[00:00:05]
>> Thank you all for being here I know it's very busy at the beginning of the semester so it's great to see you all this morning can I see a show of hands how many of you are students at Georgia Tech or elsewhere OK And any visitors who are outside of Georgia Tech.

[00:00:22]
OK Couple people thank you there's good in all the audience so I know many of the students here are in the supply chain program so what I'll talk about today is not exactly 100 per cent in supply chain but we will have a lot of connections I'll try to point out those connections around possible some of the discussion or examples are also going to use on the lid thinks more broadly operations resource management science tools and even of the application may not be directly in supply chains hopefully.

[00:00:54]
We'll we'll be able to make some connections that and if you end up not liking the seminar very much please come back for the next fall because I promise the next one is going to be much more focused to solve lightings which may be a little belittling good with your interest but hopefully this one is also going to tell the some new things so I am the I'm A Think of the number in the School of Industrial Engineering and I also direct the Center for Health and Human it is systems it's an entire discipline that it is sort sounder and I call them Sister Sanders it is because we work together quite a bit with A C L in complementary and a little bit of overlapping areas so I structured this talk in 2 parts the 1st part is going to focus on some of the applications of operations resource management science and quantitative methods and in general in health systems and then in the 2nd part we are going to look at a humanitarian systems so it in a point in time I know this room is kind of.

[00:01:54]
Intimidating but please do raise your hand stop me and ask questions. We can make this interactive so don't be shy maybe to engage in discussions or questions as we go along and I don't have to finish all the slides either so we can stop in a word. So health systems.

[00:02:16]
We can think about different kinds of activities within the health system space some of these focus on prevention basically focused on treating people healthy and out of the hospitals and clinics and so on the 2nd stream focuses on screening and diagnosis despite our best efforts sometimes we cannot prevent disease in those cases it's important to be error of these early years so that we can actually take action and then of course there is the treatment which is the part that we all think about probably going to think about health systems it's more like the sick care versus health care in in large part but there's also a battle of activity happens especially from a logistics and supply chain perspective so the decisions that are made it is different stages they impact both the individual patients health but they also have an impact on the population overall and on a locate our limited resources and vice a versa maybe even be a look at our limited resources is also going to impact how we take actions or make decisions each of these different stages so and you can imagine that we have multiple decision makers multiple objectives just like in supply chains in these systems different players sometimes all the object is Mike not 100 percent alive aligned so we still have some of these in sound of alignment issues in health systems just like we have them in supply chains.

[00:03:46]
The activities take place in different locations so again it's important to think about what we do in terms of the resources or services we provide in each of these different locations so that we achieve a good outcome from an individual perspective and also from the population perspective as a whole so we have our standard has a little research in each of these different areas so today in the interest of time I will just share with you couple of examples.

[00:04:16]
And these are coming from from these different stages than you think about the health care overall but hopefully it will give you ideas about how we use these quantitative models in our dressing some of these questions or decisions are elderly and in health systems so we'll talk about vaccinations which is on the preventive side screening decisions and then also more of the to part with an example that comes from surgery OK So our 1st example is from prevented from Catch up scheduling port childhood vaccinations.

[00:04:52]
You may remember getting vaccinated at some point when you were younger this is the recommand the vaccine schedule that post every year by the Centers for Disease Control and Prevention and also the American Academy of Pediatrics basically saying here are the vaccines that feed a command for preventable deaths in preventable diseases and also the schedule for these vaccinations so some of them are multiple dozers that need to be taken at a particularly time and if everybody followed these are commanded vaccine schedule I would just skip to the next topic in my talk but it turns out in the U.S. About half of the children fall behind with their vaccinations by age 2 so it's a pretty large percentage of children who do not follow the schedule and once the child is behind then the health care provider needs to create a personalized schedule for their child considering that the child's age as well as the vaccination history so for those of you who do work or who are familiar with machines scheduling problems in manufacturing there are little parallels in this one and those problems but also some differences as well for example in terms of the gaps that to be a law between Max's success or vaccinations and so on and so forth but it is as actually a scheduling problem.

[00:06:16]
So and it's a pretty difficult problem that historically health care providers try to solve manually often while the child is in the room they kind of scramble to all these different rules and tried to come up with a schedule and and many a times these schedules were not accurate so they were graded say on a scale of one to 66 being the best date and one being the worst or I guess the the average grade would come up around $2.00 So the sketches they created were not accurate and again they are not to blame it's just a difficult problem to solve to solve manually so some of these visible to requirements I put up there I'm not going to go through them in general but again you will see several parallels here between the end of the constraints here and what they might see say in in Mission scheduling from manufacturing so the table in the back summarizes at the very high level some of these rules and then there is this book that is updated every 3 years which contains additional information about the constraints and and so on which all need to be taken into account so Recall that were dated with the C.D.C. in this particularly project and we developed a decision support tool that takes in a child's individual history as well as aid and creates in optimal and feasible of course catch up schedule for that child and the methodology that we use here for those of you who are to present a nice way if you will you'll.

