For predictive purposes, PLS1 models were constructed for individual Y variables to increase model-specificity and reliability.
Nevertheless, a rough estimation of model specificity was indirectly and tentatively estimated by further independent techniques (i.e., chemical and histological examinations) described in the Materials and methods section.
Meanwhile, when we combined the variables of SNAP, gestational age, apnea of prematurity, PDA, and surfactant use in the final model, specificity increased to 90%, and positive likelihood ratio increased to 7.4.
The PD factor best accounted for Discrete time (31% of the total deviance) and Model specificity (18%).
The estimation for the threshold score of 12/13 fitted the model sensitivity at 87.5% and model specificity at 85.7%, for identifying major depression.
The area under ROC curve AUC was found 0.7470 and the logistic estimate for the threshold score of 8/9 fitted the model sensitivity at 76.7% and model specificity at 68.3%.
Receiver Operating Characteristic (ROC) analysis demonstrates that the coupled system (CRESLIDE) has higher specificity (94.10%) and higher sensitivity (11.36%) compared to the original SLIDE model (specificity = 93.32%, sensitivity = 10.23%) and a well-known landslide model (TRIGRS, whose sensitivity is 6.98%).
Therefore, it is neccessary to investigate multivariate and feature-selective methods to model the specificity and to extract rules from the model.
Model sensitivity, specificity, and positive and negative likelihood ratios were also calculated, and the best model cut-off points were considered to be those that corresponded to the highest sum of specificity and sensitivity.
They find that the TOMM40 SNP near APOE alone has poor predictive value; removing it from the model reduces specificity and sensitivity by only 1%%.
We evaluated model sensitivity, specificity, and percent agreement using the median of the predicted nitrate level (6.5 mg/L as nitrate-N) as a cutpoint for high and low exposure categories.
