Neuroticism is associated with future disease and mortality risks

Authors: Shuyi Huang, Yaru Zhang, Lingzhi Ma, Bangsheng Wu, Jianfeng Feng, Wei Cheng, Jintai Yu

Abstract

Background:

Neuroticism has been associated with numerous health outcomes. However, most research has focused on a single specific disorder and has produced controversial results, particularly regarding mortality risk. Here, we aimed to examine the association of neuroticism with morbidity and mortality and to elucidate how neuroticism affects trajectories from a healthy state, to one or more neuroticism-related disorders, and subsequent mortality risk.

Methods:

We included 483,916 participants from the UK Biobank at baseline (2006–2010). Neuroticism was measured using the Eysenck Personality Questionnaire. Three clusters were constructed, including worry, depressed affect, and sensitivity to environmental stress and adversity (SESA). Cox proportional hazards regression and multistate models were used. Linear regression was used to examine the association between neuroticism and immune parameters and neuroimaging measures.

Results:

High neuroticism was associated with 37 non-overlapping diseases, including increased risk of infectious, cardiometabolic, neuropsychiatric, digestive, and respiratory diseases, and decreased risk of cancer. After adjustment for sociodemographic variables, physical measures, healthy behaviors, and baseline diagnoses, moderate-to-high neuroticism was associated with a decreased risk of all-cause mortality. In multistate models, high neuroticism was associated with an increased risk of transitions from a healthy state to a first neuroticism-related disease (hazard ratio [HR] [95% confidence interval (CI)] = 1.09 [1.05–1.13], P <0.001) and subsequent transitions to multimorbidity (1.08 [1.02–1.14], P = 0.005), but was associated with a decreased risk of transitions from multimorbidity to death (0.90 [0.84–0.97], P for trend = 0.006). The leading neuroticism cluster showing a detrimental role in the health–illness transition was depressed affect, which correlated with higher amygdala volume and lower insula volume. The protective effect of neuroticism against mortality was mainly contributed by the SESA cluster, which, unlike the other two clusters, did not affect the balance between innate and adaptive immunity.

Conclusion:

This study provides new insights into the differential role of neuroticism in health outcomes and into new perspectives for establishing mortality prevention programs for patients with multimorbidity.

Introduction

Neuroticism, indicating susceptibility to psychological distress, is a significant public health concern and economic burden,^[1,2]^ correlating robustly with various mental and physical disorders and their comorbidities.^[1]^ Most research, however, has focused on the relationship between neuroticism and a single specific disorder or health outcome.^[345]^ The most contentious issue is the link between neuroticism and mortality. Several studies have observed an increased risk of mortality in individuals with higher levels of neuroticism.^[6,7]^ However, high degrees of neuroticism may also have a salutary influence on health.^[8,9]^ A previous study suggested that the relationship between neuroticism and mortality was influenced by the self-rated health status.^[10]^ After the onset of a major chronic illness, high neuroticism combined with high conscientiousness predicted positive responses such as less smoking.^[11]^ We speculate that the effects of neuroticism may vary under different health conditions. Since conflicting evidence may lead to misunderstandings about the neuroticism–health relationship, resulting in suboptimal allocation of intervention resources,^[12]^ it is necessary to clarify the relationship between neuroticism and health-related outcomes using multistate illness–death models based on objective health data.

Neuroticism, similar to other personality factors, has a hierarchical structure.^[13]^ The hierarchical clustering analysis was applied to genetic correlations among 12 neuroticism items of the Eysenck Personality Questionnaire-Revised Short Form (EPQ-RS).^[14]^ Three genetically distinct clusters were constructed corresponding to specific aspects of depressed affected, worry, and sensitivity to environmental stress and adversity (SESA).^[15,16]^ Researchers have suggested that biological understanding can be gained by studying neuroticism in genetically more homogeneous clusters.^[14]^ Nagel et al^[14]^ have found distinct and sometimes opposing genetic correlation patterns for neuroticism’s worry and depressed affect clusters. A prior study also indicated differing mortality associations with specific neuroticism aspects.^[10]^ Thus, cluster-specific analyses may further shed light on which subclusters of neuroticism are deleteriously or protectively associated with health.

