Fertility History and Risk of Cognitive Impairment Among Older Parents in the United States

Abstract

Objectives

I work from a gendered life-course perspective to examine the association between older parents’ fertility history (i.e., timing and parity) and their risk of cognitive impairment in the United States.

Methods

I analyze nationally representative data from 9 waves over 16 years of the Health and Retirement Study (2000–2016). The sample includes 14,543 respondents (6,108 men and 8,435 women) aged 50 and older at the baseline survey. I examine the relationship between parity, age at first birth, and age at last birth with risk of cognitive impairment using nonlinear discrete-time hazard models.

Results

Adjusting for the effects of full covariates, there are U-shaped relationships between women’s age at last birth and risk of cognitive impairment and between women’s parity and risk of cognitive impairment. In the sensitivity tests, the relationships remain robust when sampling weights are applied, or mortality selection is corrected.

Discussion

Fertility timing and parity are likely factors associated with the risk of cognitive impairment for older women. Understanding fertility history and its impact on cognition can help identify the most vulnerable subpopulations, so that more effective interventions can be made to improve cognitive functioning among older adults.

As the aging population increases, cognitive impairment has emerged as a rising public health concern in the United States (Alzheimer’s Association, 2020). The onset and progression of mild cognitive impairment are associated with an increased risk of developing Alzheimer’s or other types of dementia (Alzheimer’s Association, 2020). An increasing number of studies have shown that early- or middle-life experience can predict a long-term, cumulative impact on individuals’ cognitive functioning in later life (Short & Baram, 2019; Wang et al., 2019). As one of the most important life experiences in early or mid-life, fertility or childrearing can significantly shape one’s life contexts and influence health trajectories throughout the life course (Keenan & Grundy, 2019; Umberson et al., 2010). However, research on the linkage between fertility history and later-life cognitive function is limited in the United States.

Existing literature, mainly from outside the United States, shows that the linkage between fertility history (e.g., timing and parity) and cognitive trajectories may arise from both direct and indirect pathways (Bae et al., 2020; Najar et al., 2020; Ning et al., 2020; Read & Grundy, 2017). For example, pregnancy can have direct consequences on maternal health through hormonal changes and reproductive period relates to estrogen exposure, which may account for women’s brain aging and cognitive ability (Hanson et al., 2015; Karim et al., 2016; Ryan et al., 2009). Indirectly, childbearing can affect multiple dimensions of individuals’ social life, such as labor force participation, marital stability, and social integration, which have important implications on cognitive health for both men and women (Ning et al., 2020; Read & Grundy, 2017; Umberson et al., 2010). This study uses a nationally representative, longitudinal data set collected over 16 years to examine how the timing of fertility and parity are associated with the risk of cognitive impairment in the United States. I use continuous measures of fertility variables and hypothesize nonlinear relationships between fertility variables and the risk of cognitive impairment by showing gender-specific patterns. The analysis will focus on potential indirect pathways rather than female-specific factors as many efforts have been made to examine the latter.

Fertility History and Cognitive Function: A Gendered Life-Course Perspective

A life-course framework often guides research on the mechanisms linking fertility and parents’ health. People’s fertility intentions and behaviors are embedded in historical time, which could have a long-term impact on their life domains. Timing is one of the essential principles of the life-course approach (Elder, 1994). Individuals develop an expectation of a “normative life cycle,” in which certain life events occur at certain ages (Neugarten, 1979). Therefore, “off-time” transitions into parenthood often produce life stress and may negatively affect parents’ well-being, both physically and mentally (Henretta et al., 2008; Koropeckyj-Cox et al., 2007; Mirowsky, 2005). For example, early fertility timing (i.e., teenage parenthood) often indicates poor preparation for parental roles, interrupts young parents’ educational or occupational attainment, and increases the risk of singlehood and marital instability (Koropeckyj-Cox et al., 2007; Lacey et al., 2017; Mirowsky, 2005). By contrast, giving birth at a typical age or a slightly delayed age is more likely to benefit parents’ well-being because parents often have acquired social resources that help them cope with the costs and stresses of childbearing.

Parity (i.e., number of biological children) can also significantly shape parents’ social contexts and thus affect their health. The life-course perspective emphasizes “linked lives,” suggesting that the interdependence between parent and child has significant implications on parents’ social engagement and mental stimulation, and the number of children could potentially diminish or amplify this effect (Umberson et al., 2010). Low parity is related to smaller kin network and fewer of these kin to interact with due to fewer children, children-in-law, and/or grandchildren, which is likely associated with higher odds of living alone but lower odds of social engagement than peers with larger and close kin (Margolis et al., 2022). A lack of support from children may also increase older parents’ isolation and loneliness, which are related to an elevated risk of cognitive decline (Kuiper et al., 2015). High parity may negatively influence parents’ health because additional children diminish parents’ economic resources and increase perceived demands, reducing parents’ time and resources to take care of themselves (Umberson et al., 2010). Moreover, acute parenting stress caused by high parity can evoke pathophysiological metabolic effects and adverse changes in stress hormones and specific brain regions (Henckens et al., 2009; Rothman & Mattson, 2010). Stress coping strategies may also involve unhealthy behaviors (e.g., heavy smoking and drinking, inactivity) that could lead to a higher risk of cognitive impairment in later life (Hayes et al., 2016; Kirk-Sanchez & McGough, 2014; Swan & Lessov-Schlaggar, 2007).

