Atio; CI, Confidence Interval; AUC, area under the ROC curve. a

Atio; CI, Confidence Interval; AUC, area under the ROC curve. a Odds Ratio for any increase of one unit. { Eledoisin web p-value of the Wald statistic. doi:10.1371/journal.pone.0049843.tStatistical AnalysisAll the considered biomarkers were analysed as continuous variables in their original scale or after an appropriate transformation. Comparison of biomarkers distribution in cases and controls overall as well as according to stage of disease was performed by using the Kolmogorov-Smirnov test [30]. The relationship between each biomarker and the disease status was investigated by resorting to a order ML 281 logistic regression 22948146 model in both univariate and multivariate fashion [31]. In the 12926553 logistic regression model, each regression coefficient is the logarithm of the odds ratio (OR). Under the null hypothesis of no association, the value of OR is expected to be 1.00. The hypothesis of OR = 1 was tested using the Wald Statistic. For each model the biomarker that was statistically significant (alpha = 0.05) in univariate analysis was considered in the initial model of multivariate analysis. A final more parsimonious model was then obtained using a backward selection procedure in which only the variables reaching the conventional significance level of 0.05 were retained (final model). The relationship between each biomarker and disease status was investigated by resorting to a regression model based on restricted cubic splines. The most complex model considered was a fournodes cubic spline with nodes located at the quartiles of thedistribution of the considered biomarker [32]. The contribution of non-linear terms was evaluated by the likelihood ratio test. We investigated also the predictive capability (ie diagnostic performance) of each logistic model by means of the area under the ROC curve (AUC) [33]. This curve measures the accuracy of biomarkers when their expression is detected on a continuous scale, displaying the relationship between sensitivity (true-positive rate, y-axes) and 1specificity (false-positive rate, x-axes) across all possible threshold values of the considered biomarker. A useful way to summarize the overall diagnostic accuracy of the biomarker is the area under the ROC curve (AUC) the value of which is expected to be 0.5 in absence of predictive capability, whereas it tends to be 1.00 in the case of high predictive capacity [33]. To aid the reader to interpret the value of this statistic, we suggest that values between 0.6 and 0.7 be considered as indicating a weak predictive capacity, values between 0.71 and 0.8 a satisfactory predictive capacity and values greater than 0.8 a good predictive capacity [34]. Finally the contribution of each variables to the predictive capability of the final model was investigated by comparing the AUC value in the model with that of the same model without the variable itself. All statistical analyses were performed with the SASFigure 2. ROC Curves deriving from the univariate logistic analysis. ROC curves derived from the univariate logistic analysis corresponding to total cfDNA (AUC = 0.85), integrity index 180/67 (AUC = 0.76), methylated RASSF1A (AUC = 0.69) and BRAFV600E (AUC = 0.64). doi:10.1371/journal.pone.0049843.gCell-Free DNA Biomarkers in MelanomaFigure 3. ROC Curve deriving from the multivariate final logistic model. ROC curve derived from the final multivariate logistic model (AUC = 0.95). doi:10.1371/journal.pone.0049843.gsoftware (Version 9.2.; SAS Institute Inc. Cary, NC) by adopting a significanc.Atio; CI, Confidence Interval; AUC, area under the ROC curve. a Odds Ratio for any increase of one unit. { p-value of the Wald statistic. doi:10.1371/journal.pone.0049843.tStatistical AnalysisAll the considered biomarkers were analysed as continuous variables in their original scale or after an appropriate transformation. Comparison of biomarkers distribution in cases and controls overall as well as according to stage of disease was performed by using the Kolmogorov-Smirnov test [30]. The relationship between each biomarker and the disease status was investigated by resorting to a logistic regression 22948146 model in both univariate and multivariate fashion [31]. In the 12926553 logistic regression model, each regression coefficient is the logarithm of the odds ratio (OR). Under the null hypothesis of no association, the value of OR is expected to be 1.00. The hypothesis of OR = 1 was tested using the Wald Statistic. For each model the biomarker that was statistically significant (alpha = 0.05) in univariate analysis was considered in the initial model of multivariate analysis. A final more parsimonious model was then obtained using a backward selection procedure in which only the variables reaching the conventional significance level of 0.05 were retained (final model). The relationship between each biomarker and disease status was investigated by resorting to a regression model based on restricted cubic splines. The most complex model considered was a fournodes cubic spline with nodes located at the quartiles of thedistribution of the considered biomarker [32]. The contribution of non-linear terms was evaluated by the likelihood ratio test. We investigated also the predictive capability (ie diagnostic performance) of each logistic model by means of the area under the ROC curve (AUC) [33]. This curve measures the accuracy of biomarkers when their expression is detected on a continuous scale, displaying the relationship between sensitivity (true-positive rate, y-axes) and 1specificity (false-positive rate, x-axes) across all possible threshold values of the considered biomarker. A useful way to summarize the overall diagnostic accuracy of the biomarker is the area under the ROC curve (AUC) the value of which is expected to be 0.5 in absence of predictive capability, whereas it tends to be 1.00 in the case of high predictive capacity [33]. To aid the reader to interpret the value of this statistic, we suggest that values between 0.6 and 0.7 be considered as indicating a weak predictive capacity, values between 0.71 and 0.8 a satisfactory predictive capacity and values greater than 0.8 a good predictive capacity [34]. Finally the contribution of each variables to the predictive capability of the final model was investigated by comparing the AUC value in the model with that of the same model without the variable itself. All statistical analyses were performed with the SASFigure 2. ROC Curves deriving from the univariate logistic analysis. ROC curves derived from the univariate logistic analysis corresponding to total cfDNA (AUC = 0.85), integrity index 180/67 (AUC = 0.76), methylated RASSF1A (AUC = 0.69) and BRAFV600E (AUC = 0.64). doi:10.1371/journal.pone.0049843.gCell-Free DNA Biomarkers in MelanomaFigure 3. ROC Curve deriving from the multivariate final logistic model. ROC curve derived from the final multivariate logistic model (AUC = 0.95). doi:10.1371/journal.pone.0049843.gsoftware (Version 9.2.; SAS Institute Inc. Cary, NC) by adopting a significanc.

Leave a Reply