Early Prodromal Symptoms Can Predict Future Psychosis in Familial High-Risk Youth

Abstract

Background

Efforts to predict psychosis in individuals at high risk for schizophrenia have focused on the identification of sub-threshold clinical criteria and neurobiological markers, including neuropsychological assessment, structural and functional brain imaging, and psychophysiological testing. We sought to evaluate the relative utility of psychosis proneness measures for prospective prediction of psychotic disorders in a group of young relatives at familial risk for schizophrenia.

Methods

We examined the receiver operating characteristics of sub-threshold symptoms in predicting conversion to psychosis in a group of 97 young first- and second-degree relatives of persons with schizophrenia over a 2-year period. Towards this end, we utilized the Structured Interview of Prodromal Symptoms to derive measures of two of the four Scale of Prodromal Symptoms subscales (positive and disorganized) and the Chapman Magical Ideation and Perceptual Aberration scales. These four measures were, together, taken to reflect a putative index of psychosis proneness.

Results

Eleven of the 97 subjects developed a psychotic disorder over 2 years of follow-up. Seventeen of the 97 subjects tested positive on this index of psychosis proneness at baseline and of these 10 converted to psychosis. The sensitivity and specificity of the test were 91 percent and 92 percent respectively. The positive predictive value of the test was 59 percent and its negative predictive value was 99 percent. Addition of measures of cognitive or social function to the index decreased its predictive ability, reducing its specificity and/or sensitivity.

Conclusions

A relatively simple set of clinical measures can be utilized to prospectively identify familial high-risk individuals who convert to psychosis with high specificity and sensitivity. Implications for the proposed addition of an “Attenuated Psychosis Syndrome” in DSM-5 are discussed.

INTRODUCTION

Schizophrenia and related psychotic disorders typically begin in adolescence, often progress, and can cause considerable disability (Harrison et al., 2001; Tandon et al., 2010). There has been renewed interest in preventing or delaying onset of psychosis in those at high risk for developing the illness (Lee and McGlashan, 2005; McGorry, 2008). Accurate prediction algorithms have faced a number of hurdles, including non-specific early psychopathology and the multifactorial etiology of psychotic illness. Despite these challenges, studies over the past decade have generated optimism about identifying individuals at high risk. Thus far, tools to identify individuals at high risk for developing a psychotic disorder have included extensive clinical testing, neuropsychological testing, structural and functional brain imaging, and psychophysiological testing (Correll et al., 2010; Keshavan et al., 2011; Lawrie et al., 2008).

Identification of these individuals is crucial for preventing or delaying the onset of psychosis through a range of interventional approaches (Correll et al., 2010; Liu et al., 2010). At the same time, more precise risk stratification holds the potential to reduce risks of intervention in individuals who are not destined to cross the psychosis threshold. Reflecting this promise, the fifth revision of the Diagnostic and Statistical Manual for Mental Disorders (DSM-5) is considering inclusion of a risk syndrome for psychosis (Bruijnzeel and Tandon, 2011; Carpenter, 2009) based on the presence of attenuated psychotic symptoms. Although the proposal is controversial (Corcoran et al., 2010; Drake and Lewis, 2010; Woods et al., 2010), its utility rests in part on the availability of reliable and valid clinical measures with high predictive value. It is worth mentioning here the distinction between familial high risk and ultra-high risk (clinical high risk). While there has been evidence to support the accuracy of attenuated psychotic symptoms as a predictor of future conversion to psychosis in the clinical high risk population (McGorry, Nelson et al., 2009), relatively few studies have examined this question in the familial high risk group.

We therefore sought to utilize a simple clinical measure of psychosis proneness and evaluate its utility in prospectively predicting individuals at familial high risk for developing a psychotic disorder. We studied 97 first- and second-degree nonpsychotic relatives of individuals with a diagnosis of schizophrenia or schizoaffective disorder over a 2-year period to evaluate whether baseline clinical characteristics would effectively identify those who develop a psychotic disorder over this period. Results of the Structured Interview for Prodromal Symptoms (SIPS, Miller et al., 2002) and the Wisconsin Schizotypy scales (Chapman et al., 1978) at baseline were utilized to define the index of psychosis proneness.

