- Email: [email protected]
Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices
YJPSU-58821; No of Pages 5 Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
Contents lists available at ScienceDirect
Journal of Pediatric Surgery journal homepage: www.elsevier.com/locate/jpedsurg
Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices Young Mee Choi a,⁎, Kristen Campbell b, Kari Hayes b,c, Rebecca Jacobson a, Gregory Kobak b,d, Steven Moulton a,b a
Pediatric Surgery, Children's Hospital Colorado, 13123 E.16th Avenue, Aurora, CO 80045 University of Colorado School of Medicine, 13001 E 17th Pl, Aurora, CO 80045 Pediatric Radiology, Children's Hospital Colorado, 13123 E 16th Ave, Aurora, CO 80045 d Pediatric Gastroenterology, Children's Hospital Colorado, 13123 E 16th Ave, Aurora, CO 80045 b c
a r t i c l e
i n f o
Article history: Received 15 March 2018 Received in revised form 3 August 2018 Accepted 6 August 2018 Available online xxxx Key words: Gastrostomy Abdominal wall thickness Ultrasonography Enteral access device
a b s t r a c t Objectives: Abdominal wall thickness (AWT) is a key measurement when placing or replacing low profile gastrostomy devices. This measurement varies, depending on nutritional status and body habitus. We developed a mathematical model to estimate AWT using a compendium of body measurements. Methods: Ultrasonography was used to measure AWT at the initial gastrostomy site in subjects aged 22 days to 24 years old. Other body measurements (height, weight, waist circumference and distance from xiphisternum to pubis) were also obtained. Multiple linear regression was used to develop two separate models using age of 2 years to separate the groups. For analysis, AWT is log transformed. Results: Data from 97 subjects were used for analysis. The final model for those ≤24 months old is the following: ln(Estimated AWT) = −1.255 + 0.082*(1 if age 3–24 months, 0 if b3 months) + 0.022*(waist circumference in cm). The final model for those N24 months old is the following: ln(Estimated AWT) = −1.335 + 0.271*(1 if age N84 months, 0 if 24–84 months) + 0.082*(BMI) Conclusion: This model to estimate AWT is useful for determining the length of a gastrostomy device at initial placement and with subsequent changes. More data are needed to refine and further validate the model. Level of evidence: Level IV, study of prognostic test. © 2018 Elsevier Inc. All rights reserved.
Gastrostomy devices are commonly placed in children who require long-term enteral access for nutrition. Low-profile gastrostomy buttons are the preferred type of enteral access device in young children for patient comfort and easier handling [1]. A main determinant when selecting the correct length for a low-profile gastrostomy button is the abdominal wall thickness (AWT). This value should correspond to the length of the device shaft, plus approximately 2 mm for a gauze sponge and subsequent growth. Predicting the AWT in this population can be challenging, however, owing to varying degrees of malnutrition, complex comorbidities and wide age range. One would expect to find a correlation between nutritional status and the appropriate corresponding length of the gastrostomy device shaft, but individual variability –
Abbreviations: AWT, Abdominal wall thickness; BMI, Body mass index; CI, Confidence interval; MAE, Mean absolute error; PEG, Percutaneous endoscopic gastrostomy. ⁎ Corresponding author at: 13123 E.16th Avenue B245, Aurora, CO 80045. Tel.: +1 720 777 5371; fax: +1 720 777 6571. E-mail address: [email protected] (Y.M. Choi).
even in the same age or weight range – makes it difficult to estimate the AWT in such a diverse population [1]. Moreover, a nutritional status marker such as body mass index (BMI) is unreliable in children under 2 years of age [2]. Additionally, as children gain weight and their AWT increases the length of the shaft of their gastrostomy button needs to be increased, making the correct estimation of device length even more important. Ultrasonography is a widely accepted tool for measuring AWT [3,4]. Unfortunately, availability of ultrasonography is limited in outpatient settings, where most gastrostomy device changes occur. Another common method of measuring the length of a preexisting stoma tract is with a plastic sizer, but this measurement can be inaccurate owing to soft tissue inflammation and thickening [5], and can be painful especially in children with stoma-related complications. Thus, a prediction model of AWT using simple clinical parameters would be useful. Herein, we describe a mathematical model for estimating AWT using simple body measurements, for children undergoing low-profile gastrostomy device placement or replacement.
https://doi.org/10.1016/j.jpedsurg.2018.08.052 0022-3468/© 2018 Elsevier Inc. All rights reserved.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
2
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
1. Methods This study was approved by the Colorado Multiple Institutional Review Board (COMIRB). Patients aged 1 day to 24 years, who were scheduled to undergo placement of a gastrostomy device at Children's Hospital Colorado between July 2016 and May 2017, were eligible for the study. Informed consent was obtained from all study participants prior to research activity. Exclusion criteria included missing AWT measurements and previous complex or multiple abdominal surgeries that might influence abdominal wall measurements. All ultrasound measurements were obtained by the same surgery resident (Choi), who was trained by the pediatric radiologist (Hayes) in measuring AWT. A GE Healthcare Venue 50 Ultrasound was used in musculoskeletal mode to obtain the images and measurements. Patients were placed supine at the time of the measurement. A linear probe was placed over the proposed gastrostomy site, or over the rectus muscle one finger breadth below the left costal margin in transverse view, while ensuring minimal pressure. The measurements were obtained four times, and the mean of the four measurements was used for analysis. The Intra Class Correlation Coefficient was also calculated between the four measurements on the same individual and reported in Section 2.2. The images and measurements on the first three study subjects were reviewed with the pediatric radiologist to ensure the accuracy of the measurements. Other body measurements [height (cm), weight (kg), BMI, waist circumference (cm) and distance from xiphisternum to the top of the pubis (cm)] were also obtained at the time of the ultrasonography measurement. Demographic and clinical data included age, gender, comorbidities, indication for the gastrostomy device and presence of previous abdominal surgery. 