Lifetime cognitive function and timing of the natural menopause

The Medical Research Council National Survey of Health and Development.

The NSHD is a socially stratified birth cohort of 2,548 women and 2,814 men, followed 19 times between their birth during 1 week of March 1946 and the age of 43 years.16,17 At the last home visit, at age 43 years, the population interviewed was representative, in comparison with census data, of the British population of that age in most respects,18 although cohort members were more likely to be married and to have a nonmanual occupation.

In 1993 when the members of the cohort were age 47, a postal questionnaire concerning health during the climacteric was sent to 1,778 women study members with whom there was still contact, and an annual follow-up questionnaire has been sent since then.14 This represents 70% of the original female cohort; 6% of the original total of women in the cohort had died (n = 154), 9% were living abroad (n = 232) and were not in contact with the study, 12% had refused to participate in earlier follow-ups (n = 296), and 3% could not be traced (n = 88). This paper uses data collected between 1993 and 1996 on 1,572 women who completed at least one of the questionnaires (88%); 1,197 (67%) completed all four questionnaires.

Timing of menopause.

Information regarding the date of the last menstrual period was collected on each of the four questionnaires. Date of natural menopause was defined retrospectively after 12 months of amenorrhea. At the last contact, 245 of the 1,572 women (15.6%) were postmenopausal, and 724 (46.1%) were still premenopausal or perimenopausal. Periods had ceased for 288 women (18.3%) because of a hysterectomy or bilateral oophorectomy and for 21 women (1.4%) because of other surgical reasons (mostly endometrial ablation) or medical treatment (for example, chemotherapy). The menopausal status at the last contact could not be classified for 291 women (18.5%) who had started taking hormone replacement therapy (HRT) before the menopause, 1 who gave insufficient information, and 2 who were taking oral contraceptives.

Cognitive measures.

At ages 8, 11, 15, and 26 years study members took tests of verbal and nonverbal ability devised by the National Foundation for Educational Research.19,20 Tests were as follows: age 8—reading comprehension (sentence completion), pronunciation, vocabulary, and nonverbal reasoning; age 11—verbal and nonverbal intelligence (series completion), arithmetic (addition, multiplication, subtraction, and division), pronunciation (as at age 8), and vocabulary (as at age 8); age 15—Group Ability Test AH4 (verbal and nonverbal intelligence), Watts-Vernon Reading Test (sentence completion), and mathematics (arithmetic, geometry, trigonometry, and algebra); age 26—Watts-Vernon Reading Test (as given at age 15, with an additional 10 items of increased difficulty to avoid a ceiling effect). Because the content and number of intelligence tests at 8, 11, 15, and 26 years varied according to age, standardized scores were derived so that the strength of the association between cognitive function at different ages and timing of the menopause could be compared. Each test score was standardized to give a mean of 0 and a standard deviation (SD) of 1, using the total sample with a valid score for each test. Scores representing overall cognitive function in childhood, adolescence, and early adult life were obtained by averaging the individual standardized scores at ages 8, 11, 15, and 26.

At age 43 four tests devised by the NSHD were administered by a trained research nurse during a home visit: 1) Verbal memory (15-word list learning task, with three learning trials). The sum of three trials was taken to give a score out of a maximum of 45. 2) Visual memory (5-item delayed free recall task). 3) Timed visual search task (target letters P and W). The average from three trials was used to derive a speed score (total number of letters scanned in 1 minute) and an accuracy score (number of target letters correctly identified minus targets missed). 4) Timed manual peg placement. The average time, in seconds, to complete this test was based on 10 trials (5 with the right hand and 5 with the left).

Potential confounders.

Potential confounders that might affect the relationship between cognitive function and timing of the menopause were identified from previous research and from preliminary analyses. Potential confounders in childhood were measures of family background (assessed by measures of paternal social class, maternal education, birth order, and maternal age at birth of the participant) and developmental characteristics (birth weight, age at reaching developmental milestones, childhood height and weight, and age at menarche). Paternal social class was assigned using the father’s occupation (classified as professional, managerial, intermediate, skilled manual, semi-skilled manual, and unskilled) when the participant was age 11 years or, if this was unknown, occupation at 4 or 15 years.21 Maternal education was classified into those with primary or secondary education only with no formal qualifications and those with formal qualifications or any further education. Developmental milestones were taken as the age at first sitting up, standing, walking, and talking (use of language beyond “mum,” “dad,” or “nan”), reported when respondents were age 2 years. Age at menarche was based on mothers’ reports when participants were age 15 years, coded as months since birth.

