- Email: [email protected]
A risk calculator predicting postoperative adverse events in neonates undergoing major abdominal or thoracic surgery
Journal of Pediatric Surgery 50 (2015) 987–991
Contents lists available at ScienceDirect
Journal of Pediatric Surgery journal homepage: www.elsevier.com/locate/jpedsurg
A risk calculator predicting postoperative adverse events in neonates undergoing major abdominal or thoracic surgery Anne M. Stey a,b,⁎, Brian D. Kenney c, R. Lawrence Moss c, Bruce L. Hall d,e, Loren Berman f, Mark E. Cohen d, Kari Kraemer d, Clifford Y. Ko b,d, Charles D. Vinocur f a
Icahn School of Medicine at Mount Sinai Medical Center, New York, NY, USA David Geffen School of Medicine, University of California Los Angeles, Los Angeles, CA, USA Nationwide Children's Hospital, The Ohio State University, Columbus, OH, USA d American College of Surgeons, Chicago, IL, USA e Washington University School of Medicine in Saint Louis, Department of Surgery, Olin Business School, and Center for Health Policy, St Louis VA Medical Center, BJC Healthcare, Saint Louis, MO, USA f Nemours/Alfred I. duPont Hospital for Children, Jefferson Medical College, Wilmington, DE, USA b c
a r t i c l e
i n f o
Article history: Received 6 March 2015 Accepted 10 March 2015 Key words: Risk calculator Neonatal surgery
a b s t r a c t Purpose: This study sought to demonstrate the feasibility of a risk calculator for neonates undergoing major abdominal or thoracic surgery with good discriminative ability. Methods: The American College of Surgeons' National Surgical Quality Improvement Program Pediatric (ACS-NSQIP-P) 2011–12 data were queried for neonates who underwent major abdominal or thoracic surgery. The outcome of interest was the occurrence of any adverse event, including mortality, within 30-days postoperatively. The preoperative clinical characteristics significantly associated with any adverse event were used to build a multivariate model. The model's discriminative ability was assessed with the area under the receiver operating characteristic curve (AUROC). The model was split-set validated with 2013 data. Results: A total of 2967 neonates undergoing major abdominal or thoracic surgery were identified. The overall rate of adverse events was 23.3%. Sixteen variables were found to be associated with adverse events. Four variables increased the odds of adverse events at least two-fold: dirty or infected wound class [odds ratio (OR) = 2.1] dialysis (OR = 3.8), hepatobiliary disease (OR = 2.1), and inotropic agent use (OR = 2.6). The AUROC = 0.79 for development data and 0.77 on splitset validation. Conclusion: Preoperatively estimating the probability of postoperative adverse events in neonates undergoing major abdominal or thoracic surgery with good discrimination is feasible. © 2015 Elsevier Inc. All rights reserved.
Despite improvement in neonatal surgical care [1], surgical procedures in neonates are often associated with higher rates of postoperative adverse events compared to other children [2]. Neonates undergoing surgery have heterogeneous clinical and comorbidity profiles, resulting in different risks of postoperative adverse events [3,4]. Many families in the stressful weeks after their infant's birth wish for, and could benefit from, more quantitative information regarding postoperative prognosis [5,6]. This study sought to create a risk calculator model that could help preoperatively estimate patient-specific probability of postoperative adverse events in neonates undergoing major abdominal or thoracic surgery. The
specific aim was to use the American College of Surgeons' National Surgical Quality Improvement Program Pediatric (ACS-NSQIP-P) preoperative clinical and procedural data to estimate probability of postoperative adverse events with good discriminative ability. The ACS-NSQIP has built a risk calculator model for adults that can be used by physicians and patients to better estimate patient-specific probability of postoperative adverse events preoperatively [7]. The purpose of this study was to determine the feasibility of developing a similar tool in neonatal surgery.
1. Materials and methods 1.1. Data source and patient sample
Abbreviations: ACS-NSQIP-P, American College of Surgeons National Surgical Quality Improvement Program Pediatric; CPT, Current Procedure Terminology; ASA, American Society of Anesthesiologists; AUROC, Area Under the Receiver Operating Characteristic. ⁎ Corresponding author at: David Geffen School of Medicine, 10940 Wilshire Blvd, Suite 710, Los Angeles, CA 90024. Tel.: +1 310 794 2507; fax: +1 310 794 3288. E-mail address: [email protected] (A.M. Stey). http://dx.doi.org/10.1016/j.jpedsurg.2015.03.023 0022-3468/© 2015 Elsevier Inc. All rights reserved.
This observational study used 2011–2013 clinical registry data from the ACS-NSQIP-P. The setting consisted of 50 participating ACS-NSQIP-P hospitals across North America. The dataset included demographic, preoperative clinical, and procedural variables as well as postoperative adverse events in the 30 days following surgery, as described elsewhere
988
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
Table 1 Preoperative clinical characteristics of study sample of neonates undergoing major abdominal or thoracic surgery (total N = 2967). Demographic
Count
%
Male Race Caucasian African American Asian Native American Other Hispanic ethnicity Chronological age ≤28 days 29−365 days Gestational age Less than 24 completed weeks gestation 24 completed weeks gestation 25–26 completed weeks gestation 27–28 completed weeks gestation 29–30 completed weeks gestation 31–32 completed weeks gestation 33–34 completed weeks gestation 35–36 completed weeks gestation 37 or greater completed weeks of gestation Procedural variables Major abdominal Major thoracic Case Elective Emergent Urgent Inpatient Wound class Clean Clean-contaminated Contaminated Dirty/infected American Society of Anesthesiologists ASA class I ASA class II ASA class III ASA class IV ASA class V Metabolic and nutritional conditions Preoperative nutritional support Renal conditions Preoperative dialysis Cardiac conditions Cardiac risk factors No cardiac risk factors Minor risk factors Major risk factors Severe risk factors Cardiac surgery Alimentary tract conditions Esophageal gastrointestinal disease Hepatobiliary pancreatic disease Pulmonary conditions Chronic lung disease Neurologic conditions APGAR at 5 minutes APGAR score 1 APGAR score 2 APGAR score 3 APGAR score 4 APGAR score 5 APGAR score 6 APGAR score 7 APGAR score 8 APGAR score 9 Hematologic and oncologic conditions Bleeding disease Hematologic disorder Preoperative transfusion Acuity of illness Inotropic support Sepsis preoperatively No sepsis
1725
58.1
2398 489 54 21 5 347
80.8 16.5 1.8 0.7 0.2 11.7
2270 697
76.5 23.5
50 87 198 177 168 165 278 477 1365
1.7 2.9 6.7 6.0 5.7 5.6 9.4 16.0 46.0
2581 386
87.0 13.0
1381 933 653 2929
46.5 31.5 22.0 98.7
612 1811 258 286
20.6 61.1 8.7 9.6
58 502 1543 804 60
2.0 16.9 52.0 27.1 2.0
1563
52.7
27
0.9
Table 1 (continued) Demographic Systemic inflammatory response syndrome Sepsis Septic shock
Count 33 100 55
% 1.1 3.3 1.9
