Effective Exploratory Data Analysis

Motivation

“If you know the enemy and know yourself, you need not fear the result of a hundred battles.”

                                                      --- Sun Tzu, The Art of War

Understanding the data via exploratory data analysis (EDA) is arguably the most important aspect during model development, yet often overlooked. MVP provides some flexible tools to inspect and explore uploaded datasets quickly, with the goal to effortlessly:

• Perform summary checks on data on a population level
• Review and highlight profiles on an individual level

This vignette is meant to introduce a workflow of using MVP to perform quick exploratory data analysis, which all takes place in the “Data Input” page. The reader is recommended to start MVP on a separate browser screen and follow the instructions on this page.

Uploading and Filtering Data

1. The first step is upload a NONMEM-compatible (.csv) dataset, or a tab-delimited text file (.txt). For full compatibility for the rest of the app, the ID, DV, and TIME columns should be present in the dataset. At minimum, the ID column should be present to at least support plotting in the Data Exploration pages on the right hand side of the page.
   
2. There are built-in dataset cleaning options (options collapsed by default - click on the “+” icon to expand the box) that provides short-cuts to commonly used options to filter the dataset. For more custom filtering needs beyond the basics, the code editor on the left hand side of the page provides this flexibility (e.g. renaming Concentration and Time columns to be DV and TIME, respectively, if using non-NONMEM formatted datasets).

• Change from baselines (CFB) are automatically created using the DV column, per ID and CMT (DV_CFB and DV_CFBPCT for absolute and percentage CFB, respectively). If the EVID column is present, the first instance where EVID == 0 (i.e. observation row) will be taken as the baseline value per subject.

3. Click on the Apply button on the code editor when you are done with filtering via code - a dynamic table showing the filtered data should now show up on the right hand side of the screen (the “Filtered Data” tab) and able to be downloaded if required. If not, check the Console below the Code Editor for potential error messages.

💡 In general, the filters applied here should match with what eventually will be used in the NONMEM control stream files.

Summary Statistics

Quick summary statistics are now available on the “Summary Statistics” tab on the right hand side of the page, derived by using ID as a grouping variable. While summary of all variables (each column) will be provided, obviously not all are relevant or applicable, especially if a variable is a categorical (i.e. discrete) variable. However, this is a useful functionality to review summary statistics for continuous variables e.g. body weight or renal function.

💡 There is a “Transpose Table” checkbox that will further aid visualization of the results for datasets with a lot of columns.

Descriptive Statistics (NCA)

Note: NCA functionality is currently only applicable for single ascending dose (SAD) studies or its equivalent where each subject will only receive one dose in a single compartment. If that’s not the case, skip this section.

1. Column names for Time, Concentration, Dose will be pre-filled after a dataset is uploaded (user can adjust as they see fit).
   
2. Provide a “Group by Additional Column” variable to sort by the defined grouping in the table output.
   
3. (Optional) Adjust the units and provide a molecular weight (derives clearance and volume terms if present).
   
4. Click the “Calculate NCA” button to view results. The result table(s) can be transposed accordingly if required.

💡 Individual results and/or all profiles can be downloaded by clicking on their respective buttons on the bottom right of the page. For information and details on the various settings, visit the NonCompart and ncar packages guide.

General Plot

The General Plot provide a quick way to visualize multiple data types, including continuous/continuous data, continuous/discrete data (box plots), and discrete/discrete data (count plots).

Individual Plots

Most options in the “Ind. Plot” tab are carried over from the “General Plot” tab, with the exception of the “Facet by” option, since individual plots are always faceted by the ID column. Note that smoothers and median lines are not relevant for individual plots so they will also not be inserted.

1. By default, 9 subjects are displayed per page. Adjust the “Rows per Page” and “Columns per Page” accordingly to change the number of subjects per page. Note that the Page selector will be updated dynamically to reflect any changes in rows and columns.
   
2. (Optional) Dosing information is enabled by default, which searches for the EVID, ADDL, and II columns, and inserts dosing lines for each subject. The height of the dosing lines are scaled by their relative amount that is given by the “Dose Column”, and each unique dose will be labeled as text. If the RATE column is present, the dosing lines will be extended accordingly to the infusion duration. An optional unit for the dose can also be inserted. Dosing records (i.e. EVID == 1) must be present in the filtered data for this option to work!
3. (Optional) BLQ information can be inserted by providing the “LLOQ Column”, which will then be shown as horizontal dashed orange lines.
4. (Optional) Select any specific number of subjects to be viewed by using the “Select ID(s)” option. When any subjects are selected this way, the page, row and column options will be ignored.
5. (Optional) The “Flag Variable” option allows the user to select a variable, and any of its associated values, to be flagged as a red asterisk (*) on the plot, for further visualization beyond the “Color by” option.
6. (Optional) The “Flag Outliers” option allows the user to select a stratification group, and combined with the “Outlier Threshold” option, potential outliers will be automatically highlighted in their subplot ID label. The method of determining whether a subject is an outlier is done by deriving the arithmetic mean of the subject’s Y-axis values, and compare it to the entire group’s arithmetic mean. Only observation records are used (i.e. EVID == 0 when an EVID column is present), and missing values are ignored. If the subject’s mean value is above/below X% of the group, the subject will be considered an outlier.

💡 To visualize and/or highlight BLQ values, make sure the built-in filter option to remove BLQ values on the left hand side of the page is unchecked (“Built-in Dataset Cleaning Options” box - click on the “+” icon to expand the box)!

💡 All individual profiles can be downloaded by clicking on the “Generate All Individuals” button which is located directly below the plot. Plot dimensions can be adjusted by expanding the “Download Options” box on the bottom right side of the page.

Correlation Plots

Correlation between variables can be quickly assessed in this tab, using the GGally::ggpairs function (this requires the optional ‘GGally’ package: install.packages("GGally")). It is recommended to keep the number of variables <7 as the plot may become too bunched up.

💡 Any variables with more than <=10 unique values are assumed to be categorical and converted into a factor for plotting.

Covariate Histograms

Covariate distributions can be quickly evaluated in this tab to check for outliers. Note that columns of character type would be coerced into numerics to support plotting of histograms.

Summary

The Data Input page of MVP offers the ability to effortlessly review the data, assisting sanity checks and easily pinpoint outliers or potential errors. While this does not replace a more rigorous (i.e. scripted) EDA, the provided features serves as a great “first-pass” and should massively reduce turnaround time of queries back to data programmers.