Risk Factors for Late-Onset Psychosis: A Case-Control Study

Abstract

Background and Hypothesis

The onset of schizophrenia occurs after the age of 40 in up to 20% of cases. We aim to depict risk factors for first-episode psychosis after the age of 40 by comparing late-onset psychosis (LOP) patients to healthy age-matched controls.

Study Design

In this case-control study using electronic health records, 142 individuals aged 40–65 years with an encounter for a first episode of psychosis between 2013 and 2021 were included. Four controls (N = 568) were matched to each case on age, sex, race, and year of encounter. Potential risk factors for the primary analysis were captured via structured data and text-mining of medical notes. Conditional logistic regression models were used to assess the odds of LOP with potential risk factors.

Study Results

After adjusting for all variables in the main analysis, odds for LOP were increased by immigration (OR 3.30, 95% CI, 1.56–6.98), depression (OR 3.58, 95% CI, 2.01–6.38), anxiety (OR 2.12, 95% CI, 1.20–3.75), cannabis use (OR 3.00, 95% CI, 1.36–6.61), alcohol use disorder (OR 5.46, 95% CI, 2.41–12.36), polysubstance use (OR 4.22, 95% CI, 1.30–13.7), severe trauma (OR 2.29, 95% CI, 1.08–4.48), and caregiver burden (OR 15.26, 95% CI, 3.85–60.48).

Conclusions

Life stressors along with the effects of substance use and other psychiatric conditions may confer some risk to the development of LOP. Replication is required in independent prospective studies. Further research is necessary to truly parse out which of these factors belong on the causal pathway.

Introduction

The onset of schizophrenia occurs after the age of 40 in up to 20% of cases.¹ Despite numerous studies investigating the differences between late-onset schizophrenia (LOS) and early-onset schizophrenia (EOS), findings in this area are inconsistent.² Higher rates of persecutory delusion, decreased prevalence of family history of first-degree relative with psychosis, and female preponderance are well established as consistent features of LOS.¹ However, there are discrepancies in how symptom profiles and symptom severity differ between these two groups with some studies finding LOS to have less severe symptomatology and a different psychopathological profile than EOS while others studies have found EOS and LOS to have similar symptom profiles and no differences in severity.² This lack of consensus can be contributed to a number of factors. Firstly, the field did not reach agreement regarding the definition, diagnosis, and nomenclature for LOS until the year 2000.³ Secondly, comparisons of LOS and EOS yield a skewed view of the risk factors for LOS since differences between these two groups are likely driven by differences in behavior due to age. Finally, when age effects are accounted for, in studies where LOS and EOS patients are matched on age, findings may be due to illness chronicity as individuals with EOS have dealt with the illness for much longer than their LOS counterparts.²

Furthermore, just as age might drive some of the differences seen between LOS and EOS, the contrasting composition by sex between these groups may account for differences as well. Despite men having a greater risk of psychosis in general,⁴ women have consistently comprised 60%-87% of LOS samples in past research.⁵ Women exhibit a triphasic nature of onset for psychosis with an initial peak in late adolescence/early adulthood (10–23 years of age), a second peak in mid age (24–40 years of age), and the final incident peak occurring after the age of 40.⁶ This final peak aligns with the time when menopause typically occurs in women which lends support to the neuroprotective properties of estrogen.⁷ Conversely, men display a biphasic onset with only early and late incident peaks.⁶ Existing research suggests that the symptoms of psychosis in general manifest differently in men and women with women experiencing more affective disturbances and men experiencing more severe negative symptoms and social withdrawal.^4,5,8

In the current study, we aim to overcome these limitations to identifying risk factors for a first episode of psychosis after the age of 40 by using a comparison group comprised of healthy age-matched controls. When designing a case-control study, to avoid selection bias, controls should be drawn from the “base cohort,” which represents individuals in the source population during the time they are at risk to become cases.⁹ The base cohort consists of individuals between the age of 40–65 that reside in the greater Boston area. Patients with early-onset psychosis (EOP) differ from the base cohort in terms of age. Therefore, any differences in potential risk factors between late-onset psychosis (LOP) and EOP may be due to age effects (eg, substance use disorders more common in younger individuals). Potential risk factors for a first episode of LOP will be evaluated.

In a secondary analysis, LOP patients will be compared to patients hospitalized for their first episode of EOP to clarify differences between these two groups and how clinical features of LOP might be distinct from that of EOP.

Ultimately, beside female sex, the risk factors for LOP are unclear in the current literature. The use of an age-matched control group with a case-control design will shed light on the factors that may lead to greater risk of the outcome of LOP for the individuals of interest, that is, healthy individuals in their 40s without a previous episode of psychosis. Identifying risk factors is essential as it may allow for potential points of prevention and intervention for cases of first-episode psychosis occurring after the age of 40.