Researchers have proposed that the relationship between personality and health may be mediated by the activities of the immune system.^[17]^ However, the evidence for associations between neuroticism and inflammatory markers is mixed.^[18192021]^ There is a need to examine specific aspects of neuroticism and multiple measures of immune function to provide new insights. A simple and generalizable method to qualify the components of and balance between innate and adaptive immunity is the peripheral blood cell count of different leukocytes and their ratio,^[22,23]^ which are thought to be elevated in individuals with cardiovascular disease (CVD) or cancer and have been linked to morbidity and mortality in previous studies.^[22]^ Moreover, given that personality traits are linked to brain structural measures that have been linked to health outcomes,^[242526]^ examining neuroticism-related brain structural changes may help to understand the underlying mechanisms of neuroticism.

In the present study, we aimed to examine the association of neuroticism with morbidity and mortality, to elucidate how neuroticism affects the trajectories from a healthy state, to one or more neuroticism-related diseases, and the subsequent risk of mortality by using multistate illness–death models. We also examined the association of neuroticism with immune parameters and brain structures.

Methods

Study population

This study is based on the data from the UK Biobank (UKB) study, which has been approved by the National Information Governance Board for Health and Social Care and the National Health Service North West Multicenter Research Ethics Committee (No. 21/NW/0157) (https://www.ukbiobank.ac.uk/learn-more-about-uk-biobank/about-us/ethics) and provided oversight for this study. UKB is a population-based cohort of more than 500,000 participants aged 37–73 years recruited across the United Kingdom between 2006 and 2010.^[27]^ All the participants provided electronic informed consent at the baseline assessment. After excluding the participants without available neuroticism data, the participants with a history of psychiatric or personality disorders [Supplementary Table 1, http://links.lww.com/CM9/C338], and the participants who died within the first five years of follow-up, 483,916 participants remained eligible for the main analysis [Supplementary Figure 1, http://links.lww.com/CM9/C338]. The dropout rate is acceptable (10,675/494,590, 2.16%). The sample size is sufficient to detect a hazard ratio (HR) of 1.2 with at least 80% power. Analyses were conducted under UKB application No.19542.

Neuroticism and neuroticism clusters

Neuroticism (n = 393,067) was measured at the baseline with the 12-item EPQ-RS using a touchscreen-based questionnaire at the UKB assessment centers (UKB Field 20127). According to the previously established subsets of genetically homogeneous neuroticism items through the hierarchical clustering analysis of the genetic correlations among the 12 EPQ-RS items, we constructed three neuroticism cluster scores including worry, depressed affect, and SESA [Supplementary Table 2, http://links.lww.com/CM9/C338].^[15,16]^ For each neuroticism cluster, only participants with complete scores on all items were included, resulting in 441,844, 458,060, and 449,542 for worry, depressed affect, and SESA, respectively [Supplementary Figure 1, http://links.lww.com/CM9/C338]. Neuroticism (≤2, 3–5, ≥6) and its three clusters, worry (0, 1, or ≥2), depressed affect (0, 1–2, or ≥3), and SESA (≤1, 2, or 3), were grouped into tertiles.

Morbidity and mortality outcomes

Diagnoses were ascertained from hospital inpatient records, primary care data, death registry data, and self-reported medical data. Diseases were coded according to the International Classification of Diseases, 10th Revision (ICD-10) system. We focused on a predefined list of 78 common ICD-10 disease chapters and diagnostic groups constructed for outcome-wide studies by a previously published article.^[28]^ Participants were considered at risk for each diagnostic group from the date of enrollment at the recruitment center and were followed up until the date of first diagnosis, death, loss to follow-up, or December 31, 2021, whichever came first.

In the UKB, death notifications (age at death and primary ICD-10 diagnosis associated with death) were obtained through linkage to national death registries. Regarding the mortality outcome, the end of follow-up was defined as the date of death, the date of loss to follow-up, or December 31, 2021, whichever came first. Specific causes of death (including both primary and contributory causes of death) were defined by the following ICD-10 cancer (C00–C97), ischemic heart disease (I20–I25), cerebrovascular disease (I60–I69), and respiratory diseases (J00–J99).

Covariates and immune parameters

Covariates included age, sex, body mass index (BMI), ethnicity, education, Townsend deprivation index (TDI),^[29]^ physical activity, alcohol, smoking status, baseline diagnoses of hypertension, high cholesterol, chronic kidney disease (CKD), diabetes, cardiovascular disease (CVD), and cancer. Immune parameters included the counts of neutrophils, monocytes, platelets, lymphocytes, blood levels of C-reactive protein (CRP), neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), systemic immune-inflammation index (SII) (neutrophils × platelets/lymphocytes), and lymphocyte-to-monocyte ratio (LMR).^[30]^ The detailed definition and explanation of covariates and immune parameters are shown in the Supplementary Materials, http://links.lww.com/CM9/C338.

Statistical analysis

Baseline group differences were assessed by chi-squared or Kruskal–Wallis test, depending on the characteristics of the data. For prospective analysis, we used the Cox proportional hazards regression model. Proportional hazards were tested using the scaled Schoenfeld’s residuals. For each regression of the 78 health outcomes, participants with the corresponding diagnosis at baseline were excluded. To examine neuroticism in relation to the co-occurrence of neuroticism-related diseases, we constructed the outcome of multimorbidity as the presence of two or more non-overlapping neuroticism-related diseases that survived false discovery rate (FDR) corrections. Data were censored at the date of diagnosis of multimorbidity, date of death, the date of loss to follow-up, or December 31, 2021, whichever came first. The main analysis was conducted in the participants free of multimorbidity at baseline. Sensitivity analyses were conducted in healthy participants (free of any of the 78 common diseases). In the analysis of associations with morbidity and multimorbidity, models were adjusted for age, sex, BMI, ethnicity, education, TDI, physical activity, alcohol, and smoking status. In the analysis of associations with mortality, the model started with a minimally adjusted model including only age and sex as covariates (Model 1), and a second model adjusted for age, sex, ethnicity, education, TDI, BMI, physical activity, alcohol, and smoking status (Model 2). Model 3 was further adjusted for baseline diagnoses including hypertension, high cholesterol, CKD, diabetes, CVD, and cancer in addition to the covariates in Model 2. Because the proportional hazards assumption was not met for age and BMI, both the BMI and age categories were included as strata variables in Cox models to allow for a different hazard function in each group.^[31]^ Other covariates were assumed to have the same effect across strata.

Multistate models were used to determine the association of neuroticism with transitions (1) a healthy state (free of any of the 78 common diseases) to a first neuroticism-related disease (any from the predefined list); (2) a healthy state to death (in those who remained healthy during follow-up); (3) a first neuroticism-related disease to multimorbidity; (4) a first neuroticism-related disease to death; and (5) multimorbidity to death. For participants who entered different stages on the same date, we calculated the entry date of the theoretically earlier state as the date of the latter state minus 0.5 day based on the previous study.^[32]^ For comparison, we examined (post hoc analysis) the association of neuroticism with mortality risk in the same study sample (healthy state), irrespective of the incidence of neuroticism-related illness during the follow-up. In sensitivity analysis, we included anxiety and depression status and income as additional confounders to rule out their influence on the results. We then examined the association of neuroticism with transitions to multimorbidity and mortality (1) participants free of multimorbidity at baseline and (2) participants with the onset of a first neuroticism-related disease [Supplementary Figure 2, http://links.lww.com/CM9/C338]. We also performed stratified analyses as described in the Supplementary Material, http://links.lww.com/CM9/C338.

Linear regression was used to examine the associations of neuroticism with brain structures and immune parameters. A latent variable representing inflammation was estimated using confirmatory factor analysis via neutrophil count, NLR, and SII, which were significantly associated with neuroticism and its cluster. Mediation analyses were conducted to examine whether the association of neuroticism with multimorbidity was mediated by inflammation in healthy participants. Of note, the brain imaging measures were not included in the mediation analyses due to the limited sample size. Covariates of the multistate models, linear regression, and mediation models comprised age, sex, ethnicity, education, TDI, BMI, physical activity, alcohol, and smoking status. Intracranial volume was added as an additional covariate in the analysis of brain structures. FDR corrections were performed for multiple comparisons, if appropriate. All P-values were two-sided. R software version 4.1.0 (www.Rproject.org) and GraphPad Prism version 8.00 (GraphPad Software, San Diego, CA, USA) were used for statistical analyses and figure preparation.

Results

Basic characteristics of participants

A total of 393,067 participants (212,139 female, 54.0%) with available 12-item neuroticism scores were included in the analysis of neuroticism, with a mean baseline age of 56.3 years (standard deviation [SD], 8.1 years). As expected, younger age, female, lower education level, greater socioeconomic deprivation, more smoking, less fruit and vegetable intake, and more baseline diagnoses were associated with higher neuroticism. Baseline characteristics of the participants stratified by neuroticism levels are shown in Table 1 and Supplementary Tables 3–5, http://links.lww.com/CM9/C338. A general scheme of the current study is depicted in Figure 1.

Associations of neuroticism with risk of morbidity

During a median follow-up of 12.54 years, 1,680,464 incident diseases were recorded. After FDR correction, medium levels of neuroticism at baseline were associated with risks of 44 incident diseases, whereas high levels of neuroticism at baseline were associated with 58 of the 78 diseases studied [Figure 2], covering 15 ICD-10 categories. Among the 58 diagnoses, high neuroticism was significantly associated with decreased risk of cancer (HR [95% confidence interval (CI)] = 0.96 [0.94–0.98], Pfalse discovery rate [FDR] <0.001), melanoma (0.90 [0.87–0.93], PFDR <0.001), and road accidents (0.87 [0.77–0.98], PFDR = 0.034), but was associated with increased risk of other incident conditions, with the most significant association being self-harm (4.52 [3.26–6.28], PFDR <0.001). Among all the ICD-10 categories, the most significant associations with neuroticism were the mental and behavioral disorders (1.75 [1.71–1.80]) and diseases of the nervous system (1.35 [1.32–1.39], PFDR <0.001). Specifically, individuals with high levels of neuroticism have a 3.93-fold (95% CI: 3.76–4.11) increased risk of mood disorders, a 3.07-fold (95% CI: 2.96–3.19) increased risk of neurotic disorders, a 2.05-fold (95% CI: 1.73–2.44, PFDR <0.001) increased risk of psychotic disorders, and a 1.86-fold (95% CI: 1.76–1.96, PFDR <0.001) increased risk of sleep disorders. In cluster-specific analyses, these findings remained similar [Supplementary Tables 6–8, http://links.lww.com/CM9/C338]. In sensitivity analyses including anxiety and depression status and income as additional covariates, the significant diseases obtained were consistent with the main analysis.

Associations of neuroticism with the risk of multimorbidity

Further analysis omitted 12 broad neuroticism-related ICD-10 categories where specific disease outcomes within them were more strongly related to neuroticism, to focus on non-overlapping neuroticism-related disease diagnoses. Ischemic heart diseases and diseases of the liver were omitted because their associations with neuroticism were weaker than for angina pectoris and alcoholic liver disease, respectively. The unclassified symptoms and signs and diagnoses of external causes were not included in the construction of multimorbidity. Thus, the multimorbidity analyses were based on 37 non-overlapping neuroticism-related diseases [Figure 2]. Among 200,635 participants free of multimorbidity at baseline, 89,447 (44.9%) developed multimorbidity over a median (interquartile range [IQR]) follow-up of 11.79 (6.26–13.12) years. Participants with moderate (HR [95% CI] = 1.08 [1.07–1.10], P <0.001) and high (1.19 [1.07–1.10], P <0.001) levels of neuroticism had an increased risk of multimorbidity in comparison to the low neuroticism group. Results were consistent across the three neuroticism clusters [Supplementary Table 9, http://links.lww.com/CM9/C338]. In sensitivity analyses among healthy participants or with additional adjustments, the significance of the associations between neuroticism and multimorbidity was maintained [Supplementary Table 9, http://links.lww.com/CM9/C338]. Among the neuroticism clusters, SESA yielded the smallest effect size, while depressed affect showed the strongest effect.

Association of neuroticism with risk of mortality

During a median follow-up of 12.86 (12.14–13.57) years, 22,482 deaths were recorded (mortality rate, 4.45 per 1000 person-years). In the base model (Model 1), all-cause mortality was higher in patients with medium and high levels of neuroticism [Figure 3]. Further adjustment for other sociodemographic variables, physical measures, and healthy behaviors completely attenuated the effect on the association (Model 2). However, in the analysis additionally adjusted for baseline diagnoses, moderate (0.97 [0.94–1.00], P = 0.036) and high (0.95 [0.92–0.99], P = 0.005) neuroticism were associated with a decreased risk of all-cause mortality (Model 3). As with all-cause mortality, after full adjustment in Model 3, we observed an association between neuroticism and lower risk of death from cancer, ischemic heart diseases, and cerebrovascular diseases, but not from respiratory diseases. In the three models, the association of neuroticism cluster worry with mortality risk produced similar trends to the total neuroticism score. However, depressed affect was associated with a higher risk of all-cause mortality and respiratory diseases mortality in all the three models. SESA showed consistent results in all the three models associated with a reduced risk of all-cause mortality and cancer mortality.

