No, she needs the mouse to click for her. And that's good. Okay, okay, good. So, hi everyone. My name is Renata, or Renata. I am a PhD student in theoretical neuroscience at Columbia University in the city of New York. And I'm here to tell you a little bit about what we did in the Datathon in team eight with these amazing people, Si Cheng Hao, Mona Patel, Jennifer Lee, Mya Sporn, Keegan Yap, and Jamie Sturgill. And, how do I move this? Okay, so first of all we none of us have any conflicts of interest for this presentation and our team was assigned to the COVID-19 track and the initial question that we wanted to address was do COVID-19 directed therapies have similar outcomes in patients of different ethnicities and the reason behind this question is that it has been documented several disparities racial and ethnic disparities in several instances during the pandemic including higher rates of hospitalization admission to the ICU and mortality and that have been documented in black and Latino patients and so when we look at directed therapies for COVID-19 we have roughly pharmacological and known pharmacological therapies and normally pharmacological therapies like have a larger body of research on them so we started looking more on the non-pharmacological therapies and we found out that pruning is largely understood under under investigated because of the fact that pruning is not an easily trackable variable because you you don't have an order for pruning you don't have anything that ends up documented so if you don't ask professionals specifically to record that information you're not gonna have this information so it's not use of usual to have the information about pruning on databases and we had it on the virus registry so we decided to focus on pruning and also during a data thon you don't really have much time to do your research you normally have one day and a half to decide what you're gonna the question that you want to address decide how you want to address it which data you're gonna use clean your data select your core prepare your data analyze it come interpret it and make a final presentation so that's why we decided to focus on something specific so that we could do something new and contribute to the literature a little bit so we know that pruning works but what we don't know is who's being pruned and if pruning works for everyone the same way right so our questions just narrowing down to pruning we decided to focus on these two questions which was was there a difference in frequency of pruning based on race and ethnicity and question two did this have an effect on the outcomes like different pruning did pruning have a different effect a different outcome in different in patients of different ethnicities so as I mentioned we use it the virus COVID-19 registry which encompasses over 21,000 entries of ICU admissions over 21 28 countries and from these 21,000 entries we selected patients that were over 18 and that were under invasive ventilation we also decided to exclude pregnant subjects due to major differences in how COVID affects these group and also we excluded patients involved in relevant eyes Eric trials because we could not know who was under medication who was a control so we had no way to account for the effect of this potential medications and also finally we had to exclude a great number of subjects due to missingness of data of the variables that we were interested in which were first day sofa hospital discharge information missing age and race so looking at the demographics of our cohort we had a large majority of white people and also the vast vast majority of our data which ended up in 9000 patients was from the Americas and specifically from the United States so the demographics of global of the full cohort including the whole world and of America are largely very similar with the exception of a larger proportion of black patients in the American core and a smaller patient proportion of Asian patients and of this final cohort the global one how many were prone we found that we have around a third of our patients in this 9000 cohort were pruned and if we look at the demographics of the patients that were prone it does look reasonably similar but if we do a logistic regression adjusting for the severity which we use the first day of sofa age and sex we saw that black patients were prone less than the rest and interestingly we also saw that females were prone less in the global context and we also decided to look only in America and in America we don't see this difference like patients are not significantly prone lesser actually a little bit more even though the effect here is not significant so but females are still prone less and this might be due to the fact that the probability of a patient being impruned and centers outside of the United States is higher and also when we look at the probability of hospital mortality we did not see especially and when we look at overall prone we did not see an effect of brony significant in decreasing mortality this is the data from from I'm not saying that it doesn't work but this is what we saw in the data and also how can I go back okay it's not that important I was just gonna say that from that graph we can see that there were no differences also between patients of different ethnicities and this is a very preliminary study that we did in one day and a half and of course it's very limited also to the to the selectiveness of the hospitals that could participate in a virus registry and also a problem that we had was lack of detailed information from where the data come from for example we knew we only had the information of continent level so we only had America's or Asia or we knew from from the number of centers that the majority came from the United States but would be very good to be able to narrow down from country and from hospital and maybe even from state