WEBVTT 1 00:00:01.639 --> 00:00:06.400 This is parent data. I'm Emily Oster. When we talk 2 00:00:06.400 --> 00:00:11.600 about sea sections, it's often prefaced with unplanned or emergency. 3 00:00:11.720 --> 00:00:14.240 About a third of all the deliveries in the US 4 00:00:14.360 --> 00:00:17.760 are cesarean sections, and only about sixteen percent of those 5 00:00:17.880 --> 00:00:21.400 are planned, and that leaves a lot of mothers in 6 00:00:21.440 --> 00:00:25.680 a position where they're delivering differently than they planned or 7 00:00:25.720 --> 00:00:30.520 intended to. To be clear, for many women, sea sections 8 00:00:30.600 --> 00:00:34.400 are incredibly important our life saving and the World Health 9 00:00:34.479 --> 00:00:38.040 Organization says about ten to fifteen percent of birth should 10 00:00:38.080 --> 00:00:40.200 be sea sections to protect the health of the mother 11 00:00:40.479 --> 00:00:44.360 and the baby. But ten to fifteen percent isn't thirty percent, 12 00:00:44.840 --> 00:00:48.640 and abdominal surgery is never without risks. When we look 13 00:00:48.720 --> 00:00:51.360 at the difference between the share of sea sections we 14 00:00:51.360 --> 00:00:54.240 think should happen and the share that do happen, it 15 00:00:54.360 --> 00:00:58.080 seems like sometimes there are too many sea sections being performed, 16 00:00:59.080 --> 00:01:02.080 and in the US, a disproportionate number of those are 17 00:01:02.120 --> 00:01:08.000 being performed on Black women. There are a lot of 18 00:01:08.000 --> 00:01:11.240 possible reasons for the racial discrepancy, none of which are 19 00:01:11.280 --> 00:01:12.960 easily dismissible on their face. 20 00:01:13.800 --> 00:01:15.920 Black women may be at higher baseline health. 21 00:01:15.800 --> 00:01:18.920 Risk, They may be delivering in hospitals, where C sections 22 00:01:19.000 --> 00:01:22.160 are more frequent, they may request se sections more. 23 00:01:23.440 --> 00:01:25.240 None of those reasons seem to. 24 00:01:25.120 --> 00:01:29.760 Fit perfectly with the data, and understanding which of them 25 00:01:29.920 --> 00:01:33.319 is more important is core to understanding how we might 26 00:01:33.440 --> 00:01:34.880 change sea section rates. 27 00:01:35.720 --> 00:01:37.080 So how are we going to get to the root 28 00:01:37.080 --> 00:01:37.840 of what's going on? 29 00:01:38.640 --> 00:01:40.880 That's what my guest, Molly Schanell, is looking at in 30 00:01:40.920 --> 00:01:45.600 her paper, Drivers of Racial Differences in se Sections. Molly 31 00:01:45.680 --> 00:01:49.280 is an assistant professor of economics at Northwestern University who 32 00:01:49.440 --> 00:01:53.240 works on the causes and consequences of medical provider behavior 33 00:01:53.440 --> 00:01:59.360 on populations. In her paper, she finds that black mothers 34 00:01:59.360 --> 00:02:03.160 with unskeed sedule deliveries are twenty five percent more likely 35 00:02:03.400 --> 00:02:06.760 to deliver by C section than white mothers, and she 36 00:02:06.960 --> 00:02:11.280 argues that implicit racial bias among providers, or possibly even 37 00:02:11.280 --> 00:02:14.320 a financial incentive in hospitals to fill their operating rooms, 38 00:02:14.960 --> 00:02:19.680 may play a role in this racial gap. In this conversation, 39 00:02:19.720 --> 00:02:24.200 we talk in great nerdy detail about her incredibly meticulous 40 00:02:24.320 --> 00:02:28.840 methodology for separating out different explanations for trying to figure 41 00:02:28.880 --> 00:02:33.160 out what is causality versus correlation, and for whether we 42 00:02:33.200 --> 00:02:36.359 can actually figure out who's high risk and might need 43 00:02:36.360 --> 00:02:42.359 this procedure versus who isn't. On a personal note, this 44 00:02:42.440 --> 00:02:46.480 is an unsettling conversation about race and the medical system. 45 00:02:46.800 --> 00:02:50.160 Molly is doing really important work here to unpack these 46 00:02:50.200 --> 00:02:52.360 incredibly sensitive topics. 47 00:02:53.000 --> 00:02:54.680 I hope that it's a little bit of. 48 00:02:54.720 --> 00:02:59.079 Charting a path forward after the break, Professor. 49 00:02:58.680 --> 00:03:15.680 Molly Schnail, Molly Chanelle, thank you so much for joining 50 00:03:15.720 --> 00:03:16.520 me on parent Data. 51 00:03:17.200 --> 00:03:18.160 Yeah, thanks for having me. 52 00:03:19.000 --> 00:03:22.120 We're going to spend today talking about your new paper 53 00:03:22.800 --> 00:03:26.840 on race and C sections and inequality. But I would 54 00:03:26.919 --> 00:03:30.880 love if you would start by just introducing yourself. 55 00:03:30.480 --> 00:03:32.960 And saying what you do where you are. You can 56 00:03:33.000 --> 00:03:34.520 tell us about your kids if you want to. 57 00:03:35.640 --> 00:03:38.880 Yeah. So, I'm an assistant professor at of Economics at 58 00:03:38.880 --> 00:03:42.160 Northwestern and so most of my work focuses on the 59 00:03:42.200 --> 00:03:45.480 causes and consequences of provider behavior. So thinking about how 60 00:03:45.520 --> 00:03:49.240 individ visual clinicians make decisions in the presence of incentives 61 00:03:49.240 --> 00:03:51.840 and constraints and see sections is one of those decisions. 62 00:03:52.000 --> 00:03:54.440 So an interested in digiving to that into the data. 63 00:03:55.000 --> 00:03:55.440 Awesome? 64 00:03:55.720 --> 00:03:58.360 All right, So we're going to talk about this paper 65 00:03:58.400 --> 00:04:02.560 of yours, which is about racial differences in C section rates. 66 00:04:03.040 --> 00:04:04.920 But I want to sort of start at the. 67 00:04:04.880 --> 00:04:10.720 Beginning and set the stage both about what are those 68 00:04:10.800 --> 00:04:13.440 racial differences if you just sort of look out at 69 00:04:13.440 --> 00:04:16.680 the world, and the question of why this matters. So 70 00:04:16.839 --> 00:04:20.240 let's start actually with the second question. So why are 71 00:04:20.279 --> 00:04:24.280 you interested in C section rates as an important outcome 72 00:04:24.920 --> 00:04:25.440 to study? 73 00:04:26.320 --> 00:04:30.119 Yeah, so c sections are a procedure that, of course 74 00:04:30.160 --> 00:04:33.560 can be life saving for mother and baby if it's necessary. 75 00:04:34.160 --> 00:04:36.479 But we have very high rates of c sections in 76 00:04:36.520 --> 00:04:39.200 the US, rates that are so high that many would 77 00:04:39.200 --> 00:04:41.680 say are too high. And of course it's a major 78 00:04:41.800 --> 00:04:44.640 abdominal surgery that comes with the potential for a lot 79 00:04:44.680 --> 00:04:48.000 of risks and complications, both for mom and for baby. 80 00:04:48.120 --> 00:04:51.719 And so when we see really pronounced differences in rates 81 00:04:51.960 --> 00:04:54.720 of c sections between black and white moms, it leads 82 00:04:54.760 --> 00:04:56.960 to questions of whether or not that might be contributing 83 00:04:57.040 --> 00:05:00.000 to sort of persistent disparities in health outcomes between those 84 00:05:00.080 --> 00:05:00.800 two groups. 85 00:05:01.800 --> 00:05:03.560 So, what does the C section rate in the US 86 00:05:03.839 --> 00:05:05.720 look like now? On average? 87 00:05:06.400 --> 00:05:09.320 So it's in the low thirties, so somewhere between thirty 88 00:05:09.320 --> 00:05:12.840 two and thirty three percent. For white moms, it's going 89 00:05:12.839 --> 00:05:15.120 to be just below thirty percent, and for black moms 90 00:05:15.120 --> 00:05:18.279 it's closer to thirty five percent. And actually a lot 91 00:05:18.320 --> 00:05:20.200 of the data that we're going to be using for 92 00:05:20.240 --> 00:05:22.080 the study is going to come from New Jersey, and 93 00:05:22.120 --> 00:05:24.560 New Jersey has a really high C section rate, So 94 00:05:24.600 --> 00:05:27.839 it's going to be about forty five percent for black 95 00:05:27.880 --> 00:05:30.400 moms and about forty percent for white moms. 96 00:05:30.600 --> 00:05:35.720 Well, what is the correct SEA section rate and why? 97 00:05:35.800 --> 00:05:36.640 Is that a hard question? 98 00:05:37.320 --> 00:05:41.400 So it depends on who you ask, right, the who 99 00:05:41.600 --> 00:05:44.080 is going to say somewhere between ten and fifteen percent. 100 00:05:44.920 --> 00:05:47.600 That's a very difficult question to answer because it's going 101 00:05:47.640 --> 00:05:50.880 to depend on a mother's appropriateness for the procedure, right, 102 00:05:50.920 --> 00:05:53.159 So that's going to differ across populations, that's going to 103 00:05:53.160 --> 00:05:56.440 differ across countries. That could differ for a given mom 104 00:05:56.520 --> 00:05:58.800 throughout her life cycle. She might be a good candidate 105 00:05:58.800 --> 00:06:01.720 for one for a C section for one of her 106 00:06:01.720 --> 00:06:04.960 births and not for another based on her underlying conditions. 107 00:06:05.160 --> 00:06:08.280 And so it's ultimately, you know, closely tied to medical 108 00:06:08.360 --> 00:06:11.520 risk and that's something that changes. And so one of 109 00:06:11.520 --> 00:06:13.240 the goals that we're going to really look at, and 110 00:06:13.600 --> 00:06:15.960 one of the goals of this study is to use 111 00:06:16.000 --> 00:06:18.240 a lot of information on medical risk to see whether 112 00:06:18.320 --> 00:06:20.599 or not that's one of the reasons why we see 113 00:06:20.640 --> 00:06:22.640 these big differences in C section rates. 114 00:06:22.920 --> 00:06:27.039 Yeah, so when we see these, if we sort of 115 00:06:27.120 --> 00:06:29.680 establish that a C section, while safe and can be 116 00:06:29.800 --> 00:06:32.159 life saving, that the rates may be too high, and 117 00:06:32.240 --> 00:06:34.720 that it is, at least in the short term, a 118 00:06:34.800 --> 00:06:38.279 much more complicated recovery and may have contributions to morbidity. 119 00:06:39.040 --> 00:06:40.599 The way I'd often like to say this is like, 120 00:06:40.960 --> 00:06:42.840 this can be a great option if you need it, 121 00:06:43.000 --> 00:06:46.000 but it's not likely to be the first choice for 122 00:06:46.120 --> 00:06:50.320 delivery method. Yes, So when we look out at the 123 00:06:50.839 --> 00:06:55.000 world and we see differences in C section rates, either 124 00:06:55.480 --> 00:06:59.039 a cross space or across racial groups, what are the 125 00:06:59.279 --> 00:07:00.480 possible set. 126 00:07:00.120 --> 00:07:01.279 Of reasons we could see that. 127 00:07:01.320 --> 00:07:02.839 So I would say, you know, you said New Jersey 128 00:07:02.880 --> 00:07:05.520 is very high and that there are big differences across 129 00:07:05.600 --> 00:07:06.119 racial groups. 130 00:07:06.240 --> 00:07:08.440 So as a researcher, as a person who says that, 131 00:07:08.640 --> 00:07:11.320 what are the set of things that might be going on. 132 00:07:11.920 --> 00:07:15.560 Yeah, that's a that's a great question. So the the 133 00:07:15.600 --> 00:07:18.080 high sea section rates in general and the disparities by 134 00:07:18.160 --> 00:07:21.320 race are really well documented and quite persistent, and so 135 00:07:21.520 --> 00:07:24.160 that's led a lot of people to hypothesize about what 136 00:07:24.240 --> 00:07:26.760 could be driving that. It's going to be important to 137 00:07:26.840 --> 00:07:29.480 know that to be posible point to different policy solutions. 138 00:07:30.080 --> 00:07:32.360 So what if people thought, well, people have argued that 139 00:07:32.440 --> 00:07:35.360 it could be differences in what we would call maternal preferences. 140 00:07:35.680 --> 00:07:37.960 So it could be, for example, if we see that 141 00:07:38.000 --> 00:07:40.280 black moms are much more likely to have a sea 142 00:07:40.280 --> 00:07:43.520 section than white mothers, maybe it's that black mothers prefer 143 00:07:43.560 --> 00:07:45.880 to have sea sections, So maybe they're requesting them from 144 00:07:45.920 --> 00:07:48.920 their providers and that's why they're higher up rates. That, 145 00:07:48.960 --> 00:07:51.200 of course, would go against a lot of other things 146 00:07:51.240 --> 00:07:53.480 we know about how Black Americans tend to interact with 147 00:07:53.520 --> 00:07:57.520 the healthcare system. Given a history of racism within the 148 00:07:57.560 --> 00:07:59.400 healthcare system in the US, we tend to see that 149 00:07:59.440 --> 00:08:02.560 Black America want less intensive treatment as opposed to more 150 00:08:02.560 --> 00:08:05.240 intensive treatment, and so it would be surprising if c 151 00:08:05.400 --> 00:08:08.160 sections sort of when in the opposite direction. But that's 152 00:08:08.200 --> 00:08:10.880 one of the hypotheses that you'll hear of. Maybe