all right hey well good afternoon what a pleasure to be here and what an amazing opportunity to have a think about how we can impact healthcare both in New Zealand but around the world I was particularly taken by della Bohr’s example of of the work that they’re doing in Serbia and Macedonia and how cloud is driving the transformation by the data and insights that they can get from that I just astounding I’m kind of a little embarrassed that I think in New Zealand we can still do better and I think we need to challenge ourselves to think differently about the application of AI technology machine learning and how data can improve the healthcare system for all New Zealanders in the next session I really want to talk about finding signal and the noise Healthcare is awash in data it has tremendous amounts of insights to be gained but much of the data is unevenly distributed across multiple silos that we can only make an impact if we bring that together in new ways and find new ways of innovation and in our session I’m going to be I’m delighted to bring a couple of real key innovators in the space to the fore to help you understand the opportunity that we have in front of us and how our technology is being applied to improve the way we deliver health to all humanity I think such an adela our CEO says at best he says artificial intelligence probably represents one of Technology’s most important priorities but this is the part that I really like the most and healthcare is perhaps a is most urgent application and that’s what drives me that’s what makes me passionate about making an impact in health I do think technology can change the way we do things it can help save people’s lives it can help improve the way clinicians can work with their patients and it can make an indelible impact in the way that we care and and look after our communities and so it’s incredibly important about how we apply this technology to make that happen what makes that happen are the huge advances we’ve made in computing and the transition to cloud computing and the advances that we’ve made and things like deep neural networks which are enabling us to do a whole raft new things the advances we’ve made in computer vision helping machines be able to see the work we’re doing in intelligent data warehousing and analytics and some of the examples you saw before the work around helping a computer understand language understand terms and the way humans talk and the impact of those and how we can turn computers into digital assistants that can support us all and of course enabling cloud architectures that empower us to be able to do new things that we’ve never been able to do before or work with the massive datasets that things like genomics are bringing to us where that kind of compute we could never run in a hospital or under a doctor’s desk in a clinic before so do the cloud in a lie really matter in healthcare of course they do absolutely and nowhere is that more important than when I saw in a fall weeks ago in Bangkok there I met with Professor dr. Tucker from the University of Tokyo he’s made an amazing advance with his team of researchers and being able to take endoscopic camera videos and be able to analyze for tumor growths directly in the video processing of those it takes an ordinary clinician about four hours to go through one of those videos and look for abnormalities and those endoscopy cameras and many times that there’s a big delay between the endoscopy happening and the analysis being done by professionals what dr. Tucker has done at the University of Tokyo is using AI and the power of computing and what we can do today he’s able to reduce that from four hours down to two minutes of processing and that will only get faster as a time but if you think about the big problem that we have with bowel cancer screening even in this country the impact of that transition and the ability for the computer already to be able to detect at human capability and increasingly being even better than human capability our future is profound and what we’re going to be able to do in the healthcare space so from an AI for healthcare perspective there’s really three things that we think are going to be important to be able to make that happen the first is of course giving you the platforms and tools to be able to do it this that the most important one is the one in the middle which how do we empower people that make health care work there’s no point in running around going we’re going to replace doctors and nurses and do these kinds of things it’s it’s complete nonsense but what we do need to do is we need to amplify the ability of all the smart people in the world that care in this health care system and empower them with new capabilities and new tools that amplify their own human ability as well and third we want to be able to take data and develop the future of precision medicine and things like genomics and and all the things that come with it are going to empower whole new rafts of treatments that are going to help us beat cancer help us identify disease earlier and help us find new treatments to diseases we may not even know about yet so with the sea of data in healthcare what’s the what’s our what’s our real opportunity we’ve got an incredible amount of data in healthcare healthcare produces more data than any other industry you heard this morning from surco in a session that a CTS and 800 slice CT image for for the sort of head anterior 800 800 gigabytes of data a gigabyte per slice how do you process and store that data and do intelligent things with it is incredibly complicated but you combine that with all the laboratory data that pharmacy data the medication data and so forth and we’ve got a whole world of challenges in front of us around how to process that so we need to think about these health opportunities but we also need to think about how we put that health data in the hands of innovators how do we take the national collections of data that we bring together in New Zealand and New Zealand does an amazing job and creating data and storing date of health but we don’t necessarily empower that data to be used as effectively as we could and therein lies our opportunity and therein lies our opportunity for innovators like yourselves so there are a lot of opportunities for ala in healthcare no matter from from left to right or up or down the ability to be able to do those things is incredible and so what I want to do is give