In resplab/epicR: R Package for Evaluation Platform in COPD

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(epicR)

Previous Calibration

Assessing calibration of exacerbations with previous EPIC's (v0.31.3) equations:

inputs <- get_input()$values
inputs$exacerbation$ln_rate_betas = t(as.matrix(c(intercept = -3.4, female = 0, age = 0.04082 * 0.1, fev1 = -0, smoking_status = 0, gold1 = 1.4 , gold2 = 2.0 , gold3 = 2.4 , gold4 = 2.8 , diagnosis_effect = 0.9)))
validate_exacerbation(5e4, inputs)

This equation for exacerbations included a term for diagnosis. As a result, whether a person was diagnosed or not would affect the exacerbation rates, which does not have causal face validity, and is particularly undesirable for case detection studies that assess diagnosis.

If we remove the diagnosis term and go back to the previous calibration, it will look like this:

inputs$exacerbation$ln_rate_betas <- t(as.matrix(c(intercept = 1.7, female = 0, age = 0.04082 * 0.1, fev1 = -1.5, smoking_status = 0, gold1 = 0.6 , gold2 = 0.35 , gold3 = 0.08 , gold4 = -0.35 , diagnosis_effect = 0)))
validate_exacerbation(5e4, inputs)

The following recalibrations were assesses and Recalibration 5 was implemented in epicR v0.35.0

Recalibration 1:

Let's try to improve that by lowering GOLD coefficients:

inputs$exacerbation$ln_rate_betas <- t(as.matrix(c(intercept = 1.7, female = 0, age = 0.04082 * 0.1, fev1 = -1.5, smoking_status = 0, gold1 = 0.3 , gold2 = 0.1 , gold3 = 0.08 , gold4 = -0.35 , diagnosis_effect = 0)))
validate_exacerbation(5e4, inputs)

Recalibration 2:

We can now lower GOLD2 coefficient even more and perhaps increase smoking coefficient to compensate for the loss in the diagnosed patients.

inputs$exacerbation$ln_rate_betas <- t(as.matrix(c(intercept = 1.7, female = 0, age = 0.04082 * 0.1, fev1 = -1.5, smoking_status = 0.5, gold1 = 0.3 , gold2 = 0.05 , gold3 = 0.08 , gold4 = -0.35 , diagnosis_effect = 0)))
validate_exacerbation(5e4, inputs)

Recalibration 3:

Let's lower GOLD2 dramatically:

inputs$exacerbation$ln_rate_betas <- t(as.matrix(c(intercept = 1.7, female = 0, age = 0.04082 * 0.1, fev1 = -1.5, smoking_status = 0.5, gold1 = 0.3 , gold2 = -0.5 , gold3 = 0.08 , gold4 = -0.35 , diagnosis_effect = 0)))
validate_exacerbation(5e4, inputs)

Recalibration 4:

Bring GOLD2 up a bit:

inputs$exacerbation$ln_rate_betas <- t(as.matrix(c(intercept = 1.7, female = 0, age = 0.04082 * 0.1, fev1 = -1.5, smoking_status = 0.5, gold1 = 0.3 , gold2 = -0.2 , gold3 = 0.08 , gold4 = -0.35 , diagnosis_effect = 0)))
validate_exacerbation(5e4, inputs)

Recalibration 5:

GOLD2 back a bit again.

inputs$exacerbation$ln_rate_betas <- t(as.matrix(c(intercept = 1.7, female = 0, age = 0.04082 * 0.1, fev1 = -1.5, smoking_status = 0.5, gold1 = 0.3 , gold2 = -0.3 , gold3 = 0.08 , gold4 = -0.35 , diagnosis_effect = 0)))
validate_exacerbation(5e4, inputs)

Recalibration 6:

Reducing intercept and adding to the smoking term.

inputs$exacerbation$ln_rate_betas <- t(as.matrix(c(intercept = 1.4, female = 0, age = 0.04082 * 0.1, fev1 = -1.5, smoking_status = 0.7, gold1 = 0.3 , gold2 = -0.3 , gold3 = 0.08 , gold4 = -0.35 , diagnosis_effect = 0)))
validate_exacerbation(5e4, inputs)