Obesity and Mortality Risk in Critically Ill Children
BACKGROUND AND OBJECTIVES: Childhood obesity is epidemic and may be associated with PICU mortality. Using a large multicenter PICU database, we investigated the association between obesity and PICU mortality, adjusting for initial severity of illness. We further investigated whether height- and weight-based classifications of obesity compared with a weight-based classification alone alter the mortality distribution.
METHODS: This retrospective analysis used prospectively collected data from the Virtual PICU Systems database. Height, weight, age, and gender were used to calculate z score groups based on Centers for Disease Control and Prevention and World Health Organization growth curves. A random effects mixed logistic regression model was used to evaluate the association between obesity and PICU mortality, controlling for hospital, initial severity of illness, and comorbidities.
RESULTS: A total of 127 607 patients were included; the mortality rate was 2.48%. Being overweight was independently associated with increased PICU mortality after controlling for severity of illness with the Pediatric Index of Mortality 2 score and preexisting comorbidities. Mortality had a U-shaped distribution when classified according to weight-for-age or weight-for-height/BMI. When classifying patients using weight-for-age without respect to height, the nadir of the mortality curve was shifted, potentially falsely implying a benefit to mild obesity.
CONCLUSIONS: Risk-adjusted PICU mortality significantly increases as weight-for-height/BMI increases into the overweight and obese ranges. We believe that height dataare necessary to correctly classify body habitus; without such information, a protective benefit from mild obesity may be incorrectly concluded.
- ARDS —
- acute respiratory distress syndrome
- CCC —
- complex chronic condition
- CDC —
- Centers for Disease Control and Prevention
- NCCC —
- noncomplex chronic condition
- PIM2 —
- Pediatric Index of Mortality 2
- SMR —
- standardized mortality ratio
- VPS —
- Virtual PICU Systems
- WHO —
- World Health Organization
What’s Known on This Subject:
There is inconsistent evidence regarding the relationship between obesity and mortality for critically ill patients. In children, previous studies that use weight-for-age normalization generally identify the lowest ICU mortality risk in children who are mildly to moderately overweight.
What This Study Adds:
When accounting for patient height, the lowest ICU mortality risk occurred in normal weight. Risk-adjusted PICU mortality increases as weight for height/BMI increases into the overweight and obese ranges. Height data are necessary to correctly classify PICU body habitus.
Obesity is epidemic in adults and children in the United States, affecting 16.9% of children aged 2 to 19 years.1 When these children require hospitalization or critical care services, they present challenges to the health care system. Some investigators, although not all, believe that obese children have higher morbidity and mortality.2–4
In adults, higher BMI is associated with the development of acute respiratory distress syndrome (ARDS)5; for those developing ARDS, increasing BMI was associated with prolonged stay. However, increased BMI was not associated with increased mortality. This study is one of several that supports the “obesity paradox,” the notion that obesity as a comorbidity might be protective in some disease states. Interestingly, obesity has been identified as a risk factor for death from influenza A–associated ARDS,6,7 although these studies were small. Larger adult critical care studies have shown that although obesity may be associated with longer ICU stay, when controlling for severity of illness, the mortality is the same8,9 or lower10 than for nonobese adults. In non-ICU settings, a 2013 meta-analysis11 of 2.88 million adult patients found that the overweight group had a statistically significant lower all-cause mortality.
The literature is limited regarding obesity and outcomes in pediatric critical care.2–4,12,13 Existing studies identify a U-shaped distribution for mortality based on weight percentile, with both underweight and overweight children having increased mortality. There are other issues. First, it is unknown whether these studies2,3 reflect the population of US children. Second, all previous studies classified patients based on weight for age and did not take into account the patient’s height. It is unknown whether this approach results in misclassification of obesity.2,3,12,13
The goal of the present study was to examine the association between obesity and pediatric ICU mortality in US children, adjusting for initial severity of illness. We also sought to determine whether height- and weight-based classifications of obesity compared with weight-for-age alone would result in a different mortality distribution. A large PICU database was used to test these questions.