1.1. Statistical analysis Continuous variables were summarized as median and interquartile range (IQR), and categorical variables were summarized as counts and proportions. The following possible predictors were candidates in the models to predict AWT: age, gender, height, weight, BMI, waist circumference and xiphisterm to pubis length. The following process was followed to determine the best predictive model of AWT: first, univariate analyses were performed for each predictor versus AWT, and the residuals were plotted to assess the assumptions of independence, homoscedasticity, normality, and linearity. Owing to the skewness in the distribution of AWT, it was log transformed for analysis. Second, since BMI is known to be a significant predictor of AWT based on literature review, and it is only clinically proven to be a useful measurement in children more than 2 years of age [2], the patients were divided into two age groups and separate models were fit. Model 1 contained the 55 patients aged 0–24 months, while Model 2 contained the 42 patients older than 24 months. Third, univariate analyses were performed again within each subset of children. Treating age as a continuous, linear predictor violated the homoscedasticity assumption in both age groups. Thus, age was split into a categorical variable (b 3 months, 3–24 months, 24–84 months, and 84 + months). Variables that were significantly associated with AWT in the univariate analyses (p b 0.10) were considered for the final models. All hypothesis testing used a significance level of 0.05, and all analyses were performed using R v.3.4.1 (https://www.r-project.org/). 1.2. Internal validation of the models Internal validation of the prediction models was assessed using 1000 bootstrap iterations with replacement. A bootstrap sample is a random selection of patients from the original data that are selected to form a new dataset of the same size as the original, such that some patients are selected more than once and some not at all [6]. One thousand bootstrap samples were used to fit the original model and the mean absolute
error (MAE) [7] between the predicted and observed values of the natural log of AWT, both of which were calculated for each sample. The average MAE and 95% confidence interval (CI) of the 1000 iterations were reported as a measure of the difference between predicted values and observed natural log of AWT. These errors on the natural log scale can be interpreted as the mean absolute percentage error in predicting the original measurements. 2. Results 2.1. Demographics A total of 108 subjects were enrolled between July 2016 and May 2017. Ten subjects were excluded owing to incomplete AWT measurements and one was excluded owing to a history of a giant omphalocele, which required a complex closure for repair. Three subjects eventually did not undergo gastrostomy device placement, but were included for analysis as their AWT measurements were available. Analysis was performed on the remaining 97 subjects. Median age was 20 months (IQR: 3–96) and 57 were male (59%). The most common comorbidity was neuromuscular in 51 subjects (56%), and the most common indication for gastrostomy device was failure to thrive in 72 subjects (77%). Demographics of the subjects are show in Table 1. 2.2. Models for abdominal wall thickness The mean of four measurements of the AWT obtained using ultrasound was used as the outcome. The intraclass correlation coefficient between the four measurements was estimated to be 0.965 (95% CI: 0.953, 0.975), indicating high agreement between measurements on the same study subject. Model 1 contained subjects who are 0–24 months of age. The log transformed AWT was the outcome, and potential predictors were categorical age, gender, height, weight, BMI, waist circumference, and distance from xiphisternum to pubic bone. Owing to the high collinearity between height, weight, BMI, waist circumference, and distance from xiphisternum to pubic bone, the predictor that resulted in the best model fit from univariate analyses (assessed using R 2), was used in the final model. The final Model 1 included categorical age (b 3 months vs. 3–24 months) and waist circumference, and the R2 value was 0.42. The same process was followed for Model 2 with subjects 24– 298 months of age, and the final model included categorical age (24– 84 months, 84+ months) and BMI, and the model fit was better than Table 1 Subject demographics. Variables
Summary Stat (n = 97)
Age, months, median (IQR) Male, n (%) BMI, kg/m2, median (IQR) Waist circumference, cm, median (IQR) Xiphisternum to pubis length, cm, median (IQR) Comorbidities Neuromuscular, n (%) Respiratory, n (%) Ventilator dependence, n (%) Genetic/Metabolic, n (%) Cardiac, n (%) Hematologic/malignancy, n (%) Indications for gastrostomy device placementa Failure to thrive, n (%) Aspiration, n (%) Other, n (%) Operative technique Laparoscopic, n (%) Laparoscopic assisted PEG, n (%) PEG, n (%) Open, n (%) Prior abdominal surgery, n (%)
20 (3–96) 57 (59) 14.8 (13.5–16.6) 45 (39–56) 16 (13–20)
a
51 (56) 38 (39) 17 (18) 39 (40) 25 (26) 2 (2) 72 (77) 15 (15) 15 (15) 77 (82) 12 (13) 2 (2) 3 (3) 4 (4)
Some subjects had more than one indication.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
3
Fig. 1. Model 1 for subjects ≤2 years. Predicted values from the final model versus abdominal wall thickness and residual plot of the model.
in Model 1, R2 = 0.62. The predicted model fit and residuals were plotted for each of the two models and do not violate any assumptions of the model (Figs. 1 and 2). The univariate and multivariable regression models are shown in Table 2 (Model 1) and Table 3 (Model 2). The final model for those ≤2 years is the following: ln(Estimated AWT) = − 1.255 + 0.082*(1 if age 3–24 months, 0 if b3 months) + 0.022*(waist circumference in cm).
The final model for those N 2 years is the following: ln(Estimated AWT) = −1.335 + 0.271*(1 if age N84 months, 0 if 24–84 months) + 0.082*(BMI). For example, for a child who is 6 months old with a waist circumference of 46 cm, the calculation would be, ln (Estimated AWT) = −1.255 + 0.082*(1) + 0.022*(46). Therefore, the estimated AWT would be 0.85 cm.
