[00:00:00]
>> My main use evenly and I'm an associate professor at Georgia Tech instrument and can Knology at the systems engineering school and I'm also the director for the Centers for operations research in medicine and health care. I'm going to talk about some of the advances in research focus in our center and basically we look at the medical and also health care delivery system as a whole and we try to use information technology decision support them operates research and all these techniques so that we can actually at bands be medicine and health care delivery system in the United States and most of it's going to focus on the.

[00:00:43]
Medical side is that we look at from the investigation to the bedside How do you actually come up with biological advances and how do you actually put it in place into clinical practice and in this case we have a diagnosis treatment Our How do they leverage that and also the treatment outcome in the hospital management side that is to eat fish and see all of the delivery they're still organization of the health and if we're side that is the quality and improvement of processes the operations efficiency.

[00:01:13]
Information Management and also the change in the organization infrastructures so that the advances could be implemented and could be carried out. So you have a look at that business state over our research focuses that we are doing basically recombine both the medication and also medicine side with the health care delivery side and we've developed and able laying methods and tools that focus on systems modeling in the mattocks and engineering techniques.

[00:01:44]
So I'm going to focus on make I'm going to have three parts in this talk the first part will focus on the medical advances. So basically we look at how all of these cities informatics here focusing on mothering and computing is how do we make use of the various.

[00:02:00]
Information in biological or medical and also trying to extract information that our knowledge as we call it information all of those. So that we can interpret them and also be able to really come up with the in the way to ideas to assist in early detection end of engine drop delivery trim on the sign and outcome prediction.

[00:02:24]
Now as you noticed this is really quite complex. Task because we are doing with very heterogeneous type of biological and chemical data. So you can look at first of the diagnostic part. Basically we can take advantage of systems modeling and quantitative analysis a very lauch type of data set for example bottle to go imaging you know trying that medical records how do we use dose information so that we can actually and couple of discriminatory patents and I will show you some example.

[00:02:54]
This is really important in terms of looking at health risk prediction and looking in early the stand noses because it will affect how you into the in that. How do you treat them and also how do you monitor the treatment whether it is susceptible and not so you have a look at that the technology that we have developed over the past fifteen years is a very general purpose for large scale machine learning and at the official intelligence framework.

[00:03:18]
You know what we have is steam. Into a global information for example it could include gene expression data markers. It could include imaging data. It could also include patient chemical information then going through the pattern recognition module it will identify some of the very important attributes inside that then inside here.

[00:03:41]
It will. Sometimes we will repeat could I mean to give you an idea how much do these type of attributes is we have identified some of the gene signatures that is in the older all fly ten millions of them. Now if you want to predict something you really cannot use all these interviews.

[00:03:58]
Then there's a feature selection. Where it will identify this form of the right patterns. Maybe ten of them fifteen of them. How do they actually allowed us to classify these type of these into different pads. Thembi have to come up different types of patients. This could be whether the patient is have they seized it whether the patient is normal or what type of the efficiency that they have.

[00:04:20]
So just to give you an idea of some of these work and how we use mission ending in the efficient arrogance for a good mining and early diagnosis. So give your free example is that we have made some advances to be zany So for example you look at our sinus D.C.S. one of the key that notice of L. Sammis disease is that we really like to be able to determine in my own conduct impairment the state in which it is to early a state of mild cognitive impairment that can allow the doctor to actually provide medication to the patient and that will slow down all you can stop the progression into else amnesties so in this case it is really critical to be able to identify early.

[00:05:02]
So what we you list is that we use to M.R.I. imaging information and more we use actually new psychological information which is just a whole type of test where patients are being asked to draw some diagrams as about some words as about their favorite things and using those information although they are very heterogeneous and also these ferret we are able to identify a process of patients that are normal patients that have very mild cognitive impairment and as suspected to to be to be really getting into some instances in they this age as well as those patients that are already have this disease.

[00:05:40]
So this is very important and also exciting work. So the next one is about method there are mixed information. Now one of the recent advances is that we are able to actually produce spectrums of these methods met metabolized step inside the body. The difficulty. Is that with all of this information that a capture how can do and to use them to be able to actually identify certain illness that the patient may have all hidden how risk.

[00:06:09]
So basically we look at patients with various types of concentrations and for these particular one that you see on the screen that relates to the self I mean the essence deficiency. So we look at these and identified patterns within these spectrums so that we can identify individuals that are having normal level up metabolized they have excess amount of them or they are deficient among that now.

