Both sides previous revision
Previous revision
Next revision
|
Previous revision
Next revision
Both sides next revision
|
aihc [2021/10/13 17:22] nigam |
aihc [2022/09/23 15:19] nigam |
====== Stanford Medicine Program for Artificial Intelligence in Healthcare ====== | ====== Stanford Medicine Program for Artificial Intelligence in Healthcare ====== |
| |
We are working on a set of efforts collectively referred to as the [[https://stanfordhealthcare.org/stanford-health-now/ceo-report/advancing-precision-health-takes-real-smarts.html|Stanford Medicine Program for Artificial Intelligence in Healthcare]], with the mission of bringing AI technologies to the clinic, safely, cost-effectively and ethically. {{:shc_program_for_ai_in_healthcare.pdf|See brochure}}. The four key components are: | In healthcare, predictive models play a role not unlike that of blood tests, X-rays, or MRIs: They influence decisions about whether an intervention is appropriate. Whether a model is usefulness depends on the interplay between the model's output, the intervention it triggers, and the intervention’s benefits and harms. We are working on a set of efforts collectively referred to as the [[https://stanfordhealthcare.org/stanford-health-now/ceo-report/advancing-precision-health-takes-real-smarts.html|Stanford Medicine Program for Artificial Intelligence in Healthcare]], with the mission of bringing AI technologies to the clinic, safely, cost-effectively and ethically. |
| |
| {{::model-interplay.png?nolink&400|}} |
| |
| Our research evolved from the [[http://stanmed.stanford.edu/2018summer/artificial-intelligence-puts-humanity-health-care.html|effort]] in improving palliative care using machine learning [[https://jamanetwork.com/journals/jama/fullarticle/2748179?guestAccessKey=8cef0271-616d-4e8e-852a-0fddaa0e5101|Ensuring that machine learning models are clinically useful]] requires [[https://www.nature.com/articles/s41591-019-0651-8| estimating the hidden deployment cost of predictive models]] as well as quantifying the [[http://academic.oup.com/jamia/article/28/6/1149/6045012|impact of work capacity constraints]] on achievable benefit, estimating [[https://www.sciencedirect.com/science/article/pii/S1532046421001544|individualized utility]], and learning [[https://pubmed.ncbi.nlm.nih.gov/34350942/|optimal decision thresholds]]. Pre-empting [[https://www.nejm.org/doi/full/10.1056/NEJMp1714229|ethical challenges]] often requires keeping [[https://hai.stanford.edu/news/when-algorithmic-fairness-fixes-fail-case-keeping-humans-loop|humans in the loop]]. |
| |
- **Implementation**: We partner with the Technology and Digital Solutions team at Stanford Hospital to deploy predictive models in care delivery workflows. See our [[http://stanmed.stanford.edu/2018summer/artificial-intelligence-puts-humanity-health-care.html|effort]] in improving palliative care and its [[https://www.statnews.com/2020/07/01/end-of-life-artificial-intelligence/|coverage in Statnews]]. | |
- **Ensuring that models are useful**: The utility of making a prediction and taking actions depends on factors beyond model accuracy, such as lead time offered by the prediction, the existence of a mitigating action, the cost and ease of intervening, the logistics of the intervention, and incentives of making the intervention. | |
- Check out our ideas on [[https://jamanetwork.com/journals/jama/fullarticle/2748179?guestAccessKey=8cef0271-616d-4e8e-852a-0fddaa0e5101|making machine learning models clinical useful]], [[https://www.nature.com/articles/s41591-019-0651-8| estimating the hidden deployment cost of predictive models]], [[https://innovations.bmj.com/content/6/2/45|bridging the implementation gap of machine learning in healthcare]], and developing a [[https://www.nature.com/articles/s41746-020-00318-y|delivery science for AI in healthcare]]. | |
- Read about our framework for quantifying the [[http://academic.oup.com/jamia/article/28/6/1149/6045012|impact of work capacity constraints]] on achieved benefit, estimating [[https://www.sciencedirect.com/science/article/pii/S1532046421001544|individualized utility]], and learning [[https://pubmed.ncbi.nlm.nih.gov/34350942/|optimal decision thresholds]] from aggregate clinician behavior. | |
- **Safety, ethics, and health system effects**: We map the multiple groups involved in taking action in response to a prediction and study their varying perspectives, positions, stakes, and commitments to pre-empt ethical challenges. Read our perspective on [[https://www.nejm.org/doi/full/10.1056/NEJMp1714229|addressing ethical challenges]]. We believe that the use of AI can lead to good decisions if we keep [[https://hai.stanford.edu/news/when-algorithmic-fairness-fixes-fail-case-keeping-humans-loop|human intelligence in the loop]]. | |
- **Training and Partnerships**: We partner with multiple groups to figure out “what would we do differently” if we had a prediction from a model and to investigate the pros and cons of using AI to guide care. For example, we [[https://www.researchgate.net/publication/341829909_The_accuracy_vs_coverage_trade-off_in_patient-facing_diagnosis_models|examined the accuracy vs coverage trade off in patient facing diagnosis models]], and partnered with Google on efforts to enable [[https://www.nature.com/articles/s41746-018-0029-1| scalable and accurate deep learning with electronic health records]]. Stanford students, check out the [[https://stanfordmlgroup.github.io/programs/aihc-bootcamp/|AI for Healthcare Bootcamp]]. | |
If done right, the adoption of efficacious prediction-action pairs can massively improve the ability of a health system to find patients at risk and act early. As part of this effort, we search for clinical situations where the application of risk-stratification and proactive action can provide cost-effective, health system level benefits. | |
| |
---- | ---- |