Prospective self- and informant-personality associations with inflammation, health behaviors, and health indicators

Abstract

Objective:

Personality influences many aspects of the health process. It is unclear to what extent self- and informant-reports of the Big Five offer incremental validity for the prediction of inflammatory biomarkers and whether inflammation provides a unique pathway between personality and indicators of physical health, independent of health behaviors.

Methods:

Using data from older adults (N = 1,630) enrolled in the St. Louis Personality and Aging Network study, we tested whether self- and informant-reported Big Five traits show unique associations with inflammation (IL-6, CRP, TNF-α). Further, we tested whether inflammation and health behaviors indirectly link personality to health-related quality of life, body mass index, and chronic disease burden using longitudinal mediation in a structural equation modeling framework.

Results:

Self-reports, informant-reports, and general trait factors of personality predicted future inflammatory biomarker levels (unstandardized regression coefficients ranged −0.08 to 0.07 for self, −0.13 to −0.10 for informants, and −0.16 to −0.11 for general). Additionally, all assessment methods of personality were associated with the indicators of physical health through biomarker and health behavior pathways. Effects were primarily found for conscientiousness and neuroticism; IL-6 and CRP were the biomarkers with the most indirect effects; and indirect paths overall emerged more frequently through health behaviors, but this varied by outcome.

Conclusions:

Self- and informant-reports provided unique predictive validity of inflammatory biomarkers. Findings highlight the benefits of using of multiple assessments of personality and the importance of examining multiple, distinct pathways by which personality might influence health to understand the mechanisms underlying this relationship more fully.

A relationship between personality traits and health is well-established. Assessments of personality are prospectively related to health outcomes such as stroke, dementia, heart disease, and cancer (Terracciano et al., 2014; Weston et al., 2015). These personality-health associations predict outcomes decades in advance (Friedman et al., 2010), with different assessment methods (Jackson et al., 2015; Turiano et al., 2013), and across the lifespan (Hill et al., 2011; Turiano et al., 2015). Personality is believed to influence health through mechanisms such as perceptions of health, disease progression, and health behaviors (Friedman et al., 2010; Roberts et al., 2009; Weston & Jackson, 2016), but health behaviors are arguably the most well-studied.

One mechanism may involve immune system functioning, with inflammation being routinely linked to the regulation of health processes (e.g., Glaser & Kiecolt-Glaser, 2005; Petersen & Felker, 2006). Prior work relies exclusively on self-report personality measures, thus potentially missing meaningful personality variance captured by informant-reports (Jackson et al., 2015; Smith et al., 2008). The current study uses a longitudinal sample of older adults with measures of self- and informant-report personality, inflammatory biomarkers, health behaviors, and different markers of physical health status (physical health-related quality of life (PHRQoL), body mass index (BMI), and chronic disease burden (CDB)) to disentangle how immune system markers relate to the personality-health process.

Personality to Health Behaviors to Health

Of the Big Five (Goldberg, 1990) personality traits, conscientiousness is the most reliably linked to health processes. It predicts lower BMI, physical health, longevity, and mortality risk (Brummett et al., 2006; Jackson et al., 2015; Jokela et al., 2018; Hill et al., 2011). Neuroticism has been associated with negative health outcomes (Lahey, 2009), such as higher BMI (Brummett et al., 2006) and increased rates of physical disorders (Goodwin & Friedman, 2006), although studies have typically focused on mental health over physical health outcomes (Friedman et al., 2010; Ozer & Martinez, 2006). Lower levels of agreeableness predict poorer physical and subjective health (Turiano et al., 2015), premature mortality (Jackson et al., 2015), arthritis (Weston et al., 2015), and heart disease (Miller et al., 1996). Extraversion often shows both smaller positive and negative effects on health (Friedman et al., 2010; Munafo et al., 2007). Lastly, openness is associated with longevity (Jackson et al., 2015) and stroke (Weston et al., 2015).

These personality-health associations are thought to be partly driven by the initiation and maintenance of behaviors that directly impact health, known as health behaviors (Turiano et al., 2015). More specifically, people with higher levels of certain traits (e.g., conscientiousness) are more likely to engage in behaviors that promote positive health (e.g., exercise, healthy diet) and, conversely, refrain from risk behaviors (e.g., substance abuse, unprotected sex) that make poorer health outcomes more likely (Bogg & Roberts, 2004; Kern & Friedman, 2011; Lodi-Smith et al., 2010). Personality’s indirect effect on health via health behaviors can accumulate slowly over time as these behaviors are routinized or incorporated into daily habits (Hampson et al., 2007).

Despite these sensible links, health behaviors are not the only pathway linking personality with health. While longitudinal studies find similar associations to those found in cross-sectional studies, the effects are often small in magnitude and inconsistently replicate (e.g., Ploubidis & Grundy, 2009; Mottus et al., 2013; Hampson et al., 2007). For those that do, there is still much variance left unexplained (O’Súilleabháin et al., 2021; Turiano et al., 2012; Mroczek et al., 2009).

Personality to Immune Functioning to Health

One potential alternative pathway underlying the personality-health link is immune functioning (Glaser & Kiecolt-Glaser, 2005; Peters et al., 2003). Among these immune processes is inflammation, one of the body’s key defenses in response to future harm (Allen & Laborde, 2017). While in the short-term inflammation is helpful, chronic inflammation can be problematic and contribute to poor health (Tanaka et al., 2014). A number of signaling mechanisms are responsible for triggering and regulating inflammatory processes in response to foreign or harmful entities, including inflammatory biomarkers such as interleukin-6 (IL-6), C-reactive protein (CRP), and tumor necrosis factor-alpha (TNF-α; Armon et al., 2013; Denollet et al., 2008).

IL-6 is released in response to tissue damage and infections and plays a key role in inflammation and autoimmunity (Tanaka et al., 2014). Prolonged elevated levels can lead to chronic inflammation (Gillmore et al., 2001), immune dysfunction (Alonzi et al., 1998; Fattori et al., 1994), and heart disease (Kanda & Takahashi, 2004). CRP often works alongside IL-6 for inflammatory responses, as IL-6 induces its synthesis (Heinrich et al., 1990). Similar to IL-6, high levels of CRP are linked to poor health outcomes (Berton et al., 2003). TNF-α is a cell-signaling protein involved in the stress response and inflammation. TNF-α is an inducing factor for IL-6 synthesis (Walther et al., 1988).

Personality traits have been associated with these inflammatory biomarkers. High levels of conscientiousness are connected to low levels of IL-6 and CRP (Graham et al., 2018; Turiano et al., 2013). Neuroticism has been linked to higher levels of IL-6 and CRP (Graham et al., 2018; Sutin et al., 2010) and agreeableness shows similar associations (Allen & Laborde, 2017; Turiano et al., 2013), although this is not always replicated (Luchetti et al., 2014). High levels of openness are associated with lower levels of IL-6 and CRP (Graham et al., 2018; Armon et al., 2013), while extraversion often shows inconsistent or null effects (Luchetti et al., 2014, Sutin et al., 2010).