[8,9]. Inclusion criteria were the following: children accrued in ACSNSQIP-P that were designated as neonates (0–28 days old corrected for gestational age) undergoing major abdominal or thoracic surgery as specified by the current procedural terminology (CPT) codes in the ‘principal procedure’ data field (Appendix A). 1.2. Measures The primary outcome of interest was a composite 30-day adverse event variable. This composite variable was generated for two reasons. First, the rate of any single adverse event in this population was too low to model accurately in isolation. Second, the 30-day mortality rate was not common in this group of infants therefore mortality was added to the composite 30-day morbidity rate commonly used by ACS-NSQIP-P in performance benchmarking. The resulting variable was a composite dichotomous outcome variable of whether any one or more complications or mortality occurred within 30 days of the neonatal abdominal or thoracic surgery. The complications included in addition to mortality were: surgical site infection, pneumonia, reintubation, pulmonary embolism, renal insufficiency, urinary tract infection, coma, seizure, peripheral nerve injury, intraventricular hemorrhage, intracranial hemorrhage, cardiac arrest, intraoperative or postoperative transfusion, graft failure, venous thrombosis requiring therapy, sepsis, central line associated blood stream infection. The presence of complications was assessed at 30 days from the operation as defined within the ACSNSQIP-P. The predictors of interest were patient preoperative demographic and clinical variables as defined and collected by ACS-NSQIP-P. CPT codes were grouped into clinically similar categories and then converted into a linear risk variable. Case-mix was adjusted in the riskcalculator model using this previously described CPT linear risk approach (using CPT categories) [10]. 1.3. Statistical analysis
1665 820 350 132 204
56.1 27.6 11.8 4.5 6.9
2138 167
72.1 5.6
476
16.0
42 74 81 143 230 326 553 1430 88
1.4 2.5 2.7 4.8 7.8 11.0 18.6 48.2 3.0
97 449 327
3.3 15.1 11.0
224
7.6
2779
93.7
Count and frequency of preoperative demographic and clinical variables in neonates who underwent abdominal or thoracic surgery were calculated. Next, the preoperative clinical characteristics significantly associated (p b 0.05) with the composite 30-day postoperative adverse event outcome on initial bivariate logistic regression were included in the multivariate hierarchical logistic model. Stepwise selection was used to select the variables most statistically associated with the composite outcome. Parameter values from the final equation were used as the components of the risk calculator model to estimate the predicted probability of “any” postoperative adverse event for each individual infant. The predicted probability was output from the multivariate hierarchical model using only (patient-level) fixed effects. The model's discriminative ability was assessed by calculating the area under the receiver operating characteristic (AUROC) curve (the plot of sensitivity versus 1-specificity), also known as the c-statistic. Calibration was assessed with the Hosmer– Lemeshow chi-square and associated p-value. The Brier score was also calculated to assess discrimination and calibration. The Brier score is interpreted such that the lower the score the better the predictions' calibration. Finally, data from 2013 were used to split-set validate the model. The RAND CORPORATION institutional review board approved this study. All data management and analyses were performed in SAS 9.3 (SAS Institute, Cary NC).
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
2. Results A total of 2967 eligible neonates were identified as having undergone abdominal or thoracic surgery in the dataset. This was out of 6493 total cases entered in the database during the same time period, thus 45.7% of neonate cases in the original dataset. In total 76.5% of neonates who underwent abdominal or thoracic surgery had a chronological age of 28 days or less at the time of surgery (Table 1). The majority, 52.0%, of neonates were designated as American Society of Anesthesiologists (ASA) class of III. Major abdominal cases accounted for 87.0% of neonatal cases. Thirty-two percent of all neonates underwent emergent operation. 2.1. Postoperative adverse events were common in neonates The unadjusted rates of adverse events were high in neonates undergoing major abdominal or thoracic surgery—23.3% for “any” postoperative adverse event (Table 2). The most frequent of the individual adverse events was intraoperative/postoperative transfusion occurring in 8.5% of infants. The second most frequent adverse event was mortality; 4.4%. Reintubation was next, 3.9%, then surgical site infection, 3.7%, followed by postoperative sepsis, 2.7%. 2.2. Preoperative variables were highly associated with postoperative adverse events Preoperative variables highly associated with the occurrence of any postoperative adverse event in this population were identified (Table 3). A total of 16 variables were significantly associated with the occurrence of any adverse events. These 16 preoperative variables were: gestational age, primary procedure CPT category linear risk, case type, wound classification, ASA class, preoperative nutritional support, dialysis, cardiac risk factors, esophagogastrointestinal disease, hepatobiliary disease, APGAR score at 5 minutes, bleeding disease, hematologic disorder, transfusion, inotropic agent use, and sepsis (Table 3). Fifteen out of sixteen of these variables significantly increased risk, whereas the effect of gestational age depended on the category. These 16 variables became the components of the risk calculator model. 2.3. Model discriminative ability The discriminative ability of the model described in Table 3 was evaluated. The AUROC curve was 0.79 (95% CI = 0.78–0.82; Fig. 1). When this same model was split-set validated against the 2013 data of 1879 neonates undergoing major abdominal or thoracic surgery, the AUROC curve was 0.77 (95% CI = 0.74–0.79). Thus discrimination was qualified as good. The model calibration with the validation dataset was also good; the Hosmer–Lemeshow chi-square was 9.5, p = 0.31. The Brier score was 0.13 suggesting the predictions were well calibrated. 3. Discussion Preoperative patient-specific estimates of the probability of postoperative adverse events could better inform physicians and families of neonates of surgical risk. Such a data-driven quantification of surgical risk may also help set realistic postoperative expectations [11–13]. Estimation of risk can also enlighten preoperative preparations (of patient or caregivers) and inform postoperative resource or care planning. This study sought to create a risk calculator model that could help preoperatively estimate patient-specific probability of postoperative adverse event in neonates undergoing major abdominal or major thoracic surgery. It found that it is possible to preoperatively estimate patient-specific probability of postoperative adverse events in neonates undergoing major abdominal or thoracic surgery with good discriminative ability.
