Research Note: The Association Between Parity and Odds of Alzheimer’s Disease and Dementias Status

Abstract

While parity is a significant factor influencing parental health, its relationship with dementia remains underexplored. This research note advances the literature by conducting a well-powered analysis of associations between parity (i.e., number of children) and Alzheimer’s disease and dementias (AD/D) status in large-scale population data. The data contain a large number of AD/D cases (37,228 women and 19,846 men), allowing a range (1–10) of parity associations to be estimated precisely. Using proxy (adult child’s) reports of parental AD/D status, we find that both fathers and mothers with grand multiparity have decreased odds of AD/D status, and the effect sizes become larger as parity increases, with 30–40% reduction in AD/D status at parities above 7. The association is stronger for mothers than for fathers. This finding differs from much of the prior literature and likely suggests the impact of parity, as one of the important life course contexts, on people’s cognitive function and risk of having AD/D. Finally, we include population projections that consider how large changes in parity distributions over time may contribute to small elevations in AD/D rates.

Introduction

Dementia risk becomes a public health challenge in aging societies. In the United Kingdom (UK), there were almost one million people living with dementia in 2021, and this figure is projected to rise to 1.2 million by 2040 (Alzheimer’s Research UK 2023). Women are disproportionately affected, comprising two thirds of the dementia population in the UK (Benham-Hermetz 2022). Despite scientific efforts to explore risk or protective factors that may cause the onset and progression of dementia, many upstream factors are not well-understood, especially sex-specific variables. Parity—number of live births—is an important life course context that can significantly shape parents’ well-being (Umberson et al. 2010; Zhang 2022) but its connection to dementia risk remains ambiguous, considering the inconsistent findings and complex underlying mechanisms in existing research. The UK, like many countries, is experiencing a decline in fertility rates alongside a rapidly aging population, raising questions about the impact of these demographic shifts on the health and well-being of older adults. Guided by a life course perspective, this research note examines the association between parity and dementia risk from a gendered lens and discusses the mechanisms that intertwine these factors in an effort to contribute to our understanding of this critical issue.

The life course perspective underscores the importance of contextualizing social determinants of health, and gender is a key factor in understanding health disparities, including cognitive health. Gender differences exist in both dementia risk and fertility patterns. Women have a higher prevalence of dementia than men, which is partly attributed to women’s longer life expectancy, yet other factors, particularly sex-specific factors, are not fully understood. Research has pointed to hormonal fluctuations, variations in the risk of depression, high blood pressure, and disparate rates of physical activity as potential contributors (Mielke 2018). Additionally, selective attrition in men due to early mortality from cardiovascular risk factors presents a competing risk scenario for death or dementia. Fertility and parental roles display gendered patterns as well, which may likely influence the association between parity and dementia risk. For example, parent–child relationship dynamics affect mothers and fathers differently (Reczek et al. 2014; Thomas and Umberson 2018). Greater relationship strain with adult children has been linked to “higher initial levels of cognitive limitations among mothers but appeared to be protective against increasing cognitive limitations for fathers as they aged” (Thomas and Umberson 2018:1133). Moreover, mothers’ mental health shows a stronger dependency on the quality of these relationships than that of fathers (Milkie et al. 2008). This observation aligns with the life course perspective that emphasizes “linked lives,” highlighting the interdependence between parents and children (Umberson et al. 2010). Parenthood is often described as a “mixed bag” with both stress and rewards (Greenfield and Marks 2006; Nomaguchi and Milkie 2020). The number of children could potentially diminish or amplify this effect, pushing the parity–dementia linkage toward a complicated story, especially considering a gender perspective.

