 Please welcome Fuxiang Chao from Kaiser Permanente Riverside Medical Center to share an innovative framework for growth reference among very preterm infants. Hi, everyone. My name is Fuxiang Chao. I am a new natologist at Kaiser Permanente Riverside Medical Center in Southern California. I would like to thank our medicine for giving me this wonderful opportunity to share with you our recent work on an innovative framework for growth reference in very preterm infants. This project was conducted in collaboration with Dr. Yeh at Turner's Mercy Research Institute in Kansas City and Dr. Reese Clark at Pediatrics Medical Group. I have no conflict of interest to disclose. Allow me to thank Dr. Tennis Fenton first for sharing with us the LMS table for the 2013 Fenton charts, which was used for plotting and comparison in this presentation. I would also like to make a note that this presentation is not about optimal growth or nutrition research in very preterm infants. Let me introduce two terminology that are essential in newborn medicine. The first one is gestational age, or GA, which indicates the time the gestation has lapsed before an infant was born. The second one is postmenstrual age, or PMA, which is gestational age plus chronological age after birth. And it's usually reserved for describing the developmental stage since the beginning of gestation in infants born prematurely. For example, a nine day old infant born at a gestational age of 24 weeks, zero days, is now at a postmenstrual age of 25 weeks, two days. In year 2021, approximately 57,000 infants were born at a gestational age of less than 32 weeks in the United States. Many of these infants are also known as very low-birthway infants with a birth weight of less than 1,500 grams. In 1977, the American Academy of Pediatrics published a statement on the nutrition needs of low-birthway infants, asking for a prompt postnatal resumption of growth to a rate approximating intrauterine growth, because this is believed to provide the best possible conditions for subsequent normal development. Even before the statement, two intrauterine growth charts have already been published. Subsequently, several newer intrauterine growth charts have also been published, with the 2013 Fenton growth charts being the most widely used in the current clinical practice. These charts were developed by taking measurements at birth for each complete gestational week, identifying the values for each predefined percentile, connecting the dots, then applying smooth function to the connected lines. This is an example of a female infant born at 25 weeks, zero days. X-axis is postmenstrual age, and the Y-axis is weight. Left is the Fenton growth chart, and the right is converting the weight to a Z-score. The initial weight loss that almost every single newborn experiences resulted in a drop in weight Z-score or percentile. Frankly speaking, clinicians don't really know which percentile line any given infant should be following after that initial weight loss occurs, in order to be considered as having adequate growth. We also don't know if it's a problem or not if an infant is not exactly following the percentile line on an intrauterine growth chart. Here, plotting a group of female infants born at 25 weeks, zero days, we can clearly see the discrepancy between postnatal and intrauterine growth. Several factors may lead to this discrepancy, including differences in metabolic demand and supply. The growth environment, for example, the uterus is a confined space versus the isola is less of a confined space. Also, the free water status is different between fetuses and the infants. Postnatal growth is different from intrauterine growth. The question is, how do we characterize postnatal growth so preterm infants can be compared to each other more appropriately? We developed this project to model postnatal growth with the hope that the model will be able to turn into growth charts that can be used clinically. To make this happen, we know that the model has to be able to catch the initial weight loss. Also, the model has to account for variations in birth size and growth rates among infants and use that information to construct the percentile lines. We used generalized additive mix modeling techniques in this project because the technique can meet all that we want to achieve in the model. Data for modeling came from Pediatric Clinical Data Warehouse, a real-world contemporary and racially diverse data set. We identified approximately 89,000 infants that met our inclusion and exclusion criteria between 2010 and 2020. For inclusion criteria, we included infants born between a gestational age of 22 weeks, 4 days, and 30 weeks, 3 days. Who were admitted to one of the neonatal ICU run by Pediatrics Medical Group across the country within the first 7 days of life. As this is a project regarding growth reference, we want it to be as inclusive with our approach as possible. Therefore, the only exclusion criteria we have are if the infants did not have a sex assignment in the database or if there were not a single measurement data point in the database. A total of 5.4 million weight data points, 750,000 length data points, and 1.3 million test circumference data points were identified and used for modeling. We grouped the infants by sex and the nearest gestational age. A total of 16 models between 23 and 30 week groups for both sexes were developed. Because of admission age, discharge age, and variations in the interval when measurement data were entered into the database, missing values exist in this data set. We made an assumption that all missing values were missing at random due to clinical condition, not growth per se. Because generalized additive modeling in general tolerates missing at random well, no imputation was performed. Outliers were assessed after models were developed based on the predefined criteria as a leverage larger than 3 times the mean at leverage and an absolute standardized procedure larger than 2. The model includes a smooth function of day of life, a random intercept and a random slope. We use the GAMM-4 package developed and maintained by