1 00:00:00,719 --> 00:00:04,000 Exciting things are happening at Becker's Healthcare. Stay 2 00:00:04,000 --> 00:00:06,240 ahead of industry trends with the new Becker's 3 00:00:06,240 --> 00:00:08,500 CFO plus Revenue Cycle podcast, 4 00:00:08,800 --> 00:00:11,119 your go to source for insights from top 5 00:00:11,119 --> 00:00:14,224 healthcare finance leaders. Tune in wherever you get 6 00:00:14,224 --> 00:00:14,964 your podcasts. 7 00:00:15,904 --> 00:00:18,464 And don't miss the tenth annual Health IT 8 00:00:18,464 --> 00:00:22,155 plus Digital Health plus RCM conference, happening September 9 00:00:22,155 --> 00:00:24,464 30 to 10/03/2025 10 00:00:24,464 --> 00:00:25,204 in Chicago. 11 00:00:25,664 --> 00:00:27,125 Join thousands of executives, 12 00:00:27,469 --> 00:00:30,429 engage with industry leaders, and explore the future 13 00:00:30,429 --> 00:00:31,649 of health care innovation. 14 00:00:32,270 --> 00:00:36,750 Learn more about our upcoming events at beckershospitalreview.com. 15 00:00:36,750 --> 00:00:37,570 See you there. 16 00:00:38,509 --> 00:00:40,750 This is Laura Dierda with the Becker's Healthcare 17 00:00:40,750 --> 00:00:42,965 podcast. I'm thrilled today to be joined by 18 00:00:42,965 --> 00:00:45,204 Tur Guy Ayer, professor in the school of 19 00:00:45,204 --> 00:00:48,005 industrial and systems engineering at Georgia Tech. Tur 20 00:00:48,005 --> 00:00:49,045 Guy, it's a pleasure to have you on 21 00:00:49,045 --> 00:00:50,104 the podcast today. 22 00:00:50,564 --> 00:00:52,744 It's my pleasure. Thank you for having me. 23 00:00:53,204 --> 00:00:55,045 Now I'm looking forward to our conversation and 24 00:00:55,045 --> 00:00:56,725 really digging into some of the cool things 25 00:00:56,725 --> 00:00:59,179 you're doing, within your research at Georgia Tech. 26 00:00:59,179 --> 00:01:00,619 But before we dive in, can you tell 27 00:01:00,619 --> 00:01:01,899 us a little bit more about yourself and 28 00:01:01,899 --> 00:01:02,640 your background? 29 00:01:03,579 --> 00:01:04,079 Absolutely. 30 00:01:04,459 --> 00:01:07,439 So I was trained as a decision scientist. 31 00:01:08,459 --> 00:01:09,920 I did my t PhD 32 00:01:10,765 --> 00:01:15,085 thesis on breast cancer diagnosis using artificial neural 33 00:01:15,085 --> 00:01:15,585 networks. 34 00:01:17,085 --> 00:01:19,805 And I work extensively with radiologists and some 35 00:01:19,805 --> 00:01:21,885 other physicians. Even though I was trained as 36 00:01:21,885 --> 00:01:22,545 an engineer, 37 00:01:23,245 --> 00:01:25,505 I was trained almost like a pseudo doctor. 38 00:01:26,030 --> 00:01:27,950 And as for my background, I was raised 39 00:01:27,950 --> 00:01:29,950 to be a physician by my parents. They 40 00:01:29,950 --> 00:01:32,189 really wanted to want wanted me to be 41 00:01:32,189 --> 00:01:32,849 a physician. 42 00:01:33,310 --> 00:01:34,209 And up until 43 00:01:34,909 --> 00:01:37,390 junior high, I was thinking of being a 44 00:01:37,390 --> 00:01:37,890 physician. 45 00:01:38,525 --> 00:01:40,385 But my sister is a veteran medicine, 46 00:01:41,004 --> 00:01:43,805 doctor, and and I was helping her when 47 00:01:43,805 --> 00:01:46,144 I was in high school, and I fainted 48 00:01:46,204 --> 00:01:47,825 as soon as I saw some blood. 49 00:01:48,364 --> 00:01:50,364 And it made it obvious that I will 50 00:01:50,364 --> 00:01:52,444 make a good physician. So I decided to 51 00:01:52,444 --> 00:01:53,344 study engineering 52 00:01:54,079 --> 00:01:55,840 because I was good at math. I I 53 00:01:55,840 --> 00:01:57,620 liked, you know, analytical thinking. 54 00:01:58,799 --> 00:02:01,219 And I had a lot of, you know, 55 00:02:01,840 --> 00:02:03,599 I I I had a lot of trials 56 00:02:03,599 --> 00:02:06,239 and errors, in my in my studies. I 57 00:02:06,239 --> 00:02:07,680 start as a double e, 58 00:02:08,319 --> 00:02:11,075 major and then switched to computer science 59 00:02:11,694 --> 00:02:13,155 in my junior year 60 00:02:13,455 --> 00:02:14,194 in college. 61 00:02:14,814 --> 00:02:17,235 And only my senior year, I, 62 00:02:17,855 --> 00:02:20,735 figured out a discipline called operations research within 63 00:02:20,735 --> 00:02:21,794 industrial engineering, 64 00:02:22,175 --> 00:02:23,474 which is a science of 65 00:02:23,879 --> 00:02:27,319 decision making through mathematical modeling and and data 66 00:02:27,319 --> 00:02:28,219 driven approaches. 67 00:02:28,680 --> 00:02:31,319 So I ended up majoring in operations research 68 00:02:31,319 --> 00:02:32,459 and did a PhD 69 00:02:33,159 --> 00:02:37,020 in operations research slash industrial engineering focusing primarily 70 00:02:37,159 --> 00:02:38,699 on on health care applications. 71 00:02:41,615 --> 00:02:43,775 Fascinating. That's, you know, such a a cool 72 00:02:43,775 --> 00:02:45,694 journey into the career you have today and 73 00:02:45,694 --> 00:02:47,855 and really meaningful research you're doing. Can you 74 00:02:47,855 --> 00:02:49,455 share a bit about that research and why 75 00:02:49,455 --> 00:02:50,194 it's important? 76 00:02:51,294 --> 00:02:51,794 Absolutely. 77 00:02:52,415 --> 00:02:54,700 So I would summarize my research as, 78 00:02:55,099 --> 00:02:56,719 data driven mathematical 79 00:02:57,099 --> 00:03:00,860 slash computational modeling for important medical decision making 80 00:03:00,860 --> 00:03:02,240 and health policy analysis. 81 00:03:02,700 --> 00:03:05,439 Oftentimes, in clinical decision making 82 00:03:05,819 --> 00:03:08,879 and health policy making, decisions are too complex, 83 00:03:09,875 --> 00:03:12,455 and you need a framework for the systematic 84 00:03:12,594 --> 00:03:13,495 decision making 85 00:03:13,875 --> 00:03:14,694 while incorporating 86 00:03:14,995 --> 00:03:17,334 data to capture important analysis. 