I am trying to determine the minimally important change (MIC) of a frailty instrument using an anchor-based approach outlined below.
Step 1. Fit a logistic regression model between change_in_fi
(a change in the frailty measure) and srh_decline
(1/0 referring to a decline in self-reported health), which is my anchor.
Step 2. Use the predicted probabilities of the above model to fit an ROC curve.
Step 3. Identify the Youden Index, i.e. point on ROC curve which maximises the sensitivity and specificity.
Step 4. I am stuck on here - how do I find out the threshold value of change_in_fi
that the above Youden Index refers to? This value is my MIC.
Please find a mock dataset and code below:
library(pROC)
dat = structure(list(change_in_fi = structure(c(-0.05825, -0.0375,
0.04575, 0.202, 0.01675, -0.1, -0.0125, -0.04775, 0.00624999999999999,
-0.00625000000000001, -0.01875, 0.052, -0.01025, -0.025, 0.04375,
0.00825, -0.048, 0.09575, 0.073, 0), label = "FI score", class = c("labelled",
"numeric")), srh_decline = c(0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0,
1, 1, 0, 0, 0, 0, 0, 0, 1)), row.names = c(NA, 20L), class = "data.frame")
# Step 1
mod = glm(srh_decline ~ change_in_fi, family=binomial, data=dat)
# Step 2
rocobj= roc(mod$y, mod$fitted.values)
# Step 3
coords(rocobj, "best")
```