Make clinical data reviews more efficient with the interactive visualisation of Patient Profiles

From context to implementation of idea 

An important step in any clinical trial is the blind data review. This process takes place between the end of the study (when the data are collected, input, and validated) and the breaking of the blind and final data lock, after which the statistical analysis takes place. The goal is to review all the data and study events that may have an impact on the results, to make decisions on the validity of the data and the subjects to exclude from certain study populations. The review is blinded to intervention group because blind decisions introduce less potential for bias.

To optimise this process, Biofortis has developed an interactive data visualisation tool called the Patient Profiles. Using R Markdown, individual profiles are created for each subject in the clinical trial, all centralised within a single HTML interface. These profiles contain all the information and results pertaining to the subject in question, from data at inclusion (sex, age, subject characteristics, etc.), to study events (visits, adverse events, concomitant medications, protocol deviations, etc.), and data collected over the course of the study (anthropometric data, biological data, questionnaires, etc.). 

This makes it possible to detect incoherent values, investigate missing data and visualise events that are potentially linked to one another, through individual and overall charts.

An interactive and customisable data visualisation tool

Traditionally, the Patient Profiles are a simple flat file, containing all the individual information on the trial subjects. While this format contains a lot of useful information, it is not easy to navigate during data review meetings, and lacks the interactivity required to intersect data. 

This is why an interactive version of the Patient Profiles was created: the HTML files are viewable with any web browser, and allow to switch easily between data, subjects, charts, and categories. 

The Patient Profiles are split into three sections: the main section (“Information”), the population section (“Data Overview”), and the individual profiles, which are available by clicking on the subject numbers. All the information contained in the Patient Profiles is fully customisable to fit clients’ needs. 

Information

The information section is the homepage from which the individual profiles and the data overview section are available. It also contains additional information on acronyms and data sources. 

Individual profiles

The individual profiles are split in two parts. On the left-hand-side, information on subjects is listed and stored by category, in separate tabs: baseline general information (sex, age, site, smoking and dietary habits, etc.), medical history, compliance, protocol deviations, physical examination.

On the right-hand-side, each tab contains one chart: the first chart is titled Event(s), and allows to visualise subjects’ entire participation in the study, from inclusion to end of study, and includes visits, adverse events, and concomitant medications, ordered chronologically. This makes it very easy to visually associate all these items at a glance, and thus to understand their impact on the study endpoints. 

The other tabs contain figures of endpoints over time, at each visit, for the subject as well as population values: population mean, first and third quartile, and minimum and maximum. This makes it easy to detect individual outliers visually. 

In the case of kinetic measures, such as Oral Glucose Tolerance Tests (OGTT), the figure superimposes the curves for each visit, to view the evolution over time, and potentially detect atypical profiles. 

Hovering the mouse above the chart displays the values at each visit. All the figures are zoomable, downloadable and clickable: you can choose to display or hide certain curves (minimum and maximum for instance, or a certain visit in the case of kinetic curves).

Population section: “Data Overview” 

The population section contains charts for all included subjects. Most endpoints are charted using boxplots, with individual data points for each subject. This allows to view the evolution of the population distribution over time, at each visit, and flag a potential subject of interest (identified by its subject number) within the population by hovering the mouse. You can also click through to directly access the subject’s individual page. 

This interface makes it easy to detect potential outlier values.

Development using R Markdown

The Patient Profiles are coded in R language and executed using R Markdown. R Markdown is an R extension that allows to generate reproducible documents through an automatic process. It generates HTML files which contain the results (tables and graphs) for each subject, grouped and accessible via one single interface, and which can be viewed from any browser by the people involved in the data review. The tool uses several packages such as markdown, haven, dplyr, and shiny. R Markdown documents can include text, statistical results (means, percentages, etc.), code, as well as charts.

The tool has been steadily improved since its first version. For instance, we recently developed an extension for studies with microbiome analyses. New tabs have been added to visualise shotgun metagenomics. Thereby, the Data Overview section contains boxplots of alpha-diversity indices by visit and stacked barplots of taxonomic relative abundance by visit. The individual profiles contain subject-level alpha-diversity indices at each visit, as well as the population 5th and 95th percentiles in order to potentially detect outliers. Taxonomic relative abundance is also represented with stacked barplots by visit with population means and individual data. As with the clinical data, these figures are zoomable, downloadable and clickable: you can choose to display or hide certain taxa. 

Our Patient Profiles were presented through a poster at EPICLIN 2022.

Conclusion

The interactive Patient Profiles generated in HTML help you better visualise the data from your study subjects. You can easily navigate through subject information to ensure data integrity by examining the nonsensical data, extreme values and deviations and visualise all events concurrently, to best prepare the data review meeting.

Tutorial video for the Patient Profiles: