In a 1986 review, MacMahon and colleagues [4] stated that

"event rates, particularly those for fatal events and nonfatal myocardial infarction, were substantially lower in women than in men. The striking benefits of study treatments for the risk of fatal and non-fatal stroke were evident for both men and women. A reduction in total mortality could not be demonstrated for women, but the treatment effect for women was not significantly different from that in men, among whom there was an important and statistically significant reduction in mortality."

This comment was based on the results of two trials: the Hypertension Detection and Follow-up Program (HDFP) [5] and the Medical Research Council trial of treatment of mild hypertension (MRC35-64) [6]. In their 1991 analysis of data from these trials plus data from the European Working Party on High Blood Pressure in the Elderly (EWPHE) trial [7] and the Australian therapeutic trial in mild hypertension [8], Anastos and colleagues [9] concluded that

"the few data that do exist suggest that gender, like race and age, significantly influences the natural course of hypertension ... and the response to treatment ... . The data regarding aggressive treatment of white women are equivocal; there is concern that such treatment may actually be harmful."

Since these reviews were published, reports of three additional trials of antihypertensive treatment in older hypertensive men and women have appeared in print: the Medical Research Council trial of treatment of hypertension in older adults (MRC 65-74) [10], the Systolic Hypertension in the Elderly Program (SHEP) [11], and the Swedish Trial in Old Patients with Hypertension (STOP) [12]. More recently, other reviewers have stated that "antihypertensive medications do not appear to be as effective in women as in men" [13] and that "when treated, women often achieve less benefit than do men" [14].

The INDANA (INdividual Data ANalysis of Antihypertensive intervention trials) project [15] offers the opportunity to provide more evidence on the effects of antihypertensive treatment in women; results are based on individual patient data from all of the randomized, controlled trials mentioned in the preceding paragraphs. The two main objectives of the current study are to quantify the average treatment effect in each sex separately and to determine whether treatment effect differs significantly between women and men.

Methods

Top Methods Results Discussion Author & Article Info References

The INDANA project (whose rationale, objectives, and methods are described in detail elsewhere [15]) is a collaboration of representatives from most of the large randomized, controlled trials of antihypertensive drug treatment. Its results are derived from centralized files of the baseline and follow-up data available for all patients enrolled in the trials.

The Trials

Our report is based on seven trials [5-710-12, 16] (Table 1) in which both men and women were enrolled. The inclusion criteria for the trials in the INDANA project are discussed elsewhere [15]. In summary, the steering group of the project made the following decisions: The data from the Australian trial [8] were not included in the analysis because separate outcomes are not available without censoring bias; the EWPHE trial [7] data were included only for the analysis of mortality end points (separate nonfatal outcomes are not available without censoring bias); and the data from HDFP [5] were considered in a sensitivity analysis (analysis was done with and without these data because of the originality of the trial design, which compared specific antihypertensive care systems with usual care). The data from the Veterans Administration and National Heart, Lung, and Blood Institute feasibility trial [17] are available in the INDANA database but have not yet been submitted to control and extraction procedures. Thus, these data were not used in our analysis. Because this trial has only a small weight in terms of patient-years and observed events, its exclusion is unlikely to change the results presented here.

Outcomes

According to the INDANA protocol, seven outcomes were analyzed: 1) fatal strokes; 2) fatal and nonfatal strokes, excluding transient ischemic attacks; 3) fatal coronary events [including sudden death, which was defined as unexpected and unexplained death occurring within a maximal interval of 24 hours after symptom onset]; 4) fatal and nonfatal major coronary events (using criteria for major coronary heart disease obtained from patient histories in HDFP) [1]; 5) cardiovascular-related mortality, including death from pulmonary thromboembolism; 6) major cardiovascular events [combining the second, fourth, and fifth outcomes and excluding such minor cardiovascular events as angina pectoris, intermittent claudication, or nonfatal congestive heart failure]; and 7) total mortality.

Statistical Analysis

Summarized data (number of patients and number of events) were extracted from the INDANA database by sex and by trial according to the intention-to-treat principle. For the group assigned to receive active treatment, the odds ratio compared with controls was estimated by sex for each outcome according to the Peto method [18]. The odds ratio in women was compared with the odds ratio in men by determining whether the ratio was different from 1. This interaction between sex and treatment effect was checked after adjustment for the main baseline risk factors (age, baseline smoking habits, systolic blood pressure, serum cholesterol level, presence of diabetes, and history of stroke or myocardial infarction) in a multivariate logistic model [19] fitted by outcome.

For HDFP [5], we censored data at the date of the end of the trial intervention. Two deaths in the trial by Coope and Warrender [16] that were caused by pulmonary embolism were included with cardiovascular-related mortality in our analysis; one early cancer-related death in this trial was included in the analyses of total mortality because of the intention-to-treat principle.

