Statistics in medicine
-
Statistics in medicine · Feb 2000
On the misuses of artificial neural networks for prognostic and diagnostic classification in oncology.
The application of artificial neural networks (ANNs) for prognostic and diagnostic classification in clinical medicine has become very popular. In particular, feed-forward neural networks have been used extensively, often accompanied by exaggerated statements of their potential. In this paper, the essentials of feed-forward neural networks and their statistical counterparts (that is, logistic regression models) are reviewed. ⋯ In applications of ANNs to survival data, further difficulties arise. Finally, the results of a search in the medical literature from 1991 to 1995 on applications of ANNs in oncology and some important common mistakes are reported. It is concluded that there is no evidence so far that application of ANNs represents real progress in the field of diagnosis and prognosis in oncology.
-
Statistics in medicine · Jan 2000
Sample size calculation for planning group sequential longitudinal trials.
Procedures are developed in this paper for sample size calculations for planning a group sequential longitudinal trial with various correlation structures, using a test statistic based on generalized estimating equations (Lee, Kim and Tsiatis) and group sequential boundaries based on type I error spending functions (Lan and DeMets).
-
Patient non-compliance and drop-out can bias analyses of clinical trial data. I describe a parametric model for treatment cross-over and drop-out and demonstrate how the concept of ignorability, originally defined for incomplete-data problems, can elucidate sources of bias in clinical trials. I discuss some implications of the theory and present simulation examples that illustrate the potential effects of non-ignorable cross-over and drop-out on bias and power.
-
Statistics in medicine · Feb 1999
An issue of statistical analysis in controlled multi-centre studies: how shall we weight the centres?
When analysing data from a controlled multi-centre study, an important issue is how to weight each centre to assess the overall treatment effect. The unweighted analysis, which weights all centres equally, was recommended by many statisticians and the U. S. ⋯ The weighted analysis, which weights centres relative to the number of patients in them, was considered not meaningful in the presence of treatment-by-centre interaction. This paper demonstrates why we should hesitate to use the unweighted analysis as the primary statistical method of a study from a statistical power perspective. It also shows that the weighted analysis is meaningful, even in the presence of treatment-by-centre interaction, and that it is generally the preferred approach.