In sickness and in health: partner’s physical and mental health predicts cortisol levels in couples

Abstract

Individuals in stable relationships tend to be healthier than those not in stable relationships. Despite this general positive influence of relationships on health, the mechanisms for the impact of relationship quality on health are not clear. Research has focused on many factors to explain this connection, including inter- and intra-couple dynamics of physiology and behavior. To address this issue, we examined the relationship between perceived health, depressive symptoms, and relationship quality on diurnal cortisol in 30 male/female romantic dyads (N = 60). Participants provided saliva samples on two weekdays to assess total cortisol output. Females’ lower perceived physical health, lower relationship satisfaction, and higher depression scores were each related to higher cortisol output in their male partners. Males’ physical health, relationship satisfaction, and depression scores were unrelated to females’ cortisol output. Further, physical health, relationship satisfaction, and depression scores did not predict intra-individual cortisol levels for either sex. Measures of diurnal cortisol slope (DCS) were unrelated to psychosocial factors in males and females. Results provide further support for the interpersonal influence of partners’ mental and physical health on physiological outcomes and suggest females may influence their male partners more than vice versa.

Keywords:

• Couples
• covariation
• cortisol
• depression
• relationship satisfaction
• health

Introduction

Stress negatively impacts health and well-being. Social support may counteract stress effects to reduce adverse health outcomes (Cohen & Pressman, 2004). For many, romantic partners provide the greatest source of social support. The stress buffering model (Cohen & Wills, 1985) suggests that social support is protective against the negative health impacts of psychosocial stress, however, stress may covary within couples (Doerr, Nater, Ehlert, & Ditzen, 2018). For better or worse, romantic partners may co-regulate each other’s physical (Robles, Slatcher, Trombello, & McGinn, 2014) and mental health (Meyer & Paul, 2011). Yet, the process needs further exploration examining the mechanisms explaining these relationships.

Physiological changes during romantic couples’ interactions may impact health outcomes. Specific physiological alterations (e.g. reductions in cortisol) when a romantic partner is present may provide a buffering effect that ultimately results in beneficial health outcomes. Research examining cortisol covariation in dyads supports the idea that romantic partners influence each other’s cortisol responses to stress (Engert, Ragsdale, & Singer, 2018; Saxbe & Repetti, 2010). Such cortisol covariation appears to be the result of bidirectional influences between partners and may be reduced when romantic partners are happier in their relationships (Saxbe & Repetti, 2010) suggesting that stronger cortisol covariation may be indicative of troubled relationships (Saxbe et al., 2015). This could reflect a contagion among negative, but not positive moods (Saxbe & Repetti, 2010). For example, gender differences in cortisol responses may exist during couple conflict. Women may experience higher cortisol responses during conflict compared to men (Fehm-Wolfsdorf, Groth, Kaiser, & Hahlweg, 1999). Yet, another study demonstrated that men show higher acute cortisol responses to conflict whereas women showed a more prolonged recovery to baseline post-conflict (Laurent et al., 2013).

Couple dynamics may not be associated with only a transient cortisol response, but also long-term changes. Measures of the diurnal pattern of cortisol include indices of total cortisol output and diurnal cortisol slope (DCS); alterations of these measures have been linked to poor physical and mental health outcomes (see Adam et al., 2017 for review). Deviations from the normal cortisol response pattern are associated with depressive symptoms (Pariante & Lightman, 2008) and relationship dissatisfaction may precipitate a depressive episode as individuals who are chronically depressed consistently report dissatisfaction with their romantic relationship (Riso, Blandino, Hendricks, Grant, & Duin, 2002). Unhappiness with romantic relationships could be the result of changes in interactions between romantic partners. Research demonstrates a pathway between increased stress and negatively engaging with a romantic partner leading to depression (Westman & Vinokur, 1998). Yet, individuals experiencing depression display fewer positive nonverbal behaviors and are less socially responsive to their romantic partners. This may create a cycle that maintains the depression. The relationship between couple dynamics and cortisol may not be as simplistic as couple distress leading to negative symptoms.