Fig. 2. Model 2 for subjects N2 years. Predicted values from the final model versus abdominal wall thickness and residual plot of the model.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
4
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
Table 2 Model 1: patients ≤2 years old (n = 55). Natural Log of Abdominal Wall Thickness (cm) as the outcome MULTIVARIABLE (R2 = 0.42)
UNIVARIATE Estimate Intercept Age (3–24 months) Gender (Male) Weight BMI Waist Circumference (cm) Xiphisternum to Pubis (cm)
0.254 −0.025 0.046 0.064 0.027 0.042
95% CI 0.140, 0.369 −0.152, 0.101 0.027, 0.065 0.041, 0.086 0.018, 0.035 0.019, 0.065
2.3. Internal validation of the models Using 1000 bootstrap samples, the MAE between the predicted and observed log AWT was estimated to be 0.14 (95% CI: 0.11, 0.16) for Model 1 (≤2 years of age). With 95% confidence, the predicted values of the AWT deviate by between 11% and 16% of the original values of AWT. For Model 2 (N2 years of age), the MAE was estimated to be 0.23 (95% CI: 0.18, 0.29). In other words, we estimate that the predicted values of AWT deviate by about 23% of the original values of AWT. 2.4. Comparison of the model with the surgeons' measurements We compared how well the predicted values from the models compared to the surgeons' intraoperative abdominal wall measurements that were obtained in 65 subjects, using ultrasound measurements as the gold standard. We took the absolute value of the predicted measurement minus the ultrasound measurement and compared this to the absolute value of the surgeon measurement minus the ultrasound measurement. Our results showed that the predicted measurements from the model, on average, were 0.10 cm (95% CI: 0.07, 0.13) different from the ultrasound measurement, while the surgeon measurement values, on average, were 0.15 cm (95% CI: 0.12, 0.19) different. This difference was statistically significant (p = 0.03). 3. Discussion It is estimated that up to 11,000 gastrostomy devices or other types of feeding devices are placed in children each year, and the number has steadily increased from 16.6 per 100,000 children in 1997 to 18.5 in 2009 [8,9]. Low-profile gastrostomy buttons are the preferred enteral access devices for young children as they can be hidden under clothes and increase comfort [1]. These devices are available in a variety of lengths and the chosen length should correspond to the child's AWT. Unlike adults, growth and weight gain in the pediatric population require more frequent gastrostomy device size adjustment, in order to accommodate changes in weight and body habitus [1]. A gastrostomy device that is too short or tight for a child can cause excessive pressure around the stoma tract, and the pressure from the external and/or
p-value b0.001 0.695 b0.001 b0.001 b0.001 0.001
Estimate
95% CI
p-value
−1.255 0.082
−1.634, −0.876 −0.048, 0.211
b0.001 0.221
0.012, 0.033
b0.001
0.022
internal bolsters can decrease blood flow to the area and cause impaired wound healing and cellulitis [10,11]. These devices can also cause skin necrosis, erosion and shortening of the stoma tract leading to increased leakage, and buried bumper syndrome, which can be life-threatening [12–14]. Similarly, a long gastrostomy device has been shown to significantly increase the risk of leakage [15], possibly owing to widening of the tract. In the same study, children with longer gastrostomy devices (≥1.5 cm) experienced an increased incidence of granuloma formation around the gastrostomy compared to those with shorter length (≤1.2 cm) (40% versus 29%), although the difference was not statistically significant [15]. In our experience, we have noticed that gastrostomy devices that are too long for the child tend to move around the stoma tract more freely and cause peristomal irritation. It is recommended that low-profile gastrostomy devices be replaced every three to six months, depending on the design and manufacturer of the device. As a child begins to gain weight following initial gastrostomy device placement, the length of the device shaft needs to be upsized at regular intervals, to accommodate growth in the abdominal wall. It is difficult, however, to know the growth in the AWT since the time of the initial gastrostomy, given the wide range in age groups, nutritional intake and duration of follow up in the pediatric population. Furthermore, most of these children receive medical care at primary care or small hospital settings, where the need to estimate a child's AWT may be an unfamiliar task. Thus, an easy and accurate method of estimating AWT would be beneficial in this scenario. A method of measuring the AWT at the time of initial placement is to pass a fine needle through the abdominal wall at the proposed site under laparoscopic visual guidance, then clamp the needle at skin level and measure the length. This is performed intraoperatively by some surgeons to gain a sense of the child's AWT. They then add 3–4 mm when determining the overall length of the low-profile gastrostomy device that will be placed, to account for thinning of the abdominal wall during insufflation, plus stomach wall thickness and a one-layer gauze dressing. This method is only applicable at the time of initial gastrostomy device placement and cannot be used when a child presents to have his or her gastrostomy device replaced and potentially upsized. A common method for measuring AWT when changing a gastrostomy device is to insert a plastic sizer in a preexisting
Table 3 Model 2: patients N2 years old (n = 42). Natural Log of Abdominal Wall Thickness (cm) as the outcome MULTIVARIABLE (R2 = 0.62)
UNIVARIATE Estimate Intercept Age (84–298 months) Gender (Male) Weight BMI Waist Circumference (cm) Xiphisternum to Pubis (cm)
0.440 −0.090 0.023 0.089 0.025 0.055
95% CI 0.148, 0.731 −0.393, 0.213 0.018, 0.028 0.066, 0.113 0.018, 0.031 0.032, 0.078
p-value 0.005 0.564 b0.001 b0.001 b0.001 b0.001
Estimate
95% CI
p-value
−1.335 0.271
−1.703, −0.967 0.071, 0.472
b0.001 0.011
0.060, 0.105
b0.001
0.082
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
