Association between antihypertensive treatment and adverse events: systematic review and meta-analysis

1High blood pressure (hypertension) is one of the leading modifiable risk factors for cardiovascular disease worldwide, and much healthcare resource is given to reducing blood pressure. In recent years, guidelines for hypertension management have recommended lower treatment targets on the basis of trials that found benefit for cardiovascular risk reduction. In patients with frailty and multimorbidity, however, these guidelines recommend clinical judgment because of potential risks from adverse effects of treatment.

2In the UK, guidelines for managing patients with multimorbidity suggest doctors weigh the risk of diseases with the benefits and risks of treatments and make personalised treatment recommendations. Such an approach is straightforward for the benefits of treatment when data exist from numerous meta-analyses of randomised controlled trials. When attempting to judge the potential harms of treatment, however, few data are available to support decision making. Existing meta-analyses focus on the overall risk of adverse events, making it difficult to distinguish between those events that might not be considered particularly serious, such as transient electrolyte abnormalities, and those resulting in severe complications and hospital admission, such as falls or acute kidney injury.

3Currently few definitive data are available from meta-analyses of randomised controlled trials on the risks of specific harm outcomes that could be used to facilitate personalised decision making in patients with hypertension. We systematically reviewed evidence from trials and large observational studies to determine the association between antihypertensive treatment and specific adverse events such as falls, acute kidney injury, and electrolyte abnormalities.

4We performed a systematic review and meta-analysis of randomised controlled trials and large observational studies examining the association between antihypertensive treatment and adverse events. The study is reported according to the preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidelines. The study protocol was registered on PROSPERO (international prospective register of systematic reviews) and is available online (www.crd.york.ac.uk/prospero, CRD42018116860).

5To capture all randomised controlled trials reporting the association between antihypertensive treatment and adverse events we searched Embase(OvidSP), Medline(OvidSP), Cochrane Central Register of Controlled Trials (CENTRAL, Cochrane Library), and the Science Citation Index (Web of Science Core Collection). Searches were undertaken from inception of the databases until 14 April 2020, and no language restrictions were applied. In this review we focused on randomised controlled trials, which are less prone to bias from confounding by indication. We also searched for large observational studies by interrogating the bibliographies of databases of electronic health records, but as few relevant data were identified and given the limitations of observational study designs we decided not to include them in the present study. Further studies were identified through searching the references of eligible full text articles and previous meta-analyses. Supplementary table 1 shows the full search strategy.

6Eligible studies included participants aged 18 years or older, compared individuals receiving antihypertensive treatment (single agents) with those receiving placebo or no treatment, more antihypertensive drugs compared with fewer antihypertensive drugs, or one blood pressure target compared with another. Although these study designs examine different types of intervention, all compared more antihypertensive treatment with less antihypertensive treatment, enabling the potential association with adverse events to be determined. Trials were also required to present data describing the association between antihypertensive treatment and at least one adverse event. Randomised controlled trials were included if they reported 50 or more adverse events in each specific category or had at least 650 patient years of follow-up.

7To ensure study selection and data analysis remained manageable by avoiding small, early phase mechanistic studies, we specified a priori the limit on patient years of follow-up and number of outcome events. We chose the specific criteria to ensure each included study was large enough to accrue outcome events and provide reliable effect estimates. These criteria assumed an incidence of the primary outcome (falls) of 7.8 events per 100 patient years of follow-up, which would accrue at least 50 outcome events in each study.

8We excluded studies in specialist populations (children, pregnant women), and case reports, case series, or before and after studies. At least two members of the review team (AA, MS, BP, SF, CK, AD, JPS) independently reviewed study titles, abstracts, and full text articles. At each stage, the entire review team screened a proportion of articles to ensure consistency of decision making. Disagreements were resolved by a third reviewer (JPS).

9Outcomes of interest were prespecified based on those reported in recent large scale trials of blood pressure lowering treatment. The primary outcome was falls, at any time point and by any definition given in the original study. Secondary outcomes were acute kidney injury, fractures, gout, electrolyte abnormalities (changes in potassium), hypotension, and syncope (eg, fainting) at any time point during trial follow-up. Acute kidney injury was defined as any outcome reported according to the KDIGO (kidney disease: improving global outcomes) definition. All other outcomes were defined according to definitions given in the original study. Additional treatment efficacy outcomes of interest included cardiovascular death, myocardial infarction, stroke, and all cause mortality.