[00:07:56]
Find these 2 honest they really are to be used dynamic programming we show some structural results on the dominance criteria which actually enable us to solve these dynamic programs relatively quickly because in reality they may take a very long time but thanks to those results we can solve the middle to the quickly and get these optimal results within within just a 2nd.

[00:08:20]
So this is just the screenshot of the decision support all again you enter the vaccination history and and the age and then you have the option here of choosing it within schedule Gerson accelerated schedule the routine schedule is going to be similar to the extent possible to the recommended schedule of course with some differences because the child is behind that is the accelerated schedule might push the times off some of the vaccine doses a little bit earlier in time since he's appalled but to make sure that we kind of get ahead of the schedule a little bit in case we don't see this child sometime soon that will be still make sure that they they get their vaccinations in a timely in a timely manner and then the output resembles again the recommended schedule so it's easy for parents and health care providers to interpret the outcome basically showing what are they minister those are those which ones are the kitchen up those in red here and which are the only time those are so in this particular example after this point in time the child is basically back to their regular schedule because they were able to catch up with their vaccinations so the truth is available at this site H.T.T.P. is like scheduler that ours is so please visit this if you have friends and family members with young children recommended to them it actually works for children from 0 to 18 years old so even if the children are not that young they may still need to get some oxidation and.

[00:09:56]
Any feedback suggestions again will be most welcome if anybody tries it it's been up on this website and we updated they reviewed it in collaboration with the C.D.C. is the vaccination schedule changes some time from one year to the next so we do some updates about more or less the overall structure remains the same so the goal with this is to.

[00:10:20]
Help improve the vaccination rates of course reduce the disease the number of children getting sick but also create more embarrassed about vaccinations in general because the site also has a lot of links that provide detailed information about the pros and cons and vaccination so we hope that overall it also contributes to like scenarios in the country so this was an example of how we use all our M.S. methods at the individual patient level for prevention so now let's go up to the public calculate all so think about the and tired population and again let's stay within the vaccination prevention space and talk a little bit about resource allocation.

[00:11:09]
So. This is one of the areas where I think there are a lot of commonalities between what we do in house systems and what we do in supply chains in the private sector so we have certain goals that we want to achieve and we've limited resources and the question is how do we are located these remitted resources geographically and over time considering different stop elations in that area so that we achieve these goals so in sulfite chains the goals might be we have demand and we want to meet the demand as best as possible to maximize our profits So here's kind of slightly the other side of the of the coin Ron we think about the infectious diseases here you actually want to slow down the spread of the infectious diseases so it's kind of like the other side of the of the Medellin but the simple ideas are quite similar so in the infectious disease space the 1st of all of models that will help us understand the spread of the disease directly and over time and sometimes also across different age groups if you do nothing so like if you do no intervention how will this disease take its course and spread across a population.

[00:12:26]
And some of these models have some similarities to say models and marketing that it will start slow it will peak and then eventually it will die down after everybody gets sick there's nobody to infect on this it's kind of a repeated infection and then we look at various interventions strategies it can be vaccination it can be this traditional anti-virals it can be.

[00:12:50]
Educating the public so that they can take Stuart and measures to slow down the spread of the disease through their actions and so on so does interventions the goal here again we have limited resources for these interactions What is the best way to locate them geographic and over time.

[00:13:08]
To keep the disease spread as small as possible and not just for the make it go away and then we may also other reasonable occasion issues which are more along the lines of supporting So for example food suppose we have a limited availability of food because there's a big blue pandemic.

[00:13:29]
Grocery stores close and then we may need to distribute food that's not going to impact the self discipline of the disease directly but will impact how we support the population in such a situation if. History. Is. So. Mental dollars It really depends on the particular disease we are studying as well as what we try to do so I put a couple of examples here pandemic flu cholera Guinea missiles malaria so we have done quite a bit of work in the pandemic flu space so that we use a fairly detailed agent based in relation model modeling every person isn't in it as an agent in the simulation and also capturing the dynamics of the disease progression within the individual over time so you get the virus you may or may not develop symptoms and then you kind of start slowly recovering and doing this and die a period you might actually infect someone if you interact with them.

[00:14:34]
So we capture the progression within the individual but we also capture the interactions for example if you are in the same room together that causes the possibility of something spreading if you are living with people in the same household 10000000 members then the chances of spirit is higher and so on and so forth so this is kind of like the big picture framework that we use in this infectious disease modeling space developing these epidemiology models that help us.

[00:15:05]
Capture understand the spread of the disease and then combining those models with disease or so location optimization models for that we can simultaneously evaluate if we do this how is the spread going to continue or or slow down so these are some of the examples from that space and if anybody is interested in any of these and maybe to share.

[00:15:26]
Papers or other other information based on our block and others and today in the interest of time I'm only going to discuss one of these which is the spread of cholera this was the joint work with the. With the Task Force for Global Health How many of you have heard of Task Force or global health unit where the OK a couple of you so it's a it's a large pretty large actually a nonprofit organization it's based in Atlanta most of the work that they do is in Africa so that's probably why some of you may not have heard about them but submissive pretty amazing organization with significant impact in the public health space.

