I’ve recently looked a bit at clinical decision support systems. These are software systems that assist physicians and health professionals in decision making tasks. Automated diagnosis based on patient’s data is the most prominent example what such systems can do. It turns out that the idea isn’t particularly new – first systems were build already in 70s. However, when you look at the history of CDSS, or more specifically their diagnostics subset, it’s clear that most of them were discontinued quite early or they weren’t really deployed in clinical setup.
Often mentioned challenge to adoption of CDSS is so-called alert fatigue. Pulling in large number of information, tests, correlations in genetic data, drug effects and interactions and what not results in creating large number of false alerts, that quite often aren’t relevant in particular situation. Physicians warned all the time about possible and even reasonable (everybody has non-zero chance of dying of cancer, but these systems do not suggest cancer in every case) effects or diseases stop paying attention to the alerts after some time. On the other hand diagnostics errors alone are believed to cause in US between 50.000 and 100.000 deaths per year (some studies suggest that ca. 75% of them were preventable) – definitely there’s a room for improvement.
I see that as a opportunity for bioinformatics and healthcare to learn a bit from each-other. Specialists working on diagnosis decision support systems seem to say out loud: “it’s way too complex”. Bioinformaticians working on data-intensive areas such as genome-wide association studies seem to say out loud: “it’s not that simple”. Hopefully they are going to meet half way.