Reply to DR Merkle

Dear Editor:

We thank Dr Merkle for his comments, which afford us an opportunity to expand on the interpretation of the Women's Health Initiative (WHI) dietary modification (DM) trial results on coronary heart disease (CHD), overall cardiovascular disease (CVD), and total mortality, as presented in our paper (1).

As noted by Dr Merkle, overall no differences were apparent between the intervention and comparison groups for any of these clinical outcomes during the average 8.3-y intervention period, or during a substantial postintervention period, in this trial among 48,835 postmenopausal women in the United States.

Intervention-group women reported reduction in total fat intake by 8–10% of energy, and carbohydrate increases by a similar 8–10% of energy during the intervention period. The absence of adverse CHD, overall CVD, or total mortality rates in the intervention compared with the comparison group is useful information in itself, especially in view of a recent report arguing for higher fat and lower carbohydrate consumption, using data from a recent observational international study (PURE) that combined analytic and ecologic sources of epidemiologic information (2).

In an editorial related to our initial paper (3) on the DM trial and CVD, Anderson and Appel (4) requested an explanation for the paradoxical association between the intervention and CVD risk among women with prior CVD, as well as further explanation for why the CVD finding was null for the overall study population. Our recent paper (1) presented data addressing these interpretational issues, while also providing more comprehensive and longer-term outcome information for trial participants.

Specifically, we determined that the interpretation of overall trial findings, especially for CHD, were limited by postrandomization differential use of statins between the randomized groups in this (necessarily) unblinded study. Fortunately, we have medication inventories at baseline and periodically during follow-up in the WHI trials. These inventories show differential use of statins among women with prior CVD and among women with baseline hypertension, but do not suggest any such differential use among women who were without prior CVD or hypertension at baseline. This coupled with substantially different intervention to comparison group CHD HRs between the normotensive no-prior-CVD and the hypertensive no-prior-CVD groups, and between these groups and the prior-CVD group, support a focus on the 23,248 healthy normotensive women for an intervention evaluation on CVD and mortality outcomes that is unlikely to be confounded by differential statin use.

This focus is reinforced by changes from baseline in LDL-cholesterol during this time of rapid increase in population-wide statin use. For example, intervention-group women with prior CVD had increased LDL-cholesterol postrandomization in contrast to LDL-cholesterol reductions in other-trial women, strongly suggesting nonrecommended cessation of cholesterol-lowering medications in the prior-CVD group. The evidence for postrandomization confounding among baseline hypertensive women without prior CVD is more subtle: the intervention group has an ∼30% higher statin cessation rate and an ∼10% lower statin initiation rate compared with the comparison group. Given the much larger reductions in LDL-cholesterol with statins compared to the type of low-fat intervention studied here, differential statin use of this magnitude could have masked a beneficial effect of the DM intervention on CHD incidence among baseline hypertensive women. As a result, we unfortunately are constrained from using data from these women also for an evaluation of intervention effects on CHD risk.

Collectively, these considerations justify our primary focus on results based on data derived from women who were healthy and normotensive at baseline. Accordingly, one obtains a statistically valid CHD HR of 0.70 (95% CI: 0.56, 0.87) in this stratum. It seems quite valuable to have randomized controlled trial evidence of a practical dietary change leading to an estimated 30% reduction in CHD risk among healthy postmenopausal women. The value of this finding, however, may be offset by a corresponding HR of 1.29 (95% CI: 1.00, 1.66) for total stroke. This may be a chance finding, however, since a stroke elevation was not at all suggested in the hypertensive no-prior-CVD or the prior-CVD strata, and was not corroborated by related mediation analyses (1).

Concerning Dr Merkle's suggestions for improving DM trial design, we note that a trial of ∼1000 healthy postmenopausal women would have unacceptably low statistical power (i.e., < 20%) for any of the designated trial outcomes, even if eligibility requirements were strengthened to enhance homogeneity and to reduce the influence of “extraneous variables.” Also, the 40% intervention to 60% comparison group randomization ratio is statistically valid and fully accounted for in all analyses presented in DM trial reports. This ratio, rather than a 50%:50% ratio, was chosen as a means of controlling total trial costs without reducing study power, following extensive calculations at the design stage (5).