Failure to thrive secondary to inadequate nutrition in the pediatric population may result in lifelong negative impact on physical and mental health outcomes [1,2]. This is especially critical for infants and children with known risk factors, such as preterm birth, acute and chronic illnesses, and social risk factors [3-7].

Anthropometric measurements commonly used in pediatric populations to assess nutritional status include weight, length, stature, head circumference, and midarm circumference. Using appropriate growth chart references, a single measurement alone indicates growth status for age at a single time point and may provide indications for closer monitoring. With multiple measurements, growth velocity over time can be evaluated and deviation from normal growth pattern may be suggestive of suboptimal nutrition or chronic illnesses, including metabolic disorders or congenital syndromes, although suboptimal monitoring itself may impact efficacy [8].

Before more widespread availability of electronic health records, paper growth charts were commonly used, but they had limitations, including infrequent updating, restricted accessibility for multiple care providers, and the inability to exactly determine percentiles numerically between the limited discrete percentile lines displayed on the printed charts.

The development of the lambda-mu-sigma (LMS) method for describing growth charts allows a quantitative description of growth charts based on tables of parameters [9]. In these tables, parameters for anthropometric measurements of interest relate a measurement at a given age to a precisely calculated Z-score (number of SDs from the mean) and percentile. Similarly, the expected anthropometric measurement at a particular Z-score and age can also be calculated. The availability of the LMS method and parameters for an increasing number of growth charts provides an opportunity to both improve clinical care of individual patients and allow large-scale analysis of datasets, which would be difficult or impossible if using paper growth charts.

Postnatal growth failure is common in preterm infants and is known to be associated with long-term neurodevelopmental impairment [10-18]. Extending the calculation of anthropometric measurement Z-scores from individual patients to a large population of patients might yield insight into how populations of preterm infants grow during their birth hospitalization. Similarly, we hypothesized that assessing the efficacy of quality improvement initiatives targeting improved growth might benefit from an unambiguous quantitative metric based on anthropometric Z-scores.

In this paper, we describe the deployment of the publicly accessible PediTools website, which implements a suite of calculators supporting LMS-based growth charts. We hypothesized that a simple metric to assess growth outcomes—the change in weight Z-score from birth to discharge (∆Z weight)—might yield insight into growth outcome variations. We retrospectively compared outcomes at 7 Massachusetts neonatal intensive care units (NICUs) and further utilized this metric to assess the efficacy of a nutrition-based quality improvement project at one of the NICUs. In addition, we also announce the release of open-source software, which will allow others to perform large-scale LMS-based calculations more easily.

The LMS method allows a parametric definition of growth references and generation of smoothed centile curves accounting for skewness of the distribution of an anthropometric measurement [9]. The parameters lambda (L, skewness normalization via power in the Box-Cox transformation), mu (M, mean), and sigma (S, coefficient of variation) describe the distribution of the measurement (eg, weight, length, or head circumference) at a given age, and the set of LMS parameters across multiple ages parameterizes the entire growth chart. This allows convenient calculation of exact Z scores (SDs from the mean) and generation of any centile curve.

The PediTools Web calculators [34] were developed in PHP, a general-purpose scripting language well suited to Web development [35]. The website was generated using RapidWeaver version 7.5.7 (Realmac Software) [36]. Web hosting is currently provided by Bluehost Inc [37], under a shared hosting environment. Access statistics are tracked via Google Analytics. The PediTools Web server is configured to require the use of Secure Sockets Layer to encrypt traffic to and from the Web server. No data entered as inputs for the medical calculators are saved or analyzed.

For the PediTools Electronic Growth Chart, a Microsoft Excel spreadsheet template was designed to allow users to enter protected health information locally, but it would calculate the nonprotected health information values needed to generate a growth chart. Specifically, the date of birth, gender, gestational age at birth, and specific dates and measurements are entered, but only the gender and calculated postmenstrual ages and anthropometric measures are submitted to the PediTools electronic growth chart site, avoiding transmission of any specific dates.

For the PediTools electronic growth chart, LMS-based calculations were performed as above for all the measurements submitted. In addition, for sequential measurements, rate of weight change in grams per week, both observed and expected (to maintain the previous growth centile), were displayed.

