Protective Factors Predict Resilient Outcomes in Clinical High-Risk Youth with the Highest Individualized Psychosis Risk Scores

Abstract

Background and Hypothesis

Studying individuals at Clinical High Risk (CHR) for psychosis provides an opportunity to examine protective factors that predict resilient outcomes. Here, we present a model for the study of protective factors in CHR participants at the very highest risk for psychotic conversion based on the Psychosis Risk Calculator.

Study Design

CHR participants (N = 572) from NAPLS3 were assessed on the Risk Calculator. Those who scored in the top half of the distribution and had 2 years of follow-up (N = 136) were divided into those who did not convert to psychosis (resilient, N = 90) and those who did (nonresilient, N = 46). Groups were compared based on candidate protective factors that were not part of the Risk Calculator. Better functional outcome was also examined as an outcome measure of resiliency. Study Results: Exploratory analyses suggest that Hispanic heritage, social engagement, desirable life experiences, premorbid functioning and IQ are all potential protective factors that predict resilient outcomes. Reduced startle reactivity, brain area and volume were also associated with greater resilience.

Conclusions

The primary focus of CHR research has been the risk and prediction of psychosis, while less is known about protective factors. Clearly, a supportive childhood environment, positive experiences, and educational enrichment may contribute to better premorbid functioning and brain development, which in turn contribute to more resilient outcomes. Therapies focused on enhancing protective factors in the CHR population are logical preventive interventions that may benefit this vulnerable population. Future CHR research might use similar models to develop a “protective index” to predict resilient outcomes.

Introduction

Protective factors are modifiers that reduce the odds of developing a disease or poor outcome from a known risk factor. The concept of “resilience” has been studied in a variety of ways: as a capacity, a process and an outcome.¹ In the present context, protective factors predict a more resilient clinical or functional outcome in youth at clinical high risk (CHR) for psychosis. A better understanding of protective factors can inform prevention efforts in vulnerable youth.

In first-episode psychosis studies, better premorbid functioning, higher premorbid IQ, insight and a shorter duration of untreated psychosis predict symptom and functional recovery.² In children at genetic high risk for psychosis, having a higher IQ, living in a socially cohesive neighborhood, and having greater levels of social support are all associated with a reduced risk of psychosis.³ The North American Prodrome Longitudinal Studies (NAPLS) Consortium has identified protective factors associated with better functional outcomes, symptom remission and absence of conversion to psychosis.^4,5 In the NAPLS2 sample, Anglin et al.⁵ found that ethnoracial minority youth who met CHR criteria and lived in communities with a greater proportion of minorities were more likely to achieve remission of symptoms, suggesting that the relationship between childhood ethnic density and remission was partly mediated by discrimination. In CHR participants from the NAPLS3 sample, Tronick et al.⁴ reported that higher scores on a protective factors index from the investigator-rated Structured Assessment of Violence Risk in Youth (SAVRY),⁶ was associated with a lower risk of psychotic conversion, and this finding was primarily related to more prosocial involvement and resilient personality traits.

Several studies^7–10 have examined possible biomarkers of resilience in CHR youth. In a study by De Wit and colleagues,¹¹ UHR youth who were more resilient based on good functional outcomes had larger brain volumes, greater cortical thickness in frontal, temporal and parietal cortices as well as larger left hemisphere cortical surface area compared to nonresilient participants. Similarly, Collin et al.⁸ found that improvement in clinical outcomes was associated with higher network connectivity and a more typical modular connectome organization. Katagiri and colleagues⁹ reported that improvement in positive symptoms in UHR youth was associated with greater integrity of the corpus callo sum.

A few studies have examined biological markers associated with a lower risk of psychosis in the general population.^12,13 In an innovative approach, Hess and colleagues^14,15 (who have an article in this issue) identified unaffected individuals with relatively high polygenic risk scores for schizophrenia to generate a “polygenic resilience score” that excluded known risk single nucleotide polymorphisms associated with psychosis. To further elucidate neurobiological correlates of genetic resilience, Hettwer et al.¹⁶ reported a positive association between the polygenic resilience score, global cortical volume and surface area of the left fusiform gyrus in healthy subjects.