This study was approved by the Children’s Hospital Los Angeles institutional review board. This retrospective study of prospectively collected data from the Virtual PICU Systems (VPS, LLC) database was conducted from January 1, 2009, to March 31, 2013. VPS is the largest international PICU database. Participating PICUs submit prospectively collected data elements for all admitted patients. Diagnostic data and procedure codes submitted to VPS were used to identify the cohort. Data elements obtained from VPS included PICU mortality, Pediatric Index of Mortality 2 (PIM2) scores, anthropometric data, primary diagnosis, and all secondary diagnoses. All secondary diagnoses were coded into complex chronic conditions (CCC) or noncomplex chronic conditions (NCCC), as originally described by Feudtner et al14,15 and recently updated.16 This approach was used to control for preexisting comorbidities such as genetic syndromes, which are not accounted for in severity of illness scores and may confound the relationship between body habitus and mortality. VPS data were provided by VPS, LLC.
For the entire data set, weight for age and gender was used to calculate z scores from the appropriate Centers for Disease Control and Prevention (CDC) and World Health Organization (WHO) growth curves. The CDC recommends use of WHO growth curves for children aged <2 years and CDC growth curves for children aged ≥2 years. For a normally distributed population, 1 z score is 1 SD away from the US population mean. Height is not a required VPS data set element but is present in a large subset of patients. For patients who had height recorded, we used weight for height for age and gender to calculate z scores. Children aged <2 years were classified by using weight for age and height for age from WHO growth curves. Children aged ≥2 years were classified by using weight for age and BMI for age from CDC growth curves. We combined the z scores calculated by using height from the patients aged <2 years and z scores calculated by using BMI from the patients aged ≥2 years (this grouping is referred to as weight-for-height/BMI for the remainder of the article). Patients were classified into 9 z score groups using the groups proposed and validated by Prince et al3: –3.5 or lower, –3.5 to –2.51, –2.5 to –1.51, –1.5 to –0.51, –0.5 to 0.49, 0.5 to 1.49, 1.5 to 2.49, 2.5 to 3.49, and ≥3.5. Age was classified into 7 groupings based on CDC criteria: <1 month, 1 month to 1 year, 1 to 5 years, 5 to 10 years, 10 to 15 years, 15 to 18 years, and >18 years. The racial/ethnic groups included: African American, Hispanic, white/European non-Hispanic, Asian/Indian/Pacific Islander, American Indian, other/mixed, and unspecified.
The primary analysis was restricted to patients who had height measured and the results were available. To evaluate for potential selection bias from the ICUs, which record height (this element is only recorded when the ICU has agreed to submit height for all of their patients), we compared demographic characteristics and outcomes between the entire cohort and the subgroup containing height. To address the potential bias using growth curves generated from supine (aged <2 years) versus standing (aged ≥2 years) patients, the height group aged <2 years was analyzed separately (Supplemental Information).
Statistical analysis was performed by using Stata version 13.1 (Stata Corp, College Station, TX) and Statistica version 10 (StatSoft, Tulsa, OK). Continuous data were analyzed with the Wilcoxon rank-sum test because the data were not normally distributed. Categorical values were analyzed by using Pearson’s χ2 test or expected versus observed χ2 test. The association between anthropometric data and mortality was demonstrated graphically (data were separated into groupings based on age, z scores of weight for age, and z scores of weight for height/BMI). Standardized mortality ratios (SMRs) were computed within each z score group based on observed deaths versus predicted deaths from the PIM2 severity of illness score, with 95% confidence intervals using the Mid-P exact test based on Miettinen’s modifications.17
Univariate analysis was performed to identify factors associated with mortality, including: demographic characteristics, anthropomorphic data, severity of illness, presence of CCC, and presence of NCCC. We next performed multivariate analysis with the outcome of PICU mortality. This analysis used a random effects mixed logistic regression model to control for the hospital in which the patient received care18 and then added the other variables with a univariate association with mortality. The variables considered for inclusion in the model were hospital, gender, age category, race/ethnicity, z score of weight for height/BMI separated into 9 groups, PIM2 score, presence of a CCC, and presence of an NCCC.
