The effect of initiation of renin–angiotensin system inhibitors on haemoglobin: A national cohort study

To determine whether initiation of treatment with angiotensin converting enzyme inhibitors or angiotensin II receptor blockers (ACEI/ARBs) is associated with a subsequent reduction in haemoglobin in the general population.


| INTRODUCTION
Angiotensin converting enzyme inhibitors (ACEIs) and angiotensin receptor II blockers (ARBs) are widely used for the treatment of heart failure, hypertension, ischaemic heart disease and proteinuric chronic kidney disease (CKD). [1][2][3][4] In health, the renin-angiotensin system affects renal erythropoietin production and bone marrow haematopoeisis. 5 Randomised data from clinical trials suggests that ACEI/ARB use is associated with a reduction in haemoglobin. [6][7][8] However, most studies are restricted to specific populations, such as patients with heart failure, advanced CKD or erythrocytosis after renal transplantation. Observational data in this area are inconsistent and limited to restricted patient groups. [9][10][11][12][13][14] There is a lack of evidence on the effect of ACEI/ARBs on haemoglobin in routine care populations.
Renal insufficiency and elevated baseline haemoglobin may modify the effect of ACEI/ARBs on haemoglobin, and the effects of ACEIs and ARBs may differ. However, evidence in these areas is conflicting. [15][16][17][18] We examined the association between ACEI/ARBs and haemoglobin in the UK primary care population. We hypothesised that initiation of ACEI/ARBs would be associated with a subsequent haemoglobin reduction. To minimise confounding by indication (which could give rise to an association between ACEI/ARB initiation and haemoglobin reduction that is in fact due to the indication for ACEI/ARB treatment, rather than the drug itself), we compared patients commencing ACEI/ARBs with patients commencing calcium channel blockers (CCBs).

| Study design and setting
We performed a cohort study of patients starting treatment with ACEI/ARBs or CCBs, using the Clinical Practice Research Datalink (CPRD Gold) as our source population. The CPRD contains real-world data from over 600 primary care practices, covering a representative sample of 7% of the UK population. 19 The study period was 1 January 2004 to 31 December 2016. We chose this period because CKD is important in our study and serum creatinine testing became more frequent after the introduction of the Quality and Outcomes Framework in 2004. 20

| Population, exposure and outcome
We used electronic records of prescriptions and investigation results to identify patients in CPRD aged ≥18 years that received a new ACEI/ARB or CCB prescription (Table S1 in the Appendix shows all generic drug names used) during the study period, and had at least 1 haemoglobin result recorded in primary care in the 12 months before and 6 months after the date of prescription. To be confident that prescriptions were truly new and to ensure complete recording of covariates, we excluded patients with <12 months of continuous practice registration prior to their first ACEI/ARB or CCB prescription.
Our exposure was ACEI/ARB initiation; our control condition was CCB initiation. We assumed that patients initiated medications on the date of the first prescription. We treated ACEI/ARB initiation and CCB initiation in the same patient as independent events, if separated by at least 6 months. We considered patients who commenced both an ACEI/ARB and a CCB within the same 6-month period to be exposed to the first drug prescribed only.
Our outcome was a ≥1 g/dL haemoglobin reduction in the 6 months following drug initiation. We chose this outcome because we aimed to study a biological effect of ACEI/ARB initiation that could be seen at any initial level of haemoglobin, beyond the influence of natural variation or laboratory measurement error. Previous studies have demonstrated haemoglobin reductions of this magnitude after ACEI/ARB initiation. 8,21 We calculated haemoglobin change as the difference between the last result in the 12 months prior to drug initiation (pre-initiation haemoglobin) and the result closest to 115 days in the 6 months after drug initiation (post-initiation haemoglobin). We used this method because 115 days is the average lifespan of a red blood cell. 22

| Covariates
We used clinical knowledge and findings from previous research to construct a conceptual framework of the relationship between ACEI/ARB initiation and haemoglobin change (see Figure S1 in the Appendix). We used this framework to select covariates associated

What is already known about this subject
• Anaemia is associated with worse outcomes in patients with cardiovascular disease.
• Clinical trial evidence suggests that angiotensin converting enzyme inhibitors and angiotensin II receptor blockers (ACEI/ARBs) can cause a reduction in haemoglobin in restricted patient groups.
• The effect of ACEI/ARB initiation on haemoglobin in routine care populations is unknown.

