- Email: [email protected]
Diastolic hypotension due to intensive blood pressure therapy: Is it harmful?
Accepted Manuscript Diastolic hypotension due to intensive blood pressure therapy: Is it harmful? Mats Julius Stensrud, Susanne Strohmaier PII:
S0021-9150(17)31194-2
DOI:
10.1016/j.atherosclerosis.2017.07.019
Reference:
ATH 15144
To appear in:
Atherosclerosis
Received Date: 7 March 2017 Revised Date:
12 July 2017
Accepted Date: 18 July 2017
Please cite this article as: Stensrud MJ, Strohmaier S, Diastolic hypotension due to intensive blood pressure therapy: Is it harmful?, Atherosclerosis (2017), doi: 10.1016/j.atherosclerosis.2017.07.019. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
RI PT
Diastolic hypotension due to intensive blood pressure therapy: Is it harmful? Mats Julius Stensrud1,2 and Susanne Strohmaier2,3,*
Oslo Centre for Biostatistics and Epidemiology, Department for Biostatistics, University of Oslo, Oslo, Norway 2 Diakonhjemmet hospital, Oslo, Norway 3 Channing Division of Network Medicine, Brigham and Women’s Hospital, and Harvard Medical School, Boston, USA * Corresponding author ([email protected])
M AN U
SC
1
Abstract
Background and aims: Reducing the diastolic blood pressure (DBP) below a certain threshold may lead to inadequate organ perfusion. This raises some concerns, because pharmacotherapy reduces both systolic and diastolic pressure. We
TE D
aimed to investigate whether a pathway from intensive systolic blood pressure (SBP) treatment influences cardiovascular outcomes by inducing too low DBP. Methods: We had access to data from the Systolic Blood Pressure Intervention Trial (SPRINT) including 9361 patients with a SBP of 130 mm Hg or higher and
EP
an increased cardiovascular risk. In a formal mediation analysis we investigated whether the effect of intense (target SBP: 120 mm Hg) vs. standard (target SBP: 140 mm Hg) intervention on a composite endpoint would be mediated through an
AC C
indirect, potentially harmful, effect through low DBP (< 60 mm Hg). Results: Adjusting for treatment, we find that low DBP per se is associated
with poor cardiovascular outcomes (HR 1.90 (95%CI [1.46, 2.47]). However, in a formal mediation analyses we observed that the unadjusted indirect effect of intensive blood pressure treatment going through low DBP of HR 1.12 (95%CI [1.06, 1.18]) attenuates to a statistically non significant effect of HR 1.04 (95%CI [0.98, 1.10]) after adjustment for important covariates, suggesting that the mere association is considerably confounded. Conclusions:
The increased risk in subjects with diastolic pressure below 60
1
ACCEPTED MANUSCRIPT
cannot fully be explained by the intensive treatment itself, but may be due to other measured factors. More generally, this analysis shows that adjusting for mediator-
AC C
EP
TE D
M AN U
SC
RI PT
outcome confounding is essential, even in RCTs.
2
ACCEPTED MANUSCRIPT
1
Introduction
It is debated whether low diastolic blood pressure (DBP) leads to undesirable cardiovascu-
RI PT
lar outcomes. From a physiological perspective, coronary perfusion occurs mostly during
diastole, and low DBP may reduce the coronary flow 1 . Furthermore, a J-shaped relation between blood pressure and cardiovascular events has been observed 2–4 , in which DBP < 60 (hereby DBP60 ) is associated with increased risk of myocardial infarction,
SC
heart failure and cardiovascular death 5,6 . Most recently, the J-shaped relation was also observed in data from the ONTARGET and TRANSCEND trials, suggesting that DBP below 70 is associated with increased risk of cardiovascular outcomes 7 . Furthermore,
M AN U
McEvoy et al 8 found that low DBP was associated with higher levels of Troponin T, which indicate damage of the myocardium.
Intensive treatment of the systolic blood pressure (SBP) is associated with reduced risk of major cardiovascular outcomes 9,10 . However, a reduction of the SBP is intimately connected to a reduction of the DBP, as also seen in major clinical trials 9,11 . Hence, intensive blood pressure therapy could lead to excessive DBP reduction, which could be
TE D
associated with increased cardiovascular risk 7,8,12 . Motivated by the ongoing debate on intensive blood pressure therapy, we aimed to investigate whether a pathway from intensive SBP treatment influences cardiovascular outcomes through DBP60 in a formal mediation analysis using data collected within the
EP
Systolic Blood Pressure Intervention Trial (SPRINT)
Material and methods
AC C
2
2.1
Data source and study population
We were able to access individual level data from the Systolic Blood Pressure Intervention Trial (SPRINT) 9 through the SPRINT data challenge (https://challenge.nejm.org). In brief, SPRINT was a randomized, controlled, open-label, multi-center trial conducted in the United States. As reported in more detail in the original New England Journal of Medicine article 9 , 9361 participants with a systolic blood pressure of 130 mm 3
ACCEPTED MANUSCRIPT
Hg or higher and an increased cardiovascular risk, without diabetes were randomized to either an intensive or standard treatment strategy. For participants in the intensive treat-
RI PT
ment group the systolic blood pressure target was 120 mm Hg, while the target in the standard treatment group was 140 mm Hg. Participants were asked to return to the study
site on a monthly basis for the first 3 months, and every 3 months thereafter to monitor their blood pressure. Medications were adjusted monthly based on current and previ-
SC
ous systolic blood pressure values to achieve the respective target values. A composite
endpoint of myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure, or death from cardiovascular causes
M AN U
was considered as primary outcome.
The objective of the present study was to investigate whether the effect of intensive vs. standard treatment on the primary outcome is mediated by a potentially harmful indirect effect through too low diastolic blood pressure, which we defined as diastolic blood pressure below 60 mm Hg (DBP60 ). Since addressing mediation questions requires some temporal ordering (the treatment must occur before the mediator and in turn, the
TE D
mediator must be measured before the outcome), we conducted our analysis among 8301 participants who were still at risk for the primary outcome one year after randomization and who had a non-missing blood pressure measurements at that time since we consider DBP60 one year after randomization as our mediator.
EP
All participants in the SPRINT study provided written informed consent. Before entering the SPRINT data challenge we were granted permission by the Regional Committee for Medical & Health Research Ethics, South East Norway to apply to using the
AC C
data without a formal application to the committee.
2.2
Statistical Methods
Baseline characteristics and risk factors of the study population were described by mean and standard deviation (s.d.) for continuous variables, and by counts and percentages for categorical data. The assumed causal structure is displayed in Figure 1, where the relationship between
4
ACCEPTED MANUSCRIPT
the variables are depicted by arrows that indicate the direction of the effect. Our aim was to separate the total effect of intensive therapy compared to standard therapy into
RI PT
two separate pathways: i) A direct path that summarizes all effects not going through DBP60 , e.g. a beneficial effect of reducing SBP. ii) An indirect path, potentially harmful, that acts through on-treatment DBP60 .
First, we evaluated whether the criteria for formal mediation analysis were satisfied.
SC
That is, we investigated the association between treatment and DBP60 in a binomial
model to see whether the first criterion for mediation analysis was satisfied. Second, we fitted a Cox model for the primary outcome including treatment and DBP60 , assessing
M AN U
whether DBP60 is associated with the primary outcome.
Based on the results from these two models, we decided on performing a formal mediation analysis to estimate so called natural direct and indirect effects. These natural effects are defined within the counterfactual framework using non parametric structural equation models 13 which provide the theoretical framework for causal mediation problems. In particular, they allow to formulate the decomposition of the total effect of a given
TE D
intervention into a direct and indirect effect independent of any statistical model assumptions. Conceptually, for the presented analysis, the natural direct effect would compare the risk of the primary outcome for participants in the intensive treatment group compared to those in the standard treatment group if their mediator level (DBP60 ) had taken
EP
the value that would have been observed if they had been in the standard treatment group. Accordingly, the natural indirect effect would compare the risk for the primary outcome if the mediator takes the value it would naturally take under intensive therapy
AC C
compared to the risk of the primary outcome if the mediator takes the value it would take under standard therapy even though the treatment was kept at intensive therapy. For a more elaborate introduction to causal mediation analysis using these counterfactual formulations in biological applications, we refer to the recent book by VanderWeeele 14 . Even though the definitions of these effects do not make any reference to particular model assumptions, there are some assumptions required to be able to identify these effects from observed data. More particularly, one has to assume that there is no unmea-
5
ACCEPTED MANUSCRIPT
sured confounding of the intervention - mediator relationship, the intervention - outcome relationship and the mediator-outcome relationship, and that there are no intertwined
RI PT
pathways. To estimate the described effects we used a method suggested by Lange et al. 15 based
on marginal structural models. Heuristically, the natural direct and indirect effects can be directly extracted from any type of outcome model (a Cox model for the present
SC
analysis) fitted to an augmented data set where weights are assigned to each row, which
can be calculated using an appropriate model for the mediator incorporating all relevant confounders. For details, please refer to the paper by Lange et al. 15 that also provides
M AN U
code for implementation in R and SAS.
