PII-107 - A SURVEY OF REGRESSION ANALYSIS METHODS ADOPTED IN RENAL IMPAIRMENT STUDIES.

S. Yi¹, A. Al-Khouja¹, T. Zhang², S. Doddapaneni¹; ¹US Food and Drug Administration, Silver Spring, MD, USA, ²University of Pittsburgh, Pittsburgh, PA, USA.

Background: Renal impairment (RI) studies are conducted to assess the effect of RI on pharmacokinetics (PK) and inform dose adjustment in patients with RI. Since 2010, the FDA RI guidance has endorsed a regression approach, where renal function and PK parameters are treated as continuous variables, over an analysis where renal function is treated as a categorical variable, i.e., ANOVA or ANCOVA. However, in drug labeling, fold-changes of PK parameters in RI groups (i.e., normal, mild, moderate, and severe groups as a categorical variable) compared to a normal renal function group are often described. The current FDA RI guidance does not advise on how to conduct and report regression analyses. As such, we surveyed how regression analyses have been used in RI studies of approved drugs.

Methods: FDA submissions from drugs approved between the years 2015 and 2021 were screened. Submissions containing a RI study report were evaluated to explore whether and how regression analyses were conducted.

Results: Out of 323 drugs, dedicated RI studies were conducted for 96. Out of 96 studies, 24 (25%) did not perform a regression analysis, including 13/34 (38%) RI studies with a reduced design. Among the remaining studies, most (N = 64) used regression as secondary or exploratory analysis, showing a trend of relationships between PK parameters and renal function. Regression results from only 5 studies were described in the labeling, where PK parameters of a virtual patient in each RI group derived from the regression model were compared to those of a patient with normal renal function.

Conclusion: In contrast to FDA RI guidance recommendations, a regression approach was rarely used as primary analysis. Given the small sample sizes in RI studies, we suggest that the PK predictability of a regression model should be verified to support labeling.