How health care AI models are developed

Timothy M. Smith , Contributing News Writer

Health care is dominated by data, and augmented intelligence (AI)—often called artificial intelligence—can help make sense of it in powerful new ways. In fact, an AMA survey of more than 1,000 physicians found nearly two-thirds appreciate its upsides.

Still, the “fail fast, fail often” mantra that drives much of technological innovation does not work in clinical settings because medical ethics require physicians to “first, do no harm.” It is therefore essential for physicians to be aware of how AI applications are developed and know how to scrutinize their use in medicine. 

Membership Moves Medicine™

  • Free access to JAMA Network™ and CME
  • Save hundreds on insurance
  • Fight for physicians and patient rights

An AMA Ed Hub™ CME series, “AI in Health Care,” introduces learners to foundational principles in AI and machine learning, a subdomain of AI that enables computers to learn patterns and relationships from data without being explicitly programmed by people. Developed by the AMA ChangeMedEd initiative and the University of Michigan DATA-MD team and geared toward medical students, it is also suitable for residents, fellows, practicing physicians and other health professionals.

The first module in the series, “Introduction to Artificial Intelligence in Health Care,” explores big data and its impact on AI in health care, outlines the potential strengths and limitations of AI in medicine, and summarizes the process of developing AI for use in a health care setting.

The other modules in the series so far are:

Additional modules will be released in the coming months.

The benefits of AI to health care are almost too numerous to mention. Among them, it can augment clinical decision-making, alleviate cognitive burden and enable precision medicine, as well as precision education.

But there are also numerous concerns around using Al, including how algorithmic biases can amplify inequities, the potential for data-privacy breaches and that machine learning models can be exceedingly complex and difficult to understand.

“As a health care professional, you should understand the potential promise and concerns around Al as you work with other stakeholders whose values and perspectives may not always align with one another,” the module notes. “In this context, health care professionals must play an active role to ensure that Al is implemented safely and effectively in clinical practice.”

Recognizing this, the AMA has developed new advocacy principles that builds on current AI policy. These new principles (PDF) address the development, deployment and use of health care AI.

Learn more with the AMA about the emerging landscape of augmented intelligence in health care (PDF).

The Ed Hub module highlights the key questions to ask when evaluating studies of machine-learning models, namely:

  • Is the study design appropriate?
  • Has the model been validated against a gold standard?
  • Did the model undergo external validation prior to deployment?

To help physicians answer these questions, the module breaks down how AI models are developed, tuned and evaluated. It first provides definitions of key terms and concepts, beginning with the various datasets:

  • Training set: The data used by Al to develop a model.
  • Validation, or tuning, set: A subset of the training set used to observe the performance of the model and fine tune the hyperparameters, which are the parameters decided before training starts, including model type, model architecture and learning/training-related parameters.
  • Test data: Held-out data the model has not seen before used to evaluate how well a model performs on new data.

The module then explains the process used to evaluate a model’s performance, walking through an example study of a model developed with supervised learning—the process of optimizing model parameters to best match the desired model output— used to predict if someone has coronary artery disease.

Periodic knowledge checks and review sections test the user’s vocabulary and understanding of how concepts are applied.

The CME module “Introduction to Artificial Intelligence in Health Care” is enduring material and designated by the AMA for a maximum of 0.50 AMA PRA Category 1 Credit™.

It is part of the AMA Ed Hub, an online platform with high-quality CME and education that supports the professional development needs of physicians and other health professionals. With topics relevant to you, it also offers an easy, streamlined way to find, take, track and report educational activities.

Learn more about AMA CME accreditation.