Backdrop has created a novel software platform that gets the maximum predictive power from any set of electronic medical records. Backdrop aims to enable a new level of precision predictive medicine.
Address: 1648 E Federal Heights Drive
Salt Lake City, UT 84103
URL: backdrophealth.com
Modern electronic medical records (EMRs) contain a wealth of information about patients. Backdrop Health has developed machine learning and AI software that provides a wholly new way to understand any set of EMRs and make medical outcome predictions of unprecedented precision. Backdrop can deliver $ billions of value to hospitals and primary care providers, pharmaceutical companies doing clinical trials, and insurers. Backdrop’s initial target market is the thousands of Phase II, III, and IV drug trials started annually in the US alone. Backdrop enables drug trial teams to optimize trial inclusion and exclusion criteria, more easily create ‘synthetic control arms’ for real-world evidence-based trials, see if adverse effects are due to the drug or ‘medical destiny’, and more, accelerating trials and cutting their cost. Backdrop builds on a key idea in medical outcome predictions: comorbidity, or the existence of two or more medical conditions (such as diabetes and obesity) in a patient.Backdrop radically extends the idea of comorbidity to include all clinical factors: diagnoses, procedures, lab tests, and medications. Given any set of EMRs to which a customer has access, Backdrop’s software creates a comprehensive knowledge graph – the backdrop -- for those medical records. While classic comorbidities reveal part of the picture, the backdrop reveals everything. Backdrop’s software also enables fully-explainable analyses and predictions about any outcome of interest, using multiple clinical and demographic factors. Backdrop is bringing this mature technology developed over five years at The University of Utah to market under license from The U.
Technology Overview A key idea in medical outcome predictions is comorbidity: the coexistence of two or more medical conditions in a patient, such as diabetes and obesity. Comorbidities are powerful and widely used, but there are limitations in the state of the art. First, comorbidities have long been calculated based on finding relationships only between pairs of diagnosed conditions. Little of the full array of available clinical information is used in comorbidity discovery. Second, accurate comorbidity calculations often require controlling for confounding variables (like age or sex) by stratifying: reducing the data set to ensure uniformity of risk. But this results in diminished statistical power. Backdrop’s breakthrough approach is implemented in our core software called CoDE: Comorbidity Discovery Engine. CoDE takes any body of medical records which a customer has developed or licensed, ranging from sets of millions of patient records extracted from an enterprise data warehouse or medical records system to just hundreds or thousands of medical records captured in a spreadsheet. Given such data, CoDE automatically: • Finds and quantifies all comorbid relationships between all clinical variables, at massive scale; • Discovers the temporal relationships between those identified comorbid factors; • Identifies which clinical variables and demographic variables matter for any outcome of interest; • Models the influence of every patient’s individual medical history on any outcome; • Removes the need to stratify to control for confounding variables, amplifying the power of any medical data set – often by orders of magnitude; CoDE creates a rich and comprehensive knowledge graph – the backdrop -- for any body of medical data. While classic comorbidities reveal part of the overall picture, the backdrop reveals every clinical variable’s relationships to every other. This in turn enables powerful and flexible reports and queries about any outcome of interest, and about any clinical or demographic variables that contribute to that outcome. Useful predictions often require knowing the combined effects of many comorbid factors, not just a pairwise analysis. Backdrop’s unique and powerful Multimorbidity Query Engine allows predictions on multiple factors, using the AI technique of Bayesian Networks to account for all conditional dependencies among those factors. And since the contribution of each factor to any predicted outcome is clear, all predictions are fully explainable.