Comparing Cardiovascular Outcomes and Costs of Perindopril-, Enalapril- or Losartan-Based Antihypertensive Regimens in South Africa: Real-World Medical Claims Database Analysis

This was a retrospective observational study of claims data from a database of health claims from a large private sector funder in South Africa, managed by Quantium. Anonymised patient-level claims from a 5-year period (January 2015 to December 2019) were analysed to measure clinical and cost outcomes according to the type of antihypertensive regimen received. A 5-year study period was selected to allow for observation of a sufficient number of long-term complication events of hypertension.

All work was carried out in accordance with the Helsinki Declaration of 1964, and its later amendments. Ethics approval was obtained by Professor Ntobeko A. B. Ntusi from the University of Cape Town Faculty of Health Sciences Human Research Ethics Committee for the use of the dataset. Informed consent was not required for this study.

We used a private medical insurance claims database, which contains claims for approximately 3.5 million people insured with a large private medical insurance company in South Africa (estimated to include 39% of the total privately insured population in South Africa) [15]. The claims in the database cover the following healthcare services: hospital visits/admissions, drug prescriptions, medical devices, general practitioner (GP) and specialist visits, surgical procedures, radiology and pathology services, and care from allied health professionals. Patient data were anonymised, and each insured patient was assigned a single unique identifier that enabled individualised tracking of all health claims while maintaining privacy.

The database uses International Classification of Disease 10th edition (ICD-10) codes to identify specific reasons for healthcare claims. Since a disease state may be described by more than one ICD-10 code, the database also applies a unique funder-specific Disease Episode Grouper (DEG) field to each claim to apply a single disease descriptor to a group of relevant ICD-10 codes (see Supplementary Table S1 in the electronic supplementary material for details). DEG and ICD-10 codes were linked to provider codes, identifying whether the provider was a hospital, GP, specialist, or other type of healthcare professional. The World Health Organization Anatomical Therapeutic Chemical (ATC) codes were used for classifying medications according to the organ system they affect and their therapeutic, pharmacological and chemical properties [16]. A mortality indicator was used to identify patients who died during the study period.

The database was searched to identify patients who were receiving standard doses of perindopril, losartan or enalapril, alone or in combination with other agents, using ATC codes (see Supplementary Table S2 in the electronic supplementary materials) for the treatment of hypertension (based on the ICD-10 code I10) in January 2015 (the index date). These drugs were included in the analysis because they are the most commonly used RAAS inhibitors that are available within the private healthcare system in South Africa (see Supplementary Table S3 in the electronic supplementary materials). This makes them appropriate comparators and provides the rationale for investigating possible clinical differences between them under circumstances of real-world use. Claims in the preceding year (January to December 2014) were assessed, and adherence expressed as the number of prescriptions filled compared with the number expected (medicine possession ratio). Only patients with ≥ 80% adherence to perindopril, losartan or enalapril were included to minimise the number of poorly adherent patients in the analysis dataset. This approach was taken to ensure that any effects observed could reasonably be considered to be due to an effect of the patients’ drug treatment. Patients were grouped into the appropriate cohort (perindopril, losartan or enalapril). Claims over the next 5 years (January 2015 to December 2019) were assessed and analysed. Those who left the medical insurance scheme for reasons other than death within the 5-year period after the index date were excluded from the study. Patients were excluded from the analysis if they had < 80% exposure to the relevant drug-based regimen during the 5-year period of analysis based on < 80% of prescriptions filled (i.e. as a result of drug switching). Sensitivity analysis was carried out on all patients with no exclusions. Patients who died during the study period were included in the analyses, and their drug exposure adjusted accordingly.