[00:16:05]
So you don't hear or think about cholera much in the United States but it is still a pretty significant issue in other parts of the world and effecting many people causing that's just effecting the quality of life even if it doesn't kill the patients so it is treatable but sometimes access to treatment might not be available and the consensus in the public out space is that the the ultimate way to address cholera is to improve the infrastructure the water sanitation infrastructure because you can see the pathways for the for the spread of cholera it's basically not having clean water and sanitation contributes to the spread of the disease so if we can fix the infrastructure we basically for the most part take care of color but fixing the infrastructure everywhere is not necessarily feasible in the short term so we also look at some other interventions especially in case of an operation so the specific intervention that we focused on is that's a nation it's called the oral color act scene and we would like to understand how to best a look at this vaccine again to juggle if we can over time so that we can keep the spread of the disease in check so that's the question that we address so a little bit off that this think.

[00:17:33]
About cholera so what you'll see in this graph in a nutshell is that depending on the age the impact is different and it affects younger children more in terms of number of infections or the number that's compared to all the people and this is in part because of the younger children you know they put their hands into something and then they put them into the mouth and so on and so forth and also their immune system is not treatable develop so we see here disproportionately impact on younger children.

[00:18:05]
And then the colors of the bars also indicate different regions so we see different levels of prevalence of color in different regions around the world and sometimes even within the same country you might have prevalence levels that are different depending on the infrastructure water sanitation infrastructure as well is the availability or access to health care resources for treatment so if you see age based differences we see job.

[00:18:33]
Differences So this is just showing. Some examples of Harvie might see differences depending also the physical structure. Water sources and so on and so forth to be then within a particularly country so again our focus is on vaccination and when we did this work couple of years ago there was really no systematic approach out there for a locating this oral color like scene considering age and region as are the other other aspects so the methodology that we use for this is an internship program so we incorporated some of these disease dynamics into that model and we also incorporated the fact that.

[00:19:23]
Population age whoever it is in the population it changes so for example if we vaccinate a 2 year old it has an impact today and then that 2 year old is going to grow and the impact of that vaccination continues for up to 5 years that not like some of the other vaccines that protect your lifetime it protects you for up to 5 years but the impact or the protection level goes down over time so we incorporated a lot of these dynamics about the population new.

[00:19:59]
Children being born and added to the population children growing up and then also the vaccination impact decreasing over time and then we created a multi-year model to see the impact of different vaccination strategies so when it comes to vaccine a location or the location of other health care resources such as anti-virals and so on one certain age that we see.

[00:20:25]
Very frequently is called the population based or pro-rata strategies So basically you look at your limited resources proportional to the population in a particularly area this strategy is used for example going to have a shortage in the flu vaccine with it it's the seasonal flu or it may be a pandemic flu typically at the beginning of the season it's limited that's how it's.

[00:20:48]
Located to different to different regions so while this science fair in some ways. It doesn't always give us the best results in terms of keeping the disease and shake from the perspective of the entire population so. Here we test a different strategies what would be a locate by region by a by population proportion as one and so forth basically what we find is that if we simultaneously consider the region and the age with a with a particularly algorithm prioritizing we actually get the best results in terms of the number of cases prevented significant difference between some of these other searches that have been historically used in practice so what did we do about this result of the previous.

[00:21:48]
Topic I discussed with you the kids are scheduling pour for Rex in a sign that we have a decision support all that's out there are the parents and physicians have been using extensively so that broken turn into a decision support all that's used by individuals so what we have here we don't have a decision support all we have some recommendations about how to best all look at these resources.

[00:22:15]
Considering age under the engine so we will be published our work and we also contribute to the writing of a report a comprehensive integrate the strategy for color Prevention and Control which we came a guideline and we wrote. Our team wrote a particularly section of the reports specifically focusing on logistics top flight in aspects our new source the location in that space and then we also presented our results.

[00:22:47]
At the. Color. Meeting at the nation is to do so palette so that we can actually share these results with the decision makers so these are just some examples of how the resource that we do hear it through to take we try to share it with with others to an impact either it may turn into a decision support tool or it may turn into recommendations that we shared with the decision makers to inform their decisions and support their decisions.

[00:23:21]
So again I imagine couple other examples is one that we are currently working on we have actually a classroom game based on malaria and how we might all locate the limited resources for indoor is a jewel spraying we play that game in several of our classes it's Excel based and really fun and students get quite competitive they divide into subthemes and tried to come up with the best strategy but the idea is that again similar that we have limited resources we need to have trucks and equipment then chemicals and whatnot and the spread of malaria depends on the season depends on the location so how do we best to look at these new sources to again minimize the number of infections so we are currently collaborating with the C.D.C. on evaluating the impact of any interactions that you called product so this one is proactively visiting villages especially those villages which do not have easy access to health care facilities and testing people who are exhibiting malaria symptoms rather than waiting for them to show up at a health care facility which may be difficult from a logistics perspective so we are looking at the impact of this new intervention strategy and comparing with others that have been in place for some time so I'll read a tall tall deal with about the growth that we have done in pandemic flu looking at different aspects so we have one paper that got published recently specifically looking at the location of inventory again job if we can over time also considering the uptake rates or the demand for vaccines in different regions and one other called Operation B. currently is.

