Interactions between Neuroticism and Stressful Life Events Predict Response to Pharmacotherapy for Major Depression: A CAN-BIND 1 Report

Abstract

Exposure to stressful life events and individual differences in the personality trait neuroticism are important risk factors that interact to predict major depressive disorder (MDD). Less is known about their effect on treatment response in depression. Here, we examine whether stressful life events experienced prior to and during treatment interact with neuroticism to predict response to 16-week pharmacotherapy for MDD. Participants included 159 outpatients with MDD who were initially treated with 8 weeks of escitalopram. Those who responded to the initial treatment continued on escitalopram monotherapy, while non-responders received 8 weeks of adjunctive aripiprazole. Personality was assessed using the NEO-Five Factor Inventory, and stressful life events were assessed using the Life Events and Difficulties Schedule, a rigorous contextual interview that includes independent ratings of threatening life events. High baseline neuroticism was associated with a lower likelihood of response when patients experienced one or more negative life events before treatment. Secondary analyses indicated that this effect was specific to neuroticism, and not better accounted for by its self-criticism or negative affect facets. Our results suggest that assessing personality and stressful life events at baseline can help clinicians assess which patients will respond to antidepressant therapy, and which may need treatment augmentation.

In diathesis-stress models of depression, high levels of life stress exacerbate underlying cognitive or personality vulnerabilities, leading to the onset and/or maintenance of depressive illness¹. No vulnerability factor has been studied more often in diathesis-stress models than the personality trait of neuroticism, which reflects a relatively stable tendency to experience frequent and intense negative affect². In experience-sampling studies, neuroticism is associated with heightened reactivity to a variety of stressors^3–5, and magnitude of stress response is also related to risk for subsequent psychopathology⁶, supporting a role for neuroticism in a diathesis-stress account. Longitudinal studies have consistently shown that stressful life events amplify the association between neuroticism and the likelihood, severity, and temporal course of subsequent depressive illness^7–9.

While there is considerable evidence for a diathesis-stress model in which life stress and neuroticism interact to influence the onset and course of depression, less is known about the influence of these two risk factors on response to treatment for depressive illness. Baseline neuroticism typically predicts less improvement following antidepressant treatments^10–12, and changes in depressive symptoms during antidepressant trials are associated with changes in neuroticism as well^13,14. Stressful life events prior to and during treatment predict worse response to some antidepressants^15,16, though some studies have found the opposite effect¹⁷. Neuroticism may moderate the effect of stressful life events on treatment outcomes, making response more likely for some but less likely for others. For instance, Bulmash and colleagues¹⁸ found that severe stress prior to or during treatment predicts poor response, but only among individuals high in self-criticism. Psychometric research has shown that self-criticism is a lower-order facet of neuroticism¹⁹.

The purpose of the current study was to test the specific hypothesis that individuals high in neuroticism are less responsive to pharmacotherapy when they experience stressful life events prior to or during treatment. In testing this hypothesis, we examine the separate influences of independent and dependent stressful life events. Independent events occur outside the influence of one’s own behavior and choices (e.g., a natural disaster), whereas dependent events are at least partially brought about by the individual’s own actions (e.g., starting a fight). Previous treatment studies have focused on total stressful life events, and the evidence as to whether independent and dependent stressors differentially predict depressive illness is mixed²⁰. However, one previous study found that interpersonal and achievement events differentially predicted response rates following a standard 6-week antidepressant regimen²¹. Thus, the type of events experienced may be another source of discrepancies in the association between stress and treatment response. Here, we address this question using rigorous interview-based contextual life event measures to rate the presence, severity, and independence of life events that occur prior to and during treatment. Past studies have typically employed self-report life events checklists, which are common in the field but susceptible to poor reliability, inadequate coverage of stressors, and depressive recollection biases that spuriously inflate relations of life events to depressive outcomes²².

In secondary analyses, we also examined whether interactions between neuroticism and stressful life events on treatment response could be better explained by two lower-order subcomponents of neuroticism: self-reproach (akin to self-criticism from the study by Bulmash and colleagues) and negative affect²³. These analyses represent a direct test of the hypothesis that previous self-criticism/stressful life event findings were driven by the variance self-criticism shares with the broader trait of neuroticism.

Methods

Procedure

The design of CAN-BIND 1 has been described elsewhere^24,25. Data were collected across six sites in Canada. Eligible participants were prescribed eight weeks of open-label treatment with escitalopram (10–20mg daily, flexible dosage). Responders at Week 8 (defined as ≥ 50% reduction on the Montgomery–Åsberg Depression Rating Scale²⁶) continued on escitalopram monotherapy while non-responders were prescribed adjunctive aripiprazole (open-label, 2–10mg) for an additional eight weeks.

