Psychotic Experiences and Schizotypy in Early Adolescence Predict Subsequent Suicidal Ideation Trajectories and Suicide Attempt Outcomes From Age 18 to 38 Years

Abstract

Subclinical risk markers for schizophrenia predict suicidality, but little is known about the nature of the relationship. Suicidal ideation is often considered homogenous, but distinguishing passive from active ideation (ie, thoughts of death vs thoughts of killing oneself) and different temporal patterns may further the understanding of risk factors. We tested whether schizotypy and psychotic experiences (PEs) in early adolescence predict subsequent growth trajectories of suicidal ideation and suicide attempt outcomes. Participants were 1037 members of the population-representative Dunedin Study cohort. PE was measured at 11 years and schizotypy at 13 and 15 years. Outcomes were passive and active suicidal ideation, and suicide attempt, measured at 18, 21, 26, 32, and 38 years. Passive ideation was best represented by 2 trajectories, including persistent and transient ideation classes. Schizotypy predicted membership in the smaller persistent class (odds ratio [OR] = 1.21, P = .041), whereas PE was not associated with class membership. The probability of suicide attempts was 13.8% in the persistent ideation class, compared with 1.8% in the transient class. Active ideation was best represented by a 1-class model, the intercept of which was predicted by schizotypy (OR = 1.23, P = .015). Suicide attempts were predicted by schizotypy (OR = 1.53, P = .040) and PE (OR = 3.42, P = .046), and this was partially mediated by indirect effects via the active ideation trajectory. Findings indicate that adolescent schizotypy and PE are related to subsequent suicidal ideation and attempts. Suicidal ideation is heterogeneous, and schizotypy is specifically related to a persistent passive ideation subgroup.

Suicide is a leading cause of death worldwide,¹ yet there is a lack of knowledge of risk factors that cause the development of suicidal thoughts and behavior.^2,3 Reasons for limited progress include a heavy reliance on cross-sectional research and viewing suicidality as unitary in nature. Subclinical markers of schizophrenia risk are associated with suicidality, particularly in adolescence and young adulthood before the onset of severe disorder,⁴ but it is currently unclear whether this association reflects causal or noncausal mechanisms. Some have argued that although these are strongly related, the relationship is artifactual.^5–7 However, in most of the analyses, this relationship remains significant after adjustment for confounding factors.⁸

Psychotic experience (PE) refers to hallucinations and delusions that occur with or without reaching the threshold for clinical significance.⁹ PEs are thought to reflect a state of schizophrenia liability,¹⁰ although some suggest that isolated PEs are indicators of general psychopathology¹¹ or symptom severity.¹²Schizotypy refers to dispositional, personality, or individual differences that also reflect schizophrenia liability and tend to cluster with PE.¹² These terms and their associated concepts overlap, but in general, PE is restricted to hallucinations and delusions (or positive symptom-like experiences), whereas schizotypy includes positive (or cognitive-perceptual), disorganized, and interpersonal (negative) attributes. Although the theoretical distinctions are relatively clear, the operationalization and measurement of PE and schizotypy are blurred and inconsistent. Whereas schizotypy is often viewed as stable and PE as transient,¹³ schizotypy encompasses enduring traits as well as transitory PEs and thus has a dual nature.¹⁴ Indeed, in assessment, positive features of schizotypy are typically inferred from PEs. Therefore, rather than thinking about PE and schizotypy as distinct, PE can be considered one subset of a broad range of phenotypes of the latent construct schizotypy.

PEs are relatively common in the general population⁹ and are consistently associated with increased risk of suicidal ideation and behavior.^7,8,15 Schizotypy is also associated with suicidal thoughts and behaviors,^6,16–18 although this relationship has been examined infrequently compared with that between PE and suicidality. Interpersonal and disorganized attributes of schizotypy are related to suicidality,^6,16 indicating the importance of not restricting research on the schizotypy-suicidality relationship to PE only.