Despite connections between personality and inflammation, their associated effect sizes tend to be small. These small associations may occur because of the multi-faceted nature of inflammation. Alternatively, reliance on a single assessment method, such as self-report, may underestimate or fail to capture all relevant associations between personality and inflammation. It is notable that all prior studies examining the personality-biomarker association relied exclusively on self-report personality assessments. Including and analyzing a secondary source of data from multi-modal assessments helps to reduce the bias and ensures all aspects of one’s personality implicated in this process are accurately captured (e.g., Smith et al., 2008; Jackson et al., 2015).

Current Study

The current study examines three aims in a large-scale, diverse study of older adults: First, it is unclear whether informant-reports of personality provide unique information over and above self-reports of personality when predicting inflammatory biomarkers. Second, it is unclear whether personality traits’ associations with three physical health indices are mediated by IL-6, CRP, and TNF-α, as this path has not been examined. Recent work by O’Súilleabháin et al. (2021) tested if inflammation mediated the pathway between personality and mortality and primarily found associations for conscientiousness and IL-6. We seek to extend these results by examining a third biomarker (TNF-α) and three indicators of physical health. Specifically, we used PHRQoL, BMI, and CDB. Third, it is unclear if immune function and health behaviors are independent pathways between personality and health. The health behaviors related to immune functioning (e.g., exercising and smoking) are also linked to the Big Five (Bogg & Roberts, 2004; Terracciano & Costa, 2004; Malouff et al., 2006), suggesting possible redundancy. Alternatively, they may operate independently, such that health behaviors linked to personality-health outcomes are not associated with immune functioning, such as accident prevention or treating health problems before they worsen with time.

Methods

Participants

The St. Louis and Personality Aging Network (SPAN) study is an ongoing longitudinal study assessing a wide range of personality, health, social, experiential, and biological features among a sample of 1,630 older adults recruited from the St. Louis metropolitan area (ages 55-65 at baseline, M = 59.5, SD = 2.7; 55% female) using telephone records purchased from a private sampling firm. Extensive details regarding the recruitment procedures are provided in previous manuscripts (Oltmanns et al., 2014). Within certain methodological constraints (e.g., limiting the sample to people who could speak and read English and people with an address and phone number so that they could be followed over time), everything possible was done to recruit a representative sample of people living in St. Louis. Participants represented a wide range of educational and income levels. The proportions of race within the sample are representative of the St. Louis area at 64% White (n = 1040), 32% African American (n = 516), 2% Hispanic (n = 26), and 3% other (n = 48). Each participant was paid $60 to complete a 3-hour in-person assessment at baseline (N = 1,630; baseline) and has been followed up multiple times across a roughly 15-year span.

For the present study, we used data from three waves. First, personality measures (n = 1,630) were collected at the baseline assessment in 2007; measures of health behaviors (n = 1,053), covariates (n = 860), and biomarker samples (n = 791) collected roughly 7 years later in 2014; PHRQoL (n = 411) and BMI (n = 497) collected in 2017; and CDB (n = 1,627) collected initially at baseline in 2007 and any changes were noted in follow-up assessments. This study was approved by the Institutional Review Board at Washington University in St. Louis (protocol #201102523).

In addition to data from participants, informant data was collected at some assessment waves. An informant was identified as someone who knew the participant well and could provide an accurate description of their personality. Informant data was collected for 91% of the original sample; 50% of the informants were spouses/partners, 25% other family members (e.g., adult child), and 25% close friends at the baseline assessment. Participants and informants had known each other for an average of 30 years at baseline (Oltmanns et al., 2020).

Compared to those who completed all three waves of data (attrition rate was less than 30% after accounting for those that passed away), those who did not provide data at the final wave consisted of a lower proportion of White participants (χ²(1) = 30.19, p < .001), a higher proportion of Black participants (χ²(1) = 28.43, p < .001), and had lower average levels of education (t(1563.7) = −6.03, p < .001, d = −0.30). For self-reports, those who completed the final wave of personality items were slightly more agreeable (t(1476.8) = 2.55, p = .011, d = 0.13) and open (t(1527.6) = 2.11, p = .035, d = 0.11). There were no significant differences in the levels of the inflammatory biomarkers nor in ratings of engagement in health behaviors between those who provided data at the second wave in our study and those that did not provide data at the final wave.

Compared to individuals who remained alive throughout the waves of our study, participants who passed away (N = 168) were more neurotic (t(201.58) = 2.17, p = .031, d = 0.18) and less open (t(206.31) = −3.16, p = .002, d = −0.25). Only 15 of these individuals provided inflammatory marker data for all three biomarkers, and only 31 individuals provided health behavior data. Additionally, individuals who passed away consisted of a lower proportion of White participants (χ²(1) = 32.75, p < .001), a higher proportion of Black participants (χ²(1) = 36.14, p < .001), and had lower average levels of education (t(210.25) = −6.82, p < .001, d = −0.53).¹

Study participation rate at baseline was 43%, which compares favorably to that obtained in other epidemiological studies at the time of the study’s start (Galea & Tracy, 2007). The goal of the SPAN study was to obtain a sample representative of the St. Louis area; thus, even if some individuals who fit the desired demographics did not participate in the study, similar individuals who resembled them demographically were recruited instead. As a result, the final sample was representative of the St. Louis population (see Oltmanns et al., 2014 for more details). Notwithstanding the demographically representative sample, the reasons for choosing to participate in the study are ultimately not known, and thus it is possible the individuals who chose not to participate differed from those who did on important study characteristics (e.g., health). For example, individuals who participated had slightly higher household incomes than the median household income in St. Louis at the time (but resembled that of the metropolitan area) and had slightly higher average levels of education (Oltmanns et al., 2014). Additionally, the sample size at baseline was 1,630; follow-up 10 (i.e., our second wave of data) was 1,072; and follow-up 12 (i.e., our third wave) was 1,035. Reasons for people not providing data at all waves include they dropped out of the study (n = 268), died (confirmed n = 168), or they were unavailable (n = 159).

Materials

Personality.

Personality traits were measured with the NEO-PI-R (Costa & McCrae, 1992). Participants received the standard version of the measure while informants received a version with questions asking about the participant. Baseline personality scores were used in this study. Participants rated the extent to which they agreed with items asking questions related to each of the traits on a five-point scale ranging from 0 (“strongly disagree”) to 4 (“strongly agree”). Average Cronbach’s alpha values for the traits were α = .89 for self-report and α = .92 for informant-report.

Inflammatory Biomarkers.

Morning fasting blood samples were collected between 7:30-10:00 am via peripheral (primarily antecubital) venipuncture from consenting participants (n = 791) in an independent session closely following IPFU-2. Samples were not obtained from participants reporting acute illness or injury. Samples were processed according to standard operating procedures before being stored at −80° C (Tuck et al., 2009). Following collection, samples were kept upright for 40-60 minutes to allow for clot formation before being centrifuged (1300g) for 20 minutes at room temperature. Immediately after, samples were pipetted into four to eight 2 ml tubes containing volumes of 500 ug or 1 ml before being frozen at −80°C.