gastrostomy tract. Unfortunately, soft tissue thickening, skin indentation around the stoma site or erosion of the balloon into the abdominal wall is common with long-term gastrostomy devices and this method can lead to inaccurate measurements [5]. We used ultrasonography to obtain AWT measurements in our study. Ultrasonography has been successfully used to visualize and confirm proper replacement of gastrostomy devices [16,17], but its use as an adjunct for gastrostomy device size selection has not been described. Its ease of use and the ability to visualize soft tissue make it an ideal tool to measure the AWT for gastrostomy device size selection. Its limited availability in outpatient settings, however, makes its application unrealistic. Instead, we used ultrasonography to collect accurate AWT measurements from patients requiring placement of gastrostomy devices, which were, along with other patient-specific variables, used to develop a patient database to create models that can be used to determine the most appropriate gastrostomy device length in outpatient settings. Significant determinants of AWT in children have been previously reported in several studies [18–21]. Certain factors such as age and gender have been found to be associated with abdominal wall fat [18,20]. Our study, however, did not show gender to be a significant factor associated with AWT, when adjusting for other variables. Furthermore, most of our study population had varying degrees of malnutrition, which was not adjusted for, owing to a lack of nutritional data such as skinfold thickness or biochemical markers (e.g. albumin). Instead, we used body measurements as surrogate markers of nutritional status. The use of growth charts from the WHO and the CDC is recommended to assess the growth of children less than 2 years of age and children aged 2–19 years, respectively [2]. Z-scores (standard deviation scores) using a child's gender, age and weight or height are used to evaluate a child's nutritional status and growth. We had initially calculated and used weight-for-age z-scores in our model for estimating AWT, and it showed a statistically significant relationship. Nonetheless, our goal in developing this model was to be able to easily use it in a clinical setting, and calculation of z-scores in addition to the equation to predict the AWT would be a burden on clinicians and an unrealistic expectation. Therefore, we divided our study cohort into two groups, using the age of 2 years as a cutoff. This was done to avoid using BMI as a predictor of nutritional status in the younger group [2,22]. The final models (Model 1 and Model 2) to estimate AWT only require age and waist circumference for those ≤ 2 years of age, and age and BMI for those N 2 years of age. Not surprisingly, the model did not fit the data as well for the younger group compared to the older group (R2 = 0.42 vs. 0.62). This may be partly owing to potential measurement errors in the AWT, because their AWT tends to be very thin. Furthermore, our measurements did not account for variables such as fluid level or anasarca, which may increase AWT. The main limitation of this study is the small sample size, which limited our ability to perform external validation on an independent sample and to assess for relationships between comorbidities and AWT. The sample size was further reduced when the model was split into two age groups, but we decided this was important to ensure reliable measurements and clinical usability. We performed internal validation on the final models, which demonstrated that the model for younger patients performed slightly better than the model for older patients. Furthermore, 52% of our study population was aged 20 months or less. The skewed age distribution in our study reflects the true pediatric population that undergoes gastrostomy device placement. Nonetheless, data on older children would make the dataset more robust. A second limitation of this study is the fact that AWT was measured after general anesthesia was induced in some children. This may have led to variations in the degree of abdominal wall relaxation, falsely increasing the measured thickness in those who were awake and able to contract their abdominal wall muscles. Efforts were made, however, to measure AWT at the time of relaxation in those who were awake during the measurement. Another limitation is that these estimations may not be applicable for children with long-term gastrostomy devices that have caused
5
scarring or contracture around the gastrostomy sites. However, our overall goal is to avoid such severe scarring and soft tissue indentation by properly sizing the gastrostomy devices from the time of initial establishment of the gastrostomy tract and regularly upsizing the devices as the child continues to grow. The complexity of the models may limit their use in clinical settings, but simple web-based applications can be developed to perform the calculations. Lastly, we have not shown whether the implementation of these equations in determining the size of low-profile gastrostomy devices is effective at reducing the incidence of gastrostomy tract complications. Nonetheless, the aim of our current study was to develop a model to predict a child's AWT based on simple body measurements, and a large prospective trial will need to be undertaken to answer the question of its clinical effectiveness. 4. Conclusions In summary, it is of paramount importance to upsize gastrostomy devices at regular intervals for children who are gaining weight and growing. Selecting the appropriate gastrostomy device length can be a challenging task owing to difficulty estimating a child's AWT. Our mathematical models of AWT provide a useful tool using simple measurements, for those who do not have access to ultrasonography in an office, emergency department or small hospital setting. References [1] Buderus S, Adenaeuer M, Dueker G, et al. Balloon gastrostomy buttons in pediatric patients: evaluation with respect to size, lifetime in patients, and parent acceptance. Klin Padiatr 2009;221:65–8. [2] Grummer-Strawn LM, Reinold C, Krebs NF, et al. Use of World Health Organization and CDC growth charts for children aged 0–59 months in the United States. MMWR Recomm Rep 2010;59:1–15. [3] Jain N, Goyal N, Mukherjee K, et al. Ultrasound of the abdominal wall: what lies beneath? Clin Radiol 2013;68:85–93. [4] Bashir U, Moskovic E, Strauss D, et al. Soft-tissue masses in the abdominal wall. Clin Radiol 2014;69:e422–31. [5] Chang WK, Huang WC, Yu CY, et al. Long-term percutaneous endoscopic gastrostomy: characteristic computed tomographic findings. Abdom Imaging 2011;36:684–8. [6] Kuhn M, Johnson K. Applied predictive modeling. New York: Springer; 2013. [7] Willmott JC, Matsuura K. Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance; 2005. [8] Goldberg E, Kaye R, Yaworski J, et al. Gastrostomy tubes: facts, fallacies, fistulas, and false tracts. Gastroenterol Nurs 2005;28:485–93 [quiz 93-4]. [9] Fox D, Campagna EJ, Friedlander J, et al. National trends and outcomes of pediatric gastrostomy tube placement. J Pediatr Gastroenterol Nutr 2014;59:582–8. [10] DeLegge M, DeLegge R, Brady C. External bolster placement after percutaneous endoscopic gastrostomy tube insertion: is looser better? JPEN J Parenter Enteral Nutr 2006;30:16–20. [11] McClave SA, Neff RL. Care and long-term maintenance of percutaneous endoscopic gastrostomy tubes. JPEN J Parenter Enteral Nutr 2006;30:S27–38. [12] Pattamanuch N, Novak I, Loizides A, et al. Single-center experience with 1-step lowprofile percutaneous endoscopic gastrostomy in children. J Pediatr Gastroenterol Nutr 2014;58:616–20. [13] Stewart CE, Mutalib M, Pradhan A, et al. Short article: buried bumper syndrome in children: incidence and risk factors. Eur J Gastroenterol Hepatol 2017;29:181–4. [14] Segal D, Michaud L, Guimber D, et al. Late-onset complications of percutaneous endoscopic gastrostomy in children. J Pediatr Gastroenterol Nutr 2001;33:495–500. [15] Helen S, Torsten Larsson L, Arnbjörnsson E. Postoperative gastrostomy site leakage correlated to the dimension of