10AA, MH, LA, AD, and BL extracted data from eligible studies. Two reviewers independently entered outcome data into a Microsoft Excel spreadsheet (2016 version, Redmond, WA). A second reviewer then manually cross checked these, referring to the original source data when discrepancies were identified. After an initial consistency check involving extraction of data from 10 articles, one reviewer extracted study descriptive data.

11Data were extracted on populations studied, interventions tested, length of follow-up, effect measures (estimates and confidence intervals for rate ratios, odds ratios, and hazard ratios), and numbers of patients experiencing adverse events and cardiovascular or mortality outcomes.

12The methodological quality and risk of bias of individual studies was assessed using the Cochrane risk of bias tool (for randomised controlled trials).

13Summary effect estimates describing the association between all antihypertensive drug classes (combined) and adverse events were derived using a random effects meta-analysis. For uncommon adverse events (approximately less than 10% of the population experience an event), rate ratios (for rate outcomes), odds ratios (for binary outcomes), and hazard ratios (for time-to-event outcomes) were considered reasonably similar and combined provided they had the same directional interpretation. For uncommon outcomes, we label summary effect estimates as risk ratios. For more common cardiovascular disease outcomes, we synthesised rate ratios, odds ratios, and hazard ratios separately. We used restricted maximum likelihood estimation to fit the random effects model, with 95% confidence intervals derived using the Hartung-Knapp approach to account for uncertainty in heterogeneity estimates. For studies with three treatment arms, we split binary and rate outcomes for the control arm into two equal groups. This approach is not possible for the time-to-event outcomes, and therefore we made an approximate adjustment to the standard errors.

14Heterogeneity was summarised using the estimate of between study variance (τ2) and 95% prediction intervals for the treatment effect in a new study. The proportion of variability in effect estimates due to between study heterogeneity was summarised using I2.

15Sensitivity analyses were undertaken focusing on adverse events reported as a reason for study withdrawal. Meta-regression was used to examine the association between observed treatment effects and study quality. Small study effects (potential publication bias) were explored using contour enhanced funnel plots for outcomes reported in 10 or more studies. Prespecified subgroup analyses were conducted to examine the association between treatment and adverse events by antihypertensive drug class.

16No other subgroup analyses were undertaken by patient level characteristics (eg, age), owing to the risk of ecological bias. Aggregate data only allow relationships across studies to be examined, but these often do not reflect within study (participant level) relationships, because of aggregation bias and study level confounding. For example, those studies with a higher mean age might also have a longer mean follow-up or a higher dose of the drug; hence it is difficult to disentangle these different associations, and interpreting across study associations as if they were interactions at the individual level is potentially misleading.

17All analyses were undertaken using Stata version 16 (StataCorp, College Station, TX).

18This study was developed with the help of our patient and public advisor. As a member of our study advisory group, they commented on the study protocol. We also held a focus group with seven older adults during the study to discuss broader issues related to drugs for cardiovascular disease prevention and adverse events, which informed the interpretation of this work.

19A total of 15 023‬ unique articles were identified from the literature searches, of which 119 records were screened from reference lists of included articles and previous meta-analyses. After screening of the title, abstract, and full text, 63 articles originating from 58 randomised controlled trials were eligible for inclusion (fig 1). The most common reason for exclusion at full text screening was lack of adverse event reporting (n=108) or inclusion of too few patient years of follow-up (n=104).

20A total of 280 638 participants were included in the primary analyses from 58 unique randomised controlled trials. Forty eight studies compared a single drug treatment with placebo and 10 studies compared a high blood pressure target with a lower blood pressure target in the intervention and control groups (table 1). The remaining five studies either compared treatment with no treatment or compared multiple drugs with a single drug. The median duration of follow-up in the trials was 3 (interquartile range 2-4) years. Most studies were conducted in patients with at least one risk factor for cardiovascular disease in addition to hypertension.

21Supplementary table 2 presents the risk of bias assessment for individual trials. Most of the trials (n=40, 69%) had a low risk of bias (fig 2). Eight trials (14%) did not adequately blind outcome assessment of adverse events (or did not describe this adequately) and 12 (21%) did not adequately describe the randomisation process. Outcome reporting was complete in 52 trials (90%) trials.

