Social integration, marital status, and ovarian cancer risk: A 20-year prospective cohort study

Abstract

Objectives:

Low social integration and divorce/widowhood are chronic psychosocial stressors that may affect health. When assessed after cancer diagnosis, they have been associated with poorer survival, but their role in cancer development, particularly ovarian cancer (OvCA), is less understood. We investigated whether social integration and marital status were related to OvCA risk in a large population-based study.

Methods:

Women from the Nurses’ Health Study completed the Berkman-Syme Social Network Index and reported their marital status every 4 years starting in 1992 (N=72,206), and were followed up until 2012 (20-year follow-up period). Multivariate Cox regression models estimated hazard ratios (HR) and 95% confidence intervals (CI) of OvCA risk, considering relevant potential confounders, in lagged analyses whereby psychosocial indicators were assessed 4–8 (n=436 cases) and 8–12 years (n=306 cases) before diagnosis to account for the effects of pre-diagnostic symptoms on social measures. Secondary analyses evaluated stability of and cumulative exposure to these social factors on OvCA risk.

Results:

Being socially isolated versus integrated was related to an increased OvCA risk 8–12 years later (HR=1.51, 95%CI=1.07–2.13), but not 4–8 years later. Compared to married women, OvCA risk was significantly higher in widowed but not in separated/divorced individuals, with both time periods (e.g., 8–12 years later: HR_widowed=1.57, 95%CI=1.15–2.14 versus HR_{separated/divorced}=1.13, 95%CI=0.74–1.72). Estimates were comparable or stronger when investigating stability in and cumulative effects of social indicators.

Conclusions:

Results suggest higher OvCA risk among socially isolated and widowed women, particularly when such psychosocial stressors were experienced a decade before diagnosis or were sustained over time.

In 2019, 22,530 new ovarian cancer (OvCA) cases are projected among U.S. women (1), a disease that is disproportionally deadly compared to other female cancers (2). Although some risk factors have been identified, including family history of breast and OvCA, and reproductive factors (1, 3), effective strategies for primary prevention are limited. Hence, establishing additional modifiable determinants that may affect OvCA development could reduce burden from this deadly disease.

Accumulating evidence shows that the social environment plays an important role in health-related outcomes, including cancer risk (4–6). While multiple conceptualizations of social support exist, its structural and functional dimensions are well-characterized (4, 5, 7). The former includes aspects like the number and closeness of relatives and friends in a person’s network, and the frequency of his/her interaction with this network through varied activities, while the latter refers to emotional bonds with and resources/information provided by others (4, 5, 8). Individuals who are more socially integrated, which reflects greater structural support, generally have lower risk of cancer progression and mortality, whereas the functional support has had weaker associations (6, 9–13). Given its particular importance as a type of social tie, marital status has also been studied extensively as an independent form of structural support that may predict health, including cancer-related outcomes (14–17).

Most investigations about structural dimensions of social support have examined breast cancer (7). While less is known about OvCA, preliminary findings are informative. A recent case-control study of 1,180 Swedish women (239 cases/941 controls) suggested that, after adjusting for sociodemographics and health-related behaviors, the number of friends/relatives available for socializing was unrelated to OvCA onset over a 20-year period (18). However, other structural dimensions and established OvCA determinants were unavailable in this study, limiting the interpretation. Interestingly, animal models demonstrated that mice who were socially isolated (housed individually) had larger and more aggressive OvCA tumors compared to control animals housed together (19, 20). Early studies also reported higher OvCA risk among unmarried versus married women, although models were only age-adjusted (21, 22). Recent prospective research has controlled for key confounders but has mainly considered the association of marital status with OvCA survival, rather than incidence. Notably, several population-based studies revealed greater early mortality risk in unmarried versus married OvCA patients (16, 23, 24), particularly for widowed women, who had lower 5-year survival rates than separated/divorced women (23, 24).