Prior to assaying, samples were removed from −80°C and brought to room temperature for 45 minutes. Samples were then centrifuged for two minutes at 3260 xG to draw contaminants to the bottom of the tube. IL-6, CRP, and TNF-α were assayed from serum in duplicate using commercially available enzyme-linked immunosorbent assays (IL-6: Quantikine HS Human IL-6, R&D Systems, Minneapolis, MN, USA; CRP: EIA-3954 High Sensitivity C-Reactive Protein ELISA DRG International Inc., USA; TNF-α: Quantikine HS Human IL-6, R&D Systems, Minneapolis, MN, USA). Intra-and inter-assay coefficients of variation were acceptable (intra-assay CVs all < 8%, inter-assay CVs all < 14%). Samples producing unreliable measures were excluded. Inflammatory markers were log transformed and winsorized prior to analyses.

Health Behaviors.

Reports of health behaviors were assessed using the Health Behavior Checklist (HBC; Vickers et al., 1990). The HBC consists of 40 items broadly covering a person’s engagement in various protective behaviors or risk behaviors, including exercising, smoking, and regular doctor check-ups. Health behaviors were assessed at the second assessment wave and items were composited into a single variable. Cronbach’s alpha was .84. The final health behavior variable used in all analyses was standardized.

Indicators of Physical Health.

Three indices of physical health were measured via PHRQoL (i.e., what the person can physically do and how they feel) obtained from the RAND-36 Health Status Inventory (HSI; Hays & Morales, 2001), BMI, and CDB. Although PHQRoL is self-report, past research has shown that self-reports of health predict mortality (e.g., Ware et al., 1994). This measure was administered approximately 10 years after the baseline assessment. A higher-order factor for PHRQoL was calculated² (see Hays et al., 1993 for detailed scoring methods). BMI was calculated from a participant’s self-reported height and weight approximately 10 years after baseline assessment. CDB was a variable obtained by summing responses across dichotomous questions asking if the participant was diagnosed with heart disease, cancer, hepatitis, stroke, arthritis, asthma, diabetes, bleeding ulcer, epilepsy, or serious illness. Initial responses were obtained at baseline and updated each wave if and as new conditions arose. The final PHRQoL and BMI variables used in all analyses were standardized.

Covariates.

Included covariates were age at baseline, sex at baseline (coded as 0 = Male, 1 = Female), race at baseline (with three separate dummy-coded variables for Black, Hispanic, and Other), highest education level at baseline (coded as 0 = less than high school, 1 = high school or GED, 2 = some college, 3 = vocational school, 4 = 2-year college degree, 5 = 4-year college degree, 6 = master’s degree, 7 = doctoral degree, and 8 = professional degree; M = 5.00, SD = 2.07), average caffeine consumption per day at wave 2 (i.e., number of caffeinated beverages consumed in the average day; M = 2.25, SD = 1.68, Min = 0, Max = 5.5), mean arterial pressure (MAP) measured directly at the blood draw appointment during wave 2 with an inflatable blood pressure cuff (M = 102.69, SD = 13.79, Min = 65.67, Max = 159.67), and whether the participant was taking certain medications that can influence levels of the inflammatory markers. MAP was standardized and the medication variable was dummy-coded (0 = not taking any medicines, 1 = currently taking any medicine). These included ACE inhibitors (n = 303); aspirin (n = 296); any antidepressants (i.e., SSRIs, TCAs, or SNRIs; n = 175); beta blockers (n = 172); benzodiazepines (n = 50); hormonal medications (n = 72); nonsteroidal anti-inflammatory drugs (NSAIDs; n = 155); prescription pain killers (n = 98); statins (n = 307); and steroid medications (n = 102).

Analysis Plan

Analyses were carried out with the structural equation modeling program lavaan (Rosseel, 2012) within R (R Core Team, 2020). Full information maximum likelihood (FIML) was used to handle missing data for all models. Analysis scripts and expanded results are available at https://osf.io/bafvc/. First, to determine simple associations of personality with our health outcome variables of interest, confirmatory factor analyses (CFAs) were constructed to represent the association of each latent self- and informant-report personality trait with each of our outcomes. Further, a bifactor model was created to model the shared variance of self- and informant-reports. A bifactor model simultaneously models the variance of the self-and informant-reports and separates it into a general trait factor, which captures the shared variance between the two report methods; a self-report trait factor, which represents the variance unique to the self-report method; and an informant-report trait factor, which represents the variance unique to the informant-report method (see Chen et al. (2011) and Brunner et al. (2011) for a helpful and thorough description and application of these models). Then, using the CFA and bifactor models, each outcome variable was regressed onto personality, while controlling for covariates.

Then, to address Aim 1 – the assessment of whether self- and informant-reported personality is redundant in the prediction of inflammatory markers – CFAs were again constructed to represent the relationship of each latent self-report, informant-report, and bifactor personality trait. Each inflammatory biomarker was regressed onto personality, while controlling for covariates. Aim 2 – whether inflammation is a mechanism through which personality impacts health – was tested using the indirect effects of personality and the three indicators of physical health mediated by inflammatory biomarkers, with a separate test for each personality trait and biomarker. A separate set of models were run for PHRQoL, BMI, and CDB as the outcome variables. The indirect effect of personality and the respective physical health variable through inflammatory markers was the focal parameter in each set of models. To test Aim 3 – determining whether inflammatory markers capture the effects of health behaviors or carry additional information – health behaviors were introduced as a second mediator. Similar to Aim 2, we estimated separate models for each trait and biomarker, with a set for PHRQoL, BMI, and CDB. Indirect effects for the health behavior pathway and the inflammatory pathway were both calculated. For Aims 2 and 3, covariates were controlled for in all indirect paths. For an effect size for all indirect paths in the mediation analyses, in addition to the standardized regression coefficients, the proportion of the indirect effect relative to the absolute value of the total effect was calculated. Furthermore, to compensate for the number of tests, the False Discovery Rate (FDR) correction was applied to all analyses.

Results

Do self- and informant-reports of personality uniquely predict future levels of inflammation?

Descriptive statistics for study variables are available in Table 1 and correlation matrices in Tables S1–S2. First, Aim 1 was tested to see if self- and informant-reports of personality offered redundant or unique contributions in predicting inflammatory biomarkers. In text, we report significant unstandardized results from analyses of self-report, informant-report, and bifactor models of personality with covariates (see Table 2 for unstandardized and standardized results from each Big Five trait; see Tables S3–S8 for estimates of all traits along with covariates).

Table 1.

Basic Descriptive Statistics for Study Variables

Study Variable	N	M	SD	Range	Wave of Data
Demographic & Background Information
Age at Baseline	1628	59.08	2.94	13.00	1
Mean Arterial Pressure (MAP)	888	102.69	13.79	94.00	2
Average Daily Caffeine	694	2.25	1.68	5.50	2
Personality
SR Extraversion	1613	2.27	0.38	2.54	1
IR Extraversion	1464	2.30	0.45	2.94	1
SR Agreeableness	1613	2.70	0.32	2.37	1
IR Agreeableness	1464	2.59	0.48	3.17	1
SR Conscientiousness	1613	2.57	0.36	2.90	1
IR Conscientiousness	1464	2.67	0.53	3.17	1
SR Neuroticism	1613	1.51	0.43	3.08	1
IR Neuroticism	1464	1.62	0.55	3.35	1
SR Openness	1613	2.34	0.38	2.52	1
IR Openness	1464	2.21	0.40	2.67	1
Inflammatory Biomarkers
IL-6	753	0.26	0.29	1.84	2
CRP	769	0.41	0.52	2.85	2
TNF-α	651	0.15	0.21	1.44	2
Health Measures
Health Behaviors	1053	3.78	0.44	3.95	2
Physical Health-Related Quality of Life (PHRQoL)	425	62.06	9.33	44.17	3
Body Mass Index (BMI)	497	30.15	7.07	57.38	3
Chronic Disease Burden (CDB)	1627	1.29	1.19	7	1-3

Note. M = mean. SD = standard deviation. SR = self-report. IR = informant-report. IL-6 = interleukin-6. CRP = C-reactive protein. TNF-α = tumor necrosis factor – alpha.

Average daily caffeine consumption is the number of standard caffeinated beverages on an average day. The personality traits were scored on a 0-4 Likert scale. The inflammatory biomarkers are in log units. Health behaviors were scored on a 1-5 Likert scale. The BMI, MAP, health behavior, and PHRQoL variables in subsequent analyses are standardized; the raw units in this table are for descriptive purposes only.

Table 2.

Aim 1: Estimates of Baseline Personality Predicting Future Inflammatory Biomarker Levels

	IL-6			CRP			TNF-α
Model	b	95% CI	b*	b	95% CI	b*	b	95% CI	b*
Extraversion
SR	−0.02	[−0.08, 0.03]	−0.03	−0.02	[−0.12, 0.08]	−0.02	−0.04	[−0.09, −0.00]	−0.08
IR	−0.03	[−0.07, 0.02]	−0.05	−0.08	[−0.16, −0.00]	−0.07	−0.01	[−0.05, 0.02]	−0.03
BF – G	−0.03	[−0.09, 0.04]	−0.03	−0.03	[−0.13, 0.07]	−0.02	−0.04	[−0.08, 0.01]	−0.06
BF – SR	−0.32	[−3.85, 3.21]	−0.01	−0.13	[−3.59, 3.32]	−0.01	−0.05	[−0.11, 0.01]	−0.03
BF – IR	−0.02	[−0.09, 0.05]	−0.03	−0.12	[−0.24, 0.01]	−0.08	0.02	[−0.04, 0.08]	0.03
Agreeableness
SR	−0.04	[−0.11, 0.04]	−0.04	−0.13	[−0.26, 0.01]	−0.07	0.00	[−0.06, 0.06]	0.00
IR	−0.01	[−0.05, 0.03]	−0.02	−0.09	[−0.17, −0.01]	−0.08	−0.03	[−0.07, 0.00]	−0.07
BF – G	−0.04	[−0.12, 0.05]	−0.03	−0.14	[−0.29, 0.00]	−0.07	−0.01	[−0.07, 0.05]	−0.02
BF – SR	−0.05	[−0.15, 0.05]	−0.03	−0.12	[−0.31, 0.07]	−0.04	0.02	[−0.05, 0.09]	0.02
BF – IR	−0.01	[−0.06, 0.04]	−0.01	−0.08	[−0.17, 0.02]	−0.06	−0.04	[−0.08, −0.00]	−0.08
Conscientiousness
SR	−0.08	[−0.14, −0.03]	−0.11	−0.06	[−0.16, 0.04]	−0.05	−0.04	[−0.09, 0.00]	−0.08
IR	−0.08	[−0.12, −0.03]	−0.13	−0.15	[−0.22, −0.07]	−0.14	−0.03	[−0.06, 0.01]	−0.06
BF – G	−0.11	[−0.17, −0.05]	−0.11	−0.16	[−0.29, −0.04]	−0.10	−0.04	[−0.09, 0.01]	−0.06
BF – SR	−0.06	[−0.13, 0.02]	−0.04	−0.01	[−0.15, 0.12]	−0.01	−0.04	[−0.09, 0.02]	−0.06
BF – IR	−0.06	[−0.11, −0.01]	−0.08	−0.45	[−0.87, −0.03]	−0.15	−0.01	[−0.05, 0.04]	−0.00
Neuroticism
SR	0.07	[0.02, 0.11]	0.11	0.08	[−0.00, 0.16]	0.07	0.01	[−0.02, 0.05]	0.03
IR	0.05	[0.02, 0.09]	0.10	0.10	[0.04, 0.17]	0.11	0.00	[−0.03, 0.03]	0.01
BF – G	0.08	[−0.03, 0.19]	0.09	0.21	[0.03, 0.39]	0.14	−0.04	[−0.13, 0.04]	−0.07
BF – SR	0.06	[−0.02, 0.14]	0.07	−0.00	[−0.13, 0.13]	0.00	0.04	[−0.02, 0.11]	0.07
BF – IR	0.08	[−0.27, 0.44]	0.06	−0.12	[−0.73, 0.49]	−0.05	0.19	[−0.10, 0.48]	0.18
Openness
SR	−0.04	[−0.10, 0.02]	−0.05	0.01	[−0.09, 0.11]	0.01	−0.02	[−0.06, 0.03]	−0.03
IR	0.00	[−0.06, 0.05]	0.00	0.01	[−0.08, 0.11]	0.01	0.02	[−0.03, 0.06]	0.03
BF – G	−0.03	[−0.09, 0.04]	−0.03	0.01	[−0.10, 0.12]	0.01	0.00	[−0.05, 0.05]	0.00
BF – SR	−0.07	[−0.15, 0.01]	−0.05	−0.01	[−0.16, 0.14]	0.00	−0.04	[−0.11, 0.02]	−0.04
BF – IR	0.04	[−0.04, 0.12]	0.03	0.02	[−0.12, 0.16]	0.01	0.05	[−0.02, 0.11]	0.05

Note. Results are from models with covariates, but covariates are omitted from the table to reduce length. The False Discovery Rate (FDR) correction was applied to all analyses.

CI = confidence interval. b* = standardized estimate. Bold values indicate significance at FDR-corrected p < .05. SR = self-report. IR = informant-report. BF = bifactor model. BF – G = general trait factor (i.e., shared trait variance from the self- and informant-report methods). BF – SR = self-report trait factor; captures variance unique to the self-report method and not captured in the shared trait factor. BF – IR = informant-report trait factor; captures variance unique to the informant-report method and not captured in the shared trait factor. IL-6 = interleukin-6. CRP = C-reactive protein. TNF-α = tumor necrosis factor – alpha.

