top of page

#017 - 📑 Journal Club 9

 NICU journal club the incubator podcast

The articles covered on today’s episode of the podcast can be found here 👇

Hypothermia for moderate or severe neonatal encephalopathy in low-income and middle-income countries (HELIX): a randomised controlled trial in India, Sri Lanka, and Bangladesh.

Neonatal resuscitation in the NICU; Challenges beyond NRP.

Improving Compliance with a Rounding Checklist through Low- and High-technology Interventions: A Quality Improvement Initiative.

Initial Laparotomy Versus Peritoneal Drainage in Extremely Low Birthweight Infants With Surgical Necrotizing Enterocolitis or Isolated Intestinal Perforation: A Multicenter Randomized Clinical Trial.

Automated Explainable Multidimensional Deep Learning Platform of Retinal Images for Retinopathy of Prematurity Screening.

Neonatal mortality prediction with routinely collected data: a machine learning approach.


The transcript of today's episode can be found below 👇


babies, paper, nicu, infants, data, outcomes, laparotomy, algorithm, therapeutic hypothermia, diagnosis, interesting, autism spectrum disorder, artificial intelligence, work, moderate, called, rounding, resuscitation, nrp, looked


Daphna, Ben

Ben 00:04

Hello, everybody. Welcome back to the podcast. Dr. Barbeau. How are you?

Daphna 00:47

I'm good. It's been a busy weekend, my house like lots of our listeners. My little one went went back to school this this week. So that's been stressful. But the podcast has been keeping us busy, hasn't it?

Ben 01:02

That's right. For the people who want to get an inside look at what it takes to record the incubator. This is our second take for this episode, because we did record it the first time and there was some issues with the recording. So here we are re recording. So I think what it means for the audience is going to be a more polished episode, more fleshed out discuss, hopefully, but you know, consistency at all costs. We are not missing a week. We're really

Daphna 01:29

not Yeah, we're doing everything we can.

Ben 01:34

So this, this week's list of episodes of, I'm sorry, of journals and articles is quite interesting. And there's so many of them, we should not waste too much time. And we should try to get right into it. So the first paper in the first article I wanted to talk about was a very significant article that was published in The Lancet. And it was published online on August 3, and it's called hyperthermia for moderate or severe neonatal encephalopathy in low income and middle income countries. A randomized controlled trial in India, Sri Lanka, and Bangladesh. The list of authors is very long. It's the helix trial. And it's a list of authors that comprises the helix Consortium. So what was very interesting is that the objective of the paper was to examine whether therapeutic hypothermia initiated within six hours of birth, and that lasted like a normal protocol for 72 hours, reduced death or disability at 18 to 22 months when compared with usual care. And really, the reason why this question was posed was because of the fact that this will happen in the context of a low and middle income country, rather than in the US where obviously, the context at least is is different. The the trial, the helix trial was a randomized, open label multicountry trial that involved India, Sri Lanka, Bangladesh, and the units that were participating, were what we would consider level three units they were they had a lot of tools at their disposal to care for these babies. In the paper that you mentioned that they all had capabilities for assisted ventilation, cardiovascular monitoring and support and they had Tesla MRIs, and they had nitric oxide, and so on and so forth. And so I think this is the first thing that we wanted to discuss was the fact that this paper despite the fact that it's taking place in low and middle income countries, it was very thoroughly and rigorously designed very thoroughly and rigorously conducted, and the methodology was very sound. They included babies who were born at 36 weeks of gestation or later, and that weighed 1.8 kilos or more within six hours of birth. And their eligibility criteria involves two parameters, the first one being the need for continued resuscitation at five minutes of age, an Apgar score of less than six at five minutes or the or both, or an absence of crying by five minutes of age. The second parameter was evidence of moderate or severe encephalopathy at anytime assessed between one and six hours by the clinician using a sort of modified sign out scoring system to assess for potential encephalopathy. And so the first thing that pops to the attention of the reader is the fact that there's no metabolic acidosis component. And we'll get into that this was done purposefully by the group because some of these babies were born at outside or outlying hospitals or in the home, and they did not want to have to exclude potential babies because of the delay in transfer to the tertiary care center. The babies were assigned to either therapeutic hypothermia, which looked very similar to what we're used to do here in the US and around the world, which was cooling of the patients using whole body sort of cooling device that was actually a device used from the UK to 33.5 degrees Celsius for 72 hours with automated rewarming at point five degrees Celsius per hour on a servo controlled sort of device, the baby's work in the control group received what they called usual neonatal care, which included mechanical ventilation, I inotropes, avoidance of iatrogenic hyperthermia by restricting the use of overhead radiant warmers during resuscitation were possible and setting lower threshold in the server controlled warmers which means they were trying not to make them warmer than they needed to be, but they were not also getting cold to the to the to the temperature that would meet criteria for therapeutic hypothermia. And obviously, they would correct any metabolic or coagulative derangements.

Daphna 05:45

Yeah, that's that is it. That's an interesting point is that I mean, they really went out of their way to prevent hyperthermia, which we know is a big predictor of even poor outcomes.