87 00:03:18,034 --> 00:03:20,754 And and such decisions become too complex for 88 00:03:20,754 --> 00:03:22,754 human beings. To me, it's more like making 89 00:03:22,754 --> 00:03:24,694 multiplications on the order of millions. 90 00:03:25,020 --> 00:03:26,459 When you ask such a question to a 91 00:03:26,459 --> 00:03:28,000 computer, it's pretty straightforward. 92 00:03:28,459 --> 00:03:30,139 But as humans, if we try to do 93 00:03:30,139 --> 00:03:32,860 these computations in our head, it becomes pretty 94 00:03:32,860 --> 00:03:33,360 challenging. 95 00:03:33,900 --> 00:03:35,680 So I help that building 96 00:03:36,699 --> 00:03:37,759 frameworks, mathematical 97 00:03:38,219 --> 00:03:39,439 computational frameworks 98 00:03:40,335 --> 00:03:41,875 using utilizing data, 99 00:03:42,254 --> 00:03:44,514 to help it solving such important, 100 00:03:44,895 --> 00:03:47,235 pressing clinical and policy problems. 101 00:03:50,014 --> 00:03:52,289 That's fascinating to hear. And, you know, really, 102 00:03:52,370 --> 00:03:54,129 such a unique angle to be part of 103 00:03:54,129 --> 00:03:56,150 the health care space as you mentioned using 104 00:03:56,449 --> 00:03:58,949 your mathematics background and certainly, 105 00:03:59,729 --> 00:04:01,889 you know, your your talent for, 106 00:04:02,209 --> 00:04:03,810 computation to get this into, 107 00:04:04,370 --> 00:04:05,590 the health care field. 108 00:04:06,014 --> 00:04:08,495 Now I'm curious. So what challenges in health 109 00:04:08,495 --> 00:04:10,014 care are you aiming to address through this 110 00:04:10,014 --> 00:04:10,514 research? 111 00:04:11,534 --> 00:04:12,034 Absolutely. 112 00:04:12,814 --> 00:04:16,095 So in health care, the goals health care 113 00:04:16,095 --> 00:04:19,634 research, the gold standard is randomized controlled trials. 114 00:04:20,899 --> 00:04:22,680 In randomized controlled trials, 115 00:04:23,219 --> 00:04:26,180 researchers do AB testing, meaning that, you know, 116 00:04:26,180 --> 00:04:28,259 if there is a new intervention, new drug, 117 00:04:28,259 --> 00:04:29,000 new technology, 118 00:04:30,019 --> 00:04:33,404 clinical trials are designed to to assess if 119 00:04:33,404 --> 00:04:36,705 the new technology, new intervention, let's say, a, 120 00:04:36,764 --> 00:04:38,865 is better than the status quo b. 121 00:04:39,324 --> 00:04:42,045 And in such randomized controlled trials, they are 122 00:04:42,045 --> 00:04:45,245 typically expensive, very expensive, millions of dollars. They 123 00:04:45,245 --> 00:04:46,145 are time consuming. 124 00:04:46,925 --> 00:04:48,899 And and also it takes a lot of 125 00:04:49,139 --> 00:04:49,639 efforts, 126 00:04:50,660 --> 00:04:53,540 to conduct these clinical trials, recruit patients into 127 00:04:53,540 --> 00:04:56,259 those trials. And typically, they are short term. 128 00:04:56,259 --> 00:04:57,959 Typically, you know, the trials 129 00:04:58,419 --> 00:05:00,439 last for about three to five years. 130 00:05:01,459 --> 00:05:01,959 And 131 00:05:03,125 --> 00:05:06,004 by the time the trials are completed, we 132 00:05:06,004 --> 00:05:08,245 have answers to the question of whether a 133 00:05:08,245 --> 00:05:11,125 is better than b. But oftentimes, in in 134 00:05:11,125 --> 00:05:13,204 in practice, we are looking at multiple options, 135 00:05:13,204 --> 00:05:15,365 not just a versus b. We are looking 136 00:05:15,365 --> 00:05:18,104 at a versus b versus c versus d. 137 00:05:18,409 --> 00:05:21,529 And and sometimes these competitors are more than 138 00:05:21,529 --> 00:05:22,909 ten, fifteen, 20. 139 00:05:23,370 --> 00:05:25,069 And it's not plausible to 140 00:05:25,849 --> 00:05:28,889 assess those mini interventions in a clinical trial 141 00:05:28,889 --> 00:05:31,069 setting because, a, it will be extremely, 142 00:05:32,169 --> 00:05:32,669 costly, 143 00:05:33,274 --> 00:05:35,535 and, b, it will be very time consuming. 144 00:05:36,395 --> 00:05:39,274 And, c, it perhaps wouldn't be feasible to 145 00:05:39,274 --> 00:05:39,774 recruit, 146 00:05:41,194 --> 00:05:43,535 thousands of patients into such a trial. 147 00:05:45,115 --> 00:05:47,615 Because and and the last limitation of randomized 148 00:05:47,754 --> 00:05:49,970 controlled trials is that, as I mentioned earlier, 149 00:05:50,189 --> 00:05:53,009 their time span is typically short, and oftentimes, 150 00:05:53,069 --> 00:05:55,490 you would like to understand longer term implications 151 00:05:55,629 --> 00:05:56,449 of interventions. 152 00:05:57,069 --> 00:05:58,370 And at that point, 153 00:05:58,750 --> 00:06:01,389 our research chimes in and and helps with 154 00:06:01,389 --> 00:06:03,330 understanding some longer impacts 155 00:06:03,644 --> 00:06:04,384 of such 156 00:06:04,685 --> 00:06:05,185 interventions. 157 00:06:05,564 --> 00:06:08,764 We collect evidence. We basically analyze evidence coming 158 00:06:08,764 --> 00:06:11,805 out of randomized controlled trials, published studies, meta 159 00:06:11,805 --> 00:06:13,264 analysis, real world evidence, 160 00:06:13,805 --> 00:06:15,664 and build virtual trials, 161 00:06:15,964 --> 00:06:17,504 meaning that, you know, we replicate 162 00:06:18,160 --> 00:06:20,100 these trials in computer settings. 163 00:06:21,199 --> 00:06:23,699 We validate these computational models. 