Participants

Participants were 159 individuals (M Age = 35.96 years, 62% female) from CAN-BIND-1 (N = 211) who completed the LEDS-II interview (see Table S1 for demographic and clinical characterization). The primary diagnosis of MDD was assessed via the MINI International Neuropsychiatric Interview²⁷. To be included in the study, participants needed to be experiencing a current major depressive episode of at least moderate severity (for full eligibility criteria, see ^24,25).

Measures

NEO-Five Factor Inventory (NEO-FFI).

The NEO-FFI² is a 60-item self-report questionnaire that yields five subscales measuring the Big Five and thirteen lower-order subcomponent scales, including self-reproach and negative affect subcomponents for neuroticism²³. Participants rate the extent to which they agree or disagree with each item using a 5-point Likert scale prior to treatment.

Life Events and Difficulties Schedule (LEDS-II).

The LEDS-II²⁸ is a semi-structured contextual interview and rating system used to assess life events and difficulties. The interview probes 10 different domains. A timeline is used to help participants date events. Interviewers were trained to probe for contextual details regarding events, but not to query participants’ subjective responses. All interviews are audiotaped and conducted at the completion of treatment to minimize depressive bias in recall. Only treatment completers received the LEDS-II interview.

A research assistant listened to each interview and prepared vignettes for each life event, removing information related to the participant’s subjective response to the stressor. A panel of 2 to 4 raters then rated each event based on explicit rules outlined in the LEDS manual, as well as over 5,000 case vignettes that help provide anchoring and ensure standardization of the ratings. All raters and interviews were trained and supervised by KLH.

Life events were rated based on their level of contextual threat (i.e., severity) and independence. Threat was rated on a 5-point scale: 1 = marked, 2a = high moderate, 2b = low moderate, 3 = some, 4 = little/none. Independent life events included events judged to be totally or nearly totally independent of the actions of the individual. Dependent life events were at least partly dependent on the participant’s actions. Discrepancies among raters were discussed until a consensus was achieved.

We created six separate life event variables: the total number of negative (score of at least 3-some threat) events overall, negative independent events, and negative dependent events, each assessed in the 16-week period preceding treatment initiation and in the 16-week treatment trial.

Montgomery–Åsberg Depression Rating Scale (MADRS).

The MADRS²⁶ is a clinician-rated interview-based measure of depression severity consisting of ten items. Participants completed the MADRS at baseline and at the end of treatment (Week 16). MADRS response was defined as ≥ 50% symptom reduction from baseline to post-treatment.

Data Analysis

Analyses were conducted using Mplus Version 8.04²⁹ and the MplusAutomation package in R³⁰. Participants were included in analyses if they had completed the NEO-FFI and the LEDS-II interview. Of the 159 participants who met those criteria, 3 (2%) did not have response data by the conclusion of treatment at Week 16. Analyses were conducted using maximum likelihood estimation with robust standard errors to account for clustering in the data as a function of recruitment site³¹. Missing data were modeled under the assumption that data were missing at random.

Associations between pre-treatment neuroticism, stressful life events prior to and during treatment, and treatment response were evaluated with six hierarchical logistic regression models (3 models examining total, independent, and dependent stressors pretreatment and during treatment, respectively). A more conservative alpha of .008 (.05 / 6 tests) was used in primary analyses to correct for multiple comparisons. In secondary analyses, we reran any models in which there was a significant neuroticism × negative life events interaction, but substituted self-reproach and negative affect in place of neuroticism.

Continuous predictor variables were standardized. Age and sex were entered as covariates¹, followed by main effects of personality (neuroticism, or the neuroticism subcomponents), the relevant negative event variable, and finally interaction terms. Simple slopes analyses were used to probe the effect of neuroticism on treatment response at varying levels of stress exposure prior to or during treatment.

Results

Descriptive Statistics and Bivariate Correlations

Descriptive statistics for observed study variables are presented in Table S2. Correlations between primary study variables are presented in Table S3. A total of 117 participants (74%) met criteria for response by Week 16. Just over half the total sample (N = 84, 53%) experienced at least one negative life event during the 16 weeks prior to treatment (Figure S1). A smaller percentage (N = 25, 16%) experienced two or more negative events prior to treatment. Rates were similar for treatment concurrent events: 46% (N = 73) of patients experienced at least one negative event and 19% (N = 31) experienced two or more events.