you some examples of things that we’ve got coming and things that we’re doing right now to be able to make an impact in health and the first one is all about empowering people on the front lines of care so out of our Israel team who are currently doing some amazing work in bots and health bots but particularly using healthcare content to be able to build BOTS that are intelligent and are able to work with patients in a trusted and researched way with medical content that makes a real big difference and we need to do this in a private and secure way and so we’ve got working with many customers around the world right now and trialing this health bot technology you can see if it can make an impact make an impact in the area of triage and post discharged service patients reminders engaging patients in their post in their post care follow-up plans so we’re doing these kinds of things but we also want to make it a lot easier for the rest of us to be able to build the kind of AI models so that you don’t necessarily need to be a computer scientist to try and figure out how to do this we want to empower business users in healthcare to be able to use those tools and visually design the workflow processes that they will have for their own healthcare clinic the second piece of work that we’re doing which i think is incredibly exciting is the work we’re doing out of our cambridge labs our research labs particularly with a project called inner eye inner eye is trying to change the way that radio radiologists can segment images so that we can drive treatments and far more quickly today it’s a relatively arduous process to delineate organs directly inside of CT scans and if you want to start driving treatments using linear accelerators for doing radiotherapy delivery and cancer then where is where to target those beams is incredibly critical to getting the right outcome but to do that there’s a lot of work to do in the pre-planning phases and so today this is kind of how you go through a CT slice you say this is kind of organ you might have to do that for 500 slices or 600 slices and it’s arduous work to be able to bring that together we’re changing the way we’re using the artificial intelligence and machine learning so that a process that now takes two hours for a radiographer to go through and do this and the pre-planning can be done in mere minutes and the impact to there is going to be mean that we can treat more people with the same resources than we have before and the great thing about that is is that efficiency of segmentation and supporting radiographers to be able to do this really enables us to be able to drive treatments far more clearly but are also able enable us to be able to identify tumors and track tumors and the progress of treatment against those in a far more efficient way as well the third example is an example that I think is going to be incredibly impactful and this this example really takes me back to a time when I went to a hospital and in South Korea in Seoul and I sat there with the doctors and they said look missus oh this is a hospital we’re incredibly busy and our outpatient clinic has all these patients coming in all the time now it’s a bit unusual in New Zealand because in New Zealand you probably spend 10 15 minutes with your GP and you think that’s always too short but here it’s not unusual for adopters to see 160 patients in a day so think about that that’s two minute medicine and then we’re going to ask them to type the medication record into a computer that’s never going to happen so they come they have an army of support staff and nurses with them that are doing the documentation as the doctor goes from patient to patient we think that the advances in AM machine learning can make a big difference here and so we’re doing this project called project compare MD and I’ll let me show you how that works the empower MD Intelligence scribe listens to doctor/patient conversations and captures medical intelligence in the form of notes suggestions let’s see how it works hi mr. mink what brings you another question oh my god i’m stiii get tired all the time now you know i’m just so you get brave and it’ll just having trouble breathing it didn’t tired i can’t sleep and i yeah you know I’ve been doing everything I’m supposed to be doing you know that sounds like fatigue and some shortness of breath are the main things that are bringing around today the intelligence grab captures and synthesizes clinically relevant concepts this includes entities intense and synthesis phrases that naturally occur in the doctor’s speech it also captures important phrases in patience speech suggestions are mapped to the standard subjective objective assessment and plan sections of a medical note each suggestion is ranked with a confidence level from high to low at any point the doctor can choose to accept high confident suggestions into the note the doctor can map back to the original transcript for more context on the suggestions the doctor can also remove the suggestion from the note the system will learn from this the doctor can add lower confidence items to and edit the note as they please the doctor is in full control of the note and can add or remove items as they see fit the intelligent scribe learns from all of these choices once the doctor is happy with the note it’s ready to be signed the empower MD intelligent scribe empowers doctors to focus more of their time on patients this is a foundation for building a learning system that synthesizes medical knowledge at scale so I think that’s occurring that berries are going to be incredible advances to empowering and amplifying people’s ability in health care and imagine those doctors in Korea and China that are seeing such high volumes of patients the kinds of things that we can do around clinical noting and the data it generates and what we can then do with that data is incredible and empowering clinicians is really what it’s all about and so with that I’d like to invite chief architect Gareth Gareth Beaumont from Volterra to come up and talk a little bit about how they’re driving AI and machine learning to empower radiologists and mammographies to better address breast cancer and breast cancer detection Gareth please come on up I’ve been asked to give you a brief rundown on Valle para health technologies and our journey as a health tech we’re based in Wellington and taking on the world of mammography at a global scale my name is Gareth Beaumont