Artificial intelligence evangelists claim that the technology promises (or threatens) to transform the modern workforce. And most analysts back them up. They predict that occupations across the board – from customer service reps to accountants – all risk losing their jobs to the technology. Even lawyers are looking over their shoulders. All told, a 2022 Pew Research Center study found that nearly 20 percent of jobs today could be lost to AI.
Now a group of researchers at Aarhus University Hospital in Denmark have published a study that suggests that techs can engineer AI to predict the progression of psychiatric illnesses, such as schizophrenia and bipolar disorder, simply by processing routine clinical data from electronic health records (EHRs).
The team, which pored over the data generated by nearly 400,000 outpatient psychiatric visits in Denmark, discovered that AI models could predict schizophrenia and bipolar disorder, with varying degrees of success. This could mark a huge step forward in the early detection and intervention for severe mental health conditions.
Diagnosis Remains a Persistent Challenge
Schizophrenia and bipolar disorder remain two of the most severe – and life-changing – mental illnesses. And they can quickly descend into overwhelming social and occupational impairment. These illnesses almost always crop up between late adolescence and early adulthood. But tragically – and all too frequently – they can evade diagnosis for years after onset, hindering treatment that could improve long-term outcomes.
The prodromal phase, and the overlap of symptoms with other mental health conditions, play a huge part in these delayed assessments. Consequently, many of those who eventually receive a schizophrenia (or bipolar) diagnosis already received treatment for something else.
“It is a difficult clinical challenge to solve, but we have given it a try, and the results of this study show that we are on the right track,” research team leader and professor Søren Dinesen Østergaard, from the Department of Clinical Medicine at Aarhu, explained.
Harnessing AI for Early Detection
The Danes wanted to figure out whether AI-driven machine learning models could predict the probability of a patient developing schizophrenia or bipolar disorder based on their medical records.
To that end, the researchers combed through the EHRs of patients who’d attended at least two psychiatric consultations between 2013 and 2016. Then they trained their AI models to forecast a diagnostic transition to schizophrenia or bipolar disorder within five years.
The research team employed a pair of machine learning approaches: elastic net regularized logistic regression and extreme gradient boosting (XGBoost). These models used structured clinical data – and text-based data – from clinical notes to gauge risk.
AI Predicts Schizophrenia More Accurately Than Bipolar Disorder
The results revealed that AI could forecast schizophrenia with greater accuracy than bipolar disorder. The XGBoost model achieved an AUROC of 0.80 for schizophrenia prediction, much higher than the 0.62 AUROC for bipolar disorder. This, the authors argue, suggests that while AI can serve as a valuable tool for identifying those at risk of schizophrenia, predicting bipolar disorder lingers as a more daunting challenge.
Posting a predicted positive rate of 4 percent, the model demonstrated a sensitivity of 19.4 percent for schizophrenia. In layman’s terms, it accurately identified nearly one in five individuals who would later develop the disorder. Even more encouraging, the researchers reported an extremely low rate of false positives.
On the other hand, the bipolar disorder model displayed a sensitivity of just 9.9 percent, reiterating just how complex – and varied – its presentation is.
The Power of Clinical Notes in Prediction
The biggest takeaway – according to the research team – was the crucial role that text-based clinical notes played in facilitating predictive accuracy. Unlike more structured data, clinical notes feature detailed descriptions of symptoms, treatment responses, and patient-clinician interactions. These are the crucial details that help doctors recognize the early warning signs of most psychiatric disorders. By leveraging natural language processing techniques, the AI models extracted and analyzed these insights, and, in doing so, uncovered patterns that more traditional diagnostic methods typically miss.
As part of that, the researchers found that certain words and phrases within those clinical notes emerged as reliable red flags hinting at the likelihood of a future schizophrenia diagnosis. Terms related to hospitalization, social interactions, and auditory hallucinations played a notable role.
For example, mentions of “voices” often suggested auditory hallucinations, while references to “female friends” sometimes pointed to social withdrawal — a hallmark early symptom of schizophrenia.
What About Clinical Implications?
The study’s findings imply that AI models could play a crucial role in psychiatric care by helping clinicians identify high-risk individuals sooner, paving the way for more timely interventions. If implemented in clinical settings, these models could flag patients at higher risk of schizophrenia or bipolar disorder.
“If the algorithm indicates a high likelihood of developing schizophrenia or bipolar disorder within the next five years, healthcare staff can focus their examination on symptoms associated with these disorders – potentially leading to earlier diagnosis and the initiation of targeted treatment,” Søren Dinesen Østergaard added.
For schizophrenia, in particular, early intervention remains paramount. Research shows that curbing the duration of untreated psychosis leads to better outcomes, including improved symptom control and social functioning. By integrating AI predictions into EHR systems, psychiatric services could address early warning signs more aggressively, cutting down on diagnostic delays.
Even so, the researchers warn that AI should serve as a supplementary tool rather than a replacement for clinical judgment. The models can provide probability estimates, but lack the ability to make final diagnostic decisions.
AI’s Growing Role
This study adds to the mounting body of research exploring the use of AI in mental health diagnostics. Future research, the authors argue, should focus on refining predictive models, improving their versatility across different settings, while integrating additional data sources, such as genetic information and neuroimaging.
Additionally, efforts to enhance bipolar disorder prediction could involve more sophisticated AI methods, such as deep learning models trained on larger and more diverse datasets. By refining these tools, researchers hope to make early diagnosis and intervention a standard part of everyday psychiatric care.
While challenges remain, this study signals a major step toward leveraging AI in mental health diagnostics. By using machine learning to analyze EHR data, this team of researchers showed that clinicians can predict schizophrenia with reasonable accuracy, paving the way for earlier treatment. Though predicting bipolar disorder remains more difficult, ongoing advancements in AI could eventually help bridge this gap. As technology continues to evolve, the integration of AI into psychiatric care could upend what we know about identifying and managing mental health conditions.
Further Reading
Study Unveils Link Between Manic Symptoms and Schizophrenia
Lifting the Veil on Schizophrenia and Substance Use Disorders
Researchers Propose a Periodic Table of Psychiatric Disorders