it's 153 00:08:10.960 --> 00:08:15.880 differences in maternal demand. Another potential that explanation that we've 154 00:08:15.880 --> 00:08:18.440 already sort of touched on, would be differences in health 155 00:08:18.520 --> 00:08:21.239 risks at the point of delivery, right, So it could 156 00:08:21.280 --> 00:08:24.360 be that when black moms show up at hospitals ready 157 00:08:24.400 --> 00:08:27.280 to deliver that they are just much better candidates for 158 00:08:27.360 --> 00:08:30.400 the procedure. Right, envision that all black moms show up 159 00:08:30.440 --> 00:08:33.880 with babies in a breach presentation and they have justtational 160 00:08:33.880 --> 00:08:36.199 diabetes and a lot of other risk factors that would 161 00:08:36.240 --> 00:08:38.840 make them better candidates for the procedure. Well, then that 162 00:08:38.880 --> 00:08:41.040 could be what's driving the higher rates and more black 163 00:08:41.040 --> 00:08:44.400 moms than white moms. It could also be differences in 164 00:08:44.440 --> 00:08:48.400 the hospitals or the providers that mothers of different races 165 00:08:48.480 --> 00:08:51.360 go to. So we know that there are very large 166 00:08:51.360 --> 00:08:55.480 differences in C section rates across different hospitals, and given 167 00:08:55.880 --> 00:08:59.040 segregation by race, we tend to see that different populations 168 00:08:59.160 --> 00:09:02.720 go to different hospitals, and perhaps black women are just 169 00:09:02.800 --> 00:09:05.400 more likely to be going to hospitals that have higher 170 00:09:05.400 --> 00:09:06.240 C section rates. 171 00:09:06.440 --> 00:09:08.800 I think that would be the question, why does that 172 00:09:08.840 --> 00:09:12.440 hospital have let exactly, I was Alina, but I. 173 00:09:12.480 --> 00:09:14.679 Yeah, this is just going to kick the can a 174 00:09:14.720 --> 00:09:16.160 little bit, and that you know, we'd still have to 175 00:09:16.160 --> 00:09:18.559 figure out why those rates are different within those hospitals. 176 00:09:18.800 --> 00:09:21.120 But maybe once you look in the same hospital or 177 00:09:21.120 --> 00:09:24.439 within the same provider, you see no racial differences between 178 00:09:24.600 --> 00:09:27.679 C section rates, and it's really just selection into different hospitals. 179 00:09:28.400 --> 00:09:31.200 Of course, it could also be something about provider discretion, 180 00:09:31.880 --> 00:09:35.079 either that providers are more likely to do C sections 181 00:09:35.120 --> 00:09:38.199 for black moms because they're more worried about something going wrong, 182 00:09:38.760 --> 00:09:40.960 or maybe there are just general biases in the healthcare 183 00:09:40.960 --> 00:09:43.240 system that we observe in a lot of other settings 184 00:09:43.280 --> 00:09:43.800 in the US. 185 00:09:44.360 --> 00:09:47.079 I think the reason that thinking about these explanations and 186 00:09:47.120 --> 00:09:49.360 trying to distinguish them between them is so important is 187 00:09:49.400 --> 00:09:55.400 that we can't really think about solutions without understanding the problem. 188 00:09:55.760 --> 00:09:57.640 So if you said, you know, I'd like to lower 189 00:09:57.640 --> 00:09:59.400 this disparity, I'd like to lower the rates, I'd like 190 00:09:59.440 --> 00:10:03.800 to lower it. In general, just observing this very robust 191 00:10:03.920 --> 00:10:06.800 correlation or differences across groups is actually going to tell 192 00:10:06.840 --> 00:10:09.000 you nothing about how to fix it. And those explanations 193 00:10:09.000 --> 00:10:12.760 you just gave me all suggest pretty different approaches to 194 00:10:13.840 --> 00:10:16.440 whether this is a fixable problem or something we could 195 00:10:16.440 --> 00:10:18.559 affect and how we would do it. I mean, you're 196 00:10:18.559 --> 00:10:20.600 going to approach us pretty differently if it's something about 197 00:10:20.600 --> 00:10:24.480 differences across hospitals versus something about differences across provider discretion, 198 00:10:24.880 --> 00:10:28.440 and certainly if it's something about differences in risk factors. 