(A-213) Development and Validation of Claim-based Algorithms for Non-Melanoma Skin Cancer Using Optum Integrated Databases

Background: Little is known about the accuracy of ICD10 claim-based algorithms for non-melanoma skin cancer (NMSC), the most common cancer in the United States. This study aimed to develop and evaluate claim-based algorithms for NMSC.

Objectives: This study aimed to develop and evaluate claim-based algorithms for incident NMSC.

Methods: A cohort of adult patients with psoriasis were identified in both Optum claims and EHR during Oct2015-Dec2020. Inclusion criteria included one year of claim enrollment before the first diagnosis (cohort entry date), linked claims-EHRs, an overlap period during follow-up, and no prior or recent malignancy. The data was randomly split into the development and validation datasets. Broad NMSC encompassed narrow NMSC (basal cell carcinoma, squamous cell carcinoma, sebaceous cell carcinoma, and malignant neoplasms of the sweat glands), Merkel cell carcinoma, carcinoma in situ, Kaposi sarcoma of skin, cutaneous T-cell lymphoma, and cutaneous B-cell lymphoma. The five developed claim-based algorithms consisted of 1+ diagnosis code 1) at any position; 2) at the primary position; 3) by a dermatologist; 4) at any position plus treatment procedure code; 5) at primary position plus treatment procedure code. Pre-specified case definition included a patient had a diagnosis of NMSC on EHRs in any care setting with supporting evidence (treatment procedure for cancer removal, NMSC drug therapy within 30 days following the diagnosis, or confirmation from histopathology or note data) or a patient did not have a diagnosis of NMSC on EHRs, but had a biopsy record captured in the EHRs, and a combination of positive histopathology and note data for NMSC following the biopsy and had no other malignancy codes (e.g., melanoma). Criteria for true negative NMSC cases were also defined. Algorithms performance measures were estimated.

Results: Among 25,647 patients with psoriasis (mean age 51.4, female 53.6%), most were White, followed by Black and Asians. Patients in the development set and validation set were similar in demographics. A total of 321 and 281 patients with broad and narrow NMSC, respectively, were identified in the development dataset (n=12,824). Sensitivity, specificity and accuracy ranged from 54.3% to 89.9%, 100%, and 99.3% to 99.8%, respectively; PPV ranged from 98.2% to 99.1% for NMSC broad and from 96.7% to 98.5% for NMSC narrow. Algorithm 1 had the highest sensitivity, 89.9% for NMSC broad and 89.5% for NMSC narrow. Algorithm 2 had the higher PPV, 98.7% for NMSC broad and 98.5% for NMSC narrow. Both algorithms 1 and 2 met high-performing algorithm criteria with PPV greater than 90%. Algorithms 1 and 2 were validated in the validation set with similar findings.

Conclusions: This study identified and validated high performing ICD10-based algorithms for NMSC.