(069) Denosumab Data Cleaning Algorithm to Inform Exposure Measurement and Study Inclusion: Results from Two International Databases

Background: Exposure data cleaning is an important first step yet is uncommonly presented in pharmacoepidemiologic studies. Denosumab is an osteoporosis medication injected twice per year, usually via a clinician. Identifying exposures to denosumab using healthcare administrative data can be challenging, as claims can be generated in different data sources, illogical claims may arise due to appointment rescheduling, and a monthly formulation exists for cancer-related indications.

Objectives: To develop a denosumab data cleaning algorithm for osteoporosis using healthcare administrative data and apply it to Canadian and US Medicare data.

Methods: We leveraged data from 2 sources: 1) administrative data from Ontario, Canada (ON data); and 2) Medicare data for a 20% random sample of beneficiaries (US data). The first claim for denosumab between 1/2010 to 12/2022 (ON) or to 12/2020 (US) among community-dwelling older adults (aged ≥66 years) was identified using pharmacy claims (ON) and Medicare Part D (prescription drug) and Part B (outpatient) claims. Clinical experts defined a second dispensation of denosumab within 120 days as illogical for osteoporosis. We developed and applied a data cleaning algorithm for denosumab claims that used a combination of dosing, days between claims, and other administrative data. We excluded patients with 3+ claims < 121 days in a row to exclude those with use indicative of cancer ( < 1%). Among remaining, we first cleaned index date (e.g., if first claim after index was < 121, the index date was revised to the next claim date). We next applied our “ < 121 days” cleaning rules to all subsequent claims. The proportion of patients that required data cleaning, the number of claims, and the difference in days between the original and cleaned index date were estimated.

Results: We identified 192,163 older adults in ON data (mean age=78, SD=7.6; 90% female) and 97,595 in US data (mean age 77, SD=7.2; 90% female). A total of 23,696 (12%) patients in ON and 5,886 (6%) in the US required data cleaning. Of these, 36% (ON) to 71% (US) required cleaning for their index and/or last claim (ON: 4322 index, 4251 last, 238 both; US: 1995 index, 1713 last, 463 both). The mean number of days between the original and cleaned index claims was 60 (SD=47) in ON and 8 (SD=19) in the US. The median number of claims per individual after data cleaning was 5 (Q1=2, Q3=9; 14% with 1 claim, 24% with 11+ claims) in ON and 5 (Q1=2, Q3=7; 18% with 1 claim, 9% with 11+ claims) in the US.

Conclusions: Careful consideration of exposure data cleaning is important for exposure measurement and study inclusion. For denosumab, cleaning the index date may affect inclusion for exposure anchored studies of early harms and cleaning last claim dates may affect inclusion for studies estimating harms of discontinuation.