Multiple mechanisms have been postulated to explain associations between social factors and cancer-related outcomes (7). Socially integrated individuals are more likely to be influenced positively by their peers, have access to resources, and consequently, adopt healthier behaviors (e.g., smoking cessation, medical adherence); they are also less likely to be depressed (4, 7, 13, 25). Social isolation may alter biological processes too, via stress-related sympathetic nervous system activation and norepinephrine release (7, 26, 27), which can directly influence tumor progression through processes of cellular adhesion, cell migration, invasion, and angiogenesis (28–30); isolation may also reduce natural killer cell activity (30, 31) and suppress wound healing (32). Similar pathways are hypothesized for marital status (33–35). Because social isolation and transition out of marriage (e.g., divorce, widowhood) are conceptualized as “chronic” rather than “acute” stressors, it is posited that they have a cumulative, long-term toll on physical health via a sustained impact on these biobehavioral mechanisms (4, 12, 36).

Altogether, despite recent advances in our understanding of whether social integration and marital status influence cancer progression, it remains unclear whether these markers of structural social support are prospectively associated with OvCA onset specifically. Thus, we conducted a prospective study of social integration and marital status with OvCA risk in a large sample of women participating in the Nurses’ Health Study (NHS), which has detailed information about multiple established OvCA risk factors and an excellent characterization of OvCA cases. Importantly, two measures of structural social support –the Berkman-Syme Social Network Index (SNI) and marital status– were asked every 4 years over 20 years of follow-up, allowing consideration of chronic, sustained phenomena like social isolation and marital dissolution. We hypothesized that being socially isolated versus integrated, and being separated/divorced or widowed versus married would be associated with higher OvCA risk.

METHODS

Participants

The ongoing NHS cohort enrolled 121,700 female nurses in 1976 (ages 30–55 years) (37). The initial goal was to examine the relationship of oral contraceptives and breast cancer risk; because oral contraceptives were only prescribed to married women at this time, being married was an eligibility criterion for study entry (38). Women completed biennial questionnaires on sociodemographics, lifestyle, medical history and newly diagnosed medical conditions, with a response rate of >85% over decades of follow-up (39). The present study included initially cancer-free (except non-melanoma skin cancer) women with at least one ovary (and no radiation-induced menopause) who responded to at least one SNI assessment between 1992 and 2008 (N=72,206). Because of the 1976 eligibility criteria, in the current study most participants were still married/partnered in 1992 when the SNI was first queried. The study protocol was approved by the institutional review boards of the Brigham and Women’s Hospital, Harvard T.H. Chan School of Public Health, and those of participating registries as required; return of the baseline questionnaire implied consent by participants.

Measures

Social Integration.

Social networks were assessed every four years starting in 1992 with the Berkman-Syme Social Network Index (SNI) (8), a multidimensional measure that has been widely used in prior research (5). This index assesses 4 distinct dimensions of social networks: marital status (married/partnered, separated/divorced, widowed); number of close relatives and close friends, separately (0, 1–2, 3–5, 6–9, ≥10); frequency of religious activities (>1/week, 1/week, 1/month-1/year, never); and frequency of activities with community organizations (≥11 hours/week, 6–10 hours/week, 3–5 hours/week, 1–2 hours/week, no community activities). SNI responses were categorized into 4 levels (40, 41): socially isolated (individuals with low contacts—none or one of the following characteristics: married, had more than six close friends or relatives, attended weekly religious group activities, or attended weekly community activities), moderately isolated (two characteristics), moderately integrated (three characteristics), and socially integrated (all four characteristics; reference group). In sensitivity analyses (described below), a modified SNI excluding the marital status was used: socially isolated (no characteristic), moderately isolated (one characteristic), moderately integrated (two characteristics), and socially integrated (three characteristics). In this sample, SNI scores were fairly stable over study duration (intra-class correlations coefficient [ICC]=0.68, 95% confidence interval=0.68–0.69).

Marital Status.

The SNI item querying marital status was also considered separately given its prior associations with cancer outcomes (14–17), for which the response options were Married/Partnered, Separated/Divorced, and Widowed. Because women all were married at the NHS cohort inception (1976), none were categorized as “single” over the follow-up period.