Fertility could influence men’s and women’s cognitive trajectories through different mechanisms, as pregnancy is a uniquely female experience (Ning et al., 2020; Read & Grundy, 2017). For women, fertility history has been considered an indicator of estrogen exposure. Cumulative estrogen exposure has been found to be a factor associated with the prevalence of dementia among women (Peterson & Tom, 2021; Ryan et al., 2009). Moreover, physiological changes caused by pregnancy can affect mothers’ comorbidity, such as diabetes, cardiovascular disease, and depression, that have been examined as factors associated with cognitive performance in later life (Ahtiluoto et al., 2010; Deckers et al., 2017; Hanson et al., 2015; Lacey et al., 2017). However, the biological process of fertility does not operate on men. Thus, fertility could influence men’s cognition through indirect pathways (Ning et al., 2020; Read & Grundy, 2017). For example, off-time fatherhood or high parity could link to fathers’ financial strain and parenting stress. Studies found that teen fatherhood is associated with high rates of school dropout, unstable work, incarceration, and mental illness (Settersten & Cancel-Tirado, 2010). Fathers who have a first birth at older ages are more likely to be obese and smokers and to consume alcohol (Lawson & Fletcher, 2014). Noticeably, the causality between timing of fatherhood and father’s outcomes may be bidirectional and include a selection effect. This indirect pathway could also apply to women: Teenage mothers often experience more education or career penalty than teenage fathers, leading them to a cumulative disadvantage in occupational attainment, sources of income, and union stability, thus eventually having a long-term impact on women’s health (Brand & Davis, 2011; Mollborn, 2007). Mothers are also more likely to be emotionally involved in children’s lives and therefore gain stronger relationship ambivalence, which may increase parenting stress (Pillemer & Jill Suitor, 2002; Thomas & Umberson, 2018).

Empirical Evidence

Fertility Timing and Parents’ Cognitive Health

Although the association between fertility timing and parental health is fairly well researched, few studies focus on the impact of fertility timing on parents’ cognitive health. Among the limited number of studies, relatively consistent evidence suggests that early childbearing is negatively associated with, while late birth may benefit, the cognitive health of parents (Karim et al., 2016; Read & Grundy, 2017; Ryan et al., 2009). For example, a population-based study in France analyzed 996 women aged 65 years and older and found that compared with women who first gave birth between ages 21 and 29, those who first gave birth before age 21 showed an increased risk of poor cognitive performance (Ryan et al., 2009). Another study analyzed randomized clinical trial data of 830 naturally menopausal women in California and found that later age at last pregnancy (>35 years old) was associated with better verbal and cognitive performance (Karim et al., 2016). The authors argued that the benefit of late pregnancy for women’s cognition was not attributable to biological mechanisms but rather might reflect socioeconomic and lifestyle factors (Karim et al., 2016). Similarly, Read and Grundy (2017) used national data of men and women aged 50 and older in England and found that early age of parenthood (<20 for women, <23 for men) was associated with poorer cognitive functioning for both men and women, but this association was mediated by parents’ socioeconomic status (SES), health conditions, and health behaviors. Moreover, late age at last birth (>35) was associated with better cognitive function for women, net of all other covariates. However, a recent study by Gemmill and Weiss (2022) using national, longitudinal data suggested older age at last birth (>35) is associated with the incidence of dementia for women in the United States.

Parity and Parents’ Cognitive Health

Parity (i.e., number of biological children) has often been considered as a factor influencing parents’ health, and a growing body of research has examined the linkage between parity and cognitive health among the aging population (Högnäs et al., 2017; Saenz et al., 2021; Ward et al., 2009). Existing literature yields relatively consistent findings, indicating high parity is associated with an increased risk of cognitive limitation (e.g., Bae et al., 2020; Saenz et al., 2021). However, the use of study populations, sample ages, reference groups, confounding factors, and classification of low/high parity vary a lot across studies (e.g., Gemmill & Weiss, 2022; Read & Grundy 2017; Saenz et al., 2021; Song et al., 2020). For example, Bae et al. (2020) pooled data on women aged 60 or older from six population-based studies across four European and two Asian countries. They found that higher parity (5+ children) increased the risk of dementia compared with lower parity (1–4 children; Bae et al., 2020). The study by Song et al. (2020) analyzed population-based data in Singapore on women ages 45–74. They found that those who had high parity (5+ children) showed an increased risk of cognitive impairment compared to women with lower parity (1–2 children; Song et al. 2020).