[00:06:33]
This allows us to really pinpoint the health risks and that is actually very important because a lot of times before that these systems form the body actually went through some of the transformation in terms of the changes in the metabolites level changes in certain types of the. Pap weights and so we will be able to identify those at Netaji And also when those sites for the clinician to actually determine intervention in the next one.

[00:06:59]
This is about cancer and also epigenetics in human. So basically methane nation of the D.N.A. relates to the at the sheen of the C H three group into the D.N.A. and we call the C.P.G. Islands is the concentration of that part of the human genome where the concentration of C.P.G. than nucleotides is high is about six to twelve percent versus the normal three percent and in the human genome this is about to go out to more in one thousand C.B.D. islands.

[00:07:31]
So why are they important in cancer study. Basically the schematically we can look at is that the system normal cells and to be put the group the city and this is the reaction of the C.B.D. islands followed by the meth nation up the genes. That's the methylation of it for cancer patients.

[00:07:47]
If we actually go for this to be the seeds for meth nation. Well received a meth nation of the C.P.G. islands followed by hyper meth nation off the genes and now this turns out to be very important because. At Berande meth nation of usually not a not meth native city Janice Of course a lot in human can.

[00:08:08]
Human cancer. So we would like to be able to explore the importance of C.B.D. Island meth nation in cancers at the Janabi level we want to be able to develop predictive room so that we know what cost. This math nation is there Gene signatures that we can see.

[00:08:22]
And more importantly U.V. can identify those sites we can actually activate those genes that are being silenced by the words in the D.N.A. methylation and also then in this case it really allows ones to do early intervention and develop novel treatment strategies. Why are they so important. So let me show you some of these methylation S.O.C. the gene inactivation in cancer.

[00:08:44]
What you see is a very interesting. Thank you say here in the case of genes that are being silenced by methe nation. This is the site in the chromosome. And this is the tumor types. What do you see is that it covers a very broad base of cancer and also some of the cancer for example pancreatic cancer that Kamya brain cancer is among the child trial of cancer and those that are incurable with these a very critical study that one this one has to conduct because if we could actually diagnosis days thirty.

[00:09:18]
That means before the cancer is formed. We will be able to really do target intervention and we may be able to remove those cancer in which currently it is incurable into a Keurig curable state. So just to give you an idea how that with this basically this is the first study that was being done in two thousand and three in which we looked at some of the breast cancer sequences that the method they did was System normal patients that are non methylated and we want to see are there any gene signatures that allow us to predict that nation.

[00:09:51]
Basically we go through the pattern recognition face weight identifies more than ten million Gene signatures very short ones basically it's a not. G. cancer genes decide June short Gene signatures about five to twelve months and then basically the once they get the ten million of those to go through the foetus election to identify what among these huge set the ones that are really critical to really cross abide the cancer versus non cancer patients and we were able at the able to identify seven.

[00:10:26]
Of these sequences and then classify these what is most important in this study is able to actually blind create using these. Predictive grew and get ninety percent correct classification. So that is really important for. Really early disease diagnosis is that you are able to actually do a blind test and more importantly instead recently what we have done is set using these predictive really that is actually identified using breast cancer sequences instead be able to predict some of those sequences space on lung cancer patients.

[00:11:02]
Now this is really interesting as we see that gene silencing because of meth nation of course among many different type of cancers. We're able to really cross really analyzed across different type of cancer in terms of the study and this also allows us to pinpoint a twenty nine thousand C.P.U. audience in human gene them.

[00:11:24]
What are the ones to methe nation problem and then pinpoint those so that the power outages could actually validate them in the test and speed up the analysis and also can say Vance's. So after we look at some of these examples in diagnosis and going to show you some of the treatments the sign and advances that we have so what we would like to focus on treatment here is really designing individualized treatment so that we can maximize the outcome here outcome for different patients will be different.

[00:11:54]
We want to look at how do we be signed optimal drink drug regimen. How about radiation therapy. We combined modality of different type of treatment. How do we secede Indows treatment and one of the key things here is step most patients have different type off. Ready. The actions to different drugs how do you educate and Egypt or this information and be able to come up with one that is that is best for the individuals.

