Dataset Information

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.

SUBMITTER: Gravesteijn BY

PROVIDER: S-EPMC9071245 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BackgroundLow back pain is a common and disabling condition leading to large health service and societal costs. Although there are several treatment options for back pain little is known about how to improve patient choice in treatment selection. The purpose of this study was to pilot a decision support package to help people choose between low back pain treatments.MethodsThis was a single-centred pilot cluster randomised controlled trial conducted in a community physiotherapy service. We included adults with non-specific low back pain referred for physiotherapy. Intervention participants were sent an information booklet prior to their first consultation. Intervention physiotherapists were trained to enhance their skills in shared informed decision making. Those in the control arm received care as usual. The primary outcome was satisfaction with the treatment received at four months using a five-point Likert Scale dichotomised into "satisfaction" (very satisfied or somewhat satisfied) and "non-satisfaction" (neither satisfied nor dissatisfied, somewhat dissatisfied or very dissatisfied).ResultsWe recruited 148 participants. In the control arm 67% of participants were satisfied with their treatment and in the intervention arm 53%. The adjusted relative risk of being satisfied was 1.28 (95% confidence interval 0.79 to 2.09). For most secondary outcomes the trend was towards worse outcomes in the intervention group. For one measure; the Roland Morris Disability Questionnaire, this difference was clinically important (2.27, 95% confidence interval 0.08 to 4.47). Mean healthcare costs were slightly lower (£38 saving per patient) within the intervention arm but health outcomes were also less favourable (0.02 fewer QALYs); the estimated probability that the intervention would be cost-effective at an incremental threshold of £20,000 per QALY was 16%.ConclusionWe did not find that this decision support package improved satisfaction with treatment; it may have had a substantial negative effect on clinical outcome, and is very unlikely to prove cost-effective. That a decision support package might have a clinically important detrimental effect is of concern. To our knowledge this has not been observed previously. Decision support packages should be formally tested for clinical and cost-effectiveness, and safety before implementation.Trial registrationCurrent Controlled Trials ISRCTN46035546 registered on 11/02/10.

Dataset Information

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

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets