Search

Objective: Covariance analysis (ANCOVA) is a statistical procedure that enables one to compare groups on some quantitative dependent variable while simultaneously controlling for the effects of selected other continuous variables, which co-vary with the dependent. The control variables are called the "covariates". In this study, we aimed to compare the statistical power of variance analysis (ANOVA), in which covariate was omitted, and ANCOVA for different values of the correlation r between the covariate and the dependent variable in randomized studies. Material and Methods: In our study, a simulation was conducted for n = 30, 50 and 100 sample sizes by using 3 different scenarios of Cohen's small, medium and large f effect sizes. In each simulation, a data set was generated with 11 different correlations from 0 to 1 with an increase of 0.1 units between the dependent variable and the covariate, and 5000 simulations were made in each scenario. R package was used for simulation applications. The "MBESS" open source library was used to generate data and the "pwr" library was used for power analysis in the R package. Results: The results of simulation for small and medium effect sizes, showed that statistical power for ANOVA and ANCOVA started to decompose when the covariate-dependent variable correlation was higher than 0.2. ANCOVA reached much more statistical power than ANOVA when covariate-dependent variable correlation was higher than 0.5. Conclusion: The use of ANCOVA instead of ANOVA may considerably reduce the number of patients required when the covariate-dependent variable correlation was higher than 0.2. [ABSTRACT FROM AUTHOR]

Objective: This study aims to perform Turkish adaptation of the 30-item new version of the 25-item STARD criteria updated in 2015, which was originally published in 2003 and had to be filled in many international journals while submitting an article with a view to improving the quality of reporting in diagnosis accuracy studies and forming a standard. The study is believed to contribute to the improvements in publication quality of the Turkish publications by forming a standard for diagnosis accuracy. Material and Methods: STARD 2015 criteria were translated to Turkish by three authors separately, and a consensus was reached by giving the final version of each item with YY's guidance. Two doctors who are academicians checked the quality of the 15 diagnosis accuracy draft with a view to evaluating its repeatability level. Agreement among the raters was checked using GWET AC1, Intraclass Correlation Coefficient (ICC) and Bland-Altman methods. Results: GWET AC1 statistics showed that there was a full match between the evaluation results of the 21 criteria. Eight criteria were between 0.81 and 1.00 and 5 criteria were between 0.61 and 0.80. ICC value was found to be 0.871 in the total scores given to the articles by the doctors. Conclusion: It is predicted that in the near future it will be obligatory to fill in the STARD 2015 guide, which became obligatory in many international journals while submitting an article, with a view to increasing the quality of the studies on diagnosis accuracy in the journals in our country. [ABSTRACT FROM AUTHOR]

Objective: Markers which are used for classification into two groups, such as patient / healthy, benign/malignant or prediction of optimal cut off value for diagnostic test and evaluating the performance of diagnostic tests is evaluated by Receiver Operating Characteristic (ROC) curve in the diagnostic test researches. In classification accuracy research, some variables such as gender and age, commonly is not similar in groups. In these cases, covariates should be considered to estimate in the area under ROC and covariate adjustment for ROC should be performed. This study aims to introduce methods in the literature for the effect of covariate adjustment and to present an application with sample from the health field. Material and Methods: In the study, we introduced methods used in the literature for covariate adjustment and prediction of the area under ROC curves as well as an application with data from the field of urology. In this study, 105 PSA (prostate specific antigen) measurements were taken in order to examine the covariate effect for the age variable and to assess the diagnostic performance of PSA measurements with regard to pathologic methods. Results: Covariate effect were found statistically significant with 0.733 parameter estimation of the age in ROC curves analysis with PSA data (p<0.001). According to the methods (Non-parametric (empirical), non-parametric (normal), semi-parametric (empirical), parametric (normal)) that estimates of the area under ROC curves which is obtained without covariate effect were found 0.708, 0.629, 0.709 and 0.628, respectively, by using PSA measurements. Area under the curve that obtained by covariate adjustment were significantly lower as compared to the traditional ROC with estimation 0.580, 0.577, 0.582 and 0.579. Conclusion: Area under the ROC curves should be estimated with adjustment according to the covariates that could affect the markers value of diagnostic tests performed in concert with matching designs in research. [ABSTRACT FROM AUTHOR]