Covariates.

Following epidemiological evidence (3, 42, 43), covariates included established OvCA risk factors: age (continuous), oral contraceptive use duration (OC; continuous), parity (continuous), tubal ligation (no/yes), familial history of breast or ovarian cancer (Br/OvCA; no/yes), menopausal status (premenopausal or unknown/postmenopausal), hormone therapy use duration (continuous); and health-related behaviors with putative relationships with OvCA overall or with specific subtypes: body mass index (BMI; kg/m²; continuous) and smoking status (never/former/current). We also considered physical exam in the last 2–4 years (no/yes) to capture medical adherence and health care utilization. In sensitivity analyses, we adjusted for a composite binary score of depression that was created using a Boolean OR operator approach (44) based on self-reported depressive symptoms (45), physician-diagnosed depression, and antidepressant use. Covariates were self-reported at or prior to the analytic study baseline (1992) and were updated every 2–4 years until the end of follow-up.

Ascertainment of Cases.

Incident OvCA cases were identified among women who were diagnosed after return of the 1992 questionnaire to June 2012. Women who reported an OvCA diagnosis were asked permission to obtain and review their pathology reports and medical records. A gynecologic pathologist blinded to exposure status reviewed each woman’s records to confirm the diagnosis and to assess tumor characteristics (e.g., histology). Reviews of pathology records were highly concordant with surgical pathology slides (46). For cases on whom pathology reports were unavailable, appropriate cancer registries were used to confirm the diagnoses.

Statistical analyses

All analyses were conducted using SAS® 9.4 with a two-sided 0.05 p-value. Covariate distributions across baseline categories of social integration and marital status were assessed after age standardization.

Main Analyses.

Cox proportional hazards regression models assessed the hazard ratios (HR) and 95% confidence intervals (CI) of incident OvCA, until end of follow-up (2012), diagnosis of another cancer (except non-melanoma skin cancer), bilateral oophorectomy, or menopause due to radiation, whichever came first. Our primary statistical models examined the associations of social exposures with OvCA risk by incorporating latency periods of 4–8 and 8–12 years. For instance, SNI scores and marital status queried in 1992 were evaluated with OvCA incidence in 1996–2000, and in 2000–2004, respectively. Because evidence suggests the window of OvCA development averages between 7–9 years and may take up to 11–12 years (47, 48), window of 8–12 years before diagnosis examined exposures that may affect early progression of precursor lesions, whereas 4–8 years captured a key transformation period to overtly invasive disease. The association was tested in three models to evaluate the role of covariates. A minimally-adjusted model stratified by age and time period; a second model additionally adjusted for reproductive and hormonal-related variables (OC use, parity, tubal ligation, menopausal status, hormone therapy use) and family history of Br/OvCA. A third fully-adjusted model further included relevant behavior-related covariates (BMI, smoking status, prior physical exam), to distinguish potential pathways. To determine whether the relationship of social isolation with OvCA risk was monotonic (i.e., OvCA risk increases as isolation levels increase), linear p-trends were calculated in fully-adjusted models. No p-trends were calculated for marital status as it is categorical by nature. Most covariates had no missing data in our main analytic sample. Because the others had a very low proportion (≤0.2%; parity, duration of OC use, BMI, smoking, physical exam), the median value from the time assessment at which the data was missing was imputed.

Sensitivity Analyses.

Three sets of sensitivity analyses tested the robustness of these associations, using one or both latency periods (4–8 or 8–12 years) to assure sufficient statistical power in each category of exposure. First, analyses explored potential differences due to histology, as serous and non-serous tumors have both common and distinct risk factors (49, 50). Given prior research revealing an association of depressive symptoms with OvCA incidence (51) and marital status (36), sensitivity analyses further adjusted for depression in women for whom the composite described above was available (1% missing). Lastly, two distinct a posteriori analyses evaluated the association of OvCA risk with: 1) a modified SNI excluding marital status, with and without adjustment for marital status, to determine whether results obtained with the SNI were not mainly due to marital status; and 2) living arrangement categories by marital status, to disentangle whether widowhood results might instead capture functional aspects of living alone (e.g., fewer resources).

