|Home | About | Journals | Submit | Contact Us | Français|
Blood pressure (BP) guidelines that set target BP levels often rely on analyses of achieved BP from hypertension treatment trials. The objective of this paper was to compare the results of analyses of achieved BP to intention-to-treat analyses on renal disease progression. Participants (n=1,094) in the African-American Study of Kidney Disease and Hypertension Trial were randomized to either: (1) usual BP goal defined by a mean arterial pressure (MAP) goal of 102–107 mmHg or (2) lower BP goal defined by a MAP goal of ≤ 92 mmHg. Median follow-up was 3.7 years. Primary outcomes were rate of decline in measured glomerular filtration rate (GFR) and a composite of a decrease in GFR by > 50% or >25 ml/min/1.73m2, requirement for dialysis, transplantation, or death. Intention-to-treat analyses showed no evidence of a BP effect on either the rate of decline in GFR or the clinical composite outcome. In contrast, the achieved BP analyses showed that each 10 mm Hg increment in mean follow-up achieved MAP was associated with a 0.35 (95% CI 0.08 – 0.62, p = 0.01) ml/min/1.73m2 faster mean GFR decline and a 17% (95% CI 5% – 32%, p = 0.006) increased risk of the clinical composite outcome. Analyses based on achieved BP lead to markedly different inferences than traditional intention-to-treat analyses, due in part to confounding of achieved BP with co- morbidities, disease severity and adherence. Clinicians and policy makers should exercise caution when making treatment recommendations based on analyses relating outcomes to achieved BP.
A direct relationship between blood pressure (BP) and cardiovascular/renal outcomes is often shown in analyses of achieved BP in clinical trials, as well as in observational studies. Such analyses commonly document a direct, progressive relationship of BP levels with adverse outcomes at BP levels that are well below the conventional BP goal of 140/90 mmHg BP, a level demonstrated to be beneficial in randomized trials that directly compared BP targets. (1–6)
Analyses of achieved BP have commonly been used to set BP treatment targets (please see Table S1 at http://hyper.ahajournal.org). The BP guideline 130/ 80 mmHg recommended for chronic kidney disease by the American Diabetes Association,(7) the Joint National Committee on the Prevention, Detection, Evaluation and Treatment of High Blood Pressure (JNC)(8) and the Kidney Disease Outcomes Quality Initiative (9) is based on achieved BP, not randomized controlled trial evidence. For example, much of the evidence for the lower SBP goal (<130 mmHg) recommended for patients with diabetes, high risk coronary heart disease, and chronic kidney disease is based on analyses of achieved BP.(10–19) Interestingly, several of these trials (e.g. MDRD, HOT) explicitly tested the effects of two different BP goals, and the primary results of the intention-to-treat (ITT) analyses were null. (18, 20, 21) Yet, the results from analyses of achieved BP were still used to guide policy despite the inconsistency with the primary ITT analyses of these trials.
The credibility of treatment recommendations based on achieved BP is a matter of continuing controversy, particularly when they conflict with ITT results from randomized controlled trials or when such trial data are unavailable. On the surface, it may appear that analyses relating outcome to achieved BP would be more informative than ITT analyses of randomized comparisons regarding the biological effect of BP since the former, but not the latter, takes into account the patients’ actual BP levels. However, variables associated with lower levels of achieved BP may include favorable health characteristics at baseline and markers of improved prognosis during follow-up, including adherence. Hence, the apparent benefit of a lower level of achieved BP may result, in whole or in part, from confounding.
Until recently, few randomized trials specifically compared BP targets. As more randomized clinical trials comparing BP targets are becoming available, BP treatment recommendations can be based on the ITT comparisons of randomized groups rather than analyses of achieved BP. The African American Study of Kidney Disease and Hypertension (AASK) trial provides a unique opportunity to document differences in results by randomized group vs. achieved blood pressure on renal outcomes and to explore potential reasons.
AASK is a completed trial that randomized African Americans with hypertensive kidney disease to two different BP goals. As previously reported, the ITT analyses documented that there was no significant difference in renal outcomes between the randomized BP goals during the trial.(22) In this report, we compare analyses based on achieved levels of BP to the original ITT analyses, and consider whether differences in the results of these two types of analysis can be explained by the relationship of achieved BP with baseline and follow-up variables.