In the present analysis, we propose a model to explore protective factors that predict resilience to psychosis and better functional outcomes in a stratified sample of CHR participants who are deemed at the highest risk for psychosis conversion based on scores on the NAPLS Psychosis Risk Calculator.¹⁷ While meeting CHR criteria for psychosis carries a 20%-30% risk for the later development of psychosis,¹⁸ the risk calculator determines individual risk based on symptom, neurocognitive and family history indicators of highest risk for psychotic conversion. By selecting individuals from the NAPLS3 cohort who meet the CHR criteria and fall in the top half of the distribution of risk based on the calculator, we aim to explore this model as a means of identifying potential protective factors that are not a part of the calculator and are associated with resilient outcomes, as defined by a failure to convert to psychosis and better functional outcome after a 2-year period.

Methods

The methods for the NAPLS 3 study have been described in Addington et al.¹⁹ In brief, NAPLS3 is a study of individuals who meet the Criteria of Psychosis Risk Syndromes (COPS) based on the Structured Interview for Psychosis Risk Syndromes (SIPS).²⁰ All participants received clinical and biomarker assessments over a 2-year period.

Participants included N = 572 individuals meeting CHR criteria from the NAPLS3 study. All CHR participants were assessed on the NAPLS Psychosis Risk Calculator developed as part of the NAPLS2 study¹⁷ and received a “psychosis risk score” that is made up of variables associated with the highest risk of conversion to psychosis in individuals already meeting the COPS criteria: unusual thought content, suspiciousness, verbal learning, processing speed, change in social functioning and family history of psychosis. CHR participants were then divided based on a median split of their risk score to identify those at the highest risk of conversion to psychosis in the top half of the distribution (N = 295). A total of N = 136 CHR participants in the top half of the distribution were followed 2 years without conversion to psychosis (N = 90 resilient) or converted to psychosis in this time period (N = 46 nonresilient). A sample (N = 56) of normal comparison participants from the NAPLS3 sample who received startle testing was also included in the analysis of startle parameters for comparison. The nonconverted sample (N = 90) was included in the analysis of the 24-month functional outcome as a measure of resilience.

Baseline assessments and biomarkers included in the NAPLS3 battery that were not a part of the NAPLS Risk Calculator were selected as potential protective factors. These included demographic variables: parental age, race, ethnicity; functioning measures: Cannon-Spoor Premorbid Assessment of Functioning,²¹ Global Assessment of Functioning,²² Global Functioning Role and Social Scales²³; Protective Index from the SAVRY⁶; Positive life events and Social Engagement from the Daily Stress Inventory²⁴; measures of physical health and sleep. Biomarkers included measures of neurocognition, salivary cortisol, and brain imaging also described in Addington et al.¹⁹ In addition, a brief startle response paradigm was included in the electrophysiology battery.

Outcome measures of resilience included (1) failure to convert to psychosis after 2 years and (2) better 24-month functional outcome. Conversion to psychosis was assessed based on the POPS (Presence of Psychotic Syndrome) criteria from the SIPS. Twenty-four-month functional outcome was assessed with the Global Assessment of Functioning (GAF), Global Functioning: Role and Social (GF:R, GF:S) scales.

Acoustic Startle Paradigm

We used a startle-stimulus generating program (Grace Design Model m902 Amplifier and Neurobehavioral Systems Presentation software) developed by UCSD at each site. We calibrated the sound at all sites with Quest 210 Sound Level Meters and assured 70dB background noise and 115 dB for extended-length startle bursts by adjusting the Amplifier. All sites used the same Biosemi systems (Biosemi, Amsterdam, Netherlands). The startle response was measured with EMG recording by electrodes (Ag/AgCl) placed at the outer canthus and below the right eye, assuring resistance of less than 10 kΩ.²⁵ Headphones (TDH-39P) were used to present the startle stimuli binaurally. A background noise of 70 dB [A] and 10 startle pulses (115 dB [A], 40 ms noise burst) were presented. The paradigm includes a 5-min acclimation. EMG analysis was performed with Brain Vision Analyzer (Cortech Solutions, Wilmington, NC) after using a high-pass filter of 28 Hz at 12 dB/Oct. The waveform was then smoothed using a 40Hz 24 dB/Oct low-pass filter. All trials were screened for errors. Startle data was waveform averaged, baseline corrected and rectified. The highest point between 30 and 120 ms relative to baseline after onset of startle stimulus was defined as the peak startle response, or startle magnitude. Startle peak latency was also recorded. Participants with a relative lack of startle response (<5 microvolts) to pulse-alone stimuli were excluded from latency analyses per established methods.²⁵ We previously established that the startle parameters are reliable between NAPLS sites.²⁶