Just as schizophrenia liability is not unitary, neither is suicidality. Suicidal thinking and suicidal behavior appear to have distinct drivers.^19,20 Furthermore, many distinguish passive thoughts of dying (passive ideation) from active thoughts about harming oneself or plan for suicide (active ideation)—a distinction based on the individual’s agency, which is implied by the thought content (eg, “I wish I was dead” cf. “I want to kill myself”).²¹ Research on suicidality has focused on active ideation but evidence suggests that passive and active ideation are of equal value in predicting the risk of suicidal behavior.²² Traditionally, a linear process model of suicidality has been assumed, beginning with suicidal ideation, from which a proportion of individuals progress to making a suicide plan, and a smaller proportion progress to making a suicide attempt. However, this conceptualization has been challenged, with some arguing that there are multiple pathways to suicidal behavior or that not all steps are required for suicide to occur.²²

Although common, reliance on single time-point measures of suicidality belies meaningful differences in the temporal dynamics of suicidality.²³ Different temporal patterns of suicidal thinking likely have different predictors, courses, and responses to treatment. Suicidal subtypes, marked by different patterns of suicidal ideation, may reflect separate pathological processes.^24,25 Identifying classes of individuals with different suicidal thinking profiles, and testing the relationship with putative risk factors, may help strengthen inferences about the causes of suicidality.^26,27

Latent class growth analysis (LCGA) and growth mixture modeling (GMM) are statistical methods for distinguishing groups whose members display different trajectories of change over time. LCGA and GMM of suicidal thinking in clinical and nonclinical samples over different intervals have indicated that variability in suicidal ideation can be explained, at least in part, by the presence of latent groups with different growth trajectories (eg, persistent vs episodic).^28–31 Distinct suicidal ideation trajectory groups differ in their relationships with risk factors and the likelihoods of suicidal behavior. There is little agreement about the number or qualitative nature of ideation trajectories, possibly due to variability in the populations, developmental periods, and time spans that are studied. Currently, there is no empirical research on the relation of schizotypy to different suicidality trajectories, although there is some evidence that PEs are related to the persistence of suicidal ideation.^30,32

Here, our objective was to examine whether schizotypy and PE, measured in early adolescence, relate to 20-year trajectories of passive ideation (inferred from death ideation) and active suicidal ideation, from late adolescence to adulthood. We sought first to determine whether changes in suicidal thinking from age 18 to 38 years are represented by discrete patterns of growth. Second, we tested whether growth parameters (ie, intercepts and slopes) and membership in discrete trajectory classes were predicted by schizotypy, PE, or both. Third, we tested whether suicide attempts could be predicted by schizotypy and PE acting directly or indirectly via growth parameters or trajectory classes. We hypothesized that that schizotypy and PE would more strongly predict that the persistence of suicidal ideation, but given the lack of prior research, had no a priori hypotheses about the number of trajectory classes.

Methods

Participants

Participants were members of the Dunedin Multidisciplinary Health and Development Study (“Dunedin Study”), a longitudinal investigation of health and behavior in a population-representative birth cohort of 1037 individuals (91% of eligible births and 52% male) born between April 1, 1972 and March 31, 1973 in Dunedin, New Zealand (NZ). The longitudinal study was established at age 3 years based on residence in the province.³³ Assessments were conducted at birth and at ages 3, 5, 7, 9, 11, 13, 15, 18, 21, 26, 32, and 38 years, where 961 (95%) of the 1007 participants still alive took part. Each study member was brought to the research unit for a day of interviews and examinations. The cohort represents the full range of socioeconomic status on NZ’s South Island, and as adults match the NZ National Health and Nutrition Survey on adult health indicators, eg, body mass index, smoking, and general practitioner visits.³⁴ Study participants are primarily of NZ European ethnicity (approximately 93%). Written informed consent was obtained from participants, and the study was approved by the New Zealand Health and Disability Ethics Committee (NZ-HDEC).

Measures

Schizotypy and PEs.

