Primary Care Physician Characteristics Associated with Prescribing Potentially Inappropriate Medication for Elderly Patients: Medicare Part D Data

Discussion

Our study further validates existing evidence that PIM rates have been decreasing over time (Figure 1);^25,26 however, this is the first article to our knowledge to describe provider characteristics associated with PIM prescriptions using a nationally representative dataset. In terms of differences between high-prescriber and low-prescriber groups, our unadjusted results demonstrated notable differences in region and practice size.

When adjusting for patient panel and area-level characteristics, we found Medicare patient panel size had the most sizable association: the larger the Medicare patient panel size, the lower the PIM prescription rate. One reason for this difference may be that physicians with more Medicare patients have more familiarity with guidelines for this population such as Beers Criteria.

Other provider-level characteristics significantly associated with lower PIM rates were gender, degree type (MD vs DO) and primary care specialty. As compared with family physicians, general internists and general practice physicians had higher PIM rates. Although the reason for this is unclear and cannot be answered by this analysis, differences in training could play a role. Surveys of family medicine (FM) and internal medicine (IM) residency programs show that 97% of FM residency programs require training in the nursing home, compared with 65% of IM residency programs.²⁷ Furthermore, FM residents spend a larger proportion of their time training in the outpatient primary care setting, where following guidelines such as Beers Criteria can be part of a routine medication reconciliation process.²⁸ Although there was a trend toward lower PIM rates by geriatricians as compared with family physicians, this is not statistically significant. One notable similarity seen across primary care specialties was the top 10 PIM prescriptions. The most commonly prescribed PIMs across all specialties were similar with half of the top PIMs being benzodiazepines. As seen in other studies analyzing PIM usage, benzodiazepines are commonly identified as being frequently used.^29,30 An additional study is necessary to further understand the differences in PIM prescribing behavior between primary care specialties.

Other physician characteristics that are important to note are medical professional degree and gender. It is unclear why DOs had a higher PIM rate than MDs and further studies are required to explain these differences. As for gender, previous studies looking at quality of care provided by women physicians show that they tend to adhere more strongly to guidelines and are more likely to actively involve patients in their care.³¹ These attributes could explain the associations we are seeing in our data, yet further research should be done to better understand why this variability exists.

In terms of practice location, we found that, compared with PCPs who practiced in the Northeast region, those who practiced in the South had on average a higher PIM prescription rate. This is consistent with the regional trend observed among Medicare beneficiaries in a prior study with the South having the highest PIM rates.³² This pattern has been consistent over time.^33,34 Certainly, practice patterns can be regional and can be dictated by local availability of safer alternatives, having pharmacists integrated in clinical teams or other external factors. Another PCP practice characteristic that was associated with PIM prescription rate was practice size, with PCPs in larger practices having on average a lower PIM rate. This may be due to larger practices having the resources for infrastructure changes that could facilitate safer prescribing pattern, such as electronic medical records with clinical decision support systems that offer prescription alerts.

Our study has several important limitations. First, Medicare Part D PUFs only allows us to identify a prescriber if they have prescribed a particular drug greater than 10 times. There remains the possibility that we are underestimating PIM prescription rates for those who prescribe relatively infrequently and those who have relatively smaller Medicare patient panels. Second, given that not all Medicare beneficiaries are enrolled in Part D, our results may not be generalizable to those who have a relatively high proportion of Medicare patients with no Part D drug plans. However, since most Medicare beneficiaries are enrolled in Part D during our study period, this limitation is unlikely to have affected our results.³⁵ One thing to note is that we cannot distinguish whether the drug claim came from a standalone or Medicare Advantage–Prescription Drug Plan in our data. However, given that federal regulations require a certain number of drugs be covered per drug class³⁶ and that our analysis involved most commonly prescribed PIMs and all medications within each drug class, our findings are unlikely to have been driven by subtle differences across drug plan formularies. Third, due to the nature of administrative claims data, the prescription data included is limited to dispensed prescription medications. Therefore, we cannot account for unfilled prescriptions or over-the-counter PIMs that may have been recommended by physicians since these are purchased without a prescription. Fourth, we did not have patient-level health information to fully adjust for the health distribution of each patient panel. Although we adjusted for the average health status of a patient panel using Centers for Medicare and Medicaid's cost-based risk score, the distribution of health conditions would affect the PIM prescription probability and may be correlated with PCP characteristics. Finally, it is important to note that our findings do not translate directly into unsafe prescribing practice behavior. The AGS Beers Criteria are simply guidelines aimed to improve medication safety for the elderly. There are many reasons why a provider may safely prescribe a PIM including the fact that some PIMs can be prescribed for a legitimate diagnosis or as part of a deprescribing plan. We did not have patient-level diagnostic information nor medication history to assess these potential reasons behind PIM prescribing.

One of the strengths of this study is the inclusion of a national sample of PCPs providing prescription drugs to Medicare patients, which provides a more comprehensive representation of primary care prescribing patterns from 2013 to 2015. This is also the first study to analyze U.S. PCP prescribing patterns based on PIMs specific to primary care. The study findings highlight potential physician characteristics that may be correlated with PIM prescribing in the primary care setting. This has clear clinical implications. Given the cost of complications and adverse events related to PIMs in the outpatient setting, it is important that we determine factors that promote safe prescribing behavior. Identifying physician factors can guide educational and clinical interventions aimed at improving medication safety for the elderly by reducing the variation in care and overall PIM use.