For a drug to be verified as a PAP prescription medication, the pharmacy record could not have documentation of third-party payer or patient payment for that medication. The only exception made for payer and patient payment was for prescription medications provided by one particular pharmaceutical company, which required a $10 copayment for its PAP medications. Once PAP and non-PAP prescriptions were verified, they were aggregated by a unique patient identifier to yield prescriptionuse data for individual patients who were categorized as PAP users versus PAP nonusers.
Patient characteristics
Data on patient gender, race/ethnicity, age, insurance status, and primary cancer site were extracted. Race/ ethnicity was categorized as white, black, Hispanic, or Asian/other. Age was calculated as of July 1, 2006, from the patient’s birth date. Insurance status was based on the patient’s insurance status at the time of registration at MDACC and categorized as follows: no insurance (include self-payers and patients referred from the county public hospital), Medicare, Medicaid, or any of a variety of private/commercial insurances. Private insurances were combined into one category. Information on each patient’s primary cancer site was categorized as blood, breast, genitourinary, head and neck, or other (primarily brain, central nervous system, and an unknown primary site).
The gender and insurance variables had some missing data. When there were conflicting data for a particular patient’s gender, we coded gender as missing. When the insurance type was missing, data on the patient’s insurance status at the time of registration at MDACC were retrieved from MDACC’s financial department.
Prescription medication fills
Data on the prescription medication name (generic or brand) and institutional billing charges per fill were extracted from pharmacy records. Prescriptions were aggregated by generic and brand names, regardless of strength, dosage form, or method of administration, to identify the 20 most frequently dispensed medications overall and for the treatment of cancer. We then used Rxlist.com (www.rxlist.com), an online medication reference program, to identify each medication’s clinical indication(s). For example, the brand name medication Zofran would be aggregated with its generic, ondansetron, and would be considered as one medication indicated for nausea and vomiting.
We extracted patient billing charge per medication fill in dollars by the date of pickup in the outpatient pharmacy. Patient billing charge included patient copayments and did not include any payments from the patient’s payer or health plan. If the billing charge was missing for a medication fill, we applied a comparable charge from a prescription medication of the same name, dosage, quantity, date of pickup, and patient insurance status. When quantity, date, or patient insurance status differed, the lowest available charge was used. All charges were adjusted to the year 2008 using the US Bureau of Labor’s Annual Producer Price Index for pharmaceutical preparation and manufacturing.7
Data analysis
For patient-level analyses, a PAP user was a patient who received at least one medication through a PAP during the study period. We used descriptive statistics to compare patient characteristics of PAP users versus PAP nonusers. Next, we conducted separate unadjusted binary logit regression analyses (interpreted with odds ratios [ORs] and 95% confidence intervals [CIs]) to estimate the differences in the probability of being a PAP user for each of the patient characteristics. All patient characteristics that were statistically significant at P < 0.20 for the unadjusted analyses8 were included in the final multivariable model. The a priori level of significance was set at P < 0.05 for the multivariable model.
For other analyses conducted at the prescription level, a PAP medication was a medication verified as being provided through a PAP. We used descriptive statistics to compare the 20 most frequently dispensed prescription medications (overall and for anticancer agents specifically) by PAP status and clinical indication. Analyses were conducted in Microsoft Excel and STATA Version 11.9
Results
Study patients and prescription medications
During the 18-month observation period, a monthly mean of 1,550 patients received a monthly total of 19,000 noninvestigational medications in the outpatient pharmacy. Of these patients, 7.5% (n = 1,929) met study eligibility criteria for PAPs and received 1 of the 104 medications provided through PAPs. Thus, there were 979 PAP users and 950 PAP nonusers in the final study population. In total, the study population received 23.3% (n = 77,592) of all outpatient medications administered during this period, of which anticancer agents represented 4% (n = 3,105; Table 1).
Comparison of patient characteristics
In comparison to PAP nonusers, PAP users were, on average, younger (48 vs 52 years), indigent (73% vs 19%), white (50% vs 43%), and covered by Medicaid or were uninsured (75% versus 20%). PAP users also had more prescriptions fills (median = 30 vs 20) during the study period at the institution. Univariate analyses showed that all patient characteristics, except gender, significantly predicted PAP use. Given the strong correlation of indigent and insurance status to PAP use, we conducted post hoc analyses to assess the potential for multicollinearity between the two patient characteristics. The variance inflation factor (VIF = 4.57) did not indicate multicollinearity concerns.