MATERIALS AND METHODS

Subjects

The study was conducted at the Western Psychiatric Institute and Clinic, Pittsburgh. Participants were 199 first- and second-degree relatives of probands with a diagnosis of schizophrenia or schizoaffective disorder. Relatives were recruited by first approaching patients via their treating clinicians or directly through advertisements. Diagnoses of schizophrenia or schizoaffective disorder were confirmed in the index relatives using the Structured Clinical Interview for DSM Disorders (SCID, Spitzer et al., 1992). Relatives were assessed at baseline, year 1, and year 2 using the SCID for the possible emergence of psychopathology. All participants signed informed consent following a full explanation of the study. For participants < 18 years of age, the consent was provided by the parent or guardian and the subjects provided informed assent. The study was approved by the University of Pittsburgh Institutional Review Board.

For subjects to be included in the final sample, they needed to have been free of psychotic illness at baseline, received both the SIPS and the Chapman Wisconsin Schizotypy scales assessments at baseline, and be followed up for two years. Two of the 199 relatives had a prior psychotic illness at baseline and were consequently excluded from the final sample. Subjects without baseline SIPS (18) or Chapman schizotypy (9) asssessments, or neither (40), were excluded from the final sample. Thirty three of the remaining subjects were not followed up for two years, leading to the final sample size of 97 familial high-risk subjects who had complete follow-up data. Conversion to a psychotic disorder was confirmed by repeat SCID interviews, and consensus conferences with senior clinicians (MSK, DM) using all available clinical data, obtained after the conversion.

Measures

Subjects received the Structured Interview for Prodromal Symptoms (SIPS) and the Chapman Schizotypy Scales at baseline. The 19-item SIPS assesses symptoms in four domains (positive, negative, disorganized, and general symptoms) and rates symptom severity on a 0-6 scale on the Scale of Prodromal Symptoms (SOPS), with scores of 3-5 indicating prodromal or attenuated psychotic symptoms. Positive and disorganization symptoms are considered putative psychotic symptoms (Tandon et al., 2009) and are included in the proposed DSM-5 definition of “Attenuated Psychosis Syndrome” whereas negative and general symptoms are not (American Psychiatric Association, 2011). The four Chapman Schizotypy Scales include perceptual aberration, magical ideation, revised social anhedonia, physical anhedonia,. Of these, perceptual aberration and magical ideation define a positive schizotypy factor which has found to predict likelihood of psychosis in longitudinal studies of psychosis-prone subjects (Chapman et al., 1978; Kwapil et al., 2008; Lenzenweger, 1994). Conversion to a DSM-IV psychotic disorder was assessed by annual psychiatric assessments including SCID interviews over the 2-year period (American Psychiatric Association, 2000).

Both attenuated psychotic symptoms and positive schizotypy are found to indicate psychosis-proneness in high risk individuals. Even though they are correlated, the SIPS is a clinician rated measure and the Chapman scales are self-rated, and are therefore likely to provide somewhat unique, though overlapping indices of psychosis proneness. We therefore combined these measures to create a composite index of psychosis-proneness on the basis of four equally weighted items: two SOPS subscales (positive and disorganization) and the two Chapman scales (perceptual aberration and magical ideation), each with thresholds normed for the sample. The thresholds were determined by plotting the sample and calculating the minimum score for the top quartile of scores in that test (Table 2). For each test, a score of 0 was given if the individual scored in the bottom three quartiles and a score of 1 was assigned if the individual tallied in the top quartile of scores on that test. The sum of these 0/1 scores were taken for each subject as the total “psychosis proneness score”.

Table 2.

	Range Of Scores Observed	Mean	Standard Deviation	Upper QuartiLe (Threshold score)
SOPS
Positive	0-16	1.6	3.1	2
Disorganized	0-8	1.1	2.0	2
Chapman
Perceptual  Aberration	0-21	3.0	4.0	4
Magical  Ideation	0-21	5.0	4.0	7

Data Analysis

We evaluated basic receiver operating characteristics of the proposed criteria, including sensitivity and specificity of the psychosis-proneness index, as well as its positive (PPV) and negative (NPV) predictive values for predicting conversion to psychosis in an individual at familial high-risk.

The positive and negative likelihood ratios (LR+ and LR− respectively), which are a means of using sensitivity and specificity data to calculate the direct implications of the index in changing psychosis-risk estimates in individual subjects were also assessed (Sackett et al., 2000). In addition, predictive values and likelihood ratios, unlike sensitivity and specificity, reflect the prevalence of the illness in the population under study, which is of particular importance in samples with a low base-rate of illness.

