The operationally defined criteria for prodromal schizophrenia show substantial predictive validity. Thirty-five percent of individuals identified on the basis of recent onset or worsening of subsyndromal psychotic symptoms experienced conversion to psychosis after 2½ years of follow-up. To our knowledge, the current sample size of 291 is nearly 3 times larger than that of any previous study, providing greater statistical confidence in the survival estimates. This 2½-year conversion rate of 35.3% represents a relative risk of 405 compared with the incident rate of all forms of psychosis in the general population during a comparable period (ie, 0.087%, or 0.034% per annum).45
The survival curve has a decelerating trend, such that progressively fewer cases convert to psychosis with increasing length of follow-up. This finding indicates that the prodromal criteria are sensitive to risk for imminent onset and provide an empirical basis on which to time the application of preventive interventions. After 2½ years, the risk of onset of psychosis is 2.7%, still higher than the annual incidence rate of schizophrenia in the general population but significantly below the rate observed in the first year of follow-up (ie, 20%).
In the 2 largest previous studies of prodromal psychosis,15,16
a conversion rate of 35% was observed among 104 clinical high-risk subjects identified using criteria comparable to the SIPS,16
and a conversion rate of 49% was observed (after 9.6 years of follow-up) among 110 cases identified using the Bonn Scale of Basic Symptoms.15
The Bonn Scale of Basic Symptoms emphasizes changes in social, emotional, and motivational factors and is thought to ascertain individuals in a much earlier stage of developing psychotic illness.46
Prediction algorithms incorporating combinations of 3 baseline variables (genetic risk for schizophrenia with recent functional decline, higher levels of unusual beliefs or suspiciousness, and greater social impairment) resulted in dramatic increases in PPP (74%–81%) compared with SIPS criteria alone (35%). These prediction algorithms were derived empirically, rather than confirmed through hypothesis testing. A relatively conservative empirical approach was used, such that we first screened the potential predictor variables for association with conversion in multivariate models within each assessment domain and retained only those variables that contributed uniquely to prediction in an overall (cross-domain) multivariate model for consideration in combinatorial algorithms maximizing PPP. Nevertheless, because the algorithms were derived empirically, they should be confirmed in an independent study with comparable sample size and selection, assessment, and follow-up criteria, as might be possible in future collaborations of the North American Prodrome Longitudinal Study group, as well as in similar collaborative efforts in Europe.47
Genetic risk for schizophrenia with recent functional deterioration was strongly and uniquely predictive of conversion to psychosis in this sample. Although the SIPS criteria include a prodromal syndrome involving genetic risk with a decline of 30% or more on the General Assessment of Functioning Scale in the past 12 months, patients who meet these criteria exclusively, without evidence of attenuated psychoticlike symptoms, are quite rare. Nevertheless, the risk construct implied by this category appears promising given that schizophrenia spectrum disorders are specifically elevated among first-degree relatives of patients with schizophrenia.34,48,49
Thus, functional decline, although otherwise nonspecific, should be highly predictive of psychosis in those with a genetic background for the disorder. To model this possibility, we created a new genetic risk and functional deterioration metric in which the genetic component is defined as in the SIPS, but the functional deterioration requirement was relaxed to a criterion of decline of 10% or greater in social, role, or psychological functioning in the year before ascertainment, using scales developed specifically for use in adolescent and preonset samples.38
This metric proved to be a more sensitive predictor of conversion to psychosis than a family history of psychosis or schizotypal personality disorder, whose contributions to psychosis risk were not significant once the genetic risk with functional deterioration term was modeled.
Social deficits and prodromal symptom severity at baseline are also key predictors of psychosis. The present findings indicate that the poorer the social functioning and the more severe the subsyndromal symptoms at ascertainment, particularly in the domains of unusual thought content and suspiciousness, the closer the subject is to the onset of psychosis. Deficits in social functioning are among the most robust behavioral correlates of genetic risk for schizophrenia and are present in many at-risk individuals from childhood.17,50–53
Given that social deficits and prodromal symptom severity combine with a genetic risk for schizophrenia and recent functional decline in achieving maximal prediction, the onset of psychosis appears to be marked by a changing course of thinking and functioning against a backdrop of preexisting inherited vulnerability traits.17,22
In a previous study of 104 clinical high-risk patients from the Personal Assessment and Crisis Evaluation Clinic in Melbourne, Australia, the coincident requirement of meeting attenuated positive symptoms and genetic risk and deterioration criteria was associated with a PPP of 69% and a sensitivity of 31%.19
The increased sample size in the present study enabled the evaluation of specific symptom predictors and varying thresholds for functional deterioration. The predictive validity of other positive symptoms, such as perceptual abnormalities and grandiosity, is limited by their relatively low base rates in this sample.