989
Table 2 Adverse events in neonates undergoing major abdominal or thoracic surgery. Adverse eventa
Count (n)
%
Composite “any” adverse event Transfusion Mortality Reintubation Surgical site infection Sepsis Wound dehiscence Pneumonia Urinary tract infection Cerebrovascular accident Catheter associated blood stream infection Cardiac arrest Renal failure
692 252 129 116 109 80 52 40 35 31 29 28 25
23.3 8.5 4.4 3.9 3.7 2.7 1.8 1.4 1.2 1.0 1.0 0.9 0.8
a
Includes all individual complications that occurred in 1% of greater of neonates.
Preoperative variables associated with two-fold or more increased odds of adverse events were: a dirty/infected case, preoperative requirement of dialysis, hepatobiliary disease, and inotropic agent use. The clinical variables found to be associated with adverse events are similar to previous studies that have sought to predict risk of adverse events in neonates. The literature has described the impact of prematurity, APGAR scores, cardiac abnormalities and preoperative sepsis on postoperative outcomes in neonates [2,3,14]. Other studies that have looked at predictors of outcomes in neonates have often favored objective measures such as birth weight, vital signs and laboratory parameters [15–17]. Some of these parameters were available in the current study but were not used since they fluctuate greatly over time even in a single patient. Incorporation of procedural factors was an important addition in the model of neonatal adverse events presented herein. This was accomplished using the risk associated with primary procedure CPT category as an element of modeling. These procedural variables allow for the calculator model to be used to predict postoperative adverse event rates across two major system types of operations. This risk calculator model is a means of giving a patient-specific data-derived probability of postoperative adverse events to parents preoperatively. Although clinical experience gives insight as to whom is more at risk for poor outcomes, the vast majority of physicians and surgeons cannot presently give parents their child's individualized, dataderived probability of postoperative complications preoperatively. This calculator model offers a means of estimating an infant's probability of postoperative adverse events preoperatively. Model refinements are planned before making this calculator model more widely available for use. First, it is critical to refine how diagnoses of congenital disorders are included. Second, these data should be prospectively validated to demonstrate that they have value in predicting the occurrence of adverse events. This study has several limitations. The first is the potential bias in the patient sample of neonates at the self-selected, typically large academic tertiary care hospitals that participate in ACS-NSQIP-P. As a result, the findings may not be generalizable to a wider sample of pediatric surgical patients in non-ACS-NSQIP-P hospitals. A second limitation is that longterm outcomes were not included. In pediatric surgery, most important functional outcomes can only be obtained after 30 days. These data are not available in ACS-NSQIP-P in its current form. Third, procedurespecific complications are not captured. Procedure-specific complications in neonates, such as a leak or stricture after an esophageal atresia repair, are often more common than generic complications such as pneumonia, and therefore in the future might be easier to model. Nonetheless, potentially avoidable general postoperative adverse events were used as primary outcome in this analysis in order to define undesirable surgical outcomes across a wide range of procedures. These general adverse events may be less sensitive than procedure-specific
990
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
Table 3 Components of risk calculator for adverse events in neonates undergoing major abdominal or thoracic surgery. Preoperative variables
Preterm Not preterm 24 completed weeks gestation 25–26 completed weeks gestation 27–28 completed weeks gestation 29–30 completed weeks gestation 31–32 completed weeks gestation 33–34 completed weeks gestation 35–36 completed weeks gestation Procedural variables CPT category linear risk Case typea Elective Urgent Emergent Wound classificationa Clean Clean-contaminated Contaminated Dirty/infected American Society of Anesthesiologistsa ASA class I ASA class II ASA class III ASA class IV ASA class V Metabolic and nutritional conditions Preoperative nutritional support a Renal conditions Dialysisa Cardiac conditions Cardiac risk factorsa No cardiac risk factors Minor risk factors Major risk factors Severe risk factors Alimentary tract conditions Esophago-gastrointestinal disease Hepatobiliary diseasea Neurologic condition APGAR score at 5 minutesa APGAR score 1 APGAR score 2 APGAR score 3 APGAR score 4 APGAR score 5 APGAR score 6 APGAR score 7 APGAR score 8 APGAR score 9 Hematologic and oncologic conditions Bleeding diseasea Hematologic disordersa Preoperative transfusiona Acuity of illness Inotropic agent usea Sepsis postoperatively No sepsis Systemic inflammatory response syndrome Sepsis Septic shock
Estimate
Lower Upper p 95% 95% CI CI
Reference 1.10 1.48 1.00 1.31 0.77 1.15 0.78
– 0.65 0.95 0.62 0.83 0.47 0.78 0.56
– 1.87 2.32 1.62 2.08 1.27 1.68 1.09
– 0.72 0.08 0.99 0.25 0.30 0.48 0.15
1.48
1.27
1.74
b0.0001
Reference – 1.61 1.23 0.81 0.60
– 2.11 1.10
– 0.001 0.17
Reference 1.37 1.70 2.06
– 1.02 1.12 1.35
– 1.85 2.60 3.15
– 0.04 0.01 0.001
Reference 0.44 0.78 1.16 1.00
– 0.17 0.32 0.47 0.33
– 1.12 1.88 2.86 3.08
– 0.09 0.57 0.75 1.00
1.60
1.26
2.02
0.0002
3.77
1.39
10.22
0.01
Reference 1.26 1.55 1.51
– 0.99 1.14 0.93
– 1.61 2.12 2.44
– 0.06 0.01 0.10
1.13 2.05
0.87 1.37
1.46 3.08
0.37 0.001
1.42 0.98 0.67 1.87 0.85 1.28 1.23 0.77 Reference
0.77 0.45 0.35 1.04 0.52 0.88 0.88 0.58 –
2.64 2.11 1.28 3.36 1.37 1.86 1.72 1.03 –
0.26 0.95 0.23 0.04 0.49 0.20 0.22 0.08 –
1.85 1.55 1.52
1.06 1.16 1.10
3.21 2.06 2.09
0.03 0.004 0.01
2.61
1.75
3.91
b0.0001
Reference – 1.03 0.44
– 2.41
– 0.95
1.72 1.99
2.87 4.63
0.04 0.11
1.03 0.86
a All categorical variables denoted had an overall significant (p b 0.05) association with adverse events despite discrete levels within the categorical variables not being significantly different from the reference level. Bold items significant at p b 0.05.