Conscientiousness was one of two traits showing robust associations with the biomarkers (Table 2). Self-reported conscientiousness predicted lower levels of IL-6 (b = −0.08, 95% CI [−0.14, −0.03]), whereas informant-reports predicted lower levels of IL-6 (b = −0.08, 95% CI [−0.12, −0.03]) and CRP (b = −0.15, 95% CI [−0.22, −0.07]). Bifactor models, which partition the variance of self- and informant-reports into three latent factors – a general factor that is shared among self and informant, unique variance for self, and unique variance of informant – were fit next. IL-6 was predicted by the general trait factor (b = −0.11, 95% CI [−0.17, −0.05]) and CRP by the general trait factor (b = −0.16, 95% CI [−0.29, −0.04]) and bifactor informant-report (b = −0.45, 95% CI [−0.87, −0.03]). These results suggest the variance that is exclusive to informant-reports of conscientiousness has unique predictive validity above and beyond shared associations with self-reports for predicting CRP, while the shared variance of self- and informant-reports predicts IL-6 and CRP.

The other trait predicting future biomarker levels was neuroticism (Table 2). Self-report neuroticism only predicted IL-6 (b = 0.07, 95% CI [0.02, 0.11]) whereas informant-reports predicted greater levels of IL-6 (b = 0.05, 95% CI [0.02, 0.09]) and CRP (b = 0.10, 95% CI [0.04, 0.17]). After controlling for multiple comparisons, there were no remaining effects for any biomarker in the bifactor models for neuroticism.

The remaining Big Five traits as well as healthy neuroticism (i.e., the interaction of conscientiousness and neuroticism) showed no consistent effects (see Tables S3–S8). When controlling for baseline BMI, results were consistent with above findings (Table S9), highlighting the robustness of the personality-inflammation link for conscientiousness and neuroticism.

Are self- & informant-reports of personality associated with indicators of physical health?

We then tested what the direct associations were with our personality variables and each of the three indicators of health variables (Table 3). For PHRQoL, associations emerged for self-report extraversion (b = 0.50, 95% CI [0.25, 0.76]), conscientiousness (b = 0.65, 95% CI [0.40, 0.89]), and neuroticism (b = −0.53, 95% CI [−0.74, −0.32]). Additionally, from the bifactor models, the bifactor self-report for extraversion was uniquely related to PHRQoL (b = 0.62, 95% CI [0.24, 1.01]), suggesting that above and beyond the shared variance of self- and informant-reports for extraversion, the variance captured by a person’s own perception of themselves predicts PHRQoL. For BMI, associations emerged for conscientiousness and neuroticism. For conscientiousness, self-reports (b = −0.34, 95% CI [−0.58, −0.10]), informant-reports (b = −0.37, 95% CI [−0.56, −0.18]), and the general trait factor from the bifactor model (b = −0.88, 95% CI [−1.27, −0.50]) were associated with BMI. Based on the magnitude of the estimates, that variance that is shared between self- and informant-reports of conscientiousness (i.e., the general trait variance) best predicted BMI, but self- and informant-reports each provided incremental validity over and above this shared variance. For neuroticism, informant-reports (b = 0.25, 95% CI [0.09, 0.41]) and the general trait factor (b = 0.56 95% CI [0.25, 0.87) were associated with BMI. Lastly, for CDB, only associations for conscientiousness and neuroticism emerged. Self-reports for both conscientiousness (b = −0.24, 95% CI [−0.40, −0.08]) and neuroticism (b = 0.35, 95% CI [0.21, 0.48]) were associated with CDB, as well as the bifactor self-report factor for neuroticism (b = 0.35, 95% CI [0.13, 0.56]). See Tables S11–S16 for estimates with covariates as well.

Table 3.

Estimates of Baseline Personality Predicting Indicators of Physical Health Outcome Variables

	Physical Health-Related QoL			BMI (standardized)			Chronic Disease Burden
Model	b	95% CI	b*	b	95% CI	b*	b	95% CI	b*
Extraversion
SR	0.50	[0.25, 0.76]	0.19	−0.05	[−0.28, 0.18]	−0.02	−0.09	[−0.25, 0.07]	−0.03
IR	0.16	[−0.04, 0.36]	0.08	0.03	[−0.16, 0.23]	0.02	−0.03	[−0.15, 0.10]	−0.01
BF – G	0.27	[−0.02, 0.56]	0.09	−0.11	[−0.43, 0.21]	−0.03	−0.11	[−0.29, 0.07]	−0.03
BF – SR	0.62	[0.24, 1.01]	0.17	−0.07	[−0.43, 0.29]	−0.02	1.17	[−0.38, 2.72]	0.10
BF – IR	−0.73	[−1.86, 0.39]	−0.09	0.55	[0.02, 1.08]	0.10	0.06	[−0.13, 0.25]	0.02
Agreeableness
SR	0.21	[−0.14, 0.55]	0.06	−0.22	[−0.54, 0.09]	−0.07	−0.25	[−0.46, −0.04]	−0.06
IR	0.00	[−0.19, 0.20]	0.00	−0.09	[−0.27, 0.09]	−0.04	0.01	[−0.12, 0.14]	0.00
BF – G	0.19	[−0.16, 0.55]	0.05	−0.24	[−0.56, 0.08]	−0.06	−0.20	[−0.45, 0.05]	−0.04
BF – SR	0.20	[−0.22, 0.62]	0.03	−0.20	[−0.56, 0.16]	−0.04	−0.35	[−0.69, −0.02]	−0.05
BF – IR	−0.05	[−0.27, 0.17]	−0.02	−0.04	[−0.24, 0.17]	−0.01	0.05	[−0.11, 0.21]	0.02
Conscientiousness
SR	0.65	[0.40, 0.89]	0.25	−0.34	[−0.58, −0.10]	−0.13	−0.24	[−0.40, −0.08]	−0.08
IR	0.13	[−0.08, 0.34]	0.06	−0.37	[−0.56, −0.18]	−0.19	−0.15	[−0.27, −0.02]	−0.06
BF – G	0.64	[−0.05, 1.34]	0.19	−0.88	[−1.27, −0.50]	−0.24	−0.19	[−0.57, 0.20]	−0.05
BF – SR	0.90	[−0.02, 1.82]	0.17	−0.14	[−0.45, 0.17]	−0.04	−0.40	[−1.13, 0.34]	−0.07
BF – IR	−0.12	[−0.44, 0.20]	−0.05	0.01	[−0.29, 0.31]	0.00	−0.16	[−0.43, 0.12]	−0.05
Neuroticism
SR	−0.53	[−0.74, −0.32]	−0.24	0.25	[0.05, 0.44]	0.11	0.35	[0.21, 0.48]	0.13
IR	−0.16	[−0.33, 0.01]	−0.09	0.25	[0.09, 0.41]	0.15	0.13	[0.02, 0.23]	0.06
BF – G	−3.81	[−15.80, 8.18]	−1.29	0.56	[0.25, 0.87]	0.20	0.28	[−0.01, 0.57]	0.08
BF – SR	3.94	[−11.24, 19.12]	1.20	0.08	[−0.18, 0.33]	0.03	0.35	[0.13, 0.56]	0.10
BF – IR	2.04	[−5.13, 9.20]	0.96	−2.23	[−5.43, 0.96]	−0.22	−0.15	[−1.16, 0.87]	−0.02
Openness
SR	−0.03	[−0.31, 0.25]	−0.01	−0.12	[−0.37, 0.12]	−0.05	0.11	[−0.06, 0.28]	0.03
IR	−0.09	[−0.34, 0.17]	−0.04	−0.14	[−0.37, 0.10]	−0.06	0.15	[−0.00, 0.31]	0.05
BF – G	−0.09	[−0.40, 0.21]	−0.03	−0.04	[−0.33, 0.24]	−0.01	0.21	[0.03, 0.40]	0.06
BF – SR	0.04	[−0.35, 0.43]	0.01	−0.04	[−0.41, 0.33]	−0.01	0.04	[−0.20, 0.29]	0.01
BF – IR	−0.14	[−0.50, 0.22]	−0.03	−0.02	[−0.37, 0.32]	−0.01	0.19	[−0.04, 0.42]	0.04

Note. Results are from models with covariates, but covariates are omitted from the table to reduce length. The False Discovery Rate (FDR) correction was applied to all analyses.