199 00:10:29.000 --> 00:10:31.440 Yeah, absolutely, I think it. Also, you know, when you 200 00:10:31.480 --> 00:10:34.199 hear about this disparity, you sort of naturally think that 201 00:10:34.200 --> 00:10:37.080 that's probably a bad thing. But if it's differences in risk, 202 00:10:37.280 --> 00:10:39.800 or if it's differences in preferences, well then maybe we 203 00:10:39.840 --> 00:10:41.680 shouldn't be so worried about it. We should want to 204 00:10:41.720 --> 00:10:44.559 address those differences in medical risk. But at the point 205 00:10:44.600 --> 00:10:46.840 that a mother shows up to deliver, we don't want 206 00:10:46.880 --> 00:10:49.400 clinicians to reduce C sections among black moms if they 207 00:10:49.440 --> 00:10:51.440 really just are more appropriate for the procedure. 208 00:10:52.880 --> 00:10:56.000 So your goal in this research paper, which is entitled 209 00:10:56.080 --> 00:10:58.640 Drivers of Racial Differences in C sections, is to try 210 00:10:58.640 --> 00:11:02.080 to unpack a bit but what in fact. 211 00:11:01.920 --> 00:11:03.040 Is going on? 212 00:11:03.520 --> 00:11:05.440 So you're going to start this paper by trying to 213 00:11:05.520 --> 00:11:10.360 understand whether the differences across racial groups are about differences 214 00:11:10.360 --> 00:11:15.120 in pre existing risk, whether there's something that we could 215 00:11:15.120 --> 00:11:17.520 point to and say people more likely I have this 216 00:11:17.559 --> 00:11:19.760 complication that makes the C section more likely? And so 217 00:11:21.080 --> 00:11:24.280 do you do this in a very specific way using 218 00:11:24.440 --> 00:11:27.440 very fancy methods? So could you talk me through a 219 00:11:27.440 --> 00:11:30.320 little bit, you know, what you do and why it 220 00:11:30.360 --> 00:11:33.320 was important to approach it in this like quite comprehensive, 221 00:11:33.679 --> 00:11:34.480 complicated way. 222 00:11:35.240 --> 00:11:39.120 Yeah. Absolutely. So Basically, when people have looked at this 223 00:11:39.200 --> 00:11:41.360 in the past or you know, where do we get 224 00:11:41.400 --> 00:11:44.280 these statistics on sea section rates by race and over time? 225 00:11:44.440 --> 00:11:47.440 Those tend to come from birth certificate data, right, so 226 00:11:47.520 --> 00:11:50.080 you can get really good information on when you're at 227 00:11:50.120 --> 00:11:52.200 the hospital and you fill out that birth certificate. After 228 00:11:52.200 --> 00:11:55.600 your child's born, those are digitized and researchers get access 229 00:11:55.640 --> 00:11:58.079 to those and you can track things like whether or 230 00:11:58.120 --> 00:11:59.400 not people at sea sections. 231 00:11:59.520 --> 00:12:03.079 Just to be people, this doesn't have your name on it, 232 00:12:03.160 --> 00:12:05.040 so yes, people would not be able to find you 233 00:12:05.080 --> 00:12:07.640 with this, but the information from the firth certificate so 234 00:12:07.920 --> 00:12:11.199 still whatever that would like comic and go remove themselves from. 235 00:12:11.400 --> 00:12:14.400 Yes, these are anonymized, but they're very useful for researchers 236 00:12:14.440 --> 00:12:18.120 in order to track these trends over time. Now that's useful, 237 00:12:18.120 --> 00:12:20.079 and that it has a lot of information, it's not 238 00:12:20.160 --> 00:12:21.720 very useful, and that it doesn't have a lot of 239 00:12:21.720 --> 00:12:24.760 information about whether or not we think them or what 240 00:12:24.800 --> 00:12:26.679 the mother was presenting with that would cause her to 241 00:12:26.720 --> 00:12:29.160 either need to see section or not. So what we're 242 00:12:29.160 --> 00:12:31.240 going to do is we're going to get really detailed 243 00:12:31.280 --> 00:12:34.400 administrative data from the state of New Jersey. So again 244 00:12:34.440 --> 00:12:36.280 this isn't going to have individuals names, so this will 245 00:12:36.280 --> 00:12:39.079 be anonymized on the patient's side, But we're basically going 246 00:12:39.080 --> 00:12:42.120 to see everything that would be in the mother's record, 247 00:12:42.200 --> 00:12:43.880 and so we'll know sort of all of our health 248 00:12:43.880 --> 00:12:46.720 conditions leading up to birth. Will actually also be able 249 00:12:46.720 --> 00:12:49.640 to follow mothers and