Fig. 1. The area under the receiver operating characteristic curve of the risk calculator model. Each curved line represents the discriminative ability of the model with each dataset, development and validation.
procedures in pediatric surgery would be most desirable, owing to the low event rate and relatively low number of patients captured, this is not yet possible. Fifth, this calculator model does not calculate whether the risks of the procedure outweigh the benefits, but it does provide insight into such a deliberation. Finally, this risk calculator model does not predict specific adverse events rates. Once a larger sample size of neonates is available, developing a more outcome-specific calculator will make it more useful to clinicians, patients and families.
4. Conclusion This study demonstrated the feasibility of constructing a risk calculator model for preoperatively estimating patient-specific probability of postoperative adverse events for neonates undergoing major abdominal or thoracic surgery with good discriminative ability. Sixteen variables were associated with postoperative adverse events. Four preoperative clinical variables (a dirty/infected case, preoperative requirement of dialysis, hepatobiliary disease, and inotropic agent use) were associated with a greater than two-fold increased odds of postoperative adverse events. This model could be used during the preoperative counseling process to help families anticipate realistic outcomes. It could also inform preoperative preparation of the patient or caregiver team, and could shed light on potential resource needs for postoperative acute and post-acute care.
Acknowledgment outcomes and may not give a full picture of long-term functional outcomes critical in children's surgery. However, general postoperative adverse events are specific metrics of perioperative quality of care. Fourth, although a more comprehensive calculator model including all
AMS's time was supported for this publication by The Robert Wood Johnson Foundation Clinical Scholars® program and the U.S. Department of Veterans Affairs.
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
Appendix A. Major abdominal and thoracic cases accrued in ACS-NSQIP-P CPT
PROCEDURE DESCRIPTION
22900 37181-38120 38562-38780 39502 43280-43326 43520-43840 44005 44010-44025 44050 44055 44110-46748 47100-47130 47480-47620 47700-47800 48120-48146 35840, 49000-49322 49324-49421 49600-49611 50205 50220-50548 51500 51530-51585 57106-57112 31786-31805 32095-32160 32220-32402 32440-32501 32601-32665 35271-38746 39010-39400 39503-39561 43101-43352 19260
Abdominal Wall Tumor Excision Splenic Procedures Lymphadenectomy Paraesophageal Hernia Repair Fundoplication Gastric Procedures Lysis of Adhesions Enterotomy Reduction of Volvulus, Intussusception or Internal Hernia Ladd Procedure Bowel Resection and Ostomy Liver Resection Cholecystectomy Biliary Procedures Pancreatic Resection Abdominal Exploration) Peritoneal Dialysis Catheter Placement Omphalocele Repair Renal Biopsy Nephrectomy Urachal Cyst Excision Cystostomy and Cystectomy Vaginectomy Intrathoracic Tracheal Procedures Thoracotomy Procedures Decortication Lung Resection Thoracoscopic Procedures, Thoracic Lymphovascular Procedures Mediastinal Procedures Diaphragmatic Procedures Esophageal procedures Chest Wall Tumor Excision
991
References [1] Rowe MI, Rowe SA. The last fifty years of neonatal surgical management. Am J Surg 2000;180:345–52. [2] Lillehei CW, Gauvreau K, Jenkins KJ. Risk adjustment for neonatal surgery: a method for comparison of in-hospital mortality. Pediatrics 2012;130:e568–74. [3] Son JK, Lillehei CW, Gauvreau K, et al. A risk adjustment method for newborns undergoing noncardiac surgery. Ann Surg 2010;251:754–8. [4] Skarsgard ED, Blair GK, Lee SK. Toward evidence-based best practices in neonatal surgical care-I: the Canadian NICU Network. J Pediatr Surg 2003;38:672–7. [5] Perlman NB, Freedman JL, Abramovitch R, et al. Informational needs of parents of sick neonates. Pediatrics 1991;88:512–8. [6] Aite L, Zaccara A, Trucchi A, et al. Parents' informational needs at the birth of a baby with a surgically correctable anomaly. Pediatr Surg Int 2006;22:267–70. [7] Bilimoria KY, Liu Y, Paruch JL, et al. Development and evaluation of the universal ACS NSQIP surgical risk calculator: a decision aid and informed consent tool for patients and surgeons. J Am Coll Surg 2013;217:833–842.e831-833. [8] Raval MV, Dillon PW, Bruny JL, et al. American College of Surgeons National Surgical Quality Improvement Program Pediatric: a phase 1 report. J Am Coll Surg 2011;212: 1–11. [9] Bruny JL, Hall BL, Barnhart DC, et al. American College of Surgeons National Surgical Quality Improvement Program Pediatric: a beta phase report. J Pediatr Surg 2013; 48:74–80. [10] Cohen ME, Ko CY, Bilimoria KY, et al. Optimizing ACS NSQIP modeling for evaluation of surgical quality and risk: patient risk adjustment, procedure mix adjustment, shrinkage adjustment, and surgical focus. J Am Coll Surg 2013;217:336–346.e331. [11] Schwarze ML, Sayla MA, Alexander GC. A comparison of patient and physician beliefs about infrainguinal bypass operation: what role should surgical optimism play? Surgery 2007;141:239–44. [12] Lattig F, Fekete TF, O'Riordan D, et al. A comparison of patient and surgeon preoperative expectations of spinal surgery. Spine (Phila Pa 1976) 2013;38(12): 1040–8. [13] Jourdan C, Poiraudeau S, Descamps S, et al. Comparison of patient and surgeon expectations of total hip arthroplasty. PLoS One 2012;7:e30195. [14] Johansson S, Montgomery SM, Ekbom A, et al. Preterm delivery, level of care, and infant death in Sweden: a population-based study. Pediatrics 2004;113:1230–5. [15] Sacco Casamassima MG, Salazar JH, Papandria D, et al. Use of risk stratification indices to predict mortality in critically ill children. Eur J Pediatr 2014;173:1–13. [16] Richardson DK, Gray JE, McCormick MC, et al. Score for neonatal acute physiology: a physiologic severity index for neonatal intensive care. Pediatrics 1993;91:617–23. [17] Skarsgard ED, MacNab YC, Qiu Z, et al. SNAP-II predicts mortality among infants with congenital diaphragmatic hernia. J Perinatol 2005;25:315–9.