Association of neuroticism with transitions to first disease, multimorbidity, and death

Among 41,564 participants with data on neuroticism who were in healthy state at baseline, during a median follow-up of 12.94 (12.17–13.59) years, 17,527 participants developed a first neuroticism-related disease, 9026 of whom subsequently developed a second disease (multimorbidity), and 842 of whom died. Compared with low neuroticism, high neuroticism was associated with an increased risk of transitions from a healthy state to a first neuroticism-related disease (1.09 [1.05–1.13], P <0.001) and subsequent transitions to multimorbidity (1.08 [1.02–1.14], P = 0.005), but were associated with a decreased risk of transitions from the first disease to death (0.53 [0.37–0.76], P <0.001) and with a significant trend toward a decreased risk of transitions from multimorbidity to death (0.90 [0.84–0.97], P for trend = 0.006) [Figure 4]. Results for the three neuroticism clusters differed in their association with transitions to mortality. High scores for depressed affect were associated with a reduced risk of transitions from first disease to mortality (0.64 [0.45–0.92], P = 0.015), but high scores for SESA (0.70 [0.56–0.87], P = 0.001) and worry (0.84 [0.70–1.00], P = 0.044) scores were associated with a reduced risk of transitions from multimorbidity to mortality. Sensitivity analyses with additional adjustments yielded similar results [Supplementary Table 10, http://links.lww.com/CM9/C338].

In post hoc analysis, we examined the associations between neuroticism and risk of mortality without considering the incidence of neuroticism-related diseases. Of 41,564 healthy participants at baseline, 1246 died during follow-up. Baseline healthy participants with high neuroticism showed a lower risk of mortality (0.86 [0.74–1.00], P = 0.043; Supplementary Table 11, http://links.lww.com/CM9/C338). Of the three neuroticism clusters, only high SESA showed a similar result associated with a reduced risk of mortality (0.82 [0.69–0.97], P = 0.021). In the first sensitivity analysis, of 200,635 participants who were free of multimorbidity at baseline, 89,447 participants developed multimorbidity, and 6543 of them died [Supplementary Figure 2A, http://links.lww.com/CM9/C338]. Compared with low neuroticism, high neuroticism was associated with an increased risk of transitions to multimorbidity (1.19 [1.17–1.21], P <0.001) and with a decreased risk of subsequent transitions to mortality (0.88 [0.83–0.94], P <0.001), but not with transitions to mortality without developing multimorbidity [Supplementary Table 12, http://links.lww.com/CM9/C338]. The three neuroticism clusters revealed similar results in this multistate model. The target population of the second sensitivity analysis consisted of 100,548 participants diagnosed with one of the neuroticism-related diseases at baseline. Among them, 61,520 participants developed multimorbidity, and 4188 of them died [Supplementary Figure 2B, http://links.lww.com/CM9/C338]. Compared with low neuroticism, high neuroticism was associated with an increased risk of transitions to multimorbidity (1.1 [1.08–1.13], P <0.001) and with a decreased risk of subsequent transitions to mortality (0.92 [0.85–0.99], P = 0.026), but not with transitions to mortality without developing multimorbidity [Supplementary Table 13, http://links.lww.com/CM9/C338]. Among the three neuroticism clusters, only SESA showed a significant association with a reduced risk of transitions from multimorbidity to mortality (0.87 [0.80–0.95], P <0.001). Furthermore, we observed significant effect modification of neuroticism by age and TDI on transitions from healthy to first disease or from first disease to the onset of multimorbidity [Supplementary Figures 3 and 4, http://links.lww.com/CM9/C338]. The younger groups had a higher risk of developing the first disease in association with high neuroticism. Individuals with a medium level of TDI (quintile 2–4) had a higher risk of developing the multimorbidity after the onset of first disease in association with high neuroticism. The interaction P-values were not significant in the stratified analysis of associations between neuroticism and transitions from any health state to death [Supplementary Figures 5–7, http://links.lww.com/CM9/C338], indicating consistent and reliable results across strata. However, the association of neuroticism cluster depressed affect and SESA with risk of mortality after the onset of first disease differing among individuals of different smoking status [Supplementary Table 14, http://links.lww.com/CM9/C338].