MRI Paradigm

MRI methods and details regarding quality control, inter-scanner reliability and processing are described in detail in Addington et al.¹⁹ and Collins et al.²⁷ Imaging data were obtained on 3T Siemens Tim Trio systems at Emory, UNC, Harvard, UCLA, UCSF, and Yale and on 3T GE systems at Hillside, UCSD, and Calgary using the ADNI2-specified protocols for MRI. Human Connectome Project’s Minimal Preprocessing Pipelines²⁸ with the open-source Quantitative Neuroimaging Environment & Toolbox (QuNex, qunex.yale.edu) was used to process T1-weighted structural MR images. FreeSurfer v6.0^29,30 was utilized for whole-brain segmentation and surface-based cortical reconstruction. Groups were compared based on estimated intracranial volume (eICV), total segmentation volume without ventricles, total surface area, and mean cortical thickness averaged across the 2 hemispheres. Bilateral measures of cortical thickness, cortical surface area, and subcortical volume from 34 brain regions were derived and averaged together for analysis.

Statistical Analyses

All analyses in this sample were performed with SPSS 28. For the measure of resilience based on psychotic conversion, 2 groups were identified within the upper half of the sample: Resilient, who had at least 2 years of follow-up without conversion to psychosis, and NonResilient, who transitioned to psychosis during the course of the study. Resilience, as measured by functioning, was assessed based on a higher level of functioning in those nonconverted individuals who had 2 years of follow-up. For measurement of resilience based on psychosis status, univariate, multivariate, or t-test analyses were performed with normally distributed continuous variables and Chi-Square analyses were used for categorical variables. For functional resilience measures, correlational analyses were performed with continuous variables, while parametric analyses were performed to compare functional outcomes with categorical protective variables. Appropriate covariates were used for each measure. Cortisol analyses were performed with sex and age as covariates. For neurocognitive data, raw scores were used with sex and age as covariates. Time of saliva collection was used as a covariate in analyses of mean morning cortisol. In the startle paradigm, average startle magnitude was used as a covariate in analyses of startle response latency. MRI variables were analyzed using age and sex as covariates. MRI indices were not adjusted for eICV due to the possible introduction of bias in brain–behavior associations.³¹ To control the false discovery rate (FDR), the Benjamini-Hochberg method was applied to each set of 34 P values for MRI surface area, cortical thickness and subcortical volume given the number of tested regions³² (q < .05) and correlations of 24-month functioning data.

Results

Demographics and Sociocultural

The NAPLS 3 sample included 572 CHR participants who were all assessed on the Psychosis Risk Calculator. A median split of the distribution was performed, and N = 334 CHR participants were included in the top portion of the sample. For the assessment of resilience using the conversion criteria, the CHR sample included a total of N = 136, N = 90 nonconverters, who were considered the “resilient” group because they had 2 years of follow-up without a known psychotic conversion, and N = 46 converters who were considered “non-resilient,” representing a 33.8% psychotic conversion rate in the sample followed for 2 years.

The nonconversion/conversion-defined resilient outcome groups did not significantly differ in individualized psychosis risk scores or most sociodemographic, positive life events, or stress sensitivity variables (Table 1). The resilient group was more likely to be Hispanic. To determine the relationship between potential protective factors and functional outcome as the resilience measure, mean comparisons and correlations were performed using 24-month GAF, Role and Social Functioning as outcome measures in the 90 participants who did not convert to psychosis after 2 years. As seen in Table 2, Desirable Life Events and Social Engagement at baseline assessment were both significantly correlated with functioning across all domains using FDR correction. Other demographics (race, ethnicity, education, income) were not significantly associated with functional outcomes.

Table 1.