RESULTS

Of the 97 familial high-risk individuals with complete data, 11 developed a psychotic disorder over the 2-year period. Of the 11 converters, one developed schizophreniform disorder (295.4), four developed schizoaffective disorder (295.7), two developed schizophrenia undifferentiated type (295.9), and four developed psychosis not otherwise specified (298.8) (American Psychiatric Association, 2000). The characteristics on the 11 converters and 86 non-converters are compared in Table 2. There were no differences in average age, gender distribution, or race between the two groups.

We decided to begin with a total score of ≥ 2 on the 0-4 index as an indicator of psychosis-proneness. Seventeen of the 97 individuals scored ≥ 2 on the psychosis-proneness index at baseline: of these, 10 converted to psychosis over the 2-year period and 7 were non-converters. Of the 80 individuals scoring < 2, 1 converted and 79 did not.

Sensitivity and specificity are statistical measures of the performance of a binary classification test, also known in statistics as classification function. Sensitivity (also called recall rate in some fields) measures the proportion of actual positives which are correctly identified as such. Specificity measures the proportion of negatives which are correctly identified. These two measures are closely related to the concepts of type I and type II errors. For any test, there is usually a trade-off between the measures. This trade-off can be represented graphically as an Receiver Operative Characteristics (ROC) curve (Newman et al., 2001). An ROC curve demonstrates several things: a) It shows the tradeoff between sensitivity and specificity (any increase in sensitivity will be accompanied by a decrease in specificity); b) the closer the curve follows the left-hand border and then the top border of the ROC space, the more accurate the test; c) the closer the curve comes to the 45-degree diagonal of the ROC space, the less accurate the test; d) the slope of the tangent line at a cut-off point gives the likelihood ratio (LR) for that value of the test; e) the area under the curve is a measure of test accuracy (Newman et al., 2001).

The sensitivity of the psychosis-proneness index in predicting likelihood of conversion to psychosis was 91 percent (10/11) and its specificity was 92 percent (79/86). The positive predictive value of the psychosis-proneness index was found to be 59 percent and its negative predictive value found to be 99 percent. The likelihood ratio positive was found to be 11.4 and the likelihood ratio negative was found to be 0.2. The formulae for these measures can be seen in Table 3.

Table 3.

	Psychosis Positive	Psychosis Negative
Psychosis Proneness Positive	True Positive:  10	False Positive:  7	Positive Predictive Value: $\frac{TP}{TP+FP}=\frac{10}{10+7}=0.59$
Psychosis Proneness Negative	False Negative:  1	True Negative:  79	Negative Predictive Value: $\frac{TN}{TN+FN}=\frac{79}{79+1}=0.99$
	$\frac{TP}{TP+FN}=\frac{10}{10+1}=0.91$	$\frac{TN}{TN+FP}=\frac{79}{79+7}=0.92$

We then compared the utility of our test to that of other indices employed to predict likelihood of converting to psychosis in familial high-risk samples. When we utilized the 30 percent decline in GAF from baseline criterion for predicting conversion to psychosis in a genetic high-risk sample (Woods et al., 2009; Yung et al., 2008), we found both a lower sensitivity [45 percent (5/11)] and a lower specificity [87 percent (75/86)] of the test in predicting conversion to psychosis. When we utilized the 15 percent decline in GAF from baseline criterion for predicting conversion to psychosis in a genetic high-risk sample (Keshavan et al., 2011), we found both a lower sensitivity [72 percent (8/11)] and a lower specificity [73 percent (63/86)] of the test in predicting conversion to psychosis. This might be because the GAF has been shown to have poor inter-rater reliability (Grootenboer et al., 2011). When we utilized the 15 percent decline in cognitive function (i.e. increase in perseverative error score) criterion based on performance on the Wisconsin Card Sorting test (Bhojraj et al., 2011), we also found both a lower sensitivity [63 percent (7/11)] and a lower specificity [59 percent (50/86)] of the test. If we utilized a cut-off score of ≥ 3 on the 0-4 psychosis proneness index, the sensitivity of the test in predicting conversion to psychosis decreased to 64 percent (7/11) whereas its specificity increased to 96 percent (83/86). If we added the negative and general SOPS scales and utilized a cut-off score of ≥ 2, the sensitivity of the test in predicting conversion to psychosis remains 91 percent (10/11) whereas its specificity decreased to 64 percent (55/86). If instead we added the negative and general SOPS scales and utilized a cut-off score of ≥ 3, both the sensitivity of the test in predicting conversion to psychosis decreased to 64 percent (7/11) and its specificity decreased to 85 percent (73/86). Using Chapman scores as a predictor independently, we found a sensitivity of 55 percent (6/11) and a specificity of 70 percent (60/86); if we applied SOPS as a predictor independently, we found a sensitivity of 45 percent (5/11) and a specificity of 59 percent (51/86) (Table 4). All of these data were plotted on an ROC Curve (Figure 1) to evaluate the effect of utilizing different criteria for psychosis proneness.

Table 4.

Test	True Positive	TPR	True Negative	FPR
Psychosis Proneness 1	11	100	45	52
Psychosis Proneness 2	10	91	79	8
Psychosis Proneness 3	6	54	81	6
Psychosis Proneness 4	2	18	83	3
Psychosis Proneness + GS + NS 2	10	91	55	36
Psychosis Proneness + GS + NS 3	7	63	73	15
GAF 10	8	73	61	29
GAF 30	5	45	75	13
COG 15	7	63	50	41
COG 30	4	36	78	9

Figure 1.

DISCUSSION

The emergence of the early intervention model in psychotic disorders has appropriately generated optimism about the ability to “reshape the course of psychotic disorder” (McGorry, 2009; National Research Council, 2009) because it is founded on a new understanding of psychotic vulnerability in conjunction with practical preventive strategies already established in general health care (Keshavan et al., 2011; McGorry, 2010; Ruhrmann et al., 2009). The success of this strategy is, however, predicated upon the ability to precisely identify individuals at significantly elevated risk of developing a psychotic disorder and the availability of safe and effective interventions that can be specifically provided to those who are likely to benefit. Our study addressed the possibility of predicting the likelihood of developing psychosis in a group of 97 individuals already at elevated risk of such a transition because they were first- or second-degree relatives of persons with a diagnosed psychotic disorder. We found our “psychosis-proneness” index to have high specificity and sensitivity in predicting the likelihood of developing a psychotic disorder over a 2-year period in this group, with a likelihood ratio of a positive test=11.4, which indicates a test of high utility (Sackett et al., 2000). This index was found to provide higher specificity and sensitivity than other tests hitherto utilized to predict likelihood of conversion to psychosis in familial high-risk individuals and clinical high risk samples (Table 5). Additionally, the index which we utilized has the virtue of simplicity and relative ease of clinical application. There are, however, three major caveats about the applicability and potential utility of the “psychosis-proneness” index evaluated in this study.

Table 5.

Study and Authors	Sample Characteristics APS(Attenuated Psychotic Symptoms) BLIPS(Brief Limited Psychotic Symptoms) Trait (Genetic risk)	Duration and Instruments	Rates of Conversion to Psychosis
Personal Assessment and Crisis Evaluation Clinic (PACE) Nelson et al., 2011	Sample size=817 Help- seeking 83 percent with APS 27 percent with Trait	6 months Comprehensive Assessment for At Risk Mental State (CAARMS)	72/817 = 9 percent APS alone=9.4 percent APS + Trait = 8.7 percent
North America Prodrome Longitudinal Study (NAPLS) Cannon et al., 2008	Sample size=291 Help- seeking 97 percent with APS	30 months Structured Interview for Prodromal Symptoms (SIPS)	82/291 = 28 percent Conversion rate increased to 52 percent among 118 individuals with APS+Genetic Risk
Outreach and Support in South London (OASIS) Fusar-Poli et al., 2010	Sample size=152 Help- seeking Ultra-High Risk (all three groups)	24 months CAARMS	24/152 = 15.6 percent
European Prediction of Psychosis Study (EPOS) Ruhrmann et al., 2010	Sample size=245 Help-seeking Ultra-High Risk (all 3 groups) and Basic- Symptom based COGDIS	18 months SIPS and Bonn Scalefor the Assessment of Basic Symptoms (BSABS)	37/183 = 20 percent By UHR criteria 21 percent By COGDIS criteria 19 percent
Edinburgh High-Risk Study (EHRS) Johnstone et al., 2005	Sample size = 163 Young adults with family history of schizophrenia (two affected relatives)	30 months	20/147 = 13 percent Presence of schizotypy increased likelihood of conversion
UCLA Study Schlosser et al., 2011	Sample size=84 Help- seeking Ultra-High Risk	24 months SIPS	27/84 = 32 percent
Orygen Youth Health (OHY) Yung et al., 20008	Sample size=119 Help- seeking Ultra-High Risk	24 months CAARMS	19/119 = 16 percent