At age 11 years, PEs were measured with the schizophrenia section of the Diagnostic Interview Schedule for Children (DISC-C),³⁵ which was administered by a child psychiatrist.^10,36 One-quarter of the cohort was assessed at school and did not see the psychiatrist; ratings are available for 789 cohort members. Those with and without DISC-C ratings did not differ significantly in terms of schizophrenia diagnosis, suicide attempts, or any other psychiatric diagnosis by age 38 years.³⁶ The PE items were: “Some people believe in mind reading or being psychic. Have other people ever read your mind?”; “Have you ever had messages sent just to you through television or radio”; “Have you ever thought that people are following you or spying on you?”; “Have you ever heard voices other people can’t hear?”; and “Has something ever gotten inside your body or has your body changed in some strange way?” Symptoms were rated as not present (0), possibly present (1), or definitely present (2). Participants all had data on all 5 symptoms. Children were classified as having strong symptoms (rated 1 on 2 or more symptoms, or 2 on 1 or more symptoms), weak symptoms (rated 1 on 1 symptom), or no symptoms.³⁶ At age 13 and 15 years, schizotypy was assessed using items from the Revised Behaviour Problem Checklist (RBPC),³⁷ which was completed by each participant’s caregiver. RBPC ratings were obtained for n = 831 at age 13 and n = 952 at age 15. Each variable had little missing data (M = 0.35%), which was imputed using the expectation-maximization algorithm. Details of item content and psychometric properties are reported in the supplementary material.

Suicidal Ideation and Attempts.

We quantified suicidal thinking and behavior from ratings obtained by clinically trained interviewers using the Diagnostic Interview Schedule (DIS)^38,39 at ages 18, 21, 26, 32, and 38 years. For passive ideation, at ages 18 and 21 years, the item wording was, “During the last year, have you thought a lot about death (your own, someone else’s or death in general)?” and at ages 26, 32, and 38 years, “During the last year did you think a lot about your own or someone else’s death or death in general?” For past-year active ideation at ages 18 and 21 years, the item wording was, “Have you felt so low that you thought about committing suicide?” and at ages 26, 32, and 38 years, “Did you think a lot about committing suicide?” Responses were rated as yes or no. Interviewers differentiated between suicide attempts and non-suicidal self-injury by determining whether the intent to die was present.⁴⁰ Life history calendars were used to determine the timing of suicide attempts, which were then classified as occurring either before or after turning 18 years old.

Analyses

Regression Analyses.

Logistic regression models were used to test whether PE, schizotypy, or both predicted an increased risk of reporting (from ages 18 to 38 years) passive or active suicide ideation, or suicide attempts. Cumulative variables were scored as 1 if the participant had reported the outcome at any of the 5 past-year assessments. Models were adjusted firstly for sex and secondly for sex and concurrent predictors (PE and schizotypy).

Growth Mixture Modeling.

Individual variation in passive and active suicide ideation across 5 time points (from 18 to 38 years) was examined using growth mixture modeling (GMM).⁴¹ GMM is used to identify a posteriori groups or classes with distinct growth trajectories. The effects of covariates can be tested at different levels in the model (eg, within or between groups). GMM was used to identify passive and active ideation trajectories, to establish whether growth groups exist within the study cohort, and to test the effects of schizotypy and PE on the trajectories and group membership. Within all analyses, full information maximum likelihood was used to handle missing data.

First, passive ideation trajectories and active ideation trajectories were modeled separately, without covariates. One- to 3-class models were estimated with linear, quadratic, cubic, logarithmic, and exponential terms. Within-group variance was fixed. Models were compared using the Bayesian information criterion (BIC), Lo-Mendell-Rubin adjusted likelihood ratio test (LMR-LRT), and the bootstrap likelihood ratio test (BLRT). Initial model selection was based solely on the BIC, as recommended.⁴² Subsequently, the LMR-LRT and BLRT were used in a confirmatory fashion to determine whether too many or too few classes were extracted.

Second, after selecting the best fitting growth models, covariates were added individually as predictors of growth factors (intercepts and slopes), class membership, or both. The correct placement of each covariate in the model, including whether as a class-invariant or non-invariant predictor, was determined using BIC. Covariates included sex, suicide attempts before age 18 years, psychotic experiences at age 11 years, and schizotypy ratings (ages 13 and 15 average). Then, all covariates were included in a single model.