Secondary Analyses.

To consider the impact of sustained social isolation and widowhood or divorced/separated status, we also conducted two secondary analyses. First, stability was assessed by creating categories of change in the exposure, using latency periods of 4–8 and 8–12 years. These analyses were conducted in subsets of women who had completed two consecutive assessments of the exposures (e.g., changes from 1996 to 2000 in relation to risk in 2004–2008). Specifically, the SNI categories were: Remained integrated (integrated at both assessments; reference group), Became integrated (went from moderately/socially isolated to moderately/socially integrated), Became isolated (went from moderately/socially integrated to moderately/socially isolated), and Remained isolated (isolated at both assessments). The marital status categories were: Remained married/partnered (married/partnered at both assessments; reference group), Became widowed (went from married/partnered to widowed) and Remained widowed (widowed at both assessments). There were too few women entering marriage after being widowed, or experiencing separation/divorce to be included in these analyses. Second, we evaluated chronicity of exposure by evaluating 1) a cumulative SNI score averaging all available prior SNI assessments, and 2) categories representing the number of assessments for which women reported being separated/divorced or widowed (i.e., who did not re-enter marriage). To assure enough statistical power in each exposure category, these analyses were conducted with the 4–8 years lag only. Lastly, unadjusted Kaplan–Meier curves were used to depict the relation of social exposures at baseline with OvCA risk throughout the 20-year follow-up period.

RESULTS

Baseline Characteristics

At the 1992 baseline (Table 1), socially integrated women were more likely to participate in healthy behaviors (e.g., never smokers, recent physical exam) than the socially isolated women; similar patterns were observed by marital status. Otherwise, no major differences across levels of social integration and marital status categories were evident for sociodemographics, reproductive/hormonal-related and family history of OvCA, as well as other behavior-related variables.

Table 1.

Age-standardized characteristics of NHS participants in 1992, across categories of social integration and marital status^a

	Socially isolated (n=7,236)	Moderately isolated (n=14,715)	Moderately integrated (n=18,283)	Socially integrated (n=13,605)		Widowed (n=5,327)	Divorced/ separated (n=4,466)	Married/ partnered (n=44,046)
Age*	57.7 (7.1)	57.9 (7.2)	58.5 (7.2)	58.7 (7.1)		63.6 (5.8)	55.6 (6.4)	57.9 (7.1)
Duration of OC use	2.3 (3.6)	2.2 (3.5)	2.0 (3.4)	1.9 (3.3)		2.0 (3.3)	2.4 (3.6)	2.0 (3.4)
Ever parous, %	93.9	93.9	94.7	95.2		92.6	94.0	94.8
Parity among parous women	2.8 (1.5)	2.9 (1.6)	3.0 (1.7)	3.2 (1.7)		2.9 (1.7)	2.9 (1.6)	3.0 (1.6)
Tubal ligation, %	18.9	18.7	16.4	16.4		15.0	19.6	17.3
Post-menopausal, %	72.5	72.2	71.9	72.6		73.1	72.2	72.2
Family history of Br/OvCA, %	14.3	14.1	14.4	14.7		14.7	14.2	14.4
Recent physical exam, %	83.9	87.1	88.1	89.1		86.2	87.4	87.6
BMI, kg/m2	26.0 (5.3)	26.0 (5.1)	26.0 (4.9)	26.0 (4.9)		26.5 (5.4)	25.9 (5.3)	26.0 (5.0)
Never smoker, %	33.2	38.3	44.9	54.1		38.6	35.5	45.2
Former smoker, %	42.4	43.8	42.1	38.1		40.2	41.4	41.7
Current smoker, %	24.4	18.0	13.1	7.8		21.1	23.1	13.1

Notes. Values are means (SD) or percentages and are standardized to the age distribution of the study population.

^*Not age-adjusted.