and there’s indicated on the chief architect and also the information security officer for Volterra I’ve been without power for a little over two and a half years now and consider myself fortunate to be part of such an end in the vitter but also meaningful company let me show case to you what my para health technologies does for the next 20 minutes who we are what we do and how we’ve got to where we are today utilizing combinations of leading technologies including AI I believe we are revolutionizing our specific corner of the health industry I hope to give you a little insight into our journey so far and what I believe the journey will lead us to moving forward we were founded in 2009 from research originally conducted at Oxford University in the 80s well para is based in Wellington New Zealand and facilitates the early detection of breast cancer our digital health solutions enable personalized high quality breast cancer screening based on automated objective measurements of breast density and quality we hold 34 patents featuring 273 science publications hold regular occurrences including FDA and C and we are now also ISO 13485 quality assurance in 27001 information security accredited as an ace X listed company we’ve successfully raised 40 million Australian including 20 million Australian and April and May this year and currently have customers and/or research projects in 36 countries at this stage I must confess that I have an electrical engineering and software engineering background combined with the fact that my handwriting is illegible my aspiring medical career has not yet amounted to much fortunately for you I have my colleagues Gillian Marshall that and the amazing staff at Boca Raton Hospital and Florida the us to explain the essence of what volt power offers in a short video my name is Nancy Marcus I’m 62 years old I enjoy yoga I enjoy theater I enjoy life I had my yearly mammogram and the results came back that I have dense breast tissue I was aware of that my previous few mammals came back with the same results it was stated that I appeared normal and cancer-free but that I might want to follow up with an ultrasound as they couldn’t be sure my gynecologist ordered the ultrasound and something suspicious came up small I was tall chances are it’s nothing my cancer was detected I’m lucky that my gynecologist insisted on the ultrasound it was found small found early easily treated like Nancy Marcus nearly 50% of women over 40 in the US have dense breasts for most women mammography is still the most effective tool for breast cancer screening but it doesn’t work equally well in all women particularly those with high breast density not only has it been linked to an increased risk of breast cancer but high breast density also makes mammography less effective the higher the breast density the more difficult it is to find cancer in fact 35% of breast cancers go undetected by screening mammography in women with dense breasts quantifying breast density is critical for early detection but it also complicates the issue the difficulty we as radiologists have in determining breast density is that we use just our visual assessment of the mammogram to come up with a level of breast density but it has been shown that a physician being shown one woman’s mammogram may read it as hi breast density and one day and the next day look at the same mammogram and read it as low breast density so that there’s a lack of consistency a lack of reliability and repeatability and it becomes even worse if you have one radiologist read it one day and the next year another radiologist reads it the the lack of consistency is even worse how can we take subjectivity out of the equation the answer lies in volumetric density tools like the one developed by Valera solutions called Val para density it gives an objective automated score providing consistency and quality that can’t be achieved by doctors alone we’re assessing a patient’s risk and so it’s very important that we are as accurate as possible and that is the reason why we got involved with the volumetric automated breast density assessment program that helps us lose the subjectivity and become much more objective and consistent and reliable and accurate in our determination of breast density women need to know just as I know their cholesterol or their blood pressure or their blood type they need to know what their breast density is so that they can have the appropriate screening while para believes that screening saves lives using FAL para allows the clinician to personalize screening for the woman herself some modalities are less effective for women with lower breast density and more effective for those with higher breast density so we want to be careful to give the right women the right test and in the long run that will help control the cost of healthcare to make sure we’re using our resources efficiently additional solutions like the para Enterprise enable breast imaging services to continuously monitor hundreds of quality and performance measures on every screening mammogram the FDA and in particular MQ sa standards basically dictate they need to spy check images every three years and frankly the beauty of enterprise is that we look at every single image and with that we can do analysis and provide metrics on every single patient so the benefit there is that in almost a near real-time we’re giving data back to the center to provide a better patient experience better diagnostic and overall providing you know more clinical value than three-year spot-check while para enterprise software provides feedback to the technologist so they can better understand their own performance how well they did at positioning the breasts how well they did at providing a comfortable mammogram to the woman and also how effective they were controlling the radiation dose to the woman for an administrator and somebody who is passionate about quality this was a godsend to me we are able to see things about our centers our technologists our productivity that we had no way of tracking prior and it has really made a difference in the quality of imaging that we do as well as even our productivity and how we staff and even how we do quality control on some of our equipment so to me in the last thirty years I’ve been doing in breast imaging there have been certain technological advancements that really have changed our lives but having had the opportunity to use an old parent Enterprise