare drug benefits vary depending on the specific plan, so pharmacy files may be incomplete in this age range. Most health plans lack the means of assessing drugs purchased that cost less than the plan’s copay, or drugs purchased prior to the patient’s meeting the annual deductible (e.g., HMOs) or after the patient has reached the drug benefit limit. This is not a problem for Medicaid data.

Outpatient diagnosis data are available for the described health plans, but are limited in Saskatchewan to only one code per visit, and only three digits of the five-digit ICD9-CM code are used.

Access to medical records is often crucial for veri-fying diagnoses, characterizing the severity of a diag-nosis, and for obtaining data on important potential confounding variables not found in the computerized data. This access has been possible with all these databases, but is no longer feasible in Medicaid for reasons of confidentiality; other databases that rely on claims may begin to suffer from the same problems. The HMO Network has been resourceful in meeting the HIPAA requirements, and can draw on the rela-tive strengths of the participating entities as needed for specific studies. Essential requirements for their studies are carefully designed and well-coordinated planning in the preparation of the individual datasets by each participating entity.

None of these databases can assess the use of over-the-counter drugs, complementary/alternative ther-apies or physician or other professional samples. Patient compliance has not been directly measurable, although the benefit of a claims database compared with use of physician records is knowing that not only was a prescription written by the physician, it was also dispensed by the pharmacist. Prescriptions that are renewed suggest that the patient was indeed taking the drug. The extent of use of drugs taken intermittently for symptom relief is difficult to assess.

Of course, much of this will likely change over the next few years, as US Medicare begins paying for drugs for the elderly for the first time. On one hand, this represents the potential for the largest database yet created, if available to researchers. On the other hand, it may create huge gaps in the other databases. This will need to be watched closely as the programme evolves.

The growing adoption of electronic medical record systems in the US portends exciting opportunities for future pharmacoepidemiologic and clinical research. The ability to link claims from prescription fills to the physician’s issuing of the prescription will expand studies of adherence to drug therapy. Access to health indicators such as vital signs, height and weight, alco-hol consumption and smoking will enhance our capa-bility of controlling for such confounders. Although records maintained for clinical rather than research purposes have inherent biases, lessons can be learned from the experience with the UK GPRD (Gelfand, Margolis and Dattani, 2005). Systems must be estab-lished to monitor the quality of data entry by health care personnel, and other potential sources of errors in the use of electronic systems (Koppel et al., 2005). As with claims data, validation analyses and consis-tency checks must be implemented. Despite the inevitable challenges posed by an electronic medical record system, the result would be a rich comple-ment to claims data for future pharmacoepidemiologic research.

CONCLUSION

Electronic databases are useful in hypothesis testing of signals from pharmacovigilance as well as drug safety surveillance. The speed with which data can be accessed and the relatively low cost of their use make these databases excellent resources.