Effectiveness of bisoprolol versus other β-blockers and other antihypertensive classes: a cohort study in the Clinical Practice Research Datalink

CPRD is a primary care database that collects anonymized electronic patient medical records from general practitioners (GPs), covering over 50 million patients in the UK [9]. These patients are considered representative of the general population, with similar distribution in terms of age, sex and ethnicity [10]. In the UK, non emergency health-related issues, such as hypertension are managed within primary care [11]. The quality of CPRD data are enhanced through the UK ‘Quality and Outcomes Framework’, introduced in 2004, which provides financial incentives for GPs to maintain records of patients treated for hypertension and to record their BP at least every 9 months [12].

The study population comprised all patients (≥18 years) in the CPRD database who had newly initiated monotherapy with an antihypertensive drug between 1 January 2000 and 31 December 2017 (date of the first prescription for an antihypertensive drug defined as the ‘index date’). In relation to the index date, all patients had no record of being prescribed any antihypertensive drugs in the prior year; had a first diagnosis for hypertension in the prior 6 months prior and had at least 1 year of prior medical history. To select for antihypertensive monotherapy, patients had to have no record of any antihypertensive drugs other than the index drug for 14 days after the index date. To ensure research-quality data, only patients flagged as ‘acceptable’ in CPRD and who were monitored by a practice that CPRD deemed ‘up to research standard’ were included. Finally, patients were only eligible to be included in the study once, at their first qualifying event (Figure 1).

Comedication and comorbidities for adjustment of the propensity score were assessed during the baseline period defined as the year prior to the index date (included). Clinical measurement such as BMI or BP were defined as the last recorded value within the year prior the index date. Comorbidities were defined by at least one READ codes for a disease within the year prior the index date; and comedication by at least two prescriptions 90 days apart. The smoking status was defined based on the algorithm developed by Booth and colleagues [13].

Patients were followed from the index date plus 1 day, until the first occurrence of any of the following events: addition of another antihypertensive drug to the index treatment, discontinuation of the index treatment, patient death, transfer-out date or the end of the study period (31 December 2018) (Figure 1).

Patients were allocated to one of the five treatment cohorts based on the antihypertensive monotherapy initiated at the index date: bisoprolol, other β-blockers, ACEi/ARB, CCB or diuretics. The bisoprolol cohort consisted exclusively of patients who received bisoprolol as monotherapy. The ‘other β-blockers cohort’ included any other β-blocker monotherapy: acebutolol, atenolol, betaxolol, carteolol, carvedilol, celiprolol, labetalol, metoprolol, nadolol, nebivolol, oxprenolol, pindolol, propranolol or timolol. Antihypertensive drugs considered for each cohort are summarized in Supplementary Table 1.

If the duration of treatment prescription data were missing, these were calculated based on a modified version of the method used in the Observational Medical Outcomes Partnership Common Data Model CPRD Mapping Specification [14].

To capture discontinuation or switch to a combination therapy the following algorithm was implemented:

The primary outcome for this study was the average variation of SBP and DBP across cohorts and was identified by all recorded values of SBP and DBP, in the CPRD database within the follow-up period. BP was defined as all BP measurements reported and recorded by the GP, as part of their clinical routine practice.

Secondary outcomes were as follows: controlled BP state (SBP <140 mmHg or DBP <90 mmHg) and uncontrolled BP state over the follow-up period; first occurrence of T2DM; dyslipidemia; erectile dysfunction; or obesity. Algorithms combining specific diagnosis and prescription criteria were used to define these events and are detailed in Table 1. All READ codes (a clinical diagnosis terminology system used in UK general practice) [15] and product codes for medicinal drugs that were used are listed in Supplementary Appendix B.

Analytical cohort creation, including the creation of the analytical variables, was performed in the Instant Health Data platform (Panalgo, formerly Boston Health Analytics, MA, USA). Analyses were performed using R version 3.5.1.

A total of 267,352 patients fulfilled the eligibility criteria (Supplementary Figure 2). Before matching, patients in the CCB and diuretics cohorts were the oldest (median age 64 years), and those in the ACEi/ARB cohorts were the youngest (median age 53 years) (Supplementary Figure 2). A higher proportion of patients in the bisoprolol cohort presented with angina (3.9 vs 0.2–1.6% in the other cohorts) and arrhythmia (7.9 vs 0.4–0.9%); while a higher proportion of patients in the ACEi/ARB cohorts presented with diabetes (10.9 vs 1.5–2.8%). Furthermore, a higher proportion of patients in the bisoprolol cohort were prescribed platelet aggregation inhibitor (22.6 vs 9.2–10.6%) (Supplementary Table 3). Follow-up durations varied across the cohorts, ranging from a median duration of 4.6 months for the diuretic cohort, to 14.7 months for the ACEi/ARB cohort (Table 2). For each outcome, over 77% of the patients in the bisoprolol cohort were successfully matched to patients in the four comparator cohorts. Patients were well-matched, with baseline characteristics such as coronary artery disease ( e.g., angina, arrythmia or prescribed platelet aggregation) generally similar between cohorts (SMD after matching ≤0.1) (BP analysis set, Table 3; all other analysis sets, Supplementary Table 4; SMD tables for all analysis sets, Supplementary Table 5).