Ben 05:59

And so they didn't follow these babies at 18 to 22 months of age. And they and they gave the put them through the Bailey, the third edition of the Bailey and and assess their their long term neurodevelopmental outcomes. The primary outcome was death or moderate or severe disability and just for the purpose of being thorough with the Define severe disability as any of the following cognitive score of less than 70. On the Bailey, a gross motor function classification system level three to five, a profound hearing impairment requiring hearing aids or a cochlear implant or blindness. Then they had a moderate disability, which was defined as a cognitive composite score of 70 to 84. And one or more of either a gross motor function classification system to hearing impairment with no amplification, or a persistent seizure disorder assessed at 18 to 22 months of age. So they had a large number of babies 408 infants were randomly assigned. And most of them actually turned out to be from the from the units in India, so 347 from the unit in India, 33 in Bangladesh, and 28 in Sri Lanka, 202 infants were assigned to the therapeutic hypothermia group 206 were allocated to the control group. And the primary outcomes was available for 97% of the people in the hypothermia group versus 97% in the control group. And again, remember, these are outcomes at 18 to 22 months of age. That's pretty good. Yeah, amazing. Yeah. The mean age at follow up is about 20 months. And the outcomes were death or moderate or severe disability occurred in 50% of infants in the hyperthermia group versus 47% in the control group, and then of the 17 predefined secondary outcomes. Eight differed significantly between the groups, but all were worse in the hypothermia group. And these you can find in Table three of the paper and they involved intracranial hemorrhage, gastric bleeding, persistent hypertension, pulmonary hemorrhage PPH, and prolonged blood coagulation culture, positive sepsis, neck, aerated cardiac arrhythmias, thrombocytopenia, metabolic acidosis, renal failure, pneumonia, subcutaneous fat necrosis, hospitals, seatstay, length of hospital stay death before discharge, and abnormal neurological exam after at the time of discharge. And so these are the these are the main results, and I want to give them a chance to see what she thought I want to read the last part of the abstract, which is the interpretation of their results because it is extremely, it's very strongly worded right, they said therapeutic hypothermia did not reduce the combined outcome of death or disability at 18 months after neonatal encephalopathy in low income and middle income countries, but significantly increased deaths alone. Therapeutic hypothermia should not be offered as treatment for neonatal encephalopathy in low income and middle income countries, even when tertiary neonatal intensive care facilities are available. I have was baffled by this type of words. And this the way the conclusion is worded in the paper that is so influential as the Lancet. And so I'm curious to hear what your thoughts were definitely.

Daphna 09:06

Yeah, well, I think everybody is right, first prerogative is to do no harm. So you know, obviously, I think the findings were not what they anticipated. I was a little disappointed, not by the paper, because, like you said, I think they did a tremendous job. Their follow up was impeccable. And I was just kind of disappointed at the outcomes because, you know, it's unfortunate that depending on where babies are born, that outcomes can be so different. But I do think there's some things you know, for us to really look about look at the first thing is that this was a very sick cohort of babies so they excluded all the mild babies sounds like they probably excluded some moderate babies, the average cord pH for for the hypothermia group was 6.94. or, and the average for the control control group was 6.97. So they didn't use it as an inclusion criteria, but they did report it on the babies who had it, which wasn't all the babies. But the babies were quite sick. And, you know, one of the other major findings is the clinical seizures. So 74% of babies in the cooling group and 73% of babies in the control group had clinical seizures at random assignment. So that's upon arrival to the tertiary center, you know, really within those those kind of critical first critical hours. So, what that tells me is that these babies were very sick, or potentially that they even had, you know, an in utero insult that may not have been helped by therapeutic hypothermia to begin with. So that's definitely something you have to think about. I was

Ben 10:54

gonna say that this is the this is it's interesting that you're bringing this point up, because we didn't talk about the MRI findings, right? I mean, I wanted to really discuss the primary outcomes. But they, they only found in the in the MRI that was done at about 14 days of age 21% of the infant in the hyperthermia group and 26% in the control group, or had basal ganglia or Thalmic injury, which is what we most often expect to see in babies. Right? And so like you said, could that be that their pathology was radically different? And so could that explain some of their findings? I think that's a very good point. They do mention that in the discussion, too.

Daphna 11:31

Yeah, absolutely. And we know from the major cooling trials, that the babies with the most severe encephalopathy, still had poor outcomes, even when they were offered therapeutic hypothermia. And so maybe, potentially some of those moderate babies were the star not to babies who don't appear a sick initially, maybe, you know, they weren't transferred, because of the resources that that that they were having to work under. The other thing that they mentioned, is about home births, which there weren't that many home births, but there were some, but even more so than that they had a babies born at another hospital. So babies that had be transported in to these kind of tertiary care centers. So 67% of babies in the hypothermia group and 67% of babies in the control group, were all born at another hospital. And we all know that even here, you know, for example, where we are in the states that a lot can happen during resuscitation and transport before a baby gets to a major center. So I think all of those things are things we have to consider when we when we think about this cohort. The other thing about the cooled babies, the babies who received hypothermia is that they had a not insignificant amount of adverse events. And we all know that cooling comes with some risks. But you know, I, these adverse events were in greater numbers than we saw in some of the major cooling trials. And so again, that may speak to the severity of babies that were transferred, or potentially, you know, how the cooling was done, the babies get too cold, I don't know. But I was surprised to see how it kind of impressive those those events were. And so something else for us to think about. I think what I take away from the paper, is not necessarily that we shouldn't be doing it, again, I, we live in a very resource rich area. So we're quite lucky. So you know, I'm able to say that without having to deal with those consequences, but that maybe there are ways that we can you no advocate for better distribution of education and resources, and especially to kind of those outlying areas where a lot of these babies were born before they could get to higher levels of care. And that's not so different than some places, even here in the states are in other very well developed countries, where we have babies born at facilities that weren't anticipating to have a really sick baby, but they still have to care for them. And then, you know, equip the transport team. And so a lot happens in those first few hours of life that really changes outcome. So, for me, the major takeaway, especially being at a place where we get a lot of referrals, and we do transport is that, you know, we got to educate our kind of community providers and see how we can optimize those those first few hours teaching good resuscitation, and, you know, optimization and transport. What about you?