What this study adds
• Compared to initiation of a calcium channel blocker, initiation of ACEI/ARB treatment was associated with a 15% increase in the risk of a ≥1 g/dL haemoglobin reduction.
• If causal, this could impact the quality of life of many patients who take these widely prescribed medications and may influence drug choice and monitoring.
with ACEI/ARB use that are independent risk factors for haemoglobin reduction.
We used the most recent serum creatinine result in the 12 months prior to drug initiation to calculate estimated glomerular filtration rate (eGFR), using the CKD Epidemiology Collaboration equation, 23 without ethnicity adjustment. We corrected creatinine results to account for the standardisation of laboratory measurement. 24 We classified CKD stage as stage 3a, 3b, 4 and 5 (eGFR 45-59, 30-44, 15-29 and <15 mL/min/1.73m 2 , respectively). 25 We used primary care morbidity coding (Read codes) to identify comorbidities at drug initiation. These were: hypertension, diabetes mellitus, ischaemic heart disease, heart failure and conditions that cause chronic hypoxia (e.g. chronic obstructive pulmonary disease, cyanotic heart disease). Complete morbidity code lists are available at https://doi.org/10.17037/data.00001039. We used electronic records to identify concurrently prescribed medications that can cause bone marrow suppression or bleeding (see Table S1 in the Appendix).
We recorded calendar period to account for temporal changes in coding and clinical practice.

| Statistical analysis
We used t-tests and χ 2 tests to compare baseline characteristics between ACEI/ARB initiators and CCB initiators. For our main analysis, we used multivariable logistic regression to estimate the adjusted odds ratio (OR) of a ≥1 g/dL haemoglobin reduction, comparing ACEI/ARB initiators and CCB initiators. We initially adjusted for age group, sex and pre-initiation haemoglobin (<12, 12-13.9, 14-15.9, ≥16 g/dL; minimally adjusted model), before additionally adjusting for comorbidities (hypertension, diabetes, ischaemic heart disease, heart failure, CKD, chronic hypoxic conditions), coprescribed medications (oral bone marrow suppressing drugs, drugs that can cause bleeding) and calendar period (2004-2006, 2007-2009, 2010-2012, 2013-2016), and accounting for clustering at the primary care practice level with robust standard errors (fully adjusted model).
We handled missing data by performing a complete case analysis.
We performed all analyses using Stata 15.1 (StataCorp, USA) and R 3.3.3 (R Foundation, Austria). All statistical tests were 2-sided and conducted at a 5% significance level.

| Sensitivity analyses
In a series of sensitivity analyses, we examined the influence of: (i) including lifestyle factors (smoking, alcohol intake and body mass index); (ii) including patients with unknown renal function; (iii) excluding patients with heart failure and patients taking diuretics; and (iv) excluding patients initiating both ACEI/ARBs and CCBs on our results. We also restricted our study cohort by shortening our preinitiation period and then by excluding the first 4 weeks from our post-initiation period. Posthoc, we included only individuals with pre-initiation haemoglobin measured within 1 week of drug initiation (Table S2).

| Propensity score analysis
We performed a propensity score analysis to account for imbalances in confounders. We calculated propensity scores using a logistic mixed model including a practice-specific random effect, in which the outcome was treatment (ACEI/ARB vs CCB) and the covariates were the same as those in our fully adjusted model. We then used inverse probability of treatment weighting to create a pseudo-population. To ensure that we achieved a good balance between treatment groups, we estimated standardised differences for each covariate before and after propensity score weighting. We then estimated the average ACEI/ARB treatment effect using a weighted logistic regression model that included only the primary exposure and outcome.

| Additional analyses
We undertook a series of additional analyses to explore further the relationship between ACEI/ARB initiation and haemoglobin reduction.

| Effect modification
We assessed whether advanced CKD (eGFR <30 mL/min/1.73m 2 ) or elevated pre-initiation haemoglobin (≥16 g/dL) modified the effect of ACEI/ARB initiation (compared to CCB initiation) on a ≥1 g/dL haemoglobin reduction, using Wald tests. Posthoc, we tested for an interaction between sex and ACEI/ARB initiation in our fully adjusted model, as there is some evidence that sex modifies erythropoietin responsiveness. 26

| ARB vs ACEI
Because ACEIs and ARBs may affect haemoglobin in different ways, 5 we compared the odds of a ≥1 g/dL haemoglobin reduction between ARB initiators and ACEI initiators, with adjustment for the same covariates included in our fully adjusted model. We excluded from this analysis patients who started both an ACEI and an ARB during the study period.

| Secondary outcomes
To test the consistency of the relationship between ACEI/ARB initiation and haemoglobin reduction, we repeated our fully adjusted model using incident anaemia (defined as haemoglobin <13 g/dL in men and <12 g/dL in women) 27 following drug initiation as the outcome. We then used multivariable linear regression to compare absolute haemoglobin change between ACEI/ARB and CCB initiators, with adjustment for the same covariates included in our fully adjusted model. We also examined the association between ACEI/ARB initiation and diagnosed bone marrow suppression (identified using morbidity codes;see https://doi.org/10.17037/data.00001039) between 1 and 12 months after drug initiation, with adjustment for age, preinitiation haemoglobin and use of oral bone marrow suppressing drugs.

