SBIR Phase I: Combining Machine Learning with Clinical Expertise to Assess and Mitigate Risk in Healthcare

这一小型企业创新研究(SBIR)第一阶段项目的更广泛影响/商业潜力将改善与复杂程序相关的医疗保健结果。在美国，大约每7个主要外科手术中就有一个与医疗并发症有关，每年总计超过400万个并发症和800亿美元的成本。更多的并发症发生在非手术环境中。该项目将使用机器学习将经过验证的工程原理与临床专业知识相结合，以识别和解决每个程序、护理设施和患者的特定风险(考虑到从糖尿病到健康的社会决定因素等高影响风险因素)。这项技术将以一种新颖的方式，用经济高效的定制、面向流程的工具来增强现有的标准化、面向结果的质量改进工具，并设想在美国约5,100家社区医院进行初步应用。1%的并发症略有改善，每年将减少近10亿美元的成本，并将挽救4,500多人的生命。这个小型企业创新研究(SBIR)第一阶段项目将使用基于系统的方法来检查医疗保健中的流程级风险。医疗保健中的大数据结构不一致，没有进行优化以直接改善患者结果。大多数程序的大型数据集仅提供了关于风险的高级结论；它们没有准确指出高风险的提供者工作流程中的具体步骤或外部因素的作用，如合并症或设施年限。这个项目将确定在经验丰富的临床医生的监督下使用机器学习的可行性，以使用故障模式和影响分析中的原理来评估风险。该项目将开发一个概念验证机器学习系统，该系统使用专有的风险分类和修改器，将国家、州、设施和精算数据集结合在一起，为服务线的每个步骤生成风险优先级编号。然后将该系统应用于冠状动脉搭桥术，以评估其有效性和临床价值。蒙特卡洛模拟和使用利克特量表的临床医生焦点小组将确定结果的重要性。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will improve health care outcomes associated with complex procedures. Approximately 1 in 7 major surgical procedures in the US is associated with a medical complication, totaling more than 4 million complications and $80 billion in costs per year. Many more complications occur in non-surgical settings. This project will use machine learning to combine proven engineering principles with clinical expertise to identify and address specific risks for each procedure, care facility, and patient (accounting for high-impact risk factors ranging from diabetes to social determinants of health). This technology will augment existing standardized, outcome-oriented quality-improvement tools with cost-effective customized, process-oriented tools in a novel way, with an envisioned initial application for the ~5,100 community hospitals in the US. A modest improvement of 1% of complications would annually reduce costs by nearly $1 billion and will save 4,500+ lives.This Small Business Innovation Research (SBIR) Phase I project will use a systems-based approach to examine process-level risk in healthcare. Big data in healthcare is inconsistently structured and not optimized to directly improve patient outcomes. The large datasets for most procedures provide only high-level conclusions regarding risk; they do not pinpoint the specific steps in provider workflow with high risk or the role of external factors, such as comorbidities or facility age. This project will determine the feasibility of using machine learning supervised by experienced clinicians to assess risk using principles from Failure Modes and Effects Analysis. The project will develop a proof-of-concept machine-learning system that uses a proprietary risk taxonomy and modifiers to combine national, state, facility, and actuarial datasets to generate risk priority numbers for each step for a service line. This system will then be applied to coronary artery bypass graft surgery to assess its validity and clinical value. Monte Carlo simulations and clinician focus groups using a Likert scale will determine the significance of the results.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.