164 00:06:24,079 --> 00:06:26,959 And once validated that, they they they become 165 00:06:26,959 --> 00:06:28,879 a test bed for us to ask a 166 00:06:28,879 --> 00:06:30,899 lot of interesting what if questions 167 00:06:31,324 --> 00:06:32,685 in the short term. So we are no 168 00:06:32,685 --> 00:06:35,324 longer limited to three years or five years. 169 00:06:35,324 --> 00:06:38,044 We can expand the time horizon to twenty 170 00:06:38,044 --> 00:06:39,584 years, thirty years lifetime. 171 00:06:39,964 --> 00:06:42,044 We are no longer limited to, you know, 172 00:06:42,044 --> 00:06:44,384 a versus b. We can compare multiple, 173 00:06:45,569 --> 00:06:46,069 interventions. 174 00:06:46,850 --> 00:06:48,770 And and that gives us a lot of 175 00:06:48,770 --> 00:06:50,949 flexibility to, again, ask a lot of interesting, 176 00:06:51,730 --> 00:06:54,310 pressing policy or clinical level questions. 177 00:06:57,694 --> 00:06:59,714 That's amazing to hear. You know? And certainly, 178 00:06:59,935 --> 00:07:01,615 really cool to see that evolution of how 179 00:07:01,615 --> 00:07:03,875 you can approach clinical trials in a really 180 00:07:04,014 --> 00:07:05,314 meaningful way, especially 181 00:07:05,694 --> 00:07:08,414 knowing how important they are in advancing medicine 182 00:07:08,414 --> 00:07:10,899 and the treatment that is provided for patients 183 00:07:10,899 --> 00:07:13,620 across the board in a variety of, settings, 184 00:07:13,939 --> 00:07:16,100 as well as, you know, just the the 185 00:07:16,100 --> 00:07:18,199 demographics of the patients can be so different. 186 00:07:19,139 --> 00:07:21,720 How do you approach conducting research to ensure 187 00:07:21,795 --> 00:07:23,475 that it has the kind of real world 188 00:07:23,475 --> 00:07:24,915 impact on patient care? I know a lot 189 00:07:24,915 --> 00:07:26,615 of the things we've been talking about, 190 00:07:26,995 --> 00:07:28,275 you know, are looking at the design of 191 00:07:28,275 --> 00:07:29,634 the trials and those kinds of things. And 192 00:07:29,634 --> 00:07:31,335 so how do you see that impact? 193 00:07:32,115 --> 00:07:33,715 So there are a couple of ways that 194 00:07:33,715 --> 00:07:36,775 we ensure that our findings are practically useful. 195 00:07:37,490 --> 00:07:38,470 One, we 196 00:07:39,009 --> 00:07:43,430 collaborate extensively with it, end end users, stakeholders, 197 00:07:43,649 --> 00:07:45,110 typically physicians, policymakers, 198 00:07:45,410 --> 00:07:48,149 depending on the nature of the project, and 199 00:07:48,449 --> 00:07:51,935 that is useful for multiple purposes. One, we 200 00:07:51,935 --> 00:07:54,834 ensure that we capture important clinical nuances, 201 00:07:55,774 --> 00:07:57,854 medical nuances that we may not be aware 202 00:07:57,854 --> 00:08:00,574 of. So we have the clinical slash policy 203 00:08:00,574 --> 00:08:01,074 expertise. 204 00:08:02,334 --> 00:08:03,314 And two, 205 00:08:03,930 --> 00:08:05,870 because it's a co collaborative 206 00:08:06,170 --> 00:08:08,029 development, collaborative research, 207 00:08:08,730 --> 00:08:12,250 those decision makers, end users are engaged early 208 00:08:12,250 --> 00:08:13,149 on in the process, 209 00:08:13,850 --> 00:08:16,330 and we deploy our solutions to those end 210 00:08:16,330 --> 00:08:16,830 users. 211 00:08:17,245 --> 00:08:19,324 And once they believe in the value, once 212 00:08:19,324 --> 00:08:21,964 they see what these solutions tools could do 213 00:08:21,964 --> 00:08:24,444 for them, they are typically the champions to 214 00:08:24,444 --> 00:08:26,944 take take these solutions to the next step 215 00:08:27,084 --> 00:08:29,504 and, you know, bring it to clinical implementation 216 00:08:30,204 --> 00:08:32,365 or use it as a policy policy tool, 217 00:08:32,365 --> 00:08:32,865 etcetera. 218 00:08:33,700 --> 00:08:35,059 And the last thing that we do is 219 00:08:35,059 --> 00:08:37,159 we ensure that our findings 220 00:08:37,620 --> 00:08:38,440 are valid. 221 00:08:38,899 --> 00:08:41,000 Before we deploy any solutions, 222 00:08:41,379 --> 00:08:43,000 we do a lot of stress testing 223 00:08:43,539 --> 00:08:45,480 compared against any available 224 00:08:45,860 --> 00:08:46,919 real world data. 225 00:08:48,434 --> 00:08:49,095 So it's 226 00:08:49,554 --> 00:08:51,634 just to ensure that, again, the results are 227 00:08:51,634 --> 00:08:53,955 clinically meaningful, and we also have the binds 228 00:08:53,955 --> 00:08:55,174 from these end users. 229 00:08:58,434 --> 00:09:00,115 That's helpful to hear. Thank you so much 230 00:09:00,115 --> 00:09:02,370 for digging a little bit deeper there. Now 231 00:09:02,370 --> 00:09:04,049 I'm curious. What are some of the big 232 00:09:04,049 --> 00:09:05,970 obstacles that you face while you're, 233 00:09:06,290 --> 00:09:08,129 doing this type of research and trying to 234 00:09:08,129 --> 00:09:10,049 make sure that it's deployed in a meaningful 235 00:09:10,049 --> 00:09:10,549 way? 236 00:09:12,610 --> 00:09:13,509 Data challenges 237 00:09:13,889 --> 00:09:16,370 is is a big issue. Oftentimes, health care 238 00:09:16,370 --> 00:09:17,429 data is fragmented, 239 00:09:18,815 --> 00:09:21,154 and they are kept in silos. 