Data sources and sample
We used IMS Health’s Xponent™ prescription database to characterize NP, PA, and PCP prescribing of five new chronic disease medications (representing four medication classes), and IMS Healthcare Organization Services™ (HCOS) database for provider specialty and organizational affiliations. While data exist for physician characteristics, such as age, medical school, and residency program, in the American Medical Association Physician Masterfile, an equivalent dataset does not exist for NPs and PAs. As a result, we were unable to measure these provider-level characteristics in our sample. Xponent™ directly captures 86 % of all US prescriptions filled in retail pharmacies and utilizes a patented proprietary projection method to represent 100 % of prescriptions filled in these outlets . We obtained monthly provider-level data on all prescriptions of interest dispensed in Pennsylvania between January 1, 2007 and December 31, 2011. Of note, certified registered NPs and PAs in Pennsylvania are both authorized to prescribe medical and therapeutic treatments. Xponent™ contains limited patient-level information, including the source of payment (Medicare, Medicaid fee-for-service, commercial insurance, cash or uninsured) and patient age, with no individual patient identifiers. We obtained information on provider specialty, provider sex, and organizational affiliations (e.g., medical group) from IMS Health’s HCOS database. HCOS data were used to identify primary care medical groups and to specify the number of providers within each medical group.
We compiled data for four cohorts of prescribers – one for each of the medication classes under investigation – including prescribers of oral anticoagulants, antihypertensives targeting the renin-angiotensin-aldosterone system, oral hypoglycemics, and HMG-CoA reductase inhibitors among the three prescriber types (i.e., NPs, PAs, and PCPs). We excluded those who never prescribed a medication from the class of interest during the study period. Because we were interested in primary care prescribing behavior, we excluded those providers not practicing in a primary care practice as well as those with missing values for National Provider Identifier or sex.
We measured new medication adoption in our sample for each of five new medications in the four classes of interest (dabigatran, FDA approval: 10/2010; aliskiren, approved 3/2007; sitagliptin and saxagliptin, approved 10/2006 and 7/2009, respectively; and pitavastatin approved 8/2009). These drugs were of varying novelty – as assessed by their benefit/risk profiles and mechanisms of action – and differed in their order of entry in their respective therapeutic classes. For example, dabigatran was the first of a new class of oral anticoagulants with improved efficacy and potentially lower risk of bleeding compared to warfarin in patients with atrial fibrillation [15, 16]. Sitagliptin and saxagliptin were part of a new class of oral hypoglycemic agents that target a different physiologic pathway than other diabetes medications and were purported to have minimal risk of hypoglycemia and weight gain compared to sulfonylureas with comparable efficacy . Aliskiren was the first in a new class of agents for hypertension, joining two others (i.e., angiotensin-converting enzyme inhibitors and angiotensin II receptor blockers) that inhibit the renin-angiotensin-aldosterone system; comparative efficacy and safety among the antihypertensive classes is largely unknown. Pitavastatin was the seventh ‘statin’ approved in the US and represents the most minor therapeutic advance among the study drugs . This variation in medication classes and novelty allowed us to assess whether any differences between NP, PA, and PCP prescribing were specific to a medication class or consistent across multiple classes.
The decision to adopt a new drug is multifaceted. A prescriber first needs to learn about a new drug and then decide whether to use it and how frequently to prescribe it. Therefore, we constructed three measures: 1) any prescription of the newly approved medication in the final year of the study period (2011), 2) proportion of new medication among the medication class, and 3) time to first adoption of the newly approved medication. Time to first adoption – defined as the time from FDA approval to first prescription of the new drug – provides a measure of speed of adoption, while the other two measures identify the extent of provider adoption over time . In order to assess NP and PA prescribing trends in general, we also measured the share of prescribing within the drug classes of interest accounted for by NPs and PAs compared to PCPs over time.
Our analysis followed three steps. First, for each of the four chronic disease medication classes, we described characteristics of NPs, PAs, and PCPs who were regular prescribers of the medication class as a frequency (percentage) for each variable. We used Chi-square tests to assess differences in characteristics across provider types, and Wilcoxon rank sum tests for non-normally distributed variables. Second, for each of the four chronic disease medication classes, we measured the proportion of all medications within the class prescribed by the three provider types (i.e., NPs, PAs, and PCPs) across all years. Third, we estimated new drug adoption among the three provider types by assessing the three measures of adoption previously described. To assess time to first adoption, we used the Kaplan-Meier method to compute the proportion of providers who had adopted the new drug in the 15 months post-FDA approval. The date of first prescription of each newly approved medication in the dataset was used as the index date.
Given evidence suggesting the potential for sex differences in new drug adoption by providers , we conducted post-hoc sensitivity analyses. Because sex was almost perfectly correlated with provider type, we could not control for it in multivariable analysis. Thus, we repeated all primary analyses described above on only female providers across the three provider types. We used SAS version 9.3 (SAS Institute, Cary, NC, USA) for analyses and the Stata version 11.0 (StataCorp, College Station, TX, USA) for the Kaplan-Meier graphs.