Romantic relationships are protective against mental health distress (Cooper, Meyer, & Paul, 2008; Meyer & Paul, 2011) and romantic relationship satisfaction may be a primary factor contributing to the reduction in depressive symptoms (Whisman et al., 2004; Whisman, 2007). For example, the absence of conflict increased emotional support, and perceived closeness between partners predicts fewer depressive symptoms (Tower & Krasner, 2006). Relationships may modify the stress effects on the body and couples in happier relationships have lower cortisol levels (Heffner, Kieclot-Glaser, Loving, Glaser, & Malarkey, 2004). Meta-analyses have found consistent support for greater cardiovascular, endocrine, and immune health for individuals with supportive partners (Robles et al., 2014; Uchino, Cacioppo, & Kiecolt-Glaser, 1996) and reduced mortality (Holt-Lunstad, Smith, & Layton, 2010). Generally, individuals in satisfied romantic relationships are in better health than those who are not (Robles et al., 2014). For example, wives who felt supported by their husbands have lower blood pressure (Heffner et al., 2004). However, conflict between partners is stressful and physiologically arousing (Nealy-Moore, Smith, Uchino, Hawkins, Olson-Cerny, 2007). Negative couple interactions are associated with higher blood pressure, heart rate, and cortisol output (Barnett, Steptoe, & Gareis, 2005; Levenson & Gottman, 1985; Nealy-Moore et al., 2007) as well as impaired immune function (Kiecolt-Glaser, Glaser, Cacioppo, & Malarkey, 1998). Prolonged exposure to chronic stress and elevated cortisol leads to poorer physical health (McEwen, 2012); perhaps producing a synergistic effect whereby relationship stress exacerbates problematic health conditions (Jaremka, Lindgren, & Kiecolt-Glaser, 2013). Because of the difficulty of maintaining a chronic stress state, many distressed couples end their relationships (Gottman, 1999; Levenson & Gottman, 1985).

This exploratory study sought to examine the relationships among mental and physical health, relationship satisfaction, and diurnal cortisol as physiological indices of stress in romantic couples. We addressed the following questions: (1) What is the relationship between one’s own and one’s partner’s cortisol output and cortisol slope? And (2) What is the relationship between one’s own and one’s partner’s depressive symptoms, perceived health, and relationship satisfaction on one’s own and one’s romantic partner’s cortisol output and slope? We predicted cortisol responses would covary between romantic partners and one’s own and one’s partner’s depressive symptoms, perceived health, relationship satisfaction would predict diurnal cortisol levels.

Method

Participants

Using data from the Sloan Study of Youth and Social Development (a subset of the 500 Families Study; Schneider & Waite, 2008), participants included 30 male/female romantic dyads (60 total participants). The Sloan Study dataset includes 91 participants including 30 families with complete parent dyads and 31 families with data from only one parent. The second spouses from these 31 individual parent families chose not to participate in the study. Thus, considering our interests in dyadic processes, only families in which both parents participated in the study were included. All couples had teenage children. See Table 1 for sample demographic characteristics. Participants were invited to participate in this study from communities in the Midwest, Southeast, Northeast, and on the West Coast. The communities were mostly from suburban or urban settings with one rural community represented. Nonrandom sampling methods were used to recruit participants from local public schools, via newspaper advertisements, and snowball sampling methods. Data were collected through mailing or self-enumerated questionnaires and in-person interviews. See Adam (2006), Kurina, Schneider, and Waite (2004), and Schneider and Waite (2008) for details on participant characteristics and recruitment.

Table 1. Demographic and health-related characteristics and cortisol outcomes, 500 families study, Sloan cortisol study, and male–female partner dyads (n = 30 dyads).

Variables	Overall (n = 60)	Female (n = 30)	Male (n = 30)	p Value*
Age, mean (SD)	43.84 (6.22)	42.68 (5.38)	45.02 (6.86)	.011
White race, %(n)	85.00 (51)	83.33 (25)	86.67 (26)	.706
Total family income, %(n)	–	–	–	–
<$50,000	6.67 (4)	–	–	–
$50,000–$100,000	40.00 (24)	–	–	–
>$100,000	53.33 (32)	–	–	–
Employed, %(n)	93.33 (56)	86.67 (26)	100 (30)
Perceived health status (excellent/very good), %(n)	58.33 (35)	56.67 (17)	60.00 (18)	.564
Depressive symptoms, mean (SD)^a	9.19 (6.75)	9.66 (7.33)	8.69 (6.17)	.681
Marital satisfaction, mean (SD)^b	37.13 (6.32)	37.48 (6.77)	36.73 (5.88)	.480
Time weighted log cortisol average, mean (SD)	1.39 (0.30)	1.38 (0.32)	1.39 (0.27)	.802
Diurnal cortisol slope, mean (SD)	−0.13 (0.05)	−0.13 (0.05)	−0.12 (0.05)	.392

*p Value is for paired t-test (continuous) or McNemar test (dichotomous) comparing males and females.