240 00:09:21,695 --> 00:09:24,514 You don't oftentimes, you don't have national level 241 00:09:25,055 --> 00:09:25,555 databases, 242 00:09:25,934 --> 00:09:28,095 access to that data. You know, even even 243 00:09:28,095 --> 00:09:30,175 if it is local level data may be 244 00:09:30,175 --> 00:09:31,455 a challenge. We have to go through a 245 00:09:31,455 --> 00:09:33,394 little bit pool processes, and 246 00:09:34,230 --> 00:09:35,590 and in the end, it may not work 247 00:09:35,590 --> 00:09:37,429 out. So as an example, you know, we 248 00:09:37,429 --> 00:09:38,809 have recently approved 249 00:09:39,190 --> 00:09:40,970 to have access to a data 250 00:09:41,750 --> 00:09:42,409 all by, 251 00:09:43,029 --> 00:09:44,250 kept by CDC. 252 00:09:44,789 --> 00:09:47,830 And because of these communication freezes CDC is 253 00:09:47,830 --> 00:09:51,214 going through, we basically even though our our 254 00:09:51,214 --> 00:09:53,774 data access request has been approved after several 255 00:09:53,774 --> 00:09:55,394 months of back and forth communications 256 00:09:56,095 --> 00:09:59,394 and documentation, you know, report preparation, etcetera, 257 00:10:00,095 --> 00:10:01,879 basically, this is this is on freeze now. 258 00:10:02,200 --> 00:10:03,879 We won't be able to access this data 259 00:10:03,879 --> 00:10:05,740 until this uncertainty is resolved. 260 00:10:07,080 --> 00:10:08,940 So accessing the data, 261 00:10:09,399 --> 00:10:12,779 compiling, augmenting datasets, and and also 262 00:10:13,240 --> 00:10:14,860 validating, you know, these 263 00:10:15,584 --> 00:10:19,105 including missing data, validating these datasets is is 264 00:10:19,105 --> 00:10:20,964 one of the challenges. And and, 265 00:10:21,824 --> 00:10:22,964 sometimes communication 266 00:10:23,345 --> 00:10:26,004 gaps with the end users is a challenge. 267 00:10:27,985 --> 00:10:29,504 Yeah. I would I would list these two 268 00:10:29,504 --> 00:10:30,725 as the biggest challenges. 269 00:10:33,129 --> 00:10:35,230 That's helpful to know. And, you know, certainly, 270 00:10:35,289 --> 00:10:35,929 it seems like, 271 00:10:36,570 --> 00:10:38,970 having that reliable data is a challenge for 272 00:10:38,970 --> 00:10:40,649 a lot of health care organizations, as you 273 00:10:40,649 --> 00:10:43,370 said, and and really truly something that, can 274 00:10:43,370 --> 00:10:44,350 make a big difference, 275 00:10:45,105 --> 00:10:45,605 within 276 00:10:45,904 --> 00:10:48,164 the the research and development in, 277 00:10:48,625 --> 00:10:50,945 many fronts, but but certainly with the work 278 00:10:50,945 --> 00:10:53,184 that you're doing as well. I I think, 279 00:10:53,424 --> 00:10:55,924 you know, when you look at that communication 280 00:10:56,225 --> 00:10:57,120 aspect of it, 281 00:10:57,679 --> 00:10:59,679 too, you know, is there anything you found 282 00:10:59,679 --> 00:11:01,779 that has worked particularly well in, 283 00:11:02,720 --> 00:11:03,220 communicating 284 00:11:03,519 --> 00:11:05,759 some of the research and the data, to 285 00:11:05,759 --> 00:11:07,379 those who, you know, 286 00:11:07,759 --> 00:11:09,375 need to know about it or or would 287 00:11:09,534 --> 00:11:10,815 most benefit from it, 288 00:11:11,294 --> 00:11:13,215 or is that still very much a work 289 00:11:13,215 --> 00:11:13,955 in progress? 290 00:11:15,774 --> 00:11:18,495 Absolutely. So in one of our research projects, 291 00:11:18,495 --> 00:11:20,595 we work with radiologists, 292 00:11:20,894 --> 00:11:22,274 breast imaging radiologists, 293 00:11:24,379 --> 00:11:26,879 from from one of the hospitals in Wisconsin. 294 00:11:27,659 --> 00:11:30,220 And we were basically the key question research 295 00:11:30,220 --> 00:11:32,559 question we were asking was whether 296 00:11:33,500 --> 00:11:37,200 machine learning models, artificial intelligence models was useful 297 00:11:37,824 --> 00:11:41,105 in diagnosing breast cancer based on mammographic images. 298 00:11:41,105 --> 00:11:44,324 So in other words, radiologists look at suspicious 299 00:11:44,544 --> 00:11:45,044 mammograms, 300 00:11:45,745 --> 00:11:47,105 or or they look at a set of 301 00:11:47,105 --> 00:11:49,745 mammograms if if and if they notice any 302 00:11:49,745 --> 00:11:52,884 suspicion such as, you know, some some mass 303 00:11:53,750 --> 00:11:55,049 some messes, calcifications, 304 00:11:55,669 --> 00:11:56,169 distortions, 305 00:11:56,470 --> 00:11:57,610 asymmetry, etcetera, 306 00:11:58,230 --> 00:12:00,809 then they may refer them to a biopsy, 307 00:12:02,149 --> 00:12:02,649 to 308 00:12:03,190 --> 00:12:05,909 see if if patients have a tumor, a 309 00:12:05,909 --> 00:12:08,309 malignant tumor, and and they need breast cancer 310 00:12:08,309 --> 00:12:08,809 treatment. 311 00:12:09,164 --> 00:12:11,404 And it turns out that this decision making 312 00:12:11,404 --> 00:12:14,445 process is is very difficult and in more 313 00:12:14,445 --> 00:12:16,684 process. There are, like, you know, over 50 314 00:12:16,684 --> 00:12:19,105 risk factors that physicians need to consider. 315 00:12:19,884 --> 00:12:20,384 And 316 00:12:20,970 --> 00:12:22,649 and you need to again, to me, this 317 00:12:22,649 --> 00:12:24,970 is like making multiplications on the order of 318 00:12:24,970 --> 00:12:27,289 millions. And it turns out that about ninety 319 00:12:27,289 --> 00:12:29,230 eight percent of the biopsy cases 320 00:12:29,929 --> 00:12:32,490 are, false positives, meaning that out of ninety 321 00:12:32,490 --> 00:12:35,875 eight percent of all mammograms referred to biopsies, 322 00:12:36,335 --> 00:12:38,575 they are in reality, you know, just okay 323 00:12:38,575 --> 00:12:40,815 mammograms. Right? Nothing is wrong with them. But 324 00:12:40,815 --> 00:12:43,075 think about the psychology of a patient. 325 00:12:44,095 --> 00:12:45,615 Right, when you tell them that they have 326 00:12:45,615 --> 00:12:47,774 a suspicious mammogram and they need to schedule 327 00:12:47,774 --> 00:12:48,355 a biopsy, 328 00:12:48,799 --> 00:12:51,519 those biopsies are typically not scheduled for tomorrow. 