BGPT: Paper Review: Diethylstilbestrol Exposure in Utero and Depression in Women

Explore by Goal

Quick Explanation Copied

Key finding (skeptical read)

In Nurses’ Health Study II, women whose mothers reported/participants reported in utero diethylstilbestrol (DES) exposure had higher odds of depression—both history by 1993 and incident depression over follow-up (age-adjusted OR ~1.4–1.5; multivariable OR ~1.3). However, depression was not clinician-diagnosed; exposure and outcomes were self-reported, so residual confounding and reporting bias remain plausible.

Long Explanation

Paper review (critical): Diethylstilbestrol Exposure in Utero and Depression in Women

Citation: 10.1093/aje/kwq023 (Am J Epidemiol; accepted Jan 14, 2010; published Mar 23, 2010)

1) Visualize the reported effect sizes (from the paper’s numbers)

The paper reports depression-related associations using odds ratios (OR) with and without multivariable adjustment. The only reliable “from-the-paper” quantitative inputs available here are the ORs and the incidence proportions stated in the provided text.

Interpretation guardrails: ORs here summarize an observational association; OR≠causation. Depression was operationalized using self-reported antidepressant use and depressive symptoms rather than DSM clinician diagnosis.

2) Visualize the incidence proportions the authors report

Incident depression during follow-up (1995–2005) is given as a proportion: 19.7% exposed vs 15.9% unexposed.

3) What exactly did they measure?

Exposure

Population: Nurses’ Health Study II women (born during the DES-available era).
DES exposure: self-reported in a 1993 questionnaire via whether mother used DES/“other hormones” during pregnancy; a 2001 supplementary questionnaire was mailed to those reporting exposure; those “certain/somewhat certain” were classified as exposed. A subset where mothers also reported exposure was used to validate reporting; kappa reported as 0.74 agreement.

Outcome

Depression history (baseline): defined as antidepressant use reported in 1993 and depressive symptoms reported in 2001.
Incident depression: first antidepressant initiation during follow-up combined with depressive symptom reporting (using later biennial questionnaires).
Measurement constraint: not based on physician/DSM clinical diagnosis.

4) Quantitative results & attenuation with adjustment

The paper reports that adjustment for depression risk factors and DES/“other hormones”/perinatal correlates attenuated the association only moderately.

Skeptical check

Operationalization: requiring antidepressant use and depressive symptom report increases specificity for “depression-like” states, but does not eliminate diagnostic heterogeneity (antidepressants can be prescribed for other indications). The paper addresses this by excluding antidepressant users without symptoms.
Temporal structure: they claim prospective elements, but note that baseline “history” already uses later symptom reporting (2001) combined with 1993 antidepressant history; this can still be consistent with prospective inference, but it’s not a fully independent baseline clinical assessment.

5) Mechanistic plausibility the authors discuss (and what remains uncertain)

The paper proposes an explanation consistent with an estrogenic endocrine-disruptor hypothesis and discusses possible links between endocrine disruption, neural development, and epigenetic regulation (DNA methylation).

What would disprove/seriously challenge the mechanistic story?

Exposure awareness bias: if the association collapses when exposure status is determined independently of participants’ knowledge/mental-state trajectories, that would argue against a purely physiologic programming effect (authors state they cannot fully rule out awareness-related depression).
Dose-response: the paper lacks direct dose information for NHS II participants; absence of a dose–response in improved datasets would weaken “programming” claims.
Outcome validity: clinician-diagnosed depression outcomes (or validated psychiatric interviews) that reproduce the effect would strengthen causal inference; failure to replicate with better outcomes would weaken interpretation.

6) Critical methodological appraisal (evidence strengths vs blind spots)

Strengths

Large cohort and long follow-up (baseline assessment and follow-up period described).
Partial exposure validation using mother-reported DES exposure (kappa=0.74 reported).
Adjustment for multiple covariate sets: depression risk factors and pre/perinatal correlates of DES exposure, with reported attenuation rather than disappearance.

Blind spots / potential biases

Outcome misclassification: antidepressant use + symptom survey may not map 1:1 to DSM depression; measurement error can bias ORs toward or away from the null depending on differential misclassification.
Exposure misclassification: exposure derived from self-report and “certainty” categories; although agreement was reported in a subset, residual misclassification likely remains.
Awareness bias alternative: if participants’ knowledge of prenatal exposure influences later depression recognition/reporting, association could inflate; authors discuss this possibility and argue it wouldn’t explain their findings, but it cannot be entirely excluded.
Residual confounding: even with many covariates, depression is multifactorial; the paper’s covariate list is limited to variables they measured.

Bottom line (confidence-labeled):

What seems supported by the paper’s data: a statistically significant positive association between DES exposure in utero and depression-related outcomes, with modest attenuation after adjustment.
What remains uncertain: causality and mechanism, given self-report outcome/exposure and potential awareness/reporting biases.

Author reviews (bespoke BGPT links)

Feedback:

Updated: April 15, 2026