Inspecting a Machine Learning Based Clinical Risk Calculator: A Practical Perspective

Abstract

Health is reaching a point where machines are more accurate than humans, or at least as accurate but with less effort, in more and more applications. However, accuracy alone is not enough, explanation and understanding is equally important to clinicians, governments, and patients. Possibly leading to loss of health benefits potentially realized through increasingly accurate algorithms. However, various techniques exist for auditing machine learning systems via insightful visualisations. Modelling best practices, parallel computations and open source technologies facilitate implementation of these techniques. This paper leverages several of these methods to increase interpretability for a black-box clinical risk calculator, hopefully opening the door to a better adoption of modern machine learning pipelines in the healthcare sector.