329 00:12:51,519 --> 00:12:52,959 Right? You need to wait for two, three, 330 00:12:52,959 --> 00:12:54,819 four weeks, sometimes months, 331 00:12:55,199 --> 00:12:57,059 to get these biopsies done 332 00:12:58,079 --> 00:13:00,480 and on top of it further wait to 333 00:13:00,480 --> 00:13:02,485 hear about the results. So you go through 334 00:13:02,485 --> 00:13:04,045 a lot of stress. Right? You know, you 335 00:13:04,045 --> 00:13:05,485 you go through a lot of stress. Your 336 00:13:05,485 --> 00:13:07,565 your family go goes through a lot of 337 00:13:07,565 --> 00:13:08,065 stress. 338 00:13:09,004 --> 00:13:10,465 And what we did is 339 00:13:11,165 --> 00:13:11,904 we basically, 340 00:13:12,684 --> 00:13:16,144 built a an artificial neural network model, 341 00:13:16,769 --> 00:13:19,170 and we asked the key for the the 342 00:13:19,170 --> 00:13:21,350 following key question. Right? Could an artificial 343 00:13:21,810 --> 00:13:24,629 neural network model, a form of an, artificial 344 00:13:24,769 --> 00:13:26,710 intelligence, machine learning models, 345 00:13:27,170 --> 00:13:29,670 help with biopsy decision making 346 00:13:30,315 --> 00:13:34,495 using findings or extracting data from mammographic images. 347 00:13:35,434 --> 00:13:37,595 And it turned out that and and and 348 00:13:37,595 --> 00:13:39,674 our results we published our results in in 349 00:13:39,674 --> 00:13:42,475 a clinical journal, radiologist journal. We thought, you 350 00:13:42,475 --> 00:13:43,934 know, physicians, radiologists 351 00:13:44,399 --> 00:13:46,959 would be reacting to those findings because our 352 00:13:46,959 --> 00:13:48,419 findings indicated that 353 00:13:48,959 --> 00:13:51,139 our machine learning based solutions 354 00:13:51,839 --> 00:13:52,579 could perform 355 00:13:53,199 --> 00:13:55,299 better than majority of the radiologists 356 00:13:55,679 --> 00:13:57,059 in our setting. Specifically, 357 00:13:58,004 --> 00:13:58,745 seven out 358 00:13:59,285 --> 00:14:00,345 of eight radiologists, 359 00:14:00,965 --> 00:14:03,465 seven of them performed worse than the artificial 360 00:14:03,524 --> 00:14:06,165 neural network model. So meaning that a AI 361 00:14:06,165 --> 00:14:06,665 model 362 00:14:07,125 --> 00:14:08,424 was more successful 363 00:14:08,725 --> 00:14:12,644 in detecting abnormalities on mammographic images as compared 364 00:14:12,644 --> 00:14:14,399 with those seven out of eight radiologists. 365 00:14:15,019 --> 00:14:17,440 Whereas one radiologist performed significantly 366 00:14:17,899 --> 00:14:19,679 better than the AI model. 367 00:14:20,139 --> 00:14:20,539 That's 368 00:14:21,019 --> 00:14:23,580 that radiologist who stood out had thirty two 369 00:14:23,580 --> 00:14:26,299 years of domain experience, whereas the other seven 370 00:14:26,299 --> 00:14:26,799 radiologists 371 00:14:27,514 --> 00:14:30,075 who are outperformed by the AI model had 372 00:14:30,075 --> 00:14:32,575 an experience ranging from two to ten years. 373 00:14:32,794 --> 00:14:34,634 And we positioned our paper in a way 374 00:14:34,634 --> 00:14:36,335 that we we we we basically, 375 00:14:36,715 --> 00:14:38,975 you know, highlighted the key message that 376 00:14:39,690 --> 00:14:43,210 AI based solutions may complement physicians, especially junior 377 00:14:43,210 --> 00:14:45,870 physicians, to help it, you know, to to 378 00:14:45,929 --> 00:14:48,330 act as a second reviewer or a second 379 00:14:48,330 --> 00:14:48,830 reader, 380 00:14:49,690 --> 00:14:51,710 especially in the resource limited settings. 381 00:14:52,434 --> 00:14:54,915 The doubles double screening in radiology is very 382 00:14:54,915 --> 00:14:57,154 common, meaning that, you know, a radiologist first 383 00:14:57,154 --> 00:14:59,715 reads an image. And if the radiologist is 384 00:14:59,715 --> 00:15:00,295 is unsure, 385 00:15:00,754 --> 00:15:03,175 he or she consults with a second radiologist, 386 00:15:03,394 --> 00:15:04,035 and then, 387 00:15:04,595 --> 00:15:07,019 they discuss and make a decision together. But 388 00:15:07,100 --> 00:15:09,360 in the resource limited settings, that's not feasible. 389 00:15:09,980 --> 00:15:12,620 And we positioned our paper study in a 390 00:15:12,620 --> 00:15:15,259 way that we we emphasized that in resource 391 00:15:15,259 --> 00:15:16,240 limited settings, 392 00:15:16,540 --> 00:15:18,860 AI based solutions could act as a second 393 00:15:18,860 --> 00:15:20,080 screener, second reviewer. 394 00:15:21,019 --> 00:15:21,519 And 395 00:15:23,204 --> 00:15:26,184 to our surprise, pleasantly, our findings are very 396 00:15:26,245 --> 00:15:29,044 well received by by the radiology community. Right? 397 00:15:29,044 --> 00:15:31,144 Radiology community is not trained 398 00:15:31,924 --> 00:15:33,304 as as AI experts. 399 00:15:33,684 --> 00:15:35,704 But even though, you know, the findings 400 00:15:36,470 --> 00:15:38,149 could have sort of, like, to to to 401 00:15:38,149 --> 00:15:40,970 some extent threaten their their their their discipline, 402 00:15:41,509 --> 00:15:44,490 they were very positive, very enthusiastic in embracing 403 00:15:44,549 --> 00:15:47,429 our findings. And we they they ended up 404 00:15:47,429 --> 00:15:48,730 accepting our paper, 405 00:15:49,190 --> 00:15:51,254 to get published in in one of their 406 00:15:51,254 --> 00:15:53,035 flagship journals in their communities. 407 00:15:53,654 --> 00:15:55,115 This is just one example. 408 00:15:55,415 --> 00:15:57,894 I can think of several such examples where 409 00:15:57,894 --> 00:16:00,315 our findings were very well received by the, 410 00:16:00,855 --> 00:16:03,115 by by the corresponding communities. 