Haemodialysis, blood pressure and risk: at the limit of non-randomized evidence

Meta-analysis of individual-level data from 61 prospective observational studies including one million adults and 56 000 cardiovascular deaths has demonstrated a strong log-linear association between blood pressure (BP) and risk of cardiovascular mortality at all ages, without any evidence of a threshold down to at least 115/75 mmHg 1. Meta-analysis of randomized BP-lowering trials including 613 000 people and 27 000 major cardiovascular events (i.e. fatal or non-fatal coronary disease, stroke or heart failure) confirmed a one-fifth relative risk reduction for each 10 mmHg reduction in systolic BP (SBP) 2. On a relative scale, these benefits appear to be consistent across the range of SBPs to at least 130 mmHg. Subsequently, the Systolic Blood Pressure Intervention Trial of intensive versus standard BP control in at-risk adults without diabetes was stopped early following confirmation that achieving an average clinic SBP of 121 mmHg reduced the risk of major cardiovascular events by about one-quarter compared with those allocated to standard BP control (average SBP of 136 mmHg) 3. These results contributed to revised 2017 guidelines from the American College of Cardiology and American Heart Association recommending BP targets of ≤130/80 mmHg among people at ≥10% 10-year risk of cardiovascular disease, including people with chronic kidney disease (CKD) 4. In contrast to observations from apparently healthy adults, observational studies from more diseased populations have demonstrated ‘J’- or ‘U’-shaped associations between BP and risk of cardiovascular disease 5–7. Such non-randomized studies may be affected by reverse causality or other confounding, whereby cardiovascular or other diseases that lower BP and are present at the time of study assessments are incompletely measured or unmeasured. The resultant inability to account for these biases in analyses (however carefully performed) could result in optimal BP (e.g. low-normal levels) apparently being associated with increased risk of cardiovascular events compared with higher SBP. Randomized trials are not affected by such biases, and have shown that lowering BP is effective at reducing cardiovascular risk among people with cardiovascular disease in spite of such populations exhibiting U-shaped observational associations in some studies 2. Similarly, trials have demonstrated that BP lowering is effective in elderly people 3, 8, in whom some observational studies have also not always demonstrated a positive association between BP and risk of cardiovascular disease 5. U-shaped associations between BP and risk of cardiovascular disease risk are also common in advanced CKD cohorts 9–12. It is challenging to confirm that bias is distorting true associations between BP and cardiovascular risk, as comparatively few people with advanced CKD have been studied in BP-lowering trials. Nevertheless, meta-analysis by the Blood Pressure Lowering Treatment Trialists’ collaboration has demonstrated that among 8500 participants with early-to-moderate CKD, each 5 mmHg SBP reduction lowered risk of major cardiovascular events (n = 1150) by about 15%, with similar estimates of relative risk reduction in those with estimated glomerular filtration rate ≥60, ≥45 to <60, and <45 mL/min/1.73 m2 13. Benefits have also been observed in a separate meta-analysis of trials conducted among people on dialysis (eight randomized control trials: 495 events in 1679 patients), which showed that allocation to antihypertensive therapy lowered risk of major cardiovascular events by nearly 30%. These benefits were achieved with an average of just 4.5 mmHg reduction in SBP (although it is not clear when the SBP measurements that constituted this calculation were made relative to dialysis) 14. A plausible source of bias in CKD populations is again reverse causality from pre-existing cardiovascular disease (which may often be subclinical). A study of 2770 peritoneal dialysis patients from the UK Renal Registry found positive associations between BP and mortality that were stronger among those who were listed for transplantation early, or when early follow-up was effectively excluded from analyses (both methodological strategies to control for reverse causality caused by pre-existing disease) 15. Furthermore, in an analysis of 8700 people with moderate-to-advanced CKD, U-shaped associations between SBP and risk of a major cardiovascular event (n = 2187) were particularly apparent among those with elevated baseline blood troponin levels (a marker of increased risk of subclinical cardiac disease). In contrast, BP associations were positive and log-linear among non-dialysis and dialysis-dependent CKD patients who had not reported cardiovascular disease and had a low blood troponin concentration at recruitment 12. These data illustrate that the potential for biases distorting BP associations are particularly strong and difficult to control for in CKD populations. Non-randomized studies in CKD populations should therefore not be used to guide clinical practice, particularly for the selection of BP treatment targets. There are other methodological challenges for renal epidemiologists studying BP, in addition to those introduced above. Short- and medium-term within-person variability and measurement error can result in a bias referred to as regression dilution bias 16. Natural variation in BP is present in all populations and explains the phenomenon of regression to the mean, whereby a second BP measurement is more likely to be closer to the population mean BP than the first. Regression to the mean is more marked in those with an extreme value (i.e. the lowest and highest BP groups: see Figure 1 for further explanation 17). Such variation is particularly relevant in haemodialysis studies, as in-centre BP readings can be markedly affected by the rapid treatment of volume expansion that is particular to