We performed crude analyses and additionally considered different sets of mediatoroutcome confounding variables. In ’Model 1’ we adjusted for age, sex and race, in ’Model 2’ we additionally adjusted for the lifestyle characteristics, smoking status and categories of body-mass index. In ’Model 3’ we additionally adjusted for participants’ existing baseline cardiovascular disease and number of blood pressure agents, and in ’Model 4’ we
TE D
additionally adjusted for biomarkers of cardiovascular risk, creatinine and triglycerides. Cut off values for categories and units are given in the Tables 2-4. In sensitivity analyses we removed the event of stroke from the outcome definition and hence considered a composite outcome comprising of myocardial infarction, acute coro-
EP
nary syndrome not resulting in myocardial infarction, acute decompensated heart failure, or death from cardiovascular causes. In stratified analyses we evaluated total, direct and indirect effects in participants older or equal to 75 years of age and in participants younger
AC C
than 75 years. Furthermore, we stratified by baseline diastolic blood pressure, considering subgroups of participants with baseline diastolic blood pressure between 60 and 80 mm Hg and with baseline diastolic blood pressure above 80 mm Hg. Robust standard errors were used for calculation of 95% confidence intervals (CIs).
All analyses were performed with R version 3.2.0.
6
ACCEPTED MANUSCRIPT
3
Results
In Table 1 baseline characteristics of our study sample are presented. Despite the neces-
RI PT
sary restrictions for the present analysis (patients had to still be at risk for the primary event and had to have a blood pressure measurement at year 1), the distribution of the
characteristics across treatment groups is still balanced (Table 1). The median follow-up time was 3.26 years.
SC
When examining the criteria for mediation analysis, we observed the odds ratio of
DBP60 to be 3.14 (95% CI [2.74, 3.60]), comparing intensive and standard treatment groups. In a model for the outcome including the treatment and the mediator DBP60 ,
M AN U
we found a HR of 1.90 (95% CI [1.46, 2.47]) for the mediator. Since both the essential criteria for mediation were met, we proceeded with a formal mediation analysis. In a crude mediation model without adjusting for mediator-outcome confounders, the total effect of 0.70 ((95% CI [0.56; 0.86]) could be decomposed into a direct effect of HR of 0.63 (95% CI [0.50, 0.78]), and the indirect effect through DBP60 of 1.12 (95% CI [1.06, 1.18]) (see Table 2). Adjusting for basic characteristics changed the decomposition to a direct
TE D
effect of HR of 0.65 (95% CI [0.52, 0.81]) and an indirect effect of 1.07 (95% CI [1.01, 1.13]), additionally adjusting for baseline cardiovascular risk and biological markers did not change the point estimates markedly, however confidence intervals for the indirect effect included 1 (fully adjusted ’Model 4’: direct effect HR 0.66 (95% CI [0.53, 0.82])
EP
and indirect effect of 1.04 (95% CI [0.98, 1.10]). When considering the alternative outcome definition we observed a slightly stronger
AC C
total effect (HR: 0.67 (95% CI [0.53, 0.85])) and very similar patterns regarding direct and indirect effects and adjustment for different sets of confounding variables. In stratified analyses (Table 2) we observed a stronger total effect in participants 75
years or older at baseline (HR: 0.55 (95% CI [0.39, 0.76])), but the indirect effect was not found to be statistically significant in any of the considered models. In participants younger than 75 years the total treatment effect was less pronounced and not significant (HR: 0.82 (95% CI [0.62, 1.09])). When stratifying according to baseline diastolic blood pressure, we found a relatively
7
ACCEPTED MANUSCRIPT
strong total effect (HR: 0.59 (95% CI [0.44, 0.80])) among participants with a baseline diastolic blood pressure between 60 and 80 mm Hg, which could be decomposed into a
RI PT
direct effect of HR: 0.51 (95% CI [0.37; 0.69]) and an indirect effect of HR: 1.19 (95% CI [1.07, 1.31]). Adjustment for confounding altered direct and indirect effects, resulting in
a direct effect of HR: 0.54 (95% CI [0.39, 0.74]) and an indirect effect of HR: 1.11 (95% CI [0.99, 1.23]) in the fully adjusted model. Again the total effect among participants
4
Discussion
M AN U
statistically significant (HR: 0.81 (95% CI [0.58, 1.14])).
SC
with baseline diastolic blood pressure above 80 mm Hg was less pronounced and not
Our analysis of the SPRINT data does not suggest that therapy induced DBP60 strongly increases cardiovascular risk. After adjustment for mediator-outcome confounding factors, there was no significant effect of therapy induced DBP60 on the primary outcome of myocardial infarction, acute coronary syndrome not resulting in myocardial infarction,
TE D
stroke, acute decompensated heart failure or death from cardiovascular causes. To exclusively consider cardiac events, we also removed stroke from the primary outcome in a secondary analysis, but no significant effect of DBP60 was obtained. The consequences of low DBP may be particularly relevant to elderly patients, who
EP
often suffer from isolated systolic hypertension 16,17 . Hence, we also performed a subgroup analysis of subjects > 75 years at baseline, providing no clear evidence for a harmful effect of DBP60 on the primary outcome. Further, analysis within different groups of patients
AC C
defined by baseline diastolic blood pressure, showed no evidence of effect modification. The hypothesis that low blood pressure might have harmful effects has been debated
for decades 2,4,18,19 , and previous studies have also aimed to investigate the potential
harmful effect of too aggressive blood pressure treatment 7,8,20–23 . Among the most recent, McEvoy et al 8 found that DBP60 is associated with damage
to the myocardium, indicated by the release of Troponin T in the Atherosclerosis Risk in Communities (ARIC) cohort. They also found an association between DBP60 and subsequent heart disease, in particular in subjects with elevated pulse pressure. 8
ACCEPTED MANUSCRIPT
Similarly, Guichard et al 23 found that isolated hypotension was associated with heart failure in elderly patients, using propensity score matching. However, the results by
RI PT
Guichard et al are not directly comparable to our analyses since they did not have information on intensity of antihypertensive drugs.
Vidal-Petiot et al. 21 studied patients with stable coronary artery disease who were
treated for hypertension. They found that systolic blood pressure of less than 120 mm
SC
Hg and diastolic blood pressure of less than 70 mm Hg were linked to unfavorable cardio-
vascular outcomes and hence suggested that caution should be taken regarding the use of blood pressure-lowering treatment in this patient group.
M AN U
In contrast Kjeldsen et al. 22 found only too high systolic (≥ 150) and diastolic (≥ 90) blood pressure values to be associated with increased risk for cardiac events among patients treated with either valsartan or amlodipine, and hence no evidence of a J-shaped relation.
More recently, B¨ ohm et al. 7 assessed the association between different levels of SBP and DBP on cardiovascular outcomes in an observational study among patients receiving
TE D
ramipril, telmisartan, or their combination. In contrast to the SPRINT trial, their observational study found that SBP below 120 mm Hg was associated increased cardiovascular risk. They also reported that both baseline and follow-up DBP below 70 mm Hg were associated with increased cardiovascular risk, evaluating a similar primary outcome as in
EP
the SPRINT trial. It is difficult to evaluate whether their results may be confounded by background comorbidities that may yield low DBP and increase the risk of cardiovascular events based on their presented results.
AC C
Our results are in line with the results by McEvoy 8 , as we also find that DBP60 per
se is associated with severe cardiovascular outcomes in the SPRINT study. Also the results by Vidal-Petiot et al. 21 and Bohm et al 7 point in the same direction, however
with differnet cut off values for diastolic blood pressure. Still, these associations are not necessarily causal, as also recognised by McEvoy et al 8 . Furthermore, the methodological techniques in the mentioned articles cannot adequately differentiate between different causes of DBP.
9
ACCEPTED MANUSCRIPT
In particular, results from Cox proportional hazards models that include baseline covariates will not yield information on the treatment induced effect of low DBP. Such
RI PT
analyses rather indicate associations between DBP60 (possible adjusted for blood pressure treatment) and time to cardiovascular outcomes, which is a different research question. Nevertheless, we have included results from more conventional Cox models in the supplemental material, which may be more directly compared to previous studies. As for the
SC
main analysis, we observed that confounding plays a major role.
To the best of our knowledge, previous works have not approached the underlying question from a causal inference perspective. While we consider that a major strength of
M AN U
our analyses, the type of analysis provides ground for critical discussions and limitations. E.g. due to the required temporal ordering of intervention, mediator and outcome we decided to considered participants who were still in the study after one year and had a non-missing blood pressure value. Generally, these type of restrictions could introduce selection bias, especially if event and drop-out rates were markedly different between the two treatment groups during the relevant period of follow-up. However, data presented by
TE D
the SPRINT research group 9 indicate that cumulative incidence rates are fairly similar between the two groups within the first year of follow-up. Together with the comparable distribution of baseline characteristics in our analysis sample, we argue that this restriction did not greatly impact randomisation.