Contents lists available at ScienceDirect
Journal of Pediatric Surgery journal homepage: www.elsevier.com/locate/jpedsurg
A risk calculator predicting postoperative adverse events in neonates undergoing major abdominal or thoracic surgery Anne M. Stey a,b,⁎, Brian D. Kenney c, R. Lawrence Moss c, Bruce L. Hall d,e, Loren Berman f, Mark E. Cohen d, Kari Kraemer d, Clifford Y. Ko b,d, Charles D. Vinocur f a
Icahn School of Medicine at Mount Sinai Medical Center, New York, NY, USA David Geffen School of Medicine, University of California Los Angeles, Los Angeles, CA, USA Nationwide Children's Hospital, The Ohio State University, Columbus, OH, USA d American College of Surgeons, Chicago, IL, USA e Washington University School of Medicine in Saint Louis, Department of Surgery, Olin Business School, and Center for Health Policy, St Louis VA Medical Center, BJC Healthcare, Saint Louis, MO, USA f Nemours/Alfred I. duPont Hospital for Children, Jefferson Medical College, Wilmington, DE, USA b c
a r t i c l e
i n f o
Article history: Received 6 March 2015 Accepted 10 March 2015 Key words: Risk calculator Neonatal surgery
a b s t r a c t Purpose: This study sought to demonstrate the feasibility of a risk calculator for neonates undergoing major abdominal or thoracic surgery with good discriminative ability. Methods: The American College of Surgeons' National Surgical Quality Improvement Program Pediatric (ACS-NSQIP-P) 2011–12 data were queried for neonates who underwent major abdominal or thoracic surgery. The outcome of interest was the occurrence of any adverse event, including mortality, within 30-days postoperatively. The preoperative clinical characteristics significantly associated with any adverse event were used to build a multivariate model. The model's discriminative ability was assessed with the area under the receiver operating characteristic curve (AUROC). The model was split-set validated with 2013 data. Results: A total of 2967 neonates undergoing major abdominal or thoracic surgery were identified. The overall rate of adverse events was 23.3%. Sixteen variables were found to be associated with adverse events. Four variables increased the odds of adverse events at least two-fold: dirty or infected wound class [odds ratio (OR) = 2.1] dialysis (OR = 3.8), hepatobiliary disease (OR = 2.1), and inotropic agent use (OR = 2.6). The AUROC = 0.79 for development data and 0.77 on splitset validation. Conclusion: Preoperatively estimating the probability of postoperative adverse events in neonates undergoing major abdominal or thoracic surgery with good discrimination is feasible. © 2015 Elsevier Inc. All rights reserved.
Despite improvement in neonatal surgical care [1], surgical procedures in neonates are often associated with higher rates of postoperative adverse events compared to other children [2]. Neonates undergoing surgery have heterogeneous clinical and comorbidity profiles, resulting in different risks of postoperative adverse events [3,4]. Many families in the stressful weeks after their infant's birth wish for, and could benefit from, more quantitative information regarding postoperative prognosis [5,6]. This study sought to create a risk calculator model that could help preoperatively estimate patient-specific probability of postoperative adverse events in neonates undergoing major abdominal or thoracic surgery. The
specific aim was to use the American College of Surgeons' National Surgical Quality Improvement Program Pediatric (ACS-NSQIP-P) preoperative clinical and procedural data to estimate probability of postoperative adverse events with good discriminative ability. The ACS-NSQIP has built a risk calculator model for adults that can be used by physicians and patients to better estimate patient-specific probability of postoperative adverse events preoperatively [7]. The purpose of this study was to determine the feasibility of developing a similar tool in neonatal surgery.
1. Materials and methods 1.1. Data source and patient sample
Abbreviations: ACS-NSQIP-P, American College of Surgeons National Surgical Quality Improvement Program Pediatric; CPT, Current Procedure Terminology; ASA, American Society of Anesthesiologists; AUROC, Area Under the Receiver Operating Characteristic. ⁎ Corresponding author at: David Geffen School of Medicine, 10940 Wilshire Blvd, Suite 710, Los Angeles, CA 90024. Tel.: +1 310 794 2507; fax: +1 310 794 3288. E-mail address: [email protected] (A.M. Stey). http://dx.doi.org/10.1016/j.jpedsurg.2015.03.023 0022-3468/© 2015 Elsevier Inc. All rights reserved.
This observational study used 2011–2013 clinical registry data from the ACS-NSQIP-P. The setting consisted of 50 participating ACS-NSQIP-P hospitals across North America. The dataset included demographic, preoperative clinical, and procedural variables as well as postoperative adverse events in the 30 days following surgery, as described elsewhere
988
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
Table 1 Preoperative clinical characteristics of study sample of neonates undergoing major abdominal or thoracic surgery (total N = 2967). Demographic
Count
%
Male Race Caucasian African American Asian Native American Other Hispanic ethnicity Chronological age ≤28 days 29−365 days Gestational age Less than 24 completed weeks gestation 24 completed weeks gestation 25–26 completed weeks gestation 27–28 completed weeks gestation 29–30 completed weeks gestation 31–32 completed weeks gestation 33–34 completed weeks gestation 35–36 completed weeks gestation 37 or greater completed weeks of gestation Procedural variables Major abdominal Major thoracic Case Elective Emergent Urgent Inpatient Wound class Clean Clean-contaminated Contaminated Dirty/infected American Society of Anesthesiologists ASA class I ASA class II ASA class III ASA class IV ASA class V Metabolic and nutritional conditions Preoperative nutritional support Renal conditions Preoperative dialysis Cardiac conditions Cardiac risk factors No cardiac risk factors Minor risk factors Major risk factors Severe risk factors Cardiac surgery Alimentary tract conditions Esophageal gastrointestinal disease Hepatobiliary pancreatic disease Pulmonary conditions Chronic lung disease Neurologic conditions APGAR at 5 minutes APGAR score 1 APGAR score 2 APGAR score 3 APGAR score 4 APGAR score 5 APGAR score 6 APGAR score 7 APGAR score 8 APGAR score 9 Hematologic and oncologic conditions Bleeding disease Hematologic disorder Preoperative transfusion Acuity of illness Inotropic support Sepsis preoperatively No sepsis
1725
58.1
2398 489 54 21 5 347
80.8 16.5 1.8 0.7 0.2 11.7
2270 697
76.5 23.5
50 87 198 177 168 165 278 477 1365
1.7 2.9 6.7 6.0 5.7 5.6 9.4 16.0 46.0
2581 386
87.0 13.0
1381 933 653 2929
46.5 31.5 22.0 98.7
612 1811 258 286
20.6 61.1 8.7 9.6
58 502 1543 804 60
2.0 16.9 52.0 27.1 2.0
1563
52.7
27
0.9
Table 1 (continued) Demographic Systemic inflammatory response syndrome Sepsis Septic shock
Count 33 100 55
% 1.1 3.3 1.9
