Computerized Triggers of Big Data to Detect Delays in Follow-up of Chest Imaging Results.

BACKGROUND

A "trigger" algorithm was used to identify delays in follow-up of abnormal chest imaging results in a large national clinical data warehouse of electronic health record (EHR) data.

METHODS

We applied a trigger in a repository hosting EHR data from all Department of Veterans Affairs health-care facilities and analyzed data from seven facilities. Using literature reviews and expert input, we refined previously developed trigger criteria designed to identify patients potentially experiencing delays in diagnostic evaluation of chest imaging flagged as "suspicious for malignancy." The trigger then excluded patients in whom further evaluation was unnecessary (eg, those with terminal illnesses or with already completed biopsies). The criteria were programmed into a computerized algorithm. Reviewers examined a random sample of trigger-positive (ie, patients with trigger-identified delay) and trigger-negative (ie, patients with an abnormal imaging result but no delay) records and confirmed the presence or absence of delay or need for additional tracking (eg, repeat imaging in 6 months). Analysis included calculating the trigger's diagnostic performance (ie, positive predictive value, negative predictive value, sensitivity, specificity).

RESULTS

On application to 208,633 patients seen between January 1, 2012, and December 31, 2012, a total of 40,218 chest imaging tests were performed; 1,847 of the results were suspicious for malignancy, and 655 (35%) were trigger-positive. Review of 400 randomly selected trigger-positive patients found 158 (40%) with confirmed delays and 84 (21%) requiring additional tracking (positive predictive value, 61% [95% CI, 55.5-65.3]). Review of 100 trigger-negative patients identified 97 without delay (negative predictive value, 97%; [95% CI, 90.8-99.2]). Sensitivity and specificity were 99% (95% CI, 96.2-99.7) and 38% (95% CI, 32.1-44.3), respectively.

CONCLUSIONS

Application of triggers on "big" EHR data may aid in identifying patients experiencing delays in diagnostic evaluation of chest imaging results suspicious for malignancy.