[00:25:08]
Is with the Carter Center. Looking at the modeling of the guinea worm disease which is which is again one of those you know diseases that is still on force and the prevalent in Africa even though the numbers have been decreasing so they started seeing an uptick of this disease especially in dogs in recent years and the human cases the hours of human cases have also been increasing so we have been collaborating with Again understanding how the combination of different interventions 30 G.'s and all of them into the resources might help us keep this disease in check and eventually hopefully eradicated that's been their goal because well these are just some examples which I hope give you an idea about how these supply chain ideas are used in the public health space with a slightly different goal looking at the other side of the modality not keeping things in check versus you know reaching as many people as possible but similar ideas OK so now let's switch gears a little tiff you've been sleeping up on to this point so you can wake up because I'm going to start producing a different topic so we talk about prevention through vaccinations and interventions and so on so now let's talk about screening so.

[00:26:25]
In screening the idea is that. We administer some tests they could be blood tests they could be scanned maybe some combination and then based on the results of those tests the patient or the individual they are not necessarily sick the individual might be labeled as high risk or a particularly disease or low risk if the test results are just high risk which often depends on some sort of threshold that we use so here's your number and if you are above the threshold you Mark is high risk and if you are below the threshold the market is Lloris so that if you are marked as higher risk than the physician might suggest some follow up action items some of these might include invasive procedures so then there's the tradeoff because these tests are not perfect baby combine the results of these different tests and the baby choose this threshold is going to impact with that if you have how many false positives or or false negatives you might have so we don't always get 100 percent accurate result so the general goal here is can we design some of these tests in a way that minimizes these false positives and false negatives and also informs the physicians and the patients in these decisions when there is a negative or a positive result how to make the decision about whether or not to go for an inmate's of potentially Maysa procedure as a next step so fairly complicated decisions and again for the most part done in the absence of any quantitative decision support but based on the values preferences of the of the patients as good as the experience of the of the positions so this is an area where you tried to bring in some again systematic approach on quantitative methods.

[00:28:27]
So we've worked in this space on several different diseases some of some of them focus on the individual level so here's the patient and what do we do for that particularly patient and some of these decisions also focused on the population level so for example how often should we screen Think about the recommendations how we should be screened certain segments of the population for this particular disease maybe you say everybody after age 50 screened on monsignor's So these kinds of recommendations typically use decision support tools or models and methods that come from R.M.S..

[00:29:06]
They could be at that level or they could be at the individual individual level so the example I'm going to discuss today comes from. Prenatal screening for for Down Syndrome I'll skip through this quickly just again in the interest of time so we can go to the 2nd part but essentially this this is this is a condition that affects.

[00:29:28]
Again. A significant number of pregnancies It doesn't matter whether we are in the developed world or developing world it's a promise almost disease and there are some tests out there but but most of the tests that are widely used are not again 100 percent accurate so we looked at the present prenatal integrated screen this is the most commonly used test that's available there are some better ones available now but are only used for certain segments of the population and again the goal is to understand the the polls negatives and the false positives and then see depending on the age group and so on these results are affected so one of the things that we found out is that the cult of value I mentioned to you there's a threshold that's used it's a single trace hold regardless of age results in very different outcomes for different age groups so in particularly.

[00:30:27]
The post positive rate. Is is significantly higher for all the women along with the detection rate so this is one thing to keep in mind that there is not a good balance in terms of the false positives and false pill pas Paul's negatives across the age groups with there's a significant variability so we.

[00:30:51]
Looked at alternative threshold value still staying with the single threshold framework and we found that there are some alternative thresholds that would actually minimize the overall number of adverse outcomes when we think about the entire population so that's one thing to consider why use one or 270 given today's population if you could choose a different threshold that could actually overall give us better result.

[00:31:18]
And other thing to consider is why I consider only a simple threshold and why not consider 8 specific to shows because that will clearly give us a better segmentation and better outcomes in terms of. False positives and false negatives so again using Monte Carlo simulation and truth separate they can say it's of clinical trials we're all able to show that if we use 8 specific cutoff even if it's not for the individual age but say for age groups so maybe we use one cutoff for this age group another one for this age group and another one for this age group so we go let's say from one total to 3 or 4 cut offs we can still significantly improve the overall.

[00:32:02]
A curacy in the sense that minimizing some of these adverse outcomes and then finally. We know that people have very different preferences about the kinds of action they might want to take depending on the on the test result so instead of just telling them positive versus negative can begin them some more information about where they are so that they can make a more informed decision considering their own preferences about potential next steps yes I will do an invasive test or not I'm not going to write so that's kind of the main question that that is face by by everyone and family is that the state so towards that and we have a prototype a decision support tool.

