Utilizing Big Data to Predict First-Episode Psychosis and Power Preventive Care
Utilizing the latest advances in big data and machine learning, scientists from Lundbeck have developed a method that may be able to predict those individuals at risk to develop a first psychotic episode a full year before it occurs. Because development of first-episode psychosis is linked to worse long-term outcomes in people with schizophrenia,1 there is hope this methodology could one day help clinicians identify and treat people in earliest stages of the disease – before they progress to first-episode psychosis. Results of a large-scale study utilizing the methodology were recently published in The Lancet Digital Health.
“It may be possible to treat people and prevent them from developing psychosis, but the biggest obstacle is early detection,” says Dr. Paolo Fusar-Poli, faculty member of King’s College in London and co-author of the new study. “This tool might solve that problem and so might improve the lives of millions of people.”
The predictive method is known as Dynamic ElecTronic hEalth reCord deTection (DETECT) and is based on analysis of electronic health records, with data collected from standard clinical care (healthcare utilization, results of routine screening tests, prescriptions filled, prescriptions unfilled and more). 2 The new study leveraged IBM Watson Health’s Explorys Solution to analyze a subset of de-identified electronic health record (EHR) data of more than 60 million individuals. It is the largest study of its kind and was conducted by scientists from Lundbeck, IBM Watson Health and King’s College.
“The goal was to take a very generic data set covering millions of patients and look for patterns within those billions and billions of data points – relationships and patterns that are beyond human detection, but would be discernible to machine learning,” explains Dr. Bruce Kinon, Vice President for Clinical Research and Development in Psychiatry at Lundbeck US and co-author of this study.
Watch Dr. Kinon discussing the methodology in this broadcast news interview.
“This study represents an important advancement in the science and treatment of schizophrenia, a debilitating and chronic brain disorder. Our IBM Explorys Solution – based on a secure, cloud-based platform that makes billions of de-identified data points aggregated from longitudinal EHRs available to qualified researchers – helped power the analysis,” explains Anil Jain, MD, FACP, Vice President and Chief Health Information Officer at IBM Watson Health. “Researchers face immense challenges as they confront the need to generate meaningful real-world insights from complex and noisy health data. We hope that this research can be a model of how such knowledge discovery can be accelerated to help identify clinically meaningful insights for complex conditions such as those seen in behavioral health.”
The method successfully identified patients who later went on to develop their first-episode of psychosis with adequate predictive accuracy, according to the study. 2 “The predictive model looked at all various inputs that are routinely collected in the course of patient interactions within primary and secondary healthcare systems,” Dr. Kinon says. “Data from these healthcare encounters, regardless of whether for general or brain health problems, when taken together revealed a pattern that differentiated those individuals more likely to go on to a first episode of psychosis.”
Being able to predict who is at risk for first-episode psychosis is important, Dr. Kinon explains, because many people with schizophrenia go through a prodromal phase, where they may have some warning signs of the disorder but have not yet developed first psychosis.3 Once people experience first-episode psychosis, their chance of full recovery are low.1 And the longer people go without treatment for first-episode psychosis, the worse the long-term outcome.4,5
“Flagging people who are at risk for first-episode psychosis could help clinicians identify individuals who may benefit from increased follow-up,” Dr. Kinon says. “That could lead to the early start of mental health treatments that may delay the onset of psychosis or even stop it from developing in the first place.”
The tool needs further development and validation before it could be put into practice, says Lars Lau Raket, senior specialist in Biometric Sciences at Lundbeck and one of the scientists behind the method and co-author of the new study. “But the analyses suggest that DETECT has an adequate level of accuracy to offer useful information for supporting clinical decision making,” he adds. “By looking at disease patterns, our algorithms can in some cases detect the subtle signs of mental illness that could be overlooked by a doctor."
Lau Raket emphasizes that the tool cannot replace healthcare professionals but is meant to support clinical decision making by maximizing benefit of already collected data. If the method can be further developed and validated, healthcare professionals may have an extra tool to aid in early diagnosis.
“Analyses of big data will transform the healthcare system. By moving from studying hundreds of patients to thousands and even millions of patients, we can identify complex disease patterns and possibly come up with new ways to deal with the risk for mental disorders,” he says. “The perspectives of helping people through big data are promising, and this method is just an early example of what will be possible in the future.”
Sources:
1. Austin S, et al. Predictors of recovery in first episode psychosis: The OPUS cohort at 10 year follow up. Schizophrenia Research. 2013; 150:163-168.
2. L. Lau Raket, J. Jaskolowski, B. Kinon, et al. Dynamic ElecTronic hEalth reCord deTection (DETECT) of individuals at risk of a first episode of psychosis: a case-control development and validation study. The Lancet Digital Health, published online March 26, 2020.
3.Millan, M, et al. Altering the course of schizophrenia: progress and perspectives. Nature Reviews Drug Discovery. 2016 July; 15 (7): 488-515.
4. W. Cahn, N.E.M. Van Haren. Brain volume changes in the first year of illness and 5-year outcome of schizophrenia. British Journal of Psychiatry, 2006;189: 381-382.
5. Primavera et al. Does duration of untreated psychosis predict very long-term outcome of schizophrenic disorders? Annals of General Psychiatry, 2012; 11:12.
