NU 602 Unit 5 bias and confounding review example

Why an Observational Study Found Hormone Therapy Protective and a Trial Did Not: A Bias and Confounding Review of Grodstein et al. (2000)

Student Name

Department of Nursing, Herzing University

NU 602: Epidemiology, Research, and Theory

Instructor Name

February 18, 2026

Why an Observational Study Found Hormone Therapy Protective and a Trial Did Not: A Bias and Confounding Review of Grodstein et al. (2000)

The Study

Grodstein and colleagues (2000) analyzed data from the Nurses' Health Study, a prospective cohort of female registered nurses in the United States who have completed mailed questionnaires every two years since 1976. The analysis followed about 70,000 postmenopausal women for up to 20 years. Hormone therapy use was self-reported on each questionnaire and updated every two years, and women were classified as current, past, or never users. Coronary events were reported by participants and confirmed by review of medical records. After adjustment for age, smoking, body mass index, blood pressure, diabetes, cholesterol, and other factors, current users of hormone therapy had a substantially lower risk of major coronary events than never users, roughly 40% lower.

Two years later, the Women's Health Initiative randomized trial of estrogen plus progestin reported the opposite: an increase in coronary events among women assigned to hormone therapy, particularly in the first year (Writing Group for the Women's Health Initiative Investigators, 2002). This review asks which threats to validity in the observational study could explain the difference, how each would arise in this particular study, and in which direction each would push the result.

Threat 1: Confounding by Health-Conscious Behavior

How it would arise here: in the 1980s and 1990s, women chose whether to take hormone therapy, usually after discussion with a physician. Women who chose it, and continued it, tended to be those who saw physicians regularly, had more education and income, exercised, and followed medical advice generally. These characteristics independently lower coronary risk. The analysis adjusted for many measured risk factors, but health consciousness itself is difficult to measure, and adjustment for measured factors can only partly capture it.

Direction: toward apparent protection. Hormone therapy would receive credit for benefits that belonged to the kind of woman who took it.

Threat 2: Prevalent-User Bias

How it would arise here: at any questionnaire, current users included many women who had been taking hormones for years. Any harm occurring early after starting therapy would largely have happened before those women were counted as long-term users, and women who had an early event, or who stopped because of side effects, would not appear among continuing users. The trial found that excess coronary risk was concentrated in the first year after starting. An analysis dominated by long-term users would largely miss that early period.

Direction: toward apparent protection. The women counted as current users were those who had already passed through the period of highest risk.

The strongest evidence for this threat comes from a later reanalysis of the same cohort. When a later team including members of the original investigators restricted the comparison to women starting hormone therapy versus women not starting it, as a trial would, and analyzed the data to mimic a trial's comparison, the apparent protective effect largely disappeared and the results moved much closer to those of the randomized trial (Hernán et al., 2008). That reanalysis used the same women, the same questionnaires, and the same outcome data; only the comparison changed, which isolates the design choice as the source of much of the original finding.

Threat 3: Stopping Therapy When Health Changes

How it would arise here: exposure was updated every two years. A woman who developed chest pain, hypertension, or another warning condition might be advised to stop hormone therapy, and would then be counted as a past user. If she then had a coronary event, it would be attributed to the past-user group rather than to current use. This is a form of reverse causation: an early stage of the outcome influences the exposure.

Direction: toward apparent protection, by moving events out of the current-user group.

Threat 4: Misclassification of Exposure

How it would arise here: hormone use was self-reported every two years, so a woman who started or stopped between questionnaires would be misclassified for part of that interval. Because the reports were collected before outcomes occurred, the misclassification is likely to be similar in women who later did and did not have coronary events.

Direction: usually toward the null. Nondifferential misclassification of this kind tends to weaken an association rather than create one. It cannot explain a strong protective finding, and if anything it means the true association in these data was somewhat stronger than reported.

Threat 5: Detection Differences

How it would arise here: women taking hormone therapy saw physicians more often, which could lead to more nonfatal coronary events being diagnosed among users than among non-users.

Direction: toward apparent harm, the opposite of the other threats. Its presence means that the protective finding held despite a bias working against it, which shows how strong the biases pushing toward protection must have been.

Selection and Generalizability

The cohort consists of nurses, who are more health-literate than the general population and had high follow-up rates. The selection of nurses limits how far the findings can be applied to other women, but it is not in itself a source of internal bias, because the comparison was made within the cohort. High follow-up also reduces the risk of bias from loss to follow-up.

Overall Assessment

Four of the five threats examined would push the result toward apparent protection, and the fifth pushes the other way. Confounding by health-conscious behavior, prevalent-user bias, and exposure changes near the outcome together provide a plausible explanation for how a well-conducted cohort study could find a benefit that a randomized trial did not. The lesson is not that observational studies are unreliable, but that their design must mimic the comparison a trial would make, new users against non-starters from a common starting point, to avoid these particular errors. Part of the discrepancy may also reflect real differences between the populations: women in the cohort mostly started therapy near menopause, while women in the trial were on average older, and later analyses suggest the effect of hormone therapy may depend on age and time since menopause.

References

Grodstein, F., Manson, J. E., Colditz, G. A., Willett, W. C., Speizer, F. E., & Stampfer, M. J. (2000). A prospective, observational study of postmenopausal hormone therapy and primary prevention of cardiovascular disease. Annals of Internal Medicine, 133(12), 933-941. https://doi.org/10.7326/0003-4819-133-12-200012190-00008

Hernán, M. A., Alonso, A., Logan, R., Grodstein, F., Michels, K. B., Willett, W. C., Manson, J. E., & Robins, J. M. (2008). Observational studies analyzed like randomized experiments: An application to postmenopausal hormone therapy and coronary heart disease. Epidemiology, 19(6), 766-779. https://doi.org/10.1097/EDE.0b013e3181875e61

Writing Group for the Women's Health Initiative Investigators. (2002). Risks and benefits of estrogen plus progestin in healthy postmenopausal women: Principal results from the Women's Health Initiative randomized controlled trial. JAMA, 288(3), 321-333. https://doi.org/10.1001/jama.288.3.321