[00:32:51]
Actually asks the juicer to rate to 1st rank and then rate these different outcomes that are possible so for example having a healthy baby getting you know no problem would be ranked at the top for almost everybody with a high rating and then some of these other outcomes depending on the on the personal or payment preferences might differ from family to family so we asked them to a town rank these outcomes and then depending on the ranking and ratings that they put in and their test results and the age.

[00:33:28]
We are quote it particularly. Point on the risk you scale but it's not a yes or no answer 0 or one it's something along the spectrum based on what they input and then this gives them an idea about where they are currently and if for some reason what they see doesn't seem to align with what they have in mind or what they feel they can actually go back and revisit the way they entered their preferences the ranking and ratings and then they can see again how the results change so did this gives them a lot more information compared to just 01 high risk versus not high risk rowdy but it actually enables them to better as Says their own preferences and hopefully make a decision that is more in line with their physical health is that is their values so this is currently at the prototype stage we had one student last year who spent the summer at the Mayo Clinic and we worked with the positions that are to improve this and they are currently.

[00:34:34]
In the process of finalizing the steps for a pilot with their physicians and patients so we are really excited about this hopefully the pilot will start. In the coming months and then you'll get a bit more feedback directly from the field so this one from a mental picture of you know combines again simulation and some data analysis isn't nice and from these 2 large critical critical trials.

[00:35:05]
For us I'm going to pause here for just a 2nd before I go to my next topics any any questions any comments so far. OK so if you could to stop me at any point in time so the next example is organ transplant decisions for patients and physicians or this is in the treatment stage so we discussed prevention we discussed screening so now we are going to discuss an example that's on treatment so this is a pretty large team of take a chance to dance and also school overeaters from C.D.C. and more recently Emory University we've been working on this on this project for the past couple years very very exciting.

[00:35:48]
So the motivation comes from this gap growing gap between supply and demand for organ transplants so if you see here these are the number of patients on the wait list waiting for an organ transplant and then here you can see the number of. Organs recovered or the number of transplants performed and while our supply has been relatively flat our demand has been going up so we see this increasing gap between supply and demand so if the same time.

[00:36:22]
Is a significant percentage of organs reach were deemed appropriate for transplant or originally are actually discarded so the way the process works when an organ becomes available a deceased donor organ becomes available for transplant there are some algorithms out there are. Adapted by you know asked and these organizations.

[00:36:51]
For that particularly organ they will create a priority list of patients and then the organ is offered starting from the top of the list and if that patient and their physician decline we'll go to the next patient we'll go to the next patient and so on so if one of those patients accepts they undergo transplant if everything matches and if we keep going down the clock is ticking the time is passing these organs can stay viable for transplant only for a certain amount of time and if we keep hearing no no no no no the clock is ticking and really each the end of the call this time which means at that point we can no longer transplant this organ so this happens more often than you might think and actually leads to a large number of organs being discarded so one of the main reasons for this is that.

[00:37:45]
The cord quality of the organ. Really depends on the donor on the donor characteristics and the higher the quality the higher the chances of a successful transplant and a healthy life post-transplant at least if you look at the next 5 years or so so if the patient and the position drawing to the PERCY The quality is not been as good they might instead prefer to remain on the raid list until a better quality organ becomes available in the future so that's the tradeoff faced by the patients and physicians so when an organist offered we might accept.

[00:38:27]
Undergo transplant to understand the implications of this decision we need to have some idea about our post-transplant chances of survival not just within one week but maybe 3 months one year 5 years and so on imagine all these calculations are happening in your head if we decline that we need to estimate our reading the survival the future man on the waitlist let's say for another year we don't know when the next organ is going to come so we need to have some idea about Iran we might receive the next offer and have some idea about the kind of organ might be offered to us next time and then if you undergo transplanted to future time with a different type of organ the survival chances that time OK So typically the decision of except worse is decline it is the decline decision the physician has the 1st say when an offer comes in say 2 am in the morning it might have been you pick up the phone as a physician if you decide this is not a good one you may decline and you may not even discuss with your patient if you think that the organ might be a good one then you check with the patient and the patient is also OK You you move forward with the with the transplant but imagine all of this is happening in your head you're a physician and you need to juggle all this information about your patient characteristics your organ characteristics and so on and so forth and you and that's how you make that decision little That's how the decision is made so again our goal here was to put some quantitative methods primarily using machine learning methods around this.

[00:40:07]
And develop a decision support all especially also considering that some of these organs are labeled as increased risk about 20 percent of these organs they've been labeled increased risk for example in 2014 this adds one more layer of complication so the increase is donor organs they are tested for H I really A.T.V. H.P.V. the test result comes out negative but remember we said there are not 100 percent accurate there's a chance that this person may have.

[00:40:41]
May have gotten this disease they say over the past 3 days and it's also one that is not showing up in the test yet but we may still be concerned because of certain things you might know about this patients cause of death or lifestyle and so on so that's how the some of these organs are labeled increasers So there is really nothing showing up in the test but there's a very small however small possible to time that might be infection so that hurdle complicates the decision as to you know the quality looks great but how do we decide given that we know there is increased risk so we really see data a fairly large dataset from you know us that contains every single organ transplant performed in the United States over the past 20 something years and we visibly utilize that data set to develop these predictive models.