Moreover, studies examining gender differences are far from conclusive as some evidence shows that the linkage between parity and cognition varies by gender while others did not. For example, a cross-sectional study in Mexico found that high parity (>6 vs. 2–3 children) was associated with poorer cognitive function, regardless of gender, whereas low parity (0–1 child) was related to poorer cognitive ability only for mothers, not fathers (Saenz et al., 2021). Another study focusing on mid-to-old age parents in the United Kingdom found that compared with childless people, parents with two or three children showed faster response time, more accurate visual memory, and younger brain age, more so for men than women (Ning et al., 2020). The authors speculated that the lifestyle factors related to parenthood rather than the biological process of pregnancy contributed to this relationship (Ning et al., 2020). The study by Read and Grundy (2017) suggests that compared with medium parity (2 children), low (0–1 child), and high parity (>3 children) were associated with poorer cognitive functioning for both men and women.

Taken together, the present study will examine the association between fertility history—timing and parity—and risk of cognitive impairment using a longitudinal, nationally representative study in the United States. I use continuous fertility variables rather than defining atypical fertility behaviors by categorical measures. Previous studies using cut-off points (e.g., >35 = late parenthood, 2–3 kids = medium parity) may have neglected internal heterogeneity within the groups with atypical fertility behaviors. Moreover, the cut-offs are inconsistent across studies that yield challenges to compare findings (e.g., Bae et al., 2020; Read & Grundy, 2017; Saenz et al., 2021). I assess whether the associations are robust by adjusting for potential confounding effects, including parents’ SES and health factors. I also controlled for cohort as a covariate because desired age at birth and number of children can vary by cohorts (Koropeckyj-Cox et al., 2007; Mirowsky, 2002). I present gender-stratified results because men and women show differences in age patterns of fertility timing, the prevalence of cognitive impairment, and biological mechanisms linked to reproduction.

Data and Methods

Data

This study used the data from the Health and Retirement Study (HRS) 2000–2016. The HRS is a nationally representative, longitudinal data set collected biennially by the Institute for Social Research at the University of Michigan. The HRS is applicable to the current research question because of its large sample size, long-term follow-up, high response rates (81%–89%), and high-quality measures of cognitive health and family relationships among adults aged 50 and older. As the earliest wave that included completed cognition measures was taken in 2000, I selected the year 2000 as the baseline wave, including a sample of 19,579 adults and their spouses. I restricted the final sample to respondents age 50 and older (2.83% of respondents under age 50 were excluded). I further excluded missing values in key variables of analysis (3.78%), as well as likely coding errors in age at first/last birth and number of children (negative number, younger than 13, and older than 55, more than 13 children; 0.99%). The final analytical sample includes 14,543 respondents (6,108 men and 8,435 women), contributing to 61,622 person-period records across 9 waves over 16 years.

Measures

Outcome variable: cognitive impairment

For self-reporting respondents, the HRS assesses cognition using the modified version of the Telephone Interview for Cognitive Status. A final test score was calculated by summing the following cognitive items: immediate and delayed recall of a list of 10 words (1 point for each), five trials of serial 7s (i.e., subtract 7 from 100, and continue subtracting 7 from each subsequent number for a total of five trials, 1 point for each trial), and backward counting (2 points). The final score ranged from 0 (severely impaired) to 27 (high functioning; Crimmins et al., 2016; Liu et al., 2020). Respondents whose scores were 0–11 were classified as having cognitive impairment; those whose scores were 12–27 were classified as having normal cognition (Crimmins et al., 2016).

For respondents unable to complete the interviews independently, either due to illness or cognitive impairment, proxies assessed their cognitive status. In this case, respondents’ cognitive status was measured by an 11-point scale using the proxy’s assessments of (a) the respondent’s memory (0 = excellent, 4 = poor) and (b) the respondent’s limitations in five instrumental activities of daily living: managing money, taking medication, preparing hot meals, using the phone, and shopping for groceries (0–5), as well as (c) the interviewer’s assessment of the respondent’s difficulty completing the interview because of cognitive limitations (0 = none, 1 = some, and 2 = prevented completion; Liu et al., 2020). Proxy respondents with a summary score of 3–11 were classified as having cognitive impairment, and those with a score of 0–2 were classified as having normal cognition (Crimmins et al., 2016).

Independent variables: fertility history

Age at first/last birth (time-invariant) was calculated by subtracting the age of the oldest/youngest child from the parents’ current age at the baseline wave, respectively. The final sample only included parents who had biological children and excluded parents who had stepchildren, other types of children, and childless respondents. In the analysis, ages at first/last birth for men and women were centered at the gender-specific grand mean (see Table 2). There are 12.69% of respondents who had only one biological child. Thus, their age at first birth is the same as the age at last birth.

