J-Shaped Relationship between Blood Pressure and Mortality in Hypertensive Patients: New Insights from a Meta-Analysis of Individual-Patient Data
F. Boutitie,F. Gueyffier,S. Pocock,R. Fagard,J. Boissel
DOI: https://doi.org/10.7326/0003-4819-136-6-200203190-00007
IF: 39.2
2002-03-19
Annals of Internal Medicine
Abstract:Context Lowering blood pressure in hypertensive patients decreases the risk for cardiovascular events. However, clinical trials show a discrepancy between observed and expected risk reduction, possibly because of a J-shaped relationship between blood pressure and risk reduction. Contribution The authors analyzed individual-patient data from seven randomized clinical trials of both treated and untreated hypertensive patients. All-cause death rates were higher among patients with high diastolic pressure and those with low diastolic pressure. Low diastolic pressure was associated with an increased death rate, even among untreated patients. Implications Excessive antihypertensive treatment may increase the death rate. Poor health may also cause low blood pressure and increase the risk for death. The Editors The benefit of antihypertensive treatments in reducing the risk for cardiovascular events in persons with high blood pressure has been well established (1-3). Despite this evidence, epidemiologic studies have shown that after adjustment for other risk factors, treated hypertensive patients with normalized blood pressure are still at higher risk for cardiovascular diseases than normotensive persons (4). Moreover, in clinical trials the observed risk reduction for coronary events has been smaller than could be expected from the log-linear relationship between blood pressure and risk according to epidemiologic data (5). Clinical trials have reported a 5to 6mm Hg difference between the diastolic blood pressures of treatment and control groups; the risk reduction of 14% (95% CI, 4% to 22%) for coronary events contrasts with the 20% to 25% reduction noted in epidemiologic reports. These discrepancies may be related to the existence of a J-shaped relationship between blood pressure and risk, in which treated patients with low blood pressure are at increased risk for coronary events. The reports of a J-shaped relationship have come from longitudinal cohort studies of treated hypertensive patients (6-10) or clinical trial data on antihypertensive treatment groups and, in some trials, control groups (11-13). Interpretation of such results has varied. Some researchers have believed that overtreatment with blood pressurereducing drugs may compromise coronary blood flow, especially in hypertensive patients with a history of myocardial infarction (14), while others have considered that the increased risk in patients with low blood pressure is independent of treatment and may be attributed to confounding factors related to deteriorating heath (12, 15, 16) or pulse pressure (17). The INDANA (INdividual Data ANalysis of Antihypertensive intervention trials) project is a meta-analysis of individual-patient data collected from randomized clinical trials of antihypertensive medication versus placebo or no intervention, with follow-up for cardiovascular events and deaths (18). INDANA offers the opportunity to further explore the J-shaped curve. Specifically, this database allows assessment of the evolution of risk according to achieved blood pressure separately for treated and untreated patients who in all other aspects are similar because of the randomization process. Methods Study Sample Data from five randomized clinical trials in elderly patients (Coope and Warrender [13]; European Working Party on High Blood Pressure in Elderly patients [EWPHE] [19], Medical Research Council trial in older adults [MRC2] [20], Systolic Hypertension in the Elderly Program [SHEP] [21], and Swedish Trial in Old Patients [STOP] [22]) and two trials in middle-aged patients (Hypertension Detection and Follow-up Program [HDFP] [23] and Medical Research Council trial in mild hypertension [MRC1] [24]) were pooled in a common file with information on baseline patient characteristics, blood pressure measurements over time, and occurrence of major clinical events. For this analysis, we did not use the data from three other trials available at the INDANA coordinating center. First, the Multiple Risk Factor Intervention Trial (MRFIT) (25) could not directly relate a change in cardiovascular risk to a change in blood pressure because this study assessed the effects of cholesterol and smoking reduction in addition to that of blood pressure reduction. Next, we excluded the Australian National Blood Pressure Study (ANBPS) data (26) because of the risk for informative censoring (the end point was a combined criteria that included several soft events directly related to hypertension). Finally, the Veterans AdministrationNational Heart, Lung, and Blood Institute feasibility trial (VANHLBI) (27) had not been pooled in the database at the time of analysis but would have contributed only marginally to the overall results because few events were observed. Blood Pressure Data We did not include blood pressure