Protective Factors and 2-Year Psychotic Nonconversion as a Measure of Resilience

Sociodemographic, life events, and lifestyle	Resilient	Nonresilient	t or χ2	P
Psychosis Risk Score Mean (SD)	0.27 (0.13)	0.29 (0.12)	0.92	NS
Age Mean (SD)	18.08 (4.25)	18.76 (4.14)	0.90	NS
Sex F:M	43/47	18/28	0.56	NS
Education Mean Years (SD)	10.11(3.91)	10.43 (3.77)	0.45	NS
Race (%Caucasian)	51%	47%	4.15	NS
Ethnicity (%Hispanic)	30%	15%	7.10	.03
Paternal Age at Birth Mean (SD)	32.94 (7.91)	30.92 (6.70)	1.37	NS
Maternal Age at Birth Mean (SD)	30.46 (7.34)	29.33 (6.82)	0.85	NS
%Born USA or Canada	92%	89%	0.27	NS
Paternal Education (%College or >)	49%	70%	11.19	NS
Maternal Education (%College or >)	67%	63%	7.58	NS
Household Income (%>20K)	77%	83%	6.26	NS
Desirable Life Events Mean (SD)	6.67 (2.26)	6.61 (2.62)	0.13	NS
Social Engagement (SD)	3.63 (1.23)	3.32 (1.33)	-1.37	.09
Stress Sensitivity (Stress/Life Events) Mean (SD)	3.82 (1.18)	3.78 (1.33)	0.18	NS
SAVRY Protective Factor Index Mean (SD)	4.58 (1.52)	4.28 (1.56)	1.06	.15
Prosocial Involvement	69.3%	54.3%	2.95	.09
Strong Social Support	76.1%	73.9%	0.08	NS
Strong Attachment and Bonds	84.1%	84.8%	0.01	NS
Positive Attitude Intervention/Authority	89.8%	84.8%	0.72	NS
Strong Commitment to School	61.4%	63.0%	0.04	NS
Resilient Personality Traits	77.3%	67.4%	1.53	.15
Sleep Quality %Good	58.43%	55.55%	2.19	NS
Hours of Sleep (SEM)	7.01 (0.21)	7.00 (0.31)	0.00	NS
BMI (SEM)	0.24 (0.05)	0.23 (0.07)	1.16	NS
Exercise Min Past Week (SEM)	160.71 (31.91)	165.54 (45.15)	2.44	NS

Abbreviations: BMI, body mass index (corrected for age and sex); SAVRY, scale for the assessment of violence risk in youth.

Significant analyses were bolded.

The Resilient group did not convert to psychosis within 24 months, while the NonResilient group did convert in this time.

Table 2.

Baseline Protective Factors and 2-Year Functioning as a Measure of Resilience (N = 90 Nonconverted CHR Participants)

		Baseline protective factors
24 Month Functional Outcome as a Measure of Resilience		Premorbid function	GAF	GF:R	GF:S	Protective factors index	Desirable life events	Social engagement	BMI	WRAT Premorbid IQ	WASI IQ	HVLT	BACS
	24M GAF	−0.34b	0.29b	0.51b	0.38b	0.27a	0.22a	0.31b	−0.25a	0.15	0.35b	−0.19	0.24a
	24M GF:R	−0.49b	0.31b	0.54b	0.41b	0.38b	0.40b	0.33b	−0.18	0.40b	0.46b	−0.31b	0.37b
	24M GF:S	−0.41b	0.34b	0.42b	0.56b	0.45b	0.20	0.23a	−0.08	0.29a	0.34b	-0.22a	0.26a

Abbreviations: BACS, brief assessment of cognition in schizophrenia; BMI, body mass index; GF:R, global functioning role scale; GF:S, global functioning social scale; HVLT, Hopkins verbal learning test; WASI, Wechsler abbreviated scale of intelligence; WRAT, wide range achievement test.

^a P < .04,

^b P < .01 (FDR applied).

SAVRY Protective Index

Although we⁴ have previously reported that prosocial involvement and resilient personality traits from the SAVRY Protective Factors Index were associated with resilience as measured by nonconversion across the whole NAPLS 3 sample, these variables were at a trend level in this smaller subset in favor of the resilient group (Table 1). The SAVRY Protective Factors Index was significantly correlated with all 3 domains of the functional outcome measures of resilience with FDR correction (Table 2).

Lifestyle Measures

Lifestyle measures, including quality of sleep, BMI, and exercise, did not differentiate the resilient from the nonresilient groups using conversion criteria (Table 1), but baseline BMI was positively correlated with GAF at 24 months as a resilience outcome. Sleep and exercise were not related to any of the functional outcome measures.

Premorbid, Global, Social and Role Functioning

Better baseline GAF and GF:S in the past year were associated with resilience based on psychotic nonconversion in the highest-risk group (Table 3). Using 24-month functional outcome as a measure of resilience, Total Premorbid Functioning, Baseline GAF, GF:R, and GF:S were significantly correlated with functional outcome (Table 2). The Total Premorbid Functioning correlations were driven by Late Adolescent and Early Adolescent Subtotal scores that significantly correlated with 24-month functional outcome across all 3 domains, respectively (24M GAF: R’s = −0.38, −0.27,P’s = .006, .016; 24M GF:Role: R’s = −0.63, −0.46, P’s < .001; 24M GF:Social: R’s = −0.49, −0.26, P’s < .001, = .018).

