Director, Epidemiology, Safety Surveillance Research Scientist Pfizer Warren, United States
Background: A validated algorithm for identifying endometrial cancer (EC) cases using ICD-9 codes in claims data developed among estrogen-progestogen combination hormone therapy (E+P HT) users may not perform well in vaginal estrogen (VE) users or non-users of hormone therapy. Furthermore, a validated algorithm using more specific ICD-10 codes is lacking. Therefore, it is essential to develop a validated algorithm to identify EC using both coding systems separately in a US healthcare claims database for use in a post-authorization safety study aimed at examining the risk of EC among VE users compared to E+P HT users and non-users.
Objectives: To develop and assess two algorithms, one each for the ICD-9 and ICD-10 coding system, to identify EC cases in the HealthCore Integrated Research DatabaseSM (HIRD).
Methods: A sample of 700 EC cases (i.e., 300 each in VE users and non-users, 100 in E+P HT users) was identified using a screening algorithm defined as at least one ICD-9 code (182.0, 182.8) or ICD-10 code (C54.1, C54.8, C54.9) in any position (primary, secondary, etc) and in any location (outpatient, inpatient, etc) in the HIRD from January 1, 2007 to December 31, 2021. To date, one medical record per case (hospital/emergency department visit with EC or physician offices with one claim for EC) has been obtained in 623 (89%) cases. An Adjudication Committee of two clinical experts of obstetrics and gynecology is formed to review and adjudicate medical records independently. The adjudication criteria are documented EC diagnosis by a treating healthcare provider and positive results from endometrial biopsy, pathology report, or surgical procedure, or medical treatment consistent with EC care. A positive predictive value (PPV) and 95% confidence interval (CI) were calculated as the number of true (confirmed) cases divided by the total number of cases identified by the algorithm (true + false positives) among VE, E+P HT and non-users and by each coding system. Separate algorithms were generated by choosing codes and the number of encounters that will maximize the EC detection and validated in all users and non-users.
Results: Records on all 700 cases will be obtained and adjudicated by April and algorithms will be validated by June 2023. Of the 700 cases, the clinical experts have adjudicated 119 (17%) and 68 (9.7%) records, respectively. In a sample of 23 adjudicated EC cases identified using the screening algorithm, 20 were confirmed cases, with a PPV of 87% (95% CI: 73%-101%). Results based on 700 cases will be presented at the conference.
Conclusions: If validated, EC algorithms in VE, E+P HT and non-users using ICD-9 and ICD-10 codes will enable a more accurate identification of this outcome in real-world studies using claims data.
