Although clinical decision support systems using machine learning will likely add valuable improvements to healthcare processes in the next few years, developers must also consider their potential unintended consequences, three Italian researchers wrote in a JAMA op-ed.
Here are four potential unintended consequences of machine learning in medicine.
1. Reducing the skills of physicians. These tools may lead to a "deskilling" of physicians, should physicians begin to rely on automated clinical decision support systems too much. In the long term, the writers worry this phenomenon "may cause serious disruptions of performance or inefficiencies whenever technology fails or breaks down."
2. Focusing on discrete data, rather than overall context. Machine learning technologies often focus on analyzing information that can be rendered as discrete text and relies on this digital data painting an accurate representation of a medical problem. However, this information may result in a misleading interpretation of a patient's condition, should some clinical context not be represented.
3. Overlooking "intrinsic uncertainty" in medicine. Clinical decision support systems typically provide recommendations based on discrete categories. However, in everyday medical practice, individuals often disagree about diagnostic findings and outcome suggestions.
"Further research should be aimed at developing and validating machine learning algorithms that can adapt to input data reflecting the nature of medical information, rather than at imposing an idea of data accuracy and completeness," according to the writers.
4. Using the "black box" without addressing shortcomings. Machine learning algorithms are often referred to as "black box models," according to the writers, because even developers don't typically know the inner workings of how the algorithms reach their decisions. Without this understanding, subtle shortcomings of clinical decision support systems "may be difficult or impossible to prevent or detect."
Click here to view the full article.