Obesity and Mortality
There were 331 057 consecutive patient records identified between the inclusive dates, and 5732 were eliminated due to data entry errors or missing data. There were 325 325 remaining patients in the entire VPS cohort, of whom 197 718 patients did not have height recorded. The demographic characteristics of the groups with and without height are shown in Supplemental Table 4. Overall, the basic demographic characteristics were similar between the VPS cohort without height data and the height VPS cohort. More patients in the height VPS cohort had race/ethnicity reported (race/ethnicity is not a required data element for VPS). There were statistically significant differences between the height VPS cohort and the VPS cohort without height data with respect to race category, weight-for-age z score, and age; these factors were unlikely to be clinically relevant, however. There was no difference in mortality or severity of illness.
The 127 607 patients with height data from 50 US PICUs (height VPS cohort) included 3170 deaths, with a mortality rate of 2.48%. The PIM2-adjusted mortality (using SMR) for this cohort was 0.87. Using the weight-for-height/BMI classification schema, there was a steady increase in both unadjusted mortality and PIM2 SMR because the weight-for-height/BMI z score group increased above –0.5 to 0.49 (Fig 1, Table 1). The most overweight group had a PIM2 SMR 0.29 point higher than the normal weight group (expected/observed χ2 test, P < .001). A similar trend was seen as the weight-for-height/BMI z score decreased into underweight regions. There were significant differences between survivors and nonsurvivors with regard to gender, age, race, and distribution of the z score groupings for weight-for-height/BMI, presence of a CCC, and presence of an NCCC (Table 2). Age seemed to have an important effect modification on mortality (Fig 2, Supplemental Table 5), as the increased risk for death in underweight children was most prominent in children <1 year of age and children >15 years of age. Furthermore, the increased risk for death with being overweight seemed to be a consistent trend for children >1 year of age. The 95% confidence interval for the percent mortality according to age groups is presented in Supplemental Fig 5. After controlling for hospital, age group, demographic characteristics, CCC and NCCC, and PIM2 score (Table 3), being moderately underweight (z score, –2.5 to –1.51) or being in any overweight group (z score, 0.5 to 1.49, 1.5 to 2.49, 2.5 to 3.49, and >3.5) was independently associated with PICU mortality (all, P < .012).
Comparison of Weight- and Height-Based Classifications
The height VPS cohort was used to compare the distribution of z scores determined by using weight-for-height/BMI calculations versus those determined by weight-for-age. Z scores calculated according to weight for age shifted the histogram to the left compared with weight for height/BMI (Fig 3). The median weight-for-age z score was 0.42 less than the median weight-for-height/BMI z score. When examining individual patients, only 46 405 (35.5%) of 127 607 patients were classified into the same z score grouping using weight for age and weight for height/BMI (Supplemental Table 6). Assuming that weight-for-height/BMI correctly classifies patients, 50 735 were put into a lower z score group (39.8%), and 28 820 (22.6%) were put into a higher z score group.
Figure 4 illustrates the effect that the 2 methods of calculating z score groupings have on unadjusted mortality. When patients were classified by using weight for height/BMI, the nadir of mortality was in the –0.5 to 0.49 z score group, with increasing mortality as weight for height/BMI increased. When using weight for age, the nadir of mortality seemed to be spread over the –0.5 to 1.49 z score groupings before increasing.
When separately evaluating patients aged <2 years (Supplemental Information), there was a higher overall mortality rate compared with the entire cohort. There was a rise in mortality in both the highest underweight and overweight groups, similar to the entire cohort.
Being overweight or obese was associated with increased PICU mortality even after controlling for severity of illness, demographic characteristics, and comorbidities. This pattern of increased mortality with obesity seemed to appear as early as 1 year of age. Mortality also increased in underweight children, although this difference was no longer significant in the most underweight children after adjusting for severity of illness, demographic characteristics, and comorbidities. Moreover, we believe that height information is necessary to provide an accurate evaluation of anthropometric data because not using height generally results in an underestimation of the number of overweight and obese children, as well as an overestimation of “underweight” children. The misclassification of patients without using height data alters the nadir of the mortality curve and erroneously indicates a protective benefit for being mildly to moderately overweight.
Our findings are in line with pediatric studies conducted in Australia and the United Kingdom. In a single-center Australian study, Numa et al2 reviewed 6337 consecutive PICU admissions and reported increased mortality for both underweight and overweight children, using a classification based on weight for age. This U-shaped distribution of mortality in pediatric critical care was recently reported by Prince et al3 in the United Kingdom. In their study of 14 307 admissions, the investigators found that weight for age was an independent risk factor for mortality, with increased mortality in both underweight and overweight populations. They also found that the mean weight for age of patients admitted to their PICU was ∼1 SD below the UK reference population mean. There are important differences between our study and previous research. Neither of the 2 previous studies2,3 used height-based classification schema. Although we confirmed a U-shaped curve for PICU mortality, our data do not support a protective effect of mild obesity. The nadir of mortality presented by Numa et al was at the 75th percentile; the nadir of mortality for Prince et al was between 0.5 and 2.5 SDs above the weight-for-age mean in the UK study. In our study, when we did not take into account patient height and calculated weight-for-age z scores, the nadir for mortality was from –0.5 to 1.5 SDs above the CDC’s weight-for-age mean, which is more consistent with the studies of Numa et al and Prince et al. However, when we used patient height and calculated weight-for-height/BMI z scores (Fig 4), the nadir of the U-shaped mortality curve occurred at the –0.5 to 0.49 z score group, and adjusted mortality was higher for even the slightly overweight group (0.5–1.5 z scores). The difference in these classifications is partially explained by the fact that the height VPS cohort is shorter than the CDC reference population, representing a difference between patients admitted to PICUs and healthy children (median height for age z score, –0.35 [(1,3 IQR –1.63 to 0.76]) (Supplemental Table 4). These data support no protective benefit to being mildly to moderately overweight. There is no overweight-for-height/BMI z score groups in which there is a decreased risk of adjusted or unadjusted PICU mortality, thus not supporting the “obesity paradox” in critically ill US children. Previous reports of this finding may possibly be explained by misclassification of patients when height data were not available.
As with the other 2 studies,2,3 our data also demonstrate an increased risk for PICU mortality for significantly underweight patients. Interestingly, this increase in mortality did not seem to hold for the most underweight children after controlling for severity of illness. This finding may imply that severity of illness scores or the comorbidities used in the model capture the mortality risk for these extremely underweight children. It may also reflect the smaller sample sizes in this group compared with the moderately underweight children. Unfortunately, our data set cannot adequately identify malnutrition on admission nor separate this factor from underlying clinical diagnoses or syndromes associated with failure to thrive. This effect did hold after controlling for the comorbidities contained within the CCC, which may be a surrogate for syndromes associated with failing to thrive. Therefore, our data support that lowest mortality occurs at optimal weight.
We also found that the association between increasing obesity and mortality held for age groups as young as 1 to 5 years of age (Fig 2, Supplemental Table 5). Although these data are observational, they provide some evidence that even very early childhood obesity (at age <5 years) is associated with worse outcome. This interaction needs to be studied further.
We can speculate about possible explanations for our findings. First, it is important to note that although we found an association between obesity and mortality, this finding does not imply causality. Given that severity of illness-adjusted mortality was still higher for obese children (implying the excess mortality is not completely explained by higher initial severity of illness), it is possible that process of care issues in the ICU may contribute. These issues could be related to medication dosing, given that some medications are dosed to ideal body weight, some to actual body weight, and some to therapeutic effect. This issue becomes further complicated because measured weight frequently changes in ICU patients as a consequence of intravascular and extravascular fluid status. In addition, high ventilator tidal volumes, which are associated with mortality (particularly for adults with ARDS), will be “higher” for obese children if actual body weight is used instead of ideal body weight. It does not appear that pediatric practitioners are meticulous about using ideal body weight for tidal volume calculations, potentially subjecting obese children to higher risk of ventilator-induced lung injury.