Table 3.

Protective Factors and 2-Year Psychotic Nonconversion as a Measure of Resilience

Baseline functioning mean (SD)	Resilient	Nonresilient	t	P
Premorbid Functioninga
Subtotal Childhood	0.27 (0.18)	0.24 (0.18)	0.69	NS
Subtotal Early Adolescent	0.33 (0.16)	0.32 (0.16)	−0.56	NS
Subtotal Late Adolescent	0.32 (0.20)	0.34 (0.18)	−0.54	NS
Subtotal Adult	0.32 (0.25)	0.34 (0.19)	−0.16	NS
Total Premorbid Functioning	0.31 (0.15)	0.31 (0.15)	−0.12	NS
Current GAF	48.81 (9.93)	44.48 (10.18)	2.39	.009
Current GF Role	5.83 (2.18)	5.76 (2.07)	0.47	NS
Highest GF Role Past Year	7.22 (1.80)	7.00 (1.76)	0.05	NS
Current GF Social	6.21 (1.39)	5.91 (1.60)	1.12	NS
Highest GF Social Past Year	7.09 (1.34)	6.57 (1.49)	2.01	.02

^aBetter premorbid functioning is represented by a lower number on the Premorbid Adjustment Scale.

Abbreviations: GAF, global assessment of functioning; GF, global functioning.

Significant analyses were bolded.

The Resilient group did not convert to psychosis within 24 months, while the NonResilient group did convert in this time.

Biomarkers

Neurocognition: The nonconversion defined resilient and nonresilient groups did not significantly differ on any neurocognitive measure (Table 4). Several neurocognitive domains, including the WRAT Premorbid IQ, baseline WASI IQ, HVLT, and BACS were all significantly associated with functional outcome-defined resilience after correction for FDR (Table 2).

Table 4.

Protective Factors and 2-Year Psychotic Nonconversion as a Measure of Resilience

Neurocognitive and biomarkers mean (SEM)	Resilient	Nonresilient	F	P
Neurocognitive Measures
WRAT Reading	58.51 (0.80)	57.68 91.14)	0.35	NS
WASI Vocabulary	36.49 (0.78)	34.68 (1.10)	1.81	NS
WASI Matrix Reasoning	19.89 (0.60)	19.15 (0.86)	0.49	NS
WASI IQ Estimate	100.96 (1.89)	98.22 (2.67)	0.70	NS
LNS Total	20.32 (4.98)	13.33 (7.13)	0.65	NS
AM Cortisol (SEM)	0.13 (0.01)	0.14 (0.01)	0.53	NS
Startle Measures
Startle Response Magnitude (mV)	63.95 (6.24)	107.64 (21.60)	4.32	.04
Startle Response Latency (msec)	70.19 (1.52)	68.36 (2.35)	0.57	NS
MRI Measures
Average Cortical Thickness (mm)	2.88 (0.01)	2.87 (0.02)	0.11	NS
Total Surface Area (mm 2 )	168 896.06 (1890.18)	176 246.56 (2762.23)	4.79	.03
Estimated Intracranial Volume (mm 3 )	1 540 202.23 (17 607.61)	16 112 838.48 (25 731.07)	5.41	.02
Brain Segmentation Volume (mm 3 )	11 622 151.38 (13 836.01)	1 216 195.14 (20 219.41)	4.84	.03

Abbreviations: LNS, letter number sequencing; WASI, Wechsler abbreviated scale for intelligence; WRAT, wide range achievement test.

Significant analyses were bolded.

The Resilient group did not convert to psychosis within 24 months, while the Nonresilient group did convert in this time.

Cortisol: Mean morning cortisol did not differ between groups using the conversion criteria or 24-month functional outcome as resilience criteria (Table 3).

Startle Paradigm: Startle data were available for 60 resilient and 23 nonresilient participants using nonconversion criteria. The averaged startle reactivity, or the magnitude of the startle response across all 10 pulse-alone trials, was significantly smaller in the resilient sample compared to the less resilient group who converted to psychosis (Table 3). A repeated measures ANOVA of startle magnitude across all 10 startle stimuli also differentiated the 2 groups (group effect: F[1,85] = 4.32, P < .05) with significant differences between groups using follow-up t-tests (P < .05) at each stimulus, while no difference was observed between groups in startle latency (group effect: F[1,80] = 0.10, NS). In a follow-up repeated measures ANOVA to assess the resilient and nonresilient groups compared to a normal comparison sample (N = 56) from NAPLS3 on startle magnitude, there was a significant group effect (F[2,130] = 3.05, P < .05) with no significant differences between the resilient and normal comparison groups (P = 1.0) while both significantly differed from the nonresilient sample (both P = .02) (Figure 1). Neither startle magnitude nor latency was associated with functional outcome-defined resilience.