The gendered life course perspective also guides us in understanding various mechanisms linking parity and dementia risk. Specifically, a growing number of studies, mostly using female samples, have examined the association between parity and parents’ cognitive function and prevalence of dementia (Bae et al. 2020; Beeri et al. 2009; Fox et al. 2018; Jang et al. 2018). This linkage may result from two possible mechanisms. Biologically speaking, women’s number of pregnancies is a key proxy for cumulative estrogen exposure. Estrogen exposure plays an important role in regulating neuronal biochemistry, which can account for women’s brain aging and cognitive ability (Brinton 2009; Karim et al. 2016; Ryan et al. 2009). Moreover, physiological changes caused by pregnancy can affect mothers’ comorbidities (e.g., diabetes, cardiovascular disease, depression), which have been examined as factors associated with dementia risk in later life (Ahtiluoto et al. 2010; Deckers et al. 2017; Hanson et al. 2015). Socially speaking, childbearing can affect multiple dimensions of parents’ lives, such as their financial situation, labor force participation, marital stability, and social integration, which have significant implications on cognitive health for both fathers and mothers (Read and Grundy 2017; Umberson et al. 2010). In later life, adult children become pivotal in their parents’ social networks, offering economic and emotional support and often assuming the role of primary caregivers. Their involvement, including social stimulation and monitoring health behaviors, is a critical factor in positively influencing older parents’ cognitive functions (Zhang and Fletcher 2021).

Overall, the literature shows an inconclusive association between parity and dementia risk. Specifically, a sizable number of studies, most in clinical research, suggest that grand multiparity is detrimental to mothers’ cognitive function and is associated with a higher risk of dementia, early onset of Alzheimer’s disease (AD), and greater AD neuropathology (Beeri et al. 2009; Colucci et al. 2006; Jang et al. 2018; Jung et al. 2020). For example, Colucci et al. (2006) found that women with more than three pregnancies had three times the risk of developing AD and experienced AD onset four year earlier than women with three or fewer pregnancies. Moreover, several population-based studies found a U-shaped relationship or a J curve between parity and cognitive impairment in later life (most used global cognition test batteries), suggesting that either lower parity or grand multiparity may increase parents’ risk of dementia or cognitive impairment compared with a medium parity (Gong et al. 2022; Read and Grundy 2017; Saenz et al. 2021; Zhang 2022). For example, Gong et al., using UK Biobank data, found that compared with women with two children, the hazard ratio of all-cause dementia was 1.18 for women without children and 1.14 for women with four or more children, indicating a U-shaped relation between parity and dementia risk. In contrast, a small group of studies found beneficial effects of parity that may reduce parents’ dementia risk. For example, in a British study, Fox et al. (2018) reported that women (aged 70–100) with more cumulative months pregnant exhibited reduced AD risk because of the long-term neuroprotective effects of estrogen exposure. Also, several studies found that high parity was associated with better memory ability for women and younger brain age for both parents (de Lange et al. 2019; Henderson et al. 2003; Ning et al. 2020). Last, other studies reported a null association between parity and incidence of dementia (Bae et al. 2020; Gemmill and Weiss 2022; Najar et al. 2020; Ptok et al. 2002).

The current literature on this topic displays inevitable limitations because of the small number of Alzheimer’s disease/dementia (AD/D) cases, misidentification of dementia status, and the use of selective samples. Specifically, in clinical research and communitybased cohort studies, measurements of dementia status are relatively accurate but the numbers of cases are very small (often fewer than 100), which may reduce the power of analysis and render findings ungeneralizable. In national, population-based studies, the numbers of AD/D cases are relatively larger (e.g., approximately 2,000 incidents of dementia in UK Biobank (Gong et al. 2022) and 3,000 labeled dementia cases in the Health and Retirement Study (Gemmill and Weiss 2022)). Yet, dementia status in survey data is less accurately identified, and researchers often rely on cutoff points from global cognitive scores (i.e., cognition test battery in various domains) rather than clinical assessment. Still, the prevalence of dementia is often small for subgroup analysis and the use of the same cutoff points for different groups in terms of, for example, age, race, and education is questionable (Hudomiet et al. 2018). As an example, some research indicates that the accuracy of dementia assessments—often through self-report or performance-based battery—was greater among participants who were younger, higher educated, or non-Hispanic White compared with their respective counterparts (Gianattasio et al. 2019). Additionally, respondents who complete cognitive performance tests are relatively selective in many population-based studies, which yields measurement errors in cognition reporting. Participants who can complete the tests are relatively young and may not experience the onset of dementia during the study observation. For example, research using the Health and Retirement Study data found that self-reported dementia diagnosis underestimates the number of people living with a cognitive impairment consistent with dementia (McGrath et al. 2021). Moreover, because of attrition, people who are deceased or lost to follow-up are unable to provide their dementia status, which likely leads to more conservative estimates. Finally, as a result of modest analytic sample sizes, much of the literature uses categorical measures of parity that collapse grand multiparity into a single group (>4–5 kids), and doing so neglects heterogeneity among parents with multiparity.