Table 2.

Descriptive Statistics of All Analytical Variables at Baseline, HRS 2000, Mean (SD)/%

Variables	Total (N = 14,543)	Men (n = 6,108)	Women (n = 8,435)	T-test/Pr-test
Fertility history
Age at first birth (13–54)	24.22 (5.25)	25.87 (5.37)	23.02 (4.83)	*
Age at last birth (13–55)	31.47 (6.41)	33.28 (6.50)	30.16 (6.02)	*
Number of biological kids (1–13)	3.15 (1.79)	3.20 (1.77)	3.12 (1.80)	*
Age (50–104)	71.29 (9.66)	70.90 (9.00)	71.57 (10.11)	*
Race
Non-Hispanic White (ref)	77.25	79.16	75.86	*
Non-Hispanic Black	12.67	10.97	13.91	*
Hispanic	6.59	6.47	6.69
Other	3.49	3.41	3.54
Education
Less than high school (ref)	28.91	28.09	29.51
High school	34.22	30.12	37.19	*
Some college	27.68	29.06	26.69	*
College above	9.18	12.72	6.62	*
Proxy report indicator
Self-report (ref)	89.58	85.51	92.53	*
Proxy report	10.42	14.49	7.47	*
Marital status
Married (ref)	61.83	80.08	48.62	*
Unmarried	38.17	19.92	51.38	*
Number of marriages
Never married (ref)	0.86	0.43	1.17	*
Married once	79.89	78.01	81.26	*
Married twice	15.58	17.17	14.43	*
Married three times and more	3.66	4.39	3.14	*
Cohort
<1924 (ref)	25.11	19.43	29.22	*
1924–1930	11.92	10.85	12.70	*
1931–1941	51.08	56.40	47.23	*
1942–1947	11.88	13.31	10.85	*
Father’s education (0–17)	8.89 (3.27)	8.93 (3.42)	8.85 (3.15)
Mother’s education (0–17)	9.15 (3.09)	9.29 (3.18)	9.04 (3.02)	*
Household income (unit: 10k)	4.98 (7.84)	5.88 (8.94)	4.33 (6.86)	*
Wealth (unit: 10k)	43.33 (114.33)	50.35 (120.63)	38.25 (109.27)	*
Chronic condition (0–4)	1.13 (0.98)	1.17 (1.00)	1.10 (0.97)	*
Smoking
Never smoke (ref)	41.99	27.62	52.40	*
Former smoker	45.44	58.76	35.80	*
Current smoker	12.56	13.62	11.80	*

Notes: Value of wealth includes debts. Differences between men and women are tested by two-tailed t-test or proportion tes. Significant differences are marked in the fifth column. HRS = Health and Retirement Study.

*p < .05.

Number of biological children (time-invariant) was based on a question asking respondents, “How many children have you fathered/given birth to, excluding miscarriages or stillbirths and adopted or stepchildren?” In the analysis, number of biological children was centered at individual’s mean (see Table 2).

Covariates

SES (time-varying) included two variables: (a) Household income was measured by respondent’s and spouse’s income from all sources for the last calendar year (e.g., earnings, pensions, Social Security benefits, unemployment and workers’ compensation, etc.) and (b) Net value of total wealth was measured by the sum of all wealth components (e.g., the net value of primary residence, vehicles, business, stocks, etc.) minus all debt (e.g., mortgages, home loans). Because these two variables had zero and negative values, I followed the methods in previous studies to adjust them by adding a constant of $1 for both income and wealth and a year-specific constant for wealth (i.e., the minimum value at each wave). By doing this, I transformed both variables into positive values. I further divided the values of household income and wealth by the square root of household size and took the natural logs to adjust for the skewness of the distribution (Zhang & Hayward, 2006; Liu et al., 2020).

Health factors (time-varying) included two variables: (a) Smoking status: 0 = never smoke [reference], 1 = former smoker, 2 = current smoker, and (b) Chronic conditions was measured by a comorbidity index ranging from 0 to 4, with a higher score representing more comorbidities. The index summarizes four major chronic conditions—diabetes, stroke, heart disease, and high blood pressure—because prior studies have found that these chronic conditions have common links to the incidence of cognitive decline (Stampfer, 2006; Saedi et al., 2016; Deckers et al., 2017; Morley, 2017).

Other covariates included age (centered at the means for men and women, respectively), race/ethnicity (non-Hispanic white, non-Hispanic black, Hispanic, others), education level (less than high school, high school degree, some college, and college above), respondents’ marital status (married and cohabiting vs. unmarried [divorced, widowed, and never married]), number of marriage (0 [never married], 1, 2, 3, and more marriages), indicator of proxy report (self-report vs. proxy report), cohort (<1924, 1924–1930, 1931–1941, and 1942–1947), and father’s and mother’s education attainment (0–17 years). Gender, race/ethnicity, education, cohort, and parents’ education are time-invariant variables, whereas other covariates are time-varying.

