Motor racing is a funny old business. We make a new car every year, and then we spend the rest of the season trying to understand what it is we've built to make it better, to make it faster. And then the next year, we start again. Now, the car you see in front of you is quite complicated. The chassis is made up of about 11,000 components, the engine another 6,000, the electronics about eight and a half thousand. So there's about 25,000 things there that can go wrong. So motor racing is very much about attention to detail. The other thing about Formula 1 in particular is we're always changing the car. We're always trying to make it faster. So every two weeks, we will be making about 5,000 new components to fit to the car. Five to 10 percent of the race car will be different every two weeks of the year. So how do we do that? Well, we start our life with the racing car. We have a lot of sensors on the car to measure things. On the race car in front of you here there are about 120 sensors when it goes into a race. It's measuring all sorts of things around the car. That data is logged. We're logging about 500 different parameters within the data systems, about 13,000 health parameters and events to say when things are not working the way they should do, and we're sending that data back to the garage using telemetry at a rate of two to four megabits per second. So during a two-hour race, each car will be sending 750 million numbers. That's twice as many numbers as words that each of us speaks in a lifetime. It's a huge amount of data. But it's not enough just to have data and measure it. You need to be able to do something with it. So we've spent a lot of time and effort in turning the data into stories to be able to tell, what's the state of the engine, how are the tires degrading, what's the situation with fuel consumption? So all of this is taking data and turning it into knowledge that we can act upon. Okay, so let's have a look at a little bit of data. Let's pick a bit of data from another three-month-old patient. This is a child, and what you're seeing here is real data, and on the far right-hand side, where everything starts getting a little bit catastrophic, that is the patient going into cardiac arrest. It was deemed to be an unpredictable event. This was a heart attack that no one could see coming. But when we look at the information there, we can see that things are starting to become a little fuzzy about five minutes or so before the cardiac arrest. We can see small changes in things like the heart rate moving. These were all undetected by normal thresholds which would be applied to data. So the question is, why couldn't we see it? Was this a predictable event? Can we look more at the patterns in the data to be able to do things better? So this is a child, about the same age as the racing car on stage, three months old. It's a patient with a heart problem. Now, when you look at some of the data on the screen above, things like heart rate, pulse, oxygen, respiration rates, they're all unusual for a normal child, but they're quite normal for the child there, and so one of the challenges you have in health care is, how can I look at the patient in front of me, have something which is specific for her, and be able to detect when things start to change, when things start to deteriorate? Because like a racing car, any patient, when things start to go bad, you have a short time to make a difference. So what we did is we took a data system which we run every two weeks of the year in Formula 1 and we installed it on the hospital computers at Birmingham Children's Hospital. We streamed data from the bedside instruments in their pediatric intensive care so that we could both look at the data in real time and, more importantly, to store the data so that we could start to learn from it. And then, we applied an application on top which would allow us to tease out the patterns in the data in real time so we could see what was happening, so we could determine when things started to change. Now, in motor racing, we're all a little bit ambitious, audacious, a little bit arrogant sometimes, so we decided we would also look at the children as they were being transported to intensive care. Why should we wait until they arrived in the hospital before we started to look? And so we installed a real-time link between the ambulance and the hospital, just using normal 3G