In the present study, we address some of these limitations by using proxy reporting from the adult children of AD/D cases from a large-scale, population-based study. Such data can advance research in several aspects. First, the large number of AD/D cases can significantly improve the power of analysis; our sample contains 37,228 AD/D female cases and 19,846 male cases. Notably, in the aforementioned studies, the largest number of cases was fewer than 3,000 for each gender, while our sample size has approximately 10 times the number of cases. Moreover, the large sample size also enables us to estimate separate effects across a full range of parity (1–10) so that we can conduct a closer investigation of each category of parity. Also, large sample size helps improve the precision in estimating the prevalence of dementia. This is particularly significant given the challenges posed by sample attrition in longitudinal data sets, as discussed in sources such as Gemmill and Weiss (2022). Second, the mean age of respondents in our sample is about 56 years, and more than half of them experienced a parent’s death. This suggests that proxy reports of parents’ AD/D cases are very likely capturing the parents’ entire life span, which can reduce selection bias and misidentification of dementia status. Last, the large sample size and genetic information provided by UK Biobank enable us to link kinship information and find genetically related individuals (e.g., cousins) who report their parents’ AD status, allowing us to compare related individuals in the parental generation. Using “cousin” fixed-effects models, we can control for a portion of the confounding effects of parents’ environmental and genetic backgrounds. In short, this study provides new measures of AD/D status across a full range of parity reported by a large number of proxies to examine the association between AD/D status and parity.

Methods

Data and Measures

We use data from the UK Biobank (UKB) project. UK Biobank is a large-scale database containing in-depth health, genetic, and family information from half a million UK participants. The participants, aged 37–74, were originally recruited to complete touch screen questionnaires at UKB assessment centres between 2006 and 2010. Our raw sample includes 502,505 respondents. We exclude respondents who have missingness in analytic variables or report discordant cases in parent’s AD/D cases and parity. Specifically, using genetic and kinship information, we identify “co-siblings” in the sample.¹ Because co-siblings share the same parents, their reports of parents’ AD/D status and parity should be the same. Therefore, we drop the discordant reports among co-siblings to reduce measurement errors. The final sample sizes without missingness and discordant reports are 415,304 for the mothers sample (16.52% of cases were dropped for missingness and 0.99% for discordant reports) and 387,254 for the fathers sample (22.26% were dropped for missingness and 0.84% for discordant reports). Sample sizes represent the number of respondents (i.e., children) who reported valid data of parents’ AD/D status and other analytic variables. Hereafter, children = respondents, and the number of children = parity.

Outcome: Parent’s Alzheimer’s disease or dementia.

The respondents were asked, “Has/did your father/mother ever suffer from [choices of diseases]?” One of the options was “Alzheimer’s disease/dementia.” Respondents were asked to select only the diseases that they were sure about and only for blood relations. The outcome was coded as a binary variable, with 1 = yes and 0 = no.

Exposure: Parent’s parity.

The participants were asked how many sisters/brothers they have. We calculated the number of siblings by summing the number of sisters and brothers, then adding one (the respondent) to yield the total number of children that the respondent’s parents had. Having more than 10 children was coded as 10.

Covariates: Respondent’s age and parent’s alive status.

The first covariate is straightforward, and the second was obtained from the question, “Is your father/mother still alive?” Options for the latter were coded as 1 = yes and 0 = no.

Analysis

We used linear models with many levels of fixed effects (absorbing the fixed effects of respondent’s age) to examine the association between parent’s parity and AD/D status. We visualize the relationships in graphs using margins results (i.e., predicted probability of AD/D by parity). We also report marginal effects and odds ratios from logistic regression models to facilitate the interpretation. We present descriptive statistics and model results by parents’ gender. UK Biobank does not provide much parental demographic information. To control for confounding effects of genetic factors and shared family backgrounds of the parental generation, we created a “cousin” sample using kinship information and conducted cousin fixed-effects models. A further description of the cousin sample and model results are provided in the online appendix, section A.