[00:41:33]
To answer various questions and most of these are again like I said machine learning models so I'm just giving you a quick overview of the kinds of things that we have done for example for to give one organ heart liver kidney which are where it might be we enter the donor characteristics the under the patient characteristics and the decisions support all gives us the store viable curves for these different options so here we are again every decision support all I know it's hard to see but essentially it's collecting information about the patients than in Oregon.

[00:42:08]
The decision tree that I showed you earlier we have this offer in 100 now how do how do we decide considering the the future and then it all out put the survival course for the options of. Receive this organ right now and undergo transplant this is how your survival looks like or we're over the future they.

[00:42:32]
Wait for another organ so maybe you wait and you estimate it about $800.00 days you might receive another offer and if you are the go transplant at that point this is how your survival car would like it would look like so now we actually put in all of these thousands of data points into this mathematical model and instead of you doing all this calculation in your head we can actually quantify and show you what the potential implications of these different options are and of course we test these models on a lot of data for accuracy and so on so where are we with this one we published some papers already on this work prototype this is a support tool we recently this work is being supported by 2 grounds by Arnold foundation and missing trust fund they said we are grateful to them for their support and recently received the smaller seed ground from hips that you'll actually start piloting this with Emory but we all are there is even formal feedback from several positions and improved a little along the way so hopefully this will also move forward then be used more more widely in practice OK so I'm going to skip all of this and briefly tell you.

[00:43:52]
About another project this is on the. Public health space but this time instead of a locating physical resources such as vaccines and so on we are a lucky think human resources rich are also an important part of our supply chains are there in the public health space or in the.

[00:44:15]
Private sector space so this is in collaboration with the C.D.C. Task Force for Global Health and the Mozambique Ministry of Health. So here is our problem Mozambique is a country where our health care resources are very very limited and hence a location of these resources to best match the demand are very important in this environment so when health care workers whether they are nurses or doctors or other cadre Rand they finish school if they would like to work for a government post they are assigned to particularly by the government and these assignments are made considering both the demand and the supply so the demand for different types of workers in different areas and also who is already there is released who is graduating this year and then an assignment is made so historically this assignments were made primarily considering the demand and supply but not considering the preferences of the healthcare workers who just graduate from school and clearly these are people not Regents we tell them to go to particularly.

[00:45:27]
They might go but if that location is really not ideal for them not preferrable or them they ask for relocation or some of them actually do not show up to work and choose to work for 4 other organizations which of course again negatively impact the overall health system so in our collaboration with we.

[00:45:55]
It's likely to change the process or the input collection mechanism so that we actually call it information about the graduates some in some random order about where they would prefer to go and then we develop an optimization model to create this best location considering different constraints and preferences than object of that and so on.

[00:46:18]
And you can see a little bit of a screenshot I know it's too small to see but the results have been really positive the Mozambique Ministry of Health has been using this decision support all for the past 2 almost 3 years now and I also feel a bit of the results what we've achieved in terms of results so this is how they are a location look like prior to using the tool So the areas where you see red are areas where actual They are standing more people than needed to those 8 years for one reason or another and then there are other areas to reach our color light wired we are really really low in terms of meeting demand so you see quite a bit of imbalance across the country when you look at these different districts or regions and not doing a great job so now we take similar import but we are able to create in a location that is actually significantly better and more balanced.

[00:47:14]
Across the entire country so they've been very happy. With the tool and also the people that they've received from the health care workers has been quite positive that the least this input that they can provide and the majority majority of them were located to a location that was within their top 3 top 3 choices so they were able to create a bill a match between supply and demand without actually incurring extra cost or requiring X. $30.00 sources in this particular example.

[00:47:48]
OK so I'm going to skip. And briefly talk about humanitarian systems in it before I go to the next part any questions about the health systems I know there was a lot of information but any questions and comments yes. Mozambique Ministry of Health Yes So the Mozambique Ministry of Health is the one who wanted to have a bit of a way to look at disease sources so they were already called overrating on other aspects of health systems with the C.D.C. on the Task Force for Global Health and then this Georgia take came into the picture to help them with this particularly piece of optimized resource allocation of health care workers so that project or isn't this started as a class project I did master's level project class in the spring so we started the corporation through that project and then after the project under the.

[00:48:54]
We actually the continued working with them over. Teleconference a little bit looking at the screen and getting feedback and doing some training and so on to develop and finalize this in support of but we were able to bring it to a point of being actually used in practice.

[00:49:14]
Other questions come. Yet. Yes that this innocent person is of the question is Could our algorithm go through the list and actually repeat the top 10 and 20 people who are more likely to benefit from this organ so take me to yes. The current algorithm this used by you not in a way does that.

[00:49:54]
And I mean they they also you know amazing people always even working on this for some time so they typically do not change that algorithm merely often so it's very difficult to come up with something that's going to be much much better than what they do and impacting policy at that level is not that easy.

