501 – Methods for Safety, Quality, and Benefit-Risk
Modeling Censored Stability Data Using Tobit Mixed Models
Seth Clark
Merck
It is not uncommon to encounter stability data that are censored due to results being below a quantitation limit of the assay. Often various crude methods are applied to the data, such as imputing values or modeling only the portion of the stability trend that is uncensored. An alternative method is proposed utilizing Tobit regression applied in the mixed model context to account for test-to-test variation. We show the approach allows one to model the distribution mean across the entire stability profile and reduces bias due to censored values.