the gastrostomy button in children. Gastroenterol Insights 2010;2:34–6. [16] Wu TS, Leech SJ, Rosenberg M, et al. Ultrasound can accurately guide gastrostomy tube replacement and confirm proper tube placement at the bedside. J Emerg Med 2009;36:280–4. [17] Marin JR, Lewiss RE. Point-of-care ultrasonography by pediatric emergency medicine physicians. Pediatr Emerg Care 2015;31:525. [18] Brei C, Much D, Heimberg E, et al. Sonographic assessment of abdominal fat distribution during the first year of infancy. Pediatr Res 2015;78:342–50. [19] Balaguru D, Bhalala U, Haghighi M, et al. Computed tomography scan measurement of abdominal wall thickness for application of near-infrared spectroscopy probes to monitor regional oxygen saturation index of gastrointestinal and renal circulations in children. Pediatr Crit Care Med 2011;12:e145–8. [20] Butler M, Servaes S, Srinivasan A, et al. US depiction of the appendix: role of abdominal wall thickness and appendiceal location. Emerg Radiol 2011;18:525–31. [21] Sandler G, Leishman S, Branson H, et al. Body wall thickness in adults and children— relevance to penetrating trauma. Injury 2010;41:506–9. [22] Nash A, Secker D, Corey M, et al. Field testing of the 2006 World Health Organization growth charts from birth to 2 years: assessment of hospital undernutrition and overnutrition rates and the usefulness of BMI. JPEN J Parenter Enteral Nutr 2008;32:145–53.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
Contents lists available at ScienceDirect
Journal of Pediatric Surgery journal homepage: www.elsevier.com/locate/jpedsurg
Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices Young Mee Choi a,⁎, Kristen Campbell b, Kari Hayes b,c, Rebecca Jacobson a, Gregory Kobak b,d, Steven Moulton a,b a
Pediatric Surgery, Children's Hospital Colorado, 13123 E.16th Avenue, Aurora, CO 80045 University of Colorado School of Medicine, 13001 E 17th Pl, Aurora, CO 80045 Pediatric Radiology, Children's Hospital Colorado, 13123 E 16th Ave, Aurora, CO 80045 d Pediatric Gastroenterology, Children's Hospital Colorado, 13123 E 16th Ave, Aurora, CO 80045 b c
a r t i c l e
i n f o
Article history: Received 15 March 2018 Received in revised form 3 August 2018 Accepted 6 August 2018 Available online xxxx Key words: Gastrostomy Abdominal wall thickness Ultrasonography Enteral access device
a b s t r a c t Objectives: Abdominal wall thickness (AWT) is a key measurement when placing or replacing low profile gastrostomy devices. This measurement varies, depending on nutritional status and body habitus. We developed a mathematical model to estimate AWT using a compendium of body measurements. Methods: Ultrasonography was used to measure AWT at the initial gastrostomy site in subjects aged 22 days to 24 years old. Other body measurements (height, weight, waist circumference and distance from xiphisternum to pubis) were also obtained. Multiple linear regression was used to develop two separate models using age of 2 years to separate the groups. For analysis, AWT is log transformed. Results: Data from 97 subjects were used for analysis. The final model for those ≤24 months old is the following: ln(Estimated AWT) = −1.255 + 0.082*(1 if age 3–24 months, 0 if b3 months) + 0.022*(waist circumference in cm). The final model for those N24 months old is the following: ln(Estimated AWT) = −1.335 + 0.271*(1 if age N84 months, 0 if 24–84 months) + 0.082*(BMI) Conclusion: This model to estimate AWT is useful for determining the length of a gastrostomy device at initial placement and with subsequent changes. More data are needed to refine and further validate the model. Level of evidence: Level IV, study of prognostic test. © 2018 Elsevier Inc. All rights reserved.
Gastrostomy devices are commonly placed in children who require long-term enteral access for nutrition. Low-profile gastrostomy buttons are the preferred type of enteral access device in young children for patient comfort and easier handling [1]. A main determinant when selecting the correct length for a low-profile gastrostomy button is the abdominal wall thickness (AWT). This value should correspond to the length of the device shaft, plus approximately 2 mm for a gauze sponge and subsequent growth. Predicting the AWT in this population can be challenging, however, owing to varying degrees of malnutrition, complex comorbidities and wide age range. One would expect to find a correlation between nutritional status and the appropriate corresponding length of the gastrostomy device shaft, but individual variability –
Abbreviations: AWT, Abdominal wall thickness; BMI, Body mass index; CI, Confidence interval; MAE, Mean absolute error; PEG, Percutaneous endoscopic gastrostomy. ⁎ Corresponding author at: 13123 E.16th Avenue B245, Aurora, CO 80045. Tel.: +1 720 777 5371; fax: +1 720 777 6571. E-mail address: [email protected] (Y.M. Choi).
even in the same age or weight range – makes it difficult to estimate the AWT in such a diverse population [1]. Moreover, a nutritional status marker such as body mass index (BMI) is unreliable in children under 2 years of age [2]. Additionally, as children gain weight and their AWT increases the length of the shaft of their gastrostomy button needs to be increased, making the correct estimation of device length even more important. Ultrasonography is a widely accepted tool for measuring AWT [3,4]. Unfortunately, availability of ultrasonography is limited in outpatient settings, where most gastrostomy device changes occur. Another common method of measuring the length of a preexisting stoma tract is with a plastic sizer, but this measurement can be inaccurate owing to soft tissue inflammation and thickening [5], and can be painful especially in children with stoma-related complications. Thus, a prediction model of AWT using simple clinical parameters would be useful. Herein, we describe a mathematical model for estimating AWT using simple body measurements, for children undergoing low-profile gastrostomy device placement or replacement.
https://doi.org/10.1016/j.jpedsurg.2018.08.052 0022-3468/© 2018 Elsevier Inc. All rights reserved.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
2
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