Ben 14:57

Well, I agree with with what You're saying, and I think it's an interesting paper because it brings really your counter argument to all the other papers like the Toby trial and the NI CHD trial, really saying that therapeutic hypothermia did not work in this cohort. Now, like you said, I think it's a very, it's a very peculiar, very singular sort of population, they have a lot of different constraints, like you said, the outlying hospitals, the home births, but I can see though, some scenarios in which even like you said, even for us in the US, we could be placed in the similar in similar situations, I was actually talking to one of the obstetricians at our hospital who said that because of COVID, a lot of mothers have been reluctant to come to the hospital for free and have even considered like home births. So you would literally find yourself in more than usual similar position. Right? And and then, could these results then apply to us as well? Because we tend to think, oh, because it's, it's a low income country, then that doesn't apply. That's not true. So it's gonna be interesting to see more than the paper itself. I think what they've showing us is an eye opening as our eye opening findings, but it's going to be interesting to see what comes out of, of this study. And where does the pendulum go from here?

Daphna 16:11

Yeah, especially, you know, since since, you know, a lot of a lot of big groups are looking at more mild cases, early, you know, lower gestational age smaller babies. So it's really kind of a dichotomy of what's going on here, you know, from one place to the, to the other. So, you know, I'm looking forward to seeing to what those other trials bring. And again, I still took a lot from this a lot from this paper for us to work on. So.

Ben 16:40

All right, so let's jump, let's continue. And which which other article, which article that you want to go to next?

Daphna 16:47

Well, I guess if we're talking about kind of long term, developmental things, I think it would be reasonable to go to this paper and titled, preterm or early term birth and risk of autism. primary author, Dr. Casey Crump. So this was a cohort, looking at the Swedish Medical birth registry, in conjunction with Mount Sinai Hospital in New York. So the goal of this study was to really identify the risk of autism spectrum disorder. In particular, with early, early births, and looking at gestational age and risk of autism spectrum disorder. They also looked at sex. They also looked at co sibling analysis to look at some of those other parameters. So in this national cohort, they looked at over 4 million infants born in Sweden during 1973 and 2013, who survived to age one year, and then they followed using some of their ash, other national outpatient registries, into year 2015. So kind of the average length of follow up, the median length of follow up was 21 and a half years, the maximum length of follow up was 43 years. And so they did that because obviously, the the way that we have been diagnosing ASD has changed significantly in the in the last two decades. And so they report that they were using the ICD nine, which began documenting ASD codes in 1987. So they follow that that cohort, even though some people were born much earlier than those ICD nines, but to see if they, you know, triggered an ICD nine code and in some of their outpatient stays later in life.

Ben 18:58

Yeah, I think they follow them up to like 20 years of age, so, so even if the ICD nine code had not been there, like when they were two years of age, they should have, like you said, triggered it over the course of those 20 years so that you give them enough room, I think.

Daphna 19:13

Yeah, so they were particularly interested, like I said, in those preterm infants, and so what they found that preterm infants in general were more likely than term infants to be male, more likely to be firstborn more likely to have a family history of ASD. Their mothers and fathers were more likely to be at the extremes of age, and they had lower education level and their mothers were more likely to have other medical comorbidities, high BMI, preeclampsia, hypertensive disorders, and diabetes. And then when they looked at the group as a whole, they found that ASD was identified at 1.4% of their group. And that's pretty similar to previous reports of Autism Spectrum Disorder in Sweden. And just for comparison for like our listeners here in the US, our rates are about one in 54 to one and 56, which is 1.7 or 1.8%. So pretty consistent. And then they split the babies up by kind of gestational age. And again, the graphs I thought were pretty useful. So hopefully we'll be able to demonstrate some of those. But the ASD prevalences were 6.1% for the extremely preterm 2.6, for the very low moderate preterm 1.9, for the late preterm 2.1% for all preterm compared with 1.4% for term babies. And then they looked at a number of covariates. Again, there's a lot of data in this paper, and in a bunch of the papers that we're reviewing today are just packed with data. So they may take some more individual review. But even after adjusting for these covariates that are frequently associated with preterm birth, the numbers were decreased but still increased by decreasing gestational age. So meaning that the younger you are, at the time of delivery, the earlier gestational age, the higher your risk for autism spectrum disorder. And they did an analysis to show that it was about each additional week of gestation was associated with a 5% lower prevalence of ASD on average. So I thought that was actually a pretty impressive, very useful kind of piece of the data to have. They also looked at other comorbidities in particular attention deficit hyperactivity disorder, which has also been reported in the literature to be more common in the preterm population. And what they found was actually they looked at the comorbidity with ASD, and they found that 40.7% of people diagnosed with ASD also had Attention Deficit Hyperactivity Disorder. And then they looked at it again, just in the preterm population, 8.4% of ADHD for the extremely preterm 4.7, for moderate to very preterm 3.6 for late preterm, as compared to 2.8 for term births. So, so certainly both autism spectrum disorder and ADHD are more common. The earlier in gestation you are delivered, and not surprisingly, that they're frequently found is comorbidities, attention deficit hyperactivity disorder along with autism spectrum disorder. So a lot of data in this paper, I thought this was really interesting. Parents are asking us about it, you know, it's very common, certainly in the kind of lay internet. And so it's, I feel like one of the more questions that I get these days is, will my baby have autism, the one thing we didn't discuss was, you know, any association with intraventricular hemorrhage? So, you know, I'd like to see that data. But I think that, you know, we owe it to parents to add these sorts of things to our anticipatory guidance, at least, near discharge, you know, at least sometimes just so that parents have that information that they can be sure to get their follow up. And that, you know, it's something certainly for general pediatricians to be looking for.