411 00:16:04,779 --> 00:16:07,339 That's a really, powerful example, and thank you 412 00:16:07,339 --> 00:16:08,959 so much for, explaining, 413 00:16:09,339 --> 00:16:10,480 deeper there and, 414 00:16:10,940 --> 00:16:12,459 telling us a story because I think, you 415 00:16:12,459 --> 00:16:14,620 know, it's it's really cool to have that, 416 00:16:14,860 --> 00:16:15,839 frame of reference. 417 00:16:16,264 --> 00:16:18,684 Now have there been any surprising or unexpected 418 00:16:18,904 --> 00:16:20,764 findings that you've had within your research? 419 00:16:21,304 --> 00:16:24,264 Yes. Many of them. I'll just mention maybe 420 00:16:24,264 --> 00:16:25,164 one or two. 421 00:16:26,105 --> 00:16:29,404 In one project, we were looking at, hepatitis 422 00:16:29,544 --> 00:16:30,285 c elimination. 423 00:16:30,759 --> 00:16:34,120 Hepatitis c is an infectious disease, and it's 424 00:16:34,120 --> 00:16:36,360 a prevalent disease. As as a matter of 425 00:16:36,360 --> 00:16:38,139 fact, as of 02/2006, 426 00:16:38,679 --> 00:16:41,720 the number of people dying from hepatitis c 427 00:16:41,720 --> 00:16:44,120 in The US has surpassed those dying from 428 00:16:44,120 --> 00:16:44,620 HIV. 429 00:16:45,000 --> 00:16:47,615 Right? So when I when we think about 430 00:16:47,615 --> 00:16:49,215 HIV, I don't think we need to make 431 00:16:49,215 --> 00:16:50,274 a case that HIV, 432 00:16:50,975 --> 00:16:52,735 has a has a big burden on the 433 00:16:52,735 --> 00:16:53,235 society. 434 00:16:54,415 --> 00:16:56,014 And it turns out that as of as 435 00:16:56,014 --> 00:16:58,174 of, you know, as of 02/2006, 436 00:16:58,174 --> 00:17:01,054 hepatitis c has even a bigger burden on 437 00:17:01,054 --> 00:17:01,715 the society. 438 00:17:03,120 --> 00:17:03,620 And 439 00:17:04,159 --> 00:17:06,000 because it has a such a so it 440 00:17:06,000 --> 00:17:08,159 has such a big disease burden, I was 441 00:17:08,159 --> 00:17:10,419 I was interested in, you know, 442 00:17:11,039 --> 00:17:13,700 doing some modeling analysis in that space. I 443 00:17:14,079 --> 00:17:18,184 elaborated with epidemiologists and physicians and decision scientists 444 00:17:18,565 --> 00:17:19,465 in this work, 445 00:17:19,845 --> 00:17:22,244 from Emery Rowland School of Public Health, Harvard 446 00:17:22,244 --> 00:17:23,945 Medical School, among others. 447 00:17:24,484 --> 00:17:26,565 And the key research question in one of 448 00:17:26,565 --> 00:17:28,644 the projects, the key research question we were 449 00:17:28,644 --> 00:17:29,865 asking was the following. 450 00:17:30,670 --> 00:17:33,329 Does it make sense to treat hepatitis c 451 00:17:33,789 --> 00:17:36,430 in correctional facilities? Does it make sense to 452 00:17:36,430 --> 00:17:39,230 provide access to vast access to hepatitis c 453 00:17:39,230 --> 00:17:41,009 treatment in correctional facilities, 454 00:17:41,950 --> 00:17:44,025 or should it be limited in some ways? 455 00:17:44,265 --> 00:17:46,345 And the reason why this question was relevant 456 00:17:46,345 --> 00:17:47,724 is that by the time 457 00:17:48,345 --> 00:17:51,545 the initial hepatitis C treating drugs, direct acting 458 00:17:51,545 --> 00:17:54,045 agents came to market around early 2000s, 459 00:17:54,904 --> 00:17:57,404 many of the correctional health facilities were limiting 460 00:17:57,464 --> 00:17:59,244 access to hepatitis C treatment, 461 00:18:00,150 --> 00:18:01,610 because hepatitis c treatment 462 00:18:01,990 --> 00:18:03,990 by then and and even still, 463 00:18:04,390 --> 00:18:05,930 was was quite expensive. 464 00:18:06,630 --> 00:18:08,170 By the time they came to market, 465 00:18:08,470 --> 00:18:09,130 the single 466 00:18:10,390 --> 00:18:13,269 pill cost per day was about $1,000 467 00:18:13,269 --> 00:18:15,205 per pill, and patients were expected to be 468 00:18:15,205 --> 00:18:17,365 on treatment for twelve weeks. So meaning that, 469 00:18:17,365 --> 00:18:18,884 you know, by the time the treatment is 470 00:18:18,884 --> 00:18:19,384 completed, 471 00:18:19,765 --> 00:18:23,305 the per patient cost would be about $84,000. 472 00:18:24,805 --> 00:18:27,945 And even though correctional facilities receive some discounts, 473 00:18:28,005 --> 00:18:29,144 even after discounts, 474 00:18:29,740 --> 00:18:31,279 a study has shown that 475 00:18:32,539 --> 00:18:35,500 if correctional facilities provide a treatment to all 476 00:18:35,500 --> 00:18:37,279 hepatitis c infected patients, 477 00:18:37,740 --> 00:18:40,140 the total cost would be higher than the 478 00:18:40,140 --> 00:18:43,440 entire health care spending budget allocated for correctional 479 00:18:43,500 --> 00:18:44,000 facilities. 480 00:18:45,365 --> 00:18:47,605 And and and inmates are not suffering only 481 00:18:47,605 --> 00:18:50,004 from hepatitis c. Right? They could be infected 482 00:18:50,004 --> 00:18:52,424 with with HIV. They may have chronic, 483 00:18:53,044 --> 00:18:54,504 diseases such as cardiovascular 484 00:18:54,804 --> 00:18:56,264 disease, diabetes, etcetera. 485 00:18:56,644 --> 00:18:58,585 From a resource allocation perspective, 486 00:18:58,940 --> 00:19:01,039 it it it was very clear that 487 00:19:01,420 --> 00:19:03,900 correctional health systems wouldn't be able to provide 488 00:19:03,900 --> 00:19:06,299 universal access to all inmates. And it was 489 00:19:06,299 --> 00:19:08,140 a very controversial topic. As a matter of 490 00:19:08,140 --> 00:19:10,700 fact, many prison many, many inmates are suing 491 00:19:10,700 --> 00:19:11,839 state prison systems, 492 00:19:12,140 --> 00:19:14,720 for for for not providing them access to 493 00:19:15,015 --> 00:19:16,154 hepatitis c treatment. 