Table 1.
i2b2 method for defining disease phenotype algorithms.
| Steps | Task | Team Member |
|---|---|---|
| 1 | Randomly select 400 subjects with ICD-9 code | Programmer |
| 2 | Review charts, confirm diagnosis for Training Set | Domain expert |
| 3 | Create custom list of concepts relevant to disease | Domain expert |
| 4 | Extract EMR data to create codified variables | Programmer |
| 5 | Create custom list of NLP variables | Domain expert |
| 6 | Map variables UMLS concept unique identifier (CUI) | Informatician |
| 7 | Extract CUIs from narrative text in EMR using NLP | Informatician |
| 8 | Run LASSO regression with codified + NLP variables predicting disease status in Training Set | Statistician |
| 9 | Set specificity at 97%, select predicted probability among Training Set to achieve >90% PPV | Statistician |
| 10 | Apply algorithm to remaining Biobank subjects (excludingTraining Set) | Statistician |
| 11 | Randomly select 100 subjects for Validation Set | Programmer |
| 12 | Perform chart review in Test Set, define PPV | Domain expert |