Posing Causal Questions When Analyzing Observational Data-Reply.

Posing Causal Questions When Analyzing Observational Data To the Editor In their Editorial on evaluating data from observational research, Dr Goodman and colleagues1 mischaracterized the reasons for the apparent difference between the results of the Nurses’ Health Study (NHS) and the Women’s Health Initiative Trial (WHI) of hormone therapy (HT). The main reason for the difference was that in the observational studies, including the NHS, nearly all HT use began in early menopause (within a few years of onset), and most was estrogen alone, without progestin. Because coronary disease is uncommon in early menopause, the WHI targeted an older age group (mean age, 63 years [range, 50-79 years]) to increase statistical power. This created a deviation from the “like vs like” principle enunciated by Goodman and colleagues.1 When the NHS data was reanalyzed to simulate the inclusion and exclusion criteria and dose and formulation of hormone therapy used in the WHI and compared results with the younger women in the WHI,2 the hazard ratios for myocardial infarction were similarly reduced with unopposed conjugated estrogens in both studies.3 For women using conjugated estrogens plus medroxyprogesterone acetate, hazard ratios for myocardial infarction among women close to menopause onset (<10 years) were also similar in both studies. These results demonstrate that the discrepancies between the WHI and NHS could largely be ascribed to differences in timing of HT initiation in relation to age or time since menopause onset. Findings from a meta-analysis of 23 randomized trials support this timing hypothesis, such that HT significantly reduced coronary events in younger women (odds ratio, 0.68 [95% CI, 0.48-0.96]) but not older women (odds ratio, 1.03 [95% CI, 0.91-1.16]).4 Results from the Early vs Late Intervention Trial With Estradiol, which directly tested the timing hypothesis, confirmed that oral estradiol therapy was associated with less progression of subclinical atherosclerosis compared with a placebo when initiated less than 6 years after menopause onset but not when initiated 10 or more years past menopause.5 The differences in results between the NHS and WHI were not due to analytical issues but rather primarily to fundamental differences in the basic causal question—the NHS observed what happens to coronary risk in women starting (mostly unopposed) estrogen in early menopause, whereas the WHI asked about effect when starting estrogen more than a decade, on average, after menopause. Goodman and colleagues1 rightly stress the importance of causal thinking, but this also includes specification of the target population.