EP
A major criticism of causal inference approaches in general is that criteria for identification are partly relying on untestable assumptions, like the no-unmeasured confounding assumptions in the case of natural direct and indirect effects. Since the randomised
AC C
structure seems to be restored in our sample, it is reasonable to assume that there is no unmeasured confounding of the intervention-mediator and the intervention-outcome relationship. We adjusted for mediator-outcome confounding in the best possible way using measured covariates in the SPRINT study. Indeed, the results from the differently adjusted models show the importance of adjustment for mediator-outcome confounding, since randomisation of the intervention does not result in randomisation of the mediator values. However, we cannot rule out the possibility of some residual unmeasured
10
ACCEPTED MANUSCRIPT
confounding mechanism. Regarding the actual data collection for the SPRINT trial, it has been pointed out that
RI PT
the protocol for measuring the blood pressure could impact the presented results. More particularly, BP was measured by automated oscillometric devices, without any observers
in the room. In clinical practice, BP is usually measured by sphygmomanometers and
observers are normally present. The SPRINT protocol may obtain considerably lower
SC
BPs compared to the standard ambulatory recordings 24,25 . In subjects similar to the
SPRINT participants, the average difference between standard clinical BP and automated oscillometric measures recently was reported to be 12.7 mm Hg and 12.0 mm Hg for SBP
M AN U
and DBP, respectively, and the individual variation was substantial 26 . Hence, an average DBP equal to 60 mm Hg in our study may correspond to an average DBP ∼ 72 mm Hg in routine clinical practice. Nevertheless, a DBP cut-off at 70 mm Hg, rather than 60 mm Hg, has often been used to demonstrate the J-shaped relation in observational studies 7,20,21 . Still, measuring the blood pressure with cuffs, either by automatic devices or manual sphygmomanometers, may yield cuff artifacts, meaning that the true intra-
TE D
arterial pressure is lower than recorded 10 . Hence, the physiological blood pressure is not perfectly reflected in standard clinical recordings. The SPRINT study was restricted to older hypertensive patients with relatively high cardiovascular risk. Some considerable subpopulations, e.g. hypertensive patients with a
EP
history of diabetes mellitus, prior stroke, glomerular filtration rate (GFR) < 20 mm/min or proteinuria >1 g per 24 hours, were excluded from the trial. Hence, our results cannot be immediately generalized to these populations.
AC C
In conclusion, patients with DBP60 have significantly increased cardiovascular risk,
irrespective of treatment group. However, we do not find clear evidence that treatment induced DBP60 is harmful itself. We have rather shown that there exist measured con-
founders that influence both DBP and the primary outcome. Our analysis can only provide population average indirect effects that are not informative about any specific patients goups, and hence might not be immediatley applicable to individual decision making. Future research is needed to inform clinical practice on patient subgroups. The
11
ACCEPTED MANUSCRIPT
presented results suggest that an intensively treated patient with DBP60 is at increased risk of poor cardiovascular outcomes, but this increase in risk cannot be solely explained
RI PT
by an unintended treatment effect. Therefore treatment discontinuation would not be the immediate implication for a patient who presents with DBP60 . Rather, there may be
particular patient characterisitcs that are related to well tolerated diastolic hypotension
and these factors could be identified in future studies. More generally, this analysis shows
None reported.
Author contributions
M AN U
Conflict of Interest Disclosures
SC
that adjusting for mediator-outcome confounding is essential, even in RCTs.
Dr. Stensrud and Strohmaier had full access to the all data through the SPRINT data challenge and take responsibility for the integrity of the data and the accuracy of the data
TE D
analysis. Both authors contributed to formulating the research question, interpretation of the results and drafting the manuscript.
EP
Acknowledgements
We want to thank the four anonymous reviewers for their constructive feedback that
AC C
helped improving the manuscript. This manuscript was prepared using SPRINT POP Research Materials obtained form the NHLBI Biologic Specimen and Data Reppsitory Information Coordinating Center and does not necessary reflect the opinions or views of the SPRINT POP or the NHLBI.
12
ACCEPTED MANUSCRIPT
References [1] Polese A, De Cesare N, Montorsi P, et al. Upward shift of the lower range of coronary
RI PT
flow autoregulation in hypertensive patients with hypertrophy of the left ventricle.. Circulation. 1991;83:845–853.
[2] Farnett L, Mulrow CD, Linn WD, Lucey CR, Tuley MR. The J-curve phenomenon
is dangerous?. Jama. 1991;265:489–495.
SC
and the treatment of hypertension: is there a point beyond which pressure reduction
[3] Messerli FH, Mancia G, Conti CR, et al. Dogma disputed: can aggressively lowering
M AN U
blood pressure in hypertensive patients with coronary artery disease be dangerous?. Annals of internal medicine. 2006;144:884–893.
[4] Mancia G, Grassi G. Aggressive Blood Pressure Lowering Is Dangerous: The JCurveResponse to Aggressive Blood Pressure Reduction Is Dangerous: The J-Curve: Pro Side of the Argument. Hypertension. 2014;63:29–36.
TE D
[5] Banach M, Aronow WS. Blood pressure j-curve: current concepts. Current hypertension reports. 2012;14:556–566.
[6] Franklin SS, Gokhale SS, Chow VH, et al. Does Low Diastolic Blood Pressure Contribute to the Risk of Recurrent Hypertensive Cardiovascular Disease Events? Nov-
EP
elty and Significance. Hypertension. 2015;65:299–305. [7] B¨ ohm M, Schumacher H, Teo KK, et al. Achieved blood pressure and cardiovascular
AC C
outcomes in high-risk patients: results from ONTARGET and TRANSCEND trials. The Lancet. 2017.
[8] McEvoy JW, Chen Y, Rawlings A, et al. Diastolic blood pressure, subclinical myocardial damage, and cardiac events: Implications for blood pressure control. Journal of the American College of Cardiology. 2016;68:1713–1722. [9] SPRINT . A randomized trial of intensive versus standard blood-pressure control. N Engl J Med. 2015;2015:2103–2116. 13
ACCEPTED MANUSCRIPT
[10] Spence JD. Diastolic Hypotension and Myocardial Ischemia. Journal of the American College of Cardiology. 2017;69:1646–1647.
RI PT
[11] ACCORD . Effects of intensive blood-pressure control in type 2 diabetes mellitus. N Engl j Med. 2010;2010:1575–1585.
[12] Ungar A, Pepe G, Lambertucci L, et al. Low diastolic ambulatory blood pressure is associated with greater all-cause mortality in older patients with hypertension.
SC
Journal of the American Geriatrics Society. 2009;57:291–296.
[13] Pearl J. Direct and Indirect Effects. In Proceedings of the Seventeenth Conference on
M AN U
Uncertainy in Artifcial Intelligence. 2001. Pages: 411–420.
[14] VanderWeele TJ. Explanation in causal inference. Oxford University Press2015. [15] Lange T, Vansteelandt S, Bekaert M. A simple unified approach for estimating natural direct and indirect effects. American journal of epidemiology. 2012;176:190–195. [16] Kannel WB. Blood pressure as a cardiovascular risk factor: prevention and treat-
TE D
ment. Jama. 1996;275:1571–1576.
[17] Williamson JD, Supiano MA, Applegate WB, et al. Intensive vs standard blood pressure control and cardiovascular disease outcomes in adults aged 75 years: a
EP
randomized clinical trial. Jama. 2016;315:2673–2682. [18] D’Agostino RB, Belanger AJ, Kannel WB, Cruickshank JM. Relation of low diastolic
AC C
blood pressure to coronary heart disease death in presence of myocardial infarction: the Framingham Study.. BMJ. 1991;303:385–389.
[19] Verdecchia P, Angeli F, Mazzotta G, Garofoli M, Reboldi G. Aggressive Blood Pressure Lowering Is Dangerous: The J-CurveResponse to Aggressive Blood Pressure Lowering Is Dangerous: The J-Curve: Con Side of the Argument. Hypertension. 2014;63:37–40.
14
ACCEPTED MANUSCRIPT
[20] Somes GW, Pahor M, Shorr RI, Cushman WC, Applegate WB. The role of diastolic blood pressure when treating isolated systolic hypertension. Archives of Internal
RI PT
Medicine. 1999;159:2004–2009. [21] Vidal-Petiot E, Ford I, Greenlaw N, et al. Cardiovascular event rates and mortality according to achieved systolic and diastolic blood pressure in patients with stable
coronary artery disease: an international cohort study. The Lancet. 2016;388:2142–
SC
2152.
[22] Kjeldsen SE, Berge E, Bangalore S, et al. No evidence for a J-shaped curve in treated
pressure. 2016;25:83–92.
M AN U
hypertensive patients with increased cardiovascular risk: The VALUE trial. Blood
[23] Guichard JL, Desai RV, Ahmed MI, et al. Isolated diastolic hypotension and incident heart failure in older adults. Hypertension. 2011;58:895–901.
[24] Eriksson M, Carlberg B, Jansson JH. Comparison of blood pressure measurements between an automated oscillometric device and a Hawksley random-zero sphygmo-
TE D
manometer in the northern Sweden MONICA study. Blood pressure monitoring. 2012;17:164–170.
[25] Filipovsk` y J, Seidlerov´ a J, Kratochv´ıl Z, Karnosov´a P, Hronov´a M, Mayer Jr O.
EP
Automated compared to manual office blood pressure and to home blood pressure in hypertensive patients. Blood pressure. 2016;25:228–234.