The proportion of prediabetic patients (those with non diabetic increased glycemia as indicated by a fasting plasma glucose result between ≥6.1 and <7.0 mmol/l) was higher in the ACEi/ARB cohort (2.1%), compared with the CCB (1.4%), bisoprolol (1.0%), other β-blockers (0.8%) and diuretics (0.8%) cohorts (Supplementary Table 2). However, these data were missing for most (80%) patients, meaning this variable could not be added into the PS calculations.

After matching, across cohorts, median baseline SBP ranged from 160 to 163 mmHg, while median baseline DBP ranged from 94 to 96 mmHg (Table 3). Over the follow-up period, all cohorts showed similar declining trajectories in mean SBP (Figure 2) and DBP (Figure 3).

The difference in the average variation of SBP and DBP between the bisoprolol and matched cohorts over this time was no greater than 3 mmHg (Table 4). This difference was only significant in the comparison versus diuretics (PS-matched coefficient [98.75% CI] SBP 1.58 mmHg [0.54, 2.62]; DBP 2.97 mmHg [2.32. 3.61] [Table 4]). There were no significant differences in the time taken to transition from an uncontrolled to controlled BP in almost all cohorts (point estimate of the PS-matched HRs were ~1; Table 4). However, the diuretics cohort had a slightly reduced rate of achieving controlled BP compared with the bisoprolol cohort (PS-matched HR [CI: 98.75%] diuretics vs bisoprolol: 0.92 [0.85, 0.99]; Table 4). Details on the time taken to transition from a controlled to uncontrolled BP across cohorts, and the length of stay in the controlled state across cohorts are included in Supplementary Table 6.

The analysis showed that, in real-world settings, no differences were found between bisoprolol monotherapy and other therapeutic classes respectively in controlling hypertension, over a median follow-up period spanning from 4.6 to 14.7 months. Similarly, in each head-to-head comparison, no difference was seen in the time taken to present a controlled BP from an uncontrolled BP. Nevertheless, a slightly higher BP control was only observed with bisoprolol compared with diuretics.

Our findings are consistent with the results from a meta-analysis of RCTs that revealed non clinically significant differences in SBP variation, all below 2 mmHg, between the major antihypertensive classes [23]. This meta-emphasized the fact that the reduction of cardiovascular events was due to the BP-lowering per se rather than to specific drug properties. Further to the BP decrease, comparing the cardiovascular events between classes would strength the knowledge on the comparative effectiveness of these classes.

They also align with two non interventional studies that did not observe any difference in mean reduction in SBP between the major antihypertensive classes, over 6-month [4] and 3-year [24] follow-up periods. However, one limitation of the study by Bronsert et al. 2013 [4] is that BP changes were assessed as the difference between two measurements: at baseline and at the latest BP record, with the follow-up time ranging from 1 week to 1 year, ending at the earliest point that antihypertensive treatment was changed within this timeframe. The present study included all BP measurements, performed at each visit, which more accurately reflects the continuous management of patients with hypertension in a primary care setting. Furthermore, one limitation of the study by Stapff and Hilderbrand 2019 [24] is that patients who changed hypertensive therapy within the 3-year period were excluded. The study may not be representative of a real-world hypertension population; in a real-world setting (as demonstrated in this analysis), the majority of patients with hypertension discontinue their first-line therapy within the 1st year.

While several RCTs have demonstrated the efficacy and safety of bisoprolol in patients with hypertension [25–27], real-world evidence bisoprolol is more limited. A recent analysis of real-world data from UK CPRD of 539 patients prescribed bisoprolol monotherapy, showed a significantly higher survival rate that was maintained for up to 15 years over the study period, compared with other types of β-blocker and the other major antihypertensive classes [7]. Our analysis adds evidence from clinical practice that bisoprolol did not present a different BP decrease to the other antihypertensive classes, including other β-blocker subtypes.

No difference in the risk of developing T2DM was observed across most cohorts, apart from a lower risk in the bisoprolol cohort compared with the ACEi/ARB cohort. This finding should be interpreted with caution as patients with T2DM and prediabetes are more likely to be prescribed ACEi/ARB [1], and, in this study, the proportion of prediabetic patients appeared to be highest in the ACEi/ARB cohort. Moreover, as fasting blood plasma glucose measurements were missing for most (80%) patients, this variable could not be fully controlled for. Non diabetic increased glycemia may be an unknown confounder that impacted the comparison of bisoprolol versus ACEi/ARB. There are conflicting results from previous studies comparing the risk of T2DM across the major antihypertensive classes. On the one hand, β-blockers and diuretics comparatively increased the risk of T2DM in a nested case–control study [28]. However, on the other hand, a RCT found that β-blockers and CCBs were not associated with the risk of T2DM, whereas the risk was increased with diuretics [29].

