What does PA stand for?

ClinicalValidity.Rmd

@@ -233,7 +233,7 @@ We estimated the values for True Positives, True Negatives, False Positives, and
 
 2. Similarly, if the cohort definition did not include a person from the evaluation cohort, i.e. the cohort definition considered the person a “negative,” one minus the predicted probability for the phenotype for that person was the expected value of counts contributed to True Negatives and was added to it, and, in parallel, the predicted probability for the phenotype was the expected value of counts contributed to the False Negatives and was added to it. For example, PersonId 017 had a predicted probability of 1% for the presence of the health outcome (and, correspondingly, 99% for the absence of the health outcome) and 1.00 – 0.01 = 0.99 was added to the True Negatives and 0.01 was added to the False Negatives. This was repeated for all the persons from the evaluation cohort not included in the cohort definition (i.e., PersonIds 018, 020, 021, and 024).
 
-After adding these values over the full set of persons in the evaluation cohort, we filled the four cells of the confusion matrix with the expected values of counts for each cell, and we were able to create point estimates of the PA performance characteristics like sensitivity, specificity, and positive predictive value (Figure 1C). We emphasize that these expected cell counts cannot be used to assess the variance of the estimates, only the point estimates. In the example, the sensitivity, specificity, PPV, and NPV were 0.99, 0.63, 0.42, and 0.99, respectively.
+After adding these values over the full set of persons in the evaluation cohort, we filled the four cells of the confusion matrix with the expected values of counts for each cell, and we were able to create point estimates of the PA(?) performance characteristics like sensitivity, specificity, and positive predictive value (Figure 1C). We emphasize that these expected cell counts cannot be used to assess the variance of the estimates, only the point estimates. In the example, the sensitivity, specificity, PPV, and NPV were 0.99, 0.63, 0.42, and 0.99, respectively.
 
 Determining the performance characteristics of the cohort definition uses the function `testPhenotype`. This function uses the output from the prior two steps where we created the model and evaluation cohorts. We would set the modelFileName parameter to the RDS file output from createPhenoModel function, in this example, “c:/temp/lr_results_5XMI_train_myCDM_ePPV0.75_20181206V1.rds”. We would set the resultsFileName parameter to the RDS file output from createEvalCohort function, in this example, “c:/temp/lr_results_5XMI_eval_myCDM_ePPV0.75_20181206V1.rds”. To test the cohort definition we wish to use in our study, we set the `cohortPheno` to the cohort ID for that cohort definition. We can set the `phenText` parameter to some human readable description for the cohort definition, such as “MI Occurrence, Hospital In-Patient Setting”. We will set the `testText` parameter to some human readable description for the xSpec definition, such as “5 X MI.” The output from this step is a data frame that contains the performance characteristics for the cohort definition tested. The settings for the `cutPoints` parameter is a list of values that will be used to develop the performance characteristics results. The performance characteristics are usually calculated using the “expected values” as described in Figure 1. To retrieve the performance characteristics based on the expected values, we include “EV” in the list for the `cutPoints` parameter. We may also want to see the performance characteristics based on specific predicted probabilities, i.e., cut points. For example, if we wanted to see the performance characteristics of all those at or above a predicted probability of 0.5 were considered positive for the health outcome and all those under a predicted probability of 0.5 were considered negative, we would add “0.5” to the `cutPoints` parameter list. For example: