Background: Diagnoses recorded in administrative databases have limited utility for accurate identification of severe sepsis and disseminated intravascular coagulation (DIC). We evaluated the performance of alternative identification methods that use procedure records.
Methods: We obtained data for adult patients admitted to intensive care units in three hospitals during a 1-year period. Severe sepsis and DIC were identified by three means: laboratory data, diagnoses, and procedures. Using laboratory data as a reference, the sensitivity and specificity of procedure-based methods and diagnosis-based methods were compared.
Results: Of 595 intensive care unit admissions, 212 (35.6%) and 81 (13.6%) were identified as severe sepsis and DIC, respectively, using laboratory data. The sensitivity of procedure-based methods for identifying severe sepsis was 64.2%, and the specificity was 65.3%. Two diagnosis-based methods —the Angus and Martin algorithms— exhibited sensitivities of 21.7% and 14.6% and specificities of 98.7% and 99.5%, respectively, for severe sepsis. For DIC, the sensitivity of procedure-based methods was 55.6%, and the specificity was 67.1%, and the sensitivity and specificity of diagnosis-based methods were 35.8% and 98.2%, respectively.
Conclusions: Procedure-based methods were more sensitive and less specific than diagnosis-based methods in identifying severe sepsis and DIC. Procedure records could improve disease identification in administrative databases.