CI = confidence interval. b* = standardized estimate. Bold values indicate significance at FDR-corrected p < .05. SR = self-report. IR = informant-report. BF = bifactor model. BF – G = general trait factor (i.e., shared trait variance from the self- and informant-report methods). BF – SR = self-report trait factor; captures variance unique to the self-report method and not captured in the shared trait factor. BF – IR = informant-report trait factor; captures variance unique to the informant-report method and not captured in the shared trait factor.

Does inflammation mediate the relationship of personality predicting health outcomes?

Next, Aim 2 investigated if baseline personality predicting biomarker levels during the second assessment mediated the relationship of personality predicting PHRQoL, BMI, and CDB from the final measurement occasion. Across all models, IL-6 and CRP were directly associated with future PHRQoL, BMI, and CDB while TNF-α never showed this association. Here, we focus on the significant indirect effects for predicting our three indicators of physical health. Full results, including all model-provided unstandardized and standardized estimates for each Big Five trait and covariates, can be found in Table S17 for indirect effects for all Big Five traits; Tables S18–S22 for PHRQoL; Tables S23–S27 for BMI; and Tables S28–S32 for CDB.

For PHRQoL, mainly effects for conscientiousness remained after correcting for multiple comparisons (Table S20). IL-6 mediated its relationship with PHRQoL for informant-reports (b = 0.06, 95% CI [0.01, 0.10]) and the general trait factor (b = 0.15, 95% CI [0.03, 0.26]). CRP likewise mediated its relationship with PHRQoL for informant-reports (b = 0.05, 95% CI [0.01, 0.09]). One effect for neuroticism emerged: IL-6 mediated the relationship between its general trait factor and PHRQoL (b = −0.13, 95% CI [−0.24, −0.03]).

For BMI, effects emerged for agreeableness, conscientiousness, and neuroticism (Tables S24–S26). CRP mediated the relationship between the general trait factor for agreeableness and BMI (b = −0.19, 95% CI [−0.35, −0.04]). For conscientiousness, IL-6 mediated its relationship with BMI for self-reports (b = −0.09, 95% CI [−0.16, −0.02]), informant-reports (b = −0.08, 95% CI [−0.14, −0.03]), and the general trait factor (b = −0.23, 95% CI [−0.38, −0.08]). Additionally, CRP mediated its relationship with BMI for informant-reports (b = −0.12, 95% CI [−0.18, −0.05]), the general trait factor (b = −0.30, 95% CI [−0.47, −0.12]), and the bifactor informant-report (b = −0.14, 95% CI [−0.23, −0.05]). For neuroticism, IL-6 mediated the relationship between neuroticism and BMI for both self- (b = 0.07, 95% CI [0.02, 0.13]) and informant-reports (b = 0.06, 95% CI [0.01, 0.10]). CRP mediated the relationship for informant-reports only (b = 0.08, 95% CI [0.03, 0.13]).

Lastly, for CDB, effects were found for conscientiousness and neuroticism and nearly mirrored those for BMI (Tables S30–S31). For conscientiousness, IL-6 mediated its relationship with CDB for self-reports (b = −0.10, 95% CI [−0.17, −0.03]), informant-reports (b = −0.09, 95% CI [−0.15, −0.04]), and the general trait factor (b = −0.11, 95% CI [−0.18, −0.04]). CRP mediated its relationship with CDB for informant-reports (b = −0.09, 95% CI [−0.14, −0.04]) and the general trait factor (b = −0.23, 95% CI [−0.37, −0.09]). For neuroticism, IL-6 mediated the relationship between neuroticism and CDB for both self- (b = 0.08, 95% CI [0.03, 0.14]) and informant-reports (b = 0.06, 95% CI [0.02, 0.10]) as well as the general trait factor (b = 0.14, 95% CI [0.03, 0.25]). CRP mediated the relationship for informant-reports only (b = 0.06, 95% CI [0.02, 0.10]).

Are inflammatory biomarkers and health behaviors independent pathways?

The final aim investigated whether the inflammation pathway is independent of a health behaviors pathway. Health behaviors were modestly correlated with IL-6 (r = −.14 95% CI [−.21, −.06]), CRP (r = −.04 95% CI [−.11, −.03]), and TNF-α (r = −.08 95% CI [−.16, .00]). For those traits that predicted indicators of future physical health, health behaviors and biomarkers were found to often constitute separate, independent pathways in their personality-health relationship (Table 4). Furthermore, the overall number of indirect paths through health behaviors far outnumbered those through biomarkers. Specifically, for PHRQoL, 82% of indirect effects that emerged were exclusively through the health behavior path, 2% of effects were through the biomarker pathway only, and 16% occurred simultaneously (i.e., emerged in the same model). For BMI, almost 8% of indirect effects were for the health behavior path only, 77% of effects were exclusively through biomarkers, and 15% of effects co-occurred. For CDB, a third of effects were through health behaviors only, a third were through biomarkers only, and a third occurred simultaneously. Thus, overall, 59% of indirect effects were through health behaviors only, 21% were through biomarkers only, and 20% co-occurred. Here, we report representative unstandardized results from the traits with significant indirect effects through both biomarkers and/or health behaviors. Full results, including all model-provided unstandardized and standardized estimates for each Big Five trait and covariates, can be found in Tables S33–S38 for PHRQoL; Tables S39–S44 for BMI; and Tables S45–S50 for CDB.

Table 4.