Methods

Data Sources

The Mass General Brigham (MGB) Research Patient Data Registry (RPDR) is a data repository that includes electronic health record (EHR) data for all encounters at MGB. MGB is a large healthcare system that includes two large academic hospitals, a free-standing psychiatric hospital (McLean Hospital), and a network of community hospitals and outpatient clinics in the New England region of the United States. Each patient is assigned a unique identifier that tracks care across all MGB sites. Structured data includes dates of encounters, International Classification of Disease (ICD) diagnosis codes, demographics, health insurance plan, medications, and laboratory values. Sex is obtained by self-report when individuals register for MGB encounters; self-report of sex does not distinguish between assigned sex at birth or gender identity. Unstructured data consists of narrative notes for encounters. This study was approved by the MGB Institutional Review Board with a waiver of informed consent according to 45 US Code of Federal Regulations 46.116 and is reported in accordance with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for case-control studies.¹⁰

Study Population and Case-Control Definition

For cases, individuals between the age of 35 and 65 years with an encounter for a first episode of psychosis between 2015 and 2021 were included. A psychiatrist with expertise in psychotic disorders (LVM) selected ICD codes that were likely to index an encounter for an episode of psychosis.^11–13 Using structured data, patients were required to have one of these ICD codes for a principal diagnosis of psychosis which was assigned by their treating clinician (see Supplementary Table 1) and have at least 3 outpatient encounters prior to the first diagnosis of psychosis, with at least one encounter within the past year to select patients with high EHR continuity. Patients with ICD diagnosis codes for dementia or organic CNS disease were excluded (Supplementary Table 2), leaving 575 patients. Medical records were reviewed to exclude patients with a prior history of psychosis. We confirmed selection of patients with true LOP and accuracy of ICD codes by reviewing medical records where first onset of psychosis was reported by the patient and an informant (eg, family member, outpatient psychiatrist). After conducting this chart review, 142 patients with true LOP remained. There were three patients with LOP with symptoms that started prior to the study start date of 2015, starting as early as 2013, where onset of symptoms was clearly documented in notes. None of the LOP patients were younger than 40 (see Supplementary Figure 1 for flow chart).

RPDR data contains 12 diagnosis fields. Controls were patients without any past or current ICD diagnosis for psychosis in any of the diagnosis fields. Four controls were matched to each case on age, sex, race, and year of encounter, where the encounter closest to the matched date of the onset of LOP was selected. Controls had to have at least three outpatient visits prior to the match date, with at least one encounter within a year of the match date. To account for psychosis not identified by ICD codes, we performed text-mining; details can be found in a prior study.¹³ Briefly, keywords related to psychosis were curated by clinicians with expertise in psychosis. Every time one of the keywords appeared in notes within the past year, the keyword and surrounding context were extracted. These phrases were manually reviewed by raters leading to the identification of three control patients with psychosis, who were removed and replaced with controls without psychosis. A list of keywords can be found in Supplementary Table 3.

For secondary analysis, patients with LOP were compared with a group of patients with EOP. Patients with EOP were individuals between the age of 16 and 35 admitted to McLean Hospital for an initial hospitalization for a first episode of psychosis from an external study.¹³ Two controls with EOP were matched to each case on sex, race, and affective versus nonaffective psychosis.

Potential Risk Factors

The risk factors included in our model were predetermined and selected under the theoretical framework that psychosocial stressors can precipitate an episode of psychosis.¹⁴ A combination of previously established risk factors for schizophrenia at any age (eg, cannabis use and other substance use disorders, immigration, prescription amphetamine use), previous comparisons between EOS and LOS, and common psychosocial events were used to build our model.^14–16

For the primary analysis, age at encounter, sex, and year of encounter were used for matching while the covariates in the model included, race/ethnicity, insurance type (proxy of socioeconomic status), alcohol-use disorder, active cannabis use, and opioid or illicit stimulant use (cocaine or methamphetamines). Opioids and illicit stimulants were combined due to collinearity, as all patients using illicit stimulants in the LOP group were all also using opioids.

Binary covariates for the presence or absence of past psychiatric diagnoses and medications at the onset of psychosis were also included. Past psychiatric diagnoses included attention deficit hyperactivity disorder (ADHD), depression, anxiety disorders, and post-traumatic stress disorder (PTSD). Psychiatric medications on date of onset or matched encounter included antidepressants, benzodiazepines, and prescription amphetamines (all within the past 30 days). Mood stabilizers and antipsychotic medication use were not considered as risk factors as they were deemed to be the consequence of outcome.¹⁷ For example, antipsychotics may be prescribed for patients exhibiting recent onset of symptoms for psychosis. All covariates were assessed as present if either structured data reported presence of covariate (eg, ICD codes for psychiatric diagnoses, urine toxicology data) or if the covariate was identified by text-mining for relevant terms in notes. Text mining in notes from the date of the encounter and the past month was performed to identify covariates not captured in the structured data. Raters reviewed the phrases for each patient to code presence or absence of each covariate (eg, “daily cannabis user” versus “no history of cannabis use”).

Additionally, binary covariates for the presence or absence of psychosocial factors preceding initial onset of symptoms were included. Psychosocial factors were identified by text mining of notes. Information pertaining to factors that preceded symptom onset were contained in the history section of notes from the day of encounter and various notes from encounters during the previous 90 days. These factors included marital status, recent death or encounter for grief related to death of a family member, other severe trauma, unemployment or job difficulties, retirement, unstable housing, recent divorce or relationship difficulties, and caregiver burden. Psychosocial factors were not considered as risk factors if they were deemed to be the consequence of psychosis, that is, job difficulties that occurred after onset of psychosis were not counted. See Supplementary Table 4 for terms used in text mining for psychosocial factors.