Ben 23:41

Right, I think it was a very interesting paper, right? It's we know that preterm babies are at high risk of developing autism. But this paper really by grouping it in terms of gestational age, and like you said, those graphs are pretty impressive in the way that the slope of the increase towards higher frequency of autism in lower gestation babies is it's staggering. It also provides, I think, a little more one more piece to the puzzle of trying to understand the origins of autism. I think this is really one of the big questions of the 21st century trying to really understand how does that come about? And I think that by showing that there are certain gestational ages that are more at risk, it could be due, obviously, to the fact that these babies are quite sicker. But if there is anything in the development of the brain that happens at a certain amount of weeks in gestation that could explain what we're seeing later in childhood. Well, I think that's that's a terrific stepping stone for a lot of the other researchers who are looking into this topic. And like you said, I really appreciate it a nice circle that sort of pieces of the article that said each additional week of gestation is associated with a 5% lower prevalence of autism spectrum disorder, on average. These are the type of things that you can just copy paste and use in any prenatal consult and like you said, these are things that parents would like to know Oh, and we shouldn't be here for them with the data when they asked. So yeah, overall a terrific paper.

Daphna 25:06

Yeah, I think, certainly a jumping point for autism research, but but also it's me, you know, something else for us to target in kind of our developmental interventions in the NICU. So I liked it. I was glad to read it.

Ben 25:22

Yeah. Let's, let's change topics a little bit. I mean, I think before we move on to another, there was a bunch of other very important papers. I want to bring up this paper from the actor, pediatric, it's not really a paper, it's called a brief report, and it's called neonatal resuscitation in the NICU challenges beyond NRP. This was a study done out of Haifa, in Israel and in coordination with Dr. Karen Levy. nivo, from from Boston from Boston Children's Hospital. What was interesting in that brief report is that this, this group asked the question, what is the more used form of or algorithm of resuscitation used in the neonatal ICU? Considering that we're sort of in between two places where we're supposed to use an RFP, but then maybe stay in the NICU for a certain amount of time? So then, should we move on to pals? And are we even using pals? Are we using a hybrid? So they decided to try to see if this was as much of an issue as they perceived it to be? And so they sent surveys across all 25 level three and level four NICUs in Israel, asking them about their their recent codes and and what would they do to and what type of interventions that they that they used. And so it was a it was a very, it was a very thorough, I think, survey, it contains 20 questions. And so all the physicians of each institution was asked to fill it out. They received responses from at least the physician from each of the of the 20, of the 25 NICUs. And they had 75 out of 114. Physicians respond. And it's interesting, right, because you go over the different reasons for their codes. And it includes a variety of things. The most common one was obviously new authorities, bacteremia, and cardiac tamponade. And then they go over some of the things that were done, needle decompression, et cetera, et cetera. But what was also very interesting is that they looked at the different types of compression algorithms that they use. So they found that a compression to ventilation ratio of three to one was used in 87% of cases, compared to a 8.8 point 6% of other cases where they used a 15 to two ratio, which was sort of more of a of a pals sort of algorithm. And then they looked at other things that could differentiate between pals and NRP, the compression ventilation synchronization was done throughout the resuscitation in 71%. But synchronization only prior to intubation was done in 18.3% of the event and no synchronization throughout the resuscitation only an 8.4%. So what they what their conclusion was, was that among all the other respondents, they they could tell that there was no consensus. And they said that there's it's a hybrid of, of NRP, and pals, and it begs the question, should we discuss this further? Should we have an intermediate sort of algorithm to use in the NICU that would satisfy the need of our babies? And more trials are needed to sort of assess what would be the right algorithm for patients who are in the NICU? I thought this was such a relevant article, because this is something that I have struggled with in every single NICU, I've worked that there is no like we like we always joke about there's no reason for a baby that received an overdose of morphine to be dried and stimulated when they're not breathing. It's just so it doesn't make it doesn't make a lot of sense sometimes. And yet, we're all in RP certified. And we most likely don't follow an RP we don't do what I just said, we don't try and stimulate the baby when they have a pneumothorax. But yet, that really brings up the issue of then let's accept that we're either doing pals or something else, but then let's define that something else. curious to hear what you thought about that.

Daphna 29:13

Yeah, I again, some of the takeaways, I thought were impressive. Were certainly for one, how many new authorities were still still happening, especially outside of the first week of life. And so definitely something for us to remember. Some of the other things that are very much a part of the PALS algorithm, your know your H's, your T's getting those labs were done in the minority of cases. So even things like a glucose hematocrit things that we could easily fix are obviously not part of the NRP algorithm. Because you know, right at the delivery, they're not expected to be that that dysregulated but certainly in our older babies they are and then some other operators unities using obviously transillumination at the bedside, and echocardiogram, again, also used in the minority of cases. And so I think, like you said, I think it underscores that we if we're going to be keeping older babies, and nobody knows what that that threshold really is, right, but but pretty much out of the delivery room. That should we not all be trained in in pals. And the other thing is that it can't just be the physicians, right? It has to be the entire staff that would respond to a code because you can ask for whatever you want. But if people aren't prepared to give it or aren't familiar with it, then then you're not going to get it. And so I think I think it was really helpful. I mean, most of most of the codes, were using a combination anyways. And so if we all do anyway, that's how it's exactly I felt, I felt that, you know, we would have answered pretty much in a similar manner. For sure. I agree. Yeah. All right. Which,

Ben 31:08

which paper are we going to next?