Potential adult confounders were as follows: socioeconomic status (assessed by the social class of the head of the household at 43 years or earlier if this was unavailable, using similar categories to those for paternal occupational social class), own educational attainment (see below), premenopausal body mass index (measured at 36 years), parity, oral contraceptive use at age 43 years or earlier, anxiety and depression at age 43 measured by the Psychiatric Symptom Frequency scale,22 having ever smoked at least one cigarette per day for at least 1 year up to age 43,23 and potential alcohol abuse at age 43,24 represented by at least two positive items on the CAGE screen25 applied to the preceding 12 months.

The highest educational or training qualification achieved by 26 years was classified by the Burnham scale26 and grouped as follows: no qualification, below ordinary secondary qualifications, ordinary secondary qualifications (“O” levels and their training equivalents), advanced secondary education (“A” levels and their equivalents), or higher education (degree level or equivalent).

Statistical methods.

Age at natural menopause, in months from birth, provided the outcome of interest. For women who only gave an age to the nearest year (n = 62), the midpoint of that year was used. This assumption was checked using a sensitivity analysis, but the results were hardly altered and so are not presented here.

Kaplan-Meier estimates of the survivor function27 were produced and plotted to compare survival distributions, with survival being the time until menopause, for three groups of women. Three groups were used to ensure clarity of the plots, and women were allocated to one of the three groups based on their cognitive score, whereby two cut points of the cognitive score distribution were defined so that approximately equal-sized groups were obtained. Log-rank tests28 were performed to test formally the differences in survival between the three different groups.

Cox proportional hazard models29 were used to investigate the combined effects of cognitive scores across the life course. Initially each cognitive score from ages 8 to 26 was entered into the model separately and the linearity of the relationship with the log hazard ratio assessed using a likelihood ratio test to determine whether a quadratic term improved the fit of the model. A hazard ratio for an increase in one standardized unit of each score was thus obtained. Women were categorized into four approximately equal-sized groups based on three cut points of the distribution of measures of cognitive function at age 43 years. This categorization was carried out because the tests related to different types of measures of cognitive function, and were measured in different units, and because the distribution of some of the test scores was highly skewed. Each test score was then fitted either as a categorical factor with four levels or as a linear effect, with the levels of cognitive score coded from lowest to highest as 0, 1, 2, and 3. The latter coding was used if the linear effect was significant at the 10% level and there was no departure from linearity, as indicated by the likelihood ratio test comparing the models with the different parameterizations.30 Starting with the earliest cognitive score associated with age at menopause and working forward in time, a model including all scores independently associated with age at menopause was derived, with the improvement in fit due to the addition of each new cognitive score assessed using a likelihood ratio test. The earliest cognitive score was used as the starting point to assess how far back in the life course any such association may be detected. This model was further adjusted for each of the potential confounders in turn to see whether any of these attenuated the effects of the cognitive tests. The final analyses adjusted firstly for all childhood confounders, to assess whether other early life factors had an impact on the effect of cognition, and secondly for all the adult factors to assess whether any relationship with cognition was due to other risk behaviors being related to cognitive function.

In these analyses, women who had undergone hysterectomy or bilateral oophorectomy before menopause were censored at the age at which the operation took place. Women who were premenopausal or perimenopausal at their last contact were censored at the date they completed the last questionnaire. Women who started taking HRT before the menopause were censored at the date of starting HRT. The small number of women whose periods stopped for other reasons, such as surgery or medical treatment (n = 21) and those who were using oral contraceptives (n = 2) were omitted from the analyses. Also omitted were women who had given insufficient information to assess menopausal status (n = 1) or attribute a date for menopause (n = 7) or for starting HRT (n = 2).

A significant number of women who commenced HRT before their periods ceased for at least 12 months may have been approaching menopause, and therefore the assumption of independence between outcome and censoring would not hold. Further analyses using a competing risks framework31,32 were therefore carried out to assess the possible influence of the HRT users on the results. First, age at either menopause or start of HRT use was taken as the outcome, using whichever event came earlier. Then age at starting HRT was used as the outcome, censoring those who reached natural menopause at the date of this event. As in the main analyses, women who had undergone hysterectomy were censored, as were those who were still premenopausal or perimenopausal at the last contact.

The assumption of proportional hazards made in the Cox model was checked by including a time-dependent covariate in the model and using a likelihood ratio test to assess the improvement in fit due to that term. All analyses were performed using the statistical computer software SAS.33