22Seven randomised controlled trials reported data for the primary outcome of falls (fig 3). Data were available from 29 481‬ patients experiencing 1790 events. Overall, no evidence was found of an association between antihypertensive treatment and falls (summary risk ratio 1.05, 95% confidence interval 0.89 to 1.24). Little evidence was found of between study heterogeneity in this association (τ2=0.009; I2=31.5%; P=0.372). Subgroup analyses by drug type did not reveal any evidence of associations between falls and specific antihypertensive drug classes, except for thiazide diuretics, although this was based on data from just one trial (supplementary figure 1). More intensive treatment (ie, to lower blood pressure targets) was not associated with falls across four trials (supplementary figure 1).

23In analyses examining adverse events across all drug classes, antihypertensive treatment was associated with an increased risk of acute kidney injury (summary risk ratio 1.18, 95% confidence interval 1.01 to 1.39, n=15 studies; fig 4), hyperkalaemia (1.89, 1.56 to 2.30, n=26 studies), hypotension (1.97, 1.67 to 2.32, n=35 studies), and syncope (1.28, 1.03 to 1.59, n=16 studies) (table 2; supplementary figures 2-4), although statistical heterogeneity was significant for most outcomes (τ2=0.037 to 1.374; I2=42.9% to 85.1%). Evidence was unclear of an association between antihypertensive treatment and fractures (0.93, 0.58 to 1.48, τ2=0.062, I2=53.8%, n=5 studies; supplementary figure 5) and gout (1.54, 0.63 to 3.75, τ2=1.612, I2=94.3%, n=12 studies; supplementary figure 7), although confidence intervals were wide, partly reflecting large between study heterogeneity.

24Analyses of outcomes by specific drug class showed that drugs affecting the renin angiotensin-aldosterone system were associated with acute kidney injury (1.26, 1.03 to 1.56, τ2=0.030, I2=39.0%; n=9 studies; table 3, supplementary figure 8) and hyperkalaemia (2.03, 1.67 to 2.48, τ2=0.063, I2=51.0%; n=20 studies; table 3, supplementary figure 9). These effects were larger and had less between study heterogeneity than in analyses examining the association between all antihypertensive treatments and the same outcomes (table 2 and table 3). Only a small number of studies assessed the association between diuretics and hypokalaemia (three studies) or gout (five studies), and the results of these were inconclusive (table 3; supplementary figures 10 and 11). No other drug class specific associations with adverse events were observed in the stratified analyses (supplementary figures 12-14).

25On average across studies examining outcomes using time-to-event analyses, antihypertensive treatment was associated with a reduction in cardiovascular death (hazard ratio 0.92, 95% confidence interval 0.86 to 0.99, τ2=0.011, I2=54.6%, n=21 studies; fig 5), all cause mortality (0.93, 0.88 to 0.98, τ2=0.008, I2=50.4%, n=32 studies; supplementary figure 15), and stroke (0.84, 0.76 to 0.93, τ2=0.013, I2=44.8%, n=17; supplementary figure 16) (table 2). No clear evidence was found of an association between antihypertensive treatment and myocardial infarction (supplementary figure 17).

26Meta-regression examining the relation between the observed treatment effects for each adverse event outcome and study quality found no clear evidence of an association (supplementary table 3). Funnel plots showed asymmetry (potential publication bias) for hyperkalaemia and hypotension events, with smaller studies missing for smaller effect estimates, but this was not evident for other adverse events examined (supplementary figures 18-22).

27Supplementary figures 23-27 show the results of sensitivity analyses focusing on studies reporting adverse events that led to participant withdrawal from each trial (summarised in table 4). These analyses were limited to studies reporting acute kidney injury, gout, hyperkalaemia, hypotension, and syncope owing to availability of data. In these analyses, summary risk ratios for hyperkalaemia, hypotension, and syncope were increased compared with the primary analysis including all studies. However, there was no longer evidence that acute kidney injury was associated with antihypertensive treatment (table 4).

28Data from random effects meta-analyses of 58 randomised controlled trials and more than 280 000 patients with hypertension confirm the known benefit of antihypertensive treatment in reducing the risk of cardiovascular disease. These data also confirm the association between antihypertensive treatment and adverse events and show how this association varies across some drug classes and for mild (eg, hypotension without falls) and more severe (eg, acute kidney injury, syncope) adverse events. Despite a widely held belief, no association was found between treatment and falls, but an association with syncope was observed, which is important as this can have a major impact on quality of life and health service use and could even result in death.

29These data will inform shared decision making around initiation and continuation of antihypertensive treatment, especially in patients with a high absolute risk of certain adverse outcomes as a result of previous events or poor renal function. Such discussions will become increasingly important as patients age and develop frailty and multimorbidity that could put them at increased risk of adverse events.

