ttt-p1-main-analysis

This repository contains code for centralized data cleaning and the main paper for Phase I of Project TRACK to TREAT. The repo is linked to this project on the Open Science Framework (OSF): https://osf.io/c4e75/.

Table of Contents

• Project Overview
• Centralized Data Cleaning
  • Data
  • Code
  • Other Documentation
  • TODOs
• Network Analyses
• Machine Learning

Project Overview

Phase I of Project TRACK to TREAT (TTT) aims to use parameters from network models estimated from ecological momentary assessment (EMA) data to predict 3-month changes in depression symptoms and related constructs in depressed adolescents. Phase I, an observational study, consisted of a baseline Qualtrics survey, 21 days of EMA (5 pings per day), and another Qualtrics survey 3 months later.

Centralized Data Cleaning

Data, initial code, and documentation relevant to centralized data cleaning for Phase I of TTT are stored in the MSS/Schleider_Lab/jslab/TRACK to TREAT folder on the FSMResFiles server at Northwestern University Feinberg School of Medicine.

The initial centralized data cleaning code was drafted by Michael Mullarkey. The Centralized Data Cleaning section of the present repo houses Jeremy Eberle and Isaac Ahuvia's attempts to improve the code for greater reproducibility. For centralized data cleaning for Phase II of TTT, see the separate repo ttt-p2-cleaning.

Lab staff who contributed to Phase I of TTT include former research coordinators Sharon Leong (formerly Chen) and Akash Shroff, and Laura Jans (research assistant at the time).

Data

Data collection is complete.

Raw

From Qualtrics

Screening Survey

TODO

Study Surveys

Raw baseline and 3-month survey data are stored in the /TRACK to TREAT/Data/Qualtrics Data/Raw Data folder, which contains 18 CSV files obtained from Qualtrics (per Date Modified file metadata, presumably 6 files were obtained 6/16/20-5/20/21 and 12 files were obtained on 6/18/21). The Qualtrics cleaning script appears to focus on the latter 12 files (see below).

From LifePak

Raw EMA data are stored in the /TRACK to TREAT/Data/LifePak Raw Data (Do Not Modify) folder, which contains 10 CSV files obtained from LifePak (per Date Modified file metadata, presumably 8 files were obtained on 4/28/20 and 2 files were obtained on 9/28/21). The LifePak cleaning script appears to focus on the 5 files with NIS in the filename (see below); the /TRACK to TREAT/Data/readme_ttt_p1.docx file also states that files with NIS (which it defines as "notification-initiated survey") in the filename are the data to be used.

Clean

Outputs of the initial data cleaning code are stored in the /TRACK to TREAT/Data/Processed Data/2022 From Michael Mullarkey folder

Code

The present repo uses the following scripts from the /TRACK to TREAT/Code/Data Cleaning/old/2022 From Michael Mullarkey folder as a starting point for centralized data cleaning. Given that in general the Qualtrics data seem to have been cleaned before the LifePak data (after which each dataset was deidentified), in this repo the scripts have been numbered in the order to be run.

01_ttt_phase1_qualtrics_cleaning.Rmd

Inputs the following 12 raw CSV files (out of the 18 from Qualtrics)

# "dp5_b_child_p1_numeric.csv"
# "dp5_b_child_p1_choice_text.csv"
# "dp5_b_child_remote_p1_numeric.csv"
# "dp5_b_child_remote_p1_choice_text.csv"
# "dp5_b_parent_p1_numeric.csv"
# "dp5_b_parent_p1_choice_text.csv"
# "dp5_b_parent_remote_p1_numeric.csv"
# "dp5_b_parent_remote_p1_choice_text.csv"
# "dp5_3m_child_p1_numeric.csv"
# "dp5_3m_child_p1_choice_text.csv"
# "dp5_3m_parent_p1_numeric.csv"
# "dp5_3m_parent_p1_choice_text.csv"

Also inputs dp5_p1_scoring.csv

• This file, in /TRACK to TREAT/Code/Data Cleaning/old/2022 From Michael Mullarkey, was obtained by Jeremy Eberle from Michael on 10/31/23. Michael stated that he obtained the file from a Google Drive folder owned by Sharon Chen.

