Limits...
Survival analysis Part III: multivariate data analysis -- choosing a model and assessing its adequacy and fit.

Bradburn MJ, Clark TG, Love SB, Altman DG - Br. J. Cancer (2003)

View Article: PubMed Central - PubMed

Affiliation: Cancer Research UK/NHS Centre for Statistics in Medicine, Institute of Health Sciences, Old Road, Oxford OX3 7LF, UK. mike.bradburn@cancer.org.uk

AUTOMATICALLY GENERATED EXCERPT
Please rate it.

In this series of papers, we have described a selection of statistical methods used for the initial analysis of survival time data (Clark et al, 2003), and introduced a selection of more advanced methods to deal with the situation where several factors impact on the survival process (Bradburn et al, 2003)... Whereas the focus of the previous paper was to outline the purpose and interpretation of statistical models for survival analysis, we concentrate here on approaches with which to undertake the actual modelling process... In other words, the aim of this paper is to promote the correct use of the models that have been suggested for the analysis of survival data... When used inappropriately, statistical models may give rise to misleading conclusions... The table of coefficients for the full AFT multivariate model was presented in the previous paper (Bradburn et al, 2003)... A simpler model would be to consider just the performance status, cell type and treatment covariates... Removing the remaining covariates reduces the model likelihood, but not to a significant degree (χ=3.34 on 8 degrees of freedom; P=0.91)... The new time ratios, confidence intervals and P-values are presented in Table 1 They are virtually unchanged from the previous analysis, and thus the earlier conclusions remain the same... Several formal statistical tests have been proposed for assessment of proportionality of hazards... A simulation study by Ng'andu (1997) described and compared several tests in the Cox PH framework, and concluded that the (weighted) scaled Schoenfeld residuals test (Grambsch and Therneau, 1994), the linear correlation test (Harrell, 1986) and the time-dependent covariate test (Cox, 1972) were the most powerful diagnostic tools for proportionality... As with the log(−log(survival)) plot in PH models, this is a useful but limited approach as departures from linearity could be due to the AFT model being inappropriate or that one or more important covariates have been omitted... This paper has sought to demonstrate the models introduced in the previous paper in this series (Bradburn et al, 2003), to offer practical advice on how to select a method that represents the data fairly, and how to present and interpret it... In the final paper of this series, we introduce models that extend the types of models described here to incorporate recurrent events... We also present approaches to modelling continuous covariates in a nonlinear fashion, validating models and discuss alternatives when fundamental censoring assumptions do not hold.

Show MeSH
Martingale residuals plotted against (A) patient age and (B) FIGO stage; median for each stage is denoted by a horizontal line.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2376927&req=5

fig1: Martingale residuals plotted against (A) patient age and (B) FIGO stage; median for each stage is denoted by a horizontal line.

Mentions: Figures 1AFigure 1


Survival analysis Part III: multivariate data analysis -- choosing a model and assessing its adequacy and fit.

Bradburn MJ, Clark TG, Love SB, Altman DG - Br. J. Cancer (2003)

Martingale residuals plotted against (A) patient age and (B) FIGO stage; median for each stage is denoted by a horizontal line.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2376927&req=5

fig1: Martingale residuals plotted against (A) patient age and (B) FIGO stage; median for each stage is denoted by a horizontal line.
Mentions: Figures 1AFigure 1

View Article: PubMed Central - PubMed

Affiliation: Cancer Research UK/NHS Centre for Statistics in Medicine, Institute of Health Sciences, Old Road, Oxford OX3 7LF, UK. mike.bradburn@cancer.org.uk

AUTOMATICALLY GENERATED EXCERPT
Please rate it.

In this series of papers, we have described a selection of statistical methods used for the initial analysis of survival time data (Clark et al, 2003), and introduced a selection of more advanced methods to deal with the situation where several factors impact on the survival process (Bradburn et al, 2003)... Whereas the focus of the previous paper was to outline the purpose and interpretation of statistical models for survival analysis, we concentrate here on approaches with which to undertake the actual modelling process... In other words, the aim of this paper is to promote the correct use of the models that have been suggested for the analysis of survival data... When used inappropriately, statistical models may give rise to misleading conclusions... The table of coefficients for the full AFT multivariate model was presented in the previous paper (Bradburn et al, 2003)... A simpler model would be to consider just the performance status, cell type and treatment covariates... Removing the remaining covariates reduces the model likelihood, but not to a significant degree (χ=3.34 on 8 degrees of freedom; P=0.91)... The new time ratios, confidence intervals and P-values are presented in Table 1 They are virtually unchanged from the previous analysis, and thus the earlier conclusions remain the same... Several formal statistical tests have been proposed for assessment of proportionality of hazards... A simulation study by Ng'andu (1997) described and compared several tests in the Cox PH framework, and concluded that the (weighted) scaled Schoenfeld residuals test (Grambsch and Therneau, 1994), the linear correlation test (Harrell, 1986) and the time-dependent covariate test (Cox, 1972) were the most powerful diagnostic tools for proportionality... As with the log(−log(survival)) plot in PH models, this is a useful but limited approach as departures from linearity could be due to the AFT model being inappropriate or that one or more important covariates have been omitted... This paper has sought to demonstrate the models introduced in the previous paper in this series (Bradburn et al, 2003), to offer practical advice on how to select a method that represents the data fairly, and how to present and interpret it... In the final paper of this series, we introduce models that extend the types of models described here to incorporate recurrent events... We also present approaches to modelling continuous covariates in a nonlinear fashion, validating models and discuss alternatives when fundamental censoring assumptions do not hold.

Show MeSH