Aim 3: Estimates for Effects of Inflammatory Biomarkers and Health Behaviors Mediating the Personality to Physical Health-Related Quality of Life, BMI, and Chronic Disease Burden Pathway

		Indirect Effect: HBs				Indirect Effect: BMs
Model	Predictor	b	95% CI	b*	% of Effect	b	95% CI	b*	% of Effect
Physical Health-Related Quality of Life
Conscientiousness
IL-6	IR Trait	0.05	[0.02, 0.09]	0.03	23	0.05	[0.01, 0.09]	0.03	23
	BF-G Trait	0.19	[0.05, 0.33]	0.06	26	0.11	[0.03, 0.20]	0.03	15
CRP	IR Trait	0.06	[0.02, 0.10]	0.03	29	0.05	[0.01, 0.09]	0.02	24
	BF-G Trait	0.23	[0.09, 0.36]	0.07	33	0.06	[−0.00, 0.13]	0.02	9
	BF-SR Trait	0.24	[0.04, 0.43]	0.06	20	−0.01	[−0.06, 0.05]	0.00	1
	BF-IR Trait	0.00	[−0.04, 0.04]	0.00	0	0.07	[0.01, 0.12]	0.03	70
Neuroticism
IL-6	SR Trait	−0.07	[−0.12, −0.02]	−0.03	12	−0.04	[−0.08, −0.01]	−0.02	7
	IR Trait	−0.04	[−0.06, −0.01]	−0.02	16	−0.04	[−0.07, −0.00]	−0.02	16
	BF-G Trait	−0.14	[−0.24, −0.05]	−0.04	22	−0.12	[−0.21, −0.03]	−0.04	19
	BF-SR Trait	−0.07	[−0.12, −0.02]	−0.03	11	−0.04	[−0.08, 0.00]	−0.01	6
CRP	SR Trait	−0.07	[−0.13, −0.02]	−0.03	12	−0.02	[−0.05, 0.00]	−0.01	3
	IR Trait	−0.04	[−0.07, −0.01]	−0.02	16	−0.03	[−0.06, −0.00]	−0.02	12
	BF-G Trait	−0.16	[−0.26, −0.06]	−0.04	25	−0.09	[−0.17, −0.01]	−0.03	14
	BF-SR Trait	−0.08	[−0.13, −0.02]	−0.03	12	−0.01	[−0.04, 0.02]	0.00	2
BMI
Conscientiousness
IL-6	SR Trait	0.12	[0.04, 0.21]	0.05	19	−0.10	[−0.17, −0.03]	−0.04	16
	IR Trait	0.03	[0.00, 0.06]	0.02	7	−0.09	[−0.14, −0.03]	−0.04	20
	BF-G Trait	0.14	[0.01, 0.27]	0.04	19	−0.21	[−0.33, −0.08]	−0.06	29
CRP	SR Trait	0.11	[0.03, 0.20]	0.04	20	−0.06	[−0.13, 0.02]	−0.02	11
	IR Trait	0.03	[−0.00, 0.06]	0.01	7	−0.12	[−0.18, −0.05]	−0.06	27
	BF-G Trait	0.11	[−0.01, 0.23]	0.03	21	−0.18	[−0.31, −0.05]	−0.05	34
	BF-IR Trait	0.00	[−0.01, 0.01]	0.00	0	−0.14	[−0.23, −0.05]	−0.06	28
Chronic Disease Burden
Conscientiousness
Il-6	SR Trait	0.12	[0.05, 0.20]	0.04	23	−0.11	[−0.18, −0.03]	−0.03	21
	IR Trait	0.03	[0.01, 0.06]	0.01	13	−0.09	[−0.15, −0.04]	−0.04	39
	BF-G Trait	0.08	[−0.02, 0.17]	0.02	15	−0.23	[−0.35, −0.11]	−0.06	44
CRP	SR Trait	0.09	[0.02, 0.16]	0.03	19	−0.04	[−0.10, 0.02]	−0.01	8
	IR Trait	0.02	[−0.00, 0.05]	0.01	8	−0.09	[−0.14, −0.04]	−0.04	38
	BF-G Trait	0.02	[−0.07, 0.11]	0.00	5	−0.14	[−0.24, −0.04]	−0.04	38
Neuroticism
IL-6	SR Trait	−0.06	[−0.10, −0.02]	−0.02	12	0.09	[0.03, 0.14]	0.03	18
	IR Trait	−0.02	[−0.04, −0.00]	−0.01	10	0.06	[0.02, 0.11]	0.03	30
	BF-G Trait	0.08	[−0.03, 0.18]	0.02	12	0.14	[0.03, 0.25]	0.04	21
	IR Trait	−0.01	[−0.03, 0.00]	−0.01	6	0.06	[0.02, 0.10]	0.03	33

Note. Results are from models with covariates, but covariates are omitted from the table to reduce length. The False Discovery Rate (FDR) correction was applied to all analyses. The % of Effect represents the proportion of the indirect effect relative to the absolute value of the total effect.

b* = standardized estimate. Bold values indicate significance at FDR-corrected p < .05. SR = self-report. IR = informant-report. BF-G = general factor from bifactor model. BF – SR/IR = unique report trait factor; captures variance unique to that report method not captured in the general trait factor. BMI is standardized.

For conscientiousness predicting PHRQoL, informant reports appeared most consequential for inflammatory biomarkers. Both IL-6 and CRP mediated the relationship between informant-reports. However, when bifactor models were used, that variance was shared with self-report for IL-6, whereas for CRP informants provided unique information related to PHRQoL relative to self (Table 4). For neuroticism, health behaviors played a more important role than biomarkers. While self-reports and informant-reports were associated, these findings were driven by both the general trait and the self, as seen through bifactor models, rather than unique informant information.

For BMI, conscientiousness evidenced an effect through primarily biomarkers. Again, informant-reports played an important role. When using bifactor models, the effects were shared with self-reports for IL-6 but separate for CRP.

Lastly, for CDB, effects through both biomarkers and health behaviors were found for conscientiousness and neuroticism. For conscientiousness, self and informant effects were primarily redundant with one another, as the general factor evidenced an effect for both IL-6 and CRP. For neuroticism, self-rated and informant-rated effects emerged, though according to the bifactor models, they were due to shared variance.

Discussion

This study examined the association between personality from multiple perspectives and inflammatory biomarkers using a large sample of older adults. Three main findings emerged: first, we found that self-reports, informant-reports, and shared variance between these reports provide unique and non-redundant associations with inflammatory markers – indicating that multiple assessment methods are useful above and beyond single assessment methods. Second, inflammatory markers longitudinally mediated the association between all reports of personality and PHRQoL, BMI, and CDB across a 10-year period. Third, these pathways were often separate from engagement in health behaviors, suggesting that the reason personality influences later health occurs through multiple pathways. As expected, these associations mainly occurred for the traits conscientiousness and neuroticism and the biomarkers IL-6 and CRP.

Which traits and report methods predict future levels of inflammatory biomarkers?

Largely consistent with past research (e.g., O’Súilleabháin et al., 2021; Sutin et al., 2014), conscientiousness and neuroticism were the traits showing associations with the biomarkers. Overall, both self- and informant-report traits had similar associations with inflammatory markers. although informant-reports evidenced additional associations in some cases. Furthermore, the shared trait factor, specifically for conscientiousness, routinely had effects where self- and informant-reports were also significant – strengthening the claim that conscientiousness, as perceived by both the self and other, is related to inflammation. While informant-reported personality is often more strongly related to health outcomes than self-report (Jackson et al., 2015; Smith et al., 2008), it was unclear whether these were due to unique or shared attributes. The bifactor models separate these components, demonstrating the unique perspectives offered by self and informants (Chen et al., 2011; Brunner et al., 2011).