BMI=body mass index; Br/OvCA=breast/ovarian cancer; OC=oral contraceptive.

^aWhile social integration and marital status exposures are presented in the same table, they represent distinct analyses.

OvCA Risk by Latency Periods

In the analysis with a 4–8 years latency, 436 women developed OvCA over the follow-up; in the 8–12 years latency analysis, there were 306 new OvCA cases. When assessing exposure occurring 8–12 years before diagnosis (Table 2), compared to women who were socially integrated, those who were socially isolated had a significantly greater OvCA risk (HR_{fully-adjusted}=1.51, CI=1.07–2.13; p-trend=0.15), while the association was not statistically significant when the latency period was 4–8 years (HR_{fully-adjusted}=1.19, CI=0.87–1.62; p-trend=0.22). Moderate isolation/integration was unrelated to OvCA risk. However, the association for widowhood was statistically significant in both time periods, with slightly stronger estimates 8–12 years versus 4–8 years before diagnosis (HR_{fully-adjusted}=1.57, CI=1.15–2.14 versus HR_{fully-adjusted}=1.50, CI=1.17–1.92). Being divorced/separated was unrelated to risk; yet, number of incident cases were limited in this category (n_{4–8 years}=36; n_{8–12 years}=25). Unadjusted Kaplan-Meier curves considering social exposures at baseline with OvCA risk over the entire follow-up (no specific latency period) show comparable associations, although the distinction across social integration categories was somewhat less evident (Supplemental Figures 1 and 2, Supplemental Digital Content).

Table 2.

Associations of social integration and marital status with ovarian cancer risk by timing of exposure in relation to diagnosis.

	4–8 years before diagnosis				8–12 years before diagnosis
Exposure	N (cases)	Model 1 HR (95% CI)	Model 2 HR (95% CI)	Model 3 HR (95% CI)	N (cases)	Model 1 HR (95% CI)	Model 2 HR (95% CI)	Model 3 HR (95% CI)
Berkman-Syme Social Network Indexa	65,152 (436)				57,942 (306)
Socially integrated (ref.)	(67)	1.00	1.00	1.00	(59)	1.00	1.00	1.00
Moderately integrated	(118)	0.98 (0.76–1.25)	0.99 (0.77–1.27)	0.99 (0.77–1.27)	(66)	0.99 (0.73–1.33)	1.00 (0.74–1.35)	1.01 (0.75–1.36)
Moderately isolated	(138)	1.09 (0.84–1.41)	1.09 (0.84–1.41)	1.09 (0.84–1.42)	(99)	0.84 (0.61–1.16)	0.85 (0.61–1.17)	0.85 (0.61–1.19)
Socially isolated	(113)	1.17 (0.86–1.58)	1.18 (0.87–1.60)	1.19 (0.87–1.62)	(82)	1.45 (1.03–2.03)	1.46 (1.04–2.06)	1.51 (1.07–2.13)
p-trend				0.22				0.15
Marital status	65,152 (436)				57,942 (306)
Married/partnered (ref.)	(308)	1.00	1.00	1.00	(225)	1.00	1.00	1.00
Divorced/separated	(36)	1.15 (0.81–1.63)	1.16 (0.82–1.65)	1.17 (0.82–1.65)	(25)	1.12 (0.74–1.70)	1.12 (0.74–1.70)	1.13 (0.74–1.72)
Widowed	(92)	1.46 (1.14–1.87)	1.50 (1.17–1.92)	1.50 (1.17–1.92)	(56)	1.49 (1.09–2.03)	1.54 (1.13–2.10)	1.57 (1.15–2.14)

^aWhile social integration and marital status exposures are presented in the same table, they represent distinct analyses.

Notes. Model 1 is stratified on age and time period; Model 2 further controls for parity, duration of oral contraceptive use, tubal ligation, family history of breast/ovarian cancer, menopause status, duration of hormone therapy use; Model 3 further controls for BMI, smoking, recent physical exam. CI=confidence intervals; HR=hazard ratio.