and I think the major problem also was data missingness because we our initial cohort was much smaller than the number of entries in the data set due to missingness of data just going back to our main conclusions we saw that there are differences in the incidence of proning based on race globally but not in America and we saw that in general female patients were prone less in America and in the world and that race had no impact on mortality in patients who were prone in America and in the world as well so as I said this is a preliminary study but it suggests how much prone is still largely under documented and requires more study and understanding and we do plan to submit this as a manuscript and we're very glad to have met each other on the Datathon it was an amazing experience and thank you especially to all this amazing team to Ian and to everybody that made us Datathon possible and made it possible for us to be telling you this here today thank you thank you I think we'll try to have questions at the end thank you so much and so our next speaker is going to be Tiago from the University of Porto also with an internship over at MIT with a limited English does limited English proficiency affect time to death for critically ill patients who expire in the hospital this is was in our health equity track so okay so can you guys hear me I guess so so once again thank you Ian and SCCM and all the guys who worked with me at Datathon it was a pleasure to be in New York last August and to actually be here to present you some of our findings just a disclaimer all 99% of these slides all the work here where it was achieved like in less 24 hours so please as soon as you are going to make questions please be aware to you get like 24 hours to do this and we had like a very straight like moderator as Ian because you went we really had to be on time right so my name is Tiago RT is right there and all the other members actually worked tirelessly to actually achieve this work okay so before going into the hypothesis in the work let me give you like some context on on why did we decide like with this topic and theme so we know that there is some impact of of the English proficiency in health care which means that if you are native or not probably the outcomes of health care will be different another thing I'm not a clinician I'm an engineer so once again don't don't get out on me if I if I say something wrong but we know that like English proficiency will have impact for instance there are some language barriers it can actually lead to longer length lengths of stay it can actually increase the odds of more visits to the hospital and in the emergency department and of course on the other hand that could be also like different survival outcomes for both like native speakers or non-native speakers so this was the main context of course there are a lot lots of studies that you can guys can search on but in five minutes I will not like I will not tell you all of them and we all know like as a summary of all of that we know that language will affect health care in diverse ways once again it may lead to misdiagnosis or diagnostic testing or poor quality of care and on the other hand we actually want to increase patient engagement and decrease the delays in care right so we have to be on time and language will play a very important role in this and so the main goal with this project it's actually to increase the quality of time in the ICU or the comfort and their care so we know that we have this equation we have health literacy and language barriers and in this work we try to tackle language barriers because if we actually solve that problem probably you can also increase like the health literacy of your patients and so the main hypothesis that we derived in less than 24 hours was among patients who expired during admission patients with limited English proficiency will probably have increased time to that secondary to delays in goals of care these discussions so this was the hypothesis okay I'm not saying that this is true we try to tackle this in the last one so what we did we use the public database mimic 3 is a very huge database public one you can actually access it and work with it and we did some statistics and data analysis to see if we could answer or not the main question so these are pretty much the statistics we did we derived some cox proportional hazards we study the procedure utilization and at the end of life care discussion by analyzing some of the notes that are present in the in ICU I will show some some statistics and graphics that we derived with this project so we analyzed age we have more limited English for English proficiency patients actually in older ones by gender we didn't find actually a significant difference all right so that there is no it seems that there is no difference between gender or sex in terms of limiting English proficiency in race or ethnicity we actually found that there is more limited English proficiency in minorities which is a very interesting finding for us to bear in mind and of course all of these barrels also will play a goal like in the type of insurance you have and so we we end up we end up knowing that we have more LAPs in Medicaid people in Medicaid covered people as opposed to the other ones then we did a study on procedure utilization by English speaking proficiency as you can see in this graphic that's that's in there there is no actually relevant difference from what you could find and in survival probability if you like plot the probability of survival for both English and LAP once again we didn't find no significant differences so you are you are already probably seeing that like the hypothesis probably will have to change in a further study we once again we did a bunch of Cox proportional hazards models based on several variables we can discuss this we can discuss about about these later