Statistics in medicine
-
Statistics in medicine · Apr 2001
Comparative StudyComparisons of two-part models with competitors.
Two-part models arise when there is a clump of 0 observations in a distribution of continuous non-negative responses. Several methods for comparing two such distributions are available. These include the straightforward application of the z-test (or t-test), the Wilcoxon-Mann-Whitney rank sum test, the Kolmogorov-Smirnov test, and three tests that use a 2 degree of freedom chi(2) test based on the sum of the test for equality of proportions and a conditional chi(2) test for the continuous responses. ⋯ If the reverse holds, the difference in the proportion of zeros reinforces the difference in means and some single-part models (the rank sum or Kolmogorov-Smirnov) do best. In those cases, the two-part models are not far behind, although statistically significantly poorer with respect to power. Published in 2001 by John Wiley & Sons, Ltd.
-
Statistics in medicine · Dec 2000
Assessing the potential for bias in meta-analysis due to selective reporting of subgroup analyses within studies.
Subgroup analysis is frequently used to investigate heterogeneity in meta-analysis. Subgroup data are not always available, and researchers should record what data were available for each trial. If data were not available, and it is known that the subgroup data were collected, the potential for selective reporting should be considered. ⋯ The conclusion in the original review, that benefit is limited to primigravidae, was based on subgroup analysis using the three trials out of five which reported on subgroups. We developed a method of sensitivity analysis that imputes data for the missing subgroups to assess the robustness of the results and the conclusions drawn. In this particular example, our analysis indicates that the estimate of effect reported in the review is most likely to overestimate the true effect and the conclusion that benefit is limited to primigravidae may be false.
-
Statistics in medicine · Dec 2000
Historical ArticleReflections on statistics at the London School of Hygiene and Tropical Medicine 30 years ago.
A course leading to the Master of Science (MSc) degree in Medical Statistics was started at the London School of Hygiene and Tropical Medicine in 1968. The events leading up to this initiative are outlined in the context of earlier developments in statistics at the School and the general growth of opportunities in statistical education.
-
Statistics in medicine · Oct 2000
Comparative StudyTesting whether treatment is 'better' than control with ordered categorical data: an evaluation of new methodology.
A new test procedure is presented for the problem of testing whether a treatment is better than a control when there is ordered categorical data. The new test is based on the methodology developed for general 'one-sided' alternatives by Cohen and Sackrowitz. ⋯ As predicted by the theoretical work by Cohen and Sackrowitz, the new test is seen to be preferable to the Wilcoxon-Mann-Whitney test. Computer programs to assist implementation of the new test are made available.
-
Statistics in medicine · Oct 2000
Comparative StudyAnalysis of a cluster randomized trial with binary outcome data using a multi-level model.
The use of multi-level logistic regression models was explored for the analysis of data from a cluster randomized trial investigating whether a training programme for general practitioners' reception staff could improve women's attendance at breast screening. Twenty-six general practices were randomized with women nested within them, requiring a two-level model which allowed for between-practice variability. Comparisons were made with fixed effect (FE) and random effects (RE) cluster summary statistic methods, ordinary logistic regression and a marginal model based on generalized estimating equations with robust variance estimates. ⋯ Estimates of the variance components were of particular interest in this example. Additionally, parametric bootstrap methods within the multi-level model framework provide confidence intervals for these variance components, as well as a confidence interval for the effect of intervention which allows for the imprecision in the estimated variance components. The assumption of normality of the random effects can be checked, and the models extended to investigate multiple sources of variability.