Skip to main content

Table 4 Performance of full variable model against the null models using over 10-fold cross validation repeated 50 times to predict virological failure

From: Pharmacy refill adherence outperforms self-reported methods in predicting HIV therapy outcome in resource-limited settings

Model

Goodness-of-fit [avg (st.dev)] with respect to failure

Area under roc

Accuracy

Sensitivity

Specificity

Majority class (null model)

0.50 (0.00)

66.07 (2.79)

0.00 (0.00)

1.00 (0.00)

LR on all variables with AIC stepwise

0.62 (0.14)

66.79 (10.10)

0.37 (0.20)

0.82 (0.12)

LR on all variables with boost stepwise

0.64 (0.14)

67.77 (9.82)

0.39 (0.20)

0.83 (0.11)

Random Forest on all variables

0.59 (0.15)

67.80 (9.11)

0.28 (0.19)

0.88 (0.10)

Decision tree on all variables

0.55 (0.12)

65.72 (8.46)

0.21 (0.21)

0.89 (0.13)