babies following the delivery in order 250 00:12:49.679 --> 00:12:52.000 to look at how this might influence their health later on. 251 00:12:52.600 --> 00:12:55.040 But with that, we're just presented with essentially a wealth 252 00:12:55.040 --> 00:12:57.880 of information about, you know, what the clinician would have 253 00:12:57.960 --> 00:13:01.079 known about the mom at the point of making that decision, 254 00:13:01.240 --> 00:13:04.320 that decision about how to deliver the baby. And so 255 00:13:04.360 --> 00:13:06.760 we're in an age of big data and machine learning, 256 00:13:07.080 --> 00:13:09.440 and so basically what we do is we give the 257 00:13:09.480 --> 00:13:12.520 computer all of this information and we have it help 258 00:13:12.640 --> 00:13:15.080 us figure out how to predict whether or not the 259 00:13:15.120 --> 00:13:17.640 mother was going to get a sea section. And so 260 00:13:17.760 --> 00:13:19.559 what you can think of this as doing is it's 261 00:13:19.600 --> 00:13:24.120 basically figuring out over a decade worth of data of 262 00:13:24.200 --> 00:13:27.000 which characteristics are going to be most likely to be 263 00:13:27.040 --> 00:13:29.679 observed among mothers who end up having a C section. 264 00:13:30.040 --> 00:13:33.200 So you can think of this as basically capturing medical consensus, 265 00:13:33.440 --> 00:13:36.040 and so if a mom has a breach presentation, it's 266 00:13:36.320 --> 00:13:38.319 very likely that she's going to have a sea section, 267 00:13:38.679 --> 00:13:40.360 and so the model's going to tell us that that's 268 00:13:40.400 --> 00:13:43.120 an important characteristic that predicts whether or not that mother 269 00:13:43.200 --> 00:13:45.320 is going to need a sea section. We're also going 270 00:13:45.360 --> 00:13:48.520 to see things like diabetes, obesity, putting on a lot 271 00:13:48.520 --> 00:13:51.880 of weight during pregnancy, maternal age, all of these are 272 00:13:51.880 --> 00:13:54.440 going to strongly predict whether or not a mom is 273 00:13:54.440 --> 00:13:56.600 going to have a C section. And so what that's 274 00:13:56.679 --> 00:13:58.920 going to allow us to do then is for each mother, 275 00:13:59.280 --> 00:14:02.160 once we've asked maated this model using all of this data, 276 00:14:02.559 --> 00:14:05.199 we can then basically predict for each mother what her 277 00:14:05.240 --> 00:14:09.400 appropriateness is for a C section given medical consensus. So 278 00:14:09.480 --> 00:14:11.640 on average, would she had gotten one from these other 279 00:14:11.679 --> 00:14:13.560 clinicians throughout the time period. 280 00:14:14.240 --> 00:14:17.120 And so I mean this this approach it shares some 281 00:14:17.160 --> 00:14:19.720 features with a regression approach of sort of trying to 282 00:14:19.760 --> 00:14:27.480 think about basically holding constant characteristics across people, but doing 283 00:14:27.520 --> 00:14:30.720 it in a way where the thing you're holding constant 284 00:14:30.800 --> 00:14:33.760 is in some ways way way richer and much closer 285 00:14:33.800 --> 00:14:36.160 to holding constant the actual thing you care about, which 286 00:14:36.200 --> 00:14:39.800 is when this person showed up, what was the likelihood 287 00:14:40.040 --> 00:14:42.240 based on everything you would see about them when they arrived, 288 00:14:42.280 --> 00:14:44.120 what was the likelihood they could have the sea section, 289 00:14:44.240 --> 00:14:45.240 they would have a C section? 290 00:14:45.280 --> 00:14:48.160 And you sort of have that number produced by this. 291 00:14:48.760 --> 00:14:50.080 You guys use a random forest. 292 00:14:50.240 --> 00:14:52.920 We use a random forest exactly, and so it's exactly 293 00:14:52.960 --> 00:14:54.800 as you describe it. You know, we can think of 294 00:14:54.840 --> 00:14:57.120 it similar to our regression. It's just going to have 295 00:14:57.160 --> 00:14:59.560 a lot of nice properties that we tend to like 296 00:15:00.280 --> 00:15:02.880 in research, and that's going to do a very good 297 00:15:02.960 --> 00:15:04.920 job of predicting whether or not you actually are going 298 00:15:04.960 --> 00:15:05.600 to have a C section. 