Figure 1.

The Resilient and Normal Comparison Groups Had Reduced Startle Response Magnitude Across the Full 10-Pulse Startle Paradigm Compared to the More Reactive Nonresilient Group Who Converted to Psychosis. There Were No Significant Differences Between the Resilient CHR Group Who Did Not Convert to Psychosis and the Normal Comparison Sample

MRI: A total of 77 resilient and 37 nonresilient participants (using nonconversion criteria) had baseline MRI data available. In univariate analyses, groups were compared across average cortical thickness, total surface area, eICV, and total brain segmentation volume (without ventricles) to identify general patterns between groups. The resilient group had significantly smaller total surface area, estimated intracranial volume and brain segmentation volume compared to the nonresilient group, but there were no differences between groups in mean cortical thickness (Table 3). In multivariate analyses of the 34 brain regions, the resilient group had a pattern of smaller surface area and subcortical volume across all regions compared to the nonresilient group, with many regions differing at P < .01, P < .05, or trend level of significance prior to FDR correction (see Supplementary Table 1 and Figure 2). These group differences were most prominent in the surface area and volume of the paracentral (SA: F[1,95] = 6.45, P < .009, Vol: F[1,95] = 3.45, P < .034), precuneous (SA: F[1,95] = 5.35, P < .028, Vol: F[1,95] = 4.11, P < .009), superior parietal (SA: F[1,95] = 7.74, P < .008, Vol: F[1,95] = 5.38, P < .006) and area of the superior temporal (SA: F[1,95] = 7.16, P < .008) regions but none survived FDR correction (q < 0.05). The groups did not significantly differ in cortical thickness except in the transverse temporal region (F[1,95] = 7.19, P < .014, resilient > nonresilient), but this finding did not survive FDR correction. None of the general MRI measures (average cortical thickness, total surface area, eICV, and total brain segmentation volume) were significantly associated with resilience in the nonconverted sample as defined by functional outcomes.

Figure 2.

The Resilient CHR Group Had a Smaller Brain Surface Area and Brain Volume Across All Brain Regions Compared to the Nonresilient Group. No Region Survived FDR Correction, but Original Group Differences at a P < .01**, P < .05*, or a Trend Level of Significance P < .10 + in These Exploratory Analyses Are Displayed to Show the General Pattern of Results

Discussion

A better understanding of protective factors in early psychotic illness can inform the development of interventions that lead to greater resilience and modify the risk of a poor outcome in the early stages of illness. By focusing on individuals at the very highest risk of psychosis with both validated COPS criteria and high individual psychosis risk scores, we provide a model for studying protective factors in resilient and nonresilient outcome groups defined on the basis of nonconversion to psychosis after 24 months and better functional outcome. In addition, we present exploratory analyses in this highest-risk CHR sample from the NAPLS3 cohort as a means of identifying domains associated with resilient outcomes for further study.

Consistent with prior literature,^2,33,34 better psychosocial functioning was a strong predictor of resilience defined by lack of psychotic conversion and better 24-month functional outcome in the high-risk CHR sample. The finding of greater resilience to psychosis in CHR participants from Hispanic backgrounds also supports the importance of premorbid psychosocial environment as a protective factor that requires further investigation.³⁵ When using functional outcome as a measure of resilience, Premorbid Functioning (comprised of social, peer, scholastic, and social sexual domains) and the Protective Factor Index (comprised of measures of Prosocial Involvement, Social Support, Attachment and Bonds, Positive Attitude Toward Intervention, Commitment to School, and Resilient Personality Traits) were significantly correlated with 24-month functional outcome in the nonconverted high-risk CHR sample as were measures of Positive Life Events and Social Engagement. Family, peer, and community support are clearly important not only on an individual basis but also at a societal level in improving the mental health of vulnerable young people.