haemodialysis 18. Mean SBP over time for Study of Heart and Renal Protection (SHARP) participants on dialysis in categories defined by fifths of baseline SBP. This regression dilution plot uses the SHARP dialysis patients’ single SBP measurements at study visits. The cohort was first divided by fifths of baseline measurement. The mean SBP in the lowest fifth at baseline was 107 mmHg. By 2 months, average SBP in this group had increased to 121 mmHg and by 30 months, it was 128 mmHg. This is the amount by which daily BP fluctuation within individuals may result in patients being miscategorized (adapted from 17). Mean SBP over time for Study of Heart and Renal Protection (SHARP) participants on dialysis in categories defined by fifths of baseline SBP. This regression dilution plot uses the SHARP dialysis patients’ single SBP measurements at study visits. The cohort was first divided by fifths of baseline measurement. The mean SBP in the lowest fifth at baseline was 107 mmHg. By 2 months, average SBP in this group had increased to 121 mmHg and by 30 months, it was 128 mmHg. This is the amount by which daily BP fluctuation within individuals may result in patients being miscategorized (adapted from 17). If it is unaccounted for, regression dilution bias causes underestimations of the strength of BP associations with risk of cardiovascular disease. This phenomenon also needs consideration when stratifying populations and then monitoring BP change over time. For example, statistical analyses that select a group with low BP will usually observe the subsequent BPs to naturally regress up on remeasurement (Figure 1). In this issue of Nephrology Dialysis Transplantation, Zhang et al. 19 have taken on the challenge of estimating the relevance of pre-dialysis SBP and changes in SBP during dialysis to risk of death. The analyses used routinely collected haemodialysis centre data from 172 199 patients, among whom 73 529 died during a median of 2.1 years of follow-up. About one-third of the cohort was recorded as having a history of congestive heart failure (CHF) on diagnostic coding. Mean standard deviation (SD) pre-dialysis SBP was 150 (19) mmHg and mean change in SBP (calculated as post-dialysis minus pre-dialysis SBP and referred to as ‘peridialytic’ SBP change) was −10 mmHg 14. Seventy-three percent of participants’ peridialytic SBP fell, consistent with previous studies 9, 20 and a BP-lowering effect of haemodialysis. The first finding was a J-shaped association between pre-dialysis SBP and risk of death, with a steep negative association at low pre-dialysis SBP. The nadir in risk was around the population mean pre-dialysis SBP of 150 mmHg, and a pre-dialysis SBP of <100 mmHg was associated with at least a three times increased risk. This shape of association has been demonstrated in many similar previous studies 18, and the most likely explanation is confounding by disease(s) as explained above. The second finding was an unusual and complex nonlinear association between peridialytic SBP change and risk. Overall, any peridialytic SBP decline was associated with a modest reduction in mortality compared with no change, but larger peridialytic SBP declines were not associated with further reductions in risk (i.e. there was no exposure–response relationship). This finding differs from similar analyses by Park et al. in which small-to-moderate peridialytic SBP changes from 0 to −30 mmHg were associated with decreased risk, but larger peridialytic SBP declines were associated with increased mortality (i.e. Park et al. reported a U-shaped association between peridialytic SBP change with the risk, with the nadir at about 0 to −30 mmHg) 20. Thirdly, Zhang et al. found that pre-dialysis SBP modified associations. Among those with a pre-dialysis SBP of ≥110 mmHg, mortality risk increased with a peridialytic SBP increase, but among those with low pre-dialysis SBP (<110 mmHg), the association was inverse (i.e. a pre-dialysis SBP <110 mmHg and a post-dialysis SBP increase was associated with lower mortality). At low pre-dialysis SBP, a peridialytic SBP fall was also conversely associated with a poor prognosis. Some of this observation may be accounted for by pre-existing cardiac disease, as this pattern was particularly evident in analyses restricted to the population with baseline CHF (and less clear in those without). The pattern could also partially be explained by misclassification of pre-dialysis SBP at the time of the initial measurement (see Figure 1). Perhaps less likely is the possibility that these peridialytic SBP changes are a key cause of cardiovascular events. Disentangling these three explanations is challenging for any study. Zhang et al. consider that: ‘To further our understanding of the underlying pathophysiology specifically designed prospective studies with concurrent biochemical and physiological measurements are warranted.’ These studies would necessarily need to be large, and include careful assessment for pre-existing cardiovascular disease at baseline and information on cause-specific mortality. If these medium-term associations between pre-dialysis SBP and peridialytic BP change are causal, then associations with cardiovascular causes of death should be stronger than with other causes. New studies would also benefit from incorporating measures of volume expansion (e.g. with bioimpedance measures—see points below). It is unlikely to surprise a nephrologist that there is high mortality among dialysis patients with a low pre-dialysis SBP, and that this risk increases if SBP declines further during haemodialysis treatment. It may perhaps be less well-recognized that those with high pre-dialysis SBP, which increases further on dialysis, are also an important at-risk group. In other studies, peridialytic SBP increases have been associated with increased left ventricular mass 21. Pathophysiologically, Zhang et al. hypothesize that the phenomenon could be accounted for by substantial