As a visual aid to recognize excessive loss of weight Z-score between sequential measurements, after the first 10 days of life, the change in weight Z-score (∆Z) was color coded to display as red if the Z-score decreased by more than 0.06 SDs per week; yellow for decrease by more than 0.03 SD per week; and green otherwise. These thresholds were chosen somewhat arbitrarily, but over the course of a 14-week admission, each color would indicate an overall ∆Z weight of −0.84, −0.42, or less negative than −0.42 SDs, respectively.

The Vermont Oxford Network (VON) is a nonprofit voluntary collaboration of neonatal health care professionals representing more than 1200 hospitals around the world [38]. Deidentified data were obtained from 7 level 3 NICUs in Massachusetts, which participate in the VON registry. Eligibility criteria included birth year from 2006 to 2011; gestational age between 23 0/7 and 34 6/7 weeks; no severe congenital malformations; and survival to discharge. Availability of birth weight, discharge weight, and length of stay were required to calculate the weight Z-score at birth and discharge. Infants were excluded if birth or discharge weight Z-scores were less than −4 or greater than 4, as values beyond these extremes often reflected data entry error. The calculated outcome metric was the ∆Z weight from birth to discharge. For NICU C, data for neonates born between 2012 and 2017 were also obtained for postintervention quality improvement outcomes analysis.

The VON registry provides a manual of operations with data definitions and eNICQ software, which allows for the collection, error checking, and submission of infant data. These manuals, data collection forms, and electronic data submission instructions are all available on the VON website. At each hospital, individual patient-level data for that hospital were exported from eNICQ as a CSV file, with 1 row per patient and 1 column per data field. (As of 2019, eNICQ data exports are now in XML or JSON format, but they contain identical information.) The data columns abstracted for each hospital in this study included the following: birth year (BYEAR), initial gestational age (GAWEEKS, GADAYS), birth weight (BWGT), length of stay (LOS1), and discharge weight (DWGT). Additional information obtained included source of admission (inborn or outborn, LOCATE), day of life of admission (DAYADMISS), discharge disposition (home, transfer to another facility, or death, FDISP), and congenital malformations (CMAL). Gender was not obtained, as at the time the study was originally conceived, the only preterm growth chart with LMS parameters available (Fenton 2003) was not gender specific [19]. Outcomes obtained but not reported here included the following: birth (BHEADCIR) and discharge (DHEADCIR) head circumference, early (EBSEPS) or late (LBPATH) bacterial infection, oxygen requirement at 36 weeks postmenstrual age (OX36), necrotizing enterocolitis (NEC, NECSURG), and retinopathy of prematurity (ISTAGE, ROPSURG).

Analysis was performed using R, free software for statistical computing [39], using the free version of the RStudio integrated development environment [40]. For data visualizations, smoothed conditional mean curves were generated by the R ggplot [41] package via generalized additive model and cubic splines [42]. When present, the bands surrounding the smoothed curves represent the 95% CI around the mean.

Comparisons among hospitals were performed by 2-tailed t test, analysis of variance, Wilcoxon rank-sum, and Kruskal-Wallis test, as appropriate. Multihospital ∆Z analysis was performed by fitting a linear model of ∆Z versus gestational age, with interaction terms for both slope and intercept for each hospital. When multiple pairwise comparisons were performed, multiple testing adjustment was performed by the Tukey honestly significant difference method. The study was approved by the Institutional Review Boards at each of the hospitals that contributed data.

Multiple bundled growth and nutrition quality improvement interventions were essentially simultaneously implemented at NICU C, starting in late 2011. These bundled changes included the following: (1) raised awareness of baseline growth failure by educational presentations to clinicians, showing how growth outcomes differed between NICU C and NICU F; (2) development of an electronic growth chart, as described in the PediTools Web tool; (3) systematic weekly growth metric collection in a form compatible with the electronic growth chart tool; (4) formal review of all NICU patients and their interval growth at weekly multidisciplinary rounds with pediatric dieticians; (5) earlier and broader initiation of parenteral nutrition with increased protein content and more rapid advancement; (6) revision of enteral feeding advancement protocols, including earlier initiation of gut priming (trophic feeds).

Assessment of the effect of the bundled interventions was performed similar to the analysis of growth outcomes described above: ∆Z weight from admission to discharge was calculated for each patient and the results were analyzed over different birth year epochs.