Measures of premorbid IQ and neurocognition at baseline assessment were associated with 24-month functional outcome in the high-risk nonconverted sample consistent with prior reports.³⁶ Despite prior findings that better premorbid IQ may be a protective factor against psychosis in high-risk individuals³⁷ and as a predictor of conversion/nonconversion,^38,39 we did not observe neurocognitive differences in the current sample who did not convert to psychosis. There are several possibilities as to why we did not observe the same association between neurocognition and nonconversion-defined resilience in this sample. First, because neurocognitive variables of verbal learning/memory and processing speed are part of the Psychosis Risk Calculator that was used to select the sample presented here, there was a reduced range of performance on neurocognitive measures. The median split also reduced the size of the sample and the power to detect smaller differences at one end of the distribution. Nevertheless, environmental enrichment, including early childhood education, educational support and cognitive training, may be important tools to improve the clinical and functional outcome of not only individuals at risk for psychosis but youth in general.

Baseline BMI was significantly associated with 24-month GAF score. This finding is consistent with a prior report in a CHR sample⁴⁰ that found that better metabolic profiles, including lower BMI were associated with better functioning and fewer symptoms. This finding suggests that early environments that support healthy lifestyles may contribute to better outcomes. Other demographic, sociocultural and lifestyle measures were not associated with functional outcome resilience measures or nonconversion defined resilience in the current sample. These findings (or lack thereof) are consistent with a 2019 meta-analysis of risk and protective factors carried out by Oliver et al.⁴¹ who identified only employment as a potential protective factor that was associated with reduced risk of transitioning to psychosis.

Among the biomarkers assessed, reduced/normal startle reactivity was associated with greater resilience defined as nonconversion but not functional outcome in this highest-risk sample, with startle magnitude of the resilient group identical to the normal comparison sample. We did not observe differences in startle response latency between groups as we did in the NAPLS2 study in which CHR participants who later developed psychosis (nonresilient) had longer latencies suggestive of slow neural processing.⁴² Reactivity to startle stimuli is a heritable trait⁴³ that has been found to be increased in patients with chronic schizophrenia relative to their first-degree relatives and normal controls in the Consortium on the Genetics of Schizophrenia (COGS).⁴³ The startle response magnitude is increased in humans and animals in contexts of perceived dangers.^44–47 Startle response potentiation in response to aversive stimuli has been linked to amygdala and insula activation in anxiety disorders^48–50 and is responsive to psychotherapeutic interventions.^44,51 Therefore, CHR participants with lower startle reactivity may represent a group that is less reactive to stress or perceived danger in their environment. Despite differences between groups in startle reactivity, there were no differences in cortisol values or a measure of stress reactivity, based on the number of stressors and level of distress, between the resilient and nonresilient samples. The cortisol result contrasts with prior studies in the NAPLS2 sample that demonstrated a positive association between cortisol, symptom severity, life event exposure and conversion to psychosis.^52,53 Anxiety disorders, Post Traumatic Stress Disorders and a history of childhood adversity are prevalent in the CHR population^19,54 and are potentially important targets for intervention as a means of enhancing resiliency in youth at high risk for psychosis.

The cross-sectional neuroimaging findings of smaller brain surface area and volume in the more resilient CHR sample who did not convert to psychosis differ from previous findings of increased cortical thickness, surface area and volume in CHR participants with better outcomes¹¹ as well as the observed slower rate of cortical thinning over time in CHR participants who do not convert to psychosis.²⁷ This data should therefore be interpreted with caution given that the sample is relatively small and the analyses are cross-sectional and exploratory. Interestingly, in normal development, the cortical surface area is larger in more intelligent children before decreasing at a faster rate compared to those with lower IQ.⁵⁵ In contrast, children with higher IQ have a slightly thinner cortex at age 10 with faster thinning of the cortex into adolescence. In the period of cortical development between ages 10 and 21, full cortical expansion is reached at an earlier age in more intelligent adolescents before a more accelerated contraction of surface area and a more pronounced thinning by young adulthood.⁵⁵ Given the neurodevelopmental origins of many forms of mental illness, including psychosis, it is possible that those individuals who are more resilient with higher IQ and more normative neurodevelopment may have smaller surface area and volume by late teens and at an earlier age relative to those at higher risk of a psychotic outcome. Neurodevelopmental differences are of course difficult to capture in cross-sectional analyses in a transitional age CHR population. The NAPLS¹⁹ and AMP SCZ (Accelerating Medicines Partnership Schizophrenia)⁵⁶ consortia, with larger CHR samples and longitudinal structural, resting state and diffusion tensor imaging data, are important to better assess more resilient trajectories of brain development.