Dr. Simon Wood for model development. For the smooth function, we manually assigned the not number in order to capture the initial weight loss. With manual not number assignment, we could overfit the model. Therefore, it was essential for us to examine overfitting. Root mean square error, mean absolute error, and r squared were metrics that were used to examine overfitting using a 7-3 data split strategy. Total variance was calculated by summing the variances from random slope, random intercept, or variance between random slope and random intercept, as well as the error term. Variance values were extracted using the var core function from the LME-4 package. Standard deviation was calculated under Gaussian assumption. Because total variance is a function of day of life, standard deviation is also a function of day of life. Modeling was performed in R version 4.1.1 on a Linux machine. Here is a demographic summary. No major concerns for overfitting in all models developed for both male and female infants. The open circles were for training data sets, and the blue circles were for the validation data sets. Outlier percentage was low in all models, so no data were discarded in the final models. Here, this is an example of the postnatal growth model for female infants born in a 25-week gestational age group. X axis is post menstrual age, and Y axis is weight. Our model was able to capture variations in birth size and postnatal growth. Here shows the complete plots for weight, length, and hesoconference estimates, along with percentile lines. From bottom to top are 3rd, 10th, 50th, 90th, and 97th percentiles. Plotting against the 2013 Fenton growth charts, we can see that postnatal growth is vastly different from intrauterine growth, starting from birth. We have a manuscript characterizing the differences between postnatal and intrauterine growth in great detail currently under review, so I won't go into details here. For clinical application, first we need to horizontally shift the curve to fit the birth gestational age of each individual infant. For example, this is for female infants born at 25 weeks, zero days. We shift the curves to the left by three days, for infants born at 24 weeks, four days. We can also shift the curves to the right by three days, for infants born at 25 weeks, three days. For infants born at 25 weeks, four days, the 26-week group curves will be used. To interpret each measurement data point, we can characterize and calculate z-score and corresponding percentile based on the estimate and the standard deviation for that specific day. Also, since we already have the estimates and the standard deviation for each day of life, we can calculate each percentile line very easily based on these values. As an example, here is a percentile line for 8.12 percentile, and here is a percentile line for 19.96 percentile. With the ability to calculate and plot every single percentile line as we wish. In addition to calculating z-score and percentile for each individual data points, we now can also summarize a series of data points into one representative percentile value. We gave this value a name, and it's called trajectory percentile. In this case, the trajectory percentile is calculated based on the least mean squared error, and it's 86 percentile. We can also use a subset of data points to calculate trajectory percentile. For example, this infant developed a growth-affecting complication of prematurity called necrotizing entrocholitis, or NEC, at two weeks of age. We can use the pre-nec data points to calculate the trajectory percentile, which is 70 percentile in this case. And we can use this percentile value to guide post-disease nutrition and growth. Here, this baby had terrible growth after the complication when the baby reached term equivalent age of around 40 weeks post-menstrual age. The trajectory percentile can also be used for transitioning an infant to WHO child growth standards at term equivalent age. Note that the WHO child growth standards were developed based on chronological age of term infants born between a gestational age of 37 weeks zero days and 41 weeks six days. The x-axis in the WHO growth standards is age after birth, not post-menstrual age. Because pregnancy due date is calculated at 40 weeks zero days, it is common practice to want to plot the beginning of the WHO growth standard curve at the post-menstrual age of 40 weeks zero days. However, if we do so, the percentile may not match up. For example, if we transition the infant who's tracking on 86 percentile to the WHO child growth standards at 40 weeks zero days, the percentile line will drop to 67th percentile for a matching weight of 3445 grams. But we can now identify the correct transitioning point based on the trajectory percentile. In this case, if we transition the infant at around 3760 grams at 41 weeks two days post-menstrual age, the infant will remain on the same percentile line on both charts. Because of the complexity of the postnatal growth charts, we developed a shiny web app to demonstrate its use and to allow clinicians and neonatal dieticians to gain access to these charts. The goal of the web app is easy access and easy use, efficient data entry, no need to enter private health information. Without grouping strategy, the minimum information that is needed in order to render growth curves is sex and gestational age. Here is an example of weight plots. We can enter gestational age and sex and the corresponding growth charts will be rendered starting at the entered gestational age at birth. We use the DT package to create editable data table with all rows pre-generated and the PMA and the Day of Life info all filled in. So that users can enter measurement values directly into the table in the correct row. And in memory proxy data table was also created at the time of table display for actual data entry, Z score and percentile calculation. That way the on-screen table does not blink every time an event occurs to the table. The DT package has a built-in proxy table function. I believe the concept is similar. And this proxy table was the table that was used for plotting