The AASK trial design and procedures have been described previously.(22, 23) The institutional review board at each study site approved the study protocol. Written informed consent was obtained from each participant. Trial participants were African Americans, aged 18 to 70 years, with hypertensive CKD as defined by a diastolic BP > 95 mmHg and a GFR between 20 and 65 ml/min/1.73m2 measured by 125I-iothalamate clearance. Individuals were randomized in a 2×3 factorial design to one of two BP goals: (1) usual BP goal with a mean arterial BP (MAP) goal of 102–107 mmHg or (2) lower BP goal with MAP goal of < 92 mmHg. In addition, they were randomized to an initial drug therapy with the ACE inhibitor, ramipril; the calcium channel blocker, amlodipine; or the beta blocker, metoprolol.(22) Figure 1 displays the allocation of participants to the two randomized BP goals, as well as an overview of analyses.
Three consecutive seated BPs were measured at each study visit with a Hawksley random zero sphygmomanometer after 5 minutes rest. (23, 24) The mean of the last two BP readings were recorded. Glomerular filtration rate (GFR) was measured by the renal clearance of 125I-iothalamate twice at baseline then at months 3, 6, and every 6 months thereafter.(25) Mean follow-up BP levels were computed as the mean of all study BP measurements (except for BP measured at GFR visits) starting in month four of follow-up through each participant’s final BP measurement. Because the study protocol required titration of BP based on mean arterial pressure (MAP), we regard the mean follow-up MAP as the primary measure of achieved BP.
This report focuses on two main renal outcomes: the chronic GFR slope and a clinical composite defined by a confirmed decline in GFR by >50% or >25ml/min/1.73m2 from the baseline GFR, ESRD (dialysis or transplantation), or death. The chronic GFR slope is defined by the rate of change in GFR beginning at 3 months after randomization, when the maximal hemodynamic effect on GFR resulting from antihypertensive treatment is achieved. The chronic slope corresponds to the long-term rate of progression of renal disease.(22)
Participant characteristics were summarized by randomized BP group and by level of mean follow-up BP using means and standard deviations for continuous variables and by frequencies and percents for categorical variables. Analysis of variance or chi-square tests were used as appropriate to compare patient characteristics between the BP subgroups. Kernel density curves (26) were used to compare the distributions of mean follow-up MAP between the randomized BP groups.
We applied two approaches for analyses relating achieved BP to renal outcomes. In the first approach, we applied an as-treated strategy to relate renal outcomes to each participant’s level of achieved BP, irrespective of the participants’ randomized BP assignment. In the as-treated approach, which mimics analyses of achieved BP in observational studies and previous randomized trials,(27–29) the total variation in achieved BP between participants reflects a combination of the separation between the two BP intervention groups and differences among participants’ achieved BP levels within each BP group. The phrase “as-treated” indicates that the analysis relates outcomes to the treatment (BP) actually received (achieved), in contrast to intent-to-treat analyses, which compare outcomes according to randomized assignment.
In the second approach, we compared outcomes between participants who achieved the same BP level in spite of being randomized to different BP interventions. Here we took advantage of the fact that some participants randomized to the lower BP goal had achieved MAP levels at or near the target range of the usual BP goal (see Figure 2). By comparing renal outcomes for participants with the same achieved MAP that was out of range for low-goal participants but in range for the usual BP goal participants, we could examine whether failure to attain a given BP target might itself be associated with poor renal outcome, independent of biological effects of BP.
For as-treated analyses of the chronic GFR slope, we applied mixed effects models with random intercepts and slopes to relate the follow-up GFRs to mean follow-up MAP while controlling for the randomized drug assignment, clinical center, and 5 pre-specified covariates: baseline proteinuria (log urine protein/creatinine ratio), history of cardiovascular disease, baseline mean arterial pressure, sex, and age. For graphical representations, we used a 3-slope linear spline model with separate slopes for MAP < 95 mmHg, 95–104 mmHg, and ≥ 104 mmHg, where the knot points represent approximate tertiles of follow-up MAP. For analyses relating chronic GFR slope jointly to both mean follow-up MAP and randomized BP assignment, we extended the above mixed effects models by including both an indicator variable for the assigned BP group and separate spline terms for the effect of mean follow-up MAP within each BP group. Reflecting the MAP targets for the two BP goals, the joint models allowed separate slopes for MAP ≤ 92 mmHg, 92 – 102 mmHg, 102 – 107 mmHg, and > 107 mmHg.