No difference in the risk of dyslipidemia between the bisoprolol and other β-blockers cohort was observed; however, there was an increased risk compared with diuretics, and a trend-wise increase in risk compared with the CCB cohort. The comparison with the ACEi/ARB cohort was inconclusive. The increased bisoprolol-related risk of dyslipidemia compared with diuretics was unexpected. As a reminder, dyslipidemia is commonly associated with an increase in cholesterol, but as well in triglycerides [30]. Diuretics are associated with an increase in total cholesterol [31], bisoprolol has no known effect on total cholesterol, HDL cholesterol or LDL cholesterol [32,33]. Regarding triglycerides, different β-blockers have been reported to have different effects on triglycerides [34]. Furthermore, some studies have shown an increase in triglycerides after 2 years of bisoprolol treatment, while other studies found no difference after 3 years [33]. Therefore, this finding may be explained by a bisoprolol-related increase in triglycerides, rather than an increase in cholesterol. Further research should be conducted to investigate the differences between this study and the existing literature.

This study did not show any difference in the risk of developing obesity (defined as BMI ≥30 kg/m²), between patients with hypertension treated with bisoprolol compared with other antihypertensive classes. Low sympathetic nervous system activity has been suggested as a risk factor for weight gain and developing obesity. Metabolic and sympathetic signals are transmitted via β-2 adrenoceptors in adipose tissue [35]. Since bisoprolol is highly selective for β-1 receptors [7,8], it is not expected to affect adipocytes or to otherwise interfere with body weight regulation. This aligns with a study which reported that patients on carvedilol gained an average of 2.3 kg over 6 months of treatment, whereas those on bisoprolol did not experience a significant change in body weight [36]. Our analysis suggests that bisoprolol is not associated with an increased risk of developing obesity compared with other antihypertensive classes.

In this study, no difference in the risk of erectile dysfunction was observed between cohorts, but the number of events was low, with only six events reported out of the 1111 patients in the bisoprolol cohort. It is important to note that this observation may be affected by possible under-reporting, as the prevalence of baseline erectile dysfunction was much lower in this study (0.7%) compared with the results from a survey of the wider UK population (13%) [37], and hypertension is one of the most common risk factors for erectile dysfunction [38]. The low number of events and therefore lack of statistical power for this analysis limits interpretation. Our findings conflict with a systematic review that found diuretics and β-blockers were associated with erectile dysfunction while CCB and ACEi/ARBs were not [39]. However, the association between β-blockers and erectile dysfunction may be partially explained by psychological factors, as demonstrated by a study from 2003 which found that patients who experienced anxiety after being informed that β-blockers may cause erectile dysfunction, were more likely to experience this effect [40]. This information may help to ease a common patient anxiety regarding the detrimental impact of β-blockers on erectile function, improving medication adherence [39,41].

The strengths of the present study include the use of high-quality CPRD data and the appropriateness of the statistical analysis proposed with linear mixed model and the multistate Markov model. It allows to analyze all available longitudinal data while considering the censoring process as well. Likewise, the additional competing risks analysis accounted for patient discontinuation and switching of drugs, which reflects the real world setting compared with RCTs. And finally, the use of PS matching minimized confounding effects by lessening baseline differences between cohorts. Therefore, the proposed ad-hoc statistical method permitted robust evaluation of the long term, real-world BP variations and safety outcomes of different antihypertensive classes. Our study also used larger populations and much longer follow-up timeframes than those typically reported in clinical trials [25,26]. There were also several limitations to this study. As previously discussed, non diabetic increased glycemia may be an unknown confounder in this study, which could have impacted the T2DM outcome analysis. Several well-known cardiovascular risk factors, including physical inactivity, are not accurately recorded in CPRD; therefore, it is not possible to rule out the possibility that unmeasured risk factors could still confound the result, although E-values were calculated to account for and quantify residual confounding effects [21]. PS matching was done to minimize confounding bias; however, this only matched patients at treatment initiation, and this benefit could have been lost over the follow-up periods. Finally, follow-up time and discontinuation rate varied between cohorts: the bisoprolol cohort had a higher discontinuation rate, especially compared with ACEi/ARB; the shorter follow-up period for bisoprolol may have had a moderate impact on the short-term outcomes such as BP variation, but an even greater impact on long-term outcome such as the development of T2DM. But the robust methodology that were applied was able to utilize the time at risk on each subject with censored data to provide unbiased survival estimates.