so basically we try to see what kinds of variables will actually be more important to the main outcomes and we have these plots and then we try to analyze also the end of life conversations and that's where like the team and of RT Beatrice and Vedant are like playing a very important role right now once again we have some interesting findings that we can discuss afterwards and we also did try to see like some kind of gender certification and here we could see that the most pronounced difference like in end of life conversations are found among female patients which is also very interesting finding so for you to get some some takeaway messages we understood that like actually limited English proficiency is actually less impactful on time to death oppositely to the initial hypothesis we know that older patients and Medicaid ones are underrepresented in the data set which is which presents limited of the data and we also find in these 24 hours that female patients with LAPs actually have more frequent and later conversations compared to other groups which is something that it's interesting to like begin more in the future let me just take you like with a motivational let's say quote if you talk to men language in the sense that he goes to he said if you talk to him in his language probably that goes without and this will play a role in patients so if you actually can communicate with your patients probably the healthcare outcomes will be different just me let me just give you some limitations of the study we once again we did one of the at 124 hours one database mimic it's something it's like it's a very huge database so if you work with it you will understand our pains with 24 hours we didn't like some of the descriptors actually do not change over time it was an assumption and it isn't find out in the off course yet like others of others other vials that we actually have to would like to know to study and investigate so that means we need more databases and this one is also very limited to Boston Massachusetts and of course these will change across like demographic regions and there is something that is already being tackled right so we can these were like the next step so increased sample size advanced modeling techniques more analysis try to see bias in sentiment sentimentality analysis and these was the next steps when we finish the challenge so let me finish the presentation right here and knowledge all the team of course for all for that for the amazing we can we have this is the code if you want to fix up some of the analysis that we did we will I mean increase the quality of the GitHub as well and then after these are the event and Beatrice start working on more stuff and now they are actually increasing more analysis and we make from from NLP techniques to LLMs she can talk to you about you later of course increasing the data size in the population so we are also studying like the Wake Forest School of Medicine once again where actually works and we we have more hypotheses that one study but since I'm getting out of time I'll just like leave this for a Q&A and once again thank you all the corresponding author of this study will be RT and that we will expect you in Chicago next July to work more on this thank you thank you so much Tiago I do appreciate that I suppose we were a little tight on time during the data fun so you're very that is on me so thank you so much and our last presenter is nuclei Anderson so then and we'd like to talk about again this so he was our we had a third track which was in patient safety and so through that funding we were able to also sponsor a third prize in addition to beyond what we expected so this was a third prize again this is Nikolai Anderson in disparities in glucose measurements Nikolai yeah hello everybody and thanks Ian I was also fortunate enough to get to attend the data fun in back in New York and I represent team one I know it says team eight on the on the program but I think it's team one anyways we won the patient safety category and I want to talk about how we examined the disparities in glucose measurements so our particular research question was what are the key factors influencing glucose variability in point of care and serum measurements now before I get into the methodology we used to study this I want to provide some background for this type of research and how we wanted to approach this so recently there's been more research revolving around the concept of digital determinants of health now we know that social determinants of health are the social and environmental factors outside of the medical factors that influence the health of the patients right in digital determinants of health we talk about the inequalities and the differences in patient outcomes that are not due to the social factors in themselves but more due to how the digital technologies that we use interact with those social factors so a lot of research a lot of brilliant research has come out recently that talks about identifying these digital technologies that demonstrate an evidence of bias and such pitfalls generally arise from insufficient consideration of patient diversity now we see examples of this in pulse oximetry where the measurements differ based on the skin color of the patient and we see this for e.g. where the conductivity quality measures based are different based on the type of hair on the patient and I can't speak for the entire team but I know that this is definitely the overall motivation for our research is to look at some of these factors because differences in glucose has been studied for 2013 years in terms of how we measure that initial glucose measurement with a finger stick at the point of care versus the more so but more accurate results we get from the serum blood tests from the lab so we look at when those two measurements differ in a short span of time what causes those