The Role of Quality Data in Utilizing Machine Learning Models

In a recent video interview with Applied Clinical Trials, Cameron Breze, product manager, Inovalon, discussed challenges in patient recruitment for clinical trials. Traditional recruitment methods, such as manual outreach, are now less efficient due to accessibility issues. Breze highlighted technology as a way to address recruitment challenges, which can be leveraged to automate and streamline processes, reducing labor intensity and errors.

ACT: With so many data sources now in the mix in clinical research, how important is the role of quality data in identifying patient populations?

Breze: Quality data is something that is top of mind for everyone in the in the data technology space. Many people like to use the age-old adage of garbage in garbage out when it comes to machine learning models and a lot of the data modeling that's available where if you're feeding it with inaccurate information, you're going to get a result that really can't be transposed and moved and translated into anything that's really applicable and that ends up being a waste of institutional resources and not the best way to serve patients. I can share a little anecdote as well on that. I published a research paper with the MIMIC for critical care database, where we looked at resuscitation, so fluid and IV administration in the hospital with organ failure, and found out through lots of computational modeling that essentially, we had a great record of the fluids that were put in, but not a great record of the fluids put out, so the clinical staff were not recording that, not because they were missing it or they were making some error, but in the scheme of making that particular research impactful, if you don't have that information, you can't make designations on that person's fluid status, and seeing that in aggregate across hundreds of thousands of patients, you get to see what a potential impact that could have had, had that policy or procedure been updated or changed, so I think a lot of those decisions help inform the way that site staff are trained and data is collected over time. Beyond the basics of setting up a system that is going to collect quality data and allow the researchers to interact with it, there's a lot of data vetting that has to go through, so relying on subject matter experts and folks with experience in those particular areas, it's an interesting intersection, where sometimes the person who may be best able to speak to the quality of the data might come from a tech background or a healthcare background where you wouldn't expect that to be the case, typically, so combining interests and having that overlap and Venn diagram and cross pollination of ideas is something that inspires me to work every day and have that overlap and really collaborative effort with a lot of people that I work with and a lot of my other peers in the industry.