Demystifying Analytics and Machine Learning in Clinical Trials: Key Takeaways
On October 16, Medidata hosted Demystifying Analytics and Machine Learning in Clinical Trials—a webinar featuring an expert panel that considered the flaws of traditional data management methods, how advanced analytics and machine learning initiatives can help improve or maintain data quality, and ways to realize ROI (return on investment) on such initiatives. Below are a few key takeaways from the webinar:
Good data quality begins with a good end-to-end data strategy
Data quality is important in all facets of our lives, ranging from international security to the safety and efficacy of our products. The ultimate key to better data quality is enabling better processes, technology,
and people. It’s important to consider an environment that supports multiple stakeholders with multiple perspectives that require timely, more efficient solutions. Master data management provides a reliable, single source of truth for all.
This approach to standardization must go far beyond just removing duplicates or mass data maintenance, incorporating rules that eliminate or prevent data incorrectness that find their way into our environment. Once you have this sort of master data management construct in place then it's much easier, more efficient, and more effective in terms of being able to make decisions based on what you thought and expected to be truly permissible. At Medidata, we first ingest, aggregate, standardize, and clean our customers’ data. Then, we tap into our MEDS (Medidata Enterprise Data Store) data repository—that single source of truth housing both clinical and operational data—to generate deeper insights into their clinical research, trials, sites, and patients.
“Find” vs “search”: the power of machine learning and AI
In a poll we ran during the webinar trying to understand which data errors were hardest to find, half of the attendees agreed that in a typical study, it’s most difficult to find inconsistencies across related data variables using traditional data cleaning methods—which are manually intensive and error-prone. The inconsistencies we don't find or anticipate create potential quality or compliance problems. In machine learning, we apply AI (artificial intelligence) to essentially teach a computer system to learn from inputs or experiences, and then generate outcomes. It can be either supervised or unsupervised machine learning.
In supervised machine learning, we are essentially providing somewhat large volumes of data to test the algorithm, giving an input and some type of outcome—so coding might be an example. In unsupervised learning, it might look like processing omic data to generate clusters or in data quality, like, with Medidata Centralized Statistical Analytics (CSA), making associations by looking at an entire database, in an unassisted way, identifying the relationship between two data points.
An ideal use case would be where you would go to a data review meeting and really be there to resolve and do root cause analysis on the issues that have already been identified by the machine learning to help us make ourselves better, cheaper, and ultimately faster. Advanced analytics coupled with machine learning can provide the best of both worlds: a holistic overview of all the data at a macro level, and the total view of every data point.
ROI from advanced analytics and machine learning
As an industry, we are notoriously guilty of applying new technologies & capabilities to the old way we work. Some of our customers are evolving beyond the traditional methods that aren't adding value; applying advanced analytics to create a well thought-out visualization of a patient’s data is a great example. There's a lot of application and room for improvement in using analytics at the patient level, as well with Big Data.
It’s really important that a company understands its tolerance for investment [in an analytics initiative], and what needs to happen sequentially in order to see ROI. Some customers use analytics to help them bridge gaps in processes, or as a more cost-effective way to achieve better data quality. Others, with infrastructure in place, might use analytics as extra assurance that things are working well. For larger companies with many ongoing trials in their portfolio, analytics is proving an effective way to boost their ROI. For smaller companies, the promise of future returns might be too large a gamble in the short-term. Mastering data standardization is a luxury they can worry about later.