Analytical Strategy

Preliminary results based on the total sample show that the association between the fertility variables and risk of cognitive impairment displayed nonlinear relationships. Hence, I used nonlinear discrete-time hazard models to estimate the risk of cognitive impairment by the fertility history variables for men and women, separately. The nonlinear discrete-time hazard model is specified as follows:

$${\displaystyle \begin{array}{l}{\displaystyle \log \frac{h(t_{ij})}{1-h(t_{ij})}={\displaystyle \underset{j=1}{{\displaystyle \sum ^9}}}{\alpha }_j{\mathbf{D}}_{ij}+{\mathrm{\beta }}_1{\mathbf{X}}_i+{\mathrm{\beta }}_2{\mathbf{X}}_i^2+{\mathrm{\beta }}_3{\mathbf{Y}}_{ij}}\end{array}}$$

where $h(t_{ij})$ indicates the discrete hazard (i.e., conditional probability) of cognitive impairment for individual i at time j; $\underset{j=1}{\sum ^9}{\alpha }_j{\mathbf{D}}_{ij}$ represents the set of intercepts for the nine waves of the HRS from 2000 to 2016; ${\mathbf{X}}_i$ indicates the vector of time-invariant covariates including fertility history variables; X_i² is the quadratic term of fertility history variables; Y_ij indicates the vector of time-varying covariates; and β _1, β _2, and β ₃ are corresponding coefficient vectors.

I first presented the correlation between three fertility variables and descriptive statistics of analytical variables at baseline. Next, I reported three panels of models, in which each panel includes four models for both men and women. Models 1 and 5 only adjust for the effects of age (centered). Models 2 and 6 add demographic characteristics. Models 3 and 7 add SES (i.e., household income and net wealth). Models 4 and 8 add health factors (i.e., smoking and chronic conditions). This step-by-step model-building strategy follows methods in previous literature (Read & Grundy 2017), aiming to look at changes of main results by adding demographic information, SES, and health-related factors. This can help examine which factors increase or attenuate the association between fertility and cognition.

Three fertility factors were tested separately in different models (i.e., age at first birth, age at last birth, and number of children are not included in one model simultaneously). To assist in interpreting the nonlinear results, I created figures to display the predicted probability of cognitive impairment by the fertility variables. I reported significant results in the full models in the figures. To test the robustness of the results, I added three sensitivity tests in Table 4: (a) To adjust for the complex study design and generalize the findings to the whole population, I tested weighted results using personal level sampling weights; (b) to address the issue of selective mortality, I conducted a two-step Heckman approach to control for mortality selection (Heckman, 1979); and (c) because the fertility variables correlate with each other, I tested the results including three fertility variables in the same model. I used Stata 15 to estimate the models and generate the figures (StataCorp, 2017).

Table 4.

Sensitivity Test Results

	Test 1 Weighted results		Test 2 Two-step Heckman		Test 3 Fertility variables in one model
Variables	Men	Women	Men	Women	Men	Women
Panel A
Age at first birth (centered)	0.990	0.994	0.990	0.992	0.991	0.999
	(0.006)	(0.006)	(0.005)	(0.005)	(0.006)	(0.006)
Quadratic age at first birth	1.001	1.001	1.001	1.001	1.001*	1.001
	(0.001)	(0.001)	(0.001)	(0.001)	(0.000)	(0.001)
Panel B
Age at last birth (centered)	0.992	0.991*	0.993	0.991*	0.992	0.990*
	(0.004)	(0.004)	(0.004)	(0.004)	(0.005)	(0.005)
Quadratic age at first birth	1.001	1.001**	1.000	1.001**	1.001	1.001*
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Panel C
Number of children (centered)	0.987	0.960*	0.997	0.954**	0.992	0.960*
	(0.019)	(0.016)	(0.018)	(0.015)	(0.022)	(0.019)
Quadratic number of children	1.002	1.009*	1.001	1.010**	1.002	1.010**
	(0.005)	(0.004)	(0.004)	(0.003)	(0.004)	(0.003)

Notes: For both men and women, only full models were shown (controlling for all covariates). The quadratic terms were based on centered fertility measures. Standard errors in parentheses.

**p < .01, *p < .05.

Results

Table 1 shows the pairwise correlation between three fertility variables at baseline. For both men and women, age at first birth is positively correlated with age at last birth, whereas timing of birth is negatively correlated with total number of children. Table 2 displays the descriptive statistics of all analyzed variables at baseline. Significant differences between men and women, marked in the last column, were tested by two-tailed t-test or proportion tests (p < .05). Men were about 3 years older at first birth and last birth than women, on average. The mean number of biological children (i.e., average parity) was about 3 for both genders. Fathers were more likely to have higher parity than mothers. The 14,543 respondents at baseline contributed to 61,622 person-period observations over the 16 years of study length. Among 61,622 person-period observations, there were 8,252 cases of cognitive impairment (3,443 cases for men and 4,809 cases for women), yielding 13.39% incidence rate of cognitive impairment (13.69% for men and 13.18% for women).