First, in the general population, there are several groups of individuals who demonstrate an increased liability to develop a psychotic disorder: these include (i) relatives of persons with established schizophrenia or other psychotic disorders, with the degree of increased risk dependent upon the genetic proximity to the affected individual; (ii) help-seeking individuals with brief limited intermittent psychotic symptoms (“BLIPS”) which do not meet criteria for an established psychotic disorder; (iii) help-seeking individuals with attenuated ‘subthreshold’ psychotic symptoms (“APS”), which again do not meet threshold criteria; (iv) non help-seeking individuals with “BLIPS” and “APS”; and (v) individuals with schizotypal personality disorder. It is unclear if the index evaluated in our study, in this case in the first group, would have been found to be equally useful in other groups. Whereas this is an empirical question that needs to be examined in future studies, it argues for the need to carefully examine any set of “psychosis risk” criteria separately in these groups (Cannon et al., 2008; Thompson et al., 2011; Woods et al., 2009; Yung et al., 2008).

Second, thresholds to convert scores on the six tests utilized to estimate a psychosis proneness index were derived from our sample itself: the third quartile score on each of these tests was employed as the cut-off for a 0-1 mark on that test. Whether the specific scores utilized on each of these four tests (Table 1) might generalize to other familial and clinical high-risk samples remains to be evaluated. Again, the success of our approach argues for the utility of customizing the development of psychosis-proneness indices for different populations.

Table 1.

	Converters (N=11)	Nonconverters (N=86)
Gender
Male	5	42
Female	6	44
Age (Mean years, SD)	(16.1, 3.2)	(15.5, 4.6)
First-Degree Relatives
Offspring	6	40
Siblings	3	39
Second-Degree Relatives	2	7
Ethnicity
Caucasian	4	52
Latino/Hispanic	0	0
African-American	7	33
Other	0	1

Both these potential constraints suggest the utility of considering a “progressive closing-in” strategy along a continuum of risk, with additional assessments and interventions based upon the location of the individual along that risk continuum (Nelson et al., 2011; Ruhrmann et al., 2010; Tandon et al., 2009). In our study, the 10 percent likelihood of developing a psychotic disorder (based on familial risk) was “enriched” six-fold (to 59 percent) among those who additionally screened positive on the psychosis-proneness index. Subjects meeting these criteria are 59 times more likely to develop psychosis than someone in the general population. It is conceivable that additional cross-sectional and longitudinal psychophysiological and neuropsychological tests and brain imaging (Bhojraj et al., 2011 and 2011; Koutsouleris et al., 2009; Takahashi et al., 2009) as well as more distal measures of risk (such as early premorbid deficits and environmental risk factors) might add predictive value that could be utilized. Data suggest that such enrichment strategies to identify individuals at progressively higher risk of conversion to psychosis may need to be different depending upon the initial basis for their categorization as a person at higher probability for developing a psychotic disorder than the general population (Corcoran et al., 2008; McIntosh et al., 2011; Smieskova et al., 2010; Sun et al., 2009; Welch et al., 2011)

Third, it is unclear if available interventional approaches could have altered the risk of conversion to psychosis among individuals whose greater vulnerability was identified by our index. Whereas our study does not address this issue, results from other investigations generate optimism about the possibility of specifically modifying the risk of conversion in individuals at high risk (Correll et al., 2010; Dragt et al., 2011; GROUP, 2011; Preti and Cella, 2010; Schlosser et al., 2010). However, many such interventions carry substantial risk and their long-term effectiveness remains inconclusive.

One may argue that predicting psychosis by psychosis proneness scores derived from prodromal symptoms should come as no surprise. However, current efforts in predicting psychosis by prodromal characterization have typically used criteria that define the “late” prodromal phase (Yung et al 1998; Miller et 2002). It is to be noted that very few of our familial high risk subjects met the formal COPS criteria, based on SIPS ratings, for ultra-high risk (Miller et al 2002). However, the SIPS and Chapman scale scores were helpful in predicting psychosis among our familial high risk population. This suggests the value of combining sub-threshold (“early”) prodromal and premorbid indices of psychosis proneness in psychosis prediction.

Even as debates about the propriety of including an “Attenuated Psychosis Syndrome” in DSM-5 continue (Woods et al., 2010; Yung et al., 2010,) it is nonetheless prudent to assess the ability of our tools to better identify those at greater risk for developing a psychotic disorder and evaluate monitoring and interventional strategies to safely and effectively reduce the likelihood of adverse health outcomes in these individuals.