and data storage. When plotting growth reference curves, we found that it was much easier to have the tables set up wide with data points for each percentile curve occupied one column rather than having all data points in one column as in a long form. Finally, to make it easier for us and to be able to focus on developing the web apps, the web app was embedded in the Google sites as an iframe. The Google sites was used to create static content to assist the users. The web app can be accessed at nicooprolls.app. The quick plotting page has three web apps that can plot measurement values and calculate trajectory percentile. The full version web app requires registering for an account which will allow users to store and retrieve data. Here is a list of packages that I used for this project. Again, my name is Fuxian Chao. I'm a new natologist at Kaiser Riverside in Southern California. Thank you for attending this presentation. You can reach me via email or on Twitter. Wonderful. Thank you. And Fuxian, you're here with us to answer some questions and we have some time for that. The first one we have in the chat here, does it look like growth catch up is eventually complete and at what post-menstrual age? Yeah, so based on the model, the weight and hazard conference eventually tries to match the WHO child growth standards at around term equivalent age, but length is less, more of an issue in terms of catch up. And we're also catch up now to our group. It's kind of a philosophical question. Is it really catch up or do they really just follow a different trajectory? That's an open question, I think, whether these babies really catch up to the intrauterine growth or they really just follow a different growth pattern. Thank you. Next question is, do these data and lack of placental feeding suggest that there may be a role of PPN in delayed growth in early premature infants? Yeah, so I think it's definitely the parental nutrition makes plays a big role. We can only fit so much. And after birth, these very preterm infants, especially the extremely low gestational age infants less than 28 weeks, they go through this acute phase takes about at least a week, sometimes two, three weeks to stabilize and to get back to that anabolic stage. And also, we had to establish intro feeding and the pace, how it's established that definitely also affects. So that metabolic demand and supply definitely changes. It's very different between after birth. So yeah, so parental nutrition definitely play a role, but by not the whole picture. Thank you. And you're getting some appreciation in the chat here. I hope you see that. Yeah. Well, if there are no more questions, I think we can close the session. And for closing our. More of more applause is lovely. And to close out our conference. Please welcome back Stefan Kaduki from Children's Hospital of Philadelphia. All right. Well, good evening, everyone. As we draw to close on another wildly successful our medicine conference. I really find myself reflecting on this on this journey we've taken together started in 2018. And, and I followed it ever since and for the past three years I've been, I've had the privilege of serving as a chair of our medicine and that role has been both challenging and incredibly rewarding. This year, I passed the torch on to Peter Higgins and accomplished academic physician and enthusiastic teacher of our and Peter and his introductory speech beautifully encapsulated the spirit of our medicine as a forum where we share, teach, learn, help and collaborate and I couldn't agree more with his sentiments. A seamless transition and success of this year's conference is a real testament to to his capabilities and dedication. And also to the tireless work of the organizing committee, the Linux Foundation events team, and I want to specifically shout out give shout outs to Emily roof. Thank you all and Rachel Brown and our marketing team, Agüeda Chavez gamma and Jesse Kasman and their collective efforts have made this event possible and successful. I'd also like to extend our sincere thanks to our sponsors jumping rivers and positive pbc for their generous support, their contribution has been instrumental in making this conference a reality. We've broken records again. We've had 711 registrants hailing from 61 different countries. We've heard from 65 diverse speakers and facilitated 12 engaging workshops. This has been the largest our medicine conference ever, and we couldn't be more thrilled with with the turnout. Before we part ways, I want to remind everybody please fill out our super short survey just by going to this QR code right here. And as your feedback is invaluable and helping us making our medicine better and better each year. And as a small token of appreciation the first 400 of you who fill this thing out and provide your address will receive our medicine hex stickers we're not going to use your address for anything other than sending you these stickers. And don't forget recordings of the sessions will be posted on the ours our consortium YouTube channel by Friday, June the 23rd. Looking to get more involved. And we are always looking for enthusiastic volunteers to join our organizing committee, our program committee and our marketing team and if you're interested. Please do not hesitate to email us at our dash medicine dash events at Linux foundation.org and perhaps one of our hosts can drop this email into the chat for your reference. So again, we'll have recordings from all workshops and all sessions. They're going to be in the Arkansas consortium YouTube channel. And, and this is also where you can find videos and replace from the past year. I should past years. And so I so I invite you all to check it out. Finally, thank you all of you for your participation your enthusiasm and your commitment to pushing forward the use of the our ecosystem in medicine. We all love our this is a wonderful community I'm super happy to be a part of this, and that I can feel like I can give back to it. But you are what makes this conference special. And I'm excited to see what we can accomplish together in the future. See you all in 2024.