AC C
[26] Agarwal R. Implications of Blood Pressure Measurement Technique for Implementation of Systolic Blood Pressure Intervention Trial (SPRINT). Journal of the American Heart Association. 2017;6:e004536.
15
ACCEPTED MANUSCRIPT
Confounders w
~ 3 DBP60 .
& Primary Event
SC
Therapy
RI PT
8SBP
M AN U
Figure 1: Directed acyclic graph (DAG) depicting the assumed causal structure. The effect of primary interest, the indirect effect through on-treatment DBP60 is highlighted in red. ’Confounders’ comprises the set of confounding variables in the models (e.g. sex, age, etc.)., SBP stands for systolic blood pressure, DBP60 refers to diastolic blood pressure below 60 mm Hg Table 1: Baseline charcteristics of the study sample.
AC C
EP
TE D
Characteristic Female sex, n (%) Age - years, mean (s.d.) Race or ethnic group Non-hispanic black Hispanic Non-hispanic white Other Baseline blood pressure - mm Hg, mean (s.d) Systolic Diastolic Baseline cardiovascular disease *, n (%) Clinical Subclinical Serum creatinine - mg/dl, mean (s.d.) Fasting total triglycerides - mg/dl, mean (s.d.) Smoking status, n (%) Never smoked Fromer smoker Current smoker Body-mass index - kg/m2, mean (s.d.) Antihypertensive agents - average no./patients *
Intensive treatment (n= 4157) 1478 (35.6) 67.9 (9.3)
Standard treatment (n=4144) 1427 (34.4) 67.8 (9.3)
1190 (28.6) 450 (10.8) 2428 (58.4) 89 (2.1)
1229 (29.7) 430 (10.4) 2426 (58.5) 59 (1.4)
139.4 (15.6) 78.2 (11.8)
139.7 (15.3) 78.1 (11.9)
681 (16.4) 216 (5.2) 1.07 (0.34) 125.2 (81.0)
679 (16.4) 215 (5.2) 1.07(0.33) 126.9 (93.8)
1838 (44.2) 1777 (42.7) 542 (12.3) 29.9 (5.8) 1.84
1856 (44.8) 1776 (44.9) 512 (12.4) 29.8 (5.7) 1.82
Details on the definition of clinical and subclinical cardiovascular disease can be found in the online supplementary material, in which the respective criteria from the SPRINT trial are stated.
16
RI PT
ACCEPTED MANUSCRIPT
SC
Table 2: Decomposition of the total intervention effect on the primary and alternative outcome into an indirect effect through low diastolic blood pressure and a direct effect.
*
TE D
M AN U
Primary outcome *(Events=349/n=8301) HR [95% CI] Total effect Direct effect Indirect effect Unadjusted model 0.70 [0.56; 0.86] 0.63 [0.50; 0.78] 1.12 [1.06; 1.18] 0.70 [0.56; 0.86] 0.65 [0.53; 0.82] 1.07 [1.01; 1.13] Model 1 a b Model 2 0.70 [0.56; 0.86] 0.65 [0.52; 0.81] 1.07 [1.01; 1.13] 0.70 [0.56; 0.86] 0.66 [0.52; 0.83] 1.05 [0.99; 1.11] Model 3 c d Model 4 0.70 [0.56; 0.86] 0.66 [0.53; 0.82] 1.04 [0.98; 1.10] Primary outcome without stroke ** (Events=286/n=8322) Unadjusted model 0.67 [0.53; 0.85] 0.60 [0.47; 0.77] 1.12 [1.05; 1.19] 0.67 [0.53; 0.85] 0.63 [0.49; 0.80] 1.07[1.01; 1.14] Model 1 a Model 2 b 0.67 [0.53; 0.85] 0.62 [0.49; 0.79] 1.07[1.01; 1.14] Model 3 c 0.67 [0.53; 0.85] 0.64 [0.50; 0.81] 1.05 [0.99; 1.12] Model 4 d 0.67 [0.53; 0.85] 0.63 [0.49; 0.81] 1.04 [0.98; 1.11]
AC C
EP
Primary outcome: myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure or death from cardiovascular causes ** Primary outcome (footnote a) without stroke a Adjusted for sex, age (years) , race (non-hispanic black, hispanic, non-hispanic white, other) b Additionally adjusted for smoking status (never, former, current) and BMI (<18.5; 18.5-24.9, 25-29.9, 30-34.9, 35+) c Additionally adjusted for baseline clinical and subclinical cardiovascular diseae and number of blood pressure agents d Additionally adjusted for serum creatinine (mg/dl) and fasting total triglycerides (mg/dl)
17
RI PT
ACCEPTED MANUSCRIPT
SC
Table 3: Effect decompostion of the intervention effect on the primary outcome* stratified by baseline age.
Indirect effect 1.10 [1.01; 1.19] 1.08 [1.00; 1.17] 1.08 [1.00; 1.17] 1.07 [0.98; 1.16] 1.07 [0.98; 1.16]
*
TE D
M AN U
Baseline age <75 years (Events=198/n=5997) HR [95% CI] Total effect Direct effect Unadjusted model 0.82 [0.62; 1.09] 0.75 [0.56; 1.01] Model 1 a 0.82 [0.62; 1.09] 0.76 [0.57; 1.02] 0.82 [0.62; 1.09] 0.76 [0.57; 1.02] Model 2 b Model 3 c 0.82 [0.62; 1.09] 0.77 [0.57; 1.03] 0.82 [0.62; 1.09] 0.77 [0.57; 1.03] Model 4 d Baseline age ≥ 75 years (Events=151/n=2304) Unadjusted model 0.55 [0.39; 0.76] 0.52 [0.38; 0.73] Model 1 a 0.55 [0.39; 0.76] 0.52 [0.37; 0.73] Model 2 b 0.55 [0.39; 0.76] 0.53 [0.37; 0.74] Model 3 c 0.55 [0.39; 0.76] 0.54 [0.38; 0.76] Model 4 d 0.55 [0.39; 0.76] 0.53 [0.37; 0.74]
1.06 1.05 1.04 1.02 1.01
[0.98; [0.97; [0.96; [0.94; [0.93;
1.15] 1.14] 1.13] 1.10] 1.09]
AC C
EP
Primary outcome: myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure or death from cardiovascular causes a Adjusted for sex, age (years) , race (non-hispanic black, hispanic, non-hispanic white, other) b Additionally adjusted for smoking status (never, former, current) and BMI (<18.5; 18.5-24.9, 25-29.9, 30-34.9, 35+) c Additionally adjusted for baseline clinical and subclinical cardiovascular diseae and number of blood pressure agents d Additionally adjusted for serum creatinine (mg/dl) and fasting total triglycerides (mg/dl)
18
SC
RI PT
ACCEPTED MANUSCRIPT
Table 4: Effect decompostion of the intervention effect on the primary outcome* stratified by baseline diastolic blood pressure.
*
TE D
M AN U
Baseline diastolic BP between 60 and 80 mm Hg (Events=185/n=4066) HR [95% CI] Total effect Direct effect Indirect effect Unadjusted model 0.59 [0.44; 0.80] 0.51 [0.37; 0.69] 1.19 [1.07; 1.31] Model 1 ax 0.59 [0.44; 0.80] 0.53 [0.39; 0.73] 1.13 [1.01; 1.25] Model 2 a 0.59 [0.44; 0.80] 0.52 [0.38; 0.72] 1.14 [1.02; 1.27] Model 3 b 0.59 [0.44; 0.80] 0.53 [0.39; 0.73] 1.12 [1.00; 1.25] Model 4 c 0.59 [0.44; 0.80] 0.54 [0.39; 0.74] 1.11 [0.99; 1.23] Baseline diastolic BP above 80 mm Hg (Events=132 /n=3774) Unadjusted model 0.81 [0.58; 1.14] 0.78 [0.55; 1.12] 1.04 [0.95; 1.13] Model 1 a 0.81 [0.58; 1.14] 0.80 [0.56; 1.13] 1.02 [0.94; 1.11] 0.81 [0.58; 1.14] 0.80 [0.56; 1.14] 1.01 [0.93; 1.10] Model 2 b Model 3 c 0.81 [0.58; 1.14] 0.80 [0.56; 1.14] 1.01 [0.93; 1.10] 0.81 [0.58; 1.14] 0.78 [0.55; 1.12] 1.01 [0.93; 1.10] Model 4 d
AC C
EP
Primary outcome: myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure or death from cardiovascular causes a Adjusted for sex, age (years) , race (non-hispanic black, hispanic, non-hispanic white, other) b Additionally adjusted for smoking status (never, former, current) and BMI (<18.5; 18.5-24.9, 25-29.9, 30-34.9, 35+) c Additionally adjusted for baseline clinical and subclinical cardiovascular diseae and number of blood pressure agents d Additionally adjusted for serum creatinine (mg/dl) and fasting total triglycerides (mg/dl)
19
ACCEPTED MANUSCRIPT Highlights: A causal mediation analysis using the observational part of a large scale randomised controlled trial (RCT) is described.
•
There is no significant effect of therapy-induced low diastolic blood pressure on risk of certain cardiovascular outcomes.
•
Adjusting for mediator-outcome confounding is essential, even in RCTs.