Immune parameters link neuroticism to multimorbidity risk

In analysis adjusting for sociodemographic variables, physical measures, healthy behaviors, and baseline diagnosis among the total sample, we observed that neuroticism and its two clusters, worry and depressed affect, were associated with increased levels of the neutrophils, NLR, PLR, and SII, but were associated with lower levels of the lymphocytes [Supplementary Figure 8, http://links.lww.com/CM9/C338]. However, neuroticism cluster SESA was associated with lower levels of neutrophils, lymphocytes, monocytes, and CRP. To exclude the effects of multiple diseases, we repeated the analysis in healthy participants and found a robust correlation of neuroticism and its two clusters, worry and depressed affect, with increased levels of the neutrophils, NLR, and SII; whereas cluster SESA was not associated with any immune parameters [Figure 5A]. Accordingly, we estimated the latent variable representing inflammation by neutrophil count, NLR, and SII in a confirmatory factor analysis. The inflammatory markers partially mediated the associations of neuroticism and its clusters with the risk of multimorbidity in healthy participants, except for the SESA [Figure 5B].

Association of neuroticism and brain structural measures

Given the strong associations between neuroticism and neurological and psychotic disorders, we further examined the associations of neuroticism and the three clusters with brain structural measures. Neuroticism scores correlated with lower volume of the left insula and higher volume of subcortical regions, including bilateral amygdala, accumbens area, and putamen [Figure 5C and Supplementary Tables 15 and 16, http://links.lww.com/CM9/C338]. Depressed affect also correlated with these brain regions, and additionally correlated with high volume of right middle temporal, right frontal pole, and bilateral pars orbitalis lobes [Figure 5D]. However, the other two clusters, worry and SESA, did not significantly correlate with any brain structures, except for the association between SESA and lower volume of bilateral pericalcarine [Figure 5E, F].

Discussion

Briefly, our findings suggest two distinct sorts of health-related outcomes that result from neuroticism, that is, participants with high neuroticism were at greater risk for developing most illnesses and multimorbidity, but had a reduced risk of death after the onset of disease, especially after the onset of multimorbidity. The leading neuroticism cluster yielding detrimental effects on health was depressed affect, while that showing a protective role against the risk of mortality was SESA. The strong correlation between depressed affect and brain structure and the insignificant association between SESA and immune parameters further suggest the different physiological mechanisms underlying the neuroticism clusters.

Our findings on the relationship between neuroticism and morbidity were in line with previous literature, particularly the higher risk of neuropsychiatric disorders^[3,14,33]^ and CVD.^[34]^ The association between neuroticism and mortality was inconsistent across different models. In the age- and sex-adjusted analysis, higher neuroticism was associated with an increased risk of mortality. This association was completely attenuated after further adjustment for sociodemographic variables, physical measures, and healthy behaviors, suggesting that these factors may partially explain the deleterious effects of neuroticism on health outcomes, as previously reported.^[35]^ Most interestingly, however, the association between high neuroticism and mortality risk was significantly reversed after additional adjustment for baseline diagnosis, showing a protective effect. Thus, health status may significantly influence the association between the two. Neuroticism may have a salutary influence on health under certain conditions.^[12]^ For example, higher neuroticism was associated with a lower risk of mortality from all-causes and cancer among participants who rated their health as fair or poor, but not among those with excellent or good self-rated health.^[10]^ Evidence from the present study supported this notion and extended previous findings through multistate models indicating that high neuroticism was associated with a decreased risk of mortality after disease onset. Specifically, the protective effect of high neuroticism was robust for transitions from multimorbidity to mortality in both the primary and sensitivity analyses, but was not significant for transitions from health to death among those who remained healthy during follow-up. In a post hoc analysis of the same study sample, the significant association of neuroticism with transitions from healthy state to death reappeared irrespective of the presence of diseases. This further indicates that the presence of diagnosed conditions is a critical confounding factor to be reckoned with.

