OBJECTIVE: Although an algorithm to identify cases of endometrial cancer in insurance claims using International Classification of Disease version 9 Clinical Modification (ICD-9-CM) codes has been published, no such algorithm has been ascertained for ICD10-CM codes. The objective of this stidy was to determine the overall positive predictive value (PPV) of an endometrial cancer case identification algorithm using ICD-10-CM diagnosis codes and among women having type 1 or type 2 diabetes mellitus (DM), obesity, or endometrial hyperplasia.
DESIGN: Study population consisted of women aged ≥ 50 years without prior hysterectomy or endometrial ablation, with at least 12 months enrollment in a health plan prior to diagnosis of endometrial cancer from 2016 through 2020. The algorithm variant A used diagnostic codes for malignant neoplasms of uterine sites (C54.x), excluding C54.2 (malignant neoplasm of myometrium), the algorithm variant B used C54.1 (malignant neoplasm of endometrium) only. Both variants require at least 1 inpatient or 2 outpatient diagnoses (on different dates, separated by any interval). A random subsample of provisional cases was adjudicated via review of medical records as confirmed, probable, possible cases, or non-cases. Agreement in case determination among adjudicators was measured using the kappa coefficient (κ). We estimated the PPV of each variant of the case finding algorithm with exact 95% confidence intervals (CI).
RESULTS: Of 3,143 provisional cases identified by algorithm variant A, medical records for 294 unique provisional cases were obtained and adjudicated; 288 of the 294 also were provisional cases per algorithm variant B. Among the women with a provisional case (n=294), the median age was 69.0 years (25th and 75th percentiles: 63.0, 74.0) years. Among those, 49.0% were obese, 42.5% had type 1 or type 2 DM, and 27.6% had endometrial hyperplasia. Both variants identified the same confirmed cases (n=223), but differed in identification of non-cases, and possible cases. There was high level of agreement between adjudicators, κ=0.78. The overall PPV (95% CI) was 84.2% (79.2%-88.3%) for variant A and 85.8% (80.9%-89.8%) for variant B. Among obese women, PPV was 90.9% (84.7%, 95.2%) for both algorithm variants. Among women with DM, PPV was 85.7% (77.8%, 91.6%) for variant A and 87.3% (79.6%, 92.9%) for variant B. Predictive values were highest among those with endometrial hyperplasia: PPV was 96.1% (88.8%, 99.2%) for variant A and 96.0% (88.9%, 99.2%) for variant B.
CONCLUSION: Based on these results, both variants of the ICD-10-CM case finding algorithm were successful in the identification of true endometrial cancer cases. Algorithm variant B identified fewer provisional cases not determined to be true positive cases than variant A.