Third, suicide attempts made from age 18 years was added as a distal outcome. Muthén⁴³ recommended including outcomes in the model itself, instead of regressing outcomes separately from the modeling phase, to allow outcomes to influence parameter estimates of the latent classes. We examined the direct and indirect effects of schizotypy and PE on attempts from age 18 years.

Statistical analyses were performed using Mplus Version 8.1 for Mac.

Results

Table 1 presents descriptive statistics for cumulative passive ideation, cumulative active ideation, cumulative passive and active ideation, and cumulative suicide attempts (ie, reported by age 38-year assessment). Of the full sample (n = 1037), 1025 individuals provided data about suicidality at 1 or more of the 5 time points and were included in analyses. Of the 134 individuals who reported suicide attempts, 51 (38.1%) occurred before age 18, with a further 29 (21.6%) occurring between ages 18 and 21 years, 27 (20.1%) between 22 and 26 years, 20 (14.9%) between 27 and 32 years, and 7 (5.2%) between 33 and 38 years.

Table 1.

Number of Individuals Reporting Cumulative Suicidality Outcomes (ie, Binary Measure of Whether the Participant Had Reported the Outcome Once or More at any of 5 Past-Year Assessments) Between Ages 18 and 38 Years (n = 1025)

	Males (n = 530)		Females (n = 495)
Outcome (cumulative)	N	%	N	%
Passive ideation	245	46.2	291	58.8
Active ideation	93	17.5	129	26.1
Passive and active ideation	75	14.2	109	22.0
Suicide attempt	59	11.1	75	15.1

In regression analyses adjusted for sex, there was strong evidence that schizotypy scores and PE predicted suicide attempts by age 38 (table 2). Results did not differ significantly when individuals who had attempted suicide before age 18 were excluded from analyses.

Table 2.

Logistic Regression of Cumulative Suicidality Outcomes (ie, Binary Measure of Whether the Participant Had Reported the Outcome Once or More at any of 5 Past-Year Assessments From Age 18 to 38 Years) on to Schizotypy and Psychotic Experiences Measured in Early Adolescence

	Model 1a (Adj. Sex)			Model 2a (Adj Sex, Concurrent Predictors)
Outcome (cumulative)	OR	95% CI	ΔR2	OR	95% CI	ΔR2
Passive ideation
Psychotic experienceb	1.38	0.97, 1.97	.004	1.34	0.94, 1.91	.004
Schizotypyc	1.06	0.95, 1.17	.001	1.11	0.98, 1.26	.005
Active ideation
Psychotic experienceb	1.35	0.91, 2.00	.004	1.31	0.88, 1.95	.003
Schizotypyc	1.14	1.02, 1.29	.007	1.09	0.95, 1.24	.003
Suicide attempt
Psychotic experienceb	1.98	1.30, 3.03	.021	1.86	1.21, 2.86	.017
Schizotypyc	1.28	1.13, 1.45	.025	1.23	1.07, 1.43	.017

Note: R² values are pseudo R². Bold values indicate P < .05.

^aModel 1 is adjusted for sex only, and Model 2 is adjusted for sex and concurrent predictors (ie, the other of schizotypy or psychotic experience).

^bObtained at age 11 years. Measured by interviewer rating of presence of symptom (none, weak, or strong) on the Diagnostic Interview Schedule for Children (DISC-C).

^cAverage of ratings obtained at ages 13 and 15 years. Measured by sum score of parent ratings on 8 schizotypy-related items from the Revised Behaviour Problem Checklist (RBPC).

Schizotypy predicted active ideation from age 18 to 38, but not when PE was added to the model. Age 11 PE did not predict active ideation. Neither schizotypy nor PE predicted passive ideation from age 18 to 38.

Growth Mixture Modeling

Passive Ideation.

A 2-class quadratic model was chosen as the best fitting model for passive ideation (figure 1). The 3-class cubic model also provided a good fit to the data; however, as the 2-class quadratic model is more parsimonious, it was retained (supplementary material). Of the 1-, 2-, and 3-class quadratic models, the LMR-LRT and BLRT were significant (P < .0001) for the 1- vs 2-class comparison, strongly favoring the 2-class solution over the 1-class model (table 3). For the 2- vs 3-class comparison, P-values were .022 and 1.0, respectively, suggesting no improvement was achieved by going beyond 2 classes.