The findings of the protective effect of high neuroticism are unexpected but reasonable, as higher neuroticism is associated with greater use of healthcare services.^[2]^ Prompt seeking of medical advice is a possible mechanism underlying the covert protective role of neuroticism, although direct evidence is required. If so, researchers may need to reevaluate the evidence regarding the economic costs of neuroticism in terms of health-care resource utilization.^[10]^ In addition, neuroticism is thought to influence health through behavioral pathways. At baseline, participants high in neuroticism are more likely to engage in unhealthy behaviors, such as smoking more, eating fewer fruits and vegetables, and engaging in less regular physical activity. However, personality traits can affect how one responds to illness in terms of health behaviors.^[11]^ For example, high neuroticism combined with high conscientiousness, a configuration termed “healthy neuroticism”, has been suggested to predict smoking behavior after the onset of disease, not before.^[11]^ Although further evidence is needed, an alternative explanation of the paradox of neuroticism and health-related outcomes is that after the onset of multimorbidity, people who are high in neuroticism may be more vigilant about their health and more likely to modify their behaviors. By identifying the exact underlying behavioral factors, further interventions can be made on these behaviors in patients with multimorbidity to reduce the risk of death.

Examining neuroticism in genetically homogeneous clusters with different underlying biological mechanisms may further explain its controversial associations with health outcomes.^[14]^ It is worth noting that the three neuroticism subclusters all showed an association with higher risk of multimorbidity, but had different effects on mortality. The leading component of neuroticism yielding deleterious effect on health was depressed affect, which showed robust association with increased risk of all-cause and respiratory diseases mortality both in the age- and sex-adjusted base model and fully adjusted model, and showed the highest effects on the transitions to a first disease and multimorbidity. However, the leading component of neuroticism that displayed a protective role against mortality risk was SESA, which showed a consistent association with lower risk of all-cause mortality and cancer mortality both in the age- and sex-adjusted base model and fully adjusted model, and displayed robust protective effects on transitions from multimorbidity to mortality in the three multistate models. Neglecting of the differential effects of the two neuroticism clusters, SESA, and depressed affect, is likely to be responsible for the inconsistent findings of the previous studies. The two clusters, SESA and depressed affect, even showed different trends in the distribution of baseline characteristics such as physical measures and health behaviors. Differences in the distribution of these characteristics may account for the opposite direction of the associations between these two clusters and health outcomes. A previous study analyzing two facets of neuroticism constructed by an exploratory structural equation model has reported findings similar to us, in that the higher scores on a facet of neuroticism related to worry and vulnerability were associated with lower mortality.^[10]^ Consistently, the top three items with the highest loadings on the facet “worry and vulnerability” were the same as those comprising the SESA in the current study. High scores on the worry cluster also showed protective effects on mortality of all-cause, cancer, CVD, and on transitions from multimorbidity to death, but were not as robust as SESA.

Following previous clues about the relationship between personality and immune parameters,^[18,19]^ we hypothesized that immune markers might serve as a pathway between neuroticism and health outcomes. We found that healthy individuals with high neuroticism were associated with higher levels of immune parameters, including neutrophils, NLR, PLR, and SII. The measurement of neutrophils reflects innate immune responses, whereas the NLR, PLR, and SII reflect systemic inflammation and the balance between innate and adaptive immunity.^[22,23]^ The association between neuroticism and multimorbidity in healthy participants appeared to be partly explained by an imbalance in the immune system toward innate immunity. One plausible explanation is that an overactive innate immune system is associated with a failure to resolve inflammation, resulting in the development of chronic inflammation. This mechanism has been implicated in the pathogenesis of many diseases.^[36]^ Remarkably, the three clusters showed differential associations with immune parameters. Depressed affect and worry showed similarity in their associations with immune indicators, especially among healthy individuals, which showed significant positive correlations with neutrophils, NLR, and SII. However, SESA was not associated with any inflammatory parameters in healthy subjects. Regarding the relationship between neuroticism and CRP, previously reported evidence is controversial.^[20,37]^ We did not find a significant association of CRP with the total neuroticism score either in the whole cohort or in healthy participants. However, the three neuroticism clusters yielded different results. In the total sample, the depressed affect cluster was positively associated with CRP, but was negatively associated with the SESA and worry clusters. In healthy participants, the worry cluster also showed a negative association with CRP. While the depressed affect cluster and SESA clusters were not significantly associated with CRP in healthy participants, the effect estimates for both were in the same direction as in the total sample analysis. These findings may partially explain the protective role of SESA against mortality and its weak association with multimorbidity, as well as the leading effects of the depressed affect cluster on multimorbidity. It raises a possible explanation that previous inconsistent evidence on the associations between neuroticism and CRP levels may be attributed to the omission of different aspects of neuroticism.^[20,37]^ In summary, the heterogeneous findings of neuroticism clusters provide new insights into understanding the complex effects of neuroticism on health outcomes. However, the precise biological mechanisms behind the phenotypic differences resulting from the three neuroticism clusters require further investigation.