Table 1.

Pairwise Correlation Between Fertility Variables by Gender, HRS 2000, N = 14,543

	Men (n = 6,108)			Women (n = 8,435)
	Age at first birth	Age at last birth	Number of children	Age at first birth	Age at last birth	Number of children
Age at first birth	1.000			1.000
Age at last birth	0.498*	1.000		0.428*	1.000
Number of children	−0.317*	0.389*	1.000	−0.341*	0.434*	1.000

Note: *Significant correlation coefficients at the 5% level or better. HRS = Health and Retirement Study.

Table 3 presents the estimated odds ratios of cognitive impairment from the nonlinear discrete-time hazard models for men and women, separately (covariates were not shown but can be found in Supplementary Materials).

Table 3.

Estimated Odds Ratios of Discrete-Time Hazard Models, Fertility History, and Risk of Cognitive Impairment for Men and Women, HRS 2000–2016, N of person-periods = 61,622, N of respondents = 14,543

	Men (n = 25,142)				Women (n = 36,480)
Fertility variables	M1	M2	M3	M4	M5	M6	M7	M8
Panel A
Age at first birth (centered)	0.962***	0.987**	0.987**	0.989*	0.938***	0.990*	0.992	0.996
	(0.004)	(0.005)	(0.005)	(0.005)	(0.004)	(0.005)	(0.005)	(0.005)
Quadratic age at first birth	1.003***	1.001**	1.001*	1.001*	1.005***	1.001**	1.001*	1.001
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.001)	(0.001)	(0.001)
Panel B
Age at last birth (centered)	0.998	0.995	0.992*	0.992*	0.992**	0.995	0.991*	0.992*
	(0.003)	(0.003)	(0.004)	(0.004)	(0.003)	(0.003)	(0.003)	(0.003)
Quadratic age at first birth	1.002***	1.001*	1.001	1.001	1.003***	1.001*	1.001*	1.001*
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Panel C
Number of children (centered)	1.050**	1.009	0.992	0.991	1.025	0.975	0.957**	0.955***
	(0.016)	(0.017)	(0.016)	(0.016)	(0.013)	(0.013)	(0.013)	(0.013)
Quadratic number of children	1.012**	1.002	1.002	1.002	1.018***	1.009**	1.010**	1.010**
	(0.004)	(0.004)	(0.004)	(0.004)	(0.003)	(0.003)	(0.003)	(0.003)

Notes: For both men and women, Model 1 controls for centered age. Model 2 adds race/ethnicity, education, marital status, indicator of proxy report, cohort, number of marriages, and father’s and mother’s education attainment. Model 3 adds household income (logged) and net wealth (logged). Model 4 adds smoking and chronic conditions. Three fertility variables are not included in the model simultaneously. The quadratic terms were the product of centered fertility measures. Standard errors in parentheses. HRS = Health and Retirement Study.

***p < .001, **p < .01, * p < .05.

Age at First Birth

For men, Models 1–4 in Panel A show the association between men’s age at first birth (centered) and risk of cognitive impairment, adjusting for the effects of men’s age (Model 1), demographic information (Model 2), SES (Model 3), and health-related factors (Model 4). Both the linear and quadratic terms remained significant in the full model (Model 4: β ₁ = 0.989, p < .05; β ₂ = 1.001, p < .05), indicating a U-shaped (convex) curve between men’s age at first birth and odds of cognitive impairment. Figure 1A shows the results from Model 4 in Panel A (the red reference line suggests that the mean age at first birth for men is 26 years old). Both very early and delayed age at first birth were associated with increased risk of cognitive impairment for men. For women, although Models 5 and 6 also show a U-shaped relationship, the effects turned to less significant and nonsignificant in Models 7 and 8. This indicates that women’s SES and health in later life could largely explain the association between age at first birth and odds of cognitive impairment.

Figure 1.

(A) The results of Panel A Model 4: the association between men’s age at first birth (centered in the analysis) and men’s probabilities of cognitive impairment, controlling for all covariates. (B) The results of Panel B Model 4: the association between women’s age at last birth (centered in the analysis) and women’s probabilities of cognitive impairment, controlling for all covariates. (C) The results of Panel C Model 4: the association between women’s number of children (centered in the analysis) and women’s probabilities of cognitive impairment, controlling for all covariates. The red reference line in each figure shows the mean of men’s age at first birth (= 26), the average of women’s age at last birth (= 30), and the mean of women’s number of children (= 3). Only statistically significant effects are shown in figures.

