Presentation Title

New Methodology for Nonparametric Survival Comparison Applied to Kaplan-Meier and Life-tables Methods

Speaker Credentials

Associate Professor

Speaker Credentials

Ph.D.

College

Dr. Kiran C. Patel College of Osteopathic Medicine, DO

Location

Signature Grand, Davie, Florida, USA

Format

Podium Presentation

Start Date

25-4-2008 12:00 AM

End Date

25-4-2008 12:00 AM

Abstract

Objectives. 1. Identify the most appropriate test to be used when the equality of survival curves is hypothesized, and use the new taxonomy through the guidelines. 2. Define appropriate alternative hypotheses for the given data based survival curves and assess their specific characteristics. 3. Develop a reliable nonparametric survival curves comparison based on the use of the critical margins and the guidelines. Background. Hypothesis tests of the equality of Kaplan- Meier and life-table survival curves is typically accomplished using one of the available methods designed for this purpose but, without doubt, the logrank test is the one most commonly used. Perhaps the reason for the popularity of the logrank test rests in its ready availability in almost all statistical software packages. However, what many users do not appreciate is that the logrank test has very low power for some alternative hypothesis. Methods. This research used statistical power simulations for nonparametric tests under various alternatives and a meta-analysis of 172 published papers in the New England Journal of Medicine. Results. This study presents a new taxonomy for the test of the equality of survival curves that includes critical margins that are used to enhance the new created methodology. A specific guideline was created to ease the researcher tasks. The simulations led to tables of statistical comparison, specifically built for the purpose of the research. Conclusion. There is no general test that fits all comparisons, therefore the testing should be performed according to the alternative hypothesis of interest and the relationship of the hazard functions. Acknowledgement. SAS, STATA, and S-plus statistical packages where used to accomplish this research.

This document is currently not available here.

Share

COinS
 
Apr 25th, 12:00 AM Apr 25th, 12:00 AM

New Methodology for Nonparametric Survival Comparison Applied to Kaplan-Meier and Life-tables Methods

Signature Grand, Davie, Florida, USA

Objectives. 1. Identify the most appropriate test to be used when the equality of survival curves is hypothesized, and use the new taxonomy through the guidelines. 2. Define appropriate alternative hypotheses for the given data based survival curves and assess their specific characteristics. 3. Develop a reliable nonparametric survival curves comparison based on the use of the critical margins and the guidelines. Background. Hypothesis tests of the equality of Kaplan- Meier and life-table survival curves is typically accomplished using one of the available methods designed for this purpose but, without doubt, the logrank test is the one most commonly used. Perhaps the reason for the popularity of the logrank test rests in its ready availability in almost all statistical software packages. However, what many users do not appreciate is that the logrank test has very low power for some alternative hypothesis. Methods. This research used statistical power simulations for nonparametric tests under various alternatives and a meta-analysis of 172 published papers in the New England Journal of Medicine. Results. This study presents a new taxonomy for the test of the equality of survival curves that includes critical margins that are used to enhance the new created methodology. A specific guideline was created to ease the researcher tasks. The simulations led to tables of statistical comparison, specifically built for the purpose of the research. Conclusion. There is no general test that fits all comparisons, therefore the testing should be performed according to the alternative hypothesis of interest and the relationship of the hazard functions. Acknowledgement. SAS, STATA, and S-plus statistical packages where used to accomplish this research.