Neuroticism and Total Negative Events Predicting Treatment Response

Results from the first two hierarchical logistic regressions examined whether neuroticism interacted with pretreatment and treatment concurrent total negative life events, respectively, to predict treatment response (Table 1). There was no main effect of neuroticism or of total negative events on treatment response in either model before entering the interaction terms. In the model that included pretreatment events, there was a significant interaction between baseline neuroticism and total negative events (Figure 1; B = −.48, SE = .12, p < .001, OR = .62). Neuroticism was associated with a lower likelihood of treatment response among participants who had experienced at least one pretreatment negative event (B = −.40, SE = .18, p = .03, OR = .67), but had no effect on response among participants who had experienced zero pretreatment negative events (B = .09, SE = .29, p = .75, OR = 1.01). The interaction between neuroticism and pretreatment total negative events remained statistically significant after controlling for the other Big Five traits (Table S4).

Table 1.

Neuroticism and Total Negative Events Predicting Treatment Response

	Pretreatment Events				Treatment Concurrent Events
Variable	B	SE	p	OR	B	SE	p	OR
1. Age	−.24	.37	.52	.79	−.19	.37	.60	.83
2. Sex	−.29	.45	.53	.75	−.32	.47	.49	.73
3. Neuroticism	−.29	.20	.15	.75	−.26	.25	.29	.77
4. Total Negative Events	.25	.14	.08	1.28	.21	.15	.16	1.24
5. Neuroticism × Total NE	−.48	.12	< .001	.62	−.24	.38	.52	.78

Note. N = 159. Bolded effects are significant at p < .01. NE = Negative Events.

Figure 1.

Interaction between neuroticism and pretreatment total negative life events (NE) predicting the probability of response on the MADRS at Week 16. A) Effect of neuroticism on treatment response moderated by pretreatment total negative events. B) Effect of pretreatment total negative events on treatment response moderated by neuroticism.

Simple slopes analyses were also conducted to probe the effect of stress on treatment response, with neuroticism as the moderator. As depicted in Figure 1B, pretreatment stress was associated with a higher likelihood of response among those average or below average on neuroticism (< −.06 SD), but a lower likelihood of response among those high in neuroticism (> 1.51 SD).

In the model that included treatment concurrent events, there were no significant interactions between neuroticism and negative life events (Table 1).

Personality and Independent/Dependent Negative Events Predicting Treatment Response

Subsequent logistic regressions examined whether neuroticism interacted with independent or dependent life events prior to or during treatment to predict patient response. Neither independent nor dependent life events had a main effect on treatment response. There was a significant interaction between neuroticism and independent negative life events experienced prior to treatment (B = −.42, SE = .11, p < .001, OR = .66; Table 2). As depicted in Figure 2A, neuroticism was associated with worse treatment response among participants who had experienced at least one pretreatment negative independent event (B = −.72, SE = .16, p < .001, OR = .49). In contrast, there was no effect of neuroticism on treatment response among participants who had experienced zero negative independent life events prior to treatment (B = −.08, SE = .27, p = .76, OR = .92). Simple slopes analyses also examined the effect of stress on treatment response, with neuroticism as the moderator. As depicted in Figure 2B, pretreatment stress was associated with a higher likelihood of response among those low on neuroticism (< −.14 SD), but a lower likelihood of response among those very high in neuroticism (> 2.15 SD).

Table 2.

Neuroticism and Independent Negative Events Predicting Treatment Response

	Pretreatment Events				Treatment Concurrent Events
Variable	B	SE	p	OR	B	SE	p	OR
1. Age	−.22	.37	.55	.80	−.22	.38	.56	.80
2. Sex	−.31	.51	.54	.73	−.26	.49	.59	.77
3. Neuroticism	−.31	.22	.16	.73	−.29	.21	.17	.75
4. Independent Negative Events	.27	.17	.11	1.31	.29	.21	.16	1.33
5. Neuroticism × Independent NE	−.42	.11	< .001	.66	−.24	.14	.09	.79

Note. N = 159. Bolded effects are significant at p < .01. NE = Negative Events.

Figure 2.

Interaction between neuroticism and pretreatment independent negative life (INE) events predicting the probability of response on the MADRS at Week 16. Note, this effect was reduced to non-significance after controlling for conscientiousness and its interaction with pretreatment independent negative events (see Table S5 and Figure S2). A) Pretreatment negative independent events as moderator. B) Neuroticism as moderator.