Fig. 1.

Diagrams of the growth mixture models for passive (A) and active (B) suicidal ideation. SA, suicide attempts; PE, psychotic experience; SZT, schizotypy.

Table 3.

Model Fit and Classification Indices for Passive and Active Ideation

Model	Free	LL	AIC	BIC	ssBIC	LMR-LRT	BLRT	Entropy
Passive ideation
1C quad	6	−2013.13	4038.266	4067.860	4048.804
2C quad	10	−1968.57	3957.140	4006.465	3974.704	.000	.000	.340
3C quad	14	−1966.18	3960.358	4029.412	3984.946	.022	1.000	.546
Active ideation
1C lin	5	−1109.97	2229.937	2254.599	2238.719
2C lin	8	−1105.15	2226.308	2265.767	2240.359	.013	.004	.700
3C lin	11	−1100.51	2223.016	2277.273	2242.336	.001	.000	.731

Note: LL, loglikelihood; AIC, Akaike information criterion; BIC, Bayesian information criterion; ssBIC, sample-size adjusted BIC; LMR-LRT, Lo-Mendell-Rubin adjusted likelihood ratio test; BLRT, bootstrapped likelihood ratio test; C, class(es); lin, linear; quad, quadratic; cub, cubic; exp, exponential; log, logarithmic. Boldface indicates the best fit. LMR-LRT and BLRT are tests of the hypothesis that k C provides a better fit than k – 1 C.

Figure 2 illustrates the 2 growth trajectories in the sample of 1025 individuals. The largest class was characterized by a peak risk of suicidal ideation in early adulthood, followed by a decline in risk. This transient ideation class represented 79.5% of the total sample. The smaller class was characterized by a relatively stable probability of ideation over the 5 time points. This persistent ideation class represented 20.5% if the sample.

Fig. 2.

Best fitting models of passive and active suicidal ideation across 5 time points (18 to 38 years) for the sample (n = 1025). (A) a 2-class quadratic model of passive ideation. Solid line represents the transient ideation class (n = 815, 79.5%), and the dashed line represents the persistent ideation class (n = 210, 20.5%). (B) a 1-class linear model of active ideation (n = 1025).

Association With Covariates.

When tested univariately, attempt history (before 18) predicted higher intercepts in both classes (odds ratio [OR] = 2.55, P =.012). Males were significantly more likely to be in the larger transient class (OR = 2.62, P = .005). PE was not associated with class membership (OR = 0.71, P = .160) and did not predict any growth factors. Schizotypy predicted membership in the smaller persistent class (OR = 1.21, P = .041). Associations remained significant when all covariates were included in the model (attempt history predicting growth factors, and sex, PE, and schizotypy predicting class membership; figure 1).

Association With Suicide Attempt Outcome.

When controlling for direct and indirect effects of all covariates, in the persistent class, the probability of suicide attempt from age 18 was 13.8% (95% CI = 8.6, 21.2). In the transient class, the probability of suicide attempts after age 17 was 1.8% (95% CI = 0.8, 3.7).

Active Ideation

A linear model was chosen as the best fitting model for active ideation (figures 1 and 2), as fit estimates (BIC) were lower compared with models with cubic, quadratic, logarithmic, and exponential terms. The LMR-LRT and BLRT were significant (P < .05) for 1- vs 2-class comparison and for the 2- vs 3-class comparisons (P ≤ .001; table 3). However, the 2- and 3-class linear models did not improve the overall fit of the model, with higher BIC values than the 1-class model, suggesting heterogeneity in growth should not be decomposed into qualitatively distinct growth profile groups. Thus, it was decided to proceed with the 1-class model for further analyses.

Association With Covariates.