Given the current trajectory of obesity in the United States, there is a growing need to identify best practices to care for obese children. Our findings do not refute the possibility that there is a disparity in the way we care for obese children in PICUs. However, it is important to acknowledge the many unmeasured and potentially important confounders, which may explain this relationship between obesity and mortality (eg, socioeconomic status, genetic factors, host immune response, illness severity), that are not captured by severity of illness scores. Moreover, from a public health standpoint, our data do not support any protective benefit to being mildly to moderately overweight for critically ill children.
Although the very large sample size and multiple PICUs may increase the generalizability of our study, limitations remain. First, there is a possibility of selection bias when using only patients with height data because this approach does not include all of the PICUs in the entire cohort. However, we report that the cohort with height data in the VPS data set (height VPS cohort) is similar to the VPS cohort without height data available. Furthermore, evaluating mortality by using a simple age or weight-for-age classification (analysis not shown), the 2 cohorts seemed similar. We believe this finding indicates that the patients for whom height is recorded are not treated significantly differently from those without height data. Second, it is not possible to verify the accuracy of the weight and height measurements, or how and when they were obtained, due to the large sample size from >100 institutions. Equipment, measurement techniques (standing versus supine), and time of measurement (on ICU admission or before) are not standardized across institutions, and this information is not recorded in the VPS. Given the large sample size, it is unlikely that there was a systematic bias in 1 direction or another for either height or weight, which would affect our results. Third, we did not extract patient data on underlying syndromes that would cause them to be significantly underweight or overweight for their height/BMI for age, although we did use a validated surrogate with CCC and NCCC. Although this approach possibly confounds the results, it is unlikely to explain the entire association we observed. Fourth, we chose to classify our patients by using 9 z score groups to remain consistent with previously published data. It is possible that different cutoffs for percentile for weight for height/BMI would alter the results slightly. Unfortunately, in the literature on pediatric obesity, many authors have used percentile cutoffs of <3%, 3% to 85%, 85% to 95%, and >95%, and this method limits discrimination on the effect of obesity because a great number of patients are compressed into those broader percentiles. The fifth limitation was that this study was drawn from multiple US PICUs, and there may be variation in the care provided to obese or underweight patients; these variations could alter the outcome mortality. This possibility was controlled for by using (PICU) center in the multivariate models.
Based on this large retrospective study of prospectively collected data from multiple US PICUs, we believe that height data are necessary to correctly classify obesity. Without complete anthropometric data, it is possible to assume a protective benefit from mild obesity that no longer holds when using a weight-for-height/BMI classification. There seemed to be a significant increase in risk-adjusted PICU mortality as weight for height/BMI increased into the overweight and obese ranges, even for children between 1 and 5 years of age.
VPS data were provided by VPS, LLC. No endorsement or editorial restriction of the interpretation of these data or opinions of the authors has been implied or stated.
- Accepted December 10, 2015.
- Address correspondence to Patrick A. Ross, MD, Department of Anesthesiology Critical Care Medicine, 4650 Sunset Blvd, Mailstop 3, Children’s Hospital Los Angeles, Los Angeles, CA 90027. E-mail:
FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.
FUNDING: No external funding.
POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.
- Goh VL,
- Wakeham MK,
- Brazauskas R,
- Mikhailov TA,
- Goday PS
- Gong MN,
- Bajwa EK,
- Thompson BT,
- Christiani DC
- Davies A,
- Jones D,
- Bailey M, et al; Australia and New Zealand Extracorporeal Membrane Oxygenation (ANZ ECMO) Influenza Investigators
- Srinivasan V,
- Nadkarni VM,
- Helfaer MA,
- Carey SM,
- Berg RA; American Heart Association National Registry of Cardiopulmonary Resuscitation Investigators
- Feudtner C,
- Christakis DA,
- Connell FA
- Copyright © 2016 by the American Academy of Pediatrics