Daphna 31:09

So I actually, you know, we're trying to, we're trying to cover all sorts of papers. So I thought, this quality improvement paper, we can at least discuss, we have lots of listeners who do Qi work. And so this was published in the Journal of Pediatric quality and safety. It's called improving compliance with a rounding checklist through low and high technology interventions, a quality improvement initiative, lead author, Lea Carr. And so what they aimed to do was really do kind of a quality care questions, you know, each day of the week that targeted different parameters of an infant's care. So this was a questionnaire that the group was already using, but they weren't getting a lot of use out of it. So what they did is a number of PDSA cycles to see if they could improve the use. And I think there was some actually pretty important points in this paper that, you know, I'd like to cover. So basically, what they had is this kind of hardcopy checklist that they had at the computers, where the teams were doing the work, but not necessarily where the teams were doing the rounding. They didn't weren't bringing it to bedside. So they started in about November of 2018, where they initiated their first five PDSA cycles, and that is when they made those kinds of low tech interventions. So they brought in a human factors engineer to make design modifications on the checklist itself. existed, by the way, that's right. I mean, that's why are kind of cross discipline. interactions are so important, because these are human factors, engineers are being used in all kinds of other places in business, in aerospace in, you know, project management, and so we got to get them into into medicine for sure. So, what they did is they changed up the questionnaire, they made the questions shorter, easier to read used abbreviated key words, they changed the color, and I think probably most importantly, is that they moved the location from the desk to to on to kind of the rounding computers. And so those were their first PDSA cycles is making those changes, and they have their graph looking at those changes over time, and then the sixth PDSA cycle in October 2019. So nearly a year later, was actually kind of integrating the tool with their EMR. So they used rule based logic to present relevant daily questions to care teams based on patient status and day of the week, which I thought was really cool. So it auto populated kind of hard stops with appropriate questions and the right responses whenever possible. So for example, they said if the baby didn't have central access, the tool would auto populate with not applicable when they had to discuss central line, but it did prompt them to at least discuss it or not. And then, basically, for each baby, they could have a separate separate flow sheet that look to see if the checkmarks for each of the quality kind of indicators were done for the week. Something they did that was not part of a PDSA cycle, but I imagined still played an important role because I've seen that in the units where where I have a practice is that I'm starting in about February of 2019, kind of right in the middle of their low tech interventions. I mean, really started pushing out the messaging to the staff, staff and nursing leaders and asked reminders during the daily unit wide, and nurse nursing morning, huddles, and I think this is always an important part of Qi getting the rest of the team engaged. And it certainly looked like that helped them somewhat. One part of their method, their methods and kind of timeframe is unfortunately, this obviously was occurring right at the time as the COVID 19 pandemic was, was becoming such a big disruption to our medical care system. And so they talk. Yeah, right. It's too soon, too. So they talk about how that affected their, their outcomes. But I actually think that they were still able to do a great job with the, with this project, despite that. So some of their major findings is that they looked at what were the initial barriers to getting the checklist done. The number one barrier was that nobody could remember what the items were or when to do it, or on which babies to do it, which I think is a huge barrier. They had rounding interruptions, which I'm sure we're all sensitive to. And one of the smaller percentages was perceived lack of question relevance to patients. And they looked at their average completion. So before their interventions, the average completion was about 31%. And then after interventions to 80%, they also looked at average checklist completion time, which obviously, time is always a concern for people having to do extra work. And they found that actually, overtime, their ability to get through the checklist decreased from 46 seconds to 11 seconds. So obviously, 11 seconds times, you know, when you've got 2025 babies around on Yeah, adds up, but certainly improved over time. So they found that their PDSA cycles made a huge difference in their ability to use kind of this checklist. But the major point I want to say is, you know, we're all working in different kinds of units. But what I was really impressed by is that they made a major change in kind of the, where they put their rounding checklist. And I think that and it looks like when they started pushing it out to staff made a huge difference. Big, big, big increases in their completion rates, even more so than some of those high tech interventions, which I thought streamlining it with the EMR is a brilliant idea.

Ben 37:55

Yeah, doesn't it show that moving the doctors as far away from the EMR is always more efficient? That's it? And yeah, the power of the low tech tools we should it should not be it should not be underestimated. I thought I had my thoughts aligned with yours. Exactly. And and it's, it's interesting, right? I mean, even just, if you look at that paper, and you look at the different items they had on their list, I think it's a lot of inspiration as to how to organize your rounds. I'm going to steal some of their items for my rounding tool. So yeah, definitely a great paper to check out.

Daphna 38:29

Yeah, and we can put this graph on here. It's just a good reminder that you don't even have to get the high tech interventions, you just have to, you know, find what works for you what works for your unit, get everybody engaged and in, get get your items where people can see them. That's right, the big thing. Where do you want to go?

Ben 38:51

I want to go on that paper about NEC versus CIP. Yes, is laparotomy. That's that's that was such a big one. Yeah, so the paper is has a very short title. It's called initial laparotomy versus peritoneal drainage in extremely low birth weight infants with surgical necrotizing enterocolitis, or isolated intestinal perforation, a multicenter randomized clinical trial period. It was published in Annals of surgery, the list of authors is extremely long. This is a paper that was done in coordination in collaboration with the neonatal research network. So obviously, there's a long list of authors. It is a pre proof. So it should come out more formally in the next few weeks. And what was interesting about the paper is that it asked a question that we're all asking ourselves, so this was the next trial and it was designed to be the largest feasible trial evaluating the impact of initial laparotomy versus drainage on the rate of death or neurodevelopmental impairment, and whether preoperative diagnosis of neck or SIP affects the outcome. So they did a prospective randomized trial. conducted at 20 us centers that are all part of the neonatal research network between 2010 and 2017. They included babies who were born at with a birth weight of less than 1000 grams and that were less than eight weeks of age, at the time of surgery or at the time to perform surgery for suspected neck or sip. The were, they follow these babies at 18 to 22 months of age, and they use the Bayley scales of infant and toddler development third edition to assess their long term outcomes. They had some very typical definitions of neurodevelopmental impairment, and obviously they had moderate to severe. So northern mental impairment was defined as moderate to severe cerebral palsy, barely three scores of less than 85, severe bilateral visual impairment or permanent hearing loss despite amplification. The pre specified a bunch of secondary outcomes which were obviously very comprehensive intro operative complication post operative complication number of surgical procedures for each infant sepsis episodes, duration of TPN, developmental of TPN related COVID stasis length of hospitals of hospital stay rehospitalization, and each component of the primary outcome. So they did something that was interesting, right? I mean, is that they, they approach their outcomes as well, the way they presented them. They presented them in the form of either frequencies theories or Bayesian theories. And I thought that was very interesting. For people who are not familiar, there's two approaches to outcomes, you could use a frequentist method, which basically doesn't take into account too much data on the front end, and just forces you to make a guess as to what an outcome is, where you can use Bayesian theories where you take a lot of data, and then try to formulate a the most optimal guests that you can based on the data. The example that I've always heard, is, if somebody flips a coin, for example, and you can say, if before you flip the coin, what are the odds that the coin will fall heads? And you can say, well, it's 50%. And then if I do flip the coin, but I'm not really showing you the, the outcome of the of the flip, and I tell you, Well, what is what is your answer? Now? A frequentist will say, well, it's either heads or tails, there's no, it's either 100% or 0%. There's no in between, because the outcome has happened. And it's not really a guess anymore. There is an outcome that is out there in the world. And it's either yes or no. But the Bayesian person will refine their theories, they'll say, well, there was a 50% chance and the fact that you're asking me maybe mean XYZ, so I will refine my my estimate. And I will say that it's maybe 60% heads for X or Y reason. So it's interesting, because in the case of neck or sip, we try to do the best we can to figure out what is the etiology, right, we do see nematocysts you can see free air in the abdomen, and you're like, is it just a sip? Or is it neck? And and that could play a huge role depending on what the intervention is. And that's what the paper was trying to do. So they were able to enroll 310 infants, I hope any of the things I'm saying makes sense. That makes sense to me. But I hope.