30More than 15 000 articles were screened for inclusion in this review and 58 randomised controlled trials including a large number of participants and adverse events were identified. Although power was likely to be sufficient to detect associations between antihypertensive treatment and adverse events, we observed statistically significant heterogeneity across studies, and the resulting prediction intervals were wide. Such heterogeneity might preclude pooling of some treatment effects, so caution should be exercised when interpreting the results. For acute kidney injury and hyperkalaemia events, the observed heterogeneity was partly explained by pooling of different drug classes, and heterogeneity was reduced when we focused on drugs that affect the renin angiotensin-aldosterone system. For other outcomes, the observed heterogeneity could not be explained by study quality or differences in the drug class examined in individual trials; however, populations of interest, interventions, comparators, and study designs varied widely across studies, which could have contributed to the observed variation.

31As this review focused on adverse events, selective outcome reporting might also have been a problem. Evidence was found of publication bias for certain outcomes (hyperkalaemia and hypotension), confirming the findings of previous studies that showed adverse events are more likely to be reported in randomised controlled trials when they are statistically significant. This is understandable in the context of single trial reporting, but it would be better for the evidence base if all adverse events were reported in clinical trials to enable more complete meta-analyses in the future. It is a limitation of this review that original study authors were not contacted for these additional data.

32This review focused on large randomised controlled trials with the aim of including those with at least 50 adverse events (and therefore 650 patient years of follow-up). This restriction on study size was chosen to make the review more manageable in terms of screening and analysis and avoid inclusion of numerous small early phase mechanistic studies of varying methodological quality. The cut-off for this inclusion was chosen to ensure studies provided adequately powered estimates of association between treatment and outcomes. It is possible that some useful trials could have been excluded, although many relevant trials were still available for inclusion.

33Across all included trials, adverse events were poorly defined and probably varied across studies. For instance, many studies referred to syncope as an outcome, but did not say what type of syncopal event this might have included. A conservative approach to inclusion of outcomes was taken when possible, and only those explicitly stating the outcome of interest were included. For example, trials reporting hypotension or acute kidney injury were included, but those reporting hypotension or dizziness or renal impairment were excluded. Despite this approach, some studies were included that did not specify the thresholds used to define hypotension or acute kidney injury. This could have resulted in some relevant data for certain outcomes being missed, but this meant those that were included were likely to be sufficiently similar to enable pooling in a meta-analysis. Although the quality of adverse event ascertainment is likely to have varied between trials, it would not be expected to vary between treatment arms within trials. Thus it is unlikely that differences in the quality of adverse event ascertainment would have affected the relative treatment effects presented in this review.

34We prespecified adverse events of interest based on those reported in recent large scale trials of blood pressure lowering treatment. Other patient focused harm outcomes, such as weight gain, sexual dysfunction, fatigue, and exercise intolerance might exist that were reported in the original trials but not captured as part of this review. However, the reporting of these events is likely to vary because many have no standardised definitions. Some might be captured but not reported. It is also important to note that randomised controlled trials often select populations with less frailty and multimorbidity who are more likely to tolerate treatment. Therefore, fewer adverse events might have been reported in the included trials than would be expected in the general population.

35For outcomes included in meta-analyses, the time points at which they occurred varied across studies, and so the risk ratios and odds ratios provided relate to a summary across different times. We did synthesise hazard ratios when available, but these were rarely reported.

36Few previous meta-analyses have quantified the association between antihypertensive treatment and adverse events. Thomopoulos and colleagues examined the association between antihypertensive treatment and permanent discontinuation of treatment because of adverse events and found that antihypertensives were associated with a near doubling of risk (standardised relative risk 1.89, 95% confidence interval 1.51 to 2.39). This was similar to findings from our sensitivity analyses focusing on permanent withdrawal as a result of hyperkalaemia, hypotension, and syncope events. These associations were stronger than those observed in the primary analysis focusing on all adverse event reporting. It is possible that these events were more likely to be reported in the intervention group when they were considered serious enough to lead to withdrawal. Although the focus of this review was on adverse events, we found evidence for the beneficial effects of treatment on all cause mortality, cardiovascular mortality, and stroke, but not on myocardial infarction, as has been reported previously.