volume expansion, whereby ultrafiltration improves cardiac output and SBP as a patient passes back up the Frank–Starling curve. It is hypothesized that following this initial increase in SBP, further ultrafiltration should result in a drop in SBP 22, 23. Volume expansion (or ‘overhydration’) is an independent risk factor for mortality among haemodialysis patients at any level of pre-dialysis SBP (Figure 2) 24. If a peridialytic SBP increase in a ‘hypertensive’ dialysis patient is indeed a marker of volume expansion, then the analyses by Zhang et al. suggest dialysis patients with a pre-dialysis SBP of ≥160 mmHg who experience a peridialytic SBP increase should be a particular priority for a review of their dry weight 19. Association between pre-dialysis SBP and risk of death by 1-year cumulative degree of pre-dialysis volume expansion. Data are adjusted hazard ratios and 95% confidence intervals. High volume expansion (also referred to as ‘overhydration’) was defined as ≥2.5 L fluid overload, and estimated from an algorithm incorporating bioimpedance spectroscopy measurements. Those with high-volume expansion were on average 4.8 L (SD 2.8) fluid overloaded, and those with low volume expansion, fluid overloaded by 0.9 L (SD 1.5) (adapted from 24). Association between pre-dialysis SBP and risk of death by 1-year cumulative degree of pre-dialysis volume expansion. Data are adjusted hazard ratios and 95% confidence intervals. High volume expansion (also referred to as ‘overhydration’) was defined as ≥2.5 L fluid overload, and estimated from an algorithm incorporating bioimpedance spectroscopy measurements. Those with high-volume expansion were on average 4.8 L (SD 2.8) fluid overloaded, and those with low volume expansion, fluid overloaded by 0.9 L (SD 1.5) (adapted from 24). We agree with Zhang et al.’s sentiments that: ‘despite decades of research into the association between BP and outcomes in chronic HD many basic questions have not been resolved.' Identifying the optimal BP targets for people with advanced CKD with randomized evidence is overdue. The recent BP in dialysis (BID) pilot study demonstrated that some nephrologists are willing to randomize patients on dialysis with a pre-dialysis SBP ≥155 mmHg, to a standard BP control arm with a ‘standard’ pre-dialysis SBP target of 155–165 mmHg versus intensive control (e.g. pre-dialysis SBP 110–140 mmHg) 25. In this trial, investigators were first asked to review dry weights before starting new antihypertensive agents in order to attain the allocated SBP targets. Nevertheless, there was a paradoxical increase in post-dialysis weight among those allocated intensive control, while post-dialysis weight decreased in those on standard control 25. The current consensus is that a definite dry weight should be established through non-pharmacological interventions before altering antihypertensive regimes 18, and achieving this before randomization in a trial setting is important methodologically due to the interdependence of volume control and BP. Other findings from the BID trial have also increased the importance of conducting a definitive randomized BP-lowering trial in dialysis patients, as intensive control appeared to be associated with not only more cramps on dialysis, but also increased vascular access thrombosis and hospitalizations. Measurements of SBP in dialysis centres are substantially higher and more variable than home measurements 26. For example, in a trial of patient-performed home BP monitoring, baseline pre-dialysis SBP was 157 (SD 25) mmHg versus mean (SD) baseline 24-h ambulatory SBP of 144 (14) mmHg 27. The technical challenge of ambulatory and patient-performed home BP measurements (e.g. twice daily triplicate sets of BP measurements each day for a week) means current observational studies with such recordings are small. Nevertheless, there is growing consensus that the epidemiology of BP is more reliably assessed using SBP measurements mainly recorded away from haemodialysis sessions 18. Furthermore, an open-label randomized trial found that titrating home-measured average SBP to ≤135/85 mmHg versus a traditional pre-dialysis SBP target of <140/90 mmHg achieved a weekly average SBP of 144 mmHg versus 154 mmHg 27. Telemonitoring offers a more convenient method to track SBP and weight measured at home, and can be used to titrate antihypertensives between dialysis sessions 28. Such technology could be combined with systems that have been developed to centrally link up and with dialysis centres to collect trial data 29, thereby facilitating a large-scale definitive trial. Its seems implausible that optimal BP targets to reduce cardiovascular risk would differ substantially from those of general populations, as this would require a strong protective factor unique to dialysis patients that reduces the known effects of high SBP on accelerating cardiovascular diseases. It is much more likely that biases intrinsic to studying BP in diseased populations and the challenges of estimating a haemodialysis patient’s long-term average BP are distorting associations in observational studies. Given the extreme cardiovascular risk observed in dialysis populations and the modifiability of BP, reliably establishing the optimal SBP levels in dialysis populations with large randomized studies should remain a top research priority. These studies need to establish volume control before randomizing to an SBP target and consider using telemonitoring and home-based BP assessments. For now, the latest observational analyses inform us that high and low pre-dialysis SBPs that digress from the mean during dialysis are both associated with highest mortality risk, and such patients deserve careful clinical attention. None declared. (See related article by Zhang et al. Association of all-cause mortality with pre-dialysis systolic blood pressure and its peridialytic change in chronic hemodialysis patients. Nephrol Dial Transplant 2020; 35: 1602--1608)