Email Icon Facebook icon Twitter Icon GooglePlus Icon Contact

User Top Menu

Predictor Tool Uses Simple Variables to Identify Patients at High Risk of Psychosis

20 Jun 2017

by Allison Marin

Calculator

A freely available, web-based tool successfully predicts risk for later psychosis among those seeking secondary mental health care in London, using only simple clinical and sociodemographic variables, reports a new study published online March 29 in JAMA Psychiatry and led by Paolo Fusar-Poli of King’s College London, UK.

“Considering that the sample used to generate [the risk predictor] cuts across diagnostic boundaries, this tool represents a potentially important advance toward achieving the goal of personalized prediction of psychosis in a secondary mental health care setting,” wrote Ricardo Carrion of Zucker Hillside Hospital in New York City in an email to SRF. Carrion was not involved in the new work.

Some studies indicate that the administration of interventions such as antipsychotic medications and psychological treatment to individuals at high risk of developing psychosis can reduce the number of individuals who convert to the full-blown disorder, as well as improve outcomes in those who do convert (Deas et al., 2016; Fusar-Poli et al., 2016). Therefore, current psychosis prevention efforts are largely focused on identifying and treating such high-risk individuals in specialized at-risk mental state (ARMS) clinics (see SRF related news story).

“Prevention is possibly the only way to alter the course of psychosis and therefore to save many lives. What’s limiting our ability to prevent the disorder is that we can only identify a minority of individuals who are going develop psychosis,” Fusar-Poli told SRF. “This tool hopefully begins to tackle this problem and will help clinicians better identify those at highest risk,” he added.

In the new study, Fusar-Poli and colleagues generated and validated a model to estimate the risk probability that patients who have accessed secondary mental health care services will later develop psychosis. To do so, the researchers mined the health records of all patients who received a first index ICD-10 diagnosis of a nonorganic and nonpsychotic mental disorder within the South London and the Maudsley National Health Service Foundation Trust from 2008-2015.

The cohort of 33,820 individuals that was used to generate the model had a fivefold higher six-year risk of developing psychosis than the general population. In addition, while 1,001 patients in the cohort transitioned to psychosis, only 5 percent had received an ARMS designation.

These two findings underscore the need for a transdiagnostic risk predictor tool, Fusar-Poli told SRF. People accessing secondary mental health care are clearly at a higher risk of developing psychosis than the general population, but the current practice of referring patients to ARMS programs based on clinician suspicion alone is missing 95 percent of the people in this population who will later develop psychosis, he said.

The new model used four simple clinical and sociodemographic variables―the patient’s first non-psychosis mental disorder ICD-10 diagnosis, age, gender, and race/ethnicity―that were selected a priori based on meta-analytical associations with increased risk for psychosis.

Overall, the model was able to fit the data well. A measure of the model’s goodness of fit, the Harrell C index―defined as the probability that a randomly selected patient who later developed psychosis scored at higher risk of transition than a patient who did not develop psychosis―was 0.8. The model explained about 75 percent of the variation in the data. It performed similarly in a second validation cohort of 54,716 subjects.

Moving beyond simple risk probability, the researchers then performed a clinical utility analysis. Similar net benefit analyses have been used in oncology to assess whether prediction models do more harm than good in real-world clinical practice. In addition to the benefits of correct prediction, this analysis takes into account harmful factors such as unnecessary treatment. Weighing both benefits and harm, the clinical utility analysis determined that use of the risk calculator was associated with significant net benefits in real-world clinical care.

Unlike other existing personalized psychosis risk predictors, such as the one developed by the North American Prodrome Longitudinal Study (NAPLS; see SRF related news story), which utilize mostly neurocognitive predictors that require specialized training to assess and can only be used in individuals who are already classified as high risk for psychosis, the new predictor uses simple variables and can be used transdiagnostically.

“Our predictor can be easily implemented in the clinic and identifies a much larger fraction of the patients who will develop psychosis, which will allow us to [greatly] extend the benefits of preventative treatment,” Fusar-Poli said.

The risk calculator is freely available online.

Considering that there is no widely adopted or “gold standard” treatment for individuals at clinical high risk for developing psychosis, such as the NAPLS calculator, this tool can be used as a guide to select treatment and therefore can be part of the larger conversation between a mental health professional and a patient when discussing psychosis risk,” said Carrion.

References: