Weight as a Risk Factor for Mortality in Critically Ill Patients

We performed a secondary analysis of a multicenter, multinational epidemiology study of children who were critically ill and young adults (ie, Assessment of Worldwide Acute Kidney Injury, Renal Angina, and Epidemiology [AWARE] study; ClinicalTrials.gov identifier NCT01987921). In brief, the AWARE study recruited all patients between 3 months and 25 years with a PICU stay of at least 48 hours in 32 PICUs across Asia, Australia, Europe, and North America for 3 consecutive months in 2014. The primary goal of the AWARE study was to define the epidemiology of acute kidney injury (AKI) and to characterize risk factors for AKI and associated morbidity in children and young adults admitted to the PICU.²³  We chose the AWARE database for this study because it contains detailed, prospectively collected records from a large cohort of PICU patients (a total of 6821). The database provides information about patients’ epidemiology, admission diagnoses, comorbidities, and short-term outcomes; thus, it is a unique database with data collected from sites across the globe. Of the 6821 available records, subjects with complete data for sex, age, height, and weight were included in our analysis. Because sepsis in children is a significant cause of morbidity and mortality worldwide, we performed a planned subgroup analysis for patients who had a primary diagnosis of sepsis on admission to the PICU. This secondary data analysis was reviewed and approved by the Cincinnati Children’s Hospital Medical Center Institutional Review Board.

Patients were classified into 4 major categories (underweight, normal weight, overweight, and obese) on the basis of BMI adjusted for age and sex, as recommended by the Centers for Disease Control and Prevention for patients aged 2 to 19 years and by the World Health Organization for patients younger than 2 years.²⁴  For patients aged 2 to 19 years, the following Centers for Disease Control and Prevention BMI percentiles were used: <5% to define underweight, 5% to 84.9% to define normal weight, 85% to 94.9% to define overweight, and >94.9% to define obesity.²⁵  For patients younger than 2 years, the following World Health Organization BMI percentiles were used: <2.3% to define underweight, between 2.3% and 84.9% to define normal weight, between 85% and 97.6% to define overweight, and >97.6% to define obesity. For patients older than 20 years, a BMI <18.5 was used to define underweight, 18.5 to 24.9 to define normal weight, 25 to 29.9 to define overweight, and >29.9 to define obesity.²⁶ 

The primary outcome was 28-day mortality. Secondary outcomes were ICU length of stay, presence and duration of mechanical ventilation support, vasoactive agents, occurrence of AKI based on the Kidney Disease: Improving Global Outcomes (KDIGO) criteria, and need for renal replacement therapy.

The description of the study population is presented as the frequency or the median with the interquartile range (IQR), as appropriate, with bivariate associations with weight status category tested by using χ² analysis or Wilcoxon rank test, respectively. For multivariable models of 28-day mortality, variables to be tested included those significant (P ≤ .05) in the bivariate analysis used to compare surviving and nonsurviving cohort members. Logistic regression models were constructed by using backward elimination, retaining variables in the final model that were either statistically significant (P ≤ .05) or improved model fit (lower Akaike information criterion). Survival analysis was conducted with the end point being days between admission and death within the 28-day mortality window. Significance was determined by using the log-rank test with a Šidák correction for multiple comparisons. We adjusted mortality for severity of illness (SOI). Not all subjects in the AWARE database had a record of their SOI score, and because different sites used different scores that are not comparable (ie, Pediatric Index of Mortality, Pediatric Risk of Mortality, and Pediatric Logistic Organ Dysfunction), simple adjustment was not possible. To overcome this problem, we generated 3 subcohorts on the basis of the metric that was used to define SOI. We used the same multivariable model that was used to assess predictors of mortality (with and without adding SOI as a variable). All analyses were conducted by using SAS version 9.3 (SAS Institute, Inc, Cary, NC).

A total of 6821 records from the AWARE study database were screened for eligibility, and 3719 (55%) had complete data for age, sex, height, and weight and therefore met study inclusion criteria (Fig 1). One hundred eighteen records were missing survival status at day 28; therefore, 3601 had complete 28-day data for the overall cohort. The baseline characteristics of the overall cohort are depicted in Table 1. Subjects were divided as follows: 1059 (28.5%) were underweight, 1649 (44.3%) were normal weight, 423 (11.4%) were overweight, and 588 (15.8%) were obese. The median age of the overall cohort was 5.3 years (IQR: 1.6–12.5), with patients who were underweight being younger compared with the other groups (median: 4.5 years; IQR: 1.5–11.8; P = .001). Weight groups differed by racial distribution, with African American patients more likely to be overweight or obese (P = .003) and Asian patients more likely to be underweight (P = .01). Patients from East Asia and Europe were more likely to belong to the underweight group, whereas patients from Australia and North America were more likely to belong to the overweight and obese groups (P = .01). Respiratory failure on admission was significantly higher in the underweight group (P < .001). The underweight group had significantly more comorbidities and was more likely to have ≥3 comorbidities on admission compared with the other weight groups (P < .001).

A total of 542 patients (14.6%) had a primary diagnosis of sepsis on admission (sepsis cohort). Subdivision to weight groups in the sepsis cohort resulted in 188 (34.7%), 230 (42.4%), 58 (10.7%), and 66 (12.2%) patients in the underweight, normal, overweight, and obese weight groups, respectively. More patients in the sepsis cohort were underweight, and fewer were obese compared with the overall population (P = .002; Supplemental Table 4). Consistent with the overall cohort findings, the underweight sepsis group also had a higher prevalence of comorbidities and was more likely to have ≥3 comorbidities compared with the other groups (P < .001). Unlike the overall cohort, age, race, region, and admission diagnosis were not significantly different by weight status in the sepsis cohort.

A total of 129 (3.6%) patients in the overall cohort and 48 (9.1%) patients in the sepsis cohort died by day 28 after PICU admission. Overall 28-day mortality was higher in the underweight group compared with other weight groups, both in the overall cohort (P < .001; Table 2) and in the sepsis cohort (P = .003; Supplemental Table 5). In addition, a survival analysis revealed an earlier time to death for the underweight group compared with all other groups in the overall cohort (Fig 2A; P < .005 for all comparisons of the underweight group with other groups). Similarly, underweight status was associated with an earlier time to death in the sepsis cohort compared with the other groups (Fig 2B; P < .02 for all comparisons of the underweight group with other groups).

To further explore whether weight status is an independent contributor of mortality and to exclude possible confounders, such as baseline comorbidities, SOI, and others, we performed a multivariable logistic regression analysis that revealed an independent 1.8-fold (95% confidence interval [CI]: 1.2–2.8) increase in 28-day mortality for the underweight group compared with the normal weight reference group in the overall cohort (Table 3) and a 2.9-fold (95% CI: 1.2–6.9) increase in the sepsis cohort (Supplemental Table 6). Mortality was not different in the overweight or obese group compared with the normal weight group in both the overall and sepsis cohorts. Each comorbidity (excluding hematologic comorbidity) by itself and the total number of comorbidities (both as continuous and dichotomous variables representing <3 vs ≥3 comorbidities) were not found to be associated with mortality, and none could explain the increased mortality seen in the underweight group. To adjust the weight-based mortality results to SOI, we created 3 subcohorts, as explained previously, and used the same multivariable model to assess predictors of mortality (with and without adding SOI as a variable). We found that in all subcohorts, adding the SOI score to the model did not change the weight category effect on mortality. This reveals that the results are not biased by SOI.