494 00:19:17,015 --> 00:19:19,595 And our key research question was the following. 495 00:19:19,734 --> 00:19:21,654 As I mentioned, I'll be repeating myself. Does 496 00:19:21,654 --> 00:19:24,394 it make sense to expand health care spending 497 00:19:24,855 --> 00:19:26,154 for correctional facilities 498 00:19:27,015 --> 00:19:27,755 such that 499 00:19:28,679 --> 00:19:31,659 access to hepatitis c treatment could be expanded, 500 00:19:32,519 --> 00:19:34,220 for inmates behind the bars. 501 00:19:34,679 --> 00:19:38,299 And, this was a politically controversial question because, 502 00:19:38,359 --> 00:19:39,720 you know, because you know, I mean, I 503 00:19:39,720 --> 00:19:41,659 don't wanna get into too much into into 504 00:19:41,934 --> 00:19:43,934 political discussion. But as you may imagine, some 505 00:19:43,934 --> 00:19:46,255 some politicians are arguing that, you know, there 506 00:19:46,255 --> 00:19:48,815 is no point of further expanding health care 507 00:19:48,815 --> 00:19:49,315 spending, 508 00:19:49,934 --> 00:19:51,315 for correctional facilities. 509 00:19:52,335 --> 00:19:54,095 And what we did in our in in 510 00:19:54,095 --> 00:19:55,934 our analysis is we have taken a radical 511 00:19:55,934 --> 00:19:57,690 approach. Right? We have asked the following question. 512 00:19:57,690 --> 00:19:58,190 Right? 513 00:19:58,730 --> 00:20:00,829 We have taken a radical approach and asked, 514 00:20:01,529 --> 00:20:03,309 what happens if we expand 515 00:20:03,929 --> 00:20:05,630 treatment to hepatitis c 516 00:20:06,009 --> 00:20:08,990 in correctional facilities from a societal perspective 517 00:20:09,684 --> 00:20:12,025 for the for the for the community, right, 518 00:20:12,805 --> 00:20:14,025 outside the bars. 519 00:20:14,404 --> 00:20:16,184 And we have shown that 520 00:20:16,805 --> 00:20:17,625 it makes 521 00:20:18,164 --> 00:20:19,065 a lot of 522 00:20:19,445 --> 00:20:22,505 economic sense to treat hepatitis c 523 00:20:22,964 --> 00:20:23,464 while, 524 00:20:24,029 --> 00:20:26,670 in the correctional facilities because the idea is 525 00:20:26,670 --> 00:20:29,150 the average time spent in behind the virus 526 00:20:29,150 --> 00:20:30,529 is less than five years. 527 00:20:30,910 --> 00:20:34,690 Correctional facilities provide an ideal setting for identifying, 528 00:20:36,350 --> 00:20:37,570 infected inmates, 529 00:20:38,115 --> 00:20:40,835 treating them and linking them to care. Outside, 530 00:20:40,835 --> 00:20:41,494 you know, 531 00:20:42,115 --> 00:20:44,855 in the outside community, it's very harder to, 532 00:20:45,474 --> 00:20:48,434 diagnose patients. The prevalence is very lower. And 533 00:20:48,434 --> 00:20:50,934 also linking patients to care is more difficult. 534 00:20:51,234 --> 00:20:54,140 In correctional settings, it's almost like a laboratory 535 00:20:54,140 --> 00:20:56,720 environment, so linkage to care is way higher. 536 00:20:57,099 --> 00:20:58,960 And through, again, validated 537 00:21:00,140 --> 00:21:03,200 mathematical and computational models, we have shown that 538 00:21:03,980 --> 00:21:06,460 from a resource allocation standpoint, if you are 539 00:21:06,460 --> 00:21:08,484 going to spend as a society, if you'll 540 00:21:08,484 --> 00:21:11,365 spend any resources on hepatitis c treatment and 541 00:21:11,365 --> 00:21:11,865 elimination, 542 00:21:12,484 --> 00:21:14,105 it makes a lot of sense, 543 00:21:14,404 --> 00:21:17,384 rational economic sense to increase spending 544 00:21:18,005 --> 00:21:18,904 for for 545 00:21:19,445 --> 00:21:22,940 for reducing hepatitis c infections and treating hepatitis 546 00:21:23,080 --> 00:21:23,580 c, 547 00:21:24,440 --> 00:21:26,140 in the prison settings, community, 548 00:21:26,440 --> 00:21:29,080 in the in the correctional health settings, again, 549 00:21:29,080 --> 00:21:32,200 because of this intermingling happening upon release of 550 00:21:32,200 --> 00:21:33,019 these inmates. 551 00:21:34,034 --> 00:21:36,994 This was somewhat a contrary to finding in 552 00:21:36,994 --> 00:21:37,894 the sense that 553 00:21:38,194 --> 00:21:40,595 on one hand, you would think that, you 554 00:21:40,595 --> 00:21:42,514 know, it may not make sense to further 555 00:21:42,514 --> 00:21:45,634 increase health care spending for correctional stand you 556 00:21:45,634 --> 00:21:47,794 know, settings. Again, this is not my standing, 557 00:21:47,794 --> 00:21:50,179 but there are people who believe that way. 558 00:21:50,399 --> 00:21:52,319 On the other hand, we have taken this 559 00:21:52,319 --> 00:21:55,119 radical extreme approach and have shown that even 560 00:21:55,119 --> 00:21:55,779 if you 561 00:21:56,159 --> 00:21:58,559 don't care about the lives of these inmates, 562 00:21:58,559 --> 00:21:59,380 which is obviously 563 00:21:59,759 --> 00:22:01,599 unethical and doesn't make sense, but even if 564 00:22:01,599 --> 00:22:03,539 you have taken such an extreme approach, 565 00:22:04,115 --> 00:22:05,494 it still makes rational 566 00:22:05,795 --> 00:22:07,734 economic sense to prioritize, 567 00:22:08,434 --> 00:22:09,414 treatment access 568 00:22:09,715 --> 00:22:12,455 in the correctional settings from a societal perspective, 569 00:22:12,835 --> 00:22:14,215 because return on investment 570 00:22:14,674 --> 00:22:16,950 in such cases is higher because, you know, 571 00:22:16,950 --> 00:22:19,910 detecting those patients and treating these patients and 572 00:22:19,910 --> 00:22:22,809 linking them to care is easier in correctional 573 00:22:22,950 --> 00:22:23,450 settings. 