[00:12:21]
So I'm going to show us a couple of these examples. So you know that cancer is that we have about one for one point four million new cancer cases in two thousand and seven. If you look at that is set about over slightly over the Hotham median up the in deaths and half of these cancer patients actually receive some form of radiation and just in two thousand and four alone.

[00:12:46]
If you're not looking at like how fast it rapidly grows in terms of the health care costs and seventy two billion is spent on cancer treatment and alone in the United States. Now you don't get early stage prostate cancer. This is one of the cancer that is. Most common among American male and about one third of the prostate cancer patients and early stage is receiving radiation treatment in the form of palettes so this is the size of the palace retire need just like a grain of rice and it is placed inside the prostate through these localization templates and the long needle that put the seat inside.

[00:13:26]
So the key to success of this treatment is how do you actually put these suits in size so that the radiation dose can form nicely to the prostate and that it will not have too much those eight effect in the surrounding normal tissue. Now this is a difficult task because as you look at the three dimensional figure at these other to see it's what we put in and this is due to your wrist there which is a normal critical structure that is very important for the patients.

[00:13:53]
So we want to mend taint the functioning of the of these normal critical structure. So any other form of the. Then comes in as a for the breast cancer patients in this case to ease the Paschal. Impact me of the big breasts too much and then the seats will be implanted temporarily.

[00:14:14]
So again how do we edge of place to see its inside so that we get the best treatment is very important the locations of these suits incorporating biological advances that we can actually look at for example some of the information that we can get from M.R.I. as images tell us the spatial information of the cancer cells spread across the.

[00:14:34]
Tumor region how do we take advantage of those how do we actually incorporate than them make response during therapy that means when the treatment started. How do you take advantage of the string gauge of the tumor cells and so that the treatment will be more effective. So there are many challenges that we're looking at.

[00:14:51]
So just to give you an example of this is the advance every step of this research acquired many years ago that state in ninety ninety five and it took quite many years to actually go through it dances and then do the clinical trials so in the traditional minute instead across many clinical settings is that computer aided treatment planning is used to that offline trial and error approach is used to the side the configuration of the seats so as to be actually replaced this is so we actually used and also mated to and planning and that allows once actually image the tumor and add to the sign the planning right during the day of operation that overcomes a lot of the pre-planning limitations that the clinicians and count that in the past and not only that because now we can actually design the plant in seconds.

[00:15:43]
We can actually also incorporate some of this information he what we do is that instead of chasing the anti prostate as a homogeneous ness like in all the clinical sites what we do is that we look at the M.R.I. as images and we actually identify pockets of cancer cells that.

[00:16:00]
Very dense and incorporate those information for those escalation. We also look at the shrinkage of the prostate over a period of thirty days and during that the sign of the treatment be take that into a bandage so that we can actually send a much improved treatment. So what does it resell again.

[00:16:18]
Istead this system. I Laos one to solve these treatment planning problem in real time in the operating room it improved it you look intro from sixty five percent to ninety five percent. It also drastically reduced the side effect that the patients and counted. So you look at just in the operation side.

[00:16:36]
That means reducing the offline planning and make the operations actually on on nine and during the day of operation we actually reduce about four hundred fifty million dollars for prostate cancer patients alone in the United States. Now let's go to the after we look at some of these children and then we look at the outcome prediction.

[00:16:57]
The key here is stats as patients come in they would like to be able to employ evidence based practice so that we can actually provide the best recommendation to each individual patients that type of treatment that is best suited to the type of basis that they have and the demographics and the really in the medical and also micro level.

[00:17:19]
So I'm going to industry one example here. As said if you look at the vaccines a lot of times is that these and it is really difficult but one of the key things is that some patients may be allergic to certain type of exams and they will only know it when the patients are exposed to the infection.

[00:17:37]
So one of the challenge is how do we actually predict the effectiveness of these vaccines without infecting the individual. So basically the we will with the order just and in the nation's Emory Vaccine Center is set the they used yellow fever of X. in as a model and.

[00:17:56]
You know that scene is actually a very good. Mix in MA. To study because it has been used for half a billion of people over the last seventy years and that a single shot has been very effective for nearly thirty years in terms of the induced immunity but very little is known about what makes it effective.

[00:18:16]
So what the experiments here is that biologically sperm instead a group of patients there to meet up with the vaccine and the T. cells and about the extent responses are being studied the gene expression patterns in the right but cells are collected over a period of sixty days and there are about fifty thousand genes signatures that are collected and that is time dependent.