To illustrate the difference in the treatment effect between men and women, we applied two graphical approaches to the second and fourth outcomes (all strokes and all coronary events). First, each trial was represented by sex in a treatment-effect graph [20] in which the x-axis is the risk observed in the control group (Rc) and the y-axis is the risk observed in the treated group (Rt) (Figure 1). The odds ratio line, with a slope equal to the odds ratio and a null intercept, indicates the treatment effect by sex. The principal diagonal of the plane Rt x Rc represents the absence of treatment effect (Rt = Rc; odds ratio, 1). The vertical distance between the odds ratio line and the principal diagonal indicates the absolute risk reduction for a given untreated risk. Second, the absolute risk reduction attributable to treatment and its CI were computed by tertiles of individually predicted risk for each sex and were plotted against the average predicted risk in each tertile (Figure 2). The predicted risk was derived from individual scoring built on the results of a multivariate logistic model, including the major risk factors mentioned above. Tertiles were computed to contain similar numbers of events.

View larger version (14K): [in this window] [in a new window]   Figure 1. Effect of antihypertensive treatment on absolute risk for fatal and nonfatal stroke (left) and fatal and nonfatal coronary events (right). The points show subgroups in each trial by sex; the x-axis represents risk in the control group, and the y-axis represents risk in the treatment group. The risk is given as the rate for 1000 patient-years. The two dashed lines represent the odds ratios in women and men. Coope = Coope and Warrender [16]; HDFP = Hypertension Detection and Follow-up Program [5]; MRC1 = Medical Research Council trial of treatment of mild hypertension [6]; MRC2 = Medical Research Council trial of treatment of hypertension in older adults [10]; SHEP = Systolic Hypertension in the Elderly Program [11]; STOP = Swedish Trial in Old Patients with Hypertension [12].

View larger version (9K): [in this window] [in a new window]   Figure 2. Absolute risk reduction of fatal and nonfatal stroke (left) and fatal and nonfatal coronary events (right) by untreated risk level and sex. The points show the absolute risk reduction on the y-axis with 95% Cls. The x-axis represents the predicted untreated risk. The number next to each point is the number of events observed by tertile. The predicted risk was obtained from a multivariate score established on the control group with the main cardiovascular risk factors: age, baseline smoking habits, blood pressure, serum cholesterol level, diabetes, and history of stroke or myocardial infarction.

Meta-analysis computations were done using Easy-MA software [21]; data management and logistic regression were done using SAS software [22].

Results

Top Methods Results Discussion Author & Article Info References

The key features of the seven trials are presented in Table 1. Five of the trials addressed hypertension in older persons, and two studied mild to moderate hypertension in younger persons. The drugs used in the trials were primarily thiazide diuretics, ß-blockers, or both. The data for these seven trials contained in the INDANA database represent 97.5% of all existing data from all applicable trials in terms of patient-years of follow-up during the active phase of the trials. The combined trial data on risk factors by sex (Table 2) show that, on average, women were older; had a higher baseline cholesterol level, a higher systolic blood pressure, and a lower smoking rate; and less frequently had a history of myocardial infarction. Because these baseline characteristics were similar for the active treatment and control groups, these groups are combined in Table 2.

For each of the seven outcomes, Table 3 and Table 5 shows the number of events in the active treatment and control groups that occurred in men and women, both within each trial and in all trials combined.

The exclusion of the HDFP data from the analysis changes neither the direction nor the magnitude of the odds ratio for either sex and does not affect the differences between men and women. These data are therefore included in the results presented. In Table 4, the combined odds ratios for all trials are shown separately for men and women; these odds ratios were estimated using a fixed-effects method. In women, odds ratios favoring treatment were statistically significant for strokes (both fatal and either fatal or nonfatal) and major cardiovascular events but not for other outcomes. In men, odds ratios favoring treatment were statistically significant for all seven outcomes considered. No significant interaction was found between sex and treatment effect: The odds ratios between the treated and control groups did not differ between women and men, regardless of outcome. Adjustment for the major available baseline risk factors (including age, blood pressure, smoking habits, serum cholesterol level, presence of diabetes, and history of stroke or myocardial infarction) in a logistic regression model did not change this finding.

The graphical approaches illustrate the absence of interaction between sex and the size of the treatment effect for major coronary events and strokes. On the treatment effect graph (Figure 1), the size of the absolute risk reduction is indicated by the vertical distance between the odds ratio line and the principal diagonal for each sex. This size depends on the untreated risk, not on sex; for an untreated risk of 10 per 1000 patient-years, the absolute risk reduction in both women and men is approximately 1 for coronary events and 3 for strokes. For an untreated risk of 30, the absolute risk reduction in both women and men is approximately 3 for coronary events and 10 for strokes. This means that when the untreated risk is 10 per 1000 patient-years, the number of patients needed to treat over 1 year (the reciprocal of the absolute risk reduction) to avoid one coronary event is about 1000; the number needed to treat to avoid one stroke is about 300. The corresponding figures for an untreated risk of 30 per 1000 patient-years are about 300 and 100. These figures are the same in women and men.