Patients were assigned to three cohorts according to the antihypertensive agent they were receiving at the index date (January 2015) (i.e. perindopril, losartan or enalapril). Within each treatment cohort, patient baseline characteristics at the index date were extracted, including age, sex and health plan type. Comorbidities were identified using claims for DEG and ICD-10 codes (see Supplementary Table S1 in the electronic supplementary material) from the pre-index year (i.e. January 2014 to December 2014). All claims associated with hypertension and related conditions, specifically aneurysm (± dissection or rupture), diabetes (type 1 or 2), ischaemic or haemorrhagic stroke, ischaemic heart disease (IHD), heart failure (HF), myocardial infarction (MI), renal failure, hyperlipidaemia, vascular dementia and arrhythmias over the 5-year post-index period were also collected using DEG and ICD-10 codes for specific outcomes (see Supplementary Table S1 in the electronic supplementary material), and deaths were identified using the database mortality indicator.

The primary outcome was the composite of the risk-adjusted event rate of having at least one claim for MI, IHD, HF or stroke within the 5-year study period. Secondary clinical outcomes were risk-adjusted all-cause mortality rate and risk-adjusted rate of hospitalisation for any of the composite outcome events, both expressed as the rate per 1000 people over 5 years. An additional secondary outcome was the risk-adjusted rate of the first occurrence of the composite outcome over 5 years expressed as the percentage of patients in each cohort; first events were identified as those occurring in patients who had not had any claim for the same complication in the pre-index year (i.e. during 2014).

Outcomes were risk-adjusted to account for multiple cofounders following the cell-based approach described by Ellis et al [17]. Briefly, each patient was assigned to 1 of 351 risk cells based on their age, sex, health plan type (high, intermediate, low, hospital-only or other coverage), existing comorbidities and whether or not these comorbidities were a direct outcome of hypertension. The risk index for each cell was calculated by dividing the average healthcare claim costs for all patients in the risk cell by the average cost of all patients in the cohort. The overall risk index for each treatment cohort was calculated as the average risk index for all cells in that cohort. A higher risk index indicates that the cohort had a greater underlying risk profile and higher likelihood of worse clinical outcomes and higher costs.

Cost outcomes were defined as the total cost of claims paid [in South African Rands (ZAR) per life per month (PLPM)] that were linked to hypertension or a related condition as a comorbidity (aneurysm, diabetes, stroke, IHD, HF, MI, renal failure, hyperlipidaemia, vascular dementia and arrhythmias). Costs were disaggregated into hospital, GP, specialist and medication, based on identifiers in the claims data. All costs were calculated as a weighted average of the claims paid during the 5-year study period divided by exposure months (i.e. 60 months or less for a patient who died during the study) for each individual patient and averaged across all patients. Costs were adjusted by + 2.1% for inflation each year, which was the medical inflation rate in South Africa over this period, and standardised to 2019 ZAR. The average rate of exchange for the ZAR between 2015 and 2019 (the study period) was ZAR 13.70192 per $US 1 [18].

Standard descriptive statistics were used to describe outcomes. Continuous variables were summarised as means and 95% confidence intervals (CIs), and categorical variables as frequency per 1000 patients and percentage of patients per category over the 5-year study period. Two-way statistical comparisons between the perindopril group and losartan or enalapril groups were conducted using the stratified non-parametric bootstrap method for continuous variables [19], and a Chi-squared test or the z-test equal proportions were used for categorical variables. All-cause mortality and event rates were analysed using the Kaplan–Meier estimator to provide a time–treatment analysis. Cost data were analysed using the non-parametric bootstrap method, which is well established as a primary statistical analysis when making inferences about arithmetic means for moderately sized samples of highly skewed data, such as real-world costs [19]. The bootstrap method involves taking random samples of data (bootstrap samples), with each replaced back into the dataset before a subsequent sample is taken, and uses these samples to model the overall dataset. For this test, we chose 10,000 bootstrap samples with replacement for each of the groups separately, and then compared the difference in the means of the bootstrap sample groups with the observed difference in the means of the groups in the actual dataset.

Sensitivity analyses for selected clinical and cost outcomes (primary composite event rate and proportion of patients with such an event, and total cost of claims) were conducted with no exclusion criteria applied (i.e. including all patients taking perindopril, losartan or enalapril).

All analyses were performed using R version 1.1.456 and Python version 2.7.