299 00:15:06.120 --> 00:15:08.920 Do you have a sense of how good these these 300 00:15:09.000 --> 00:15:12.000 random forest models have sort of metrics of how how 301 00:15:12.040 --> 00:15:12.800 predictive they are. 302 00:15:12.840 --> 00:15:14.080 How good is your random forest? 303 00:15:14.560 --> 00:15:16.240 So what we can do is we can, you know, 304 00:15:16.480 --> 00:15:21.160 sort mothers in the testing sample into groups based on 305 00:15:21.200 --> 00:15:22.560 whether or not they need one, and we're going to 306 00:15:22.640 --> 00:15:24.640 basically be able to predict it for you. So the 307 00:15:24.960 --> 00:15:26.880 women who are not in our sample that we're using 308 00:15:26.920 --> 00:15:29.520 to train the model, we can kick out this number 309 00:15:29.520 --> 00:15:31.400 of whether or not we think you're going to have 310 00:15:31.480 --> 00:15:34.080 a C section. We can group you into different groups 311 00:15:34.120 --> 00:15:36.680 based on that average risk, and then we can compute 312 00:15:36.760 --> 00:15:39.320 the true chance that mothers in that group actually have 313 00:15:39.440 --> 00:15:41.480 a C section, and they're going to line up, they're 314 00:15:41.480 --> 00:15:44.200 going to match almost perfectly. So I think this also 315 00:15:44.480 --> 00:15:46.680 tells you there's a lot of discretion in the decision 316 00:15:46.680 --> 00:15:48.320 of whether or not to have a C section that 317 00:15:48.480 --> 00:15:50.600 I'm sure we'll get into it and in just a bit, 318 00:15:50.640 --> 00:15:52.440 and you know, we'll see some patterns in the data 319 00:15:52.440 --> 00:15:55.040 that are going to really highlight the role of discretion. 320 00:15:55.480 --> 00:15:57.120 But at the same time, there are a lot of 321 00:15:57.200 --> 00:16:00.360 characteristics that are gonna, you know, really predict whether or 322 00:16:00.400 --> 00:16:02.360 not you're a good candidate for the procedure or not. 323 00:16:02.680 --> 00:16:06.320 Yeah, okay, so let's start. Let's get into what you 324 00:16:06.640 --> 00:16:08.840 what you find. So I want to start with the 325 00:16:09.400 --> 00:16:14.000 sort of baseline difference. So there's a racial difference in 326 00:16:14.080 --> 00:16:17.760 C section rates, and you're focusing here, it's really useful to. 327 00:16:17.680 --> 00:16:20.200 Say on unscheduled c sections. 328 00:16:20.320 --> 00:16:23.880 So we're already kind of taking out the possibility, although 329 00:16:23.920 --> 00:16:27.120 I agree it's remote and not true, but the possibility 330 00:16:27.160 --> 00:16:31.200 that there are differences in kind of what people request 331 00:16:31.320 --> 00:16:34.080 or even what is sort of happening at the doctor 332 00:16:34.080 --> 00:16:36.600 beforehand in terms of how how you negotiate your things. 333 00:16:36.600 --> 00:16:39.800 These are people who come in with an expectation of 334 00:16:40.240 --> 00:16:43.440 laboring and having a vaginal birth and then there is 335 00:16:43.480 --> 00:16:47.600 a C section or not afterwards. So with that sample, 336 00:16:47.640 --> 00:16:51.160 what's the kind of headline racial difference in C section risk? 337 00:16:51.680 --> 00:16:53.680 Yes, so we're going to see that above mothers with 338 00:16:53.840 --> 00:16:57.440 unscheduled deliveries in New Jersey, Black mothers are going to 339 00:16:57.480 --> 00:17:00.240 be twenty five percent more likely to have a C 340 00:17:00.400 --> 00:17:04.040 section than white mother's And I think that's already quite striking. 341 00:17:04.520 --> 00:17:06.959 Why I agree with you that I didn't think this 342 00:17:07.040 --> 00:17:09.160 was going to be differences in maternal demand. I don't 343 00:17:09.200 --> 00:17:11.000 often hear a lot of moms sort of begging to 344 00:17:11.000 --> 00:17:13.240 have a sea section. But if you really want to 345 00:17:13.280 --> 00:17:15.520 have a C section, you can certainly find a provider 346 00:17:15.560 --> 00:17:18.280 who's willing to schedule one for you. And so that's 347 00:17:18.320 --> 00:17:21.160 the reason we're focusing on unscheduled deliveries and is also 348 00:17:21.240 --> 00:17:23.240 something that we people haven't been able to do in 349 00:17:23.240 --> 00:17:25.920 previous data. So again going back to that