Potentially Inappropriate Medication Prescribing by Nurse Practitioners and Physicians

Lin-Na Chou MS; Yong-Fang Kuo PhD; Mukaila A. Raji MD; James S. Goodwin MD

Disclosures

J Am Geriatr Soc. 2021;69(7):1916-1924. 

In This Article

Methods

Data Source

We used the medical claims and prescription filling record from 100% Texas Medicare beneficiaries in 2015 and 2016. Master Beneficiary Summary Files were used to determine beneficiary characteristics and Medicare enrollment status. Outpatient visits were defined from the Outpatient Statistical Analysis Files (OutSAFs) and Carrier files. Medication use was determined using Prescription Drug Event files. Medicare Provider Analysis and Review files, OutSAFs, and Carrier files were used to determine hospitalization and comorbidity. The University of Texas Medical Branch Institutional Review Board approved this study.

Cohort Selection

Medicare beneficiaries aged ≥66 years in 2016 and continuously enrolled in fee-for-service (FFS) and Part D plan in 2015 and 2016 were selected (Figure S1). We restricted our cohort to beneficiaries who had at least one outpatient visit to a PCP in Texas. Outpatient visit was defined by Current Procedural Terminology codes (99201–99205, 99211–99215). PCPs included primary care physicians and primary care NPs. Primary care physicians were determined with Centers for Medicare and Medicaid Services (CMS) specialty codes as 01, 08, 11, and 38. Primary care NPs were determined with CMS specialty code as 50 and the taxonomy codes 363L00000X, 363LA2200X, 363LF0000X, 363LG0600X, 363LP2300X, and 363LW0102X. The final cohort included 2,502,374 outpatient visits by 615,395 beneficiaries to 4822 PCPs (Figure S1).

Measurement

The definition of PIM prescription was based on the measure of high-risk medications in older adults from the Healthcare Effectiveness Data and Information Set (HEDIS).[24] There were 8695 National Drug Code codes listed in HEDIS® 2017 to determine high-risk medications, including 76 medications from 20 therapeutic categories according to the 2015 AGS Beers Criteria.[7,25] Table S1 shows the list of medications for each therapeutic category. An initial PIM prescription was defined as a PIM prescribed for a beneficiary who had not taken the same medication during the past year, and a refill PIM prescription was the prescription for a beneficiary who had ever taken the same medication during the past year. Only medications prescribed by the same provider and filled within 7 days after a face-to-face outpatient visit were used to estimate the PIM prescription rate, to eliminate the prescriptions received from other providers or for another medical need.

Each outpatient visit was classified as a physician visit or NP visit based on the provider specialty as defined above. Beneficiary characteristics included demographic factors (age, gender, race/ethnicity, and resident area), original medical entitlement, dual eligibility, any hospitalization in 2015, and Elixhauser comorbidity score in 2015.[26] The resident area was classified into three-level rurality (metropolitan, urban, and rural) according to the 2013 rural–urban continuum codes from the United States Department of Agriculture.[27]

Statistical Methods

The initial PIM rate and the refill PIM rate were estimated as per 1000 visits for each provider type and for different categories of beneficiary characteristics. The impact of provider type and beneficiary characteristics on the PIM prescription rate was estimated as odds ratio (OR) with adjustment for age, gender, race/ethnicity, original medical entitlement, dual eligibility, comorbidity, hospitalization, and area of residence in a multivariable logistic regression model. A full model with an interaction term of provider type and each beneficiary characteristic was applied to test the interaction effect. Stratification analyses were applied for those characteristics with a significant interaction effect.

We conducted two sensitivity analyses. First, to eliminate the involvement of shared care, we selected a subset of 2,197,124 visits from 566,053 patients who visited only the same type of provider. Second, to control for the cluster effect of provider, we applied a multilevel logistic regression model. For a patient with more than one visit during the study period, we randomly selected one visit to remove the dependency of repeat measures within the same patient. The sample size for this analysis was 615,395. All analyses were performed with SAS version 9.4 (SAS Inc., Cary, NC).

processing....