AC C
EP
TE D
M AN U
SC
RI PT
•
S0021-9150(17)31194-2
DOI:
10.1016/j.atherosclerosis.2017.07.019
Reference:
ATH 15144
To appear in:
Atherosclerosis
Received Date: 7 March 2017 Revised Date:
12 July 2017
Accepted Date: 18 July 2017
Please cite this article as: Stensrud MJ, Strohmaier S, Diastolic hypotension due to intensive blood pressure therapy: Is it harmful?, Atherosclerosis (2017), doi: 10.1016/j.atherosclerosis.2017.07.019. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT
RI PT
Diastolic hypotension due to intensive blood pressure therapy: Is it harmful? Mats Julius Stensrud1,2 and Susanne Strohmaier2,3,*
Oslo Centre for Biostatistics and Epidemiology, Department for Biostatistics, University of Oslo, Oslo, Norway 2 Diakonhjemmet hospital, Oslo, Norway 3 Channing Division of Network Medicine, Brigham and Women’s Hospital, and Harvard Medical School, Boston, USA * Corresponding author ([email protected])
M AN U
SC
1
Abstract
Background and aims: Reducing the diastolic blood pressure (DBP) below a certain threshold may lead to inadequate organ perfusion. This raises some concerns, because pharmacotherapy reduces both systolic and diastolic pressure. We
TE D
aimed to investigate whether a pathway from intensive systolic blood pressure (SBP) treatment influences cardiovascular outcomes by inducing too low DBP. Methods: We had access to data from the Systolic Blood Pressure Intervention Trial (SPRINT) including 9361 patients with a SBP of 130 mm Hg or higher and
EP
an increased cardiovascular risk. In a formal mediation analysis we investigated whether the effect of intense (target SBP: 120 mm Hg) vs. standard (target SBP: 140 mm Hg) intervention on a composite endpoint would be mediated through an
AC C
indirect, potentially harmful, effect through low DBP (< 60 mm Hg). Results: Adjusting for treatment, we find that low DBP per se is associated
with poor cardiovascular outcomes (HR 1.90 (95%CI [1.46, 2.47]). However, in a formal mediation analyses we observed that the unadjusted indirect effect of intensive blood pressure treatment going through low DBP of HR 1.12 (95%CI [1.06, 1.18]) attenuates to a statistically non significant effect of HR 1.04 (95%CI [0.98, 1.10]) after adjustment for important covariates, suggesting that the mere association is considerably confounded. Conclusions:
The increased risk in subjects with diastolic pressure below 60
1
ACCEPTED MANUSCRIPT
cannot fully be explained by the intensive treatment itself, but may be due to other measured factors. More generally, this analysis shows that adjusting for mediator-
AC C
EP
TE D
M AN U
SC
RI PT
outcome confounding is essential, even in RCTs.
2
ACCEPTED MANUSCRIPT
1
Introduction
It is debated whether low diastolic blood pressure (DBP) leads to undesirable cardiovascu-
RI PT
lar outcomes. From a physiological perspective, coronary perfusion occurs mostly during
diastole, and low DBP may reduce the coronary flow 1 . Furthermore, a J-shaped relation between blood pressure and cardiovascular events has been observed 2–4 , in which DBP < 60 (hereby DBP60 ) is associated with increased risk of myocardial infarction,
SC
heart failure and cardiovascular death 5,6 . Most recently, the J-shaped relation was also observed in data from the ONTARGET and TRANSCEND trials, suggesting that DBP below 70 is associated with increased risk of cardiovascular outcomes 7 . Furthermore,
M AN U
McEvoy et al 8 found that low DBP was associated with higher levels of Troponin T, which indicate damage of the myocardium.
Intensive treatment of the systolic blood pressure (SBP) is associated with reduced risk of major cardiovascular outcomes 9,10 . However, a reduction of the SBP is intimately connected to a reduction of the DBP, as also seen in major clinical trials 9,11 . Hence, intensive blood pressure therapy could lead to excessive DBP reduction, which could be
TE D
associated with increased cardiovascular risk 7,8,12 . Motivated by the ongoing debate on intensive blood pressure therapy, we aimed to investigate whether a pathway from intensive SBP treatment influences cardiovascular outcomes through DBP60 in a formal mediation analysis using data collected within the
EP
Systolic Blood Pressure Intervention Trial (SPRINT)
Material and methods
AC C
2
2.1
Data source and study population
We were able to access individual level data from the Systolic Blood Pressure Intervention Trial (SPRINT) 9 through the SPRINT data challenge (https://challenge.nejm.org). In brief, SPRINT was a randomized, controlled, open-label, multi-center trial conducted in the United States. As reported in more detail in the original New England Journal of Medicine article 9 , 9361 participants with a systolic blood pressure of 130 mm 3
ACCEPTED MANUSCRIPT
Hg or higher and an increased cardiovascular risk, without diabetes were randomized to either an intensive or standard treatment strategy. For participants in the intensive treat-
RI PT
ment group the systolic blood pressure target was 120 mm Hg, while the target in the standard treatment group was 140 mm Hg. Participants were asked to return to the study
site on a monthly basis for the first 3 months, and every 3 months thereafter to monitor their blood pressure. Medications were adjusted monthly based on current and previ-
SC
ous systolic blood pressure values to achieve the respective target values. A composite
endpoint of myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure, or death from cardiovascular causes
M AN U
was considered as primary outcome.
The objective of the present study was to investigate whether the effect of intensive vs. standard treatment on the primary outcome is mediated by a potentially harmful indirect effect through too low diastolic blood pressure, which we defined as diastolic blood pressure below 60 mm Hg (DBP60 ). Since addressing mediation questions requires some temporal ordering (the treatment must occur before the mediator and in turn, the
TE D
mediator must be measured before the outcome), we conducted our analysis among 8301 participants who were still at risk for the primary outcome one year after randomization and who had a non-missing blood pressure measurements at that time since we consider DBP60 one year after randomization as our mediator.
EP
All participants in the SPRINT study provided written informed consent. Before entering the SPRINT data challenge we were granted permission by the Regional Committee for Medical & Health Research Ethics, South East Norway to apply to using the
AC C
data without a formal application to the committee.
2.2
Statistical Methods
Baseline characteristics and risk factors of the study population were described by mean and standard deviation (s.d.) for continuous variables, and by counts and percentages for categorical data. The assumed causal structure is displayed in Figure 1, where the relationship between
4
ACCEPTED MANUSCRIPT
the variables are depicted by arrows that indicate the direction of the effect. Our aim was to separate the total effect of intensive therapy compared to standard therapy into
RI PT
two separate pathways: i) A direct path that summarizes all effects not going through DBP60 , e.g. a beneficial effect of reducing SBP. ii) An indirect path, potentially harmful, that acts through on-treatment DBP60 .
First, we evaluated whether the criteria for formal mediation analysis were satisfied.
SC
That is, we investigated the association between treatment and DBP60 in a binomial
model to see whether the first criterion for mediation analysis was satisfied. Second, we fitted a Cox model for the primary outcome including treatment and DBP60 , assessing
M AN U
whether DBP60 is associated with the primary outcome.
Based on the results from these two models, we decided on performing a formal mediation analysis to estimate so called natural direct and indirect effects. These natural effects are defined within the counterfactual framework using non parametric structural equation models 13 which provide the theoretical framework for causal mediation problems. In particular, they allow to formulate the decomposition of the total effect of a given
TE D
intervention into a direct and indirect effect independent of any statistical model assumptions. Conceptually, for the presented analysis, the natural direct effect would compare the risk of the primary outcome for participants in the intensive treatment group compared to those in the standard treatment group if their mediator level (DBP60 ) had taken
EP
the value that would have been observed if they had been in the standard treatment group. Accordingly, the natural indirect effect would compare the risk for the primary outcome if the mediator takes the value it would naturally take under intensive therapy
AC C
compared to the risk of the primary outcome if the mediator takes the value it would take under standard therapy even though the treatment was kept at intensive therapy. For a more elaborate introduction to causal mediation analysis using these counterfactual formulations in biological applications, we refer to the recent book by VanderWeeele 14 . Even though the definitions of these effects do not make any reference to particular model assumptions, there are some assumptions required to be able to identify these effects from observed data. More particularly, one has to assume that there is no unmea-
5
ACCEPTED MANUSCRIPT
sured confounding of the intervention - mediator relationship, the intervention - outcome relationship and the mediator-outcome relationship, and that there are no intertwined
RI PT
pathways. To estimate the described effects we used a method suggested by Lange et al. 15 based
on marginal structural models. Heuristically, the natural direct and indirect effects can be directly extracted from any type of outcome model (a Cox model for the present
SC
analysis) fitted to an augmented data set where weights are assigned to each row, which
can be calculated using an appropriate model for the mediator incorporating all relevant confounders. For details, please refer to the paper by Lange et al. 15 that also provides
M AN U
code for implementation in R and SAS.
