Last observation carry-forward and last observation analysis

Area of Science:

• Clinical Trials
• Biostatistics
• Statistical Analysis

Background:

• Drop-out in clinical trials is common and often informative, meaning drop-out groups differ from completers.
• Regulatory agencies often require intention-to-treat (ITT) analysis for informative drop-out.
• Last observation carry-forward (LOCF) is a simple ITT method but its theoretical properties are not fully understood.

Purpose of the Study:

• To investigate the theoretical properties of the LOCF analysis of variance (ANOVA) test.
• To evaluate the validity and performance of LOCF in handling informative drop-out.
• To propose a new, asymptotically valid statistical test for clinical trials with informative drop-out.

Main Methods:

• The study theoretically analyzes the asymptotic validity of the LOCF one-way ANOVA test.
• The performance of LOCF is compared to a newly proposed asymptotically valid test.
• Simulation studies are conducted to assess finite sample performance.

Main Results:

• The LOCF test is asymptotically valid when comparing two treatments with equal group sizes, irrespective of drop-out informativeness.
• In other scenarios, the LOCF test's asymptotic size deviates from the nominal size, often being too small with informative drop-out.
• This deviation leads to reduced power in detecting treatment effects.

Conclusions:

• The LOCF test has limitations in general clinical trial settings with informative drop-out.
• A novel asymptotically valid test is proposed to accurately compare treatment effects across subpopulations defined by drop-out timing.
• The proposed test offers improved power and validity compared to LOCF in complex drop-out situations.