Results

Table 1 displays descriptive statistics. The average age of respondents/children in both samples is around 56 years old. There are 37,228 respondents who reported that their mothers had ever suffered from AD/D, representing 8.96% of the sample. Fewer respondents reported that their fathers had AD/D (19,846, or 5.12% of the sample). In both samples, more than half of the respondents reported that their parents had two or three children (i.e., the respondents have one or two siblings). Although the numbers decrease as the parity increases, the high-parity groups (7 or more children) are quite sizable (e.g., more than 2,000 respondents’ parents had 10 children). Moreover, 42.05% of respondents’ mothers and 25.08% of respondents’ fathers were alive at the time of the baseline interview.

Table 1.

Descriptive statistics by parent’s gender, full sample

	Mother (N = 415,304)		Father (N = 387,254)
	Frequency	% or Mean (SD)	Frequency	% or Mean (SD)
Age	415,304	56.48 (8.13)	387,254	56.44 (8.12)
AD/D Cases	37,228	8.96	19,846	5.12
Parity
1	50,628	12.19	44,545	11.50
2	135,429	32.61	127,756	32.99
3	103,543	24.93	97,599	25.20
4	56,807	13.68	53,404	13.79
5	29,543	7.11	27,535	7.11
6	16,466	3.96	15,278	3.95
7	10,088	2.43	9,349	2.41
8	6,337	1.53	5,794	1.50
9	3,937	0.95	3,664	0.95
10	2,526	0.61	2,330	0.60
Parent’s Still Alive (ref. = no)	174,643	42.05	97,129	25.08

Note: AD/D = Alzheimer’s disease/dementia.

Table 2 presents results from the linear and logistic regression models while absorbing age fixed effects (which broadly controls for parental birth cohort) and controlling for parents’ mortality status. We see that the marginal effects are very similar between these two sets of models (differences in coefficients are no greater than 0.003). For the mothers’ results, we find a negative association between parity and the odds of AD/D, suggesting that the AD/D odds decrease as parity increases. Moreover, as parity rises, the negative association between parity and AD/D becomes persistently larger. Specifically, compared with mothers with only one child (the base group), those who had two children did not show a significant difference in AD/D odds, but mothers with three or more children showed significantly lower odds of AD/D. The effect size is −0.006 among mothers with three children, but the coefficient is 6 times as large among those with 10 kids (−0.038). To facilitate the interpretation, the odds ratios from the logistic model suggest that mothers with three children had a 6% lower odds of AD/D ([1 – 0.940] * 100, p < .001) whereas those with 10 children had a 40% lower odds ([1 – 0.603] * 100, p < .001). Using the linear model results, the top panel of Figure 1 displays the consistently declining odds of AD/D as parity increases (the y-axis shows the predicted probability of AD/D).

Table 2.

Age fixed-effects models from linear models and logistic models (with total Ns) by parents’ gender

