Impact of Increasing Enrollment Requirements in Administrative Claims Data
Background: When claims-based real-world data sources are used for pharmacoepidemiology research, a minimum period of observability prior to cohort entry should be established to increase the likelihood of evidence of chronic conditions and minimize the potential for misclassification of baseline characteristics used for confounding control. While a generally accepted rule of thumb is to require ≥6 months (mo), longer periods may be desired; however, tradeoffs associated with longer enrollment requirements such as impacts to sample size and generalizability have not been fully characterized.
Objectives: Evaluate the impact of increasing the minimum enrollment requirement on sample size and the distribution of baseline patient characteristics using 2 claims-based cohorts.
Methods: We used IBM® MarketScan® commercial and Medicare supplemental claims data to identify 2 cohorts of patients with 1) incident rheumatoid arthritis (RA; index date defined as the first of ≥2 RA-related claims ≥7 days apart) and 2) incident chronic obstructive pulmonary disease (COPD; index defined as first inpatient or first of ≥2 outpatient COPD-related claims 1-365 days apart), excluding patients with prior RA and COPD diagnoses, respectively. Sub-cohorts comprised patients in each population with ≥ 6, 12, 18, and 24 mo of continuous enrollment prior to index. We assessed demographics on index and a selection of comorbidities over each respective baseline period (exclusive of index) for each sub-cohort. We compared patient characteristics among those with 12, 18, and 24 mo to 6 mo baseline via absolute standardized differences (ASD), considering an ASD >0.1 to be a meaningful difference.
Results: Among the patients with RA (N=73,595) and COPD (N=211,877), the median [IQR] days of available baseline enrollment was 460 [64, 992] and 827 [298, 1,121]. The added 6 - 24 mo baseline requirements decreased the RA and COPD cohorts by 42.1% - 58.3% and 22.8% - 41.2%, respectively. Relative to patients with 6 mo required baseline in the RA and COPD cohorts, patients with 24 mo were more likely to be older [mean (SD) age, 57.82 (12.04) vs 56.56 (11.78) years, ASD 0.11; 66.11 (12.93) vs 64.63 (12.90) years, ASD 0.11] and have Medicare benefits (ASD 0.09, 0.11), acute respiratory infection (ASD 0.67; 0.51) and hyperlipidemia (ASD 0.49; 0.51).
Conclusions: In addition to increasing capture of chronic conditions in baseline, requiring a longer minimum enrollment for capturing baseline characteristics reduces the sample size and may also impact the demographic makeup of a cohort. Descriptive analysis to compare cohort characteristics can help to better understand the bias-precision tradeoff for lengthening enrollment requirements.