医院如何利用机器学习挽救生命,节省开支

作者: Derrick Harris (JUL. 14,2014)

关键词: 医院机器学习

大数据文摘翻译/整理: 兔八哥

转载请保留

Researchers at theUniversity of Washington, Tacoma, have built a machine learning system capableof predicting readmission risks for congestive heart failure patients. It hasshown good results in a pilot deployment, and now the team hopes tocommercialize the technology.

位于塔科马的华盛顿大学的研究人员，建立了一个机器学习系统能够预测充血性心脏衰竭患者的再住院风险。它在试点部署显示了良好的效果，现在研究小组希望将该技术商业化。

Whenmost people leave the hospital after a lengthy stay, they probably assume theywon’t be coming back again soon to deal with the same problem. Unfortunately,that’s often just wishful thinking. In fact, re-admissions — sometimes within justa couple weeks — are such a big problem that the Affordable Care Act (akaObamacare) includes measures toaddress the problem.

大部分人在经过漫长的住院而出院后, 一般都认为自己不会在短期内因为相同的原因再来医院. 但这只是人们一厢情愿的想法. 事实上,”再次就诊” — 有时两次就诊间隔短短几个礼拜 — 是<平价医疗法案> (又称Obamacare)所针对解决的一个重要问题.

Putsimply, the law provides financial incentives for hospitals to improvereadmission rates and financial sticks with which to punish hospitals where theproblem persists. Improve the problem, get more funding. Keep readmittingpatients within short windows after discharge, don’t get paid for treatment.The latter scenario is bad for patients and bad for hospitals.

简单来说, ”法案”利用财政奖励来鼓励医院降低再入院率, 并对那些再入院情况没有得到改善的医院实行财政处罚. 一种情况是: 改善存在的问题, 获取更多的资金; 另一种情况是: 对短期再入院情况置之不理, 从而无法得到就诊的费用. 第二种情况明显对病人与医院都很不好.

According to studies, about a quarter ofMedicare patients treated for heart failure are readmitted within 30 days, andheart-failure re-admissions alone cost Medicare about $15 billion a year.Predictions about how many of those are preventable range from less than 20percent up the the Department of Health and Human Services estimate of 75percent.

据研究显示，美国Medicare(译者注：Medicare为美国常见的针对残疾人和65岁以上老年人的保险)医保患者中治疗心脏衰竭的患者有四分之一的人平均在30天内又再次入院，而单单因为心脏衰竭患者再次入院所消耗的医疗保险费用每年约有150亿美金。据美国卫生部(Department of Health and Human Service)估算，这其中有20%到75%的花费可以避免。

“Ifyou can predict that, that’s a huge, huge cost saving for the hospitals,” saidAnkur Teredesai, who manages the Center for Data Science at the University ofWashington, Tacoma.

在华盛顿大学塔科马校区的数据科学中心的AnkurTeredesai说道： “如果你能预测的话，这将给医院节省十分巨大的开支。”

However,help might be on the way thanks to a research project by Teredesaiand his Center for Data Science colleagues. It’s called the Risk-O-Meter, andit’s already being used by one hospital system in the Seattle area. Now, theresearchers who created it are looking to commercialize it, either by licensingthe access to the cloud-based service or by starting their own company.

Teredesai以及他在数据科学中心的同事正在做一项研究，来帮助这个进程。这项研究成果叫做“风险零米（Risk-O-Meter）”。它已经在西雅图地区的医院系统开始使用。现在，这些研究者们正在试图将它商业化，采取方式是要么通过授权基于云技术的服务，要么开办自己的公司。

Underthe hood of the web and mobile applications that allow doctors to enter patientinformation and receive a risk score is a machine learning system that analyzesmore than 100 attributes about each patient. These range from standard stuffsuch as vital signs, lab results and medical history to more-personal stuffsuch as a patient’s demographic information and living conditions.

运行在网络和移动应用程序后台的是一个机器学习系统，它允许医生输入病人信息，分析病人100多个不同属性，给出风险评分。这些病人属性包括一些标准的东西，如生命体征，实验室结果和病历，以及一些比较私人化的东西，如病人的人口学数据和生活条件。

However,the Risk-O-Meter has much more utility than simply as a one-off risk-scoringapp, Teredesai explained. Risk scores change as patients progress throughtreatment, helping doctors to evaluate treatment options on an ongoing basis.Even after patients leave the hospital, hospital staff can benefit from alertsindicating it’s a good time to check up on a patient, or to call with remindersabout taking medication.

然而，Teredesai解释，"风险零米（Risk-O-Meter）"不仅只是个一次性的风险评分应用程序，而是有更多的用途。风险评分会随着病人的治疗进展而改变，帮助医生随时动态地评估可选治疗方案。甚至在病人离开医院后，医院工作人员还可以根据系统提示来确定现在是一个很好的时间来检查病人，或电话提醒病人服药。

Doctorscan also drill down into the data in order to figure out what factors arecausing a score to spike. This type of analysis is important because a highscore could be caused by a non-medical factor that’s easy enough to account foronce a patient is discharged. For example, Teredesai said, “The chances of thengetting readmitted are higher — much higher — if [patients] live alone. … Themodels actually show that.”

医生还可以直接查看数据来弄清楚究竟是什么因素造成某些分数过高。这种分析非常重要，因为一个高的分数可能是由于非医疗因素导致的, 这种因素在病人出院后很容易解决。例如，Teredesai说，“如果病人独自生活, 获得再次入院的几率则更高 - 并且高很多。该模型清楚表明了这点。“

Onthe treatment front, he noted that although doctors will often keep patientsfor 13 days after heart failure, the project’s models show that risk scoresdrop significantly after 10 days, but then rise again after 13 days. So perhapsthose three extra days aren’t really necessary, and might even help cause areadmission, Teredesai said.

在治疗方面，他指出，虽然医生会让患者心脏衰竭后住院13天，该项目的模型显示，在10天后的风险分数大幅下降，但13天后再次上升。因此，也许这些额外的三天住院不是必要的，甚至可能导致再次入院。

The Risk-O-Meter team is also working on away to help doctors evaluate the effectiveness, or risk, of variousintervention options, although Teredesai noted that’s still a work in progress.It’s potentially a very valuable tool, though, because a system that can modelspecific treatments against the profiles of specific patients could result inmore effective treatment plans.

风险O米(Risk-O-Meter)的团队也正在研究一种方法，帮助医生评估各种干预方案的有效性或风险，虽然Teredesai也提到，这仍然是一个进展中的工作。这可能是一个非常有价值的工具，如果有这样一个系统，可以分析特定患者的资料来定制具体的治疗方法，就可以帮医生找到更有效的治疗方案。

Theteam also wants to integrate the risk-scoring system into existing electronicmedical records systems in order to cut down on the number of tools doctorshave to use.

该小组还希望将风险评分系统集成到现有的电子病历系统，以减少医生不得不使用的工具数量。

Teredesaisaid the system was designed with privacy in mind, which is why it runs on aHIPAA-compliant cloud platform (Microsoft Azure) and lets hospitals choose whatdata they want send for analysis. Whatever they send, though, it remainsanonymous and is discarded once it’s scored against the models. The systemdoesn’t keep any patient info, he said.

Teredesai表示，系统在设计之初就考虑到了隐私问题，这也就是为什么启用了HIPPA 兼容的云平台(Microsoft Azure)，并且让医院选择他们愿意发送分析哪些数据。医院发送的任何数据都是匿名的，并且会在模型计算打分之后立刻销毁。Teredesai说“ 这套系统不保留任何患者信息 ”。

We’veseen a lot of activity — and a lot of capital investment – at the intersectionof big data and health care, but many efforts, including IBM Watson and startup Lumiata, are morediagnostic in nature. A lot have been strictlyacademic research or the results of individual hospitalsystems analyzing their own data.

在大数据与医疗保健的交汇处我们已经看到过很多项目在实施——更准确地说的很多资本投资活动，但是包括IBM Watson和startup Lumiata在内的项目很多都只是侧重于诊断。还有很多工作只是严谨的学术研究，或者是私人医院系统在分析他们自己的数据。

Whatmight be most appealing about the Risk-O-Meter is that it’s a broadlydeployable cloud service that promises better patient outcomes while alsohelping hospitals where it matters most to them — their bottom lines. HospitalCIOs and administrators know they need to do both, and anything that canplausibly deliver has to at least get a serious look.

风险O米(Risk-O-Meter)最有吸引力的可能是，它是一个广泛部署的云服务，承诺改善患者治疗效果，同时也帮助医院解决他们最需要的问题 - 这是他们的底线。医院CIO和管理者知道他们需要两者兼顾，任何能够满足这种需求的产品/服务都值得一看。