37Frey and colleagues focused on data from seven original studies investigating the harms of intensive blood pressure lowering targets (≤130 mm Hg) versus usual care (<140 mm Hg). Although this number of studies was insufficient to conduct a meta-analysis, the descriptive summary suggested that intensive blood pressure lowering might be associated with higher rates of serious adverse events. The present analysis included all trials of blood pressure lowering treatment enabling meta-analyses of the association between antihypertensive treatment and adverse events and how this association varies across mild and more severe adverse events. We identified an increased risk of acute kidney injury, hyperkalaemia, hypotension, and syncope with antihypertensive treatment.

38Stratified analyses by drug class suggested that associations with acute kidney injury and hyperkalaemia were mostly driven by the use of drugs that affect the renin angiotensin-aldosterone system (eg, angiotensin converting enzyme inhibitors, angiotensin II receptor blockers, and direct renin inhibitors). However, no evidence was found of an association with this class of drug and falls, fractures, gout, or hypokalaemia. In analyses that focused on patients prescribed diuretics, a 10-fold increase in the risk of hypokalaemia was observed, but this association was derived from only three trials, with high between study heterogeneity. The pooled effect had large confidence intervals and was not statistically significant. This null finding contrasted with previous studies that recommend routine monitoring of potassium to detect hypokalaemia in patients prescribed diuretics. This could be explained by the small number of included trials examining this drug class.

39Much debate exists in the literature on the association between antihypertensive treatment and falls. Most data showing an association originate from observational studies, which are prone to bias from confounding by indication. Despite conflicting evidence, a wide held belief remains that antihypertensive treatment increases the risk of falls. This study found no evidence for an association between treatment or lower blood pressure targets and falls, but an association was found with syncope. Although syncope is a common cause of falls, not all falls are caused by syncope and therefore not all falls will be related to blood pressure lowering treatment. In addition, reporting of falls might vary among participants (ie, not all participants will be admitted to hospital or see their primary care doctor after a fall) and participants might be more likely to be withdrawn from a trial when experiencing events that could be considered precursors to falls and fractures (eg, hypotension). If this were the case and hypotension events are not dealt with by treating doctors, the incidence of serious falls and fractures associated with antihypertensive treatment could be greater in routine clinical practice.

40The present data clearly show the benefits and harms of antihypertensive treatment for specific cardiovascular outcomes and adverse events. The data also highlight that certain adverse events might be specific to certain drug classes (eg, renin angiotensin-aldosterone system drugs and acute kidney injury or hyperkalaemia). This detail is important because some adverse events reported in randomised controlled trials might be considered relatively mild and worth the risk when weighed against the substantial benefits of treatment. These new data will allow patients and clinicians to take into consideration these benefits and risks, as has been recommended in clinical guidelines. This is particularly important now that guidelines for the management of hypertension across the world increasingly recommend more intensive treatment, but with conflicting blood pressure targets, meaning a personalised approach is required for each patient.

41The present data should ideally be combined with information about an individual’s absolute risk of each harm outcome to make informed, personalised treatment decisions. This process is complex and requires real time data, which suggests that tools embedded in electronic health records will be the way forward. Further work is needed to understand better the results of this meta-analysis (which summarises average risk ratios across all participants and studies) in the context of individualised absolute risks so that treatment initiation and discontinuation can be targeted at those with the most to gain. In the absence of such information, doctors should focus on patients who have experienced previous adverse events or have poor renal function.

42This review found no evidence of an association between antihypertensive treatment and falls (primary outcome) or fractures but did show a variation in the association between antihypertensive treatment and mild (eg, hypotension without falls) and more severe (eg, acute kidney injury, syncope) adverse events. Some effects were found to be specific to the drug class used. In patients at high risk of drug harms because of previous adverse events or poor renal function, these data should be used to inform shared decision making between doctors and patients around initiation and continuation of antihypertensive treatment.

43Many meta-analyses exist of randomised controlled trials that examine the efficacy of antihypertensive treatment, but few have studied potential harms

44Existing meta-analyses have focused on the association between antihypertensive treatment and all adverse events, grouping mild and more serious outcomes

45The association between antihypertensive treatment and specific adverse events is unclear

46In a meta-analysis of 58 randomised controlled trials, including 280 638 participants, no evidence was found of an association between antihypertensive treatment and falls (primary outcome) or fractures

47Evidence was, however, found of an association between antihypertensive treatment and potentially both mild (hypotension) and more severe (acute kidney injury, syncope) adverse events

48These data might be used to inform shared decision making between doctors and patients about the benefits and harms of initiation and continuation of antihypertensives, especially in those at high risk of harm because of previous adverse events or poor renal function