Covariates for the secondary analysis were identical to the primary analysis with the addition of presence or absence of a first degree relative with psychosis and the removal of marital status and retirement as psychosocial factors. Academic difficulties were also included when coding the covariate for unemployment or job difficulties. Because of the difficulty of determining whether phrases related to psychosocial difficulties were precipitants or sequelae of EOP, notes were manually reviewed to capture psychosocial stressors that preceded EOP.

Secondary Analysis: Symptom Profiles

In the secondary analysis, notes were manually reviewed to record and compare symptoms of psychosis between LOP and EOP patients. Binary covariates for the presence or absence of specific psychotic symptoms were included. Symptoms included nondelusional paranoia, auditory hallucinations, visual hallucinations, olfactory hallucinations, and delusions divided by type: persecutory, grandiose, thought dissemination, somatic, delusions of parasitosis, and misidentification; no other types of delusions (eg, erotomanic) were identified in either LOP or EOP groups.

Statistical Analysis

Primary Analysis: LOP vs Outpatient Controls

A first episode of LOP was the primary outcome. A conditional logistic regression model was used to assess the odds of LOP comparing individuals with LOP to age-matched controls (also matched on sex, race/ethnicity, and year of encounter) and included the set of potential risk factors/covariates defined above. Unadjusted and adjusted odds ratios (ORs) with 95% confidence intervals (CIs) are presented.

Secondary Analyses: LOP vs Matched EOP

Symptom profiles of psychotic symptoms were compared between LOP and EOP patients using chi-square analyses.

A conditional logistic regression model was used to assess the odds of LOP, comparing individuals with LOP to those with EOP matched on sex, race/ethnicity, and affective vs nonaffective psychosis. Potential risk factors/covariates were included in the model. Unadjusted and adjusted OR with 95% CI, are presented.

For both the primary and secondary analysis, missingness of data was restricted to two variables (immigration and family history of first-degree relative with psychosis) and was less than 1%, where we assumed that the absence of any phrases for these covariates indicated they were not present. Due to this small amount, there was little concern for missing data impacting the results of this study.

Exploratory Analysis

Prior to matching LOP to EOP patients, we compared LOP patients to a full dataset of 1,221 unmatched patients with first onset psychosis admitted to McLean to examine risk factors for LOP vs EOP. This analysis was performed to identify whether factors related to psychosis that were included as matching variables, such as sex and race/ethnicity, differed between LOP and EOP patients. A logistic regression model was used to assess the odds of LOP with potential risk factors. Unadjusted and adjusted ORs with 95% CIs are presented.

Exploratory Subgroup Analyses

Subgroup analyses were performed separating groups by sex. The primary analysis was replicated in men and women separately. The secondary analysis was also replicated, comparing individuals with LOP to those with EOP, in men and women separately. Notably, for the replication of the secondary analysis, psychosocial stressors were collapsed into a single variable as collinearity was observed in the male group.

Finally, three additional analyses investigating the impact of sex on symptom profiles of psychotic symptoms were performed using chi-square analyses. Profiles of psychotic symptoms were compared between LOP and EOP patients in men and women separately. Additionally, symptoms profiles were compared between men and women in analyses restricted to LOP patients.

Results

Primary Analysis: LOP vs Outpatient Controls

Demographic and clinical characteristics of cases and controls can be found in table 1. After adjusting for all variables, increased odds for LOP were associated with immigration (OR 3.30, 95% CI, 1.56–6.98), depression (OR 3.58, 95% CI, 2.01–6.38), anxiety (OR 2.12, 95% CI, 1.20–3.75), cannabis use (OR 3.00, 95% CI, 1.36–6.61), alcohol-use disorder (OR 5.46, 95% CI, 2.41–12.36), opioid/illicit stimulant use (OR 4.22, 95% CI, 1.30–13.70), other severe trauma (OR 2.29, 95% CI, 1.08–4.48), and caregiver burden (OR 15.26, 95% CI, 3.85–60.48). Additionally, multivariable analysis identified death of a family member (OR 0.33, 95% CI, 0.12–0.90) as associated with decreased odds for LOP. Overall, the amount of variance explained by the model was 38% (pseudo R²). See table 2 for unadjusted and adjusted primary analysis.

Table 1.

Demographic and Clinical Characteristics of Late-Onset Psychosis (LOP) Cases and Matched^a Controls