	Mother			Father
	Linear (N = 415,304)	Logit (N = 415,297)		Linear (N = 387,254)	Logit (N = 387,245)
	Marginal Effect	Marginal Effect	Odds Ratio	Marginal Effect	Marginal Effect	Odds Ratio
Parity (ref. =1)
2	−0.002 (0.001)	−0.001 (0.001)	0.985 (0.017)	0.001 (0.001)	0.001 (0.001)	1.020 (0.025)
3	−0.006*** (0.002)	−0.005** (0.002)	0.940*** (0.017)	0.001 (0.001)	0.001 (0.001)	1.025 (0.026)
4	−0.009*** (0.002)	−0.009*** (0.002)	0.901*** (0.019)	−0.001 (0.001)	−0.001 (0.001)	0.981 (0.028)
5	−0.017*** (0.002)	−0.016*** (0.002)	0.812*** (0.021)	−0.004* (0.002)	−0.004* (0.002)	0.914* (0.032)
6	−0.020*** (0.003)	−0.020*** (0.002)	0.775*** (0.026)	−0.007** (0.002)	−0.007** (0.002)	0.867** (0.038)
7	−0.022*** (0.003)	−0.021*** (0.003)	0.762*** (0.031)	−0.008** (0.003)	−0.008** (0.002)	0.848** (0.046)
8	−0.028*** (0.004)	−0.027*** (0.003)	0.696*** (0.036)	−0.007* (0.003)	−0.006* (0.003)	0.871* (0.058)
9	−0.031*** (0.005)	−0.029*** (0.004)	0.666*** (0.043)	−0.014*** (0.004)	−0.014*** (0.003)	0.720*** (0.064)
10	−0.038*** (0.006)	−0.035*** (0.005)	0.603*** (0.050)	−0.014** (0.005)	−0.014** (0.004)	0.723** (0.079)
Parent’s Still Alive (ref. = no)	−0.018*** (0.001)	−0.018*** (0.001)	0.793*** (0.011)	−0.004*** (0.001)	−0.004*** (0.001)	0.915*** (0.019)
R 2	.019	—		.004	—
AIC	129,909.7	241,342.2		−73,493.49	154,898.9
BIC	130,030.0	241,801.5		−73,373.95	155,344.5

Notes: Standard errors are shown in parentheses. Several age categories in logistic models were omitted because of small sample size (n < 5) causing perfect prediction. The logistic model controls for child’s age but is not shown. AIC = Akaike information criterion. BIC = Bayesian information criterion.

^*p < .05;

^**p < .01;

^***p < .001

Fig. 1.

Association between number of children and parent’s odds of having Alzheimer’s disease or dementias (AD/D). Results are linear predictions based on the results from Model 1 in Table 2.

For fathers, Table 2 also shows a negative association between parity and the odds of AD/D, with the odds becoming smaller as parity increases, but there are two differences compared with mothers’ results. First, the effect sizes are much smaller for the fathers’ results. Second, there is no significant difference between having one child and medium parities (i.e., 2–4 children). The negative association becomes statistically significant only for fathers with five or more children. Odds ratios suggest that, compared with fathers with one child, those with 10 children had a 28% lower odds of AD/D ([1 – 0.723] * 100, p < .001). The bottom panel of Figure 1 displays the declining trend of AD/D odds over parity.

Sensitivity Tests and Projection of AD/D Rate by Cohorts

First, robustness tests using a “co-cousins” sample (see the online appendix, section A, for further description) suggest that the negative association between parity and AD/D remains for mothers but is mostly nonsignificant for fathers (see Model 1 in Table A2 in the online appendix). Noticeably, the negative association between grand multiparity and AD/D odds persists for both fathers and mothers, although it is nonsignificant. The analysis suggests that there is some scope for genetic and environmental confounding of the relationship between parity and AD/D status, but we also continue to observe important associations with parity. In unreported results, we also controlled for place of birth fixed effects to examine whether geographic differences in parity or AD surveillance might affect our results, but we found no changes in the main effects.

Second, the participants in the UKB were not randomly selected, indicating a “volunteer bias” (van Alten et al. 2022). According to van Alten et al. (2022), individuals who participated in the UKB tend to be older, female, and of higher socioeconomic status. To reduce the selection bias associated with volunteering, we adopted van Alten et al.’s method, generated inverse probability weighting, and used it in our analyses. Table B1 in the online appendix suggests robust results.