We identified 108,820 patients who were taking one of the designated antihypertensive agents in January 2015, of whom 63,489 (58.3%) met the study eligibility criteria after exclusion of patients who had < 80% adherence prior to entry and those who lived for 5 years but did not have 5 years of continuous health plan enrolment after the index date (Fig. 1). The mean [standard deviation (SD)] age of patients was 62.6 (12.6) years, and 49% were female (Table 1). Patients in the perindopril treatment cohort tended to be younger (mean age 61.8 years) than those receiving enalapril (63.8 years) or losartan (63.2 years) and were less likely to be female (45.3% vs. 51.9% and 53.8%, respectively; P < 0.001), but had a higher incidence of IHD as a comorbidity (14.5% vs. 9.9% and 12.5%, respectively; P < 0.001). A higher proportion of patients in the perindopril group (48.4%) had the highest level of health insurance coverage compared with the enalapril group (32.2%) or losartan group (42.9%; P < 0.001). Risk index values were 1.03 for the perindopril cohort, 1.01 for losartan and 0.94 for enalapril (P < 0.001). The majority of patients in each of the cohorts was receiving additional antihypertensive drug classes; overall, similar proportions of patients in each group were receiving diuretics, CCBs and beta-blockers as the only adjunctive agent and as combination adjunctives (see Supplementary Table S6 in the electronic supplementary materials).

The risk-adjusted rate of at least one claim for MI, IHD, HF or stroke within the 5-year study period (primary outcome) was 205 events per 1000 patients with perindopril compared with 221 events per 1000 patients with losartan (P < 0.001) and 223 events per 1000 patients with enalapril (P < 0.001). A breakdown of each of the specific cardiovascular events such as MI, IHD, HF or stroke has been provided (Table 2). Overall, 15.2% of patients receiving perindopril had a first primary outcome event during the study period compared with 16.7% of those receiving losartan (P < 0.001) and 17.2% receiving enalapril (P < 0.001). The risk-adjusted Kaplan–Meier curve also demonstrated a difference in the rate of first composite outcome complications over time (log-rank P < 0.005; Fig. 2a).

The risk-adjusted all-cause mortality rate over 5 years was 100 deaths per 1000 patients with perindopril compared with 139 deaths per 1000 patients with enalapril (P = 0.007) and 94 deaths per 1000 patients with losartan (P = 0.650) (Fig. 2b). A combined all-cause mortality and cardiovascular event rate was 305 events per 1000 patients with perindopril and 362 events per 1000 patients with enalapril (P < 0.001) and 315 events per 1000 patients with losartan (P = 0.034). Non-risk-adjusted outcomes are shown in Supplementary Table S5 in the electronic supplementary materials.

Over the 5-year study, the risk-adjusted mean (95% CI) health claims paid PLPM for hypertension and related conditions were ZAR 1341 (87–8973) in the perindopril group compared with ZAR 1465 (104–9365; P = 0.0044) for losartan and ZAR 1540 (77–10,546; P < 0.001) for enalapril (Fig. 3 and Table 3). The mean cost of claims in the perindopril cohort was therefore 8.5% lower than in the losartan cohort and 12.9% lower than in the enalapril cohort.

Hospital (inpatient) costs constituted the greatest proportion of the healthcare claims in these cohorts (Table 3). Mean (95% CI) hospital costs were highest in the group receiving enalapril [ZAR 911 (0–8810)], intermediate in the group receiving losartan [ZAR 799 (0–7586)] and lowest in the perindopril group [ZAR 748 (0–7452)0); the difference between perindopril and enalapril was statistically significant (P < 0.001), but the difference between perindopril and losartan was not (P = 0.113). The perindopril group also had lower mean claims paid for GP visits and for other services (radiology, pathology and allied healthcare services) compared with enalapril (P < 0.001 and P = 0.015, respectively) and losartan (P < 0.001 and P = 0.001, respectively), and lower pharmacotherapy costs compared with losartan (P < 0.001), but not compared with enalapril (P = 0.096). There was no significant difference in specialist costs between the perindopril group and either the losartan or the enalapril group.