to me I am so excited about what it’s going to do for our practice and how we’re going to be able to improve our efficiencies and our effectiveness it touches so many different areas with the data mining that people have done in other industries now that we’re able to pull from the data that we have that’s been hidden here for many many years and to be able to improve our practice our effectiveness and our efficiencies you can just going to improve the outcomes that we have we owe that to our patients and clarify some of the points in the video as well as setting the context to the markets for para and seventy-five million woman a screened each year globally using mammography 39 million of which are currently in the US as it financial year end approximately 3.7 percent of these 39 million in the US are processed by vipera software that’s only 1.5 million woman we still have a long way to go in making a difference in both the US and here around the world the primary medical principle that our software is based on breast density strongly correlates to increased risk of cancer the density specifically been that a fiber Colangelo tissue they should convert to fatty tissue with age by nature however of fiber ganger tissue it shows up in x-rays unfortunately similar to actual cancers now to fully understand the technology and where and what we are doing I believe it is important to realize where the technology of mammography has come from the premise of x-rays in general was only invented in 1895 by vilem Rajan previously diagnosis was possible was only possible by physical examination and the doctors best guess in 1913 a German surgeon Albert Solomon first began x-raying breast tissue specimens detecting a visible difference between healthy and unhealthy tissue micro calcifications in tumors in particular being denoted this was effectively the beginning of mammography as we know today from that point on it became a technology race the first compression mammography machine was invented in 1966 as with photography of the time this was on physical hardcopy film film eventually gave way to do a digital mimic Moulton sorry a digital mammography only however becoming comparable to dessert to traditional x-ray methods in 2004 with machines improving both quality and patient experience it is worth pointing out that this is only a potential maximum digital data set of 14 years the digital images today are still predominantly reviewed manually by radiologists in 1989 Ralph Honam our CEO first ventured the idea of using AI to assist with reading and screening of mammography images to Professor Sir Mike Brady and Oxford together they pioneered what is the foundations of opera health technologies today since then with the eventual digitization of images and the ever-growing processing and storage capabilities that the cloud now provides for para is making constant advances we are not directly helping woman yet but the more we understand about Bruce density the better the understanding we have about breast cancer the hoping that one day we will be able to assess the risk of a woman in their 30s start those screenings earlier and the treatments earlier and save lives as mentioned in the video our legacy and core business has been volt power density to date a rating a percentage of breast density calculated using our painters at algorithms this was carried out on service at a customer’s site using algorithms that we had crafted in-house the result is visually supplied to the radiologists interpreting each patient’s images as a scorecard not dissimilar to the one shown the intention being that this is an assistive metric and now viewing the radiologists to decide whether further more expensive modalities such as MRI or ultrasound to be employed on a patient by patient basis the following diagram is illustrative of the data flow that DICOM imagery flows from a hospital or clinic employing volterra analytics DICOM is a medical industry imaging format comprised of both imagery and associated metadata and it’s this data rich file that we push to Azure and consume before displaying in many different visual forms to the users the key takeaway from this diagram is that we are taking large amounts of data which has been processed and hosted and a scalable yet secure manner globally to clarify my previous comments about data rich files a DICOM image file consists of the expected but met imagery but also can be seen from this slide a wealth of metadata this data ranges from image descriptors all the way through to the voltage that the x-ray tube was consuming at the time from this digital imagery we have found that a multitude of other measures can be quantified and qualified our new flagship product for parent Enterprise analytics is that offering providing data visualizations on key metrics that our hospital and mammography clinics had not been able to contemplate collating and digesting themselves before one of these is position and quality as mentioned in the video this is a key offering that the Analects product assists technologists benefiting the patients and both their screening experience but also the reduction and number of retake images and screening sessions required this serves dual purposes the first being timeliness and having accurate answers as a patient is clearly a benefit and secondly in the u.s. an FDA enforce program called equipt mandates the monitoring for mammography quality vulparia analytics helps provide that quality reporting and to assist the technologists and improving day on day on day we are slowly but surely deriving other benefits from the start also data that is exponentially growing in volume preventative maintenance equipment room and staff utilization and more importantly temporal analysis all of these additional benefits adding to the value of our products and understanding towards early detection artificial intelligence AI is certainly something we’re incorporating more and more into our products these days one of their motivations for getting this right is the ever-growing prison pressure on the trained and skilled radiologists currently in the field their numbers are reducing globally due to ageing workforce reasons automation and an assistant manner being one of our solutions to help aid this problem our approach is simple we try to mimic our radiologist would like would logically utilize multiple data sources discreetly whilst reaching an evaluation decision as opposed to the