1231 Unlocking insights: Lesson learned from the tokenization of a psoriasis trial

A randomized controlled trial (RCT) of risankizumab and apremilast for moderate plaque psoriasis was selected as a pilot study. Patient-specific tokens were generated after obtaining written informed consent, which outlined the use of PII for token generation, from all participants at trial enrollment from US sites. We used the ConcertAI claims dataset of both open (payer agnostic) and closed (payer specific) claims (01/2018 to 06/2023) and the de-identified tokens to estimate the overlap (proportion of trial participants with information in claims) and examine any discrepancies between the RCT and RWD datasets.

A total of 105 trial participants were included and tokenized. The overlap was approximately 87% (91/105) in open claims and 41% (43/105) in closed claims. Among the demographics, race was consistently underreported in claims for all four race categories of the trial, with a higher reporting of Hispanic or Latino ethnicity compared to non-Hispanic or Latino ethnicity. Out of the 43 trial participants with plaque psoriasis identified in closed claims, only 65% had a psoriasis diagnosis in claims before the administration of trial treatment. Baseline comorbidities in the RCT and RWD populations were similar, with minor discrepancies in 10 out of the 20 analyzed conditions. Larger discrepancies were observed in renal disease (2 RCT vs 0 RWD), psoriatic arthritis (6 RCT vs 3 RWD), and atopic dermatitis (0 RCT vs 2 RWD). The under-reporting in closed claims may reflect a lack of insurance coverage (enrollment), as participants had a median cumulative enrollment duration of 19 (IQR 9, 31) months, and 40% (17/43) of them had continuous enrollment for at least 6 months before and including the treatment administration date.

The initial results of integrating RCT data and RWD yields opportunities to gather insights on trial participants. Careful consideration of limitations due to reduced sample size after overlap, segmented enrollment and inherent biases of available data sources is necessary when designing studies integrating RCT data and RWD.