Other variables associated with increased 28-day mortality in the overall cohort by using a multivariable model were as follows: admission diagnosis of shock (odds ratio [OR]: 3.4; 95% CI: 2.2–5.2), cardiovascular-related admission diagnosis (OR: 2.5; 95% CI: 1.4–4.7), respiratory failure on admission (OR: 1.6; 95% CI: 1.1–2.4), central nervous system–related admission diagnosis (OR: 2.4; 95% CI: 1.5–3.9), having an hematologic comorbidity (OR: 2.9; 95% CI: 1.7–4.9), need for vasoactive support (OR: 5.2; 95% CI: 3.3–7.9), and stage 3 KDIGO criteria AKI (OR: 4.3; 95% CI: 2.6–7.3) compared with no AKI (Table 3).

In the overall cohort, the ICU length of stay was longer for the underweight group (median: 3 days [IQR: 1–6 days]) compared with normal weight, overweight, and obese groups (median of 2 days for each group; P < .001) (Table 2). In addition, more patients who were underweight required mechanical ventilation support (P < .001) and had fluid overload >10% at day 3 (P < .001). Among the sepsis cohort, the only difference in secondary outcomes by weight category was a fluid overload rate >10% at day 3, which was higher in underweight and normal weight groups than in the obese group (P < .001). Other secondary outcomes (ie, length of mechanical ventilation, need for vasoactive support, AKI at day 3, and need for renal replacement therapy) did not differ between weight groups. A sensitivity analysis used to assess the association between fluid overload at day 3 and higher mortality rates among patients who were underweight was not significant (P = .12), suggesting that higher mortality was not due to fluid overload alone.

In this study, we performed a secondary analysis of a multicenter, multinational epidemiology observational study (ie, the AWARE study) of children and young adults who were critically ill, with the aim to explore the association between weight status and short-term outcomes of patients admitted to the PICU at centers worldwide. We found that being underweight, but not overweight or obese, increases the risk for mortality within 28 days of PICU admission, even after adjustment for possible confounders, such as admission diagnosis, comorbidities, and SOI. Higher mortality risk in subjects who were underweight was found in the overall cohort, and even more so in the sepsis cohort.

Contrary to our initial hypothesis, obesity was not associated with worse outcomes. Additionally, we did not find evidence to support the obesity paradox in the pediatric population because no survival benefits were found in patients who were obese or overweight compared with patients who had a normal weight. Our mortality results were somewhat different from those reported by Ross et al,¹⁸  who suggested a U-shaped distribution of mortality for pediatric patients who were critically ill according to weight categories, with a higher mortality risk in subjects who were underweight, overweight, and obese in comparison with subjects who had a normal weight. Differences in study design, study population, data collection methods, and weight group stratification may explain the discrepancy in mortality results between our study and the study by Ross et al.¹⁸  Consistent with our findings, Davis et al¹⁷  reported that being overweight or obese was neither protective nor a risk factor for mortality in children who were critically ill. Of note, in all studies so far (including ours), authors looked at the association between the state of obesity and outcomes while ignoring the concept of duration of obesity, which can be an important variable effecting and mediating pathologic states related to obesity, as seen in adults.²⁷ 

The strong association between underweight and increased mortality, as seen in our study, may be mediated by malnutrition as a cofactor. Malnutrition is often caused by long-term chronic diseases and is a common finding among the PICU population.^28,29  Malnourished children are often characterized by low weight, anorexia, muscle and fat wasting, low energy reserves, and deranged metabolic response to stress, as occurs in critical illness.^30,31  Unfortunately, data regarding the nutritional state of our study population before and during their PICU stay were not available to us; hence, we cannot determine if malnutrition plays a role as a mediator. We found higher rates of baseline comorbidities among the underweight group compared with the other weight groups but also found that adjustment of mortality to comorbidities and SOI on admission did not abolish the association between being underweight and increased mortality. This suggests that underweight is an independent risk factor for mortality not biased by the baseline comorbidities, chronic illnesses, or SOI. Recently, in a retrospective cohort analysis of a large clinical data repository from 139 hospitals in the United States, Pepper et al¹¹  reported increased short-term mortality in underweight adult patients with sepsis. The adjusted OR for mortality was 1.6 for underweight relative to normal weight patients. Our study reveals the same association between underweight state and increased mortality in the pediatric population, with an adjusted OR for mortality of 1.8 in the overall cohort and 2.9 in the sepsis cohort.