1. Methods This study was approved by the Colorado Multiple Institutional Review Board (COMIRB). Patients aged 1 day to 24 years, who were scheduled to undergo placement of a gastrostomy device at Children's Hospital Colorado between July 2016 and May 2017, were eligible for the study. Informed consent was obtained from all study participants prior to research activity. Exclusion criteria included missing AWT measurements and previous complex or multiple abdominal surgeries that might influence abdominal wall measurements. All ultrasound measurements were obtained by the same surgery resident (Choi), who was trained by the pediatric radiologist (Hayes) in measuring AWT. A GE Healthcare Venue 50 Ultrasound was used in musculoskeletal mode to obtain the images and measurements. Patients were placed supine at the time of the measurement. A linear probe was placed over the proposed gastrostomy site, or over the rectus muscle one finger breadth below the left costal margin in transverse view, while ensuring minimal pressure. The measurements were obtained four times, and the mean of the four measurements was used for analysis. The Intra Class Correlation Coefficient was also calculated between the four measurements on the same individual and reported in Section 2.2. The images and measurements on the first three study subjects were reviewed with the pediatric radiologist to ensure the accuracy of the measurements. Other body measurements [height (cm), weight (kg), BMI, waist circumference (cm) and distance from xiphisternum to the top of the pubis (cm)] were also obtained at the time of the ultrasonography measurement. Demographic and clinical data included age, gender, comorbidities, indication for the gastrostomy device and presence of previous abdominal surgery. 1.1. Statistical analysis Continuous variables were summarized as median and interquartile range (IQR), and categorical variables were summarized as counts and proportions. The following possible predictors were candidates in the models to predict AWT: age, gender, height, weight, BMI, waist circumference and xiphisterm to pubis length. The following process was followed to determine the best predictive model of AWT: first, univariate analyses were performed for each predictor versus AWT, and the residuals were plotted to assess the assumptions of independence, homoscedasticity, normality, and linearity. Owing to the skewness in the distribution of AWT, it was log transformed for analysis. Second, since BMI is known to be a significant predictor of AWT based on literature review, and it is only clinically proven to be a useful measurement in children more than 2 years of age [2], the patients were divided into two age groups and separate models were fit. Model 1 contained the 55 patients aged 0–24 months, while Model 2 contained the 42 patients older than 24 months. Third, univariate analyses were performed again within each subset of children. Treating age as a continuous, linear predictor violated the homoscedasticity assumption in both age groups. Thus, age was split into a categorical variable (b 3 months, 3–24 months, 24–84 months, and 84 + months). Variables that were significantly associated with AWT in the univariate analyses (p b 0.10) were considered for the final models. All hypothesis testing used a significance level of 0.05, and all analyses were performed using R v.3.4.1 (https://www.r-project.org/). 1.2. Internal validation of the models Internal validation of the prediction models was assessed using 1000 bootstrap iterations with replacement. A bootstrap sample is a random selection of patients from the original data that are selected to form a new dataset of the same size as the original, such that some patients are selected more than once and some not at all [6]. One thousand bootstrap samples were used to fit the original model and the mean absolute
error (MAE) [7] between the predicted and observed values of the natural log of AWT, both of which were calculated for each sample. The average MAE and 95% confidence interval (CI) of the 1000 iterations were reported as a measure of the difference between predicted values and observed natural log of AWT. These errors on the natural log scale can be interpreted as the mean absolute percentage error in predicting the original measurements. 2. Results 2.1. Demographics A total of 108 subjects were enrolled between July 2016 and May 2017. Ten subjects were excluded owing to incomplete AWT measurements and one was excluded owing to a history of a giant omphalocele, which required a complex closure for repair. Three subjects eventually did not undergo gastrostomy device placement, but were included for analysis as their AWT measurements were available. Analysis was performed on the remaining 97 subjects. Median age was 20 months (IQR: 3–96) and 57 were male (59%). The most common comorbidity was neuromuscular in 51 subjects (56%), and the most common indication for gastrostomy device was failure to thrive in 72 subjects (77%). Demographics of the subjects are show in Table 1. 2.2. Models for abdominal wall thickness The mean of four measurements of the AWT obtained using ultrasound was used as the outcome. The intraclass correlation coefficient between the four measurements was estimated to be 0.965 (95% CI: 0.953, 0.975), indicating high agreement between measurements on the same study subject. Model 1 contained subjects who are 0–24 months of age. The log transformed AWT was the outcome, and potential predictors were categorical age, gender, height, weight, BMI, waist circumference, and distance from xiphisternum to pubic bone. Owing to the high collinearity between height, weight, BMI, waist circumference, and distance from xiphisternum to pubic bone, the predictor that resulted in the best model fit from univariate analyses (assessed using R 2), was used in the final model. The final Model 1 included categorical age (b 3 months vs. 3–24 months) and waist circumference, and the R2 value was 0.42. The same process was followed for Model 2 with subjects 24– 298 months of age, and the final model included categorical age (24– 84 months, 84+ months) and BMI, and the model fit was better than Table 1 Subject demographics. Variables
Summary Stat (n = 97)
Age, months, median (IQR) Male, n (%) BMI, kg/m2, median (IQR) Waist circumference, cm, median (IQR) Xiphisternum to pubis length, cm, median (IQR) Comorbidities Neuromuscular, n (%) Respiratory, n (%) Ventilator dependence, n (%) Genetic/Metabolic, n (%) Cardiac, n (%) Hematologic/malignancy, n (%) Indications for gastrostomy device placementa Failure to thrive, n (%) Aspiration, n (%) Other, n (%) Operative technique Laparoscopic, n (%) Laparoscopic assisted PEG, n (%) PEG, n (%) Open, n (%) Prior abdominal surgery, n (%)
20 (3–96) 57 (59) 14.8 (13.5–16.6) 45 (39–56) 16 (13–20)
a
51 (56) 38 (39) 17 (18) 39 (40) 25 (26) 2 (2) 72 (77) 15 (15) 15 (15) 77 (82) 12 (13) 2 (2) 3 (3) 4 (4)
Some subjects had more than one indication.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
3
Fig. 1. Model 1 for subjects ≤2 years. Predicted values from the final model versus abdominal wall thickness and residual plot of the model.
in Model 1, R2 = 0.62. The predicted model fit and residuals were plotted for each of the two models and do not violate any assumptions of the model (Figs. 1 and 2). The univariate and multivariable regression models are shown in Table 2 (Model 1) and Table 3 (Model 2). The final model for those ≤2 years is the following: ln(Estimated AWT) = − 1.255 + 0.082*(1 if age 3–24 months, 0 if b3 months) + 0.022*(waist circumference in cm).
The final model for those N 2 years is the following: ln(Estimated AWT) = −1.335 + 0.271*(1 if age N84 months, 0 if 24–84 months) + 0.082*(BMI). For example, for a child who is 6 months old with a waist circumference of 46 cm, the calculation would be, ln (Estimated AWT) = −1.255 + 0.082*(1) + 0.022*(46). Therefore, the estimated AWT would be 0.85 cm.
