Ishanu Chattopadhyay PRO
ML Data Science Biomedicine Social Science Faculty
Instantaneous Questionnaire-free Primary-Care Screening
To Predict Suicidal Ideation/Attempt in near future
ZCoR-S Screening
Zero Burden Co-morbid Risk*
Out-of-sample performance plots
Datasets
MarketScan nationwide Insurance claims dataset
(5.2k Cases, 737k Control patients, Age 25-55)
Case Definition: Patients with record of SI/SA in EHR
Performance
Methodological Insight
AI discovers co-morbidity patterns found in patient's EHRs to infer the SI/SA risk.
- Does not depend on known risk factors
- Effectively screens lowrisk populations
- Can screen entire VA population at scale
Positive Likelihood ratio ~ 35
AUC > 87%
Posterior odds of SI/SA
in flagged population:
13 in 20
Prior odds of SI/SA
in general population:
1 in 20
Confidential: Do not circulate publicly
3 out of 13 true flags have no prior history of mental disorders
Instantaneous Universal Screening at Primary Care.
No blood tests, no questionnaires, just diagnostic codes.
Works even for patients without history of mental disorders.
Copy of template_zed
By Ishanu Chattopadhyay
Copy of template_zed
Predictive modeling of crime and rare phenomena using fractal nets