One of the interesting findings of this study was the fact that subjects who were underweight, in both cohorts, were more likely to develop >10% fluid overload during their PICU stay. The association between fluid overload and morbidity in children who are critically ill has gained increasing attention in the last decade. Arikan et al³²  reported positive fluid balance to be associated with worse oxygenation index, longer duration of ventilation, and longer ICU and hospital length of stay. Consistent with the findings of Arikan et al,³²  patients who were underweight in our study (in the overall but not in the sepsis cohort) also had higher ICU length of stay, and more of them needed mechanical ventilation support. To evaluate whether a higher rate of fluid overload was directly responsible for the higher mortality rate among the underweight group, we ran sensitivity analyses, but they failed to establish a clear association between the two. However, we believe that a closer look at the relationship between fluid accumulation and underweight status is warranted.

The large prevalence of underweight status on admission to the PICU, approaching 30%, was unexpected and highlights a potential important ICU population. Data on underweight prevalence and trends in children are scarce. Prince et al,³³  in a prospective single-center study from the United Kingdom, used a different anthropometric measurement (weight for age) and also reported an overrepresentation of patients in the PICU at the extremes, with 18% of patients having a weight for age >2.5 SDs below the general UK population mean. As far as we know, in our study, we are the first to estimate the prevalence of underweight in pediatric patients who are critically ill in a multinational cohort. The reason for the high prevalence of underweight on ICU admission may be related to the higher prevalence and burden of baseline comorbidities in this group, as reflected in this study.

The key strength of this study is our use of detailed, prospectively collected data of children and young adults from >30 PICUs worldwide. The amount of data provided enabled us to generate multilayered analyses to explore associations between weight status and short-term outcomes while avoiding possible confounders and reducing statistical bias. The study is not without limitations. First, we used US-based cutoffs to define weight categories in different populations, which may not always be appropriate given known differences in body habitus, particularly in East Asian countries.³⁴  Because ∼80% of the data from this study came from North America, however, using cutoffs suitable for this population should decrease the likelihood of misclassification. Second, this is a secondary analysis of the AWARE database, which was not originally designed to investigate weight categories. Many of the AWARE subjects did not have sufficient data (missing height, weight, or sex) for inclusion in our analysis. Therefore, although we still maintained a large sample size, it is possible that bias may occur for patients who have complete data included. Finally, the classification of sepsis was not explicitly defined in the AWARE study protocol during patient enrollment; however, we suspect that most, if not all, sites used the international pediatric sepsis consensus conference definitions for sepsis published by Goldstein et al³⁵  in 2005. Also, admission diagnosis of sepsis, rather than discharge diagnosis, was captured to meet the AWARE study goals; hence, we suspect an underestimation of the actual prevalence of sepsis in this cohort (missing subjects who were diagnosed with sepsis during their ICU stay). We also did not have data regarding specific pathogens for patients who were septic.

In this study, we found that patients who are underweight make up a significant proportion of all patients in the PICU. Patients who are critically ill and underweight have a higher short-term mortality rate, higher morbidity, and greater burdens of comorbidities. More work is necessary to understand and explore mechanisms that affect mortality and morbidity among patients who are underweight during critical illnesses.

A complete list of investigators in the AWARE study is provided in the Supplemental Information.

Dr Kaddourah’s Pediatric Acute Care Nephrology and Dialysis Fellowship at Cincinnati Children’s Hospital Medical Center was supported by an educational grant from Gambro Renal Products.

AKI

acute kidney injury

AWARE

Assessment of Worldwide Acute Kidney Injury

Renal Angina

and Epidemiology

CI

confidence interval

IQR

interquartile range

KDIGO

Kidney Disease: Improving Global Outcomes

OR

odds ratio

SOI

severity of illness