telephony to send that data so that the ambulance became an extra bed in intensive care. And then we started looking at the data. So the wiggly lines at the top, all the colors, this is the normal sort of data you would see on a monitor -- heart rate, pulse, oxygen within the blood, and respiration. The lines on the bottom, the blue and the red, these are the interesting ones. The red line is showing an automated version of the early warning score that Birmingham Children's Hospital were already running. They'd been running that since 2008, and already have stopped cardiac arrests and distress within the hospital. The blue line is an indication of when patterns start to change, and immediately, before we even started putting in clinical interpretation, we can see that the data is speaking to us. It's telling us that something is going wrong. The plot with the red and the green blobs, this is plotting different components of the data against each other. The green is us learning what is normal for that child. We call it the cloud of normality. And when things start to change, when conditions start to deteriorate, we move into the red line. There's no rocket science here. It is displaying data that exists already in a different way, to amplify it, to provide cues to the doctors, to the nurses, so they can see what's happening. In the same way that a good racing driver relies on cues to decide when to apply the brakes, when to turn into a corner, we need to help our physicians and our nurses to see when things are starting to go wrong. So we have a very ambitious program. We think that the race is on to do something differently. We are thinking big. It's the right thing to do. We have an approach which, if it's successful, there's no reason why it should stay within a hospital. It can go beyond the walls. With wireless connectivity these days, there is no reason why patients, doctors and nurses always have to be in the same place at the same time. And meanwhile, we'll take our little three-month-old baby, keep taking it to the track, keeping it safe, and making it faster and better. Thank you very much. (Applause)
赛车是一项充满乐趣的经典运动。 每年我们都造出新的赛车, 然后在这一年剩下的时间中 分析和理解新赛车的特性, 改进它,让它更快。 然后第二年,重新开始这个过程。 你现在看到的这辆车是非常复杂的。 光底盘就用了超过一万个零件, 引擎的零件数量超过六千个, 电控部分有大概八千五百个组件。 总零件数超过2.5万个,每个零件保证不能出错。 所以赛车运动需要非常注重细节。 F1方程式赛车的另一个特别的地方, 是我们一直在更新和改装赛车。 我们持续性的改进它使它更快。 所以平均每两周时间, 我们就会替换掉车内大约5000个零件。 赛车中大约5-10%的零部件 在一年中每两周就会被更换一次。 那么我们是怎么做到的? 我们的改进过程从比赛开始。 我们用大量的传感器来记录车辆运行状态。 你们面前的这辆车在比赛时, 会携带大约120个不同的传感器。 它们记录赛车运行时的所有数据。 数据被记录并保存。 我们的数据系统包含约500个内部参数, 1.3万个车况信息及事件记录器, 当系统的某些地方工作出现问题时, 我们将这些数据回传到车库进行分析 无线传送速率能达到2M到4M每秒 所以两个小时的比赛过程后,每辆车发送的数字 数据量超过7.5亿。 我们每个人一辈子说过的单词总数加起来, 还不到其中的一半。 这是很多的数据。 但是仅仅测量和记录这些数字还不够。 你需要利用这些数据做出点什么。 所以我们花了大量的时间和精力, 赋予这些数据以意义, 使我们能够知道,引擎的状态如何, 轮胎磨损的程度如何, 油耗的情势如何? 我们所做的这一切, 就是将数据转换成能够知道我们工作的知识。 现在,让我们看看一眼原始数据。 再看看从一个三个月大的 病人身上采集到的一些数据。 这是个孩子,而你现在看到的都是真实数据, 注意最右边的曲线, 情况开始变得有点糟糕, 患者开始出现心跳骤停的症状。 这被认为是一件无法预先判定的事件。 没有人预见到这次心脏病发作。 但是当我们仔细看这些曲线, 我们能够看到在心跳骤停的5分钟左右, 记录仪的数据出现了一些征兆。 我们能够看到心率等数据中 出现的一些细小的改变。 这些改变的幅度很小,常规的检测值 无法甄别出这些改变。 那么问题变成了,为什么我们注意不到这个? 这个事件可以用来预测么? 是不是我们对于数据中模式的分析越详尽 就可以预测的越准? 这个孩子 跟台上的这台赛车一样大, 三个月了。 这个孩子患上了心脏病。 现在,看着屏幕上的这些数据, 有心率、脉搏、血氧量、呼吸频率, 它们跟正常的小孩子相比都存在差异, 但是对于病房的小朋友们来说很正常, 所以医疗诊断的挑战之一, 就是我如何通过观察眼前的病人, 通过从她身上获取的特有的一些数据, 能够在情况开始恶化之前, 检测到苗头? 这点上病人与赛车是相似的, 当事情恶化时,你只有很短的时间, 来避免事态扩大。 我们的方法是使用一个数据系统, F1赛车每两周运行一次的数据系统, 我们把该系统安装在伯明翰儿童医院的 内部电脑上。 我们将医院儿童重症监护室中病床周围的设备 接入到我们的系统中, 这样我们就可以实时的查看到这些数据, 并且更重要的,我们将这些数据长期保存, 使我们能够从中寻找规律。 然后,我们在系统上使用了一款软件, 能够帮助我们实时的将数据中包含的模式 梳理出来,让我们看到事情发展的过程, 让我们能够确定情势何时开始发生了变化。 在赛车比赛中,我们都怀有野心, 无所畏惧,有时候还会有些鲁莽, 所以我们觉得我们应该在孩子被送往医院的路上, 就开始对他们进行数据采集和分析。 为什么要等到他们到了医院才开始 数据采集和分析呢? 于是我们在救护车和医院之间 搭建了一个实时数据传送的链接, 采用普通的3G通信网络传送数据, 现在急救车也变成了移动版的 重症监护室。 然后我们开始分析这些数据。 上面的这些弯弯绕绕、五颜六色的线条, 跟你在病床监视器上看到过的数据是一样的—— 心率、脉搏、血氧含量, 以及呼吸速率。 现在,底下的两条红色和蓝色的曲线, 是我们关心的。 红线表示的是伯明翰儿童医院 正在使用的早期预警阈值 这是(通过数据分析)自动生成的。 从2008年就开始运行了, 已经在医院内成功的阻止了多次 心搏骤停引发的悲剧。 蓝色的曲线向我们指示出 病人的情势何时开始发生变化, 让我们甚至不需要进行临床诊断 就能立即而直观的, 看到数据自己向我们传达的信息。 数据告诉我们出问题了。 这些红色和绿色的小球体, 表示的是来自不同样本群体的 同一种类型的数据。 我们通过绿色的小球表示的数据来学习“正常状态” 我们称之为“常态云”。 然后当情势开始变化, 当病情开始恶化, 数据就跳到了红色的范围。 这并不复杂。 我们只是将已经存在的数据用另一种方式呈现, 放大数据的差异,供医护人员诊断, 让他们更容易看出情势的变化。 这跟一个优秀的赛车手, 依赖于各种线索来决定何时刹车, 何时转向是一个道理, 我们需要帮助我们的医生和护士 在病情恶化的开始提前发现和处理。 我们有一个很雄伟的计划。 我们在这场比赛中需要打破常规。 我们目标远大,我们在做正确的事。 如果我们的方法是可行的,那么没有任何理由, 将这种方法的应用范围局限在医院内。 它可以有更广的使用范围。 无线网络连接在今天已经无处不在, 没有理由还要求病人、医生和护士 一定要在同一个时间,出现在 同一个地点。 与此同时,我们将继续改进我们的赛车, 确保其车况良好,确保其安全, 并且使之更快、更好。 谢谢大家。 (掌声)