SUNDAY CLERKING
Kanye’s X Rampage: A Case for AI in Early Bipolar Detection?

Why does Kanye feel the need to call himself a Nazi? No clue.

Why does Kanye feel the need to disrespect transgender people? Couldn’t tell you.

Why does Kanye feel the need to go on a tweet storm, then vanish? Now that—that I can explain.

In 2018 Kanye revealed upon the release of his album, ’Ye’ that he had bipolar disorder. This disorder is marked by cycles of mania or hypomania—possibly reflected in his erratic social media behavior—followed by episodes of depression, which he frequently explores in his music.

We cannot undo what is already done. Kanye West is cooked, regardless of whether he made the legendary album Graduation. 

But, what if we have got to him earlier than 2016. The symptoms of Bipolar Disorder and Schizophrenia come on insidiously and as a result are diagnosed quite late. 

Dr. Ostegaard & crew at Aarhus University in Denmark conducted a cohort study to test whether machine learning could predict these conditions developing using previous electronic health records (EHRs)—before their official diagnosis. Potentially allowing for earlier intervention —kind of like forecasting a mental health storm before it fully hits.

They took over 24,000 patients receiving psychiatric care between 2013-2016, and used their electronic health records(EHR’s) to predict who would go on to develop schizophrenia or bipolar disorder within the next five years. The model took structured data(meds, diagnoses, admissions etc.) and clinical text notes.

What were the results? The model called XGBoost performed the best. Prediction had a median lead time of 1.1 years before the official diagnosis was made irl. This is good, however the Schizophrenia prediction was way more accurate than BPD(AUROC = 80 and 0.62 respectively).

Dr Ostegaard suggests this is due to schizophrenia presentations being more homogenous (psychotic events) compared to the more varied presentation of Bipolar Disorder(severe mania vs severe depression). 

Additionally, whilst the study did suggest feasibility, the positive predictive value wasn’t high enough for clinical use. Finally, clinical notes yielded better prediction data that the structured EHR data—suggesting barely legible Drs notes may hold more truth than pre-made checkboxes.

So what does this all mean for Mr West? Not much…yet. We can’t diagnose anyone from their chart, let alone their tweets. But it shows promise that one day, we can diagnose manic episodes before they go public.

Sources:
Hansen, L., Bernstorff, M., Enevoldsen, K., Kolding, S., Damgaard, J.G., Perfalk, E., Nielbo, K.L., Danielsen, A.A. and Østergaard, S.D. (2025). Predicting Diagnostic Progression to Schizophrenia or Bipolar Disorder via Machine Learning. JAMA Psychiatry. [online] doi:https://doi.org/10.1001/jamapsychiatry.2024.4702.

I like medical news… but only when it’s interesting. So I’ll try and make it more interesting for you too. Not to be taken too seriously, but memorable enough that you can reference them to sound clever and well-read to your consultant.

Soo…Did you like it? Did you hate it? Give me a reply, so I can improve for next weeks issue 🤝.

See you on Friday for the first Weekly Handover