We performed crude analyses and additionally considered different sets of mediatoroutcome confounding variables. In ’Model 1’ we adjusted for age, sex and race, in ’Model 2’ we additionally adjusted for the lifestyle characteristics, smoking status and categories of body-mass index. In ’Model 3’ we additionally adjusted for participants’ existing baseline cardiovascular disease and number of blood pressure agents, and in ’Model 4’ we
TE D
additionally adjusted for biomarkers of cardiovascular risk, creatinine and triglycerides. Cut off values for categories and units are given in the Tables 2-4. In sensitivity analyses we removed the event of stroke from the outcome definition and hence considered a composite outcome comprising of myocardial infarction, acute coro-
EP
nary syndrome not resulting in myocardial infarction, acute decompensated heart failure, or death from cardiovascular causes. In stratified analyses we evaluated total, direct and indirect effects in participants older or equal to 75 years of age and in participants younger
AC C
than 75 years. Furthermore, we stratified by baseline diastolic blood pressure, considering subgroups of participants with baseline diastolic blood pressure between 60 and 80 mm Hg and with baseline diastolic blood pressure above 80 mm Hg. Robust standard errors were used for calculation of 95% confidence intervals (CIs).
All analyses were performed with R version 3.2.0.
6
ACCEPTED MANUSCRIPT
3
Results
In Table 1 baseline characteristics of our study sample are presented. Despite the neces-
RI PT
sary restrictions for the present analysis (patients had to still be at risk for the primary event and had to have a blood pressure measurement at year 1), the distribution of the
characteristics across treatment groups is still balanced (Table 1). The median follow-up time was 3.26 years.
SC
When examining the criteria for mediation analysis, we observed the odds ratio of
DBP60 to be 3.14 (95% CI [2.74, 3.60]), comparing intensive and standard treatment groups. In a model for the outcome including the treatment and the mediator DBP60 ,
M AN U
we found a HR of 1.90 (95% CI [1.46, 2.47]) for the mediator. Since both the essential criteria for mediation were met, we proceeded with a formal mediation analysis. In a crude mediation model without adjusting for mediator-outcome confounders, the total effect of 0.70 ((95% CI [0.56; 0.86]) could be decomposed into a direct effect of HR of 0.63 (95% CI [0.50, 0.78]), and the indirect effect through DBP60 of 1.12 (95% CI [1.06, 1.18]) (see Table 2). Adjusting for basic characteristics changed the decomposition to a direct
TE D
effect of HR of 0.65 (95% CI [0.52, 0.81]) and an indirect effect of 1.07 (95% CI [1.01, 1.13]), additionally adjusting for baseline cardiovascular risk and biological markers did not change the point estimates markedly, however confidence intervals for the indirect effect included 1 (fully adjusted ’Model 4’: direct effect HR 0.66 (95% CI [0.53, 0.82])
EP
and indirect effect of 1.04 (95% CI [0.98, 1.10]). When considering the alternative outcome definition we observed a slightly stronger
AC C
total effect (HR: 0.67 (95% CI [0.53, 0.85])) and very similar patterns regarding direct and indirect effects and adjustment for different sets of confounding variables. In stratified analyses (Table 2) we observed a stronger total effect in participants 75
years or older at baseline (HR: 0.55 (95% CI [0.39, 0.76])), but the indirect effect was not found to be statistically significant in any of the considered models. In participants younger than 75 years the total treatment effect was less pronounced and not significant (HR: 0.82 (95% CI [0.62, 1.09])). When stratifying according to baseline diastolic blood pressure, we found a relatively
7
ACCEPTED MANUSCRIPT
strong total effect (HR: 0.59 (95% CI [0.44, 0.80])) among participants with a baseline diastolic blood pressure between 60 and 80 mm Hg, which could be decomposed into a
RI PT
direct effect of HR: 0.51 (95% CI [0.37; 0.69]) and an indirect effect of HR: 1.19 (95% CI [1.07, 1.31]). Adjustment for confounding altered direct and indirect effects, resulting in
a direct effect of HR: 0.54 (95% CI [0.39, 0.74]) and an indirect effect of HR: 1.11 (95% CI [0.99, 1.23]) in the fully adjusted model. Again the total effect among participants
4
Discussion
M AN U
statistically significant (HR: 0.81 (95% CI [0.58, 1.14])).
SC
with baseline diastolic blood pressure above 80 mm Hg was less pronounced and not
Our analysis of the SPRINT data does not suggest that therapy induced DBP60 strongly increases cardiovascular risk. After adjustment for mediator-outcome confounding factors, there was no significant effect of therapy induced DBP60 on the primary outcome of myocardial infarction, acute coronary syndrome not resulting in myocardial infarction,
TE D
stroke, acute decompensated heart failure or death from cardiovascular causes. To exclusively consider cardiac events, we also removed stroke from the primary outcome in a secondary analysis, but no significant effect of DBP60 was obtained. The consequences of low DBP may be particularly relevant to elderly patients, who
EP
often suffer from isolated systolic hypertension 16,17 . Hence, we also performed a subgroup analysis of subjects > 75 years at baseline, providing no clear evidence for a harmful effect of DBP60 on the primary outcome. Further, analysis within different groups of patients
AC C
defined by baseline diastolic blood pressure, showed no evidence of effect modification. The hypothesis that low blood pressure might have harmful effects has been debated
for decades 2,4,18,19 , and previous studies have also aimed to investigate the potential
harmful effect of too aggressive blood pressure treatment 7,8,20–23 . Among the most recent, McEvoy et al 8 found that DBP60 is associated with damage
to the myocardium, indicated by the release of Troponin T in the Atherosclerosis Risk in Communities (ARIC) cohort. They also found an association between DBP60 and subsequent heart disease, in particular in subjects with elevated pulse pressure. 8
ACCEPTED MANUSCRIPT
Similarly, Guichard et al 23 found that isolated hypotension was associated with heart failure in elderly patients, using propensity score matching. However, the results by
RI PT
Guichard et al are not directly comparable to our analyses since they did not have information on intensity of antihypertensive drugs.
Vidal-Petiot et al. 21 studied patients with stable coronary artery disease who were
treated for hypertension. They found that systolic blood pressure of less than 120 mm
SC
Hg and diastolic blood pressure of less than 70 mm Hg were linked to unfavorable cardio-
vascular outcomes and hence suggested that caution should be taken regarding the use of blood pressure-lowering treatment in this patient group.
M AN U
In contrast Kjeldsen et al. 22 found only too high systolic (≥ 150) and diastolic (≥ 90) blood pressure values to be associated with increased risk for cardiac events among patients treated with either valsartan or amlodipine, and hence no evidence of a J-shaped relation.
More recently, B¨ ohm et al. 7 assessed the association between different levels of SBP and DBP on cardiovascular outcomes in an observational study among patients receiving
TE D
ramipril, telmisartan, or their combination. In contrast to the SPRINT trial, their observational study found that SBP below 120 mm Hg was associated increased cardiovascular risk. They also reported that both baseline and follow-up DBP below 70 mm Hg were associated with increased cardiovascular risk, evaluating a similar primary outcome as in
EP
the SPRINT trial. It is difficult to evaluate whether their results may be confounded by background comorbidities that may yield low DBP and increase the risk of cardiovascular events based on their presented results.
AC C
Our results are in line with the results by McEvoy 8 , as we also find that DBP60 per
se is associated with severe cardiovascular outcomes in the SPRINT study. Also the results by Vidal-Petiot et al. 21 and Bohm et al 7 point in the same direction, however
with differnet cut off values for diastolic blood pressure. Still, these associations are not necessarily causal, as also recognised by McEvoy et al 8 . Furthermore, the methodological techniques in the mentioned articles cannot adequately differentiate between different causes of DBP.
9
ACCEPTED MANUSCRIPT
In particular, results from Cox proportional hazards models that include baseline covariates will not yield information on the treatment induced effect of low DBP. Such
RI PT
analyses rather indicate associations between DBP60 (possible adjusted for blood pressure treatment) and time to cardiovascular outcomes, which is a different research question. Nevertheless, we have included results from more conventional Cox models in the supplemental material, which may be more directly compared to previous studies. As for the
SC
main analysis, we observed that confounding plays a major role.
To the best of our knowledge, previous works have not approached the underlying question from a causal inference perspective. While we consider that a major strength of
M AN U
our analyses, the type of analysis provides ground for critical discussions and limitations. E.g. due to the required temporal ordering of intervention, mediator and outcome we decided to considered participants who were still in the study after one year and had a non-missing blood pressure value. Generally, these type of restrictions could introduce selection bias, especially if event and drop-out rates were markedly different between the two treatment groups during the relevant period of follow-up. However, data presented by
TE D
the SPRINT research group 9 indicate that cumulative incidence rates are fairly similar between the two groups within the first year of follow-up. Together with the comparable distribution of baseline characteristics in our analysis sample, we argue that this restriction did not greatly impact randomisation.