[00:18:40]
So just to show you some pictures here is said these are the T. cells response over a period of time and then you see the difference in the T. cells and they reached how they response. So we think he pot here is we want to be able to identify do signatures that can predict the manito of T. cells response and by response induced by the Eckstein we want to know if it's a high or a low response in the.

[00:19:05]
T. cells and how do we actually big to be cells stand about ticks response it's second thing more important is that we want to know once we identify these signature they can predict those response. Can we actually pretty big that on a set of independent patients the response that these patients have so these is very important because this allows us to validate precisely the predictive rule is correct.

[00:19:29]
So what we have is very exciting results we are able to identify these think Gene signatures that discriminate the level of T. cell response and everybody expects fans to do induced by the vaccine and in the brine prediction on an independent set of patients it us as high as ninety percent correct classification.

[00:19:47]
So this is very exciting. This is the first kind first study of this kind of critical clinical significance. Regarding to the zine evaluation and effectiveness of the vaccine from the. Clinical significance side is that it allows one to really predict the immunity and effect an off the big scene and that means if the city rep at the Santa vixen.

[00:20:09]
S.Y.S. evaluation of it. It also ident I allow us to identify individual school radiation and receive these to have a big C in May So that actually are allowed as to prevent the harm that may be done to the vaccine by the breaks into individuals that shouldn't have taken those facts in it and some very fundamental questions that need to bed breaks in Nations and prevention of these.

[00:20:31]
The next step that the group is working on now is to design effective flu vaccine. Here we understand that it takes about six months or more to identify an effective flu vaccine and using this study we can speed up the process because I'm not up to predictive rule actually identified Gene signatures that is within seven days of the vaccination which is very very important because we can identify at very early stage that means we can actually alter it.

[00:20:58]
This is the sign and improve the. Vixen outcome. So the rep at the sign and that they vet edition of an effective vaccine is really critical in the emergency response in infectious disease celebrate it before through a pandemic or biological attack. So no you don't look at the what I have shown you is the and that is we have a look at his diagnosis treatment and the outcome prediction but no one would like to see if you look at chronic disease management only if you look at.

[00:21:27]
A lot of the. Like management he says management that commissions have to look at is that it would really be helpful. You want in that you look at the Anti of these these cycles and be able to come up with different weights of looking at the. Potential deficiency of the care process and really try to improve it.

[00:21:47]
So just to show you one of these is that this is an I.V. T.C.'s cycle that we have developed the in this column it shows you the how in the within the population individual edge of contract. Type of the Seas. This is the diagnostics part. This is the D.C. staging and this is the outcome where the the the individual receive this different up treatment and then they state part that is a successful treatment and that is the fair treatment that means the patients may die as a reseller so for this type of model it's very logically the key part is to be able to have the complication of Pollard that allow clinician to look at different staging and be able to run the tests so that they can look at different interventions and how to improve the clinical outcome.

[00:22:33]
So I haven't finished the first part on the medical process side or now we're going to look at the their liver side that means. How do you actually really perform some of these health care operations so that you provide the best care when you're actually trying to deliver all these advances in translation of side to the patients.

[00:22:53]
So basically the vision here is really to transform the American healthcare system following the Institute of Medicine into one that is safe. You fat the efficient timely equitable and patient centered. Now these are the six factors that is identified by instead of Medicine ten years ago and that is really the goal of what we would like to see.

[00:23:13]
So the mission here is really that we must would industrial partners in this case that means health systems expression to hospitals and clinicians and also physicians group and by working together with them. We want to be able to really improve the health care delivery system. So some of the projects that we have here on the operation side for example.

[00:23:33]
Optimal care delivery model we try to optimize the health care delivery by the creasing the practice variations improve the quality of care and also equalize that length of stay of. Of the different type of patients for example we can look at a certain type of the seas and we look at how long does it take for this particular patient to be treated across different house systems and we see a huge variations in the practice in terms of the.

[00:24:00]
Time that they have to receive the treatment from that notice all the way to actually recovery. So this type of. Practice variance is very important to reduce because it is important for the patients to receive a standard quality of care within the same organization as us across different cities in the sector Monday look at different type of medication error.