The R code used for the calculation of anthropometric measure Z scores from LMS parameters was bundled [43] into the R peditools package, and this will be hosted on GitHub [44] and shared under the Massachusetts Institute of Technology (MIT) License. The peditools package can be installed using the devtools package [45], with the command install_github(“jhchou/peditools”).

All growth charts described in this work are supported by the R peditools package, with the exception of the Fenton 2013 growth chart [21], for which the LMS parameters are available from the author by license only. If the Fenton 2013 parameters become publicly available in the future, they will be added to the peditools R package. In the meantime, the Olsen 2010 [25] or gender nonspecific Fenton 2003 [19] charts can be used to analyze preterm growth.

The first PediTools Web calculator was developed in 2011 as an in-house tool to improve documentation of anthropometric measurements of premature newborns cared for at hospital C by allowing calculation of Z-scores and percentiles, using the Fenton 2003 preterm growth chart [19], for which LMS parameters were published in 2007 [20]. The webpage was moved to public hosting in January of 2012. A screenshot of a representative Web-based PediTools growth calculator is shown in Figure 2.

Although accessible to the general public, the target audience and purpose of the PediTools website are pediatric clinical providers’ bedside use. PediTools is agnostic to which growth charts are made available and does not provide recommendations as to which charts are appropriate for which populations. The users of the website are expected to exercise their own professional clinical judgment to determine suitability for their purposes.

Additional growth chart calculators have subsequently been added to PediTools, including support for the Fenton 2013 preterm chart [21], CDC infant and pediatric [22,23], World Health Organization (WHO) infant [24], Olsen 2010 preterm [25], WHO arm circumference and triceps and subscapular skinfold [26], CDC triceps and subscapular skinfold [27], Olsen 2015 BMI for preterm [28], Zemel 2015 Down syndrome [29,30], Abdel-Rahman 2017 midupper arm circumference [31], and Mramba 2017 midupper arm circumference [32].

As PediTools Web calculators were intended to be used by clinicians at the point of care, features in addition to reporting percentiles and Z-scores were integrated to promote ease of use and clinical relevance. For example, with the preterm calculators, a gestational age calculator was integrated to allow entry of either the postmenstrual gestational age of interest or any combination of last menstrual period, due date, delivery date, or chronologic age. For assessment of obesity, the CDC pediatric growth calculator includes both the Z score for BMI and updated categorization of extreme obesity, defined as BMI≥120% of the 95th percentile or ≥35 kg/m² [46]. Both international and imperial units are supported. For infant calculators, calculations at both chronologic age and age corrected for prematurity can be reported, which is beneficial when assessing for timely attainment of developmental milestones. To help set goals for future growth, the calculators report the expected amount that anthropometric measures should increase over time to maintain the current Z-score (ie, equivalent to growing along the current percentile curve).

As poor growth trajectory might exacerbate long-term neurodevelopmental impairment, particularly for those patients born the most preterm, NICU C embarked on a multifocal quality improvement project to reduce the loss in weight Z-score from birth to discharge. Bundled interventions introduced in 2011 targeting factors potentially contributing to poor growth included the following:

These interventions were associated with a significant reduction in loss of weight Z-score, particularly at the lowest gestational ages, which was sustained and progressive (Figure 7). Compared with the preintervention epoch 2006 to 2008, there was a marked reduction in loss of birth weight Z-score in the postintervention epoch from 2012 to 2014.

Using the 2006 to 2008 epoch as a reference and fitting a linear model of ∆Z weight versus gestational age at birth, with interaction terms for both slope and intercept (centered at 29 weeks) for the remaining 3 epochs of 2009 to 2011, 2012 to 2014, and 2015 to 2017, showed significant differences in both the ∆Z weight at 29 weeks (differences in intercept, P=.03, <.001, and <.001) and association with gestational age at birth (differences in slope, P=.015, <.001, and <.001, respectively).

In fact, because of concern that the degree of reduction of weight Z-score loss at the lowest gestational ages might be excessive in the 2012 to 2014 epoch, potentially contributing to future development of the metabolic syndrome [12], a less aggressive approach was adopted. The subsequent growth target was a ∆Z from birth to discharge of roughly −0.6 for all gestations, resulting in the more flattened curve of epoch 2015 to 2017 (Figure 7).