Daphna 43:15

I think once you talk about the, I think once you get this data that it will. Yeah.

Ben 43:21

So they have 310 infants that were randomized to either receiving laparotomy or Penrose drain placement, and they were able to follow 96% of them at 18 to 22 months. The primary outcome occurred in 69% of infants after initial laparotomy and that is death or no detrimental impairment. So they were noticing death or no other mental impairment in 69% of infants after laparotomy versus 70%. After Penrose replacement mortality at 18 to 22 months was 29% Overall 28% With laparotomy 30% With Penrose placement, when we're looking at things in a more granular way 94 infants assessed at 22 months with a preoperative diagnosis of neck. So that so that's so out of that group, specifically 69% died or had impairment after initial laparotomy versus 85% After drainage. And then they looked at this Bayesian post probability, and they said that the Bayesian post probability that laparotomy resulted in a lower rate of death or impairment than did the Penrose was 97% which means that in the cases where neck was highly suspected, or laparotomy, was beneficial when it comes to the primary outcome 97% of the time. Among the babies who had sip. Death or NDI occurred in 69%, after initial laparotomy versus 63%, with a Penrose placement, and when they did this Bayesian posterior probability, the rate of death or NDI with an initial laparotomy was 18%. So, a few more results. And then we can we can talk about the other outcomes mortality occurred at 18 to 20 months occur in 46%. With a preoperative diagnosis of neck versus 21%, with a preoperative diagnosis of SIV among infants who were diagnosed with NEC the mortality was 40%. With initial laparotomy versus 51%, with the Penrose placement among infants with a preoperative diagnosis of CIP 23%, died after laparotomy versus 19% with initial drainage. So before we go into the before we go into the secondary analysis, I think the primary outcomes are very important, right? I mean, it's showing us that if we the PRI, it's all hinting at the fact that if you suspect a diagnosis of an EC, placing a Penrhos is not going to be beneficial down the road, the laparotomy seems to be more beneficial. In the case of sip, it doesn't seem like the data answers the question. And at the end of the results there, they were also mentioning the fact that among infants who had an initial laparotomy, and had the intra operative diagnosis data, which was 138 babies, the intra operative diagnosis was concordant, with the preoperative diagnosis in 64%, of cases. So I think it's interesting for us to talk about this, right? Because you tend to think, well, you know, I think we can differentiate pretty easily between the answer's no, it's still not super clear. And with a preoperative diagnosis of NEC, the concordance rate was 78%. And with sip, it was 59%. And I think this is where it gets a little bit tricky. I think NEC really declares itself when it does pretty clearly, but a SIP can always be misleading. So in their main conclusion, and I'll let you discuss what your thoughts were, and we can even talk about the about the secondary outcomes. But their conclusion was that there was no overall difference in depth or NDI rates at 18 to 22 months corrected age between initial laparotomy versus drainage. However, the preoperative diagnosis of NEC or CIP modified the impact of the initial treatment. Alright, I'm gonna stop talking. Yeah.

Daphna 47:05

Well, again, this is another study that enrolled, you know, they had almost 1000 babies that were eligible, they randomized 300 babies, there's a whole group of babies that weren't randomized. And they, they speak to that, that that analysis is coming. So I'm looking forward to seeing that data, why those babies weren't randomized. But regardless, they had these babies, and they had pretty good follow up also, with this group, even though, you know, these babies were all over the country. So this was quite a collaborative effort with, obviously, some of the major institutions here in the States. And, you know, I think, talking about some of the secondary outcomes are a piece of the puzzle, right? Because if we just say, well, maybe they're not different, but one has much, much worse outcomes, you know, then we have to pick one over the other. But this, this was not what I anticipated it to show either. So for example, initial laparotomy. So those babies who are randomized to surgery versus Penrhos placement, did not increase the time to full feedings, duration, the mechanical ventilation, TPN, or length of hospitalization. The total number of operations was similar for treatment groups, with initial laparotomy. Two and those babies with initial drainage 2.1. But subsequent laparotomy was performed much more often after the initial drain 50% Versus if the baby got a laparotomy. First, they only needed, you know, repeat laparotomy a quarter of the time, which is still a lot of laparotomies. But a pretty big difference. And so, you know, I thought that was pretty impressive. I think it shows that we need to get better at deciding what what is our diagnosis, right. And the truth is, sometimes we don't know. But all we can do is use the information that we have, I'm hopeful is we are better at refining the diagnosis, we can use some new tools, hopefully ultrasound will be part of our diagnostic criteria. That that maybe we will eat get even better with our kind of pre operative diagnosis to give us even more information about to do what to do surgically.