Variable	LOP (N = 142)	Controls (N = 568)	P
Age, mean (SD)	53.7 (6.0)	53.6 (5.9)
Male, no (%)	44 (31.0)	176 (31.0)
Race/ethnicity
White	91 (64.1)	364 (64.1)
Asian	6 (4.2)	24 (4.2)
Black or African American	21 (14.8)	84 (14.8)
Hispanic	23 (16.2)	92 (16.2)
Decline/unknown	1 (0.7)	4 (0.7)
Public/private insurance	58 (40.9)	157 (27.6)	.002
Immigration	43 (30.3)	117 (20.6)	.014
ADHD	12 (8.5)	21 (3.7)	.016
Depression	82 (57.8)	130 (22.9)	<.001
Anxiety	66 (46.5)	121 (21.3)	<.001
PTSD	14 (9.9)	5 (0.9)	<.001
Other psychiatric diagnosis	11 (7.8)	11 (1.9)	<.001
Prescription amphetamineb	12 (8.5)	20 (3.5)	.011
Antidepressantb	60 (42.3)	154 (27.1)	<.001
Benzodiazepineb	41 (28.9)	109 (19.2)	.011
Cannabisb	29 (20.4)	38 (6.7)	<.001
Alcohol-use disorderb	30 (21.1)	22 (3.9)	<.001
Polysubstance useb	14 (9.9)	9 (1.6)	<.001
Marital status			.119
Married	65 (45.8)	319 (56.2)
Divorced	29 (20.4)	80 (14.1)
Separated	45 (31.7)	158 (27.8)
Widowed	3 (2.1)	11 (1.9)
Death of family	9 (6.3)	41 (7.2)	.714
Other severe trauma	38 (26.8)	39 (6.9)	<.001
Unemployed or job difficulties	58 (40.9)	118 (20.8)	<.001
Retirement	11 (7.8)	21 (3.7)	.037
Unstable housing	7 (4.9)	8 (1.4)	.009
Recent divorce or relationship difficulties	12 (3.7)	21 (8.5)	.016
Caregiver burden	8 (5.6)	5 (0.9)	<.001

The bolded values are any p-value that is less than 0.05 and therefore significant.

^aControls matched 4:1 to LOP cases on age, sex, race/ethnicity, and year.

^bPast month.

Table 2.

Conditional Logistic Regression Results: LOP vs Matched^a Controls

Column1	Unadjusted		Adjusted
Variable	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
Public/private insurance	1.92 (1.28, 2.86)	.001	1.13 (0.65, 1.95)	.67
Immigration	2.67 (1.51, 4.72)	.001	3.30 (1.56, 6.98)	.002
ADHD	2.49 (1.17, 5.28)	.018	0.94 (0.25, 3.60)	.93
Depression	4.78 (3.19, 7.17)	<.001	3.58 (2.01, 6.38)	<.001
Anxiety	3.29 (2.21, 4.90)	<.001	2.12 (1.20, 3.75)	.009
PTSD	13.44 (4.41, 40.94)	<.001	3.50 (0.79, 15.44)	.10
Other psychiatric diagnosis	4.22 (1.78, 9.97)	.001	1.96 (0.53, 7.33)	.32
Prescription amphetamine use in past 30 days	2.49 (1.19, 5.20)	.015	1.84 (0.52, 6.53)	.35
Antidepressant	2.09 (1.40, 3.12)	<.001	0.68 (0.36, 1.26)	.22
Benzodiazepine	1.80 (1.15, 2.81)	.009	0.74 (0.39, 1.41)	.36
Cannabis	3.90 (2.23, 6.83)	<.001	3.00 (1.36, 6.61)	.006
Alcohol-use disorder	7.11 (3.80, 13.30)	<.001	5.46 (2.41, 12.36)	<.001
Polysubstance use	6.74 (2.82, 16.12)	<.001	4.22 (1.30, 13.70)	.017
Marital status
Married
Divorced	1.88 (1.12, 3.17)	.018	0.85 (0.42, 1.75)	.67
Separated	1.49 (0.95, 2.33)	.084	1.10 (0.61, 1.95)	.76
Widowed	1.35 (0.36, 5.05)	.653	2.08 (0.44, 9.86)	.36
Death of family	0.87 (0.41, 1.84)	.713	0.33 (0.12, 0.90)	.031
Other severe trauma	5.44 (3.18, 9.29)	<.001	2.29 (1.08, 4.84)	.030
Unemployed or job difficulties	2.66 (1.79, 3.96)	<.001	1.35 (0.78, 2.33)	.29
Retirement	2.57 (1.10, 6.02)	.030	2.50 (0.79, 7.89)	.12
Unstable housing	3.74 (1.30, 10.76)	.014	0.77 (0.18, 3.23)	.72
Recent divorce or relationship difficulties	2.46 (1.16, 5.20)	.019	1.58 (0.56, 4.46)	.39
Caregiver burden	6.40 (2.09, 19.56)	.001	15.26 (3.85, 60.48)	<.001

The bolded values are any p-value that is less than 0.05 and therefore significant.

^aControls matched 4:1 to LOP cases on age, sex, race/ethnicity, and year.

Secondary Analysis: Symptom Profiles of LOP vs EOP

Demographic and clinical characteristics of LOP and matched EOP patients can be found in table 3. For symptomatology, LOP patients were more likely to have persecutory delusions (P < .001), somatic delusions (P = .001), and delusions of parasitosis (P < .001) compared to EOP patients. EOP patients, on the other hand, had higher rates of grandiose delusions (P < .001), thought dissemination (P < .001), nondelusional paranoia (P < .001), and auditory hallucinations (P = .032). See table 4 for a comparison of symptom frequency between LOP and matched EOP patients.

Table 3.