574 00:22:24,150 --> 00:22:26,309 So this is one example. I can I 575 00:22:26,309 --> 00:22:28,470 can keep going? I can share more examples 576 00:22:28,470 --> 00:22:29,115 if you'd like. 577 00:22:29,994 --> 00:22:31,434 If I can share a second one and 578 00:22:31,434 --> 00:22:32,255 stop that. 579 00:22:32,634 --> 00:22:35,295 A second example is is in another project, 580 00:22:35,914 --> 00:22:37,054 we looked at, 581 00:22:38,234 --> 00:22:42,335 Medicare Advantage Plans. Medicare Advantage Plans are basically, 582 00:22:42,474 --> 00:22:43,855 you know, Medicare plans 583 00:22:44,390 --> 00:22:47,110 offered by private insurances. What what what happens 584 00:22:47,110 --> 00:22:48,890 is Medicare sort of outsources 585 00:22:49,430 --> 00:22:51,930 the provision health care of health care services 586 00:22:52,070 --> 00:22:55,509 to to private entities, such as UnitedHealthcare may 587 00:22:55,509 --> 00:22:58,035 be providing coverage for Medicare patients, and it's 588 00:22:58,035 --> 00:23:00,194 part it becomes part of the Medicare Advantage 589 00:23:00,194 --> 00:23:00,694 Plans. 590 00:23:02,115 --> 00:23:04,434 And it is well established that it is 591 00:23:04,434 --> 00:23:07,554 somewhat well understood in the published literature that, 592 00:23:07,794 --> 00:23:09,575 Medicare Advantage Plans 593 00:23:10,299 --> 00:23:12,940 tend to do risk selection, meaning that they 594 00:23:12,940 --> 00:23:14,480 select profitable patients, 595 00:23:15,900 --> 00:23:17,920 through some selection mechanisms. 596 00:23:19,420 --> 00:23:21,339 And the well under there is this well 597 00:23:21,339 --> 00:23:23,740 understood notion that, you know, they may be 598 00:23:23,740 --> 00:23:24,240 spending 599 00:23:24,859 --> 00:23:25,359 money 600 00:23:25,755 --> 00:23:26,255 allocated 601 00:23:26,795 --> 00:23:29,674 for high risk patients towards treatment of sorry. 602 00:23:29,674 --> 00:23:30,734 Money allocated 603 00:23:31,035 --> 00:23:33,994 for low risk patients towards treatments of high 604 00:23:33,994 --> 00:23:36,714 risk patients, and it's referred to as cross 605 00:23:36,714 --> 00:23:37,214 subsidization. 606 00:23:38,819 --> 00:23:40,900 In in one of our research papers, we 607 00:23:40,900 --> 00:23:41,799 have shown that 608 00:23:42,259 --> 00:23:44,339 the other way around is also correct, meaning 609 00:23:44,339 --> 00:23:47,460 that medic Medicare Advantage Plans tend to utilize 610 00:23:47,460 --> 00:23:49,799 resources allocated for high risk patients 611 00:23:52,005 --> 00:23:54,984 to affect low risk patients into their plans, 612 00:23:55,045 --> 00:23:57,684 you know, and through mechanisms such as such 613 00:23:57,684 --> 00:23:58,164 as, 614 00:23:58,805 --> 00:24:01,384 reduced gym memberships. Right? You know, some Medicare 615 00:24:01,445 --> 00:24:03,625 Advantage Plans offer reduced gym memberships. 616 00:24:03,924 --> 00:24:05,940 And if you think about reduced gym memberships, 617 00:24:05,940 --> 00:24:07,380 it sounds great. Right? You know, you would 618 00:24:07,380 --> 00:24:10,039 think that, you know, this is a great 619 00:24:10,820 --> 00:24:12,820 add on perk that they're offering, but when 620 00:24:12,820 --> 00:24:14,759 you carefully think about it, 621 00:24:15,859 --> 00:24:18,339 carefully think about who would be interested in, 622 00:24:18,820 --> 00:24:19,640 gym membership 623 00:24:20,054 --> 00:24:22,454 among elderly population. Right? We are talking about 624 00:24:22,454 --> 00:24:25,115 Medicare patients, patients over age 65. 625 00:24:26,534 --> 00:24:29,274 An individual suffering from multiple chronic conditions 626 00:24:29,575 --> 00:24:30,875 and not more you know, 627 00:24:31,894 --> 00:24:33,575 not easy to you know, not having a 628 00:24:33,575 --> 00:24:35,750 hard time with movements, etcetera, 629 00:24:36,130 --> 00:24:37,590 wouldn't be so much interested 630 00:24:38,369 --> 00:24:39,269 in gym membership. 631 00:24:39,890 --> 00:24:41,910 But healthy physically healthy, 632 00:24:42,210 --> 00:24:43,269 physically active, 633 00:24:43,570 --> 00:24:44,789 yeah, older individuals, 634 00:24:45,730 --> 00:24:48,369 would be more attracted to gym membership, reduced 635 00:24:48,369 --> 00:24:49,109 gym memberships. 636 00:24:49,505 --> 00:24:51,424 Even though at first, it sounds like a 637 00:24:51,424 --> 00:24:52,244 great perk, 638 00:24:52,785 --> 00:24:53,525 in reality, 639 00:24:53,904 --> 00:24:55,904 it is a selection mechanism. Right? It is 640 00:24:55,904 --> 00:24:58,305 a it is a mechanism primarily designed for 641 00:24:58,305 --> 00:24:58,805 attracting 642 00:25:00,305 --> 00:25:04,164 relatively healthy healthier older adults into their population. 643 00:25:04,650 --> 00:25:06,570 And we have shown that they are using 644 00:25:06,570 --> 00:25:10,509 basically these resources allocated for these relatively healthier, 645 00:25:11,289 --> 00:25:11,789 individuals, 646 00:25:12,570 --> 00:25:14,430 towards treatment of sicker patients. 647 00:25:18,244 --> 00:25:20,644 Wow. That's fascinating to hear. You know? And 648 00:25:20,644 --> 00:25:21,384 really, really, 649 00:25:22,005 --> 00:25:25,125 interesting to understand the applications of your research 650 00:25:25,125 --> 00:25:27,365 and dig a little bit deeper here. Jorge, 651 00:25:27,365 --> 00:25:28,804 thank you so much for joining us on 652 00:25:28,804 --> 00:25:30,884 the podcast today. I I really appreciate your 653 00:25:30,884 --> 00:25:33,140 time, and I look forward to, connecting with 654 00:25:33,140 --> 00:25:35,140 you again soon. Thank you so much for 655 00:25:35,140 --> 00:25:35,799 the opportunity.