For dichotomous outcomes, true risk probabilities cannot intrinsically be known.
In binary outcomes (true/false) in diagnostic tests e.g. for disease we can have four outcomes: true positives (TPs), false positives (FP), true negatives (TN) and false negatives (FN).
A diagnostic algorithm can theoretically generate 4 different outcomes: true positive, false positive, true negative, and false negative.
In our model, the sum of the probabilities of true-positive outcomes, false-negative outcomes, true-negative outcomes, and false-positive outcomes is equal to one.
The screening mammogram can have four possible outcomes: true-positive, false-negative, true-negative, and false-positive, from which the model predicts the two primary events CDR and recall rate.
When combined in a multivariate Poisson regression model with LAM sensitivity as the outcome (true positives coded as "0", false negatives coded as "1"), the influence of HIV status and sex was less prominent than in separate univariate models.
Using this procedure, we calculated the number of true positives (invasive prediction and invasive outcome), false positives (invasive prediction and DCIS outcome), true negatives (DCIS prediction and DCIS outcome), and false negatives (DCIS prediction and invasive outcome) at all possible thresholds between 0 and 100%.
It should be noted that, whilst such dichotomization is useful in demonstrating associations with outcome, true utility of ERβ variant mRNA measurement will only be demonstrated with larger patient cohorts and may be better achieved by treating mRNA quantitation as a continuous variable, as in other RTPCR based outcome predictors [ 24].
In the simplest case, when only outcome, true, and surrogate exposures are considered, the corrected estimate of risk is given by, and the variance of the corrected estimate is given by Rosner et al. (1990) developed computational details for models with multiple covariates measured with and without error.
The issues also rise when substituting auxiliary outcome for true outcome; see Ellenberg and Hamilton ([1989]), Prentice ([1989]) and Fleming ([1992]).
ROC curves were generated based on the predicted outcome and true outcome.