at study entry in the analysis because these values were heavily conditioned by the trial-specific selection criteria for blood pressure. At least one blood pressure measurement per each year of follow-up was available. For trials that obtained several measurements per year, we used in our analysis the measurement obtained closest to the annual visit. Because systolic blood pressure in HDFP was measured only after 5 years of follow-up, this trial was not included in statistical analyses involving systolic blood pressure or pulse pressure. Clinical Events and Follow-up After study entry (baseline), the trials in the database monitored patients for occurrence of myocardial infarction, stroke, and death during a mean follow-up period ranging from 2.2 years in STOP to 5.8 years in MRC2. To make our results more robust, we focus on fatal events (classified as cardiovascular or noncardiovascular death). We did not analyze nonfatal events because they were monitored and reported less consistently than fatal events (for example, HDFP did not note the time to occurrence of strokes and EWPHE did not record nonfatal events after the occurrence of a nonfatal end point). In the analysis, follow-up for mortality started 1 year after randomization. For each subsequent yearly period, the risk for death was assessed according to the systolic and diastolic blood pressures obtained closest to the beginning of the year. Statistical Analysis As a first descriptive approach, we estimated the adjusted death rates by treatment group in successive categories of ongoing diastolic blood pressure ( 65, 66 to 75, 76 to 85, 86 to 95, 96 to 105, 106 mm Hg) and systolic blood pressure ( 120, 121 to 130, 131 to 140, 141 to 150, 151 to 160, 161 to 170, 171 to 180, 181 mm Hg) while controlling for any confounding effect of age, sex, study, and year of blood pressure measurement since study entry (28). For all patients, each successive 1-year period of follow-up was used to relate the blood pressure level at the beginning of that period to the risk for death during that period. This method allowed us to consider the influence of blood pressure on risk as it evolved over time. We used a Poisson linear regression model to relate the risk for dying (on a log scale) to the set of covariates (29). In addition to blood pressure, age, and time since entry (with each updated at every period), the other covariates were sex, history of medical events (myocardial infarction, stroke, and diabetes) before study entry, and smoking habits. Occurrence of nonfatal myocardial infarctions during follow-up was a further time-dependent covariate. To determine whether the relationship between blood pressure and risk was J-shaped, we fitted a model with both a linear and a quadratic (squared) term for blood pressure. This model assumes a curved relationship, with the risk for death decreasing to a minimum value with decreasing blood pressure before eventually increasing. Next, we estimated the minimum (nadir) and associated confidence interval from the coefficients of the linear and quadratic terms and their variance (30). Finally, to verify that an increase in risk with lower blood pressure was not a result of constraints of the quadratic model, we fitted blood pressure measurements at the left and right of the nadir with two independent curvilinear relationships, as proposed by Goetghebeur and Pocock (31). We repeated such modeling by using increasing nadir values (at intervals of 2 mm Hg for diastolic blood pressure and 5 mm Hg for systolic blood pressure) and determined the optimum nadir by the model with the best goodness of fit. We performed all statistical analyses by using the SAS software package for Windows, version 6.12 (SAS Institute, Inc., Cary, North Carolina). Role of the Funding Source The funding source had no role in the collection, analysis, and interpretation of the data or in the decision to submit the paper for publication. Results Of the 40 777 patients randomly assigned to active treatment or control groups in the seven trials, we analyzed data on the 40 233 total patients alive at 1 year after study entry (Table 1). The mean follow-up was 3.9 years. Among the patients in our analysis, 48.9% (n = 19 692) were men and 27.7% (n = 11 107) were current smokers; 1312 patients (3.3%) had a history of myocardial infarction, 464 (1.1%) had a history of stroke, and 1420 (3.5%) had a history of diabetes mellitus. We could not include data on 31 995 1-year periods of follow-up (18.7%) in assessing a possible relation between fatal events and diastolic blood pressure because of missing blood pressure measurements. Overall, our analysis comprised data on 1655 deaths (56% of which were caused by cardiovascular events) during 126 908 total patient-years of follow-up. Table 1. Patient Events in Each Study Recorded during Follow-up between 1 Year after Study Inclusion and End of Study Follow-up One year after study entry, mean diastolic blood pressure in each trial was significantly lower in the active treatment group than in the control group; the mean 1-year difference in diastolic blood pressure between treatmen