[8,9]. Inclusion criteria were the following: children accrued in ACSNSQIP-P that were designated as neonates (0–28 days old corrected for gestational age) undergoing major abdominal or thoracic surgery as specified by the current procedural terminology (CPT) codes in the ‘principal procedure’ data field (Appendix A). 1.2. Measures The primary outcome of interest was a composite 30-day adverse event variable. This composite variable was generated for two reasons. First, the rate of any single adverse event in this population was too low to model accurately in isolation. Second, the 30-day mortality rate was not common in this group of infants therefore mortality was added to the composite 30-day morbidity rate commonly used by ACS-NSQIP-P in performance benchmarking. The resulting variable was a composite dichotomous outcome variable of whether any one or more complications or mortality occurred within 30 days of the neonatal abdominal or thoracic surgery. The complications included in addition to mortality were: surgical site infection, pneumonia, reintubation, pulmonary embolism, renal insufficiency, urinary tract infection, coma, seizure, peripheral nerve injury, intraventricular hemorrhage, intracranial hemorrhage, cardiac arrest, intraoperative or postoperative transfusion, graft failure, venous thrombosis requiring therapy, sepsis, central line associated blood stream infection. The presence of complications was assessed at 30 days from the operation as defined within the ACSNSQIP-P. The predictors of interest were patient preoperative demographic and clinical variables as defined and collected by ACS-NSQIP-P. CPT codes were grouped into clinically similar categories and then converted into a linear risk variable. Case-mix was adjusted in the riskcalculator model using this previously described CPT linear risk approach (using CPT categories) [10]. 1.3. Statistical analysis
1665 820 350 132 204
56.1 27.6 11.8 4.5 6.9
2138 167
72.1 5.6
476
16.0
42 74 81 143 230 326 553 1430 88
1.4 2.5 2.7 4.8 7.8 11.0 18.6 48.2 3.0
97 449 327
3.3 15.1 11.0
224
7.6
2779
93.7
Count and frequency of preoperative demographic and clinical variables in neonates who underwent abdominal or thoracic surgery were calculated. Next, the preoperative clinical characteristics significantly associated (p b 0.05) with the composite 30-day postoperative adverse event outcome on initial bivariate logistic regression were included in the multivariate hierarchical logistic model. Stepwise selection was used to select the variables most statistically associated with the composite outcome. Parameter values from the final equation were used as the components of the risk calculator model to estimate the predicted probability of “any” postoperative adverse event for each individual infant. The predicted probability was output from the multivariate hierarchical model using only (patient-level) fixed effects. The model's discriminative ability was assessed by calculating the area under the receiver operating characteristic (AUROC) curve (the plot of sensitivity versus 1-specificity), also known as the c-statistic. Calibration was assessed with the Hosmer– Lemeshow chi-square and associated p-value. The Brier score was also calculated to assess discrimination and calibration. The Brier score is interpreted such that the lower the score the better the predictions' calibration. Finally, data from 2013 were used to split-set validate the model. The RAND CORPORATION institutional review board approved this study. All data management and analyses were performed in SAS 9.3 (SAS Institute, Cary NC).
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
2. Results A total of 2967 eligible neonates were identified as having undergone abdominal or thoracic surgery in the dataset. This was out of 6493 total cases entered in the database during the same time period, thus 45.7% of neonate cases in the original dataset. In total 76.5% of neonates who underwent abdominal or thoracic surgery had a chronological age of 28 days or less at the time of surgery (Table 1). The majority, 52.0%, of neonates were designated as American Society of Anesthesiologists (ASA) class of III. Major abdominal cases accounted for 87.0% of neonatal cases. Thirty-two percent of all neonates underwent emergent operation. 2.1. Postoperative adverse events were common in neonates The unadjusted rates of adverse events were high in neonates undergoing major abdominal or thoracic surgery—23.3% for “any” postoperative adverse event (Table 2). The most frequent of the individual adverse events was intraoperative/postoperative transfusion occurring in 8.5% of infants. The second most frequent adverse event was mortality; 4.4%. Reintubation was next, 3.9%, then surgical site infection, 3.7%, followed by postoperative sepsis, 2.7%. 2.2. Preoperative variables were highly associated with postoperative adverse events Preoperative variables highly associated with the occurrence of any postoperative adverse event in this population were identified (Table 3). A total of 16 variables were significantly associated with the occurrence of any adverse events. These 16 preoperative variables were: gestational age, primary procedure CPT category linear risk, case type, wound classification, ASA class, preoperative nutritional support, dialysis, cardiac risk factors, esophagogastrointestinal disease, hepatobiliary disease, APGAR score at 5 minutes, bleeding disease, hematologic disorder, transfusion, inotropic agent use, and sepsis (Table 3). Fifteen out of sixteen of these variables significantly increased risk, whereas the effect of gestational age depended on the category. These 16 variables became the components of the risk calculator model. 2.3. Model discriminative ability The discriminative ability of the model described in Table 3 was evaluated. The AUROC curve was 0.79 (95% CI = 0.78–0.82; Fig. 1). When this same model was split-set validated against the 2013 data of 1879 neonates undergoing major abdominal or thoracic surgery, the AUROC curve was 0.77 (95% CI = 0.74–0.79). Thus discrimination was qualified as good. The model calibration with the validation dataset was also good; the Hosmer–Lemeshow chi-square was 9.5, p = 0.31. The Brier score was 0.13 suggesting the predictions were well calibrated. 3. Discussion Preoperative patient-specific estimates of the probability of postoperative adverse events could better inform physicians and families of neonates of surgical risk. Such a data-driven quantification of surgical risk may also help set realistic postoperative expectations [11–13]. Estimation of risk can also enlighten preoperative preparations (of patient or caregivers) and inform postoperative resource or care planning. This study sought to create a risk calculator model that could help preoperatively estimate patient-specific probability of postoperative adverse event in neonates undergoing major abdominal or major thoracic surgery. It found that it is possible to preoperatively estimate patient-specific probability of postoperative adverse events in neonates undergoing major abdominal or thoracic surgery with good discriminative ability.
