№ files_lp_3_process_9_28720
Technical specification detailing the structure, feature definitions, regression formula, filter criteria, and performance characteristics of a machine-learning phenotype algorithm for classifying bipolar disorder cases from electronic health record data.
Document Version: 1.0.00-001
Document Identifier: BD_filterpositive_10May18
Date: 10May18
Institution: Harvard eMERGE
Subject: Bipolar Disorder (BD) phenotype algorithm
Type of Document: Algorithm implementation specification
Model Type: Penalized logistic regression (LASSO)
Data Source: Electronic Health Records (EHR)
Training Set: 200 chart-reviewed subjects
Gold Standard Method: ICCBD Diagnostician Review for Diagnosis & Confidence
Performance Metrics: PPV 0.918; TPR 0.563; FPR 0.05; NPV 0.687
Specificity Threshold: 0.95
Features Included: Age; ICD diagnosis codes; Medication codes
Filter Criteria: ≥3 ICD codes on three separate days; ≥1 clinical note; ≥1 BD diagnosis code
Outcome Definition: Predicted probability of bipolar disorder
Price: 8 / 10 USD
The file will be delivered to the email address provided at checkout within 12 hours.

Don’t have cryptocurrency yet?

You can still complete your purchase in a few minutes:
  1. Buy Crypto in a trusted app (Coinbase, Kraken, Cash App or any similar service).
  2. In the app, tap Send.
  3. Select network, paste our wallet address.
  4. Send the exact amount shown above.
After sending, paste your TXID (transaction ID) and your email to receive the download link. Need help? Contact support and we’ll guide you step by step.