The interaction between neuroticism and independent negative events survived when controlling for the other Big Five traits and their interaction with independent events, with one exception: a significant interaction between conscientiousness and independent negative events (B = .72, SE = .25, p = .004, OR = 2.05) suppressed the neuroticism effect when added to the model, reducing it to non-significance (Table S5). In the full model that included all traits, higher conscientiousness predicted better treatment response when participants had experienced at least one independent negative event before treatment (B = 1.16, SE = .30, p < .001, OR = 3.20; Figure S2A). In contrast, there was no effect of conscientiousness on treatment response when participants had experienced zero negative life events prior to treatment (B = .10, SE = .29, p = .73, OR = 1.11). Pretreatment independent events were associated with better treatment response in individuals scoring ≥ .33 standard deviations above the mean on conscientiousness. Overall, this pattern of results indicates that any interaction between neuroticism and pretreatment independent negative events on response is driven primarily by shared variance between neuroticism and conscientiousness.

Additional analyses found no evidence for an interaction between neuroticism and treatment concurrent independent events, or dependent negative events prior to or during treatment (Tables 2, S5–S7).

Subcomponents of Neuroticism and Negative Life Events Predicting Treatment Response

Lastly, we examined whether the interaction between neuroticism and pretreatment total negative life events was driven specifically by the self-reproach subcomponent of neuroticism. Self-reproach was substituted for neuroticism in an additional logistic regression model. There was a significant interaction between self-reproach and total pretreatment negative events on treatment response (Table 3). However, the interaction between self-reproach and negative events was reduced to non-significance when the negative affect subcomponent of neuroticism was added to the model. This suppression suggests that the interaction between self-reproach and total pretreatment negative life events is driven primarily by shared variance between the two subcomponents (i.e., neuroticism), rather than by variance unique to self-reproach.

Table 3.

Neuroticism Subcomponents and Total Pretreatment Negative Events Predicting Treatment Response

	Self-Reproach Only				Negative Affect Only				Self-Reproach & Negative Affect
Variable	B	SE	p	OR	B	SE	p	OR	B	SE	p	OR
1. Age	−.24	.36	.51	.79	−.22	.38	.57	.80	−.24	.36	.52	.79
2. Sex	−.24	.49	.63	.79	−.27	.49	.58	.76	−.29	.46	.53	.75
3. Self-Reproach	−.23	.19	.22	.80	---	---	---	---	−.16	.21	.44	.85
4. Negative Affect	---	---	---	---	−.24	.20	.24	.79	−.18	.20	.37	.83
5. Total Pretreatment Negative Events	.18	.14	.21	1.19	.21	.10	.03	1.23	.25	.13	.05	1.29
6. Self-Reproach × Total Pretreatment NE	−.36	.15	.01	.70	---	---	---	---	−.30	.16	.07	.74
7. Negative Affect × Total Pretreatment NE	---	---	---	---	−.32	.17	.06	.73	−.25	.17	.15	.78

Note. N = 159. Bolded effects are significant at p < .05. NE = Negative Events.

Discussion

The results of this study extend diathesis-stress models of depression by providing new evidence that neuroticism interacts with the degree of stress in one’s environment to predict not only the onset and course of depression, but also its responsiveness to pharmacotherapy. Consistent with our primary hypothesis, neuroticism was negatively related to treatment response, but only among patients who had experienced one or more negative life events prior to treatment. Importantly, focusing on treatment response (≥50% symptom reduction) as a primary outcome means that our results speak specifically to clinically meaningful change. As seen in Figure 1, individuals in our study who scored two standard deviations above the mean on neuroticism, and who also experienced a negative life event prior to treatment, were about 20% less likely to respond to antidepressant therapy than individuals who had the same score on neuroticism but experienced no negative life events preceding treatment. Follow-up analyses indicated that this effect was not better accounted for by variance unique to neuroticism’s self-reproach or negative affect subcomponents.

One explanation for these findings is that life stress prior to treatment amplifies depressive cognitions and behaviors among individuals already prone to high negative affect, in turn disrupting the therapeutic effects of pharmacotherapy. Such a model would be consistent with personality theories positing that variability in neuroticism is most apparent in the presence of environmental threats and punishments^32,33. Previous findings from CAN-BIND-1 have also highlighted improvements in early reward sensitivity as an important marker of treatment response³⁴. Stressful life events dampen reward-related activation in the left ventral striatum among those high in neuroticism, and this dampening is associated with history of depression⁹. Thus, blunted and/or recalcitrant reward sensitivity may be another pathway by which neuroticism and stressful life events combine to hamper treatment response. Finally, our results indirectly add to evidence implicating baseline serotonergic functioning in antidepressant response³⁵. Neuroticism has been consistently linked to serotonin functioning in drug challenges, include those involving acute citalopram administration^36,37, and in PET studies examining serotonin receptor binding³³. Changes in neuroticism also mediate the effects of some SSRIs^13,14. Stressful life events prior to treatment interact with variation in the serotonin transporter gene to predict antidepressant response¹⁷, though the effect is often inconsistent and may be small in nature³⁸. Taken together, these results suggest that neuroticism may act in part as a proxy for serotonergic responsiveness, both to environmental stressors and to the therapeutic effects of antidepressants.