Age at Last Birth

For men, Models 1–4 in Panel B do not show consistent nonlinear associations between men’s age at last birth and odds of cognitive impairment. However, for women, Model 8 in Panel B indicates a robust association between women’s age at last birth and odds of cognitive impairment, regardless of women’s SES and health (β ₁ = 0.992, p < .05; β ₂ = 1.001, p < .05). Figure 1B displays the results from Model 8 in Panel B, indicating a U-shaped like relationship: approximately, either very early or very delayed age at last birth was associated with increased risk of cognitive impairment for women (the red reference line suggests that the mean age at last birth for women is 30 years old).

Parity

For men, Models 2–4 in Panel C do not show significant associations between men’s parity and odds of cognitive impairment. In contrast, for women, Models 5–8 in Panel C show consistently significant nonlinear relationships between parity and women’s odds of cognitive impairment (e.g., Model 8: β ₁ = 0.955, p < .001; β ₂ = 1.010, p < .01). Figure 1C displays the results from Model 8 in Panel C. This pattern demonstrates that both low and very high parity were associated with increased probabilities of cognitive impairment for women, regardless of women’s SES and health in later life.

Sensitivity Tests

Sensitivity tests were shown in Table 4. First, Test 1 suggests that compared with unweighted results (Panel A Model 4 in Table 3), the association between men’s age at first birth and odds of cognitive impairment does not persist in the weighted model. In contrast, the associations between women’s age at last birth/parity and cognitive impairment remain robust in weighted results. Test 2 presents the results using the Heckman approach to correct mortality selection. Estimates on cognitive impairment risk should be interpreted as adjusted for the unobserved factors that may affect cognition and tendency for death. The results show similar findings to the weighted results. Test 3 displays the results controlling for the other two fertility variables in the same model (linear terms only), which do not alter the main findings.

Discussion

A growing body of research has examined the relationship between fertility history and cognitive function among the aging population (e.g., Gemmill & Weiss, 2022; Ning et al., 2020; Read & Grundy, 2017). This study adds to this group of research using a nationally representative, longitudinal data set to explore the association between continuous fertility measures and risk of cognitive impairment among older men and women in the United States. Guided by a gendered life-course perspective, this study examines nonlinear relationships between fertility history and the risk of cognitive impairment for both older men and women, highlighting the effects of timing and parity on cognitive health. The findings also advance a recent study by Gemmill and Weiss (2022) using the same data sets and measures in several aspects.

First, the relationship between age at first birth and risk of cognitive impairment displays a U-shaped pattern for men, indicating both early and delayed first birth were associated with increased risk of cognitive impairment in later life. However, this relationship does not persist when sampling weights are applied, or mortality selection is fixed. Because no prior evidence shows such association, this is the first study that focuses on how the timing of fatherhood may link to fathers’ cognitive health, which calls for more replication work to test the robustness of the results. Moreover, readers should be cautious in interpreting the results and generalizing to the whole population. For women, age at first birth is not associated with their risk of cognitive impairment, likely due to women’s economic resources and health conditions in later life. This overall null association for both genders, as well as potential mediating effects of SES and health, is consistent with prior literature using population-based data sets in the United States and United Kingdom (Gemmill & Weiss, 2022; Read & Grundy, 2017). For example, Gemmill and Weiss (2022) found no significant association between age at first birth and risk of dementia for both genders when covariates (including SES and health) were adjusted. Read and Grundy (2017) also found that the significant association between early parenthood and poorer cognitive functioning does not persist when SES and health were considered. One possible reason may be the “survivor effect” among parents who had off-time first birth but live longer. Prior research indicates that early or late motherhood often relates to a higher risk of mortality and morbidity (Henretta et al., 2008; Lacey et al., 2017; Mirowsky, 2005; Pirkle et al., 2014). Therefore, mothers with atypical timing of first birth in the final sample are “survivors” with better health or more agency and resources to cope with the adverse effects of the off-time parenthood.

Second, off-time last birth was associated with increased risk of cognitive impairment for women, regardless of women’s SES and health in later life, whereas this association was not found among men. This finding is consistent with the study by Gemmill and Weiss using the same data, which suggests that older age at last birth (>35 years old) was associated with increased incident dementia for women but not for men (see Author Note 1; Gemmill & Weiss, 2022). This finding may indicate a direct, biological process linking late last birth and cognitive health for women. Late fertility can increase mothers’ health risks, with high rates of stillbirths, miscarriage, and maternal morbidity, because aging leads to decreased fecundity and deterioration of physiological functions (Lisonkova et al., 2017). Moreover, the present study indicates that the increased risk is more serious among women with very late last birth, especially those close to typical age at menopause (45–55). A delayed last birth could relate to late age at menopause or a longer reproductive period, which are factors associated with an increased risk of dementia (Najar et al., 2020). Some evidence shows protective effects of late fertility on parents’ cognition because of the interaction with young children (e.g., reading, playing games; Ryan et al., 2009; Karim et al., 2016; Read & Grundy, 2017). However, this effect may not apply to women with a very late last birth because mothers who begin to have aging-related health conditions may experience greater stress in parenting minor children. Moreover, this nonlinear (U-shaped) relationship also indicates that women’s early last birth could relate to increased risk of cognitive impairment. This result is robust when controlling for age at first birth and number of children. Having last birth at an early age could be due to even earlier first birth as well as infertility at an early age, which is likely related to hormone imbalance or anovulation and could be a mechanism linked to later-life cognitive functioning (Song et al., 2020). The finding of the U-shaped relationship advanced Gemmill and Weiss’s work as well as previous research by showing the risk of cognitive impairment of early last birth and risk heterogeneity of late last birth (35–45 vs. 45–55).

