Modern Learning from Big Data in Critical Care: Primum Non Nocere

Abstract

Large and complex data sets are increasingly available for research in critical care. To analyze these data, researchers use techniques commonly referred to as statistical learning or machine learning (ML). The latter is known for large successes in the field of diagnostics, for example, by identification of radiological anomalies. In other research areas, such as clustering and prediction studies, there is more discussion regarding the benefit and efficiency of ML techniques compared with statistical learning. In this viewpoint, we aim to explain commonly used statistical learning and ML techniques and provide guidance for responsible use in the case of clustering and prediction questions in critical care. Clustering studies have been increasingly popular in critical care research, aiming to inform how patients can be characterized, classified, or treated differently. An important challenge for clustering studies is to ensure and assess generalizability. This limits the application of findings in these studies toward individual patients. In the case of predictive questions, there is much discussion as to what algorithm should be used to most accurately predict outcome. Aspects that determine usefulness of ML, compared with statistical techniques, include the volume of the data, the dimensionality of the preferred model, and the extent of missing data. There are areas in which modern ML methods may be preferred. However, efforts should be made to implement statistical frameworks (e.g., for dealing with missing data or measurement error, both omnipresent in clinical data) in ML methods. To conclude, there are important opportunities but also pitfalls to consider when performing clustering or predictive studies with ML techniques. We advocate careful valuation of new data-driven findings. More interaction is needed between the engineer mindset of experts in ML methods, the insight in bias of epidemiologists, and the probabilistic thinking of statisticians to extract as much information and knowledge from data as possible, while avoiding harm.

Prognostication in Neurocritical Care

“…Some experts in statistical modeling have suggested that machine learning has no advantage over standard regression modeling 54 . Future research will be necessary to determine whether machine learning techniques have the potential to mitigate existing pitfalls in statistical modeling, including biases arising from withdrawal of life-sustaining treatments and the associated self-fulfilling prophecy 55 …”

Identifying novel subgroups in heart failure patients with unsupervised machine learning: A scoping review

“…Clustering algorithms commonly are performed in a static way with baseline data and/or outcome data. They are useful to answer descriptive questions (10,11). In the early attempts, unsupervised clustering analysis algorithms were used on clinical laboratory indexes and demographic data characteristics of patients with heart failure to make homogeneous inductive groups (12,13).…”

Navigating the Ocean of Big Data in Neurocritical Care

“…Practitioners often struggle when reviewing the ever-expanding panoply of publications applying machine learning techniques to data analysis, not knowing how to critically review the methodological machinations behind the curtain of these algorithms. Gravesteijn and colleagues [20] provide a valuable review of machine learning's strengths and pitfalls, including the use of clustering to find endotypes (as proposed in the articles cited above) and the critical need for skepticism and validation of findings from machine learning studies. In addition, we must remember that an ethical and equity-minded perspective must be applied when collecting and implementing data-driven decision making in medicine [21].…”