Time-dependent Cox regression was used to relate the hazard for the clinical composite outcome (GFR event, ESRD, or death) to the prior mean follow-up MAP, while controlling for randomized drug group and the five pre-specified covariates, with stratification for clinical center. Analogous to the analysis of chronic GFR slope, we performed both a standard as-treated analysis in which BP assignment was not included in the model, and a joint analysis with both BP assignment and linear spline terms for the relationship of the log hazard function with achieved MAP.
To investigate possible sources of differences in renal outcomes between participants achieving the same MAP in the two BP groups, we used t-tests or chi-square tests as appropriate to compare baseline characteristics between usual BP goal and lower BP goal participants whose mean follow-up MAP was between 100 mmHg and 107 mmHg. Here we broadened the MAP range to 100–107 from the usual BP goal target of 102–107 mmHg to increase the sample size for the lower BP participants.
To investigate the sensitivity of the results to the selection of the outcome variable and to the statistical model, we considered a) analyses of an exclusively renal composite including only GFR events and ESRD, while censoring deaths, b) multivariable models not including baseline MAP as a covariate, c) extended models including adjustment for additional baseline covariates, and d) “step-function” models (rather than splines) with follow-up MAP subdivided by MAP < 92 mmHg, 92–102 mm Hg, 102–107 mmHg, and > 107 mmHg.
Figure 2 displays the distribution of mean follow-up achieved MAP by randomized BP group. Although the distribution is bimodal, there was substantial overlap in achieved MAP between the randomized groups. The baseline characteristics of the participants stratified by randomized BP group assignment and whether or not the mean achieved BP was at goal is shown in Table 1. As previously reported, there were no significant differences in the baseline characteristics between the two randomized groups.(22)
In contrast, when we compared participants within each randomized group whose achieved BPs were at goal vs. out of goal, there were significant differences in several important baseline demographic and clinical characteristics. Participants in the Lower BP group who achieved the BP goal (in goal) had fewer co-morbidities, lower baseline mean systolic and diastolic BPs, and lower prevalence of left ventricular hypertrophy than those who were above their BP goal. Participants in the Lower BP group, who achieved goal BP, were also more adherent (92.6 % ± 7.66 vs. 83.0% ± 14.7, p < 0.001) to antihypertensive medications and required fewer antihypertensive medications than those who failed to achieve their goal. Similarly, in the usual BP group those participants whose achieved BPs were at or below their MAP goal also had fewer co-morbidities and antihypertensive medications, lower mean baseline BPs, lower prevalence of LVH, and greater adherence to antihypertensive medications (90.8 % ± 13.2 (below goal), 91.4 % ± 8.75 (at goal) vs. 78.5% ± 19.1 (above goal), p <0.001) when compared to those whose achieved BPs were above their goal.
Figure 3 compares the results of intention-to-treat and achieved BP analyses of GFR slope and clinical outcomes. The intention-to-treat analyses showed no evidence of a BP effect on either the chronic GFR slope (panel a) or the clinical composite outcome (panel b). In contrast, the results of the achieved BP analyses show that at progressively higher levels of achieved BP, mean GFR decline steepened (panel c) and the rate of the clinical composite outcome increased (panel d). In the achieved BP analyses, each 10 mm Hg increment in follow-up MAP was associated with a 0.35 (95% CI 0.08 – 0.62, p = 0.01) ml/min/1.73m2 faster mean GFR decline and a 17% (95% CI 5% – 32%, p = 0.006) increase in the risk of the clinical composite outcome. These relationships appeared to strengthen at higher levels of achieved MAP (Figure 3, panels c and d).
Figure 4 displays the results of analyses relating chronic GFR slope to both randomized BP group and achieved MAP. As shown, for participants assigned to the lower BP goal, where MAP ≤ 92 mm Hg was in-range and MAP > 92 was out of range, higher MAP levels were associated with faster GFR decline. In contrast, among participants assigned to the usual BP goal, renal outcomes appeared similar in participants with achieved MAP near the usual goal target range of 102–107 mm Hg and participants who achieved lower MAP levels. Importantly, within the 102–107 mm Hg range, after adjusting for the pre-specified baseline covariates the chronic GFR slope was steeper among lower BP goal participants than usual BP goal participants, at the same level of achieved MAP. At the midpoint of this interval at 104.5 mm Hg, the adjusted mean (±SE) GFR slope was 0.99 ± 0.41 ml/min/1.73m2/yr faster among lower BP goal participants than among usual BP goal participants (p = 0.02); the difference in adjusted mean slope tended to increase further towards the lower end of the interval (−1.33 ± 0.53 ml/min/1.73m2/yr, p = 0.01) where MAP = 102 mm Hg, and ameliorate towards the upper end point where MAP = 107 mm Hg (−0.65 ± 0.47 ml/min/1.73m2/yr, p = 0.16). Similar results were obtained for the time-dependent Cox regression analyses; at the MAP goal interval midpoint of 104.5 mm Hg, the adjusted hazard ratio (95% CI) for the comparison of the clinical composite outcome between low and usual BP goal patients was 1.58 (95%CI: 1.10, 2.27) p=0.014.