989
Table 2 Adverse events in neonates undergoing major abdominal or thoracic surgery. Adverse eventa
Count (n)
%
Composite “any” adverse event Transfusion Mortality Reintubation Surgical site infection Sepsis Wound dehiscence Pneumonia Urinary tract infection Cerebrovascular accident Catheter associated blood stream infection Cardiac arrest Renal failure
692 252 129 116 109 80 52 40 35 31 29 28 25
23.3 8.5 4.4 3.9 3.7 2.7 1.8 1.4 1.2 1.0 1.0 0.9 0.8
a
Includes all individual complications that occurred in 1% of greater of neonates.
Preoperative variables associated with two-fold or more increased odds of adverse events were: a dirty/infected case, preoperative requirement of dialysis, hepatobiliary disease, and inotropic agent use. The clinical variables found to be associated with adverse events are similar to previous studies that have sought to predict risk of adverse events in neonates. The literature has described the impact of prematurity, APGAR scores, cardiac abnormalities and preoperative sepsis on postoperative outcomes in neonates [2,3,14]. Other studies that have looked at predictors of outcomes in neonates have often favored objective measures such as birth weight, vital signs and laboratory parameters [15–17]. Some of these parameters were available in the current study but were not used since they fluctuate greatly over time even in a single patient. Incorporation of procedural factors was an important addition in the model of neonatal adverse events presented herein. This was accomplished using the risk associated with primary procedure CPT category as an element of modeling. These procedural variables allow for the calculator model to be used to predict postoperative adverse event rates across two major system types of operations. This risk calculator model is a means of giving a patient-specific data-derived probability of postoperative adverse events to parents preoperatively. Although clinical experience gives insight as to whom is more at risk for poor outcomes, the vast majority of physicians and surgeons cannot presently give parents their child's individualized, dataderived probability of postoperative complications preoperatively. This calculator model offers a means of estimating an infant's probability of postoperative adverse events preoperatively. Model refinements are planned before making this calculator model more widely available for use. First, it is critical to refine how diagnoses of congenital disorders are included. Second, these data should be prospectively validated to demonstrate that they have value in predicting the occurrence of adverse events. This study has several limitations. The first is the potential bias in the patient sample of neonates at the self-selected, typically large academic tertiary care hospitals that participate in ACS-NSQIP-P. As a result, the findings may not be generalizable to a wider sample of pediatric surgical patients in non-ACS-NSQIP-P hospitals. A second limitation is that longterm outcomes were not included. In pediatric surgery, most important functional outcomes can only be obtained after 30 days. These data are not available in ACS-NSQIP-P in its current form. Third, procedurespecific complications are not captured. Procedure-specific complications in neonates, such as a leak or stricture after an esophageal atresia repair, are often more common than generic complications such as pneumonia, and therefore in the future might be easier to model. Nonetheless, potentially avoidable general postoperative adverse events were used as primary outcome in this analysis in order to define undesirable surgical outcomes across a wide range of procedures. These general adverse events may be less sensitive than procedure-specific
990
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
Table 3 Components of risk calculator for adverse events in neonates undergoing major abdominal or thoracic surgery. Preoperative variables
Preterm Not preterm 24 completed weeks gestation 25–26 completed weeks gestation 27–28 completed weeks gestation 29–30 completed weeks gestation 31–32 completed weeks gestation 33–34 completed weeks gestation 35–36 completed weeks gestation Procedural variables CPT category linear risk Case typea Elective Urgent Emergent Wound classificationa Clean Clean-contaminated Contaminated Dirty/infected American Society of Anesthesiologistsa ASA class I ASA class II ASA class III ASA class IV ASA class V Metabolic and nutritional conditions Preoperative nutritional support a Renal conditions Dialysisa Cardiac conditions Cardiac risk factorsa No cardiac risk factors Minor risk factors Major risk factors Severe risk factors Alimentary tract conditions Esophago-gastrointestinal disease Hepatobiliary diseasea Neurologic condition APGAR score at 5 minutesa APGAR score 1 APGAR score 2 APGAR score 3 APGAR score 4 APGAR score 5 APGAR score 6 APGAR score 7 APGAR score 8 APGAR score 9 Hematologic and oncologic conditions Bleeding diseasea Hematologic disordersa Preoperative transfusiona Acuity of illness Inotropic agent usea Sepsis postoperatively No sepsis Systemic inflammatory response syndrome Sepsis Septic shock
Estimate
Lower Upper p 95% 95% CI CI
Reference 1.10 1.48 1.00 1.31 0.77 1.15 0.78
– 0.65 0.95 0.62 0.83 0.47 0.78 0.56
– 1.87 2.32 1.62 2.08 1.27 1.68 1.09
– 0.72 0.08 0.99 0.25 0.30 0.48 0.15
1.48
1.27
1.74
b0.0001
Reference – 1.61 1.23 0.81 0.60
– 2.11 1.10
– 0.001 0.17
Reference 1.37 1.70 2.06
– 1.02 1.12 1.35
– 1.85 2.60 3.15
– 0.04 0.01 0.001
Reference 0.44 0.78 1.16 1.00
– 0.17 0.32 0.47 0.33
– 1.12 1.88 2.86 3.08
– 0.09 0.57 0.75 1.00
1.60
1.26
2.02
0.0002
3.77
1.39
10.22
0.01
Reference 1.26 1.55 1.51
– 0.99 1.14 0.93
– 1.61 2.12 2.44
– 0.06 0.01 0.10
1.13 2.05
0.87 1.37
1.46 3.08
0.37 0.001
1.42 0.98 0.67 1.87 0.85 1.28 1.23 0.77 Reference
0.77 0.45 0.35 1.04 0.52 0.88 0.88 0.58 –
2.64 2.11 1.28 3.36 1.37 1.86 1.72 1.03 –
0.26 0.95 0.23 0.04 0.49 0.20 0.22 0.08 –
1.85 1.55 1.52
1.06 1.16 1.10
3.21 2.06 2.09
0.03 0.004 0.01
2.61
1.75
3.91
b0.0001
Reference – 1.03 0.44
– 2.41
– 0.95
1.72 1.99
2.87 4.63
0.04 0.11
1.03 0.86
a All categorical variables denoted had an overall significant (p b 0.05) association with adverse events despite discrete levels within the categorical variables not being significantly different from the reference level. Bold items significant at p b 0.05.