approach of feeding all the data in at once and expecting AI to provide The Smoking Gun solution one of the single tasks we use AI for as the early detection of artifacts and the images that we are processing we now with a factor in then image has anomalies present such as pacemakers chemo ports or even eyeglasses means our calculations for density and not impacted another single task for AI is the detection and segmentation of the pectoral muscle one of the key challenges when interpreting a mammography image as the density of a pectoral muscle being to that of tuna sales the appearance of this and a medial lateral oblique mlo view will increase false positives accurately being able to segment us out for further calculations clearly been a benefit one class of objects that we are working to classify and detect also implants for our purposes has been able to identify and convey that a patient has implants work directly assists with positioning during the screening session once again being able to exclude the implant from density calculations also ensuring accuracy phantom images are used daily to evaluate their imaging performance of mammography systems prior to real patients these images range from straightforward non-biological and appearance through to swirl and biopsy images detecting these and being able to exclude them from processing and more importantly aggregate results we callate for research reasons as the key purpose of the single task image blur is something we have just had fda clearance for and are currently developing at valcara this is literally where the image due to patient movement or poor positioning has blew the image in some way being able to detect and advise us with the patient’s still present in the exam room means a lower recall rate and improved image quality clearly a complex task for AI and with human interpretation also been highly subjective this will be a great use of AI and our products if you recall the diagram regarding data flows one new exciting project we are currently working on as i/o HT appliances internet a health things this will revolutionize the on-premise element of our products by providing enhanced security maintainability and serviceability this will also extend our ability to carry out this machine learning and timely load balanced fashion whether it is from the appliance at the source often referred to as on the edge or alternatively from the cloud achieved fundamentally with the same code this is what we are working on at Valle para and this is what we’re trying to achieve ultimately reducing breast cancer globally thank you yeah thank you so much and thank you for everything that volper is doing for advancing breast cancer screening we’re happening right here in Wellington New Zealand amazing research is having amazing impact for Paris doing incredible work with our teams across Taiwan Singapore Thailand and and and beyond and so the impact from from Wellington New Zealand in this space as global and in the u.s. I know that business is going incredibly well as well thank you for all the work that you guys do and keep up the innovation it’s great to see so we’re also making significant advances in computer vision and you saw some great examples of computers being able to analyze breast tissue and being able to do those kinds of things but there is also many other scenarios with this computer vision and these deep neural networks that it can understand what they’re seeing and be able to analyze that can also have significant impact as all of us know with elderly parents or people with with they were starting to suffer from eyesight issues or just trying to remember what a certain pill does can be very very challenging if I have difficulty in seeing them what we’re doing at the moment is we’re building and of our research labs some work around computer vision for real-time pedalled identification so imagine you can take your iPhone get a camera you can hold it up to the pills and the medications that you’ve been given and if you don’t know what they are you can put your pill in front of them analyze them and it can go and can go and find out what that Pilar’s and be able to give you information and insight on that now for many of us that doesn’t matter we just sort of get them in a little pill packs and we take them and we just assume it’s all it’s all right but it’s not like that all over the world and also if you’re if you can’t see or you have trouble seeing or making out color and so forth the computer can then become an assistive agent and being able to make that happen and our technology behind the scenes to make that happen in the cloud becomes incredibly powerful to do this so being able to take Azure being able to take in containers some of our AI work you can actually go and build these types of applications very very quickly and if you’ve got ideas around how you want to innovate for healthcare we believe we have we now have a workbench of tools that really empowers you to be able to drive this innovation to make that happen we know that these things are going to be important some of these projects are not going to be as useful as others but it’s important that as a company we also research and support the efforts of innovators like yourselves to be able to take this and take the ideas and take the learnings from them so we can all we can all share the outcomes to build up to build a greater outcome the other area that I spoke about earlier is the area of precision medicine and the ability to take the masses of amounts of data that we did build in healthcare and really do impactful things with it and I think we’re really starting to be able to do things that we weren’t able to five years ago ten years ago being able to take these massive data sets and rethink how we do clinical research and how we support that we’ve been doing work with st. Jude Children’s Hospital and research and we’re trying to take their whole genomic program and advance it in the cloud traditionally this is a very complex space with a lot of data being able to sequence DNA takes a long long time and through some advances in computer science and work and the work that Microsoft Research has done with with the biotech sand and bioinformaticians who wrote the algorithms we’ve managed to increase the time at take that reduced the time it takes to to sequence DNA by 7.8 times so just by applying computational advancements we can rapidly bring reduce the time