The graph by tertile of predicted risk (Figure 2) shows that the size of the absolute risk reduction is the same in women and men whose untreated risk is similar and that the reduction increases with untreated risk. In the control group, the untreated risk distribution for stroke is similar in men and women; in men, the distribution for coronary events and the corresponding absolute risk reduction shifted to higher levels in men.

Discussion

Top Methods Results Discussion Author & Article Info References

Our analysis was based on data from 7 of 15 trials [5-810-1216, 17, 23-28] that compared antihypertensive drug interventions with placebo or no (or minimal) treatment and enrolled both men and women. It is unlikely, however, that the trials that have not been integrated into the INDANA database would change the results presented here: The analyzed data represent more than 97% of the patient-years of follow-up for these 15 trials.

We selected the outcomes on the basis of several criteria, including the availability of the outcomes, their homogeneity across trials, the reliability of their notification, and their clinical relevance. In particular, we did not use softer criteria, such as nonfatal heart failure, angina pectoris, or intermittent claudication, which may have been affected by antihypertensive drugs. Thus, the number of patients needed to treat to avoid an event may not represent all patients in whom treatment is beneficial. Rather, this number is specific for the considered outcomes.

The drugs used in the trials analyzed here were primarily thiazide diuretics, ß-blockers, or both. We cannot extrapolate these results to other antihypertensive drugs without further research. Trials comparing drug classes are currently under way, and a prospectively planned review [29] of these trials should indicate whether our results are reproducible if drugs other than thiazide diuretics and ß-blockers are used as first-line intervention.

In men, odds ratios favoring treatment were statistically significant for all outcomes considered. The fact that statistical significance for coronary events or total mortality was not reached in women may have a simple explanation: The underlying rate of these events was lower in women than in men, which means that the analyses of data in women had less statistical power. Indeed, the number of coronary events observed in women was roughly half that observed in men. In contrast, when the control groups of the seven trials are considered together, men and women had similar numbers of observed strokes and similar numbers of patient-years of follow-up (Table 3 and Table 5).

No evidence of interaction was found between sex and treatment effect in relative terms; this means that the odds ratios for treatment should be the same for men and women. However, the absolute risk reduction is not the same for women and men; as shown in the figures, this reduction depends on the untreated risk level. Both absolute risk reduction and its reciprocal number needed to treat are estimates that are more helpful for understanding the size of treatment benefit: When a clinician states that treatment prevents half the strokes, this half could be close to 0 and therefore practically negligible if the untreated risk is very low. In contrast, this effect could be clinically meaningful if the untreated risk exceeds a given threshold. This threshold is arbitrary, however, as is the definition of hypertension. Absolute risk reduction does not reflect only the proportionate treatment effect (for example, a 50% reduction in the rate of stroke); it also contains information on the untreated risk.

Our results are based on post hoc analyses; in none of the included trials was the comparison of the treatment effect size between men and women a main objective. It would not be correct, therefore, to conclude that treatment effect size does not differ whatsoever between men and women. We have shown only that the available data to date do not suggest such a difference.

Extrapolating the size of the treatment benefit by using predicted untreated risk is a way to individualize antihypertensive drug treatment [30, 31]. However, the validity of such extrapolation requires that there be no interaction between baseline characteristics and the treatment effect in relative terms. It is impossible to be sure from the seven trials alone that no such interaction exists, but the available data do not support it.

Our meta-analysis includes almost all of the data from trials in which antihypertensive treatment was assessed in both sexes. It indicates that with treatment, odds ratios for cardiovascular risk are similar in women and men. However, the quantification of benefit in terms of absolute risk reduction shows that for women, the benefit is seen primarily for strokes, whereas in men treatment prevented as many coronary events as strokes. These findings suggest that the absolute benefit for the two main categories of cardiovascular events cannot be predicted across the sexes and that physicians need appropriate tools for predicting untreated risk in order to individualize antihypertensive and other preventive therapies.

Dr. Pocock: Medical Statistics Unit, Department of Epidemiology and Population Sciences, London School of Hygiene, Keppel Street, London WCIE 7 HT, United Kingdom.

Dr. Coope: Bollington Medical Centre, The Waterhouse, Bollington near Macclesfield, Cheshire SK 10 5JL, United Kingdom.

Drs. Cutler and Friedman and Ms. Schron: Division of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, 2 Rockledge Centre, MSC 7936, 6701 Rockledge Drive, Bethesda, MD 20892.

Dr. Ekbom: Lund University, Department of Community Health Sciences Dalby/Lund, Helgeandsgatan 16, S-223 54 Lund, Sweden.

Dr. Fagard: UZ Pellenberg, Weligerveld 1, 3212 Pellenberg, Belgium.

Dr. Perry: Washington University, Box 8048, 660 South Euclid Avenue, St. Louis, MO 63110.

Dr. Prineas: Department of Epidemiology and Public Health, University of Miami School of Medicine, 1550 Northwest 10th Avenue, Miami, FL 33136.