[00:24:20]
Reduction and basically try to actually met the end tie a process of the. Medication during a minister to bring so that we know exactly what is causing this error and also the relative plans to mitigate that and in the last one is about optimizing the electronic medical record usage and how did you actually use the systems beyond just the opt in them and be placing the paper bread.

[00:24:45]
So I'm going to show you a little bit of these one because it is actually one of the very important aspect of health information technology is how the optimize that you are trying that medical record usage beyond just at Docking the usage of keeping records. So the background.

[00:25:00]
Instead the kind of the date of meat. I turn the medical records from that poetry data and also from imaging really provide of wealth of information for advancing diagnosis optimizing have good delivery operations and also improving patient care. So one of the fundamental difficulty is really to develop clinical informatics system that that is at to continue to allow to one to capture all these evolution with the in the clinical setting.

[00:25:25]
So we edge of developed the clinical data management system so that we can actually at the opt to the evolving data as well as a kind of practice. Now one of the things we look at the E.M.R. is to look management. This basically is that whenever there are questions regarding medications or the clinical practice and that the system they indicates that the clinician actually prescribe something that is different than the standard practice it will actually fire out some of these records but the problem instead within a week we actually get sixty thousand of these.

[00:26:00]
And that costs a lot of the teeth that means if there is really a real problem. The clinician may not be able to notice it because they're just too many of these alarms setting off. So what we do this is a very complex analysis that we have actually looked at him or not.

[00:26:14]
The systems that consists of multiple clinics is that we look at all of the alerts trying to look at those at a future that means they actually are not important alerts and they shouldn't be as shown on the screen the ones that I'm through. They're looking at the different types of the space on drug.

[00:26:30]
Alerts allergies the dose alerts and duplications and they actually come up with a very interesting decision support system that allowed condition to actually determine what they have to do based on these alerts so and now the last part that I'm going to talk about is still the public health and human agency response in a sense that both Part One and Part two are gritty integrated as you know this if you look at Yemen.

[00:26:56]
You can add to us and all this information to advance to medical care process by looking evidence based diagnosis chairman and the Senate and everything in the last part about public how be looking at population health. But look at how do we actually deal with such a capability.

[00:27:13]
How do you deal with the responses and how do you diagnose from the chemicals syndrome make and also and momento detection like if there's about a logical infectious and how do we actually handle those things. So basically this relates to the population health monitoring about city readiness and emergency response and it in really really involved in multilevel strategic and operational planning all of the public health and health systems infrastructure and because for these don't just treat sick patients.

[00:27:45]
We also have to protect our large population that are well but that we have to dispense prophylactic treatment to protect them. So some of these challenges to really include the strategic stockpile medical supply distribution locations of these Stenson facilities. In addition to existing health care systems that are there.

[00:28:03]
How do you optimize the city the staffing when you have very limited resources and how do you route a population and also all the logistics related to transportation and this Mensing that of these so what we have actually developed some of these systems along with C.D.C. is T. is a large Go Public Health Informatics decision support systems and it is actually used in many different ways.

[00:28:29]
I will show you in the next one. That's the real opt is to software system. It is a modeling and information decision support systems for operations planning and logistics for clinically signed basically allowed you to design clinic layout so that you can actually have the best throughput It allows you to do resource allocation time motion study of the services.

[00:28:50]
It allows you to track this is probably gauge and which is really important when you look at the health care facilities how do you mitigate the this is probably geishas so that you can avoid. Overwhelmed of the facilities with patients that get infected and then you also want to be able to do Blue time decision making.

[00:29:09]
So the system has been used since two thousand and three for anthrax and also. Different type of vaccinations across the country. So that's this one actually relates to the anthrax. Drill exercise and that involves basically be dispensing up and body aches and this one is actually a real flu vaccine and is to say during season the X. you know how do you actually manage these individuals and then this one is actually for a drive through of an infectious disease outbreak Brit sample you have the smallpox or or like a flu.

[00:29:49]
You don't really want patients to come in to get a big scene. You would like them just to go through drive through and these actually is a real hepatitis drive through clinic that. Set up in Buffalo. So basically this is actually a response to one confirmed case of hepatitis A in the grocery store worker and that triggers ten thousand individuals that have to be vaccinated.

[00:30:13]
So these are really really event and basically it really is a very complex system and all the way from the clinical setting to the operations side to be had to really look at it in great detail. So these conclude the presentation and this is the website and the e-mail for.

[00:30:32]
But a communication. Thank you.