| Haemoglobin variation
We explored whether variation in haemoglobin biased our results, with two further analyses. Firstly, we predicted that patients with decreasing haemoglobin prior to drug initiation would be more likely to experience a haemoglobin reduction after drug initiation. If the haemoglobin trajectory prior to drug initiation differed between ACEI/ARB and CCB initiators, our outcome definition could incorrectly attribute a haemoglobin reduction to drug initiation. We assessed this by comparing the average (unadjusted) haemoglobin change prior to drug initiation (defined as the coefficient of the regression line of all haemoglobin results in the 12 months prior to drug initiation) between ACEI/ARB and CCB initiators. We restricted this analysis to patients with at least 3 haemoglobin results in the preinitiation period, so that we had sufficient data to determine a trend.
Any between-group differences in this analysis would suggest that our outcome definition biased our estimate of the association between ACEI/ARB initiation and haemoglobin reduction.
Secondly, we hypothesised that a post-initiation haemoglobin reduction would be more likely in patients with greater natural haemoglobin variation (i.e. variation over time), due to the phenomenon known as regression to the mean. 28 If ACEI/ARB initiators had greater natural haemoglobin variation than CCB initiators, the likelihood of a ≥1 g/dL haemoglobin reduction (our primary outcome) F I G U R E 1 Flowchart of cohort creation ACEI/ARB, angiotensin converting enzyme inhibitor/angiotensin receptor blocker; CCB, calcium channel blocker; CPRD, clinical practice research datalink would be higher in ACEI/ARB initiators at any time in our study period, irrespective of drug initiation (see Figure S2 in the Appendix for a graphical illustration of this). We investigated this by comparing the odds of a ≥1 g/dL haemoglobin reduction after an arbitrarily chosen date (1 January 2010), between patients who initiated ACEI/ARBs or CCBs at another time in the study period. We restricted this analysis to patients who only initiated an ACEI/ARB or a CCB and had at least 1 haemoglobin result in the 12 months before and 6 months after 1 January 2010. We adjusted this analysis for age, sex and initial haemoglobin (the most recent result before 1 January 2010). Any between-group difference in this model would be attributable to natural variation alone, and therefore argue against a causal drug effect in our main analysis.

| RESULTS
The study cohort comprised a total of 146 610 drug initiation events (86 652 ACEI/ARB and 59 958 CCB) in 136 655 individual patients.
Of these, 9955 patients initiated both an ACEI/ARB and a CCB, and therefore contributed 2 drug initiation events. Figure 1 shows the development of the cohort.
The mean age at drug initiation was 64.7 years. Male sex, younger age, ischaemic heart disease, heart failure and diabetes were more common in ACEI/ARB initiators than CCB initiators ( Table 1). The distribution of lifestyle variables (smoking, alcohol, body mass index) did not differ meaningfully between ACEI/ARB initiators and CCB initiators (see Table S3 in the Appendix).
The mean number of haemoglobin results in the pre-initiation and post-initiation periods was similar in ACEI/ARB initiators and CCB initiators. Pre-initiation haemoglobin was 0. 19 Table 3. Table S4 in the Appendix shows the mutually adjusted ORs of all covariates in the fully adjusted model).

| Sensitivity analyses
Our results did not differ meaningfully in any of our sensitivity analyses, although restriction of the pre-initiation period attenuated the association between ACEI/ARB initiation and haemoglobin reduction.
In the posthoc analysis restricted to individuals with a pre-initiation haemoglobin within 1 week of drug initiation, a ≥1 g/dL haemoglobin reduction was associated with CCB initiation (Figure 2; Table S2 in the Appendix).

| Propensity score analysis
Our propensity score analysis showed that, after accounting for potential confounders, the OR of a ≥1 g/dL haemoglobin reduction after drug initiation comparing ACEI/ARB initiators to CCB initiators was 1.15 (95% CI 1.11-1.20. Table S5 in the Appendix shows cohort characteristics before and after propensity score weighting).

| Secondary outcomes
The incidence of anaemia (defined as haemoglobin <13 g/dL in men and <12 g/dL in women) 27

| Haemoglobin variation
There were 16 040 ACEI/ARB initiators and 12 065 CCB initiators with 3 or more haemoglobin results in the pre-initiation period. Previous studies have found a stronger association between ACEI/ARBs and haemoglobin reduction. 9 This may be due to restriction to specific patient groups with a higher prevalence of comorbidities, such as the elderly 10 or patients with heart failure, 7 CKD,8 or diabetes. 12 However, it is noteworthy that advanced CKD did not modify the effect of ACEI/ARB initiation on haemoglobin reduction in our study. The greater effect that we observed in ARB initiators compared with ACEI initiators conflicts with the findings of a previous study in primary care. 15 Our findings of an attenuated association between ACEI/ARB use and haemoglobin reduction among patients with higher preinitiation haemoglobin, and a more substantial effect in women compared to men, warrant further examination.
A causal association between ACEI/ARB initiation and haemoglobin reduction is plausible. Alterations in intrarenal haemodynamics could inhibit erythropoietin secretion by increasing oxygen delivery to renal fibroblasts. A small study demonstrated a reduction in serum erythropoietin (but not haemoglobin) in healthy volunteers taking ACEIs, 30 which supports this mechanism. Experimental studies have also suggested a direct ACEI/ARB effect on bone marrow. 5,31,32 It is not possible to determine which of these explanations might underpin our findings.
Anaemia in older patients is associated with cognitive decline, decreased quality of life and increased mortality. 33  In summary, we found a modest increase in the risk of haemoglobin reduction after initiation of ACEI/ARB treatment. For some patients, this degree of haemoglobin reduction could have clinical implications. Further study could identify patients at higher risk of this adverse outcome, who may benefit from closer monitoring.