Table 1 Algorithm description
From: Natural language processing to identify lupus nephritis phenotype in electronic health records
Algorithm name | Classification model | Description |
|---|---|---|
Baseline algorithm | Rule-based | A patient is confirmed to have lupus nephritis if he/she has proteinuria > 0.5 mg in laboratory test or has ICD 9/10 diagnosis code for lupus nephritis. |
Full MetaMap model (binary) | L2-regularized logistic regression | Features are the non-negative mention of MetaMap CUIs. We treated CUIs as binary variables and fitted L2-regularized logistic regression to predict lupus nephritis. |
Full MetaMap model (count) | L2-regularized logistic regression | The same as the full MetaMap model (binary) except that MetaMap CUIs are treated as numeric variables representing the count of instances each concept is mentioned in the clinical text. |
MetaMap mixed model | L2-regularized logistic regression | There are 13 features in this model including 7 CUI features, 5 RegEx concepts, and 1 feature from structured data. |