it does to do this which means we can derive better insight and so we make those services now available in our is your cloud so that you guys can use them as a service and so this is how the cycle of innovation comes together to be able to make that impact we’ve also recently made an investments in a company on a Seattle called adaptive bio technologies these these guys are really at the forefront of immunotherapy and being able to use data to treat cancer and to be able to find better treatments for for complex cancer cases we’ve had some amazing cases in New Zealand recently where the immunotherapy is really curing people that were otherwise not able to be cured and for many of those people today that still means a journey to the United States but I’m hopeful that over the next five five years these types of treatments will become pervasive all around the world and that the data science and technology to be able to treat cancer and be able to bring these things to market are going to happen so with that speaking of Seattle I want to bring another partner up who’s been really a thought leader in being able to take the large data sets in healthcare and really have a big impact in trying to help hospitals have health insurers public health organizations become much more effective with the data they have and to be able to use things like machine learning and artificial intelligence to be able to predict the future and guide hospitals with insights rather than dashboards so with that I’d like to invite Jeff Lumpkin from from Kensei to come up and present to you the work that they’re doing to revolutionize the way hospitals and healthcare organizations can use data so in 2015 I took on a new role I joined the power bi engineering team and my specific group was called the cat team so the customer advisory team and it was a dream because I’d left finance after 15 years so just getting our finance was was great for me but what I got to do was travel around the world and help companies use data effectively and and my medium was power bi but I became more and more engaged in integrating data science into power bi so using data science and visualizing it to help drive change within organizations I came across this company in Seattle I actually presented to them and I was intrigued first by their model which is death versus data science that’s a pretty excuse-me ballsy motto for a company to have I went down and I met with them and I what really caught my interest was the level of collaboration that existed between the computer scientists the clinicians so the five doctors and one nurse that are on staff and the computer and the computer engineers what Kensei does is to take data from hospitals or payers and produce analytics and little predictions that can be used to save lives and that’s that’s a tremendous mission that we have a Kensei and I’d want it to be a part of it so nine months ago I left my dream job at Microsoft to come to Kansai and take on a role of vice president of analytics where’s the slide advancer thank you go all right so we have two basic groups one is out of Microsoft our CEO Sameer manager 8 was part of the Azure team so he reported directly to so Joseph Soroush the other founder CTO and Kurtis I came out of academia and is actually still a professor at the University of Washington where he ran the data science program so we have a strong belief in peer-reviewed papers in making our code available to anyone who wants to see it so we publish all our code up on github so we wish to be an open book about the type of data science that we’re doing we’ve worked with a lot of different organizations both in the US but also around the world Kaiser Permanente in Beaumont in the US Singapore’s health promotion board where we’re doing a study on disease progression for diabetics National Health both in Scotland and in UK University RN in England University of Washington rush Medical Center in Chicago and then the Center for Disease Control in the United States when I’m thinking about what chem side does we work with three types of systems the first of these is a system of record so this can be health records it can be claims data internal cost or operational data we feed our we feed this data into a system of intelligence so this is our our AI work and then we move that data back into systems of engagement to really make our AI actionable and usable and this is the really important part as a as someone who did a lot of work in power bi power guy was cool but it was it was an indirect way of driving action what we wish to do is take our output and integrate it right back into the systems of wrecked systems of engagement so that it can be used automatically by clinicians or care providers so we start out with variation analysis which I think dolly board described as just understanding what is going on with data so this is prescriptive analytics we bring that into the Kensei platform which is based on Azure and we have two areas of focus right now one is utilization so who are high utilizers of a healthcare system or high utilizers of prescription drugs we also focus on patient flow so there’s gonna be who is coming into emergency departments from there who progresses into an inpatient setting might they go into the ER or the ICU and then how do they get out what is the length of stay and what is the risk of readmission once they’re laid out in the hospital so this describes all of the different topics that we can cover with our our AI solutions starting with variation analysis very simply then looking at acute care management case management again utilization or member engagement so this is for a fordable care or accountable care organizations who need to not only look at how they treat people but how efficiently they treat people we have three systems that we think about one is Kensei IO so this is our data platform again as your based second is Kensei AI so this is where we make information usable to clients and then third is Ken site SDK and one of our primary goals is to enable customers to do this data science themselves so make our data platform usable not just by Ken size engineers but by data scientists on-site with a customer with that I’m going to jump into some demos and show you how we’re actually delivering information to customers it’s been asleep so long I need to sign in again can we switch over please Oh sir can we switch to this demo machine got it thank you okay thanks