EP
A major criticism of causal inference approaches in general is that criteria for identification are partly relying on untestable assumptions, like the no-unmeasured confounding assumptions in the case of natural direct and indirect effects. Since the randomised
AC C
structure seems to be restored in our sample, it is reasonable to assume that there is no unmeasured confounding of the intervention-mediator and the intervention-outcome relationship. We adjusted for mediator-outcome confounding in the best possible way using measured covariates in the SPRINT study. Indeed, the results from the differently adjusted models show the importance of adjustment for mediator-outcome confounding, since randomisation of the intervention does not result in randomisation of the mediator values. However, we cannot rule out the possibility of some residual unmeasured
10
ACCEPTED MANUSCRIPT
confounding mechanism. Regarding the actual data collection for the SPRINT trial, it has been pointed out that
RI PT
the protocol for measuring the blood pressure could impact the presented results. More particularly, BP was measured by automated oscillometric devices, without any observers
in the room. In clinical practice, BP is usually measured by sphygmomanometers and
observers are normally present. The SPRINT protocol may obtain considerably lower
SC
BPs compared to the standard ambulatory recordings 24,25 . In subjects similar to the
SPRINT participants, the average difference between standard clinical BP and automated oscillometric measures recently was reported to be 12.7 mm Hg and 12.0 mm Hg for SBP
M AN U
and DBP, respectively, and the individual variation was substantial 26 . Hence, an average DBP equal to 60 mm Hg in our study may correspond to an average DBP ∼ 72 mm Hg in routine clinical practice. Nevertheless, a DBP cut-off at 70 mm Hg, rather than 60 mm Hg, has often been used to demonstrate the J-shaped relation in observational studies 7,20,21 . Still, measuring the blood pressure with cuffs, either by automatic devices or manual sphygmomanometers, may yield cuff artifacts, meaning that the true intra-
TE D
arterial pressure is lower than recorded 10 . Hence, the physiological blood pressure is not perfectly reflected in standard clinical recordings. The SPRINT study was restricted to older hypertensive patients with relatively high cardiovascular risk. Some considerable subpopulations, e.g. hypertensive patients with a
EP
history of diabetes mellitus, prior stroke, glomerular filtration rate (GFR) < 20 mm/min or proteinuria >1 g per 24 hours, were excluded from the trial. Hence, our results cannot be immediately generalized to these populations.
AC C
In conclusion, patients with DBP60 have significantly increased cardiovascular risk,
irrespective of treatment group. However, we do not find clear evidence that treatment induced DBP60 is harmful itself. We have rather shown that there exist measured con-
founders that influence both DBP and the primary outcome. Our analysis can only provide population average indirect effects that are not informative about any specific patients goups, and hence might not be immediatley applicable to individual decision making. Future research is needed to inform clinical practice on patient subgroups. The
11
ACCEPTED MANUSCRIPT
presented results suggest that an intensively treated patient with DBP60 is at increased risk of poor cardiovascular outcomes, but this increase in risk cannot be solely explained
RI PT
by an unintended treatment effect. Therefore treatment discontinuation would not be the immediate implication for a patient who presents with DBP60 . Rather, there may be
particular patient characterisitcs that are related to well tolerated diastolic hypotension
and these factors could be identified in future studies. More generally, this analysis shows
None reported.
Author contributions
M AN U
Conflict of Interest Disclosures
SC
that adjusting for mediator-outcome confounding is essential, even in RCTs.
Dr. Stensrud and Strohmaier had full access to the all data through the SPRINT data challenge and take responsibility for the integrity of the data and the accuracy of the data
TE D
analysis. Both authors contributed to formulating the research question, interpretation of the results and drafting the manuscript.
EP
Acknowledgements
We want to thank the four anonymous reviewers for their constructive feedback that
AC C
helped improving the manuscript. This manuscript was prepared using SPRINT POP Research Materials obtained form the NHLBI Biologic Specimen and Data Reppsitory Information Coordinating Center and does not necessary reflect the opinions or views of the SPRINT POP or the NHLBI.
12
ACCEPTED MANUSCRIPT
References [1] Polese A, De Cesare N, Montorsi P, et al. Upward shift of the lower range of coronary
RI PT
flow autoregulation in hypertensive patients with hypertrophy of the left ventricle.. Circulation. 1991;83:845–853.
[2] Farnett L, Mulrow CD, Linn WD, Lucey CR, Tuley MR. The J-curve phenomenon
is dangerous?. Jama. 1991;265:489–495.
SC
and the treatment of hypertension: is there a point beyond which pressure reduction
[3] Messerli FH, Mancia G, Conti CR, et al. Dogma disputed: can aggressively lowering
M AN U
blood pressure in hypertensive patients with coronary artery disease be dangerous?. Annals of internal medicine. 2006;144:884–893.
[4] Mancia G, Grassi G. Aggressive Blood Pressure Lowering Is Dangerous: The JCurveResponse to Aggressive Blood Pressure Reduction Is Dangerous: The J-Curve: Pro Side of the Argument. Hypertension. 2014;63:29–36.
TE D
[5] Banach M, Aronow WS. Blood pressure j-curve: current concepts. Current hypertension reports. 2012;14:556–566.
[6] Franklin SS, Gokhale SS, Chow VH, et al. Does Low Diastolic Blood Pressure Contribute to the Risk of Recurrent Hypertensive Cardiovascular Disease Events? Nov-
EP
elty and Significance. Hypertension. 2015;65:299–305. [7] B¨ ohm M, Schumacher H, Teo KK, et al. Achieved blood pressure and cardiovascular
AC C
outcomes in high-risk patients: results from ONTARGET and TRANSCEND trials. The Lancet. 2017.
[8] McEvoy JW, Chen Y, Rawlings A, et al. Diastolic blood pressure, subclinical myocardial damage, and cardiac events: Implications for blood pressure control. Journal of the American College of Cardiology. 2016;68:1713–1722. [9] SPRINT . A randomized trial of intensive versus standard blood-pressure control. N Engl J Med. 2015;2015:2103–2116. 13
ACCEPTED MANUSCRIPT
[10] Spence JD. Diastolic Hypotension and Myocardial Ischemia. Journal of the American College of Cardiology. 2017;69:1646–1647.
RI PT
[11] ACCORD . Effects of intensive blood-pressure control in type 2 diabetes mellitus. N Engl j Med. 2010;2010:1575–1585.
[12] Ungar A, Pepe G, Lambertucci L, et al. Low diastolic ambulatory blood pressure is associated with greater all-cause mortality in older patients with hypertension.
SC
Journal of the American Geriatrics Society. 2009;57:291–296.
[13] Pearl J. Direct and Indirect Effects. In Proceedings of the Seventeenth Conference on
M AN U
Uncertainy in Artifcial Intelligence. 2001. Pages: 411–420.
[14] VanderWeele TJ. Explanation in causal inference. Oxford University Press2015. [15] Lange T, Vansteelandt S, Bekaert M. A simple unified approach for estimating natural direct and indirect effects. American journal of epidemiology. 2012;176:190–195. [16] Kannel WB. Blood pressure as a cardiovascular risk factor: prevention and treat-
TE D
ment. Jama. 1996;275:1571–1576.
[17] Williamson JD, Supiano MA, Applegate WB, et al. Intensive vs standard blood pressure control and cardiovascular disease outcomes in adults aged 75 years: a
EP
randomized clinical trial. Jama. 2016;315:2673–2682. [18] D’Agostino RB, Belanger AJ, Kannel WB, Cruickshank JM. Relation of low diastolic
AC C
blood pressure to coronary heart disease death in presence of myocardial infarction: the Framingham Study.. BMJ. 1991;303:385–389.
[19] Verdecchia P, Angeli F, Mazzotta G, Garofoli M, Reboldi G. Aggressive Blood Pressure Lowering Is Dangerous: The J-CurveResponse to Aggressive Blood Pressure Lowering Is Dangerous: The J-Curve: Con Side of the Argument. Hypertension. 2014;63:37–40.
14
ACCEPTED MANUSCRIPT
[20] Somes GW, Pahor M, Shorr RI, Cushman WC, Applegate WB. The role of diastolic blood pressure when treating isolated systolic hypertension. Archives of Internal
RI PT
Medicine. 1999;159:2004–2009. [21] Vidal-Petiot E, Ford I, Greenlaw N, et al. Cardiovascular event rates and mortality according to achieved systolic and diastolic blood pressure in patients with stable
coronary artery disease: an international cohort study. The Lancet. 2016;388:2142–
SC
2152.
[22] Kjeldsen SE, Berge E, Bangalore S, et al. No evidence for a J-shaped curve in treated
pressure. 2016;25:83–92.
M AN U
hypertensive patients with increased cardiovascular risk: The VALUE trial. Blood
[23] Guichard JL, Desai RV, Ahmed MI, et al. Isolated diastolic hypotension and incident heart failure in older adults. Hypertension. 2011;58:895–901.
[24] Eriksson M, Carlberg B, Jansson JH. Comparison of blood pressure measurements between an automated oscillometric device and a Hawksley random-zero sphygmo-
TE D
manometer in the northern Sweden MONICA study. Blood pressure monitoring. 2012;17:164–170.
[25] Filipovsk` y J, Seidlerov´ a J, Kratochv´ıl Z, Karnosov´a P, Hronov´a M, Mayer Jr O.
EP
Automated compared to manual office blood pressure and to home blood pressure in hypertensive patients. Blood pressure. 2016;25:228–234.
AC C
[26] Agarwal R. Implications of Blood Pressure Measurement Technique for Implementation of Systolic Blood Pressure Intervention Trial (SPRINT). Journal of the American Heart Association. 2017;6:e004536.