^aCES-D Scale (20 items), scores range from 0 to 60.

^bEnrich Marital Satisfaction Scale (15 items), scores range from 15 to 75.

Procedures and measures

Salivary cortisol sampling and assay procedures. Procedures for this standardized study were conducted by a well-established team of researchers. Investigations using the cortisol data have been widely published. Detailed salivary cortisol methods for the Sloan Study of Youth and Social Development (Schneider & Waite, 2008) may be found in Kurina et al. (2004). Briefly, over two consecutive weekdays, participants were directed to provide six saliva samples beginning immediately upon awakening, immediately before bedtime, and four other pseudorandom times across the day when signaled by a beep emitted from a reminder watch given to participants by the research team. This schedule was constrained such that three samples were collected in the morning and three were collected in the afternoon or evening. Please see Kurina et al. (2004) for evidence of cortisol diurnal sampling. Signals from the reminder watch were scheduled 20 min after a diary entry and excluded sampling after meals. The diary entries contained information such as: other individuals present, current activities, and psychological state. Kraemer et al. (2006) noted that two sampling days results in reliable diurnal cortisol measurement; a more recent meta-analysis (Adam et al., 2017) suggests that studies that collected salivary cortisol only 1 d showed adequate relationships with health outcomes so long as a sufficient number of within-day samples (>3/d) were collected.

To increase saliva volume, participants chewed a piece of gum and expelled saliva via a straw into a sterile vial. Participants stored saliva samples in their home freezer before returning frozen samples (−20 °C) by courier to the research lab. Researchers then sent samples to a laboratory where the samples were frozen to −70 °C until assay. All samples were assayed in duplicate for salivary cortisol by high-sensitivity enzyme immunoassay using the Salimetrics LLC salivary cortisol assay (Salimetrics, State College, PA, see Adam (2006) and Kurina et al. (2004) for further details of assay). This assay has a range of sensitivity from 0.007 to 1.8 µg/dL, and average intra- and inter-assay coefficients of variation were less than 5 and 9%, respectively.

Measure of perceived health status. Participants were asked to rate their health using the following sentence stem, “Basically, would you say your health is (circle one).” Perceived health status was measured on a five-point scale with anchors of 1 = poor, 2 = below average, 3 = average, 4 = above average, and 5 = excellent. The use of one item to measure health perception is commonly used (Ware & Sherbourne, 1992) and has demonstrated reliability (Lundberg & Manderbacka, 1996) and validity (Quesnel-Vallée, 2007). For analysis, perceived health was measured as a dichotomous variable where 1 = excellent or above average health and 0 = average, below average, or poor perceived health.

Measure of depression. Depressive symptoms were measured with the Center for Epidemiologic Studies Depression Scale (CES-D; Radloff, 1977). This instrument contains 20 items answered for frequency via a four-point Likert scale. All questions referred to how participants felt about themselves or their lives in the past week (e.g. “I was happy” and “I felt sad”). This is a reliable and well-validated measure of depression (Corcoran & Fischer, 1987).

Measure of relationship satisfaction. Relationship Satisfaction was measured using the Enrich Marital Satisfaction Scale (EMS; Fowers & Olson, 1993). This 15-item test asks individuals to rate their agreement on a five-point Likert scale. Questions are based on either relationship satisfaction or conventionality (e.g. “I have some needs that are not being met by our relationship” and “My partner and I understand each other completely.”). The EMS is a reliable measure with evidence of validity (Fowers & Olson, 1993).