When entered into the model individually, having made an attempt before 18 predicted a higher intercept of the active ideation trajectory (OR = 9.29, P < .001), and being male predicted a lower intercept (OR = 0.57, P = .008). PE did not predict slope or intercept growth factors. Schizotypy significantly predicted a higher intercept on its own (OR = 1.23, P = .015) and with sex and prior attempts included in the model (OR = 1.20, P = .038), but was not significant when PE was included in the model (OR = 1.13, P = .458).

Association With Suicide Attempt Outcome.

There were significant direct and indirect effects of covariates on suicide attempts from age 18. Being male did not significantly predict suicide attempts from 18 (OR = 0.69, P = .505). Attempts before 18 predicted attempts from 18 (OR = 121.63, P < .001), and this relationship was significantly mediated via the intercept of the active ideation trajectory (P < 0.001). When this indirect effect was modeled, there was no evidence of a direct effect of attempts before 18 on attempts from 18 (OR = 1.18, P = .825). Of the total effect of attempts before 18 on attempts from 18, 96.5% was mediated, principally via the intercept growth factor (92.5%, P = .003), not the slope growth factor (4.0%, P = .495).

Schizotypy was a significant predictor of attempts from 18 (OR = 1.53, P = .040), as was age 11 PE (OR = 3.42, P = .046), but specific indirect effects of schizotypy and PE via the active ideation trajectory or specific direct effects were not statistically significant. This indicates that the effect of schizotypy and PE on suicide attempts after 18 was likely due to a combination of direct effects and indirect (mediated) effects via active ideation.

Discussion

Using growth mixture modeling, we found that a 2-class quadratic model best represented trajectories of passive ideation, inferred from death ideation ratings, from age 18 to 38. One class was characterized by a peak in risk of ideation around age 21, followed by a rapid decline in risk (transient ideation), whereas the other was characterized by the higher, relatively stable risk of ideation across the 5 time points (persistent ideation). When covariates were included in the passive ideation model, schizotypy predicted membership in the persistent ideation class. PE did not predict class membership, inconsistent with previous research findings that PEs were related to the persistence of ideation.^30,32 The difference in findings between schizotypy and PE was unexpected as the schizotypy and PE measures overlapped significantly in terms of content. This discrepancy may reflect the differences in time point, types of measurement, observer effects, the smaller sample of people with age 11 data, or a combination of these factors.

Membership in the persistent ideation class was associated with a higher probability of attempting suicide after age 18 than membership in the transient ideation class, consistent with previous findings.²⁹ Although suicide attempts from 18 were treated as an outcome, attempts occurred at any point from age 18 to 38; ie, the predictor and outcome were not temporally contiguous but overlapping. This derivation of the attempt outcome score may affect the interpretation of findings.

For active ideation, a 1-class linear model was determined to be the best fitting model. This indicates that, in our sample, variability in active ideation appears not to be due to the existence of distinct trajectory groups. In contrast, others have reported finding 2 or 3 growth trajectories for active ideation when modeling a binary measure of active ideation in a general population sample.^31,44 The difference between the current study findings and previous research may reflect the longer time period (total duration and assessment intervals) examined in the current study. Patterns of growth likely differ depending on timeframe and time windows between assessments. Given suicidal states are time-limited,⁴⁵ the long time windows between assessments in the present study may be less sensitive to these dynamic processes.

Schizotypy significantly predicted a higher risk of experiencing active ideation, whereas PE did not. However, when schizotypy and PE were both included as covariates, the relationship between schizotypy and active ideation was nonsignificant. This diminished effect may reflect overlap in the content of schizotypy and PE measures or observer differences (caregiver vs interviewer). Neither schizotypy nor PE predicted the trajectory of active ideation. These findings indicate that people who report schizotypy may have a higher risk of active ideation, but the course of active ideation across time is not different from people without schizotypy.

Suicide attempts from age 18 were predicted by attempts before 18, schizotypy, and PE. The association between suicide attempts before age 18 and suicide attempts from 18 was almost entirely mediated via increased risk of active ideation and not through any direct effects on suicide attempts. The influence of schizotypy and PE on suicide attempts from 18 appeared to be partly due to direct effects, and partly mediated through the influence of schizotypy and PE on the active ideation trajectory, though these associations were not statistically significant. This indicates that schizotypy and PE increase the risk of suicide attempts both directly and by increasing the risk of experiencing active ideation, which, in turn, increases the risk of suicide attempts.