[00:50:17]
But the the current algorithms all do the talk to a large extent do that and they create create that prioritized list for a specific organ so again when we think about the priority list it's not a static list so every time an organ becomes available for example if you consider the location and patients who are religious 3rd in that region they jump higher are patients who are younger usually are ranked higher he saw again depending on the match but it really looks at the combination of this particularly organ house of the prior to this looked like for this particularly organ so it's created specifically for that organ that's what the current algorithms value most.

[00:50:58]
Or. Nothing's likely actually so that's that was part of the issue with the mismatch before there are some people who actually want to go to some of the smaller areas because they have families that are and they prefer to go so some of those people with the previous matching algorithm may have ended up in a big city there is somebody who want to be in a big city may end up in a similar place so so busy could buy eliciting these preferences could you can do a better match compared to what you could do otherwise.

[00:51:57]
And again if you if you look at the numbers that are the person there are people who are able to go to their top 3 choice is actually pretty high so I mean I agree with you the army to some preference about the big cities but it's not across the board.

[00:52:16]
Other question comment yes. So it will 1st and foremost look at the demand supply match that's kind of the primary criteria and then as a secondary criteria it looks at the if. It's the preferences so what we said just a to the to the government is to. Keep keep a record of these people who end up not going to their top 3 choices and then when they receive relocation requests for example they know that this person was not assigned to their top 3 choices now they could prioritize the people who didn't end up in their top 3 choice this was also something that wasn't possible before because they didn't even know and that's a way to prioritize the real location requests and if you're already in your top choice for well there were reason if you have to move you say after 5 years than Again this is good information where you were in your number one choice for a while now story about we need to move you elsewhere because they also do some moment of the of the existing workers sometimes not as much as 2 years so that's a good question.

[00:53:46]
So this one is the badge because they graduate about the same time and then they just see their assignments but then you can envision some of our Media years if there are a lot of changes moments and whatnot they could still do it or they could do it every time there's like large enough batch that graduates but not at the individual level.

[00:54:30]
So the the match so what what he's saying is that we have 90000 people in the weightlessness a 4 kidney but then we have still thousands of kidneys being discarded so for the most part the reason they are discarded and again so when we talk about this card so there are some that simply are not liable for transplant we are not talking about those because they are there out of the system anyway so the ones that are viable for transplant the technical crew they've been transplanted medically speaking still are not accepted by anybody on the list within the limited amount of time and then eventually they are discarded so that in large part is happening because of this concern that the quality is not good and for kidney here in this country we have dialysis so then you evaluate again without the support of a quantitative tool you think you know I'm on there also I can stay on their list as for a couple more months no problem but what is not that is the 2 like recognize with our limited human brain is that your health is deteriorating while you are in the aisles is your quality of life is is not as great so what we find actually with with this work is this increase risk donor organs as an example they are discarded at a higher rate than.

[00:55:54]
Those they are that are not increased risk and they typically come from donors who are younger so Dorgan quality quote unquote tends to be better on average. And yet there are discarded at the higher rates so our work basically shows that many of these organs that are currently discarded could be lifesaving.

[00:56:13]
So we have some again. Results that in terms of numbers if you transplant the you know X. of these you know what what the result could be not only in terms of saving the lives for the people who receive transplant but also clearly you can envision you are taking some people out of the Riddler So now it's also easier to essentially run increasing the supply utilized supply Yeah.

[00:56:49]
So the algorithm takes as input an individual or or donor donor characteristics and particularly patients characteristics who is being offered that organ and then for that specific pair it generates this survival course if you say yes right now what happens what are the you know what's the survival Caruthers survival career look like and then if you say no than what the survival career look like while you're on the wait list and if you receive a transplant say later with an average organ.

[00:57:32]
I mean it's again it's soon that you. You respond within the. Feasible time for that organ to be still viable outside of the time limit of course you can do it anyway but this in a way to help also with these decisions because typically you the physicians the patient joins the mayor up to an hour let's say to respond and again it's very difficult to make that decision just in your head.

[00:58:21]
Yet there they do work out of time so the they give up to an hour but many times they may receive an answer much sooner some physicians actually put into the system for example they may say if the organ has the following characteristics not automated so in that case it will automatically jump bypass and jump go to the next person so for some organs which are not of high quality because of these rules put into the system by physicians for that specific patient we may actually bypass Iowa 1000 patients and then go down so sometimes you see some organs was actually one of the things that we are looking at they may go to someone who is like 10000 in the list and then others go to someone who is in the top on top 100 so it's really very it and up and up.

[00:59:35]
So the 2 cases why organs get this is kind of the same because each each person has a limited amount of time during which they need to respond the problem is that if too many people say no. We keep going down right on and every time we go to a new person we give them about an hour to think about even if they respond in 15 minutes and eventually say no the clock is still ticking so it's kind of the same so nobody gets on the meter the amount of time to give an answer everybody has a limited amount of time the problems of 2 men in those even if they respond quickly still time passes.