Outputs (though both are commented out) yb_lsmh_ids_dates.csv and cleaned_qualtrics_ttt_phase_1.csv. Moreover, outputs cleaned_qualtrics_ttt_phase_1_fixed_220604.csv, but this does not appear to be used later in data cleaning pipeline (seems later scripts just input cleaned_qualtrics_ttt_phase_1.csv)

• Isaac Ahuvia stated that he revised the cleaning script in May 2022 just to keep a variable that had been deleted or something similar, so cleaned_qualtrics_ttt_phase_1_fixed_220604.csv may relate to this. Both this script and a separate script ttt_phase1_qualtrics_cleaning_fix.Rmd (which is not on the present repo but in /TRACK to TREAT/Code/Data Cleaning/old/2022.06 From Isaac Ahuvia and which outputs a CSV file with a different date, cleaned_qualtrics_ttt_phase_1_fixed_220606.csv) were last modified on the same date (Date Modified metadata of 6/6/22).
• Note: 02_ttt_phase1_lifepak_cleaning.Rmd below inputs cleaned_qualtrics_ttt_phase_1.csv and then overwrites it after correcting some participant IDs

02_ttt_phase1_lifepak_cleaning.Rmd

Inputs the following 5 raw CSV files (out of the 10 from LifePak)

# "3T_P1_V1_NIS_2020_Mar_02.csv"
# "3T_P1_V2_NIS_2020_Mar_13.csv"
# "3T_P1_V2_NIS_21200_958251_Download2.csv"
# "3T_P1_V2_NIS_21200_958251_Download3.csv"
# "3T_P1_V4_NIS.csv"

Also inputs cleaned_qualtrics_ttt_phase_1.csv (presumably originally from 01_ttt_phase1_qualtrics_cleaning.Rmd)

Outputs cleaned_combined_qualtrics_lifepak_ttt_phase_1.csv (does not appear to be used later in data cleaning pipeline), cleaned_lifepak_ttt_phase_1.csv, and cleaned_qualtrics_ttt_phase_1.csv

03_deid_ttt_phase_1.Rmd

Inputs cleaned LifePak data (cleaned_lifepak_ttt_phase_1.csv) and cleaned Qualtrics data (cleaned_qualtrics_ttt_phase_1.csv)

Outputs deidentified data (deid_cleaned_lifepak_ttt_phase_1.csv and deid_cleaned_qualtrics_ttt_phase_1.csv)

Other Documentation

The following files in the MSS/Schleider_Lab/jslab/TRACK to TREAT folder appear relevant to data cleaning

General

• TRACK to TREAT/Data/readme_ttt_p1.docx
• TRACK to TREAT/Data/Processed Data/2022 From Michael Mullarkey/README.rtf
• TRACK to TREAT/Code/Data Cleaning/README_ttt_p1_data_cleaning.docx
  • Points to present repo as most recent data cleaning effort

LifePak

• TRACK to TREAT/Data/3TP1_LifePak_Version_IDs.xlsx
• TRACK to TREAT/Data/README info from Laura Jans folder
  • See contents of this folder for info from Laura Jans re (a) 7 participants who have LifePak data for fewer than the expected number of beeps (see yellow highlights in 2024.04.03 Email with Laura Jans re EMA slider and missing EMA data.pdf) and (b) whether EMA slider items could be skipped (see orange highlights).
  • For the main TTT paper, these data are treated as missing (see 02_networks/code/02_further_clean_data_align_obs.R in the present repo)

TODOs

General

• TODO: As of 12/3/24, Jeremy can reproduce cleaned_lifepak_ttt_phase_1.csv (and deid_cleaned_lifepak_ttt_phase_1.csv) per identical(x, y, FALSE, FALSE, FALSE, FALSE). However, he cannot reproduce cleaned_qualtrics_ttt_phase_1.csv.
  • Specifically, he can reproduce the clean LifePak data when using R 4.1.1 (latest version available on 9/28/21; see below) and the most recent versions of tidyverse, skimr, glue, and janitor available on 12/3/24 (loaded via library()). He tried to use the groundhog package to load the latest available package versions on 1/7/22 (date that output files were saved to server; see below) but could not use groundhog to load tidyverse as tidyverse depends on knitr, which is "already in use" as it is used to execute Rmd files.