Sensitivity Analyses

When considering histology (Supplemental Table 1, Supplemental Digital Content), risk estimates appeared slightly stronger for serous versus non-serous type; however, caution is warranted given the small number of cases in these analyses. Results were highly similar to those obtained in the main models when further adjusting for depression (e.g., 8–12 years lag fully-adjusted models: HR_{socially isolated versus integrated}=1.54, CI=1.09–2.19; HR_{widowed versus married}=1.65, CI=1.21–2.25). When using the modified SNI and adjusting for marital status, estimates were comparable to those of the main models: socially isolated versus integrated women had an elevated OvCA risk a decade later only (fully-adjusted: HR_{8–12 years}=1.41, CI=0.95–2.10; HR_{4–8 years}=1.00, CI=0.69–1.46), with no evidence of a linear relationship (p-trends>0.05); analogous results were obtained without marital status adjustment. Compared to women who were married/partnered and living with family, both widowed women living with family (HR_{fully-adjusted}=2.20, CI=1.45–3.34) and those living alone (HR_{fully-adjusted}=1.32, CI=0.99–1.77) had increased OvCA risk. Associations with incident OvCA were non-significant for separated/divorced women, regardless of whether they were living with their family or alone.

Stability of Psychosocial Indicators

Estimates were non-significant when considering change/stability in SNI categories with OvCA risk 4–8 years later (Supplemental Table 2, Supplemental Digital Content). However, associations became stronger when considering marital status. Notably, compared to women who remained married/partnered over two consecutive assessments, those who remained widowed (i.e., who did not get remarried) had a significantly higher OvCA risk 4–8 years later (HR_{fully-adjusted}=1.61, CI=1.17–2.23) but not 8–12 years later, whereas those who went from married/partnered to widowed had a significantly higher OvCA risk in both time periods (e.g., 8–12 years later, HR_{fully-adjusted}=2.22, CI=1.28–3.85).

Cumulative Effects of Psychosocial Indicators

When examining cumulative average SNI scores up to 4–8 years before diagnosis, being socially isolated versus integrated was related to higher OvCA risk (Table 3; HR_{fully-adjusted}=1.38, CI=1.00–1.92; p-trend=0.09). Similarly, compared to women who were never widowed, OvCA risk was significantly increased for those who remained widowed for a larger number of time assessments (i.e., who did not get remarried; fully-adjusted models: HR_{3–8 versus 0 assessments}=1.60, CI=1.17–2.20); the estimates were elevated but failed to reach statistical significance for women in the intermediate category (i.e., 1–2 assessments; p-trend=0.002). Long-term separation/divorce was not associated with OvCA risk, although statistical power was limited (i.e., 16–26 cases across strata; p-trend>0.05).

Table 3.

Associations of cumulative average social integration scores over follow-up and long-term marital status with ovarian cancer risk using a 4–8 years latency period between exposure assessment and diagnosis

Exposure	N (cases)	Model 1 HR (95% CI)	Model 2 HR (95% CI)	Model 3 HR (95% CI)
Cumulative scores on the Berkman- Syme Social Network Indexa	65,152 (436)
Socially integrated (ref.)	(56)	1.00	1.00	1.00
Moderately integrated	(117)	1.03 (0.81–1.32)	1.03 (0.81–1.32)	1.04 (0.81–1.32)
Moderately isolated	(148)	1.08 (0.83–1.40)	1.09 (0.84–1.41)	1.09 (0.84–1.42)
Socially isolated	(115)	1.35 (0.98–1.87)	1.37 (0.99–1.89)	1.38 (1.00–1.92)
p-trend				0.09
Number of assessments at which women were widowed	60,534* (405)
0 (remained married/ partnered; ref.)	(312)	1.00	1.00	1.00
1–2	(39)	1.29 (0.92–1.82)	1.31 (0.93–1.84)	1.32 (0.93–1.86)
3–8	(54)	1.54 (1.13–2.10)	1.59 (1.16–2.17)	1.60 (1.17–2.20)
p-trend				0.002
Number of assessments at which women were separated/divorced	58,819** (382)
0 (remained married/ partnered; ref.)	(340)	1.00	1.00	1.00
1–2	(16)	1.18 (0.70–1.97)	1.21 (0.72–2.01)	1.20 (0.72–2.01)
3–8	(26)	1.11 (0.74–1.67)	1.12 (0.74–1.68)	1.12 (0.74–1.68)
p-trend				0.47

^aWhile social integration and marital status exposures are presented in the same table, they represent distinct analyses.