Strengths, Limitations, and Conclusions

This study has several strengths. The sample is a representative birth cohort with very low rates of attrition and the method involved a long follow-up period with multiple assessments of suicidality. Schizotypy and PE were measured with different modalities of assessment (caregiver- and interviewer-reported), which strengthens inferences as each type of assessment introduces different types of measurement error.^9,46

Study limitations include, first, that the time between assessments was relatively long (3 to 6 years) and irregular, whereas suicidality questions referred to past-year experiences. Irregularity of assessment intervals can compromise the accurate detection of growth trajectories.⁴⁷ Irregularity may have been one of several measurement factors that contributed to relatively low classification accuracy of the GMM model for passive ideation. Second, passive ideation was inferred from questions about preoccupation with death, that may have been endorsed in the absence of suicidal ideation. A preoccupation with death and dying need not entail thoughts about one’s own death; the question was not restricted to a desire for death, which is regarded as the core feature of passive ideation.^21,48 It is feasible that a broader death ideation construct could relate more generally to psychopathology, rather than specifically to suicidality. However, passive ideation did predict suicide attempts, indicating that these constructs were related. Third, the active ideation item wording differed slightly across assessments. We cannot rule out that wording differences affected responding, introducing measurement error. Further, measures may have been interpreted differently at different developmental stages or participants at different ages may have different thresholds for responding. Fourth, although we selected RBPC items to quantify schizotypy on the basis of semantic content (eg, alignment with Meehl’s definition of schizotypy), 2 items we selected have been used by others to assess acting out (“Not liked by others; is a ‘loner’ because of aggressive behavior”) and anxious withdrawal (“Shy, bashful”) albeit within clinical samples (supplementary material). Nevertheless, we found that the RBPC item subset predicted schizophrenia diagnosis in adulthood, indicating it is a valid measure of liability for schizophrenia. Last, other psychopathology variables were not included as covariates in the models. These were omitted because of the difficulty delineating nuisance covariates from key variables of interest. Indeed, the statistical control of variables that are not true confounders may generate misleading results.⁴⁹ That is, including covariates that are peripheral to the research question but nevertheless integral within a causal framework (eg, childhood maltreatment) risks removing meaningful variance upon which the research question depends. This notwithstanding, by including attempt history as a predictor in models, the influence of unmeasured or omitted confounders is mitigated. Over control will not yield a better test of the research question.

The findings lead to implications for understanding whether schizotypy and suicidality are causally related. Findings support the idea that there is heterogeneity in patterns of suicidal ideation, and that this variability can be partially explained by latent groups or suicidal subtypes.^24,50 For passive ideation, schizotypy related not simply to a quantitative increase in the probability of ideation, but specifically to a qualitatively distinct subgroup of passive ideation, indicating some specificity of the schizotypy-suicidality relationship. In future research, investigators should consider ideation trajectory types. If schizotypy causally influences only one type of suicidal ideation—distinguished by trajectory—effects may not be distinguished when different trajectory types are commingled.

Schizotypy and psychotic experiences measured in adolescence are related to subsequent suicidal ideation and attempts. People who display schizotypal indicators, including hallucination- and delusional-like experiences, interpersonal difficulties, and disorganized speech, in early adolescence are more likely to experience persistent passive ideation across their lifetime. Experience of transient passive ideation is common, but a persistent pattern of chronic passive ideation was associated with a higher likelihood of having attempted suicide. Schizotypy was also associated with an increased risk of active ideation but did not predict the course of active ideation across the lifetime.

Funding

The Dunedin Multidisciplinary Health and Development Research Unit is supported by the New Zealand Health Research Council and has also received funding from the New Zealand Ministry of Business, Innovation and Employment. Funding support was also received from the US National Institute of Aging grant (AGO32282) and the UK Medical Research Council grant (MR/KOO38IX). Additional support was provided by the Jacobs Foundation.