Propensity score matching enables comparison of rare disease product registry data with an external study comparator


Topic:

Other

Poster Number: 116

Author(s):

Richard Able PhD, Shelley Johnson DBA, MBA, Francesco Bibbiani MD, Joel Jiang PhD, Allan Kristensen PhD, Panayiota Trifillis PhD, Claudio Santos MD

Institutions:

1. PTC Therapeutics, 2. PTC Therapeutics, Inc., 3. PTC Therapeutics, Inc., 4. PTC Therapeutics, 5. PTC Therapeutics, 6. PTC Therapeutics, 7. PTC Therapeutics

Product registries are important for studying the real-world and long-term effectiveness of treatments, especially those for rare diseases. However, unlike randomized controlled trials, observational studies lack comparators, and baseline characteristics vary. Propensity score matching (PSM) is a statistical technique used for estimating treatment effects when random assignment of patients is not possible. PSM balances patient characteristics between a study population and an external comparator cohort to generate comparable groups. STRIDE (NCT02369731) is an ongoing registry assessing ataluren use in nonsense mutation Duchenne muscular dystrophy (nmDMD) patients. The CINRG DNHS (NCT00468832) was a longitudinal natural history study of DMD patients receiving standard of care. PSM was used to identify CINRG DNHS patients who were comparable to STRIDE patients to enable assessment of treatment effect, in this case that of ataluren, in the STRIDE study.

Propensity scores were estimated using a logistic regression model with the following predictors of DMD progression as covariates: age at first symptoms, age at first corticosteroid use, duration of deflazacort use; and duration of other corticosteroid use. A local optimal (greedy) algorithm based on the nearest neighbour approach without replacement was used to find matching controls.

STRIDE Registry data were extracted on July 9, 2018. A comparison between the STRIDE effectiveness population (N=181) and the unmatched CINRG DNHS population (N=400) revealed differences between the two cohorts. Propensity matching of the CINRG DNHS population yielded a comparable population (N=181) to the STRIDE effectiveness population (N=181) with respect to age at first symptoms (P=0.4766), age at first corticosteroid use (P=0.8607), duration of deflazacort use (P=0.3642) and duration of the use of other corticosteroids (P=0.8976).

To assess treatment effect, PSM enabled comparison of rare disease product registry data with an external study in terms of age at loss of ambulation, an important disease milestone of DMD progression.