Notes. Model 1 is stratified on age and time period; Model 2 further controls for parity, duration of oral contraceptive use, tubal ligation, family history of breast/ovarian cancer, menopause status, duration of hormone therapy use; Model 3 further controls for BMI, smoking, recent physical exam. CI=confidence intervals; HR=hazard ratio. To avoid misclassification in the reference group, time assessments at which women reported being *separated/divorced or **widowed were skipped from these analyses.

DISCUSSION

This study investigated the prospective association of social integration and marital status, two structural dimensions of social support, with OvCA risk in a large cohort with 20 years of follow-up. Findings revealed that being socially isolated, compared to being socially integrated, about a decade before OvCA diagnosis was associated with significantly increased risk of disease incidence. Widowhood was also associated with higher OvCA risk, in both the shorter and longer latency periods. In our models, the magnitude of these effects was comparable to those obtained from certain traditional OvCA determinants (e.g., hormone therapy use, familial history of breast/ovarian cancer) (3). Women with moderate integration or isolation levels, and those who were separated/divorced, did not have a higher risk than their socially integrated and married counterparts, respectively. Overall, these findings show that social isolation and widowhood are associated with OvCA onset and underline the importance of considering exposure many years before diagnosis. Interestingly, findings from secondary analyses also suggest that these effects are independent, since social isolation as measured with a modified SNI excluding marital status was still associated with an elevated OvCa risk 8–12 years later, even after controlling for marital status. Cumulative social isolation or long-term widowhood without re-entering marriage was also related to a higher OvCA risk, supporting the concept that chronicity of distress related to these structural social factors may play an important role in disease development.

Our findings are consistent with those of prior work showing an association of social isolation and unmarried status with cancer-related outcomes (9, 14–17). Of note, our risk estimates are of somewhat larger magnitude; while this could be due to the nature of the outcome (i.e., risk of cancer-related mortality versus OvCA incidence); it may also reflect stronger effects when psychosocial stressors are chronic or occur many years before diagnosis. Interestingly, recent evidence suggests that the window of development from a key precursor lesion to invasive high-grade serous OvCA–the most common histotype–averages between 7–9 years and may take up to 11–12 years (47, 48), hinting to a role of social stressors in early progression to overt disease. More broadly, our results are also aligned with findings for cardiometabolic outcomes (52), physical/mental functioning (53) and overall mortality (5, 6).

Several different mechanisms could explain these associations. For instance, individuals in a woman’s social network, including her spouse/partner, may encourage her to adopt favorable behaviors (e.g., healthier lifestyle habits, timely response to emerging symptoms, adherence to medical recommendations), as well as facilitate the gathering of health information and access to health care (11, 54) that lead to lower risk. Accordingly, socially isolated women and those without a partner may not engage as much or may progressively disengage from these protective habits over time, compared to their more integrated counterparts. A strength of our study was the ability to adjust for concurrent relevant health-related behaviors, namely BMI, smoking, and recent physical exam. Such statistical adjustment did not alter the associations, suggesting these may not be predominant pathways for OvCA risk, partly because such behaviors (e.g., smoking) are not strong OvCA determinants overall (1, 3). Similarly, further adjustment for depression status, which could be a potential pathway relating social isolation and widowhood to OvCA risk or a confounder of the relationship, did not substantially alter the associations.

Social integration may also act through alteration of biological pathways. For example, activation of the sympathetic nervous system leading to up-regulation of catecholamines, specifically norepinephrine, and downstream activation of β₂-adrenergic pathways is hypothesized as a primary mechanism of the distress-OvCA linkage (20, 29). Several studies in OvCA patients have shown that tumor norepinephrine is higher (26, 27, 55) and the expression of genes related to β-adrenergic signaling is upregulated (56) among those with poor functional social support, although the association of such biomarkers with structural support remains understudied. Grief as a response to widowhood also alters specific physiological systems, including dopamine release (55), which have been found to influence ovarian tumorigenesis in animal models (20, 57). While a prior study conducted in a small subset of OvCa patients from this cohort did not reveal that widowhood was differentially associated with risk of β₂-adrenergic receptor positive versus negative OvCA tumors (58), further work in larger samples is needed to determine how structural social dimensions, particularly widowhood, affect stress response systems that can influence ovarian carcinogenesis.

Interestingly, risk was not significantly higher for separated/divorced women. These results are consistent with other findings showing that while separation/divorce has a detrimental impact on some health outcomes (54), individuals who are widowed report worse self-rated health (59), exhibit uniquely compromised biological functions (e.g., inflammation, metabolism) (60), and have shorter cancer survival (14, 17) compared to those who are single but non-widowed (e.g., divorced, never married). Further, the impact of widowhood on OvCA risk was elevated both for women living alone or with family, suggesting structural (e.g., losing one’s significant tie, like a spouse) versus functional (e.g., living arrangement, including the related resources shared with others living in the same household) social isolation may be more important for OvCA development. Altogether, these findings support the idea that experiencing widowhood may foster a critical level of distress that increases susceptibility to developing OvCA in healthy women.

This study has several strengths, including the use of a large sample size with a prospective and longitudinal design. Updated assessments of social integration and marital status with a 20-year follow-up duration accounted for changes in structural support over time while minimizing recall bias. Similarly, using latency periods in our statistical analyses reduced potential for reverse causation (whereby preclinical OvCA symptoms can affect a woman’s interpersonal relationships, or their reporting on them). Moreover, relevant confounders and potential mediators were considered. Limitations included the relatively homogeneous sample regarding sociodemographics, notably marital status as all women were married at the larger cohort baseline (1976); while homogeneity enhances the internal validity, it does limit the results generalizability. Furthermore, detailed assessment of functional social support and quality of the relationships were not queried in this cohort, besides workplace-related perceived social support (61); hence it remains unclear whether unhappy marriages, reduced financial assistance, or other emotional/functional aspects of social support relate to OvCA development. Accumulating evidence indeed shows that supportive versus strained marital relationships have a positive impact on (some, but not all) health-related outcomes (36, 62). Lastly, complex social integration measures (i.e., assessing multiple dimensions), like the widely-studied SNI, have been associated more strongly to health outcomes compared to unidimensional binary items (5, 6); however, more nuanced information about a network’s structure remain uncaptured (7). For instance, whether disease risk differs for a woman who reports seeing her sole close friend several times a week compared to another one who sees each of her 10 close friends once a month is unclear.

The current findings suggest that socially isolated and widowed women are at a greater risk of developing OvCA, independent of known confounders and potential mediators. Associations were particularly strong when these stressors occurred many years before diagnosis and were sustained over time, underlining the importance of long follow-up periods and chronic psychosocial exposures when studying cancer determinants. These results are of particular importance, considering that in the U.S. >25% of the adult population lives alone (63) and ~50% of women are widowed by the age 70 (versus ~25% of men) (64). Given the documented disparities in OvCA outcomes (2), further research should examine whether structural dimensions of one’s environment are associated to OvCA incidence in populations who may be at greater risk due to having smaller or less stable social networks (e.g., lower socioeconomic status) (11). Generating such empirical evidence is aligned with the priorities of major health organizations that now recognize social connectivity as a significant health determinant and may contribute to the development of successful cancer prevention strategies (5, 7).

Acronyms:

BMI

body mass index

CI

confidence intervals

HR

hazard ratios

NHS

Nurses’ Health Study

OC

oral contraceptive

OvCA

ovarian cancer

SNI

Berkman-Syme Social Network Index