iwCLL 2023 | The implementation of a data-driven CLL-TIM algorithm into an EPIC-based electronic health record

For the last ten years, we have seen implementation of electronic health record systems across the world. This means that we get access to more automatized extracted data. Unfortunately, we are still at the very first early steps of using those data to actually inform our daily clinical practice. If you imagine sitting in front of your patient in the outpatient clinic and looking through the medical record and your screen in front of you and trying to grasp the pattern of data, both including the medical history for the patient or prescription drugs for that patient, the laboratory results that usually would go decades back, and also the diagnosis and prior lines of treatment or risk of infections for this specific patient that would be recorded in terms of blood cultures drawn, the microbiology results, but also types of antibiotics and enzymatic profiles prescribed to that patient.

So to use these data we need pattern recognition or machine learning algorithms, essentially data-driven hematology. We need to use all this routine data available to us in the electronic health record and develop it into decision support tools. That’s what we have done as a proof of concept with the CLL treatment infection model. But still you have to enter 80 different variables, and a lot of them actually plot results as time series, took at least an hour per patient to enter these data, and we had the risk of typos. So that made us focus on implementing it as an automatized prediction that you just get as a decision support tool within your electronic health record system. And after a little more than one and a half years, we succeeded to do that, validating from our research data set all the variables that we needed in the electronic health record system, so remapping the research data into the electronic health record system, and after that, validating for more than a thousand patients that we got exactly the same prediction in the research data set as in the EHR system.

And that’s where we have now, in an EPIC-based Danish health record system covering half of the Danish population, implemented it and for all patients diagnosed with CLL, you will see that prediction given as a decision support tool. At the same time, we have implemented the algorithm, the CLL treatment infection model, into a clinical trial testing whether we can actually improve the immune dysfunction for the patients with a high risk of a serious infection by treating them with three months of acalabrutinib and venetoclax in combination, not to cure the disease, not to get the patients into very deep, undetectable MRD status, but to improve the immune function and reduce the risk of serious infections for these patients. So this is a proof of concept that we can actually implement data-driven hematology into commercial electronic health record systems, but on a standalone this is a Python based platform on a standalone platform where we as clinicians and researchers have full control of the algorithms.