A history of substance abuse also predicted conversion, although in multivariate analyses no specific substance class of the 7 tested (ie, alcohol, hypnotics, cannabis, amphetamines, opiates, cocaine, and hallucinogens) was significantly associated with risk. It is possible that larger studies will be needed to determine whether specific substances are associated with psychosis in prodromal cases. Although the low base rate of substance abuse severely limits sensitivity, its association with conversion risk is theoretically important because a drug-related mechanism may be capable of producing psychosis-promoting changes in brain function in some high-risk patients. Furthermore, this association, if confirmed, suggests that abstinence from drugs may help to lower the risk of psychotic illness in this population.
Although rates of conversion were higher among cases ascertained in earlier years of the study than more recently, after controlling for other predictors, this effect was not significant. Given that most prodromal research programs have increasingly engaged in community outreach and education efforts to increase awareness of early warning signs, decrease stigma, and stimulate referral, a higher proportion of more recently recruited patients may be ascertained in an earlier phase of risk, when symptoms are less severe.54
A decreasing transition rate could also reflect the increasing and/or more effective application of pharmacologic and psychosocial interventions in prodromal clinics and the community. Three preliminary studies support the notion that early intervention with either or both approaches is associated with prodromal symptom reduction and possibly with reduced or delayed risk for onset of psychosis.42,55,56
Although most investigators in the prodromal field advocate a highly conservative approach to drug treatment of clinical high-risk individuals, whereby antipsychotic drug therapy is initiated only after symptoms have reached a fully psychotic level of intensity, community-based physicians may sometimes be less conservative. In addition, it is not unusual for clinical high-risk patients to receive psychosocial interventions in the community or in the prodromal research programs themselves because these interventions are generally indicated to address presenting complaints (eg, low motivation, social withdrawal, and school failure) and have a lower risk of adverse events than drug treatment.
In this study, antipsychotic drug treatment was found to be associated with a significant increase in risk of conversion to psychosis at the univariate level, most likely reflecting the fact that most of the patients were treated in naturalistic circumstances in which physicians prescribe antipsychotics in the presence of greater severity of positive symptoms. This effect disappeared in the cross-domain multivariate analyses controlling for symptom severity, and accounting for this treatment variable did not modify the predictive relationships between the other study variables and conversion risk. Thus, the predictive relationships between other risk indicators and conversion, the decelerating survival function, and the 35% conversion rate observed in this study appear to be statistically independent of the application of such treatments. A more rigorous basis for dissociating the effects of treatment from natural factors influencing the risk of conversion to psychosis may be possible in a formalized treatment study with random assignment of patients to an active treatment vs placebo. However, patients who consent to participation in randomized, placebo-controlled studies of antipsychotic drugs may differ in substantial ways from those who are willing to be followed up longitudinally while retaining choice over interventions received. More restrictive exclusion criteria (eg, owing to diagnostic comorbidities or the need for conjoint treatments) and attrition owing to the adverse effects of drug treatments further limit generalizability of prediction findings from samples drawn from randomized treatment studies.
In general, the multivariate algorithms, while achieving a considerably higher PPP than any of the univariate models, were associated with much lower sensitivity. This pattern reflects the lower base rates of coincident occurrences of risk factors. Allowing for noncoincident combinations of risk factors resolves this problem, yielding excellent sensitivity but at the sacrifice of PPP, which falls to the level of the univariate models. Sensitivity may be increased in multivariate algorithms integrating quantitative measures that may have more favorable distributional properties than clinical ratings, such as indicators of brain anatomy or physiology or neurocognitive performance.57–62
Attrition was unrelated to the primary variables that predicted conversion to psychosis in this sample. Although more male than female patients were lost to follow-up, conversion to psychosis did not vary according to sex, suggesting that this asymmetry is neutral with respect to the prediction results.
The present results apply to a treatment-seeking population that is recruited and screened for psychosis risk indicators. The results are not expected to be useful in general population screening. Moreover, the present criteria for a prodromal state reflect emerging clinical symptoms and signs that are thought to be on a continuum with fully psychotic states. Thus, the prediction results apply to a population that is already to some extent ill, rather than to a completely clinically unaffected population, and thus it is more appropriate to view prediction in this context in relation to risk of progression and increasing severity of illness than to the risk of illness per se. It is hoped that the knowledge gained from using this approach to monitor neurobiological changes over time will lead eventually to risk ascertainment criteria that can identify at-risk cases before emergence of subpsychotic clinical features.
The shape of the survival function suggests that the initial 2½ years after ascertainment represents a critical window of opportunity for evaluating changes in brain functioning that may underlie the development of psychosis and for the application of interventions that could attenuate or prevent the emergence of psychotic symptoms and functional disability. The present results thus provide a benchmark for the shape and rate of conversion risk against which to compare in future studies assessing comparable populations provided with a standardized intervention program. The use of prediction algorithms with 80% PPP will enable more selective recruitment into prevention programs (minimizing exposure of false-positive cases to potential adverse events) and facilitate studies attempting to elucidate neural and other changes proximal to the onset of psychosis.22,63,64