Social support and memory in postmenopausal women: a careful null

The idea that an active social life protects the ageing brain is popular, plausible and supported by a good deal of observational research. A secondary analysis published in Menopause on 1 September tested a sharper version of it — whether specific kinds of social support map onto specific cognitive domains in older women — and returned a result worth reporting precisely because it is negative [s1].

Where the data came from

The analysis draws on the ELITE trial (Early versus Late Intervention Trial with Estradiol), a randomised, double-blind, placebo-controlled study of oral 17β-estradiol versus placebo in postmenopausal women, designed mainly to test whether the timing of hormone therapy affects atherosclerosis and cognitive decline [s1][s2]. The present work is a secondary, explicitly hypothesis-generating look at a psychosocial substudy conducted from 2009 to 2012, in which 448 women completed psychosocial assessments every six months [s1].

Cognition was measured at baseline, 2.5 and 5 years using composite scores for executive function, verbal memory, visual memory and global cognition, derived from 14 standardised tests [s1]. Perceived social support was captured with the Medical Outcomes Study Social Support Survey on a 0–100 scale [s1]. The paired analysis included 298 women (mean age 60.3 years) contributing 322 paired cognitive-psychosocial visits, in models adjusted for age, marital status, blood-pressure medication, BMI, education, income, race and randomised treatment [s1].

What it found — and didn't

On first pass, two associations looked positive. Instrumental social support was associated with visual memory (β = 0.0044, SE 0.0022; 95% CI −0.00002 to 0.0088; p = 0.051), and positive social interaction with global cognition (β = 0.0092, SE 0.0046; 95% CI 0.0001–0.0182; p = 0.047) [s1]. In age-stratified analysis, the positive-interaction-and-global-cognition link reached significance among women under 65 (p = 0.042) but not those 65 or older (p = 0.34) [s1].

Then comes the decisive sentence: no association survived false-discovery-rate correction for multiple comparisons [s1]. With four cognitive domains, several support dimensions and repeated tests, some p-values near 0.05 are expected by chance alone, and once the authors accounted for the number of comparisons, the apparent signals dissolved [s1]. Their own conclusion is that no association between specific dimensions of social support and specific cognitive domains was statistically significant after correction [s1].

Why a null like this is useful

This is a small, honest negative result, and it is valuable for two reasons. First, it is a correction against over-reading: had the authors reported only the uncorrected p-values below 0.05, the paper could have been written as "social support boosts memory in postmenopausal women," which the full analysis does not support [s1]. Second, it illustrates the multiple-comparisons trap that inflates a great deal of psychosocial and nutritional research, where many exposures are tested against many outcomes and the winners are reported as if they were the hypothesis all along.

The age-stratified hint is a useful case study in this. Taken alone, a significant association among women under 65 that vanishes in those 65 and older (p = 0.042 versus p = 0.34) reads like a story about a critical window when social engagement matters most [s1]. But splitting an already small sample by age multiplies the number of tests and shrinks each subgroup, which is exactly the setting in which a chance finding can look like a pattern — and none of these survived the correction the authors applied [s1]. Because the parent ELITE trial was built to test the timing of hormone therapy on vascular ageing and cognition rather than social ties, the psychosocial substudy was never powered for the question examined here, which the authors flag by calling the work hypothesis-generating [s1][s2].

Limits and what it does not say

The null does not prove social ties are irrelevant to cognition. The substudy is modest in size, its participants were healthy middle-to-older women in a hormone-therapy trial rather than a general sample, and the follow-up and instruments were designed for other questions [s1]. A true but small effect could easily be missed here. The reasonable takeaway is narrow: in this well-characterised sample, once the statistics were done properly, specific dimensions of perceived social support did not clearly track specific cognitive domains — and claims to the contrary need stronger evidence than uncorrected subgroup findings.

This article is informational and does not constitute medical advice.

Sources

• [s1] Social support and cognitive function in postmenopausal women. Menopause, 1 September 2026. https://doi.org/10.1097/gme.0000000000002877
• [s2] Vascular Effects of Early versus Late Postmenopausal Treatment with Estradiol (ELITE). New England Journal of Medicine, 30 March 2016. https://doi.org/10.1056/NEJMoa1505241