The association of neuroticism with the amygdala and insula volume is reasonable. The amygdala plays an important role in the pathophysiology of depression as a brain structure specifically involved in emotion regulation and fear conditioning;^[38]^ and the insula is a critical region involved in anxiety disorders through its mediation of interoceptive processing.^[39]^ Changes in these brain structures associated with mood disorders in people with high neuroticism appear to represent a biological mechanism linking neuroticism to health outcomes. Although the decreased connectivity between the two regions and the increased amygdala activation in response to specific stimuli have been associated with neuroticism,^[404142]^ the association between the macroscopic brain structure and neuroticism is inconclusive.^[43,44]^ Moreover, the three neuroticism clusters differed in their association with brain volume. The depressed affect cluster showed similar results to the total neuroticism score, whereas worry and SESA showed little significant association with brain structure. This suggests that the neurobiological mechanisms of neuroticism may differ between neuroticism clusters or facets.

Strengths of our study include the large sample size, long follow-up, and extensive objective diagnostic data, which allow us to comprehensively analyze the association of neuroticism with a broad range of common health conditions and health state transitions. Another strength of this study is that we further examined the association of neuroticism with the interplay between innate and adaptive immunity based on blood cell ratios. There are some limitations. First, the study population did not include children and younger adults. Because the spectrum of diseases and the causes of death in these age groups differ from that of middle-aged and elderly people, further research is warranted to examine whether our findings generalize to all age groups. Second, no data were available on personality traits other than neuroticism, particularly conscientiousness. Although there is evidence that conscientiousness may influence the relationship between neuroticism and mortality,^[8]^ we were unable to examine its potential impact in our multistate models, which may yield very interesting results. Moreover, the three neuroticism clusters, derived to be genetically homogenous, will have strong correlations with each other because of their associations with the general factor of neuroticism,^[14]^ potentially resulting in power reduction.

In conclusion, higher neuroticism was associated with a greater risk of developing most illnesses. The relationship between neuroticism and the risk of mortality differs among health status. Higher neuroticism was associated with an increased risk of developing a first neuroticism-related disease and multimorbidity, but with a decreased risk of mortality after the onset of a first disease or after the onset of multimorbidity. Besides, neuroticism was correlated with increased levels of neutrophils, NLR, and SII, which mediated the association between neuroticism and risk of multimorbidity. Notably, the three clusters showed differential associations with risk of mortality in multistate models. The leading neuroticism cluster that associated with higher risk of health–illness transitions was depressed affect, which correlated with higher amygdala volume and lower insula volume. The protective effect of neuroticism against mortality was mainly contributed by the SESA. In contrast to the other two clusters, SESA was not associated with biomarkers of an imbalance between innate and adaptive immunity. Further research on the behavioral changes in people with high neuroticism before and after multimorbidity, as well as on the genetic and biological mechanisms related to SESA, may help to provide guidelines for mortality prevention in people who already have multimorbidity and to elucidate the mechanisms underlying neuroticism’s covert protection against death.

Acknowledgments

The authors gratefully thank all the participants and professionals contributing to the UK Biobank.

Funding

This study was supported by grants from the Science and Technology Innovation 2030 Major Projects (No. 2022ZD0211600), the National Natural Science Foundation of China (Nos. 82071201 and 82071997), Shanghai Municipal Science and Technology Major Project (No. 2018SHZDZX01), Research Start-up Fund of Huashan Hospital (No. 2022QD002), Excellence 2025 Talent Cultivation Program at Fudan University (No. 3030277001), Shanghai Talent Development Funding for The Project (No. 2019074), the Shanghai Rising-Star Program (No. 21QA1408700), and ZHANGJIANG LAB, Tianqiao and Chrissy Chen Institute, and the State Key Laboratory of Neurobiology and Frontiers Center for Brain Science of Ministry of Education, Fudan University.

Conflicts of interest

None.

Supplementary Material