Domain specific composite scores for cognitive impairment

This repository contains:

• Code to calculate the domain composite scores in new data, assign a MCI subgroup and predict progression from MCI to dementia within different time windows.
• Code to reproduce the analysis and results presented in the manuscript "Quantification of cognitive impairment to characterize heterogeneity of patients at risk of developing Alzheimer’s disease dementia"

Built With

• R (version 3.6.3)
• Neuropsychological data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database. See http://adni.loni.usc.edu/

Calculating scores for new data

Pre-requisites

• Sub-scores from neuropsychological tests matching the ADNI convention for naming variables. You can input your data to the example/example_data.csv file.
• dplyr R library should be installed to process data and calculate scores.
• If you want to make predictions about progression from MCI to dementia, you will need the randomForest library and the list of pre-trained classifiers in example/pretrained_RF_MCIprogression_prediction.RData.

Run example

In the R console

1. Set the working directory in the repository folder

setwd("~/neuropsycho_adni")

2. Load the required functions and lists of neuropsychological tests

source("scripts/example_functions.R")
load("example/tests_lists.RData")

sel_tests <- c("ADAS", "MMSE", "MOCA", "CLOCK", "TMT", "LM", "CATFL", "AVLT", "BNTMINT")
it <- unname(unlist(selitems[sel_tests]))

3. Read the example data and reverse sub-scores that should be reversed

A <- read.csv("example/example_data.csv")
A <- reverse_scores(A, selitems)

4. Calculate the Standardized Regression Based (SRB) z-scores using the parameters estimated from a sample of cognitively unimpaired participants in ADNI

srbcoef <- read.csv("results/srb_parameters.csv")
A_srb <- calculate_srbz(A, it, srbcoef)

5. Calculate the domain-specific composite scores using the learnt parameters

load("results/cfa_estimates.RData")
namesfac <- colnames(LW$L)
S <- composite_scores(newdata = A_srb, weights = LW$W, centerit = meansub)
S <- cbind(dplyr::select(A_srb, -all_of(it)), S)

6. Assign one MCI subgroup to each subject in example_data. This is done by calculating the distance to each subgroup's representant (medoids) and selecting the closest one.

grmeds <- read.csv("results/medoids_domainscores_MCIsubgroups_k4.csv")
MCIGR <- distance2meds(S, LW$Cz, grmeds, namesfac)
S <- merge(S, select(MCIGR, ID, GR))

7. Predict progression from MCI to dementia within the next 1 to 5 years

library(randomForest)
load("example/pretrained_RF_MCIprogression_prediction.RData")
PRED <- predict_MCIprogression(S, RFlist, namesfac)
S <- merge(S, select(PRED, ID, ends_with("prediction")))

Reproducing results and plots in the manuscript

Pre-requisites

To fully reproduce all the analyses you will need the following R libraries:

library(ADNIMERGE)
library(dplyr)
library(reshape2)
library(ggplot2)
library(tidyr)
library(lavaan)
library(NbClust)
library(cluster)
library(survival)
library(survminer)
library(randomForest)
library(ROCR)

Folder organization

• Scripts for each of the following steps are in the folder scripts.
• Processed ADNI data for each step is saved in a folder named processed_data, not included in this public repository but can be generated if the user has access to the ADNIMERGE R package.
• Estimated parameters and results are saved in the results folder.
• Graphics in EPS and PNG are saved in plots.

Steps

1. Pre-process neuropsychological data in ADNIMERGE package with preprocess_neuropsychological_adni_data.R
2. Reverse some of the neuropsychological sub-scores with reverse_scores_neuropsychological_adni_data.R
3. Pre-process diagnostic data in ADNIMERGE package with preprocess_diagnostic_adni_data.R
4. Select first visit with complete information and split data for analyses with select_and_split_neuropsychological_adni_data.R
5. Calculate Standardized Regression Based (SRB) z-scores with srb_zscores_neuropsychological_adni_data.R
6. Perform Confirmatory Factor analysis and calculate domain specific scores with cfa_with_srb_neuropsychological_adni_data.R
7. Cluster analysis using domain scores with mci_clustering_domain_scores.R, it includes survival analysis to compare progression risk between MCI sub-groups.
8. Automated prediction of progresion from MCI to AD dementia with mci_progression_predicton_domain_scores.R.

License

Distributed under the MIT License. See LICENSE for more information.

Contact

Diana L. Giraldo - @diagiraldo - [email protected]