This is a longitudinal, retrospective study using interrupted time series (ITS) to examine out-of-pocket payments and MoH expenditures for HTN and DM treatment. The main outcomes addressed are rates of HTN and DM treatment coverage, number of individuals in FP, total expenditures, percentages paid by MoH, treatment cost per capita and out of pocket payment.
The Brazilian National Ethics Committee, by the National School of Public Health – Fiocruz – Brazil and the WHO ERC, approved the ISAUM-Br project, which is the basis for this paper.
Interventions
The study interventions are two changes in patient cost sharing in AFP. The April 2009 AFP-II policy involved a reduction in reference prices for most FP medicines by an average of 24.5%, coupled with administrative changes aiming to improve accountability. In February 2011, the “Saúde não tem preço” (SNP) program was implemented, under which all covered medicines for HTN and DM were dispensed free of charge to patients. FP private pharmacies were reimbursed according to a set of negotiated prices, while in government-owned pharmacies, medicines were fully subsidized. Only FP private pharmacies are addressed in this paper.
Data source and study population
The FP information system is the first widespread governmental administrative system on medicines dispensing in Brazil. The FP information system in contracted pharmacies is managed by the Unified Health System Informatics Department (DATASUS). Data include patient unique identification number (CPF), price paid, date of purchase, prescribed daily dose and amount procured. CPF allows linking to data on gender and date of birth. In the majority of cases, the buyer corresponds to the patient for patients over 18 years old. Other administrative systems cover a small set of medicines (e.g. ARVs, high cost medicines) and are not integrated at national level.
FP program eligibility criteria have remained unchanged during the program: all medicines are sold only if a national ID and a valid prescription are presented. During the study period medicines were dispensed on a monthly basis, although prescriptions were valid for 120 days. Over time, the number of participating private sector pharmacies expanded substantially, especially in some regions [3].
Data are derived from an electronic point-of-sales dispensing program implemented in 2006 in FP retail pharmacies and then integrated online by DATASUS. Available data include patient and pharmacy identifiers, patient age and gender, geographic location of the pharmacy, date of dispensing, name and quantity of medicine dispensed, daily prescribed dose, amount of MoH reimbursement, and patient copayment.
We use data on dispensing of HTN and DM medicines from October 2006 to December 2012. All patients with at least one dispensing during the study period were included in this analysis. Dispensing data are of good quality and relatively complete, with duplicate cases accounting for less than 0.005% and individual-level missing data at less than 0.05%. We excluded encounters with missing data on any outcome variables from all analyses.
Medicines covered by the program include four oral antidiabetic medications (glibenclamide 5 mg, and metformin 500 mg, 850 mg, and slow release 500 mg formulations), insulin NPH and regular and six antihypertensive medications (atenolol 25 mg, propranolol 40 mg, hydrochlorothiazide 25 mg, captopril 25 mg, enalapril 5 mg, and losartan 50 mg).
Analysis
We analyzed five study outcomes related to FP program coverage, MoH expenditures, and affordability, as follows:
1) Monthly number of individuals who received at least one dispensing at AFP pharmacies;
2) Total monthly program expenditure in reais (Brazilian currency), including total MoH expenditure and total patient payments;
3) Monthly percentage of expenditure paid by the MoH;
4) Monthly expenditure per treatment (per capita), which is the total monthly expenditure divided by the number of individuals in the program; and.
5) Average monthly out-of-pocket payment, which is the average amount paid by patients per treatment.
Annual inflation was a relatively stable 3 to 7% during the study period. We performed a monthly inflation correction for all financial outcomes [11]. We report all financial outcomes in 2012 inflation-adjusted Brazilian reais; the exchange ratio during the study period was roughly 2 Brazilian reais to 1 US dollar [12].
As an indicator of potential program sustainability, we estimated the level of expenditure that would be needed to fully cover all individuals in Brazil with DM and HTN through the FP program, and calculated the percentage that would represent of total MoH expenditures on medicines, yearly from 2006 to 2012.
It has been demonstrated that most people with HTN and DM diagnoses, respectively 95 and 85%, are under pharmacological treatment in Brazil [13]. Thus, it seems fair to use national prevalence to estimate potential FP costs, assuming that all patients were treated through the program. The costs per individual treated in the program consider the average cost per capita per type of disease HTN or DM.
To create this sustainability measure, we first developed two measures estimating annual FP program utilization: a) Number of unique individuals with at least one dispensing within a given year; b) Average number of individuals receiving at least one dispensing per month, averaged across 12 months in a given year (i.e., allowing individuals to repeat across months). We used these to construct annual and monthly estimates of program coverage, where the denominator of each measure is an estimate of the annual prevalence of each disease in Brazil, as a proxy for the number of individuals who should be under treatment [14].
In addition to deriving yearly coverage estimates using FP program data, we also used the FP coverage estimates reported in the following surveys: National Program for Improving Access and Quality of Primary Health Care (Programa Nacional de Melhoria do Acesso e da Qualidade da Atenção Básica - PMAQ-AB) [15], Brazilian Survey on Medicine Access, Utilization and Rational Use of Medicines (Pesquisa Nacional sobre acesso e utilização e promoção do Uso Racional de Medicamentos – PNAUM) [13], Surveillance of risk-factor for chronic diseases through telephone interviews (Vigilância de fatores de risco e proteção para doenças crônicas por inquérito telefônico – VIGITEL [14] and National Health Survey (Pesquisa Nacional de Saúde – PNS) [16].
We did not adjust the monetary values used in this analysis for inflation, since we are comparing the proportions of expenditures in each year, and not actual expenditures themselves.
Statistical methods
To analyze the impact of Farmácia Popular interventions on affordability and MoH expenditures, we used ITS segmented linear regression models to determine the effect of the FP policy changes on the study outcomes. In estimating effects, ITS models adjust for pre-existing trends in the period before the policy change [17]. Segmented linear regression models were constructed using the “prais , corc” command in STATA v12 [18], we analyzed linearity and autocorrelation. ITS considered to be one of the strongest quasi-experimental design to evaluate longitudinal effects of interventions, while segmented regression analysis is a commonly used statistical method for estimating intervention effects in ITS studies [17,18,19,20,21,22].
Our ITS models included three segments, baseline and one for each of the two program periods, with 29, 22, and 23 monthly observations, respectively. The segmented regression model was specified as follow [17, 20]:
$$ {Y}_t={\beta}_0+{\beta}_1\ast {month}_t+{\beta}_2\ast {AFP II}_t+{\beta}_3\ast months\ {after\ AFPII}_t+{\beta}_4\ast {AFP}_t+{\beta}_5\ast months\ after\ {SNP}_t+{e}_t $$
In this model, time (t) is a continuous variable indicating time in months from the start of the observation period; Yt = outcome variable in month t; β0 = level at the start of the observation period (intercept); β1 = baseline trend; montht = number of months from start of observation; AFPIIt = whether month t is after AFPII; β2 = level change after the AFPII; β3 = trend change after the AFPII; SNPt = whether month t is after SNP; β4 = Level change after the SNP; β5 = trend change after the SNP; et = residual error.
The baseline segment was fit with an intercept and a variable estimating trend. We estimate each policy effect by a variable representing the change in level of the outcome immediately after the policy and a second representing the change in trend of the post-policy segment. Patients would experience changes in copayment only when they presented to fill a prescription after the policy change. We thus defined a post-policy implementation period of 2 months for the program to take effect; these periods were excluded in the ITS models so that we could estimate stable post-intervention effects. Additionally, we performed a sensitivity analysis considering the possibility of autocorrelation, assessing the significance of the Durbin-Watson statistic. We found that all outcomes have some level of autocorrelation, we compare the use of “prais” alone, “prais, var rhotype (dw)”, and “prais var, corc” [18]. We made an option to use Cochrane-Orcutt procedure “prais var., corc” since it presented the better adjustment. The sensitivity analysis showed that small autocorrelation did not impact the direction, significance of the findings. (Additional file 1).
We retained all parameters in the models regardless of statistical significance. We highlight the results with p < 0.05. To create single number summaries of policy effects, we calculated estimates of the relative changes in outcomes compared to expected values based on prior trends in April 2010 and February 2012, about 1 year after the two copayment interventions.