Demographic and Clinical Characteristics of LOP and Matched EOP Patients

Variable	LOP (N = 142)	EOP (N = 284)	P
Age, mean (SD)	53.7 (6.0)	24.3 (5.0)
Male, no (%)	44 (31.0)	88 (31.0)
Race/ethnicity
White	91 (64.1)	182 (64.1)
Asian	6 (4.2)	12 (4.2)
Black or African American	21 (14.8)	42 (14.8)
Hispanic	23 (16.2)	42 (14.8)
Other/decline/unknown	1 (0.7)	6 (2.1)
Public/private insurance	58 (40.9)	38 (13.4)	<.001
Immigration	43 (30.3)	48 (16.9)	.001
First degree relative psychosis	12 (8.5)	24 (8.5)	1.000
ADHD	12 (8.5)	60 (21.1)	.001
Depression	82 (57.8)	96 (33.8)	<.001
Anxiety	66 (46.5)	61 (21.5)	<.001
PTSD	14 (9.9)	18 (6.3)	.194
Other psychiatric diagnosis	11 (7.8)	42 (14.8)	.038
Amphetamine use in past 30 days	12 (8.5)	36 (12.7)	.194
Antidepressant	60 (42.3)	78 (27.5)	.002
Benzodiazepine	41 (28.9)	36 (12.7)	<.001
Cannabis	29 (20.4)	123 (43.3)	<.001
Alcohol-use disorder	30 (21.1)	44 (15.5)	.148
Polysubstance use	14 (9.9)	37 (13.0)	.342
Death of family	9 (6.3)	18 (6.3)	1.000
Other severe trauma	38 (26.8)	95 (33.5)	.160
Unemployment/job or academic difficulties	58 (40.9)	143 (50.4)	.064
Unstable housing	7 (4.9)	4 (1.4)	.031
Recent divorce or relationship difficulties	12 (8.5)	36 (12.7)	.194
Caregiver burden	8 (5.6)	4 (1.4)	.013

The bolded values are any p-value that is less than 0.05 and therefore significant.

Table 4.

Comparison of Symptom Frequency LOP vs Matched EOP

Symptoms	LOP (N = 142)	EOP (N = 284)	P
Affective psychosis, no (%)	37 (26.1)	74 (26.1)
Delusions	120 (84.5)	229 (80.6)	.327
Persecutory	95 (66.9)	123 (43.3)	<.001
Grandiose	13 (9.2)	69 (24.3)	<.001
Though dissemination	7 (4.9)	53 (18.7)	<.001
Somatic	23 (16.2)	17 (6.0)	.001
Parasitosis	14 (9.9)	4 (1.4)	<.001
Misidentification	6 (4.2)	8 (2.8)	.442
Paranoia	11 (7.7)	88 (31.0)	<.001
Auditory hallucinations	52 (36.6)	135 (47.5)	.032
Visual hallucinations	23 (16.2)	48 (16.9)	.854
Olfactory hallucinations	7 (4.9)	5 (1.8)	.062

The bolded values are any p-value that is less than 0.05 and therefore significant.

Secondary Analysis: Risk Factors for LOP vs EOP

After adjusting for all variables, increased odds for LOP compared to EOP were found for public insurance (OR 5.97, 95% CI, 2.83–12.59), depression (OR 3.14, 95% CI, 1.68–5.87), anxiety (OR 2.03, 95% CI, 1.04–3.97), benzodiazepine use (OR 2.50, 95% CI, 1.09–5.76), alcohol-use disorder (OR 2.87, 95% CI, 1.29–6.42), and caregiver burden (OR 9.90, 95% CI, 1.24–78.14). Additionally, multivariable analysis identified decreased odds of LOP compared to EOP for ADHD (OR 0.19, 95% CI, 0.04–0.90), other psychiatric diagnosis (OR 0.24, 95% CI, 0.08–0.73), cannabis use (OR 0.38, 95% CI, 0.19–0.75), other severe trauma (OR 0.38, 95% CI, 0.18–0.78), and unemployment/job or academic difficulties (OR 0.41, 95% CI, 0.23–0.75). Overall, the amount of variance explained by the model was 43% (pseudo R²). See table 5 for unadjusted and adjusted secondary analysis.

Table 5.

