Artificial Intelligence to Enhance Clinical Value: Key Points
- Gill SK, Karwath A, Uh HW, et al., on behalf of the BigData@Heart Consortium and the cardAIc group.
- Artificial Intelligence to Enhance Clinical Value Across the Spectrum of Cardiovascular Healthcare. Eur Heart J 2023;Jan 11:[Epub ahead of print].
The following are key points to remember from this state-of-the art review on artificial intelligence (AI) to enhance clinical value across the spectrum of cardiovascular health care:
- A step-wise framework for applying AI can improve the use of AI in clinical research. This should start by determining the appropriate methods and data sets to address the research hypotheses. It is recommended to develop a clear hypothesis when selecting the data to be included in the analysis. Data type and collection method can be considered unstructured, semi-structured, structured, and structured as information such as demographics or blood biomarkers. Semi-structured includes text-based or continuously measured data, while unstructured includes imaging data.
- Adequate data pre-processing is essential for allowing AI methods to deliver accurate, meaningful results. The output of any AI model is only as good as the data inputted. Missingness is important to consider in this step. Considering standardizing or normalized data, imputation, and dimensionality reduction is also important.
- Choosing the appropriate machine learning (ML) approach is based on the clinical question or hypothesis and the setting in which learning is performed. Supervised learning uses ML approaches with labeled data to train in prediction, while unsupervised learning’s computational algorithms have no ground truth for comparison. Overfitting is a concern for prediction models during the training stage of any ML approach. A wide variety of ML algorithms are available, including decision trees, random forests, neural networks, deep neural networks, and autoencoders.
- Validation is a critical next step to allow an understanding of how generalizable the findings are between data sets and real-world clinical practice. External validation is recommended for all AI studies. Currently, many studies do not validate their findings. It is recommended that investigators provide clear details on their methods, guided by the FAIR principles (Findability, Accessibility, Interoperability, and Reusability), and allow others to independently validate their findings.
- Investigators must be mindful of relevant social constructs, data privacy, and health care equality.
Keywords: Algorithms, Artificial Intelligence, Biomarkers, Delivery of Health Care, Intelligence, Machine Learning, Knowledge Management, Privacy, Treatment Outcome, Validation Study
< Back to Listings