A cost-aware framework for the development of AI models for healthcare applications.
A cost-aware framework for the development of AI models for healthcare applications.
复制标题
DOI：
10.1038/s41551-022-00872-8
复制
发表时间：
2022-12
影响因子：
28.1
通讯作者：
中科院分区：
文献类型：
--
作者：
最近出现的用于疾病诊断的精确人工智能（AI）模型提出了基于AI的临床决策支持可以大大降低医疗保健提供者工作量的可能性。然而，要实现这一点，人工智能预测模型的输入数据，即，患者的特征本身必须是低成本的，即，高效、廉价或低努力地获得。当收集数据的时间或财政资源有限时，例如在急诊或重症监护医学中，使用数千个患者特征的现代高精度人工智能模型可能是不切实际的。为了解决这个问题，我们开发了CoAI（成本感知AI）框架，以支持任何类型的AI预测模型（例如，深度神经网络、树集成模型等）在给定少量低成本功能的情况下做出准确的预测。我们表明，CoAI大大降低了预测院前急性创伤性凝血病，重症监护死亡率和门诊死亡率相对于现有风险评分的成本，同时提高了预测准确性。它在预测性能、模型成本和训练时间方面也优于现有的最先进的成本敏感预测方法。将这些结果外推到美国的所有创伤患者中表明，在固定的假阳性率下，CoAI可以提醒提供商比其他风险评分更多的数万个危险事件，同时将提供商的数据收集时间减少约90%，导致每年节省20万个累积小时。我们推断CoAI在重症监护中的临床效用也有类似的增加。这些优势源于几个独特的优势：首先，CoAI使用公理化特征归因方法，可以精确估计特征重要性。其次，CoAI是模型不可知的，允许用户选择最适合预测任务和手头数据的预测模型。最后，与许多现有方法不同，CoAI在给定预算内找到高性能模型，而无需调整成本与性能的权衡。我们相信CoAI将极大地改善医学领域的患者护理，在这些领域中，需要在有限的时间和资源下进行预测。
The recent emergence of accurate artificial intelligence (AI) models for disease diagnosis raises the possibility that AI-based clinical decision support could substantially lower the workload of healthcare providers. However, for this to occur, the input data to an AI predictive model, i.e., the patient’s features, must themselves be low-cost, that is, efficient, inexpensive, or low-effort to acquire. When time or financial resources for gathering data are limited, as in emergency or critical care medicine, modern high-accuracy AI models that use thousands of patient features are likely impractical. To address this problem, we developed the CoAI (Cost-aware AI) framework to enable any kind of AI predictive model (e.g., deep neural networks, tree ensemble models, etc.) to make accurate predictions given a small number of low-cost features. We show that CoAI dramatically reduces the cost of predicting prehospital acute traumatic coagulopathy, intensive care mortality, and outpatient mortality relative to existing risk scores, while improving prediction accuracy. It also outperforms existing state-of-the-art cost-sensitive prediction approaches in terms of predictive performance, model cost, and training time. Extrapolating these results to all trauma patients in the United States shows that, at a fixed false positive rate, CoAI could alert providers of tens of thousands more dangerous events than other risk scores while reducing providers’ data-gathering time by about 90 percent, leading to a savings of 200,000 cumulative hours per year across all providers. We extrapolate similar increases in clinical utility for CoAI in intensive care. These benefits stem from several unique strengths: First, CoAI uses axiomatic feature attribution methods that enable precise estimation of feature importance. Second, CoAI is model-agnostic, allowing users to choose the predictive model that performs the best for the prediction task and data at hand. Finally, unlike many existing methods, CoAI finds high-performance models within a given budget without any tuning of the cost-vs-performance tradeoff. We believe CoAI will dramatically improve patient care in the domains of medicine in which predictions need to be made with limited time and resources.
登录
查看更多内容
影响因子：
64.8
作者：
Esteva A;Kuprel B;Novoa RA;Ko J;Swetter SM;Blau HM;Thrun S
通讯作者：
Thrun S
影响因子：
8.8
作者：
Davenport R;Manson J;De'Ath H;Platton S;Coates A;Allard S;Hart D;Pearse R;Pasi KJ;MacCallum P;Stanworth S;Brohi K
通讯作者：
Brohi K
影响因子：
8.8
作者：
Johnson, Alistair E. W.;Kramer, Andrew A.;Clifford, Gari D.
通讯作者：
Clifford, Gari D.
影响因子：
4.5
作者：
Friedman, JH
通讯作者：
Friedman, JH
影响因子：
158.5
作者：
Celli, BR;Cote, CG;Cabral, HJ
通讯作者：
Cabral, HJ