There are various reasons why self or informant reports of personality may provide better predictive utility, such as the degree of observability of a trait or biases that are often present in self-report measures (e.g., inaccurate self-knowledge, self-presentation; Vazire, 2010; Smith et al., 2008). For example, self-reports might be biased by a participant not wanting to appear as irresponsible or moody – markers used to assess undesirable levels of conscientiousness and neuroticism, whereas informants may be more comfortable rating these attributes. Indeed, informant-reports of personality have also been shown to be advantageous over self-reports in some cases for predicting objective health outcomes such as mortality (Jackson et al., 2015). Overall, these findings, in addition to past literature, further highlight the need for and benefits from incorporating multiple assessment methods of personality.

Inflammatory biomarkers and health behaviors as explanatory processes

Among the Big Five traits related to our three health variables, inflammation only mediated the personality-physical health pathway for conscientiousness and neuroticism. There are a few reasons why these effects would emerge. Inflammation has been linked to consequences of heightened stress responses, which could result from a tendency to perceive situations as threatening or routine exposure to stressors (Stewart et al., 2009; Miller et al., 2009). Conscientiousness is associated with physiological reactivity, stress response, and exposure to environmental stressors (e.g., those that are career- or income-related; Zobel et al., 2004; Sutin et al., 2009); while neuroticism is associated with a person’s negative emotion reactivity, HPA axis activation, and responses to environmental stressors (e.g., interpersonal issues; Stewart et al., 2009; Miller et al., 2009). Additionally, people with maladaptive personality characteristics, qualities often most related to neuroticism, can lead to stress exposure through their behavior or situation selection (Conway et al., 2018).

When looking at the personality-biomarker-health pathway in our study, indirect effects were found for IL-6 and CRP whereas TNF-α failed to show any associations. Past research has more repeatedly linked IL-6 and CRP to a broader variety of both personality traits and health outcomes while TNF-α is typically more limited in its associations, usually to links with agreeableness (Denollet et al., 2008; Petersen & Felker, 2006). Also, IL-6 and CRP often function concurrently in the immune system, as IL-6 signals the synthesis of CRP and they are involved in similar inflammatory responses, while TNF-α often fulfills different roles and is involved in separate cell-signaling mechanisms (Sesso et al., 2007). In our study, the moderate correlation between IL-6 and CRP further supports the linked nature of these inflammatory markers.

Distinct, independent indirect effects were found for inflammatory markers and health behaviors, with both offering different reasons as to why personality is prospectively related to physical health. Establishing independence is important insomuch as to demonstrate that health behaviors are not the only pathway to poor health. Current theories about why personality is associated with biomarkers often revolve around behaviors to reduce infections and stress and avoiding acquiring health outcomes linked to increased levels of inflammatory markers (e.g., Hamer et al., 2012). But, based on the current findings, inflammation-associated processes are not captured under standard health behavior indices. Indirect pathways through health behaviors emerged more frequently overall; although, this was not the case for BMI and CBD. For BMI, exclusive biomarkers effects constituted majority of effects while the proportion of effects for biomarkers and health behaviors was equivalent for CDB. The reason for the high proportion of health behavior effects for PHRQoL relative to the other two outcomes could, in part, be due to the shared method variance. However, those effects that are evidenced by informant-reports or the general trait factors suggest that, although the shared method variance might lead to more frequent effects for self-reports, these effects in general occur in instances where shared method variance is not as concerning of a confounding factor.

Aside from statistical significance, examining the practical significance of these effects offers another helpful perspective on their utility. For the indirect effects, two effect sizes were presented: the percentage of the absolute value of total effect their magnitude constituted and the standardized regression coefficient. Previous investigations of effect sizes in health outcomes, particularly with multimodal assessments, have found that the significant effects explained around 1% of variance in the outcomes (Miller et al., 2016). Small effects (which are not actually “small” in current literature; Funder & Ozer, 2019) have important downstream consequences (Götz et al., 2021), especially for clinically relevant or health-related outcomes and consistent attributes such as personality traits. That is, across the lifespan, these effects can accumulate in impact over time and have great potential to influence the health process. Thus, collectively, for even those effects that are small in percentage of the total effect and/or in magnitude for the standardized regression coefficients, their practical implications for an individual’s life are likely not negligible.

Limitations

While longitudinal, we did not have multiple measures of our study variables, which would have allowed for a stricter test of our pathways. Therefore, the current directions of influence implied in our study, although separated in time, might not reflect the true mediational processes. For instance, health behaviors or biomarkers could first influence personality, or they could reciprocally influence one another across time (e.g., McClendon, 2020). Additionally, while our health outcomes are longitudinally separated in time from personality, health behaviors, and inflammatory markers, it does not mean that we can separate these processes temporally.

Next, we only measured normal-range levels of personality traits and the Big Five traits; past research showing associations with maladaptive personality and stress (e.g., Gleason et al., 2014) and other trait models such as the interpersonal circumplex with mortality (e.g., Chapman et al., 2020) suggests measuring this form of personality could offer valuable insight. Also related to personality, half of the informants were spouses. While spouses presumably know their partners quite well, past research examining objective health outcomes find associations with friends as well (Jackson et al., 2015). Thus, while the variety in type of informants suggests the effects are more generalizable compared to a single informant type, it would be useful in the future to examine how informant type impacts predictive validity. As a final note on personality, it is an assumption that the self- and informant-report factors from the bifactor models still reflect some “true” trait variance for their respective traits (i.e., not purely method variance). Although this assumption cannot be strictly tested in our study, the associations that occurred were for the expected traits and in the expected directions, speaking to the underlying trait variance likely driving these associations. However, future work should aim to confirm these assumptions.

Further, there were limitations regarding the health behavior variable. First, to be broad and objective, we included all the items in our measure, even those that might not be directly theoretically relevant. This point especially could explain the lack of indirect effects through health behaviors for predicting BMI, as health behaviors such as looking both ways before crossing a street and wearing a seatbelt are unlikely to be directly related to BMI but were nevertheless included in our health behavior measure. Second, instead of broadly categorizing health behaviors, examining narrower dimensions of protective and risk behaviors could even more so explain how personality-health associations arise (Lodi-Smith et al., 2010).

Lastly, while the participants in our sample were demographically representative of the St. Louis area, those who chose to participate may systematically differ from those who did. For example, individuals who view themselves as relatively healthy might be more willing to participate in a study collecting health data than individuals who view themselves as less healthy. Reasons for lack of participation in the study are unknown, thus it must be kept in mind this sample could consist of adults who are healthier than what is typical of the St. Louis area when attempting to generalize these results to other samples. However, reluctance to participate in research could stem from many factors, including increases in telemarketing and other research requests. The sampling procedure was conservative (i.e., if one person in a household was not willing to participate, no one in that household was eligible, even if they were willing), suggesting there is less of a role for individual differences to determine study participation. Nevertheless, these factors should be considered when interpreting and/or generalizing results.

Conclusion

This study provided the first examination of whether multiple methods of personality are uniquely associated with inflammatory biomarkers. Findings indicate that multiple perspectives are important in understanding the personality-physical health process. Moreover, findings supported that personality is associated with three indices of physical health indirectly through biomarkers and health behaviors, with these two pathways operating independently. Future research that might otherwise only focus on health behaviors as a mediating mechanism could benefit from more often incorporating physiological indicators of health, such as inflammatory biomarkers, to understand the personality-health pathway more thoroughly.