Ben 49:34

Yeah, I think that was that was what was interesting, right? You you we can anticipate articles that says oh laparotomy versus Penrose in NEC, right. Right. So this is where it's interesting is that it's CIP or NEC laparotomy or Penrose, and then let's do all the math and let's do all the stats. So that was that was really really cool to be able to say well, if you believe that it is neck and you are correct, then laparotomy is more beneficial or has less negative side effects down the road. versus when it is set? Well, you really don't really know what it is your your chance of getting it right is actually very low. And then it is not really clear whether laparotomy is the way to go versus a Penrose. It's just fascinating to move away from P values and have this more uncertain, uncertainty centered approach. I think a lot of the papers sometimes you're like, Yeah, but I don't know the diagnosis 95% of the time, because I'm not sure. And then obviously, by the time I do know, the diagnosis, it's too late. So to be able to say, Well, let's think about the moment where you don't have all the information, and you need to make a call, what would be where do we do good? Where we do? Where are we usually where do where our faults? I think that was fascinating. It's a great, great paper.

Daphna 50:42

Yeah, and that's how we do medicine, right, we get a little bit of data, and we make a clinical decision, we get a little bit more data, and we make another clinical decision. So you know, at some point in time, our research methods have to reflect

Ben 50:56

as as you may have guessed, I'm a Bayesian type of thinker, and so I could not be happier. And we're very excited that one of our guests is going to be a strong advocate of that, of that, of that mentality, and that in that approach to science, so I'm not going to reveal who that is, but

Daphna 51:11

we're excited. Well, I I'm, I'm sensitive to the time and I know that you want to talk about some of these machine learning papers. Yes. So I want to give you the opportunity to do that.

Ben 51:26

So, papers that we selected, that are looking at machine learning and artificial intelligence in neonatology. The first one, and we'll talk about this one first is it's called it's it has it has an has a long title. It's in JAMA. It's called Automated, explainable multi dimensional deep learning platform of retinal images for retinopathy of prematurity screening. And now it's G Wang, I know if you're an if you were barely interested in reading it, the title just throws you off right off the bat, you're like I'm not doing but basically what they did is that they looked at whether machine learning algorithms could interpret pictures of of the retina have done during ROP screening, and measure whether that's the sort of interpret these images for ROP. And they compare that head to head with with the reading of of of technologists. And so they had 14 1000s, or have eyes of about 1000 preterm infants. And they had a deep learning based ROP screening platform that could identify retinal images using five classifier including image quality stages of ROP, intraocular, hemorrhage, pre pre plus and plus disease and posterior retina, the platform achieved an area under the curve of 98.3% to 99.8%. That is staggering. And the referral system, which was that they if the machine detected a certain threshold, then it would refer to an ophthalmologist had an area under the curve of 99% to 99.56%. I mean, that's just bragging at this point, the platform achieved a coin K of point eight, six 2.98 compared with point nine, three 2.98 by the ROP experts. So number one, this is really cool, because sorry, this is really cool. Because if you don't know much about artificial intelligence, it's showing you that computers can read images, just like you have eyes, they can look at an image and interpret. And it's not very hard to design. There's a lot of algorithms out there that can read images. It's been done in radiology, many in over many, many papers. And you could you could read that, before we go to more into the machine learning algorithm, we can talk about the other article that was published, this one was in BMC pediatrics, and it's called neonatal mortality prediction with routinely collected data in machine learning approach. And this was a study out of Brazil. And basically, I'm going to explain this rather plainly, they used they had several machine learning algorithms, which they fed data from 2012 to 2016. And they allowed the machines to train on different babies, their morbidities their outcomes. And the machines were trained to learn, were trained to to perform prognostic sort of measurements. And then they used their data from 2017 to actually test the machine. And you can go into the paper, they have tons of different variables. And then they also used another set of variables which was which were the ones for that identify neonatal mortality risk based on the who, and that was maternal age, place of delivery, mode of delivery and weight and gestational age at birth. So they did, they did go over the different areas under the curve, which was the accurate which represents pretty much the accuracy of the system. And they had three machine learning algorithms, the gradient boosting trees, the light GBM and the cat boost that achieved respectively 97 1% 96% and 97% accuracy. What's interesting is that you can see how this paper highlights the importance of good data. Obviously, then they they ran these algorithms with the WHO criteria, and the area under the curve was 90.5%. And then they started introducing a bit more data points, like the five minute Apgar, and that increased the area under the curve to 95%. Then they added whether the baby had congenital anomalies, that increased to 97%. And so you can see that when you tweak the data, you can get a system that gets better and better train. I think this, these are important papers for people to read, because I'm not exactly sure when artificial intelligence and machine learning will actually roll out into the NICU. But before that can happen, we're going to need to collect data, we're going to collect excellent data. Because if we want to train the machines to perform calculations that are accurate and precise, they need good data to train on. And you can start seeing out of California out of Duke people are publishing what they call just data registries so that you could use them to train the machine to practice but I do feel like you will not really be able to get the best results until you train the machines with your own population, we're seeing that there's every population is different. So if you train a machine in the US on the outcomes of patients in Japan, well, they may not get it perfectly, right, because there's a lot of inheritant difference between the two populations based on their geographical difference. So I think these are very interesting. And you can see that the machines can actually do a pretty good job. The one thing I do want to mention is that the machine learning algorithms that they used in this Brazil study actually gave you two extremes. So they gave you like the 5% of the highest risk of mortality. And so for the 5% births, with the highest predicted risk of neonatal death, it was including 90% of the patients who had died during the year 2017. And on the other hand, there were no deaths among the 5% birth with the lowest predicted risk. So you could see that the machine on the extremes did extremely, extremely well. And if you think about this, from your own perspective, and say, Well, if I could be told when a baby's admitted, whether they're in the top 5%, for neonatal death, I would like to know that because my approach would probably be different. So it's really, really cool. I get really, I geek out on these papers.

