Development and Validation of an Endometrial Cancer Algorithm in US Claims Data.

PURPOSE: This study developed and validated algorithms to identify the incidence of endometrial cancer using ICD-9-CM and ICD-10-CM coding systems as part of a post-authorization safety study (EU PAS: EUPAS45602) of vaginal estrogen, estrogen and progestin combination hormone therapy, and non-users of vaginal estrogen.

METHODS

Using national claims data in the HealthCare Integrated Research Database, provisional endometrial cancer cases were identified using a screening algorithm comprising at least one endometrial cancer diagnosis code at any location and position between January 1, 2007 and December 31, 2021. One medical record per case was obtained for 700 provisional cases. Two experts in obstetrics and gynecology adjudicated the medical records classifying each patient as a confirmed case, confirmed non-case, or unknown. Positive predictive value (PPV), conditional sensitivity assuming all cases were identified by the screening algorithm, and their respective 95% confidence intervals (CI) were calculated. Algorithms were generated by choosing the codes and number of encounters to balance endometrial cancer detection and false positives. Algorithms were trained in a sample of non-users of vaginal estrogen and tested across treatment groups.

RESULTS

Among the non-user training set, two visits with ICD-9-CM code 182.0 or 182.8 had a PPV of 91.2% (85.4%-95.2%) and sensitivity of 99.3% (95.9%-100.0%). Two visits on separate days with ICD-10-CM code C54.1, C54.8, or C54.9 had a PPV of 97.0% (93.6%-98.9%) and sensitivity of 99.5% (97.2%-100.0%). PPVs and sensitivities were over 75% across testing set cohorts for both algorithms.

CONCLUSION

Two visits with an endometrial cancer diagnosis code accurately identified endometrial cancer cases in both coding systems. There were very few false positive endometrial cancer diagnoses. PPVs and sensitivities were high and similar in all cohorts.