Data analysis

All analyses were conducted using SAS version 9.4 (SAS Institute, Cary, NC). Both DCS and time-weighted log cortisol average were selected to account for changes in slope across data points and the total cortisol output on each day of collection to capture not only cortisol output for healthy subjects, but also the unique cortisol response those with dysregulated HPA functioning exhibit. The amount of cortisol was quantified using a time-weighted log cortisol average calculated using the following equation for each day and then finding the mean for the two days, where n is number of samples in a given day, x_i is the log (i.e. natural log) of cortisol value for sample number i, and t_i is the corresponding sampling time in 24-h decimal notation (see Kurina et al., 2004): $$\frac{\frac{1}{2}\mathrm{*}{\sum }_{i=1}^{n-1}\left(x_i+x_{i+1}\right)\mathrm{*}\mathrm{ }\left(t_{i+1}-t_i\right)}{t_n-t_1}$$

Cortisol slope was calculated using a least squares model with the same time-weighted log cortisol value (Kurina et al., 2004).

Variables were summarized overall and by gender using frequencies and percentages or means with standard deviations. Paired samples t-tests for continuous variables and McNemar tests for 2 × 2 contingency tables were used to compare male and female demographic, predictor, and outcome variables at the dyad level. Correlations were calculated to ensure the predictor variables were independent constructs (see Tables 2 and 3 for details). Separate actor–partner interdependence model (APIM) analyses (Kenny, Kashy, Cook, & Simpson, 2006) were conducted on DCS and time-weighted log cortisol average to assess how one’s own and one’s partner’s predictor variables affect one’s own cortisol output. Six separate bivariate APIM models assessed the relationship of perceived health status, depressive symptoms, and marital satisfaction on cortisol output and slope. Dyads were treated as distinguishable with the distinguishing variable being gender. A two-intercept model approach was used to obtain actor-partner effects per level of gender. The models were run within the multilevel framework of SAS PROC MIXED, with a repeated statement and heterogeneous compound symmetry covariance symmetry (allowing variances to differ across dyad members). Models were created so male and female effects were calculated simultaneously. Eight effects are given in these models: (1) Female intercept; (2) Male intercept; (3) Female actor effect (e.g. effect of female’s own predictor variable on cortisol output); (4) Female partner effect (effect of male’s predictor variable on female’s cortisol output); (5) Male actor effect (effect of male’s own predictor on own cortisol output); (6) Female partner effect (effect of male’s predictor variable on female’s cortisol slope); (7) Male actor effect (effect of male’s own predictor on own cortisol slope); and (8) Male partner effect (effect of female’s predictor variable on male’s outcome).

Table 2. Correlations of primary independent variables among males (n = 30).

Variables (males)	Perceived health status	Depressive symptoms
Perceived health status	–	–
Depressive symptoms	−0.22	–
Marital satisfaction	0.23	−0.48*

*p<.01,

Table 3. Correlations of primary independent variables among females (n = 30).

Variables (females)	Perceived health status	Depressive symptoms
Perceived health status	–	–
Depressive symptoms	−0.22	–
Marital satisfaction	0.49**	−0.43*

*p<.05,

**p<.01,

Results

Table 1 shows a summary of demographic, health-related characteristics, DCS, and time-weighted average cortisol output for all participants and separated by gender. Tables 2 and 3 show correlations among predictor variables. The correlations do not reach the multicollinearity threshold of 0.8 (Pallant, 2005). Thus, our correlations demonstrate the variables are unique constructs, they do not represent a latent construct, and they contribute uniquely to the outcome variables. Tables are divided by gender showing correlations for males and females, respectively.

Table 4 shows results of three bivariate APIM models examining the relationships among psychosocial variables and time-weighted log cortisol average between dyads. Female partners’ perceived health status was associated with time-weighted average cortisol output of males such that males showed a −0.14 lower log time-weighted average cortisol level if paired with female partners reporting excellent or very good health status compared to those paired with females reporting good, fair, or poor perceived health. Greater female depressive symptoms were associated with higher cortisol output in male partners: males showed a 0.01 higher log time-weighted average cortisol level if paired with a female partner with more depressive symptoms. Similarly, female partners’ lower marital satisfaction was associated with higher cortisol levels in their male partners such that males showed a 0.01 higher log time-weighted average cortisol if paired with a female partner with lower marital satisfaction. In contrast, we found no such relationships between males’ psychosocial variables and their female partners’ cortisol output. Finally, no significant intra-individual relationships were noted between one’s own psychosocial variables and one’s own cortisol levels.

