- We used logistic regression with main effects only for each covariate and the outcome is conversion to psychosis.
- Subscales were included assuming a linear relationship with the logit.
- A lasso penalized fitting procedure was used to perform variable selection and estimation of the model. 5-fold cross validation (folds constructed to reflect the ~30% case rate in the sample) was used to determine the penalization parameter. This was carried out using functions from the R package glmnet.
- Validation of the model is based on bootstrapping the performacne measures (AUC, Breir Score) - using procedure outlined in Harrell 1996 Stat in Medicine Vol 15 361 - 387 (Tutorial in Biostatistics: Multivariate Prog. Models: Issues…)
Model Coefficients
- Tables show
- Beta = estimated coefficient for corresponding predictor (note that predictors are NOT standardized)
- Std_Beta = estimated coefficient for corresponding predictor when predictor is standardized (table is ordered by increasing magnitude of this Std_Beta)
- Prop_Sel_BS = proportion of times that the predictor was selected to remain in the model using the bootstrapping procedure
- Selected Variables
P4v |
-0.30 |
-0.43 |
0.99 |
G2 |
-0.18 |
-0.29 |
0.95 |
P1 |
0.28 |
0.28 |
0.90 |
P5 |
0.20 |
0.26 |
0.90 |
Idea_Sev_Base |
0.57 |
0.26 |
0.95 |
race_bin..c.is.0..non.c.is.1 |
0.49 |
0.24 |
0.92 |
N1 |
0.14 |
0.23 |
0.90 |
Behav_Sev_Base |
0.82 |
0.20 |
0.90 |
GAF |
-0.03 |
-0.18 |
0.86 |
P1PD |
0.08 |
0.09 |
0.73 |
G3 |
0.05 |
0.08 |
0.73 |
SI_Base |
0.17 |
0.04 |
0.67 |
Trauma_Sexual |
0.11 |
0.03 |
0.74 |
N5 |
0.02 |
0.03 |
0.66 |
D3 |
0.02 |
0.03 |
0.69 |
GFS..Social |
0.02 |
0.03 |
0.66 |
SB_Base |
0.09 |
0.02 |
0.57 |
X1..no.is.0..yes.is.1 |
0 |
0 |
1 |
Female |
0 |
0 |
1 |
P2 |
0 |
0 |
1 |
famhx1..0.no..1.yes |
0 |
0 |
1 |
P1NP |
0 |
0 |
1 |
Age |
0 |
0 |
1 |
N3 |
0 |
0 |
1 |
P1OB |
0 |
0 |
1 |
P1FR |
0 |
0 |
1 |
G1 |
0 |
0 |
1 |
Trauma_NonSexual |
0 |
0 |
1 |
GFS..Role |
0 |
0 |
1 |
D4 |
0 |
0 |
1 |
D1 |
0 |
0 |
1 |
P1SNG |
0 |
0 |
1 |
G4 |
0 |
0 |
1 |
P3 |
0 |
0 |
1 |
N2 |
0 |
0 |
1 |
schizotypal..scz.is.1..non.is.0 |
0 |
0 |
1 |
N6 |
0 |
0 |
1 |
N4 |
0 |
0 |
0 |
D2 |
0 |
0 |
0 |
P4a |
0 |
0 |
0 |
P4 |
0 |
0 |
0 |
Frequency Distribution of Model-Based Predicted Risks Among Converters and Nonconverters (in sample)

ROC Curves