[01:00:16]
So this is especially for increased risk organs because you can see quantitatively that they're actually pretty good from a survival perspective if they take that into consideration yes you expect less of these special interest donor organs less of them being discarded because they come to be again on Everett the quality than the other once consistent with donors age and so on.

[01:00:54]
Yes. Yes So the question does the mother consider the location so the model just assumes that the match is been made and the algorithm by your nose or B.T.N. that does the match it does consider the location so if the organ is offered to a particularly patient most likely they are.

[01:01:14]
In the same vicinity. Only if we go down down down for March down the list and nobody in the region wants this organ at that point it's over to some other people that are maybe a little bit farther away considering is that enough time. It is different for every organ actually right so I mean the vigilance is not different so you are on the wait list it's one big wait list but the prioritized list for a given organ who is the top priority patient for this organ was the 2nd approach that is customized for that specific organ at the time.

[01:01:57]
So the weight is the way the rate is changes is I mean people get added to the wait list and then they are inserted some are for human organ based on their condition and location or people leave the wait list they may die they may get too sick to the transplant they may have a transplant so the biggest itself changes dynamically but the sort of data list for a given organ it's for a given organ so every time a new organ becomes available we create a new sort of weightless specific for that organ if that makes sense.

[01:02:51]
So the. Every time an organ becomes available there is a new sorting there is a new sorting because the pans OK so where is the donor there are the patients so people who are nearby are going to get to the top of the list for death particularly Orgon.

[01:03:12]
The age of the donor the age of the recipients that impacts the condition of the people in the waitlist So somebody maybe was doing great is today today suddenly their health condition deteriorated maybe if we received this new organ yesterday they would have been here in the weightless clotted A They are higher up if that makes sense so it's really we receive a new organ becomes available and then the prioritized list is created specifically for that organ at that time.

[01:03:43]
Think again. So the the the very time of the patients factors into it how long they've been in the wait list their current condition age all of these factors into the into the prioritization Yeah. OK that's an excellent question so suppose there is a patient who is more or less always somebody to top of the wait list currently in the sort of greatness and then they keep saying no no no no no do they get penalize in any way the answer is no they don't so if.

[01:04:29]
Right but there is no penalty for that so if if you are you know if you are in a critical condition but still able to wait a little bit you know that most likely when the next organ becomes available you will still be high up in the weightless so it matches be in your advantage to say no and wait for a bill or organ and you are not penalize for that currently Yeah let's.

[01:04:53]
Just say. Thank you. So. So this is a very common so let me maybe tell you just one slide if you have time about some of the things that we do in the humanitarian system space and also in going to put some information here so. This is some information about our center some of the workers we do and this is a law there conference that we are called overeating and supporting just the Sears I'll put that information here that's going to be here in Atlanta so in the humanitarian side we do projects in 2 streams one stream focuses on nature land manmade disasters there are many of them and unfortunately not slowing down or going away and then the other of course is these ongoing we call them development problems of food water shelter and so on and so forth and the environments year are multiple decision makers there's uncertainty there are destructions So in a lot of ways you see the similarities between what's happening here in the space and also what their plans typically in chains so as I mentioned different players the owners the recipient agency which may be different than the agency that actually does the delivery in terms of delivering the aid and then of course we have the demand which is the people in need and we also sometimes see these other players that impact you know what how certain decisions are made not always necessarily in the best interests of the of the recipients or the system but the impact of some of these decisions like the media.

[01:06:52]
So whether we think this is a disaster response situation or we are thinking about an ongoing development for example establishing a water and sanitation infrastructure. We need some assessment what do we need where how much which changes over time we need to mobilize resources people resources financial resources physical resources transport ation and of course distribution for.

[01:07:18]
And all these different decision makers so you can see these are all fairly challenging Buddhistic supply chain problems in a slightly different context than what you might currently look at in the in the private sector so very briefly in terms of disasters re think about disasters in 3 stages or 4 so we have the preparedness or mitigation stage there are several activities that can be done that are so that we actually minimize the impact that we see after the disaster there is the immediate response phase after something happens for example a hurricane it's a particularly Rhea we respond to help the population in that area and then comes the post disaster recovery which kind of loops back into this mitigation and preparedness because it's really covered we don't necessarily really build exactly the same way that we had before so we kind of consider the future in mind as well so I won't go into the details here but we have done work in this stage for example prepositioning inventory how do we preposition different types of in MT Henri in different areas around the world or in a particularly region and how do we mobilize that in one Tory post disaster to have the best impact on the population of you've done some work in the space I mean of course there's quite a bit of work in this area as well about here.

[01:08:44]
Maybe also across the intersection they really management is an area where we did a lot of work post disaster they re management. Yard it actually is working on a project that's motivated by that So lots of the reasons related after a disaster and it impacts people leaving the area trying to escape the area it impacts how we deliver goods and services into the area so again given our limited resources resources for clearance how do we allocate those resources to best.

[01:09:18]
Clear out this debris again all the time and geographically to see the best impact so those are some of the areas that you wrote on and again I'm a bit the child off line I know we are pretty much at the end of our time but. It was there on the Internet if there is interest Thank you thank you very much.