Table 4. Actor–partner interdependence models (APIM) for time weighted log cortisol average among males and females (n = 30 dyads).

Outcome model	Female cortisol average		Male cortisol average
Bivariate predictor	B (SE)	p Value	B (SE)	p Value
Perceived health status
Intercept	1.51 (0.13)	<.0001	1.58 (0.07)	<.0001
Actor effect	−0.10 (0.13)	.433	−0.09 (0.07)	.184
Partner effect	−0.09 (0.13)	.471	−0.14 (0.07)	.040
Depressive symptoms
Intercept	1.45 (0.13)	<.0001	1.36 (0.07)	<.0001
Actor effect	0.002 (0.008)	.817	−0.0002 (.005)	.969
Partner effect	−0.009 (0.01)	.381	0.01 (0.005)	.014
Marital satisfaction
Intercept	1.91 (0.50)	.001	1.84 (0.27)	<.0001
Actor effect	−0.01 (0.01)	.213	0.002 (0.006)	.740
Partner effect	−0.002 (0.01)	.874	−0.01 (0.005)	.024

Table 5 shows results of three additional bivariate APIM models examining the relationships among psychosocial variables and DCS between dyads. These models demonstrated no significant intra-individual or interpersonal relationships between the psychosocial variables and DCS.

Table 5. Actor–partner interdependence models, two intercept approach, for diurnal cortisol slope among males and females, 500 families study, and Sloan cortisol study (n = 30 dyads).

Outcome model	Female cortisol slope		Male cortisol slope
Bivariate predictor	B (SE)	p Value	B (SE)	p Value
Perceived health status
Intercept	−0.13 (0.02)	<.0001	−0.12 (0.02)	<.0001
Actor effect	0.004 (0.02)	.823	0.03 (0.02)	.213
Partner effect	−0.007 (0.02)	.734	−0.02 (0.02)	.287
Depressive symptoms
Intercept	−0.14 (0.02)	<.0001	−0.10 (0.02)	<.001
Actor effect	0.001 (0.001)	.483	−0.0004 (0.002)	.779
Partner effect	0.00003 (0.001)	.986	−0.001 (0.001)	.332
Marital satisfaction
Intercept	−0.02 (0.07)	.765	−0.20 (0.08)	.015
Actor effect	−0.0009 (0.001)	.494	0.003 (0.002)	.089
Partner effect	−0.002 (0.001)	.190	−.0006 (0.001)	.695

Discussion

We found that females’ perceived health status, depressive symptoms, and marital satisfaction were associated with their male partners’ daily time-weighted average cortisol output. By contrast, males’ perceived health, depressive symptoms, and marital satisfaction did not predict their female partners’ cortisol output. Consistent with previous research (Liu, Rovine, Cousino Klein, & Almeida, 2013; Saxbe et al., 2015), these results support the idea that romantic partners may influence each other’s cortisol levels and suggest within-partner influence is not uniform between sexes. The inter-individual connections between psychosocial reports and partner’s cortisol output suggest subtle correlations with partners’ physiological functioning that may have health implications.

Recent research demonstrated the influence of the observation of another’s stress on cortisol response (Buchanan, Bagley, Stansfield, & Preston, 2012; Engert, Plessow, Miller, Kirschbaum, & Singer, 2014; Engert et al., 2018). In laboratory settings, viewing both strangers and romantic partners under stress resulted in a cortisol reaction in both the target and observer, suggesting that witnessing another person’s psychological state can influence one’s own physiological response. Specifically, those who are more empathic (Buchanan et al., 2012) or more intimately connected (Engert et al., 2014, 2018) experienced a greater shared stress response by observing a target in distress. There is evidence these findings transfer beyond laboratory settings to naturalistic settings (Engert et al., 2018). Thus, one could conclude indices of cortisol may reflect both one’s own and romantic partner’s experiences of stress.

In contrast to the inter-individual effects, we found one’s own social, mental, and physical health did not predict one’s own cortisol levels. This suggests individuals may have difficulty estimating the influence of their physical or psychological well-being on a physiological indicator of physical health. This could be for two reasons. First, we know that individuals may respond to psychosocial questions in a socially desirable manner. Second, self-report measures of stress are often not associated with cortisol output (Chan, Sauvé, Tokmakejian, Koren, & van Uum, 2014; Dettenborn, Tietze, Bruckner, & Kirschbaum, 2010; Karlén, Ludvigsson, Frostell, Theodorsson, & Faresjö, 2011). Thus, stress perception and physiological stress may not be related due to differences in chronic stress that may alter diurnal cortisol reactions compared to day-to-day stress which may increase cortisol levels but not enough to produce lasting changes in diurnal responses.

The influences of cortisol on mental health are well-documented within individuals (Pariante & Lightman, 2008; Thomas et al., 2012). Yet, our findings demonstrate one’s romantic partner’s depression may be related to their own cortisol levels and suggests a directional pathway for how interpersonal regulation could exert its effects. The increase in cortisol levels may reflect the challenges romantic partners face when watching their beloved hurting (Thomas et al., 2012). It could be that when romantic partners are mentally struggling, it may leave healthy partners feeling helpless. The outcome, dysregulated cortisol, may be related to the psychological consequence of witnessing the partner’s pain.

Considerable research has demonstrated that relationship satisfaction buffers the negative impact of stress on health (Robles et al., 2014; Uchino et al., 1996). This supportive relationship may buffer against overactivation of the HPA axis, which may result in lower cortisol output and reduced negative health effects caused by cortisol hypersecretion. The current results highlight the directionality of the connection between supportive relationships and health. Just as a supportive partner may buffer the effects of stress on health and reduce total cortisol output, negative aspects of a partner’s health and relationship may negatively impact physiological functioning and health outcomes. Male partners may be more vulnerable – for good or for ill – to a romantic partner’s biopsychosocial variables than females, suggesting that men may be internalizing their partners’ stress levels. Meyer and Paul (2011) noted married couples report higher perceived stress than non-married individuals (but see Chin, Murphy, Janicki-Deverts, & Cohen, 2017). On the contrary, positive partner interventions may benefit men. For example, when women were enrolled in a health intervention, their male partners also received health benefits (White et al., 1991). The potential asymmetry of the stress sensitivity deserves further research attention.

The lack of a connection from male biopsychosocial variables to female’s cortisol levels is surprising. Despite the lack of support in this research, as females tend to live longer than males (Kochanek, Murphy, Xu, & Tejada-Vera, 2016) and most individuals are in romantic partnerships, females may gain health benefits via other mechanisms. It could be that women receive cardiovascular benefits. For example, women more satisfied in their romantic partnerships have lower blood pressure in the presence of their spouse compared to women in distressed romantic relationships (Carels, Sherwood, Szczepanski, & Blumenthal, 2000). But men also receive cardiovascular benefits from romantic relationships (Kiecolt-Glaser & Newton, 2001; Robles et al., 2014). It is plausible that the differences reflect that women have larger social networks than men. Women may communicate more with friends and family compared to males. Many social relationships may contribute to women’s health, while men may have a more restricted social support network (Szell & Thurner, 2013) and, therefore, be more influenced by partners’ health status.

The negative relationship between female partners’ relationship satisfaction and male partner’s cortisol levels replicates previous research (Saxbe et al., 2015). While our findings may support a similar underlying mechanism, our findings are unique. We found support for females’ biopsychosocial factors being associated with male romantic partners’ outcome variables. This may be reflective of Gottman’s (1999) work that notes the importance of husbands accepting their wives influence as a key element of conflict management. We may be uncovering relationship dynamics in socially healthy, heterosexual couples that are applicable beyond disagreements to the biopsychosocial model of health. Yet, it cannot be ignored that these findings might be related to partner perception that emerged as gender differences. For example, when individuals have more favorable perceptions of their partner compared to themselves, they report greater relationship satisfaction (Busby, Holman, & Neihus, 2009).