Daphna 57:18

I know you do. And you know, I'm still having trouble with my iCalendar. So but, but I understand that this is the this is the future of medicine, right? And if it if it helps us take better care of babies, I was particularly impressed by the ROP paper and how well that it stood up to you know, they had a number of experts also read the pictures. How well it stood up to that I was really impressed by that. And so I think that it can potentially provide more access to care, especially for places who may not have certain resources. I think that's really interesting. i Do you also think you bring up a good point about different kind of how to our kind of racial, ethnic changes, how does that impact our data sets? And actually, maybe our next guest will be able to, to speak to that a little bit. But how does that interplay with our ability to use artificial intelligence, but it's the future of medicine, again, all of the other major disciplines in the world are using artificial intelligence. And so I just had

Ben 58:34

a big keynote address where they actually are showing how it artificial intelligence is pretty much going to be the route of the self driving car. So I mean, it's coming to us, I think it's very freaky, I'm doing some some courses in artificial intelligence, and to teach your computer to recognize numbers drawn by hand, meaning I could draw a set of numbers and the computer be like you're writing a tune out. And it's like, this is very freaky, because they're not really, it just knows, like, you're just taking a picture of the image. And the compare was like, Yeah, that's a three. And even if you scribble it, it's it's crazy. And when you start seeing, and I'm just doing this in my in my bedroom, you can imagine in more advanced laboratories, where they actually teach the computers to read X rays, ultrasounds, it's nuts. If the what if you could take an ultrasound and the machine tells you Yeah, that's that's neck right there. Yeah, be cool. And,

Daphna 59:26

I mean, maybe yeah, maybe they'll put us out of a job. But if it

Ben 59:30

I don't know, I don't know. Because at the end of the day, the machine is only as good as the but I mean, listen, at the end of the day, it's a long discussion, I can only recommend you can go on YouTube. And you can look for the the documentary called AlphaGo and it's an it's a documentary that goes over the artificial intelligence that was used to train and, and take on the world champion in the game of Go and it's a fascinating look into Do what artificial intelligence can do. And it's a very poetic documentary on the concept of man versus machine. So it's free, I think you can it's not not hacked, you can just go type in Alpha Go on YouTube, and it should be there for free. Just take a look at it. It's very,

Daphna 1:00:15

very cool. Even I've heard of that one.

Ben 1:00:18

And it's pretty cool. And then when you look at, and I'm sorry to say, but when you look at how much parameters and a lot of the complexity of the code and the in the in the algorithms, and then you see that they're carrying this in like a laptop, right? So again, it's just very, very easy. It's cool. Yeah, it's amazing. All right. There's there's a few more articles that some of them were pre approved, but we can definitely leave them for next time. And we'll go over them. I think we went over the Main articles that we wanted to touch on, right?

Daphna 1:00:48

Well, there's always something but can't seem to get to everything can we write? That's fine.

Ben 1:00:53

I think the articles we discussed some very significant articles. Should we do a little feedback back Sunday when we do? So the highlight of this week came from Frank Han, and it was not really a feedback, but it was a response to somebody else on Twitter. Dr. Caroline Bartman asked the question on Twitter, why are 99% of the podcast interviews with academic scientist I see with male scientists. And Frank Hahn said, Well, Dr. Daphna and Dr. Ben are turning the tide. And I was very happy to see yet. I don't we don't really pick guests based on on sex, gender or anything. I just we just tried to pick interesting people. But I'm very happy to see that people are listening and that the guests are being appreciated that I think that's really

Daphna 1:01:38

yeah, we've got too many too many great people to choose from, actually.

Ben 1:01:42

Yeah, it's becoming we're becoming a victim of our own popularity. But I think we have a lot of good guests coming up, do you want to tease of who our next guest.

Daphna 1:01:52

So it's definitely a good problem to have. And I am honored love, would love to talk about our next guest, who, who I guess speaks to what Dr. Hahn was saying. But this is Dr. Diana Montoya Williams. She's an attending Neonatologist at CHOP. She's a clinician scientist and chops Policy Lab, and then an Assistant Professor of Pediatrics at UPenn School of Medicine. And so Diana is admittedly a personal friend of mine, one of my closest friends. But she's doing really tremendous work on the drivers of racial and ethnic inequities in infant health outcomes. So we are really excited to have her on.

Ben 1:02:38

I'm very excited to have her on, I think, mostly for myself being being very good to stick right now. But because of the fact that I'm not the most knowledgeable person on that topic, and I'm very eager to talk to somebody of that caliber, who will be able to break it down for us and really show us where are the areas of improvements, and so on and so forth. So definitely check it out. And yeah, I'm looking forward to recording it. Yeah,

Daphna 1:03:02

this will be a special episode. I mean, I think Diana will help shine some light on a lot of the things that we can we can all be doing to support the work.

Ben 1:03:13

Awesome. All right, definitely. Well, that was fun. Hopefully, we don't have to record this a third time.

Daphna 1:03:19

Okay. All right. I'll see you next time. Thanks, everybody.

Ben 1:03:24

Thank you for listening to this week's episode of the incubator. If you liked this episode, please leave us a review on Apple podcast or the Apple podcast website. You can find other episodes of the show on Apple podcasts, Spotify, Google podcasts, or the podcast app of your choice. We would love to hear from you. So feel free to send us questions, comments or suggestions to our email address, Nicu You can also message the show on Instagram or Twitter at NICU podcast. Personally, I am on Twitter at Dr. Nikhil spelled Dr. NICU. And Daphna is at Dr. Dafna MD. Thanks again for listening and see you next time. This podcast is intended to be purely for entertainment and informational purposes and should not be construed as medical advice. If you have any medical concerns, please see your primary care practitioner. Thank you


bottom of page