very much so this is the Kansai platform and this is the model Bank the model Bank is where we store all of the machine learning models produced by our data science most of this work is done in Python some is done in R and some is done on Azure machine learning platform but this is where a user can come in and look at all of the different predictions that we’ve developed within Microsoft and right here you can see with whom or or who actually did these predictions so we have an open platform meaning right now we have predictions from University of Washington from Microsoft from Fullerton Health and Singapore as well as others and just scrolling down you can see that we are covering a lot of different areas one example might be the risk of readmission prediction so here’s where we have the model name the author the description the academic references that were used to produce this model and then the model version how accurate it is and then links to the github repository work with the code for this data one example of an analytic that we’ve done recently is a population health analytic so this is looking at the opiate or the problem with drug addiction in the United States this is a data set produced by NC HS cover 16 years of data about mortalities from drug overdoses so there are two key metrics that we have in this data set the first of those is the mortality rates so this is deaths per 100,000 of population you can see it started in 1999 at six deaths per 100,000 of population and increased almost 3x to 17 same number of same rise in terms of number of actual fatalities so 55,000 per year in the United States that this map is showing six different regions of the US when I first put this data set together I believed that the highest rates mortality rates would be in the southeast in Appalachia so when I select the southeast part of the United States this mortality rate jumped from 17.1 to 17.7 I turned out to be wrong the region most severely affected is in the southwest where the rate is 19 and then my nexus was which stay is going to be the worst and California jumped out to me as the obvious one selecting California is actually below the national average what’s what was really high was New Mexico at twenty six point nine so Madison over New Mexico I have two bits of data here one is showing the average mortality rate across the 16 year time period those are the blue columns the red line is showing those states I’m sorry those counties within New Mexico that have exceeded the top bucket so you can mortality rate buckets where 0 to 2 2 to 4 4 to 6 all the way up to 28 to 30 the last bucket was greater than 30 deaths per 100,000 so in order to calculate fatalities I converted 2 to 4 into the average of 2 & 4 which was 3 greater than 30 I called 31 so this line of 31 means the number of counties in New Mexico that had hit this top level bucket and it turns out that more than half of the counties were at this very high rate so this then became kind of a metric that I started to focus on counting the number of counties which had exceeded or had hit this top bucket so what this line is showing us is that the progression of counties that were greater than 31 deaths per 100,000 so in 2001 there were 2 in 2005 there were 4 2010 there were 30 in 2015 148 2016 192 so this is a logarithmic progression right here and the implication is that while in the United States we started to pay some attention to this problem we’re not going to turn it around quickly because you just can’t stop acceleration like that on a dime and in the 4 days that I’ve been meeting with organizations in New Zealand I learned it bit and I knew us before you don’t have an opiate problem like we do in the US but I think you do have a methamphetamine problem that is pretty severe and so it would be I think a worthwhile exercise to take a look at that data and see if there’s a similar dynamic taking place I wanted to be able to map this so much like da LIBOR did I used a choropleth map and it was a really great way to understand rapidly how this this spread started so if we start in the first of the data set there were two parts of the country that had greater than 31 deaths per 100,000 one was Rio Arriba County in New Mexico the second was McDowell County in West Virginia in West Virginia so let’s focus on the southeast and I’m just gonna progress through years and what I want to call out is how this is spreading like a virus spreads and I’ve shown this to you know physicians and they say it’s spreading just like a disease spreads a hyper spreading disease so let’s jump to 2005 and you can see adjacent counties are now infected to 2010 it spreads much more rapidly 2015 and then 2016 this is all actual data and and so that what the spread is really severe I wanted to do a forecast so I did a regression forecast with this data to go out to 2025 and see what it looked like so I when I come to the forecast page what I forecast was a hundred sorry 1758 counties by 2025 so almost more more than half of the counties in the u.s. at this super high rate what did that look like from a mapping perspective when we select 2016 and will come in on the choropleth map this is the 2016 view for the US getting to 2025 it’s virtually the entire country is looking this way now of course there can be interventions which can change the directory trajectory but right now it doesn’t it doesn’t look very optimistic from my point of view so this is one example looking at population health and this has been a very effective tool that I’ve used to motivate people in power so politicians to take action in the US and at least within the Washington State area where I’ve been working we have been able to start impacting and then driving some very serious thing about how to resolve this problem I’m going to switch now to a much more clinical view so coming back to the kansai platform to look at a different model so this is a platform that we designed for clinicians to use in order to risk stratify patients and determine which patients they need to focus on very quickly so we’re combining three different analytical models the first of those is low ejection fraction ejection fraction is a measurement of how efficiently your heart is pumping so below 50% that means your you need some help and it ends up being a very good predictor for heart disease within one to three years so this has been in terms of charting CHF we use low ejection fraction a lot we also look