Fig. 2. Model 2 for subjects N2 years. Predicted values from the final model versus abdominal wall thickness and residual plot of the model.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
4
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
Table 2 Model 1: patients ≤2 years old (n = 55). Natural Log of Abdominal Wall Thickness (cm) as the outcome MULTIVARIABLE (R2 = 0.42)
UNIVARIATE Estimate Intercept Age (3–24 months) Gender (Male) Weight BMI Waist Circumference (cm) Xiphisternum to Pubis (cm)
0.254 −0.025 0.046 0.064 0.027 0.042
95% CI 0.140, 0.369 −0.152, 0.101 0.027, 0.065 0.041, 0.086 0.018, 0.035 0.019, 0.065
2.3. Internal validation of the models Using 1000 bootstrap samples, the MAE between the predicted and observed log AWT was estimated to be 0.14 (95% CI: 0.11, 0.16) for Model 1 (≤2 years of age). With 95% confidence, the predicted values of the AWT deviate by between 11% and 16% of the original values of AWT. For Model 2 (N2 years of age), the MAE was estimated to be 0.23 (95% CI: 0.18, 0.29). In other words, we estimate that the predicted values of AWT deviate by about 23% of the original values of AWT. 2.4. Comparison of the model with the surgeons' measurements We compared how well the predicted values from the models compared to the surgeons' intraoperative abdominal wall measurements that were obtained in 65 subjects, using ultrasound measurements as the gold standard. We took the absolute value of the predicted measurement minus the ultrasound measurement and compared this to the absolute value of the surgeon measurement minus the ultrasound measurement. Our results showed that the predicted measurements from the model, on average, were 0.10 cm (95% CI: 0.07, 0.13) different from the ultrasound measurement, while the surgeon measurement values, on average, were 0.15 cm (95% CI: 0.12, 0.19) different. This difference was statistically significant (p = 0.03). 3. Discussion It is estimated that up to 11,000 gastrostomy devices or other types of feeding devices are placed in children each year, and the number has steadily increased from 16.6 per 100,000 children in 1997 to 18.5 in 2009 [8,9]. Low-profile gastrostomy buttons are the preferred enteral access devices for young children as they can be hidden under clothes and increase comfort [1]. These devices are available in a variety of lengths and the chosen length should correspond to the child's AWT. Unlike adults, growth and weight gain in the pediatric population require more frequent gastrostomy device size adjustment, in order to accommodate changes in weight and body habitus [1]. A gastrostomy device that is too short or tight for a child can cause excessive pressure around the stoma tract, and the pressure from the external and/or
p-value b0.001 0.695 b0.001 b0.001 b0.001 0.001
Estimate
95% CI
p-value
−1.255 0.082
−1.634, −0.876 −0.048, 0.211
b0.001 0.221
0.012, 0.033
b0.001
0.022
internal bolsters can decrease blood flow to the area and cause impaired wound healing and cellulitis [10,11]. These devices can also cause skin necrosis, erosion and shortening of the stoma tract leading to increased leakage, and buried bumper syndrome, which can be life-threatening [12–14]. Similarly, a long gastrostomy device has been shown to significantly increase the risk of leakage [15], possibly owing to widening of the tract. In the same study, children with longer gastrostomy devices (≥1.5 cm) experienced an increased incidence of granuloma formation around the gastrostomy compared to those with shorter length (≤1.2 cm) (40% versus 29%), although the difference was not statistically significant [15]. In our experience, we have noticed that gastrostomy devices that are too long for the child tend to move around the stoma tract more freely and cause peristomal irritation. It is recommended that low-profile gastrostomy devices be replaced every three to six months, depending on the design and manufacturer of the device. As a child begins to gain weight following initial gastrostomy device placement, the length of the device shaft needs to be upsized at regular intervals, to accommodate growth in the abdominal wall. It is difficult, however, to know the growth in the AWT since the time of the initial gastrostomy, given the wide range in age groups, nutritional intake and duration of follow up in the pediatric population. Furthermore, most of these children receive medical care at primary care or small hospital settings, where the need to estimate a child's AWT may be an unfamiliar task. Thus, an easy and accurate method of estimating AWT would be beneficial in this scenario. A method of measuring the AWT at the time of initial placement is to pass a fine needle through the abdominal wall at the proposed site under laparoscopic visual guidance, then clamp the needle at skin level and measure the length. This is performed intraoperatively by some surgeons to gain a sense of the child's AWT. They then add 3–4 mm when determining the overall length of the low-profile gastrostomy device that will be placed, to account for thinning of the abdominal wall during insufflation, plus stomach wall thickness and a one-layer gauze dressing. This method is only applicable at the time of initial gastrostomy device placement and cannot be used when a child presents to have his or her gastrostomy device replaced and potentially upsized. A common method for measuring AWT when changing a gastrostomy device is to insert a plastic sizer in a preexisting
Table 3 Model 2: patients N2 years old (n = 42). Natural Log of Abdominal Wall Thickness (cm) as the outcome MULTIVARIABLE (R2 = 0.62)
UNIVARIATE Estimate Intercept Age (84–298 months) Gender (Male) Weight BMI Waist Circumference (cm) Xiphisternum to Pubis (cm)
0.440 −0.090 0.023 0.089 0.025 0.055
95% CI 0.148, 0.731 −0.393, 0.213 0.018, 0.028 0.066, 0.113 0.018, 0.031 0.032, 0.078
p-value 0.005 0.564 b0.001 b0.001 b0.001 b0.001
Estimate
95% CI
p-value
−1.335 0.271
−1.703, −0.967 0.071, 0.472
b0.001 0.011
0.060, 0.105
b0.001
0.082
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052
Y.M. Choi et al. / Journal of Pediatric Surgery xxx (xxxx) xxx–xxx
gastrostomy tract. Unfortunately, soft tissue thickening, skin indentation around the stoma site or erosion of the balloon into the abdominal wall is common with long-term gastrostomy devices and this method can lead to inaccurate measurements [5]. We used ultrasonography to obtain AWT measurements in our study. Ultrasonography has been successfully used to visualize and confirm proper replacement of gastrostomy devices [16,17], but its use as an adjunct for gastrostomy device size selection has not been described. Its ease of use and the ability to visualize soft tissue make it an ideal tool to measure the AWT for gastrostomy device size selection. Its limited availability in outpatient settings, however, makes its application unrealistic. Instead, we used ultrasonography to collect accurate AWT measurements from patients requiring placement of gastrostomy devices, which were, along with other patient-specific variables, used to develop a patient database to create models that can be used to determine the most appropriate gastrostomy device length in outpatient settings. Significant determinants of AWT in children have been previously reported in several studies [18–21]. Certain factors such as age and gender have been found to be associated with abdominal wall fat [18,20]. Our study, however, did not show gender to be a significant factor associated with AWT, when adjusting for other variables. Furthermore, most of our study population had varying degrees of malnutrition, which was not adjusted for, owing to a lack of nutritional data such as skinfold thickness or biochemical markers (e.g. albumin). Instead, we used body measurements as surrogate markers of nutritional status. The use of growth charts from the WHO and the CDC is recommended to assess the growth of children less than 2 years of age and children aged 