Conditional Logistic Regression: LOP vs Matched^a EOP

	Unadjusted		Adjusted
Variable	Odds Ratio (95% CI)	P	Odds Ratio (95% CI)	P
Public/private insurance	4.82 (2.84, 8.19)	<.001	5.97 (2.83, 12.59)	<.001
Immigration	2.42 (1.43, 4.09)	.001	0.82 (0.39, 1.73)	.607
First degree relative psychosis	1.00 (0.48, 2.06)	1.000	0.88 (0.28, 2.73)	.826
ADHD	0.33 (0.17, 0.64)	.001	0.19 (0.04, 0.90)	.035
Depression	2.54 (1.69, 3.82)	<.001	3.14 (1.68, 5.87)	<.001
Anxiety	3.39 (2.11, 5.43)	<.001	2.03 (1.04, 3.97)	.039
PTSD	1.69 (0.79, 3.64)	.178	1.40 (0.41, 4.82)	.586
Other psychiatric diagnosis	0.44 (0.21, 0.93)	.031	0.24 (0.08, 0.73)	.012
Amphetamine use in past 30 days	0.62 (0.31, 1.25)	.183	2.17 (0.40, 11.84)	.372
Antidepressant	1.99 (1.29, 3.08)	.002	1.30 (0.67, 2.56)	.439
Benzodiazepine	2.84 (1.68, 4.80)	<.001	2.50 (1.09, 5.76)	.031
Cannabis	0.32 (0.20, 0.53)	<.001	0.38 (0.19, 0.75)	.005
Alcohol-use disorder	1.45 (0.87, 2.41)	.153	2.87 (1.29, 6.42)	.010
Polysubstance use	0.73 (0.38, 1.40)	.344	0.53 (0.21, 1.37)	.192
Death of family	1.00 (0.44, 2.30)	1.000	0.83 (0.25, 2.79)	.766
Other severe trauma	0.73 (0.47, 1.14)	.165	0.38 (0.18, 0.78)	.009
Unemployment/job or academic difficulties	0.70 (0.47, 1.04)	.075	0.41 (0.23, 0.75)	.004
Unstable housing	3.50 (1.02, 11.96)	.046	1.67 (0.25, 11.36)	.599
Recent divorce or relationship difficulties	0.62 (0.31, 1.25)	.183	1.07 (0.43, 2.69)	.884
Caregiver burden	4.88 (1.28, 18.6)	.020	9.90 (1.24, 78.14)	.030

The bolded values are any p-value that is less than 0.05 and therefore significant.

^aEOP matched 2:1 to LOP cases on sex, race/ethnicity, and affective/nonaffective psychosis.

Exploratory Analysis: Risk Factors for LOP vs EOP (Full Dataset)

Demographic and clinical characteristics of LOP and the full dataset of EOP patients can be found in Supplementary Table 5. After adjusting for all variables, odds for LOP compared to odds of EOP using the full dataset of patients with EOP (unmatched) were increased by public insurance (OR 3.31, 95% CI, 2.14–5.12), depression (OR 2.17, 95% CI, 1.40–3.37), anxiety (OR 2.19, 95% CI, 1.38–3.48), and benzodiazepine use (OR 1.75, 95% CI, 1.05–2.92). Additionally, multivariable analysis identified male sex (OR 0.31, 95% CI, 0.20–0.47), Asian race (OR 0.31, 95% CI, 0.12–0.83), ADHD (OR 0.27, 95% CI, 0.11–0.66), other psychiatric diagnosis (OR 0.42, 95% CI, 0.20–0.88), and cannabis use (OR 0.27, 95% CI, 0.17–0.45) as associated with decreased odds for LOP compared to EOP. See Supplementary Table 6 for unadjusted and adjusted exploratory analysis.

Exploratory Analysis: LOP vs Outpatient Controls With Collapsed Psychosocial Variable

The primary analysis was replicated in which the seven psychosocial variables were collapsed into a single variable that indexed any psychosocial stressor. After adjusting for all variables, increased odds for LOP were associated with immigration (OR 3.10, 95% CI, 1.52–6.33), depression (OR 3.29, 95% CI, 1.90–5.69), anxiety (OR 2.16, 95% CI, 1.24–3.75), cannabis use (OR 2.71, 95% CI, 1.27–5.76), alcohol-use disorder (OR 5.06, 95% CI, 2.36–10.88), opioid/illicit stimulant use (OR 4.21, 95% CI, 1.31–13.51), and any psychosocial stressor (OR 2.63, 95% CI, 1.59–4.33). See Supplementary Table 7 for unadjusted and adjusted primary analysis with a collapsed psychosocial variable.

Exploratory Analysis: LOP vs EOP With Collapsed Psychosocial Variable

The secondary analysis was replicated in which the seven psychosocial variables were collapsed into a single variable that indexed any psychosocial stressor. After adjusting for all variables, increased odds for LOP were associated with public insurance (OR 5.52, 95% CI, 2.78–10.97), depression (OR 2.43, 95% CI, 1.37–4.31), and alcohol-use disorder (OR 2.46, 95% CI, 1.16–5.20). Additionally, multivariable analysis identified decreased odds of LOP for ADHD (OR 0.18, 95% CI, 0.04–0.75), other psychiatric diagnosis (OR 0.25, 95% CI, 0.09–0.72), cannabis use (OR 0.31, 95% CI, 0.16–0.60), and any psychosocial stressor (OR 0.42, 95% CI, 0.23–0.78). See Supplementary Table 8 for unadjusted and adjusted secondary analysis with a collapsed psychosocial variable.

Exploratory Subgroup Analysis: LOP vs Outpatient Controls by Sex

After adjusting for all variables in the subgroup analysis for females, odds for LOP were associated with immigration (OR 2.94, 95% CI, 1.15–7.49), depression (OR 2.77, 95% CI, 1.38–5.55), anxiety (OR 2.03, 95% CI, 1.05–3.95), cannabis use (OR 2.87, 95% CI, 1.02–8.06), alcohol-use disorder (OR 10.83, 95% CI, 3.14–37.27), other severe trauma (OR 3.54, 95% CI, 1.41–8.90), retirement (OR 2.29, 95% CI, 1.08–4.48), and caregiver burden (OR 16.87, 95% CI, 3.25–87.47). Additionally, multivariable analysis identified death of a family member (OR 0.17, 95% CI, 0.04–0.62) as associated with decreased odds for LOP.