We compared the clinical and demographic characteristics of participants with achieved mean MAP of 100–107 mmHg between the two BP groups (Table 2). Among participants whose achieved mean MAP was 100–107 mmHg, those assigned to the lower BP group had greater prevalence of LVH, higher baseline BP and required more antihypertensive medication as compared to those in the usual BP group. Antihypertensive medication adherence was also lower among those in the lower BP group.
Similar results to those summarized above were observed for the renal composite including only ESRD or declining GFR events, with death censored, and for multivariable models in which baseline MAP was not included as a covariate. In step function models comparing lower and usual BP goal participants with MAP between 102–107 mm Hg, the mean chronic slope was 1.52 ml/min/1.73m2 steeper (95% CI: 0.13 – 2.9, p =0.03) for participants assigned to the lower BP goal, and the hazard ratio of the clinical composite including GFR events, ESRD, and death was 2.15 (95% CI: 1.27, 3.63, p = 0.004). Similar results were also obtained in the extended models including covariate adjustment for randomized treatment assignment, and baseline LVH on ECG, smoking, fasting glucose, and history of cardiovascular disease in addition to the pre-specified covariates. Using the linear spline models, with more extensive covariate adjustment the adjusted mean GFR slope at MAP = 104.5 mm Hg was 1.09 ± 0.41 ml/min/1.73m2/yr faster in the low than the usual BP group (p = 0.008), and the hazard ratio for the clinical composite outcome was 1.58 (95% CI, 1.10 – 2.29, p = 0.01) for the low vs. the usual BP group.
This report illustrates the challenges of using analyses of achieved BP to guide treatment recommendations. The AASK trial is one of a few studies that directly compared different BP goals and is ideally suited to this evaluation. In this paper, we document that analyses based on achieved BP lead to markedly different inferences than ITT analyses. Specifically, ITT analyses comparing randomized groups documented that the rate of decline in kidney function did not differ between the lower and usual BP goal groups. However, in analyses based on achieved BP, lower levels of BP were associated with more favorable renal outcomes. Further analyses documented that the likely reason for this discrepancy was confounding of achieved BP with co-morbidities and adherence. When results were stratified by achieved MAP, participants in the lower BP group who were above goal had a faster decline in renal function than did participants in the usual BP group who were within goal but had the same achieved MAP. However, this group also had poorer prognostic factors at baseline and worse adherence during follow-up.
This pattern has been seen in other trials comparing the effect of BP goals on clinical outcomes. A similar discrepancy between the ITT comparison of randomized BP groups and analyses of achieved BP occurred in the Modification of Diet in Renal Disease (MDRD) trial, in which participants were randomized to lower BP goal group (MAP ≤ 92mmHg) or a usual BP goal group (MAP ≤ 107 mmHg, < 113 mmHg in those age > 61). (21) As in the AASK analyses, when the randomized groups were combined and analyzed by achieved mean MAP, there was an association between higher levels of mean MAP during follow-up and decline in GFR, even after controlling for baseline characteristics.(18)
No difference in CVD or renal outcomes was noted in the overall cohort in the Hypertension Optimal treatment Trial (HOT), a large clinical trial (n=18,790) which included a subset of hypertensive participants with mild renal disease at baseline randomized to one of three diastolic BP goal groups, (≤90 mmHg, <85 mmHg and, ≤80 mmHg).(30) In contrast to the ITT analysis, lower levels of achieved diastolic BP were associated with a reduced risk of CVD in this trial, though not a significant difference in serum creatinine after 3.8 years of follow-up. Of note, this study randomized participants based on diastolic rather than systolic BP goals, and only two serum creatinine measurements were available (baseline and final visit).(30)
The limitations of as-treated analyses, in which outcomes are related to treatment received rather than randomized treatment assignment, have been documented in the nephrology and general clinical trials literature.(29, 31–33) This report demonstrates these limitations in the setting of analyses of achieved blood pressure. Recently, the Action to Control Cardiovascular Risk in Diabetes Trial (ACCORD) documented that a SBP goal <120 mmHg did not significantly lower the rate of composite or cardiovascular events compared to a traditional SBP goal of < 140 mmHg in patients with type 2 diabetes.(34) Such results will likely encourage post-hoc analyses to identify those that may have benefited from the lower SBP goal. However, in view of results from AASK, we urge caution, especially if analyses based on achieved BP are conducted. A trial similar to ACCORD, the Systolic Blood Pressure Invention Trial (SPRINT), in non diabetic hypertensive patients (40% with CKD) which is underway by NHLBI will offer similar appeal.