2–19 years, respectively [2]. Z-scores (standard deviation scores) using a child's gender, age and weight or height are used to evaluate a child's nutritional status and growth. We had initially calculated and used weight-for-age z-scores in our model for estimating AWT, and it showed a statistically significant relationship. Nonetheless, our goal in developing this model was to be able to easily use it in a clinical setting, and calculation of z-scores in addition to the equation to predict the AWT would be a burden on clinicians and an unrealistic expectation. Therefore, we divided our study cohort into two groups, using the age of 2 years as a cutoff. This was done to avoid using BMI as a predictor of nutritional status in the younger group [2,22]. The final models (Model 1 and Model 2) to estimate AWT only require age and waist circumference for those ≤ 2 years of age, and age and BMI for those N 2 years of age. Not surprisingly, the model did not fit the data as well for the younger group compared to the older group (R2 = 0.42 vs. 0.62). This may be partly owing to potential measurement errors in the AWT, because their AWT tends to be very thin. Furthermore, our measurements did not account for variables such as fluid level or anasarca, which may increase AWT. The main limitation of this study is the small sample size, which limited our ability to perform external validation on an independent sample and to assess for relationships between comorbidities and AWT. The sample size was further reduced when the model was split into two age groups, but we decided this was important to ensure reliable measurements and clinical usability. We performed internal validation on the final models, which demonstrated that the model for younger patients performed slightly better than the model for older patients. Furthermore, 52% of our study population was aged 20 months or less. The skewed age distribution in our study reflects the true pediatric population that undergoes gastrostomy device placement. Nonetheless, data on older children would make the dataset more robust. A second limitation of this study is the fact that AWT was measured after general anesthesia was induced in some children. This may have led to variations in the degree of abdominal wall relaxation, falsely increasing the measured thickness in those who were awake and able to contract their abdominal wall muscles. Efforts were made, however, to measure AWT at the time of relaxation in those who were awake during the measurement. Another limitation is that these estimations may not be applicable for children with long-term gastrostomy devices that have caused
5
scarring or contracture around the gastrostomy sites. However, our overall goal is to avoid such severe scarring and soft tissue indentation by properly sizing the gastrostomy devices from the time of initial establishment of the gastrostomy tract and regularly upsizing the devices as the child continues to grow. The complexity of the models may limit their use in clinical settings, but simple web-based applications can be developed to perform the calculations. Lastly, we have not shown whether the implementation of these equations in determining the size of low-profile gastrostomy devices is effective at reducing the incidence of gastrostomy tract complications. Nonetheless, the aim of our current study was to develop a model to predict a child's AWT based on simple body measurements, and a large prospective trial will need to be undertaken to answer the question of its clinical effectiveness. 4. Conclusions In summary, it is of paramount importance to upsize gastrostomy devices at regular intervals for children who are gaining weight and growing. Selecting the appropriate gastrostomy device length can be a challenging task owing to difficulty estimating a child's AWT. Our mathematical models of AWT provide a useful tool using simple measurements, for those who do not have access to ultrasonography in an office, emergency department or small hospital setting. References [1] Buderus S, Adenaeuer M, Dueker G, et al. Balloon gastrostomy buttons in pediatric patients: evaluation with respect to size, lifetime in patients, and parent acceptance. Klin Padiatr 2009;221:65–8. [2] Grummer-Strawn LM, Reinold C, Krebs NF, et al. Use of World Health Organization and CDC growth charts for children aged 0–59 months in the United States. MMWR Recomm Rep 2010;59:1–15. [3] Jain N, Goyal N, Mukherjee K, et al. Ultrasound of the abdominal wall: what lies beneath? Clin Radiol 2013;68:85–93. [4] Bashir U, Moskovic E, Strauss D, et al. Soft-tissue masses in the abdominal wall. Clin Radiol 2014;69:e422–31. [5] Chang WK, Huang WC, Yu CY, et al. Long-term percutaneous endoscopic gastrostomy: characteristic computed tomographic findings. Abdom Imaging 2011;36:684–8. [6] Kuhn M, Johnson K. Applied predictive modeling. New York: Springer; 2013. [7] Willmott JC, Matsuura K. Advantages of the mean absolute error (MAE) over the root mean square error (RMSE) in assessing average model performance; 2005. [8] Goldberg E, Kaye R, Yaworski J, et al. Gastrostomy tubes: facts, fallacies, fistulas, and false tracts. Gastroenterol Nurs 2005;28:485–93 [quiz 93-4]. [9] Fox D, Campagna EJ, Friedlander J, et al. National trends and outcomes of pediatric gastrostomy tube placement. J Pediatr Gastroenterol Nutr 2014;59:582–8. [10] DeLegge M, DeLegge R, Brady C. External bolster placement after percutaneous endoscopic gastrostomy tube insertion: is looser better? JPEN J Parenter Enteral Nutr 2006;30:16–20. [11] McClave SA, Neff RL. Care and long-term maintenance of percutaneous endoscopic gastrostomy tubes. JPEN J Parenter Enteral Nutr 2006;30:S27–38. [12] Pattamanuch N, Novak I, Loizides A, et al. Single-center experience with 1-step lowprofile percutaneous endoscopic gastrostomy in children. J Pediatr Gastroenterol Nutr 2014;58:616–20. [13] Stewart CE, Mutalib M, Pradhan A, et al. Short article: buried bumper syndrome in children: incidence and risk factors. Eur J Gastroenterol Hepatol 2017;29:181–4. [14] Segal D, Michaud L, Guimber D, et al. Late-onset complications of percutaneous endoscopic gastrostomy in children. J Pediatr Gastroenterol Nutr 2001;33:495–500. [15] Helen S, Torsten Larsson L, Arnbjörnsson E. Postoperative gastrostomy site leakage correlated to the dimension of the gastrostomy button in children. Gastroenterol Insights 2010;2:34–6. [16] Wu TS, Leech SJ, Rosenberg M, et al. Ultrasound can accurately guide gastrostomy tube replacement and confirm proper tube placement at the bedside. J Emerg Med 2009;36:280–4. [17] Marin JR, Lewiss RE. Point-of-care ultrasonography by pediatric emergency medicine physicians. Pediatr Emerg Care 2015;31:525. [18] Brei C, Much D, Heimberg E, et al. Sonographic assessment of abdominal fat distribution during the first year of infancy. Pediatr Res 2015;78:342–50. [19] Balaguru D, Bhalala U, Haghighi M, et al. Computed tomography scan measurement of abdominal wall thickness for application of near-infrared spectroscopy probes to monitor regional oxygen saturation index of gastrointestinal and renal circulations in children. Pediatr Crit Care Med 2011;12:e145–8. [20] Butler M, Servaes S, Srinivasan A, et al. US depiction of the appendix: role of abdominal wall thickness and appendiceal location. Emerg Radiol 2011;18:525–31. [21] Sandler G, Leishman S, Branson H, et al. Body wall thickness in adults and children— relevance to penetrating trauma. Injury 2010;41:506–9. [22] Nash A, Secker D, Corey M, et al. Field testing of the 2006 World Health Organization growth charts from birth to 2 years: assessment of hospital undernutrition and overnutrition rates and the usefulness of BMI. JPEN J Parenter Enteral Nutr 2008;32:145–53.
Please cite this article as: Choi YM, et al, Model to estimate abdominal wall thickness in children undergoing placement or replacement of gastrostomy devices, J Pediatr Surg (2018), https://doi.org/10.1016/j.jpedsurg.2018.08.052