Email Icon Facebook icon Twitter Icon GooglePlus Icon Contact

User Top Menu

Narrowing Criteria Improves Prediction of Psychosis

19 Oct 2015

October 20, 2015. A new set of clinical high-risk criteria may improve the prediction of patients likely to develop a psychotic illness, reports a study led by Barbara Cornblatt of the Feinstein Institute for Medical Research and the Zucker Hillside Hospital, Department of Psychiatry, North Shore-LIJ Health System. The findings were published October 1 in the American Journal of Psychiatry.

Although previous studies of people deemed at high risk typically included all patients, the group narrowed down its participants to include only individuals with attenuated positive symptoms in the hope that a more homogenous clinical population would lead to a more accurate predictor model. The researchers identified four variables that predicted the rate of conversion of their selected high-risk population with an 81.8 percent positive predictive validity and calculated a risk index score for each participant.

Cornblatt is careful to state that this risk assessment should not be applied to the general population. "We had a very homogenous selected population, even for a clinical population, and in that group, our risk predictors worked really well," she told SRF. She also emphasizes that it's not enough for a model to work for one sample. But by specifying criteria, they've made it possible for others to replicate their work.

Improving the prediction of high risk

Attenuated positive symptoms can begin to appear early in those who will develop schizophrenia or bipolar disorder during a pre-psychotic stage referred to as the prodrome. The presence of these symptoms puts an individual at high risk for developing full psychosis. Beginning treatment early may delay or even prevent conversion, making it critical to identify high-risk patients who are truly in the prodrome at this early stage.

The current high-risk criteria only accurately predict about one-third of patients who will convert (Fusar-Poli et al., 2013). Thus, many more who are identified as high risk do not convert, raising ethical concerns of how to begin early treatment intervention without risking unnecessary exposure for those who may not need it.

Several groups around the world have been trying to improve the prediction of high-risk individuals for the past 20 years, and a few studies have managed a positive prediction value of about 80 percent (see SRF related news report; SRF news report). However, because variables differ across clinic populations, each study produces a predictive model tailor made for its own set of participants. This highlights the necessity of replication.

"If you can get common threads between them, then you can really start building something that's going to be the height of accuracy," said Cornblatt, adding that replication of previous work is a strength of the current study. Updated criteria based on replicated predictors could markedly improve identification of individuals who will convert to psychosis and better inform decisions on intervention.

Narrowing predictive variables

Participants for this study were recruited during Phase 1 of the Recognition and Prevention (RAP) Program of the North Shore-Long Island Jewish Health System in New York. Cornblatt and colleagues assessed 92 high-risk adolescents and 68 healthy comparison participants for neurocognitive, behavioral, functional, and clinical measures. They obtained follow-up clinical ratings until conversion to psychosis or until follow-up was completed for those who did not convert.

The researchers started with predictor variables generated by previous models, combined with standard predictors such as clinical characteristics and demographic variables. They then narrowed these down and found that the most powerful factors in predicting conversion were disorganized communication, suspiciousness, verbal memory deficits, and declining social functioning. With these four variables, the researchers produced a model with a positive predictive value of 81.8 percent.

Of the variables used, Cornblatt highlights their inclusion of cognitive deficits in the model, which few studies have used before. "We've been talking about cognitive problems for years, and so it makes sense that there are cognitive predictors," said Cornblatt. Their findings also demonstrate social problems as important predictors. Social problems have been identified clinically across many studies in high-risk individuals but have not been well established as predictors of conversion.

Assigning a risk score

In the current study, the researchers used their model to generate a risk index score for each individual, similar to the prognostic score by Rurhmann and colleagues (Rurhmann et al., 2010). "What we were able to do was to generate an index score that accurately categorizes somebody as being at high risk, intermediate risk, or low risk," said Cornblatt. The prediction was highly associated with conversion status for patients on either end of the risk spectrum—those with high-risk scores converted to psychosis, and those with low-risk scores did not convert—but was less reliable for individuals falling in the middle of the spectrum.

Producing a risk index score for an individual would help guide decisions for necessary intervention. Many physical illnesses such as cancers and heart disease already have risk calculators. Cornblatt thinks that a risk calculator for psychosis will be the next big advancement in the field. The risk index score would guide the intervention patients receive, ranging from aggressive therapies for high-risk scores and close follow-up for intermediate scores to no intervention for low-risk scores.

Far to go before an impact in the clinic

In addition to the need to refine a prediction model for larger use, another barrier prevents this work from having an immediate or bigger impact: It requires catching people at this early but critical high-risk state before their lives and futures fall apart. "Only a minority of patients currently present to a clinic during the pre-psychotic period," said Patrick McGorry of the University of Melbourne, adding that the vast majority still only seek help during a first psychotic episode. McGorry, who was not involved in the study, said that we have some way to go before these predictive models will be ready for more widespread use for the young people they are intended to help.

He also stressed that there is more to the use of a predictive model than simply establishing on what side of a conversion cutoff a patient stands. "The main goal is to not predict just psychosis (while this is obviously important), but to identify people with a need for care," he said.

In a follow-up email to SRF, McGorry wrote, "We also need to ensure we do not apply these particular prediction rules beyond help-seeking clinical samples, which are enriched for risk so the rules can work. If we were to use them in the general population, this would result in a much higher proportion of false positives. This could discredit this crucial preventive approach, which holds the key to reducing the impact of schizophrenia and related psychoses."—Lesley McCollum.


Cornblatt BA, Carrion RE, Auther A, McLaughlin D, Olsen RH, John M, Correll CU. Psychosis Prevention: A Modified Clinical High Risk Perspective From the Recognition and Prevention (RAP) Program. Am J Psychiatry. 2015 Oct 1;172(10):986-94. Abstract