After adjusting for all variables in the subgroup analysis for males, increased odds for LOP were associated with immigration (OR 13.48, 95% CI, 2.26–80.33), depression (OR 25.18, 95% CI, 3.98–159.07), PTSD (OR 177.58, 95% CI, 2.00–15764.49), alcohol-use disorder (OR 22.36, 95% CI, 2.76–181.22), widowed marital status (OR 132.25, 95% CI, 1.16–15075.88), and recent divorce or relationship difficulties (OR 13.24, 95% CI, 1.03–170.71). Additionally, multivariable analysis identified benzodiazepine use (OR 0.11, 95% CI, 0.01–0.95) as associated with decreased odds for LOP. Estimates for males were less precise due to smaller number of males with LOP in the study. See Supplementary Tables 9 and 10 for unadjusted and adjusted subgroup analyses.

Exploratory Subgroup Analysis: Secondary Analysis

After adjusting for all variables in a subgroup analysis comparing females with LOP and EOP, increased odds for LOP compared to EOP were found for public insurance (OR 5.11, 95% CI, 2.21–11.82) and depression (OR 2.37, 95% CI, 1.19–4.73). Additionally, multivariable analysis identified decreased odds of LOP compared to EOP for ADHD (OR 0.02, 95% CI, 0.0004–0.75), other psychiatric diagnosis (OR 0.18, 95% CI, 0.05–0.68), and cannabis use (OR 0.32, 95% CI, 0.23–1.14).

After adjusting for all variables in the subgroup analysis comparing males with LOP and EOP, increased odds for LOP compared to EOP were found for public insurance (OR 8.41, 95% CI, 1.80–39.29), and alcohol-use disorder (OR 8.41, 95% CI, 1.32–53.61). Additionally, cannabis use (OR 0.17, 95% CI, 0.04–0.73) decreased odds for LOP. See Supplementary Tables 11 and 12 for unadjusted and adjusted subgroup analyses of LOP vs EOP.

See Supplementary Tables 13–15 for additional subgroup analyses for the impact of sex on psychotic symptom profiles in the current study.

Discussion

This study investigated risk factors for a first episode of psychosis after age 40 by comparing patients with LOP to healthy age-matched controls. The primary analysis found nine covariates associated with increased risk of LOP including immigration status, depression, anxiety, cannabis use, alcohol-use disorder, opioid/illicit stimulant use, other severe trauma, and caregiver burden.

Overall, these findings are consistent with existing literature. Previous studies have shown that immigrants have higher incidence of schizophrenia compared to nonimmigrant populations.^18,19 Furthermore, previous research has identified elevated levels of depression, anxiety disorders, and substance use in those with schizophrenia compared to the general population.²⁰

Previous studies have shown lower rates of substance use, cannabis use, and alcohol use in LOS compared to EOS.^21,22 However, by using a healthy age-matched comparator group, our study showed that cannabis use, alcohol-use disorder, and opioid/illicit stimulant use conferred an increased risk for LOP, consistent with multiple studies demonstrating an increased risk of developing schizophrenia with cannabis use.²³ Additionally, a systematic review investigating a wide age range identified cannabis, opioids, and illicit stimulants use to be associated with schizophrenia spectrum disorders (SSD).²⁴

Congruent with the current study’s findings, previous studies have shown that past trauma confers increased risk of SSD and that LOS patients experienced significantly more trauma within the past three years compared to healthy controls.^22,24

Our PTSD findings require replication. PTSD (OR = 3.50) increased odds for LOP but was not significant due to small number of patients (n = 19) with PTSD. This finding is consistent with severe trauma (OR = 2.29) increasing odds of LOP which was significant due to larger number of patients (n = 77) with severe trauma. Similarly, in the secondary analysis, there was no significant difference in rates of PTSD between LOP (n = 14) and EOP (n = 18) due to the small number of patients while severe trauma was sufficiently powered and yielded a significant difference.

We uncovered a novel association between caregiver burden and risk of LOP. Increased risk of LOP in caregivers may be due to negative consequences beside the burden of care itself such as social isolation, financial stress, feelings of lack of agency, and depression.²⁵

The secondary analysis, comparing LOP patients to matched EOP patients, found six covariates associated with an increased risk for LOP including public insurance, depression, anxiety, benzodiazepine use, and caregiver burden. Conversely, ADHD, other psychiatric diagnosis, cannabis use, other sever trauma, and unemployment/job or academic difficulties were found to decrease risk for LOP. Many of these findings are likely driven by differences in age between groups. Additionally, the findings that cannabis use and ADHD decrease the odds of LOP might be attributed to higher prevalence of substance use and ADHD in younger age groups.^26,27 Conversely, caregiver burden increasing risk of LOP may be attributed to older individuals being more likely to have elderly parents needing extensive care compared to their younger counterparts.

Novel findings for the secondary analysis included public insurance, depression, anxiety, and benzodiazepine use increasing the odds of LOP with other severe trauma, and unemployment/job or academic difficulty decreasing the odds of LOP compared to EOP. Although these findings are significant, they should be interpreted with caution as they may be due to differences in ascertainment of groups (all inpatient EOP vs inpatient and outpatient LOP), severity of illness and the associated consequences (potentially more comorbidities), as well as differences due to age.

