MPH 510 Module 6 Bias and Confounding Example

When Observation Misleads: Bias, Confounding and the Hormone Therapy Reversal

Student Name

Master of Public Health Program, Aspen University

MPH 510: Epidemiology in Public Health

Instructor Name

Month Day, Year

When Observation Misleads: Bias, Confounding and the Hormone Therapy Reversal

Epidemiologists must constantly ask whether an association is real or produced by bias or confounding. Few episodes illustrate the stakes better than menopausal hormone therapy. For years, observational studies suggested it protected women's hearts; then a large randomized trial found the opposite. This paper examines how bias and confounding explain the reversal and what methods can prevent similar errors.

The Observational Evidence

By the early 1990s, dozens of observational studies had compared women who used estrogen after menopause with women who did not. A quantitative review of this evidence estimated that estrogen users had about 44% lower risk of coronary heart disease, a relative risk of about 0.56 (Stampfer & Colditz, 1991). Many physicians prescribed hormone therapy partly to prevent heart disease.

The Trial

The Women's Health Initiative enrolled 16,608 postmenopausal women between 50 and 79 years old and randomized them to combined estrogen and progestin or to a placebo. Its safety board halted the estrogen-plus-progestin arm at a mean follow-up of 5.2 years when harms outweighed benefits (Writing Group for the Women's Health Initiative Investigators, 2002). The table shows the main results.

Bias Defined

Bias is a systematic error in how participants are selected or information is gathered that distorts the association. Selection bias arises when the people studied differ in ways related to both exposure and outcome. Information bias arises when exposure or outcome is measured inaccurately or differently between groups.

Healthy User Bias

Women who chose hormone therapy in observational studies tended to be healthier, wealthier and more educated, with better access to care, than women who did not. They were more likely to exercise, eat well and adhere to other treatments. This pattern, known as healthy user bias, made hormone users look healthier for reasons unrelated to the hormones themselves.

Confounding

A confounder is a factor associated with the exposure that independently affects the outcome and is not on the causal pathway. Socioeconomic status is a classic confounder here: it was linked to both hormone use and lower heart disease risk. Observational studies adjusted for measured confounders, but unmeasured or poorly measured factors, such as health consciousness, could not be fully removed.

Timing and Time Since Initiation

A further problem was how observational studies handled time. Many compared current users, often long-term users who had tolerated therapy well, with nonusers. Women who had problems early on tended to stop, leaving healthier survivors among current users. The trial, by contrast, compared women from the start of therapy, capturing early harms.

Reconciling the Evidence

A reanalysis of a large nurses' cohort tried to emulate the trial by comparing women who started hormone therapy with similar women who did not, and following both groups from that point. When analyzed this way, the observational results moved much closer to the trial's findings, suggesting that differences in analysis, particularly comparing prevalent users with nonusers, explained much of the discrepancy (Hernán et al., 2008).

Methods to Control Confounding

In design, investigators can randomize, restrict the study to a narrow group or match on key factors. In analysis, they can stratify, use multivariable regression, apply propensity scores or use methods that emulate a target trial. No analytic method can fully remove confounding by unmeasured factors, which is why randomized trials remain the strongest test when they are feasible and ethical.

Why Observational Studies Still Matter

The reversal does not mean observational studies are unreliable. Many important findings, such as the link between smoking and lung cancer, came from observational research and have been confirmed repeatedly. The lesson is that observational findings suggesting a benefit from a treatment people choose deserve particular caution, because the choice itself is often tied to health.

Public Health Consequences

After the trial's results were published, hormone therapy use fell sharply. Guidelines now recommend against using it to prevent chronic disease and support it mainly for managing menopausal symptoms using the smallest dose that works for as brief a period as needed. The episode changed how researchers and clinicians interpret observational evidence on preventive treatments.

Information Bias

Information bias also played a role. In some observational studies, hormone use was recorded at one point and assumed to continue, although many women stopped or started later. Misclassifying exposure in this way can distort associations in either direction. Heart disease outcomes, too, may have been detected differently in women who saw physicians more often, as hormone users did.

Randomized Trials and Their Limits

Randomization balances both measured and unmeasured confounders on average, which is why the trial could isolate the effect of the hormones. Yet trials have limits: the Women's Health Initiative enrolled women who were on average about 63 years old, many years past menopause, so its results may not apply fully to women starting therapy near menopause. Later analyses suggested that timing may modify risk.

Lessons for Evaluating Interventions

Public health professionals evaluating programs face the same threats. People who enroll in a wellness program may already be healthier than those who do not, making the program look more effective than it is. Comparing enrollees with similar nonenrollees from the point of eligibility, rather than comparing current participants with everyone else, helps avoid this error.

Communicating Uncertainty

The reversal damaged some public trust in medical advice. Explaining the strength of evidence behind recommendations, and noting when advice rests mainly on observational studies, can help the public understand that changing guidance reflects better evidence rather than confusion.

Reverse Causation

A related threat is reverse causation. Women who developed early symptoms of heart disease or other illness might have been advised to stop hormone therapy, leaving healthier women among current users. The apparent protection would then reflect illness causing the change in exposure rather than exposure preventing illness.

Conclusion

The hormone therapy reversal shows how healthy user bias, confounding and the handling of time can make a treatment appear protective when it is not. Understanding these threats, and the methods that address them, allows public health professionals to read observational studies critically and to value randomized evidence where it can be obtained.

References

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

Stampfer, M. J., & Colditz, G. A. (1991). Estrogen replacement therapy and coronary heart disease: A quantitative assessment of the epidemiologic evidence. Preventive Medicine, 20(1), 47-63. https://doi.org/10.1016/0091-7435(91)90006-P

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