When performance is not enough-A multidisciplinary view on clinical decision support

Abstract

Scientific publications about the application of machine learning models in healthcare often focus on improving performance metrics. However, beyond often short-lived improvements, many additional aspects need to be taken into consideration to make sustainable progress. What does it take to implement a clinical decision support system, what makes it usable for the domain experts, and what brings it eventually into practical usage? So far, there has been little research to answer these questions. This work presents a multidisciplinary view of machine learning in medical decision support systems and covers information technology, medical, as well as ethical aspects. The target audience is computer scientists, who plan to do research in a clinical context. The paper starts from a relatively straightforward risk prediction system in the subspecialty nephrology that was evaluated on historic patient data both intrinsically and based on a reader study with medical doctors. Although the results were quite promising, the focus of this article is not on the model itself or potential performance improvements. Instead, we want to let other researchers participate in the lessons we have learned and the insights we have gained when implementing and evaluating our system in a clinical setting within a highly interdisciplinary pilot project in the cooperation of computer scientists, medical doctors, ethicists, and legal experts.

Fig 1. Overview of the data flow from the patient database to the model.

The last part about the patient-level data split depends on the evaluation. In the case of internal evaluation, the data split represents the 10-fold cross-validation. In the case of the reader study, the split has been applied just once.

Fig 2. Overview of the dashboard which shows the recent risk score (red arrow), together with the previous risk scores of the patient over time depicted in a graph.

Moreover, the dashboard shows the most important local features, which had a direct influence on the current scoring of the model, as well as the most relevant global features, which had generally a strong influence on the model.

Fig 3. Overview of the two parts of the reader study.

On the left, the setup with medical doctors (MD) alone. The MD has up to 30 minutes to 1) study the patient at a given date in his/her life and then 2) make a risk estimation for infection for the next 90 days. The right-hand side depicts the setup in which 1) the MD also receives the patient data and the risk estimation of the decision support system. After analyzing both for up to 30 minutes, the physician 2) makes a risk estimation.