Fig. 1. The area under the receiver operating characteristic curve of the risk calculator model. Each curved line represents the discriminative ability of the model with each dataset, development and validation.
procedures in pediatric surgery would be most desirable, owing to the low event rate and relatively low number of patients captured, this is not yet possible. Fifth, this calculator model does not calculate whether the risks of the procedure outweigh the benefits, but it does provide insight into such a deliberation. Finally, this risk calculator model does not predict specific adverse events rates. Once a larger sample size of neonates is available, developing a more outcome-specific calculator will make it more useful to clinicians, patients and families.
4. Conclusion This study demonstrated the feasibility of constructing a risk calculator model for preoperatively estimating patient-specific probability of postoperative adverse events for neonates undergoing major abdominal or thoracic surgery with good discriminative ability. Sixteen variables were associated with postoperative adverse events. Four preoperative clinical variables (a dirty/infected case, preoperative requirement of dialysis, hepatobiliary disease, and inotropic agent use) were associated with a greater than two-fold increased odds of postoperative adverse events. This model could be used during the preoperative counseling process to help families anticipate realistic outcomes. It could also inform preoperative preparation of the patient or caregiver team, and could shed light on potential resource needs for postoperative acute and post-acute care.
Acknowledgment outcomes and may not give a full picture of long-term functional outcomes critical in children's surgery. However, general postoperative adverse events are specific metrics of perioperative quality of care. Fourth, although a more comprehensive calculator model including all
AMS's time was supported for this publication by The Robert Wood Johnson Foundation Clinical Scholars® program and the U.S. Department of Veterans Affairs.
A.M. Stey et al. / Journal of Pediatric Surgery 50 (2015) 987–991
Appendix A. Major abdominal and thoracic cases accrued in ACS-NSQIP-P CPT
PROCEDURE DESCRIPTION
22900 37181-38120 38562-38780 39502 43280-43326 43520-43840 44005 44010-44025 44050 44055 44110-46748 47100-47130 47480-47620 47700-47800 48120-48146 35840, 49000-49322 49324-49421 49600-49611 50205 50220-50548 51500 51530-51585 57106-57112 31786-31805 32095-32160 32220-32402 32440-32501 32601-32665 35271-38746 39010-39400 39503-39561 43101-43352 19260
Abdominal Wall Tumor Excision Splenic Procedures Lymphadenectomy Paraesophageal Hernia Repair Fundoplication Gastric Procedures Lysis of Adhesions Enterotomy Reduction of Volvulus, Intussusception or Internal Hernia Ladd Procedure Bowel Resection and Ostomy Liver Resection Cholecystectomy Biliary Procedures Pancreatic Resection Abdominal Exploration) Peritoneal Dialysis Catheter Placement Omphalocele Repair Renal Biopsy Nephrectomy Urachal Cyst Excision Cystostomy and Cystectomy Vaginectomy Intrathoracic Tracheal Procedures Thoracotomy Procedures Decortication Lung Resection Thoracoscopic Procedures, Thoracic Lymphovascular Procedures Mediastinal Procedures Diaphragmatic Procedures Esophageal procedures Chest Wall Tumor Excision
991
References [1] Rowe MI, Rowe SA. The last fifty years of neonatal surgical management. Am J Surg 2000;180:345–52. [2] Lillehei CW, Gauvreau K, Jenkins KJ. Risk adjustment for neonatal surgery: a method for comparison of in-hospital mortality. Pediatrics 2012;130:e568–74. [3] Son JK, Lillehei CW, Gauvreau K, et al. A risk adjustment method for newborns undergoing noncardiac surgery. Ann Surg 2010;251:754–8. [4] Skarsgard ED, Blair GK, Lee SK. Toward evidence-based best practices in neonatal surgical care-I: the Canadian NICU Network. J Pediatr Surg 2003;38:672–7. [5] Perlman NB, Freedman JL, Abramovitch R, et al. Informational needs of parents of sick neonates. Pediatrics 1991;88:512–8. [6] Aite L, Zaccara A, Trucchi A, et al. Parents' informational needs at the birth of a baby with a surgically correctable anomaly. Pediatr Surg Int 2006;22:267–70. [7] Bilimoria KY, Liu Y, Paruch JL, et al. Development and evaluation of the universal ACS NSQIP surgical risk calculator: a decision aid and informed consent tool for patients and surgeons. J Am Coll Surg 2013;217:833–842.e831-833. [8] Raval MV, Dillon PW, Bruny JL, et al. American College of Surgeons National Surgical Quality Improvement Program Pediatric: a phase 1 report. J Am Coll Surg 2011;212: 1–11. [9] Bruny JL, Hall BL, Barnhart DC, et al. American College of Surgeons National Surgical Quality Improvement Program Pediatric: a beta phase report. J Pediatr Surg 2013; 48:74–80. [10] Cohen ME, Ko CY, Bilimoria KY, et al. Optimizing ACS NSQIP modeling for evaluation of surgical quality and risk: patient risk adjustment, procedure mix adjustment, shrinkage adjustment, and surgical focus. J Am Coll Surg 2013;217:336–346.e331. [11] Schwarze ML, Sayla MA, Alexander GC. A comparison of patient and physician beliefs about infrainguinal bypass operation: what role should surgical optimism play? Surgery 2007;141:239–44. [12] Lattig F, Fekete TF, O'Riordan D, et al. A comparison of patient and surgeon preoperative expectations of spinal surgery. Spine (Phila Pa 1976) 2013;38(12): 1040–8. [13] Jourdan C, Poiraudeau S, Descamps S, et al. Comparison of patient and surgeon expectations of total hip arthroplasty. PLoS One 2012;7:e30195. [14] Johansson S, Montgomery SM, Ekbom A, et al. Preterm delivery, level of care, and infant death in Sweden: a population-based study. Pediatrics 2004;113:1230–5. [15] Sacco Casamassima MG, Salazar JH, Papandria D, et al. Use of risk stratification indices to predict mortality in critically ill children. Eur J Pediatr 2014;173:1–13. [16] Richardson DK, Gray JE, McCormick MC, et al. Score for neonatal acute physiology: a physiologic severity index for neonatal intensive care. Pediatrics 1993;91:617–23. [17] Skarsgard ED, MacNab YC, Qiu Z, et al. SNAP-II predicts mortality among infants with congenital diaphragmatic hernia. J Perinatol 2005;25:315–9.