Finally, we consider the combined effects on AD/D risk of shifts in the parity distribution over the birth cohorts represented in our sample. Our parental sample is from older cohorts (i.e., those with children born between 1940 and 1970), in which high parity is more common than in more recent cohorts. To project the impacts of the changing parity distribution on AD/D rates of more recent cohorts, we applied the estimated coefficients/odds ratios from the logistic regression models in Table 2 to calculate the AD/D rates of the respondents’ cohorts (i.e., those born between 1940 and 1970) in UKB data and also in a larger set of cohorts with UK and U.S. census data (see the online appendix, section C). We set the AD/D rates of the childless group (who are not included in our UKB analysis) within an estimated range based on prior population-based evidence that shows that childless older adults may have approximately a 10–20% higher risk of cognitive impairment (Sundström et al. 2016; Zhang and Fletcher 2021). Using the UKB respondents’ cohort, we calculated comparisons of the AD/D rates between the younger and older cohorts. The key finding is a 3–4% increase in population-level AD/D rates as fertility declined in younger cohorts; our UK Census analysis shows a 1–2% increase between the 1940 and 1970 birth cohorts. Using U.S. census data, we calculated a 5–6% increase in predicted population-level AD/D rates as fertility declined in younger cohorts, reflecting the larger shifts in parity that have occurred in the United States between 1940 and 1970 than those in the UK.

Conclusion and Limitations

In this study, we found negative associations between parity and the odds of AD/D for both fathers and mothers, and the associations are stronger for mothers. Our study advances the literature by using a large enough number of AD/D cases to statistically detect heterogeneity by parity status. The findings may resonate with some prior studies that found neuroprotective effects of estrogen exposure on mothers’ cognitive function (e.g., Fox et al. 2018) and may also speak to life course studies that suggest social pathways linking parenting/intergenerational relationships and parental well-being through social interaction and caregiving (Keenan and Grundy 2019; Reczek and Zhang 2016; Umberson et al. 2010; Zhang 2022). Our prediction across different cohorts indicates a potential increase in dementia risk associated with declining fertility rates in recent cohorts. Among the potential mechanisms driving this trend, a notable consideration for public policy is the increasing unmet care needs of older adults, which may be exacerbated by shrinking family sizes. However, a challenge in current research is the difficulty of studying the association between fertility rates and dementia risk in more recent cohorts, as they are often too young for dementia onset to be tracked in survey data. Our findings indicate that an association between higher parity and lower dementia risk, observed in older cohorts (aged 75+), does not appear in recent cohorts. For example, analysis of the Health and Retirement Study family data (respondents aged 50+) presents mixed outcomes: some studies found no significant association (Gemmill and Weiss 2022), while others identified a negative impact of high parity on cognitive functions (Read and Grundy 2017; Saenz et al. 2021). In short, our study highlights that parity is an important life context that may correlate with parents’ exposure to dementia status and can also help in understanding sex-specific pathways that underlie the large sex differences in AD/D. Also, we explored whether the impact of these trends may vary across different cohorts, and we advocate for increased public policy focus on the long-term effects of declining fertility on caregiving and health outcomes in later life.

While use of the UK Biobank has several important advantages over earlier work, one major limitation is the lack of information about the sociodemographic characteristics of the parents. Therefore, we are unable to test other specific mechanisms or confounders hypothesized in the literature, such as the quality of parent–child relationships, parents’ education levels, and sex hormones. Future work is needed to examine the mechanisms and how the linkage between parity and dementia risk varies by subpopulations. Additionally, use of the UK Biobank, which operates on a volunteer-based model, inherently presents challenges in participant representativeness. Therefore, it is reasonable to guess that this selective participation might influence the representation of reporting parental dementia status. Moreover, our study, which centers on children reporting their parents’ dementia, excludes childless individuals, thus limiting our understanding of dementia risk differences between parents and nonparents. Finally, there is considerable missingness in parents’ AD/D status (>15% of the raw sample). Additional tests suggest that parents who are alive and have fewer children are less likely to have missing data for AD/D status. However, respondents (i.e., children) who are male, younger, of non-European ancestry, with less education, and with less income are more likely to have missing data in parental AD/D status. We also view the missingness as having a complex set of sources; for example, children may not know the dementia status of each parent, even if the parent is old enough to be classified as having or not having dementia, or the parent may have died relatively early and therefore is not eligible to have dementia as an older person. Other reasons are children may not be in contact with their parents, or the child may have failed to respond to all questions about parental health status. Although the reweighting results provided in Table B1 in the online appendix help, future research can provide more evidence to demonstrate the robustness of the results. ■

Supplementary Material

ELECTRONIC SUPPLEMENTARY MATERIAL The online version of this article (https://doi.org/10.1215/00703370-11585876) contains supplementary material.