at risk of readmission so one of the what is the probability of having to come back into the hospital within 30 days after having been let out and then finally risk of mortality so in this view we’re looking at a cohort of patients from one week who are coming into a large hospital group in the center part of the u.s. 5,000 patients 1,300 of whom are at risk of readmission so this is a 26% readmission rate at $12,000 per readmission you looking at 15 million dollars in risk I can cut it very simply by say Hospital group here and see that the risk of readmission fell slightly from the average or by major diagnostic category in this case major organs and systems when I select that the risk of readmission jumps to twenty-eight point seven so well above the overall average now I wanted to use this in a couple ways one way was to be able to sit with a decision-maker at a customer site say a CFO and say this is of significant risk and cost to you let’s understand what savings you might generate by driving a risk of readmission reduction program so I’m going to drill through now to the cost mitigation page this is where I would see with the CFO and say currently you are at 28.7% readmissions Medicare in the US says that we should be able to cut that by at least half with two-thirds is kind of the upper boundary of efficiencies that can be had so if we drop this readmission rate from 28% down to 14% we’re going to generate more than two million dollars in savings just for this one week’s worth of patients to go even further and say what happens if we are able to reduce by two-thirds and get down to 9% or well above three million in savings and the point here is to make power bi something that that we can use to help decision-makers get to decisions very very rapidly the next is how do we work this in a clinical setting so now we’re going to come to this free model page where we’re looking at cohorts of patients and let’s hope it renders there we go so we have we’re looking at patients by risk of readmission low ejection fraction and mortality we have 500 patients I’m going to select major organs and systems and so we’re going to filter down now to 1500 patients and now I want to select those patients who are at high risk of readmission and high risk of low ejection fraction and so now we’re down to 43 patients and these are the highest risk patients so these are the ones that I know I need to focus on and this alone is really important because one of the biggest challenges for clinicians or care providers is figuring out which patients should they focus their energies on and this allows them to do that very very rapidly so I talked about the importance of being able to to take actions very very quickly one of those actions would be to be able to sit with a patient and help the patient understand the risk that they are facing and what they can do to offset that risk so here I’ve selected one of the highest risk patients and I’ve drilled through to the patient risk mitigation page and this is where we might sit with Stephanie a 68 year old woman we see her provider we see what her died most recent diagnoses were and we also see that she has a predicted risk of readmission score of 84 but by improving her follow-up compliance meaning coming in to the doctor or the therapist when she should we can drop her risk by a point if we get her to take her medication as prescribed we’re gonna bring her down by another three points if we can get her to improve her diet to normalize blood glucose we’re gonna bring her down to 77 and if Stephanie can stop smoking we’re gonna bring her out of the high-risk category and into the lowest category and what we want to do this is help patients understand that they can have significant influence over their own health outcomes to really motivate people to take better care of themselves and better control of their lives and then lastly we integrated a power apps visualization right in here I just want to see if it renders otherwise I have to switch to a different view okay so I’m gonna I’m going to switch over to this view and the at-risk patients here and so the power app I hope will function maybe it’s just too far from from home but this is a power and what I wanted to show with this was how we would might integrate back into an EHR or how we might allow a clinician to take very rapid actions based on understanding risk so we had these same 43 patients I’m going to select off all of these 43 patients and then two simple actions that I program one of which is to create a PDF on my onedrive for business so in the US this is HIPAA compliant meaning it’s safe and legal to do we’re ensured that this data is safe and secure so here it’s going to render this PDF and we can come back to it but I can then take this PDF and share it on teams or charity via other means such as a fire API back into the EHR and then the other thing that I program is just to send an email in this case to myself so this now shows up in my email but this is a very simple example of how people can take action immediately in order to to most efficiently serve and treat customers this is an example of you know pretty simple example of what we do but I think our goal is to make data real and usable as quickly as possible so with that I’ll bring Dava back up on station thank you thank you so much Jeff I think it really goes to show the unreasonable effectiveness of data and how we can actually take the data that healthcare already generates and really start taking advantage of it to change the lives of all of us you’ve heard from Val Parra doing amazing work and breast cancer screening you’ve heard from Ken Tsai about the work that they’re doing to be able to predict the future and help clinicians and hospitals and healthcare organizations serve their customers better and you’ve heard how Microsoft is driving the research agenda and investing heavily in healthcare to think about the problems not just of today but how we’re going to solve some of the hard challenges that we’ve got coming in the future and continuing to build it out and build that infrastructure in our cloud so innovators like yourselves can take those technologies and make a difference I’m a big believer that healthcare with its data set has an amazing opportunity and the amazing opportunity to build the future for us now is really a upon us and I think the the time is now to really start taking those problems and start doing things with them