Table 3 Features used by the model: Pearson’s correlation between physician workload with the high- vs. low-workload indicator (n = 43,730), as measured by workload over the prior week
From: A machine learning model using clinical notes to identify physician fatigue
Feature | Correlation with | P value |
|---|---|---|
physician workload | ||
Word Un-predictability | ||
Perplexity (log) | −0.092*** | 5.34e-83 |
Cognitive words (LIWC) | ||
Insight (fraction) | −0.09*** | 4.34-75 |
Certainty (fraction) | 0.075*** | 1.07e-56 |
Pronouns (LIWC) | ||
First person singular pronouns (fraction) | −0.07*** | 7.85e-48 |
Impersonal pronouns (fraction) | −0.049*** | 3.82e-22 |
Affect (LIWC) | ||
Anger (fraction) | 0.025*** | 1.86e-07 |
Readability | ||
Flesch-Kincaid grade | −0.05*** | 4.88e-24 |