15
ACCEPTED MANUSCRIPT
Confounders w
~ 3 DBP60 .
& Primary Event
SC
Therapy
RI PT
8SBP
M AN U
Figure 1: Directed acyclic graph (DAG) depicting the assumed causal structure. The effect of primary interest, the indirect effect through on-treatment DBP60 is highlighted in red. ’Confounders’ comprises the set of confounding variables in the models (e.g. sex, age, etc.)., SBP stands for systolic blood pressure, DBP60 refers to diastolic blood pressure below 60 mm Hg Table 1: Baseline charcteristics of the study sample.
AC C
EP
TE D
Characteristic Female sex, n (%) Age - years, mean (s.d.) Race or ethnic group Non-hispanic black Hispanic Non-hispanic white Other Baseline blood pressure - mm Hg, mean (s.d) Systolic Diastolic Baseline cardiovascular disease *, n (%) Clinical Subclinical Serum creatinine - mg/dl, mean (s.d.) Fasting total triglycerides - mg/dl, mean (s.d.) Smoking status, n (%) Never smoked Fromer smoker Current smoker Body-mass index - kg/m2, mean (s.d.) Antihypertensive agents - average no./patients *
Intensive treatment (n= 4157) 1478 (35.6) 67.9 (9.3)
Standard treatment (n=4144) 1427 (34.4) 67.8 (9.3)
1190 (28.6) 450 (10.8) 2428 (58.4) 89 (2.1)
1229 (29.7) 430 (10.4) 2426 (58.5) 59 (1.4)
139.4 (15.6) 78.2 (11.8)
139.7 (15.3) 78.1 (11.9)
681 (16.4) 216 (5.2) 1.07 (0.34) 125.2 (81.0)
679 (16.4) 215 (5.2) 1.07(0.33) 126.9 (93.8)
1838 (44.2) 1777 (42.7) 542 (12.3) 29.9 (5.8) 1.84
1856 (44.8) 1776 (44.9) 512 (12.4) 29.8 (5.7) 1.82
Details on the definition of clinical and subclinical cardiovascular disease can be found in the online supplementary material, in which the respective criteria from the SPRINT trial are stated.
16
RI PT
ACCEPTED MANUSCRIPT
SC
Table 2: Decomposition of the total intervention effect on the primary and alternative outcome into an indirect effect through low diastolic blood pressure and a direct effect.
*
TE D
M AN U
Primary outcome *(Events=349/n=8301) HR [95% CI] Total effect Direct effect Indirect effect Unadjusted model 0.70 [0.56; 0.86] 0.63 [0.50; 0.78] 1.12 [1.06; 1.18] 0.70 [0.56; 0.86] 0.65 [0.53; 0.82] 1.07 [1.01; 1.13] Model 1 a b Model 2 0.70 [0.56; 0.86] 0.65 [0.52; 0.81] 1.07 [1.01; 1.13] 0.70 [0.56; 0.86] 0.66 [0.52; 0.83] 1.05 [0.99; 1.11] Model 3 c d Model 4 0.70 [0.56; 0.86] 0.66 [0.53; 0.82] 1.04 [0.98; 1.10] Primary outcome without stroke ** (Events=286/n=8322) Unadjusted model 0.67 [0.53; 0.85] 0.60 [0.47; 0.77] 1.12 [1.05; 1.19] 0.67 [0.53; 0.85] 0.63 [0.49; 0.80] 1.07[1.01; 1.14] Model 1 a Model 2 b 0.67 [0.53; 0.85] 0.62 [0.49; 0.79] 1.07[1.01; 1.14] Model 3 c 0.67 [0.53; 0.85] 0.64 [0.50; 0.81] 1.05 [0.99; 1.12] Model 4 d 0.67 [0.53; 0.85] 0.63 [0.49; 0.81] 1.04 [0.98; 1.11]
AC C
EP
Primary outcome: myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure or death from cardiovascular causes ** Primary outcome (footnote a) without stroke a Adjusted for sex, age (years) , race (non-hispanic black, hispanic, non-hispanic white, other) b Additionally adjusted for smoking status (never, former, current) and BMI (<18.5; 18.5-24.9, 25-29.9, 30-34.9, 35+) c Additionally adjusted for baseline clinical and subclinical cardiovascular diseae and number of blood pressure agents d Additionally adjusted for serum creatinine (mg/dl) and fasting total triglycerides (mg/dl)
17
RI PT
ACCEPTED MANUSCRIPT
SC
Table 3: Effect decompostion of the intervention effect on the primary outcome* stratified by baseline age.
Indirect effect 1.10 [1.01; 1.19] 1.08 [1.00; 1.17] 1.08 [1.00; 1.17] 1.07 [0.98; 1.16] 1.07 [0.98; 1.16]
*
TE D
M AN U
Baseline age <75 years (Events=198/n=5997) HR [95% CI] Total effect Direct effect Unadjusted model 0.82 [0.62; 1.09] 0.75 [0.56; 1.01] Model 1 a 0.82 [0.62; 1.09] 0.76 [0.57; 1.02] 0.82 [0.62; 1.09] 0.76 [0.57; 1.02] Model 2 b Model 3 c 0.82 [0.62; 1.09] 0.77 [0.57; 1.03] 0.82 [0.62; 1.09] 0.77 [0.57; 1.03] Model 4 d Baseline age ≥ 75 years (Events=151/n=2304) Unadjusted model 0.55 [0.39; 0.76] 0.52 [0.38; 0.73] Model 1 a 0.55 [0.39; 0.76] 0.52 [0.37; 0.73] Model 2 b 0.55 [0.39; 0.76] 0.53 [0.37; 0.74] Model 3 c 0.55 [0.39; 0.76] 0.54 [0.38; 0.76] Model 4 d 0.55 [0.39; 0.76] 0.53 [0.37; 0.74]
1.06 1.05 1.04 1.02 1.01
[0.98; [0.97; [0.96; [0.94; [0.93;
1.15] 1.14] 1.13] 1.10] 1.09]
AC C
EP
Primary outcome: myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure or death from cardiovascular causes a Adjusted for sex, age (years) , race (non-hispanic black, hispanic, non-hispanic white, other) b Additionally adjusted for smoking status (never, former, current) and BMI (<18.5; 18.5-24.9, 25-29.9, 30-34.9, 35+) c Additionally adjusted for baseline clinical and subclinical cardiovascular diseae and number of blood pressure agents d Additionally adjusted for serum creatinine (mg/dl) and fasting total triglycerides (mg/dl)
18
SC
RI PT
ACCEPTED MANUSCRIPT
Table 4: Effect decompostion of the intervention effect on the primary outcome* stratified by baseline diastolic blood pressure.
*
TE D
M AN U
Baseline diastolic BP between 60 and 80 mm Hg (Events=185/n=4066) HR [95% CI] Total effect Direct effect Indirect effect Unadjusted model 0.59 [0.44; 0.80] 0.51 [0.37; 0.69] 1.19 [1.07; 1.31] Model 1 ax 0.59 [0.44; 0.80] 0.53 [0.39; 0.73] 1.13 [1.01; 1.25] Model 2 a 0.59 [0.44; 0.80] 0.52 [0.38; 0.72] 1.14 [1.02; 1.27] Model 3 b 0.59 [0.44; 0.80] 0.53 [0.39; 0.73] 1.12 [1.00; 1.25] Model 4 c 0.59 [0.44; 0.80] 0.54 [0.39; 0.74] 1.11 [0.99; 1.23] Baseline diastolic BP above 80 mm Hg (Events=132 /n=3774) Unadjusted model 0.81 [0.58; 1.14] 0.78 [0.55; 1.12] 1.04 [0.95; 1.13] Model 1 a 0.81 [0.58; 1.14] 0.80 [0.56; 1.13] 1.02 [0.94; 1.11] 0.81 [0.58; 1.14] 0.80 [0.56; 1.14] 1.01 [0.93; 1.10] Model 2 b Model 3 c 0.81 [0.58; 1.14] 0.80 [0.56; 1.14] 1.01 [0.93; 1.10] 0.81 [0.58; 1.14] 0.78 [0.55; 1.12] 1.01 [0.93; 1.10] Model 4 d
AC C
EP
Primary outcome: myocardial infarction, acute coronary syndrome not resulting in myocardial infarction, stroke, acute decompensated heart failure or death from cardiovascular causes a Adjusted for sex, age (years) , race (non-hispanic black, hispanic, non-hispanic white, other) b Additionally adjusted for smoking status (never, former, current) and BMI (<18.5; 18.5-24.9, 25-29.9, 30-34.9, 35+) c Additionally adjusted for baseline clinical and subclinical cardiovascular diseae and number of blood pressure agents d Additionally adjusted for serum creatinine (mg/dl) and fasting total triglycerides (mg/dl)
19
ACCEPTED MANUSCRIPT Highlights: A causal mediation analysis using the observational part of a large scale randomised controlled trial (RCT) is described.
•
There is no significant effect of therapy-induced low diastolic blood pressure on risk of certain cardiovascular outcomes.
•
Adjusting for mediator-outcome confounding is essential, even in RCTs.
AC C
EP
TE D
M AN U
SC
RI PT
•