Despite the implicit association between diurnal slope and time-weighted average cortisol output, we found no relationship among our variables using DCS as the dependent variable. Individuals with dysregulated HPA functioning exhibit blunted cortisol slopes contributing to greater cortisol output across the day (Adam et al., 2017). This implies a two-step process whereby the use of the variable time-weighted log cortisol average in the main analysis captured individual differences in cortisol slope and total cortisol output. Perhaps, suggesting that for healthy populations, DCS may not be the best unit of measurement. Despite that total cortisol output and slope are generally correlated because couples could influence one another throughout the course of the day, it could be that using slope was merely capturing the morning and evening partner influences and our use of total cortisol output captured a comprehensive potential of partner influence.

The stress buffering hypothesis is well supported through correlational studies (Robles et al., 2014), but lacks mechanistic explanations for why social relationships may produce lasting benefits. Our findings may partially explain how social relationships could contribute to long-term health effects. Cortisol affects many bodily systems (e.g. cardiovascular, immune, and endocrine) that contribute to the well-being of individuals. Interpersonal variables, such as partner’s health and well-being may exert subtle, but significant effects on physiological functioning and health over the course of a relationship. As such, results from this study and others support testing couple interventions as an avenue to alter both partners’ physical and mental health concerns. Systemic treatment enhancing the couple relationship could be a form of preventative healthcare for both partners. Strengthening emotional ties between romantic partners could supplement treatments to hopefully, produce a synergistic effect on traditional healthcare treatment to improve clinical outcomes.

These results should be interpreted with caution due to methodological limitations. First, the sample was small with only 30 heterosexual couples and all couples were middle-aged with teenage children. Having children of this age may be challenging for the couple relationship (Meyer, Robinson, Cohn, Gildenblatt, & Barkley, 2016), influencing health factors, depression, as well as cortisol output. Second, measurement of perceived health included only one question. The use of a single question limits our ability to generate a comprehensive understanding of health. Third, the majority of the sample reported middle to high socioeconomic backgrounds. Lower-income couples report lower relationship satisfaction (Hawkins & Erickson, 2015). Furthermore, lower socioeconomic status is associated with increased cortisol and epinephrine (Cohen, Doyle, & Baum, 2006). Therefore, our results could be biased and not generalizable to couples from lower socioeconomic strata. Fourth, the women in the study varied in phase of the menstrual cycle and menopausal transition. Women in menopause may experience higher cortisol secretions overnight (Woods, Mitchell, & Smith-DiJulio, 2009). Research supports variance in cortisol levels may be specifically related to differences in menstrual cycles among females (e.g. menstrual cycles with follicular or luteal phases longer than 14 d; Nepomnaschy et al., 2011). Thus, the gender difference found in this study may be influenced by menstrual/menopausal conditions. This study is limited by the number of days cortisol was sampled, which, along with a host of genetic and environmental factors, results in considerable within-person variability in diurnal cortisol (as noted by Davis (2017) and by Kurina et al. (2004), using the same dataset). As cortisol was measured over only 2 days, this is not as reliable as measuring cortisol for a longer duration. Further, the saliva sampling regimen was not designed to assess the cortisol awakening response (Pruessner et al., 1997), which has been associated with a range of clinical disorders (see Clow, Thorn, Evans, & Hucklebridge, 2004) and may have demonstrated a different pattern of covariation compared to the diurnal cycle collected throughout the day.

Future research should address these relationships with larger samples providing sufficient power to control for menstrual cycle and menopausal symptoms, socioeconomic status, and include greater cultural diversity, such as the types of couples included (e.g. same-sex couples), and include more variety in children’s ages as well as include couples without children. We know relationship satisfaction differences are present between those with and without children (Meyer et al., 2016). Including the presence of children in the relationship in future research may add to the generalizability of these findings. Future research may be best assessed through longitudinal studies following newly committed couples across the lifespan, paying close attention to cortisol, relationship satisfaction, mental health, physical health, and partner influences. Such work may help to reveal how and whether female partners benefit from romantic relationships.

Despite these limitations, our results suggest the importance of nurturing romantic partnerships to help buffer stress responses especially in men. The findings may provide the groundwork for explaining the mechanisms underlying the stress buffering hypothesis and serve as an impetus for future research to test strengthening emotional support in romantic relationships as preventive healthcare.

Disclosure statement

No potential conflict of interest was reported by the authors.