errors now we know that there are errors right because there are measurement errors in the finger stick and the directionality of those errors and how it affects people in general based on the timeline has been studied but we wanted to see if we could find other key factors that influence this variance so we use the EICU database which is a really extensive multi-center ICU database and of course we want the measurements that we look at to be as close to each other in time so that the point-of-care finger stick glucose measurement is taken really close to the actual blood sample that's sent to the lab so the way we curated our data set was we selected a time span of 15 minutes before or after the different types of sampling so we aligned these pairs that we have of glucose measurements with a lot of different other variables regarding the patients such as the demographic variables of course since this is our main focus and also treatment variables such as insulin, steroids, vasopressors, mechanical ventilation etc everything that also could affect this variance in some way now we started out with a hundred and thirty nine thousand patients and we excluded ninety or seventy three thousand to begin with that had no pairs within this time span or no pairs at all sorry and then we excluded 7,000 pairs more because we wanted these 15 minutes yeah and then we also excluded 11,000 pairs from hospitals with less than a thousand patients so that left us with 46,000 patients and a hundred ninety thousand pairs of glucose measurements now when we're investigating bias in general and how different errors might disproportionately affect certain parts of the population we also are really considerate of how that data got into the data set to begin with so a really important part is to a really important thing is to look for patterns of missingness in the data and this is what's called an offset plot where we can see the different combinations of missing variables and how they're distributed across a cohort so here we can see that we had a lot of missingness for all the insulin, vasopressors, lactate and then the different combinations of missingness and we also looked at how these were missing and we found that they were missing at random across the demographics which is also really important to consider when looking at sampling bias now the we found out when after selecting these pairs that the general error was really nicely normally distributed so there was no directionality in the way these measurements were taken were wrongfully taken if that makes sense and we can see that the distribution based on ethnicity is also there are no obvious differences here so what we do is we specify an outcome called occult hypoglycemia that is we're interested in patients where the point-of-care glucose show above 70 milligram per deciliter and then the serum glucose shows below that so we don't want we don't want I mean we accept some error in these finger stick measurements right but we don't want them to cause any harm disproportionately to a certain part of a patient demographic because they're affected by the directionality of this error so how can we model this well we fit a patient logistic regression model for each hospital to look at if we can predict this occult hypoglycemia so who are measured as being higher than 70 milligram per deciliter from the finger stick and then when the serum value comes back we see that that was actually much lower so when we look at the posterior estimates here we can see that each line represents a hospital and if you're familiar with patient statistics these are basically the parameter estimates of the effect on ethnicity for these things so instead of getting one parameter in typical frequency statistics we get we can account for the we can account for the uncertainty of that parameter estimation basically so we see a lot of hospitals where there's no effect on race ethnicity on these occult hypoglycemia harmful outcome but then we also see a subset of hospitals with a really high probability of unequal harm now we shouldn't it's it's hard to interpret this in the way that we should go out and point to these hospitals and say you're you're disproportionately harming certain patients because of course there are a lot of things that can explain this variability but at least it's it's a star for us we were also having like a busy time during the data fun and those two days right so we don't want to draw any conclusions too fast but it's definitely some interesting preliminary results that we have here of course those hospitals that were like flaring up in terms of the effect parameters could have certain devices they could all share some protocols or practices that are interesting to look into and we can see well if these devices are consistently consistently giving error in a certain direction then well we can do something about it right and this is also the benefit of working with these large data sets so for future work and diving deeper into this project we really want to look at newer data data look at these particular hospitals model them in more sophisticated ways and then also adjust for some of the co-founders we collected also from the EICU so thank you this was team sugar data and I really want to thank all of the members on the team Andrew, Enrique, Joao, Rishika, Vishwa, yeah this was far from a solo effort I mean everybody did a lot and I think if you have any questions from the clinical aspect I'm a data science student you should definitely ask dr. Burrows down here because I think he was really annoyed with me in the end of the data time because I was asking questions like 24-7 what is this value mean what is that what is this yeah so yeah thank you so thank you so much

racial disparities

COVID-19 outcomes

proning effect

limited English proficiency

healthcare disparities

glucose measurement bias