birth certificate 350 00:17:26.000 --> 00:17:28.840 data that again is anonymized, it will just say whether 351 00:17:28.880 --> 00:17:30.959 you had a C section or not. It's not going 352 00:17:31.000 --> 00:17:33.480 to say whether that c section happened after a trial 353 00:17:33.520 --> 00:17:36.040 of labor, whereas in our data we can see that 354 00:17:36.160 --> 00:17:39.159 mom shows up in labor, has a trial of labor, 355 00:17:39.400 --> 00:17:42.040 and that ultimately ends in a C section. Those are 356 00:17:42.080 --> 00:17:43.760 the cases that we're going to look at to sort 357 00:17:43.800 --> 00:17:46.480 of exclude these women who already scheduled the C section 358 00:17:46.560 --> 00:17:48.560 in advance, they knew they were coming in to have 359 00:17:48.600 --> 00:17:51.480 a C section. Why we think if you show up, 360 00:17:51.680 --> 00:17:54.440 you didn't schedule a CE section, you show up to 361 00:17:54.600 --> 00:17:56.560 have a baby and it ends in a C section. 362 00:17:56.920 --> 00:17:58.919 Those are the mothers who have signaled that their preferred 363 00:17:59.000 --> 00:18:01.440 method would have been a bat delivery and then ultimately 364 00:18:01.480 --> 00:18:02.440 proceeded to a C section. 365 00:18:03.680 --> 00:18:06.399 So you see an increase, and we should say it's 366 00:18:06.440 --> 00:18:08.680 twenty five it's about twenty five percent, not twenty five 367 00:18:08.720 --> 00:18:09.480 percentage points. 368 00:18:09.560 --> 00:18:10.560 It's twenty five. 369 00:18:10.640 --> 00:18:14.480 So it's four point two percentage points over the base 370 00:18:14.520 --> 00:18:16.200 for white moms is twenty five percent. 371 00:18:16.720 --> 00:18:19.320 So I think the thing that's really striking for me, 372 00:18:19.440 --> 00:18:23.720 even the baseline statistics, is that this difference is in 373 00:18:23.800 --> 00:18:28.280 percentage terms, much larger for people who's like ex Ante, 374 00:18:28.680 --> 00:18:32.560 coming in risk of a C section is lower. So 375 00:18:32.920 --> 00:18:34.679 there's a set of people you have like kind of 376 00:18:34.760 --> 00:18:37.639 they came in, you thought maybe they were unlikely to 377 00:18:37.680 --> 00:18:40.000 have a sea section, and then for black moms, that 378 00:18:40.119 --> 00:18:42.919 group is just way elevated and their risk relative to 379 00:18:42.920 --> 00:18:46.080 the group that came in and it seemed likely they 380 00:18:46.119 --> 00:18:48.560 have a C section where there's an elevation but it's smaller. 381 00:18:49.320 --> 00:18:51.280 Yeah, I think this is one of the most striking 382 00:18:51.320 --> 00:18:53.840 findings that just comes out of the data that if 383 00:18:53.840 --> 00:18:56.000 you we talked about how we get that measure of 384 00:18:56.080 --> 00:18:59.679 risk for moms, and if we group moms into risks, 385 00:18:59.680 --> 00:19:01.639 so we have very low risk moms, so think in 386 00:19:01.680 --> 00:19:03.879 the lowest quintile and then moms you know, all the 387 00:19:03.920 --> 00:19:06.879 way up to the highest risk. There there is a disparity, 388 00:19:06.920 --> 00:19:09.600 but it's very small for mothers in you know, once 389 00:19:09.640 --> 00:19:12.840 you get to the highest risk quintile. Why most moms 390 00:19:12.840 --> 00:19:14.879 that are getting sea sections right as they should and 391 00:19:14.920 --> 00:19:17.760 so there's not much of a disparity. What is driving 392 00:19:17.920 --> 00:19:21.199 the underlying racial disparity? We see it's low risk moms. 393 00:19:21.400 --> 00:19:23.199 So if you show up to the hospital, these low 394 00:19:23.280 --> 00:19:25.880 risk moms essentially have nothing in their record that would 395 00:19:25.920 --> 00:19:28.119 suggest that they're going to need a sea section. We 396 00:19:28.160 --> 00:19:30.359 see that black moms are going to be almost one 397 00:19:30.440 --> 00:19:33.400 hundred and fifty percent more likely to have an unscheduled 398 00:19:33.400 --> 00:19:36.639 sea section, so one point five times more likely, and 399 00:19:36.680 --> 00:19:38.440 that's what's driving that overall disparity. 400 00:19:38.800 --> 00:19:41.520 I mean, this paper is really really good, but it's 401 00:19:41.560 --> 00:19:43.919