Table 1.

Example clinician performance scores. An observed vs. expected 30-day morality contingency table was constructed for each clinician based on the set of patients they were responsible for in 2010–2013. A patient was counted as “dead” if their death occurred within 30 days of their last recorded admission order. The observed dead (D_O) and alive (A_O) counts can be deduced directly from admission order and mortality timestamps. The expected dead (D_E) and alive (A_E) counts are predicted using 30-day mortality probabilities generated by a L1-regularized logistic regression model trained on 2008–2009 patient data. In these examples, Clinician A has a lower O-to-E ratio than Clinician B, but a larger quantity of data that yields more confidence and thus an equivalent final score. Clinicians B and C have the same observed-to-expected mortality ratio, but their scores differ due to varying confidence in the estimate. Higher magnitude scores thus reflect a greater effect size or certainty of deviation from the expected mortality rate.

Observed to Expected Ratio = DO/DE		Clinician Example A O-to-E Ratio = 1.5 Score = −0.22		Clinician Example B O-to-E Ratio = 3 Score = −0.22		Clinician Example C O-to-E Ratio = 3 Score = −2.89
Observed Dead (DO)	Observed Alive (AO)	9	70	3	17	30	170
Expected Dead (DE)	Expected Alive (AE)	6	73	1	19	10	190