Prognostic Groups by Tree-Based Partitioning and Data Refinement Methods
John Crowley, Antje Hoering in Handbook of Statisticsin Clinical Oncology, 2012
Figure 28.5 shows a tree grown on the IMF using the logrank test statistic for splitting, with a constraint of a minimum node size of 5% of the sample size. The tree has 15 terminal nodes. The logrank test statistic and permutation p-value are presented below each split in the tree. The p-value is calculated at each node by permuting the responses over the covariates and recalculating the best split at that node 1000 times and then calculating the proportion of logrank test statistics greater than the observed statistic. At each terminal node, the logarithm of the hazard ratio relative to the leftmost node and the number of cases falling into each terminal node are presented. The logarithm of the hazard ratio is obtained by fitting a Cox (1972) model with dummy variables defined by terminal nodes in the tree. The worst prognostic group are patients with very high B2M (B2M ≥ 8.89) and high CREAT (CREAT ≥ 3.872) and corresponds to an estimated logarithm of the hazard ratio relative to the best prognostic group equal to 74. While the minimum node size was set to be quite large (5% of the sample or approximately 134 observations), the logrank test statistics near the bottom of the tree (and permutation p-values) indicate there may be several nodes that should be combined to simplify the model.
Classical Survival Analysis
Catherine Legrand in Advanced Survival Models, 2021
The most popular test to compare the survival curves between two groups is certainly the logrank test. The idea behind the logrank test is to compare, at each event time, the observed number of events with the expected number of events under the null hypothesis. The information available at each event time is often represented in a contingency table as in Table 2.4.
Clinical Trial with Survival Endpoint
Mark Chang, John Balser, Jim Roach, Robin Bliss in Innovative Strategies, Statistical Solutions and Simulations for Modern Clinical Trials, 2019
The logrank test statistic compares estimates of the hazard functions of the two groups at each observed event time. It is constructed by computing the observed and expected number of events in one of the groups at each observed event time and then adding these to obtain an overall summary across all-time points where there is an event.
Prognostic Markers of Overall Survival in Cancer Patients Attending a Cachexia Support Service: An Evaluation of Clinically Assessed Physical Function, Malnutrition and Inflammatory Status
Published in Nutrition and Cancer, 2021
Kelcey A. Bland, Eva M. Zopf, Meg Harrison, Matthew Ely, Prue Cormie, Enwu Liu, Anna Dowd, Peter Martin
Descriptive statistics were used to report patient demographic and medical characteristics. Single variable Cox proportional hazard regressions were used to estimate a hazard ratio (HR) and 95% confidence interval (CI) for all-cause mortality for all independent variables. Any variable with a p-value of <0.10 from the single variable analysis was included in a multivariable Cox regression. A stepwise backwards elimination procedure was then used to reduce the number of predictors and improve model fit. Prior to conducting the Cox regression, to account for approximately 8% of observations missing at random, multiple imputation by fully conditioned specification was performed and ten imputations were applied. Due to the non-normal distribution of some outcomes from the FAACT and EORTC questionnaires, a log transformation was applied. The Kaplan-Meier product-limit method was then used to describe survival probabilities for significant prognostic markers identified in the multivariable Cox regression. The logrank test was used to describe potential incremental differences in the unadjusted survival probabilities between groups for each variable. SPSS version 26.0 (IBM, Corporation, Armonk, New York) was used to perform the analyses. Due to the exploratory nature of this analysis, adjustments for multiple comparisons were not made and statistically significant outcomes were defined as p < 0.05.
Effect of Dietary Methylseleninic Acid and Se-Methylselenocysteine on Carcinogen-Induced, Androgen-Promoted Prostate Carcinogenesis in Rats
Published in Nutrition and Cancer, 2022
Maarten C. Bosland, Michael J. Schlicht, Yibin Deng, Junxuan Lü
Lesion incidences were compared using the Fisher exact test (one-sided or two-sided as appropriate and indicated in text and tables) and a Chi Square test when comparing more than two groups. Continuous data such as body and organ weights and age at death were compared using one way ANOVA with a post hoc test (Dunnett’s test for comparison with control and Tukey test for comparisons of all groups). For data that were not normally distributed or had significantly different standard deviations, we used a Kruskal–Wallis ANOVA with Dunn’s post hoc test or the Mann–Whitney test for two sample comparisons. Differences in survival were analyzed using a logrank test. Prizm software (GraphPad, San Diego, CA) was used for these analyses
Fish Oil, Plant Polyphenols, and Their Combinations Have No Tumor Growth Promoting Effects on Human Lung and Colon Carcinoma Xenograft Mice
Published in Journal of Dietary Supplements, 2023
Tapas Das, Svyatoslav Dvoretskiy, Cheng Chen, Menghua Luo, Suzette L. Pereira
Prism (GraphPad) was used for graphical and statistical analysis. The logrank test, which evaluates overall survival experience, was used to analyze the significance of the differences between the TTE values of two groups. Logrank analysis includes the data for all animals in a group except those assessed as NTR deaths. Two-tailed statistical analyses were conducted at significance level p = 0.05. The statistical tests were not adjusted for multiple comparisons. Because tests of statistical significance do not provide an estimate of the magnitude of the difference between groups, all levels of significance were described as either significant or not significant within the text.
Related Knowledge Centers
- Clinical Trial
- NONparametric Statistics
- Null Hypothesis
- Proportional Hazards Model
- Statistical Hypothesis Testing
- Survival Analysis
- Censoring
- Cochran–Mantel–Haenszel Statistics
- Data Monitoring Committee
- Likelihood-Ratio Test