The PediTools website has met a need for clinicians wishing to analyze data for an individual patient at a time. In contrast, the multisite comparison of growth outcomes of thousands of patients at 7 NICUs yielded additional insight into growth patterns of preterm newborns and prompted a successful quality improvement project at 1 NICU. Here, we describe the release of open-source computer code to permit others to conveniently do similar analysis, which should be useful for both clinical research and quality improvement monitoring.

The peditools R package will be made available on GitHub and provide functions to work with LMS-based anthropometric charts, including all LMS parameters possible [44]. All growth charts available on the PediTools website (listed in Table 1), with the exception of the Fenton 2013 preterm growth chart, are included. At this time, the Fenton 2013 chart LMS parameters are available by license only and are restricted from being shared. As additional charts are added to the PediTools website, the peditools R package will also be updated to include the new charts. The peditools R package will be released under the relatively permissive MIT License, which allows for commercial use, modification, redistribution, and sublicensing.

The primary package tool is the peditools::x_to_z() function, which takes as inputs a vector of anthropometric measurements, a vector of ages, a vector of genders, and a uniquely specified chart and measure, and which outputs a vector of Z-scores. In addition, helper functions peditools::recode_von() and peditools::recode_von_xml() are included to easily import VON datasets (exported as a CSV or XML file) into the R environment for analysis.

Previous work has analyzed growth of large populations of preterm newborns. Horbar et al [66] drew on data obtained from the full Vermont Oxford Registry on 362,833 newborns born between 2000 and 2013, with birth weight from 501 to 1500 grams. In this large, aggregate population, they reported improvements of growth velocity and a decrease in discharge with growth failure and severe growth failure (defined as discharge at less than the 10th and 3rd percentiles), across the time period from 2000 to 2013. Similarly, Griffin et al [67] reported on 25,899 infants born in California, with birth weight from 500 to 1500 grams or gestational age from 22 to 32 weeks, born between 2005 and 2012. They demonstrated a reduction in fall in weight Z-score between birth and discharge over the time period, as well as a reduction in the proportion of infants discharged home below the 10th percentile for weight or ∆Z weight less than −1. We see similar improvements in less negative ∆Z weight over time, comparing birth years 2006 to 2008 versus 2009 to 2011 (Figure 5).

Although both Horbar et al [66] and Griffin et al [67] report outcomes by birth weight (binned into categorical groups of 250 gram increments and which would therefore include both large for gestation more premature and small for gestation less premature newborns), previous studies have not reported the association described here between growth outcomes as ∆Z weight and gestational age at birth (both continuous variables).

Both of these studies report on the important findings of overall population-level improvement in growth outcomes in preterm newborns across a time period ranging from 2000 to 2013, which likely reflects clinical practice changes across the field of neonatology as a whole, which may result in improved long-term outcomes. However, this information is less helpful for clinicians attempting to assess outcomes at the local hospital level. In addition, the previous studies have not shared the tools needed to make it convenient to perform this analysis on new populations. In fact, without tools to easily assess growth, it is not easy for clinicians to even recognize that there might be an issue with growth outcomes in their patient populations.

A major goal of this study is to make tools available, allowing others to perform their own large-scale growth outcomes analysis, facilitating future research to better describe ideal growth that will lead to optimal long-term outcomes. More than 1200 hospitals around the world participate in the VON registry, and an increasing number of clinical sites use electronic health records from which anthropometric data can be extracted. With this increased availability of growth data, clinicians can easily replicate this analysis in a very small amount of code, using free and open-source tools, including the R statistical programming language, the RStudio integrated development environment, the ggplot2 R visualization package [39-41], and the peditools package described here [44].

Tools to perform LMS-based growth chart calculations have been made available on a public website and are highly utilized by clinical caregivers worldwide. Applying these methods to a large population of preterm newborns demonstrated widespread overall loss in weight Z-score from birth to discharge; that the magnitude of loss was associated with increasing prematurity, the population at the highest risk of poor neurodevelopment outcomes; and that there was significant interhospital variation in growth outcomes. At 1 site, these tools provided a convenient and reliable outcome measure for a clinical quality improvement project targeting growth. With this report, release of open-source code that implements LMS-based calculations will allow other clinicians and investigators to conveniently perform similar analyses with the promise to improve long-term outcomes in these high-risk pediatric patients.