Given the above findings, it is likely that therapeutic interventions that emphasize family and peer support, stress reduction, and target comorbid conditions will make a difference in bolstering resilience in this population. A focus on lifestyle factors, including healthy diet, exercise, adequate sleep, and avoidance of drugs and alcohol, will contribute to healthy brain development and better physical health. Several research groups are focusing on interventions that specifically bolster protective factors in high-risk populations. Shi et al.⁵⁷ used systemic therapy, based on systemic-constructivist and psychosocial resilience theory, in a small sample of CHR youth and noted improvement in symptoms, social support, and self-esteem in those who received the intervention. DeTore et al.^58,59 examined a Resilience Training intervention in high-risk youth focused on emotion regulation, self-perception and social interaction to address loneliness in college students. They found that Resilience Training increased resilience-related capacities and decreased loneliness that was associated with more resilience, mindfulness, and self-compassion.⁵⁸

In this exploratory study, we used the outcome measures of nonconversion to psychosis as well as 24-month functional outcomes in nonconverted participants as a means of measuring resilient outcomes. Other studies of resilience utilize symptom reduction, functional improvement, or measures of resilient characteristics⁶⁰ and a focus on these metrics may have yielded different results. Emphasizing the perspective of service users, Petros et al⁶¹ have developed a Good Outcome checklist based in part on the input of patients that emphasizes variables such as well-being, level of functioning, distress from positive symptoms as well as the absence of negative symptoms. A core outcomes assessment set that incorporates all the various outcome measures would be an important tool for future clinical trials, allowing specific questions to be addressed based on the intervention being investigated.^60,62

A number of caveats and limitations are important to acknowledge. One of the challenges of studying protective factors and resilience in CHR samples is that for many proposed measures, they are the opposite of risk factors or simply the absence of a deficit or trait. In the current model, individuals already known to be at the highest risk for psychosis based were identified for the examination of protective factors that were not part of the psychosis risk profile used to define the sample. The identification of biomarkers in which resilient individuals perform better than or equal to that of the general population is clearly important in identifying markers of neuroprotection that lead to resilience. The NAPLS consortium, as well as many CHR cohort studies, typically follow participants for 2-3 years to assess psychotic conversion and functional outcomes, and it is unknown if there are further conversions beyond this brief period or whether functional trajectories change. We are currently performing a long-term (5-20 years) follow-up of past NAPLS participants and identifying later conversions and outcomes in the past 5 years in this sample. Utilizing the long-term follow-up data may enhance our ability to identify risk and protective factors with a larger number of known conversions to psychosis. In addition, we will have detailed long-term functional outcome data in the converted sample in order to examine protective factors associated with resilience as defined by better functional outcomes in this population. The exploratory analyses in this study were performed on a relatively small sample and included many analyses. Given that for some analyses, only a few variables or biomarkers met strict criteria for identifying true group differences due to power limitations, we chose to show the data and look for trends that might be informative in a larger dataset.

The stratification model presented and assessed in the NAPLS3 cohort provides a template for the investigation of protective and risk factors in those at highest risk within the CHR cohort. With larger CHR samples, as in the AMP SCZ consortium, it will be possible to use the NAPLS Psychosis Risk Calculator or other similar tools that incorporate biomarker data to further subdivide the population to assess interventions in the highest risk group or to target known biomarker-linked mechanisms of disease. The use of subgroups within the CHR sample also provides a means of reducing heterogeneity and the high rate of false positives within the sample depending on the research questions at hand. Within the NAPLS sample, as well as AMP SCZ, there is longitudinal data that includes not only the protective measures described here but additional biomarkers (genetic, imaging, electrophysiological, inflammatory) that will also be important to investigate as we continue to refine tools to identify treatment targets. The wealth of biomarker data can then be used in precision medicine studies as we pursue the ultimate goal of prevention of severe mental illness and disability.

Funding

This study was supported by the National Institute of Mental Health (grant U01MH081984 to Dr Addington; grant U01MH081928 to Dr Stone; grant U01MH081944 to Dr Cadenhead; grant U01MH081902 to Drs Cannon and Bearden; grant U01MH082004 to Dr Perkins; grant U01MH081988 to Dr Walker; grant U01MH082022 to Dr Woods; grant U01MH076989 to Dr Mathalon; grant UO1MH081857 to Dr Cornblatt).

Conflicts of interest

None of the authors report any conflicts of interest.