Specific

• TODO: Determine what R version and package versions should be used for each script
  • 01_ttt_phase1_qualtrics_cleaning.Rmd lists 6/17/2021 as the Date; 02_ttt_phase1_lifepak_cleaning.Rmd lists 9/28/2021 as the Date. The output files cleaned_qualtrics_ttt_phase_1.csv and cleaned_lifepak_ttt_phase_1.csv have Date Modified metadata of 1/7/22. Thus, the scripts used R and package versions prior to these dates.
  • Note: Michael stated that he cannot guarantee he always used the most up-to-date packages, but he endorsed using these dates as a starting point for determining which R and package versions he used
  • Also find way to control version of knitr (needed for Rmd files) or do not use Rmd files (see issue above)
• TODO: Determine what packages are needed, load only those, and load all needed packages at top of script
  • Although many packages are loaded, only a few appear used by each script (see lists below)
  • Moreover, one of the loaded packages (doMC, for parallelization) is Unix only and unavailable for Windows
  • Some packages (datapasta, fuzzyjoin) are loaded partway through script rather than at top

# "01_ttt_phase1_qualtrics_cleaning.Rmd" packages: "tidyverse", "glue", "janitor", "fastDummies", "diffdf", "datapasta", "fuzzyjoin"
# "02_ttt_phase1_lifepak_cleaning.Rmd" packages:   "tidyverse", "skimr", "glue", "janitor"

• TODO: Remove parallelization, as it does not seem needed
• TODO: Treating . as the parent folder for the present repo, create local raw data folders (./data/raw/qualtrics and ./data/raw/lifepak), put raw CSV files in those folders, and use relative file paths to load raw data and output clean data (vs. storing raw data, clean data, and code in same folder)--see below for example. We can then describe the directory structure in this README.

example_raw_table <- read.csv("./data/raw/qualtrics/example_raw_table.csv")

clean_path <- "./data/clean/"
dir.create(clean_path)
write.csv(example_clean_table, paste0(clean_path, "example_clean_table.csv"))

• TODO: Load dp5_p1_scoring.csv at top of 01_ttt_phase1_qualtrics_cleaning.Rmd (vs. partway through script)
• TODO: Clearly reflect what 01_ttt_phase1_qualtrics_cleaning.Rmd should output (see description of its outputs above for various issues)
• TODO: Remove extraneous code/comments
• TODO: Avoid hard-coding (e.g., use specific LifePak filenames rather than the code below in Lines 103-109 of 02_ttt_phase1_lifepak_cleaning)

files <- list.files(pattern = "*.csv")
lifepak_files <- files[c(1:5)]

• TODO: Consider checking for data quality (e.g., see Exclusion Criteria from Phase II study registration)
  • Obvious lack of English fluency in open response questions
  • Responding with random text in open response questions
  • Duplicate responses from the same individual in baseline or follow-up surveys
• TODO: Put deidentified clean data in ./data/clean folder on OSF project linked to the present repo

Network Analyses

Data

Analyses input cleaned and deidentified LifePak data in deid_cleaned_lifepak_ttt_phase_1.csv

Code

Initial network analyses were run by Sebastian Castro-Alvarez and Laura Bringmann using code in TODO. Analyses were revised by Josip Razum and Jeremy Eberle.

TODO: Jeremy to document scripts

Machine Learning

TODO: Jeremy to update this section

Data

Analyses input cleaned (but not deidentified) LifePak data in cleaned_lifepak_ttt_phase_1.csv and cleaned (but not deidentified) Qualtrics data in cleaned_qualtrics_ttt_phase_1.csv.

• TODO: Consider revising code to import deidentified data instead)

Code

Initial machine learning analyses were run by Michael Mullarkey using code located at TODO. Analyses were revised by Yama Chang.

Michael's

TODO: Compare code on server with code on Michael's ttt-main-analyses repo to help determine if code on server is Michael's original code

• See R:\MSS\Schleider_Lab\jslab\TRACK to TREAT\Code\Primary Analyses

Yama's

TODO: Process and describe Yama's latest code in folder below

• See R:\MSS\Schleider_Lab\jslab\TRACK to TREAT\Code\Primary Analyses\2024.08.01 From Yama Chang