Contributed Commentary by Michael Phillips

September 13, 2019 | The essential goal of a clinical trial is to collect a dataset of sufficient quality to statistically test the hypothesis stated in the trial protocol. A lot of data are collected, including repeated observations on each patient, laboratory tests and other technical measurements, clinical outcome assessment questionnaires, and even data from devices used by the patients in their own homes. At the same time, there are a lot of procedural requirements, not least keeping the trial participants safe.

All of this has to be monitored and managed. Not so long ago, clinical data analysis was confined to two main activities: checking data entry at the individual trial sites and programming systematic checks against the collected data. These tasks remain important, but advances in electronic data capture, integration, analytics, and visualization have introduced new ways to monitor the conduct of the trial in more holistic and insightful ways.

Central Monitoring

The current guidelines for good clinical practice place a strong emphasis on integrated quality and risk management because this is the best way to ensure that everyone involved in the running of a clinical trial is focused on the areas that matter most. Checking the transcription of patient data into electronic data capture systems and programming edit checks are not the best way to find underlying trends in the trial data; these activities will not reveal much about potential or emerging risks. You need to engage central monitors, whether they be specialized data analysts looking for risk signals, or physicians evaluating safety and other clinical trends.

Central monitors need information; they need data analysis tools. They are empowered by technology that allows them to interactively interrogate integrated datasets, cross reference, pivot, filter and sort. It helps if they have online access to the data, automated data refreshes, and some degree of self-service configuration of the analyses. Their work is characterized most by the use of data visualization not only to find signals and trends but to communicate their findings for decision making. A picture paints a thousand words, and a good data visualization brings to life countless bytes of data processing.

Data Listings Versus Visualization

What can we do with the raw data extracted from clinical trial data capture systems? We can program a data listing for review and perhaps filter and sort in a spreadsheet. What if we want to evaluate an adverse event distribution, quickly select a trending adverse event based on frequency and severity and drill into the medical history, lab results, and vital signs of the patients participating. What if we want to see those lab results or blood pressure measurements plotted as time courses so we can visually inspect the trend? We can’t expect the central analyst to build all this functionality from scratch; they need prebuilt visualizations that allow them to move from high-level views to details, across data domains, to pick out the key signals and share them with others in the study team.

It is possible to define key risk indicators, such as the rate of adverse event reporting, and provide the central monitor with a visualization that highlights sites that deviate from the rate that is associated with the trial as a whole.  It is also possible to visualize how consistently related clinical outcomes measures do in fact correlate, thereby detecting potential lapses in clinical methodology that could threaten the integrity of the trial’s primary endpoints.

In a single heat map for the whole study, plotting each patient’s compliance with a daily procedure, such as a pain diary or wearing a device, coloring by compliance rules and using dendrograms and filters provides the ability to find patterns and quickly interrogate and drill into details.

For oncology trials, plotting the change in tumor size over time, colored by tumor assessment—the so-called spider plot—reveals clear patterns to medical reviewers that cannot be seen in a data listing. A sorted bar chart of best overall tumor response by patient—the “waterfall plot”—provides an instant sense of the overall promise to a particular combination of drugs or drug doses in a multi-cohort trial, helping quicker and more insightful decision making to ensure better safety and efficacy for patients.

Central monitors like details, so listings are essential. Simple filtering and sorting greatly aid data interrogation. However, well-constructed, interactive visualizations that collate and relate data across multiple domains of clinical data reveal so much more and allow the central monitor to process significantly greater volumes of data with a low risk of missing important signals.

The dynamic nature of data visualization tools has changed the approach to surveillance. Expert monitors are now getting involved in a discussion that inspires a scientific line of inquiry and produces quick answers. They can challenge the data, look at combinations of parameters and search for trends in a way that they could never do with data listings. Interactive data visualizations match the way subject matter experts think about data.

Michael Phillips, Director Clinical Risk Management, ICON plc, has worked in IT and business intelligence for over 17 years. He has a strong mix of scientific, business, and technical experience. After gaining a PhD in biochemistry, Michael worked as a general science and medicine editor for 10 years before moving into IT and subsequently into clinical risk management. For the past 8 years, he has been designing and building clinical informatics solutions. His team at ICON develops and maintains risk-based monitoring and medical data review solutions to support the monitoring of clinical trials. He can be reached at Michael.Phillips@iconplc.com