QALYs in adults with cerebral palsy: mapping from the San Martin Scale onto the EQ-5D-5L instrument
Author
Nova-Díaz, D.M., Adin, A., and Sánchez-Iriso, E.
Published
Jun/2025
This repository contains the original data and R code to reproduce the analyses presented in the paper entitled “QALYs in adults with cerebral palsy: mapping from the San Martin Scale onto the EQ-5D-5L instrument” (Nova-Díaz et al., 2025).
Data
Data were collected from 72 participants in the Study of Adults Cerebral Palsy in Navarre, Spain (EPCANA), who completed both the St. MQoL-S and the EQ-5D-5L survey instruments simultaneously.
In all cases, ANOVA F-tests and stepwise regression are employed to evaluate alternative models and identify the most appropriate predictor variables.
Mapping model performance
The script 3_Mapping.R fits several direct mapping model candidates (using ordinary least squares, generalized linear model, and Tobit regression analysis) to estimate EQ-5D-5L utilities, with St. MQoL-S total and domain scores as explanatory variables.
It also generates the following figures and tables:
Table 2: Goodness-of-fit results of model candidates
Table S3: Estimated regression coefficients of model candidates
Figure S2: Scatterplot between observed and predicted EQ-5D utility scores (Model 1)
Figure S3: Scatterplot between observed and predicted EQ-5D utility scores (Model 2)
Figure 5: Empirical cumulative density functions of observed and predicted EQ-5D utility scores (Model 1)
Cross Validation
To ensure generalization for out-of-sample individuals, the script 4_CrossValidation.R employs a repeated 5-fold cross-validation approach to select the mapping model with the best predictive performance.
Results are summarised in the figures and tables:
Table 3: Goodness-of-fit using repeated 5-fold cross validation
Figure 4: Goodness-of-fit using repeated 5-fold cross validation
Examples of the mapping algorithm
Or best mapping algorithm (OLS - Model 1) can be described as
After fitting the model, predictions of EQ-5D-5L utility values for an adult person with cerebral palsy can be obtained as follows:
## Fit the core model providing the best mapping algorithm: OLS - Model 1 ##Model.final <-lm(EQ.INDEX ~ STMartin.INDEX + AGE + SEX + CPT, data=Database)
## EXAMPLE 1:## - Prediction for a female (SEX="0") who has a St. MQoL-S total score of 106,## 30 years old, and has a CP of the spastic type (CPT="1")newdata <-data.frame(STMartin.INDEX=106, SEX="0", AGE=30, CPT="1")newdata$EQ.INDEX <-predict(Model.final, newdata, interval="confidence")print(newdata)
STMartin.INDEX SEX AGE CPT EQ.INDEX.fit EQ.INDEX.lwr EQ.INDEX.upr
1 106 0 30 1 0.3357393 0.2738009 0.3976776
## EXAMPLE 2:## - Prediction for a male (SEX="1") who has a St. MQoL-S total score of 100,## 40 years old, and has a CP of the dyskinetic type (CPT="2")newdata <-data.frame(STMartin.INDEX=100, SEX="1", AGE=40, CPT="2")newdata$EQ.INDEX <-predict(Model.final, newdata, interval="confidence")print(newdata)
STMartin.INDEX SEX AGE CPT EQ.INDEX.fit EQ.INDEX.lwr EQ.INDEX.upr
1 100 1 40 2 0.3388744 0.1857931 0.4919557