Somewhat surprisingly, pretreatment negative events predicted better treatment response among individuals with average to low neuroticism. A few previous studies have also reported a positive association between stress and treatment response^17,21. Our results suggest variability in neuroticism may explain the divergent results in the existing literature. Depressive symptoms elicited by pretreatment stressors may be less persistent in individuals low in neuroticism, who are adept at coping with stress and limiting its deleterious effects. Future studies that include an untreated depressive comparison group could test this hypothesis explicitly by examining which of our effects apply specifically to treatment response and which are more generally related to the persistence of depressive symptoms over time.

We found little evidence that the interaction between neuroticism and stressful life events varies according to the type of stressful life event (independent or dependent) experienced. Initially, there was a significant interaction between neuroticism and independent pretreatment events on treatment response, but this effect was later reduced to non-significance once we covaried for conscientiousness and its interaction with independent events. Although not hypothesized, this conscientiousness × pretreatment independent events interaction is consistent with research suggesting that conscientious individuals respond to stress actively, often via the use of healthy coping strategies and health-promoting behaviors that reduce subsequent distress^39,40. Overall, this pattern of results suggests that shared variance between neuroticism and conscientiousness (which may reflect one’s ability to maintain stability in life more generally;³⁹) helps to promote adaptive responses to stressors that arise independent of an individual’s behavior and/or decision-making.

Notably, neuroticism did not significantly interact with life events during treatment to predict outcome. This was unexpected given that the interaction reported by Bulmash and colleagues¹⁸ was robust to whether the stressor occurred prior to, or during treatment. The discrepancy may be attributable to study design or sample differences. Though the stress × self-criticism interaction in Bulmash et al. was not further moderated by treatment type, descriptively the effect appeared to be primarily associated with the psychotherapy (not pharmacotherapy) group. Stressors that occur during treatment may amplify neurotic behaviors that impede a patient’s ability to participate in intensive psychotherapy, whereas the same behaviors could have less relevance for pharmacotherapy. More broadly, personality × stress interactions may vary across treatment types, and even pharmacological agents. Our results are most pertinent to a course of pharmacotherapy that mimics CAN-BIND’s design (i.e., an initial trial of an SSRI, followed by an option for augmentation).

The current study utilized a methodologically strong pharmacotherapy protocol, a rigorous semi-structured interview to code life events, and a well-validated measure of personality. Nonetheless, limitations were present. Most of our sample was single, White, and female. Results held when covarying for age and sex, but the generalizability of our findings to other racial/ethnic groups and socioeconomic strata will need to be addressed in future research, particularly in light of research showing that antidepressant treatment is less common among racial/ethnic minorities⁴², despite stressful life events occurring more often⁴³. Few participants in our study experienced multiple negative life events. This may have been due to our modest sample size, suggesting a need to replicate these results in larger samples. Relatedly, if self-criticism exerts a small effect independent of neuroticism, we may have been underpowered to detect it (note the effect of self-criticism × pretreatment negative events was reduced to non-significance in the model with negative affect, but still marginal). Some of our results may also have been affected by the design of the trial. Though MADRS response was selected a priori as the outcome of interest in CAN-BIND-1, studies that incorporate measures of clinical functioning, or that use multiple measures to assess latent depressive symptoms, would be helpful in replicating our findings. Likewise, the CAN-BIND trial included an augmentation medication for initial nonresponders, which may have increased the likelihood that Week 16 nonresponders were more treatment resistant than nonresponders in other studies. Additional research will be needed to determine whether our effects generalize across trials that employ alternative treatment algorithms.

Overall, our findings provide a promising path forward for researchers and clinicians interested identifying early markers of response to pharmacotherapy for depression, as they indicate that the predictive validity of baseline individual difference variables can be improved via the measurement of contextual variables. An important next step in this line of research will be to examine the feasibility of assessing stress and personality in clinical settings. Stressful life event interviews can carry a significant time burden to both the interviewer and clinician, but new measures that incorporate computerized adaptive testing to reduce interview time may have promise in clinical settings⁴⁴. Our own results suggest that assessing stress and personality at the outset of treatment can yield actionable information for treatment planning and personalization. Future research can help determine whether there are feasible and efficient ways to gather this information and use it to effectively guide decision-making.