An important issue is the extent to which confounding may account for the relationship of blood pressure with clinical outcomes in observational studies. Our sense is that the problem of confounding, while potentially present in observational studies, is magnified in the setting of achieved BP analyses conducted in the setting of a clinical trial. Although their effects might be reduced, the confounding relationships demonstrated in the AASK as-treated analyses seem likely to also occur in the observational setting, where national and international standards stipulate maximum acceptable BP levels. If such a bias does exist, then regardless of where BP targets are set, patients with higher observed BP levels may always seem to do worse. When BP level is explicitly targeted, as in the AASK Trial, variations in co-morbidities and behavioral factors may account for a greater proportion of the variation in achieved BP than in the observational setting, where observed BP level also depends on variations in practice patterns, access to treatment, and other factors. Thus, the degree of confounding in achieved BP analyses from the AASK Trial is likely amplified due to the controlled conditions of the trial. Nonetheless, residual confounding likely occurs in observational analyses, especially recent studies in which there are substantial efforts to control BP to recommended BP goals.
The AASK Trial, with distinct randomly assigned BP goals but partially overlapping achieved BP levels, provided a unique opportunity to investigate potential biases in analyses of achieved BP by comparing patients with similar BP levels who were in-goal for one treatment arm but out of goal for the other. Potential limitations include reduced statistical power in analyses based on subgroups of participants. However, major strengths include a well-characterized study population with extensive data on potential confounders, both at baseline and during follow-up; a long duration of follow-up; and a large, sustained contrast in BP between randomized groups.
In summary, analyses based on achieved BP can lead to markedly different inferences than traditional intention-to-treat analyses. A major reason for this discrepancy appears to be confounding of achieved BP with co morbidities, disease severity and adherence. Clinicians and policy makers should exercise caution when making treatment recommendations based on analyses relating outcomes to achieved BP.
Funding/Support: Study was supported by a cooperative agreement from the National Institute of Diabetes and Digestive and Kidney Diseases and the NIH Office of Minority Health Research and in part by the following institutional General Clinical Research Centers and other National Institutes of Health grants: UL1 RR024989, 5M01 RR-00071, M01 00032, P20-RR11145, M01 RR00827, M01 RR00052, 2P20 RR11104, and DK 2818-02.1
Role of Sponsor: The funding organizations were independent of the design and conduct of the study; collection, management, analysis and interpretation of the data; and preparation, review or approval of the manuscript.
Publisher's Disclaimer: 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 citable 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.
Author Contribution:Study concept and design: Appel, Davis, Greene, Wright.
Acquisition of the Data: Appel, Greene, Pogue, Rahman, Wang, Wright
Analysis and interpretation of the data: Appel, Davis, Greene, Rahman, Wang, Wright,
Drafting of the manuscript: Appel, Davis, Greene, Wang, Wright,
Critical revision of the manuscript for important intellectual content: Appel, Astor, Davis, Greene, Lipkowitz, Pogue, Rahman, Toto, Wang, Wright
Statistical Analysis: Greene, Wang
Obtained funding: Appel, Greene, Lipkowitz, Toto, Wright
Administrative, technical or material support: Davis, Wang, Wright
Study Supervision: Appel, Davis, Greene, Lipkowitz, Toto, Wright.
1 Potential conflict of interests and disclosures: J.T.W. is a consultant/advisory board member for Sanofi-Aventis (modest), Novartis (modest), Daiichi-Sanyo (modest), Take Care Health (at least $10 000), Noven Pharmaceuticals (modest), NiCox Pharmaceuticals (modest), and CVRx DSMB (modest). M.S.L has received honoraria (<$10,000); T.G, is a consultant/advisory board member for Eli Lilly&Co (modest) Keryx Biopharmaceuticals (modest) Amgen inc (modest), Cormedix Inc (modest), Nephrogenex inc (modest).