Of note, while many studies have shown weaker genetic predisposition for psychosis in LOP patients as indexed by family history,^1,21,28 the current study failed to find a significant difference in family history for LOP and EOP patients.

Considering the inconsistent reporting of differences in symptom profiles between LOS and EOS, our study hoped to shed light on these discrepancies. Overall, our findings were congruent with evidence that these groups exhibit different symptoms profiles. LOP patients exhibited a higher frequency of persecutory delusions, somatic delusions, and delusions of parasitosis. Conversely, EOP patients were found to have more grandiose delusions, thought dissemination, nondelusional paranoia, and auditory hallucinations. Our findings are consistent with previous research demonstrating that LOS patients exhibit more persecutory delusions and decreased prevalence of hallucinations compared to EOS.^2,29 Findings in the current study of greater somatic delusions or delusions of parasitosis in LOP patients are consistent with previous research indicating that onset of delusions of parasitosis generally occurs in older adults (56.9 years).³⁰ Prior work has found increased prevalence of tactile hallucinations in those with LOS.^3,31 Tactile hallucinations may precipitate a delusional interpretation, where patients might interpret the subjective sensation as evidence of a parasitic infestation.³²

In exploratory analyses, when comparing LOP patients to full set (unmatched) EOP patients, male sex decreased the odds for LOP which is a well-established demographic feature of LOP in the literature.^31,33,34

In models that collapsed psychosocial stressors into a single variable, associations found in the primary and secondary analysis remained significant and did not substantively change our conclusions, indicating the number of covariates in our model did not impact stability.

With our sample being roughly 67% female, we performed additional subgroup analyses by sex. Results for females in the primary analysis were consistent with those from the entire cohort. Replicating the primary analysis in males yielded inconclusive results with wide confidence intervals as the analysis was unpowered. The current study did not have data on hormone levels or postmenopausal status in women. Replication of current work with the addition of this data might be a valuable direction for future research.

Similarly, comparisons of LOP vs EOP patients by sex did not indicate substantive risk factor differences in women. Results in males were inconclusive due to an underpowered analysis. Finally, no clear sex differences were found in symptomatology within LOP patients beside females experiencing significantly more somatic delusions compared to LOP males.

The primary limitation of this study was the use of EHR as notes can vary in detail and suffer from inconsistent documentation which can lead to misclassification of covariates. Reverse causality may be responsible for findings, as risk factors could be due to the prodromal phase of LOP (eg, cannabis’ effects precipitating an episode of psychosis versus patients self-medicating with cannabis for prodromal symptoms). Additionally, some differences between LOP and EOP patients may be due to differences in the LOP group’s mix of inpatient/outpatient status compared to an all inpatient EOP group. Previous research has found that when comparing EOS to LOS patients with inpatient status, LOS had more severe thought disorder in symptoms while another study in outpatients found LOS patients to exhibit less severe psychotic symptoms.² Unmeasured confounding may lead to biased estimates, as notes may lack data on factors that influence onset of LOP, such as education and other social determinants of health. Furthermore, association of caregiver burden and polysubstance use with LOP were based on a small number of patients; these finding require replication in independent prospective studies. Finally, as white individuals comprised 64% of the sample, findings may not generalize to areas with different demographics.

Despite these limitations, there were several strengths of this study. To our knowledge this is the largest study investigating LOP (n = 142). Additionally, EHR allowed for sourcing data from medical notes on the day of encounter and any note from the previous 90 days prior to onset of psychosis. Information contained in both the history section of notes from the day of encounter which indexed events preceding symptom onset and a variety of notes from encounters during the previous 90 days were used to refine the timing of factors. This combination allowed for more confidence in using the label risk factors rather than comorbid factors. Moreover, while previous studies have often compared LOP to EOP or healthy controls, this study’s contribution to the existing literature is use of an age-matched control group that allows for identification of risk factors without bias due to age effects. Comparisons between younger EOP patients and older LOP patients have the potential to misrepresent risk factors since these groups’ behavior is inherently different due to age. Some studies have compared LOP vs age-matched EOP patients, but these studies are limited by the confounder of chronicity of disease with EOP patients suffering from illness much longer than LOP patients.² In the current study, LOP and EOP patients presented for a first episode of psychosis, overcoming the limitation of differences in illness chronicity between LOP and EOP patients in prior research.

In conclusion, many important risk factors exist for LOP. Despite a lack of consensus in the literature, life stressors along with the effects of cannabis use, other substance use, and psychiatric conditions are associated with increased odds of LOP. Further research is necessary to parse out whether these factors are on causal pathway or are related to the prodromal phase LOP. Future research should focus on conducting sex specific analyses to reveal how sex plays a role in shaping the epidemiology of LOP. Understanding risk factors for LOP allows for potential points of prevention and intervention for cases of first-episode psychosis occurring after the age of 40.

Supplementary Material

Supplementary material is available at https://academic.oup.com/schizophreniabulletin/.

Conflict of Interest

Dr. Moran is an employee of Sage Therapeutics (start date after manuscript was written and edited), unrelated to the subject of this study. All other authors have declared that there are no conflicts of interest in relation to the subject of this study.

Funding

This work was supported by the National Institutes of Mental Health (R01MH122427).