Last, both low and high parity were associated with increased risk of cognitive impairment for women, with a very high parity makes this effect stronger, which is consistent with prior evidence (Bae et al., 2020; Read & Grundy, 2017; Saenz et al., 2021; Song et al., 2020). Low parity is related to smaller family size and social network and likely reduce later-life social contacts with respect to the interaction with children or grandchildren. It has been well-recognized that a lack of social engagement and contact can reduce mental stimulation and brain reserve (Kuiper et al., 2015). This association between high parity and cognitive impairment may suggest a direct, biological process: Changes in estroprogestinic ratio during pregnancy or increased exposure to progesterone and/or estrogen can escalate the risk of cognitive decline or Alzheimer’s disease (Najar et al., 2020). High parity may also influence mothers’ cognitive health through social processes related to parenting stress (Umberson et al., 2010). A greater number of children may increase mothers’ time spent on parenting and housework and even alter mothers’ employment history and occupation choices. The “linked lives,” from a life-course perspective, emphasize the interconnectedness between family members. Mothers are often more socially and emotionally involved in children’s lives and experience relationship ambivalence. Parenting a greater number of children could amplify this effect. Indeed, a study by Thomas and Umberson (2018) found that relationship strain with children protected against fathers’ cognitive function but not for mothers because conflict with children may place more stress on mothers than fathers. Noticeably, this finding is not consistent with the study by Gemmill and Weiss that found no relationship between parity and risk of dementia for both genders (Gemmill & Weiss, 2022). This is likely due to smaller dementia cases than cognitively impaired cases, reducing their analytical power. It could also lie in that a categorical measure of parity may neglect the within-group heterogeneity.

This study is not without limitations. First, to calculate parents’ age at first birth, the analytical sample was restricted to parents who had biological children. Nonparents who potentially had fertility history (i.e., miscarriage, stillbirth) were excluded. Future studies can use a broader definition of fertility history to test whether this relationship exists among a larger population. Second, although dementia cannot be reversed, a mild cognitive impairment may be due to treatment. The recurrent cases (i.e., a normal cognition coming after an impaired status) do exist in the HRS longitudinal data. There is a lack of information to distinguish between measurement errors, misclassification of status, and real recovery. Future studies can explore why there are recurrent cases and to what extent they may influence dementia or mortality risk. Third, the sample in this study is composed of specific cohorts born before the 1950s in the United States. It cannot be known whether these findings apply to all Americans. With the changing social norms and increasing use of assisted reproductive technology, the timing of birth may become less important to individuals’ health in younger cohorts. Last, the association between fertility and cognition could be affected by unmeasured factors, making the study assumption biased. Future research can test the robustness of the results by adding more covariates, such as female-specific fertility measures (e.g., age at menarche/menopause, use of oral contraception pills/hormone therapy), early life exposures, employment history, social interaction, genetic factors, and personality traits.

Conclusion

Despite these limitations, this study has several strengths. It is one of the first studies to use nationally representative, longitudinal data to examine the association between fertility history and risk of cognitive impairment among older men and women in the United States. This study advances recent work by Gemmill and Weiss (2022) using the same data set in several aspects, including the use of continuous fertility measures, the findings of U-shaped relationships between women’s age at last birth/parity and risk of cognitive impairment, and an emphasis on internal heterogeneity within those with atypical fertility. The findings also echo the life-course theory by highlighting the timing of life events and its long-term impact on population health. Given the changing fertility patterns and the increasing prevalence of cognitive impairment, understanding fertility history and its association with cognition over the life course will help identify the most vulnerable subpopulations, so that more effective interventions can be made to offset the risk.

Author Note

1.This finding was only found in the subdistribution hazard model.

Funding

This research was supported by the National Institute on Aging, grants R01AG061118 and RF1AG062765. The authors also gratefully acknowledge use of the facilities of the Center for Demography of Health and Aging at the University of Wisconsin–Madison, funded by the National Institute on Aging Center Grant P30 AG017266.

Author Contribution

The author conducted the data analysis and drafted the full manuscript.

Conflict of Interest

The author declares that there is no conflict of interest.