Red Blood Cell Distribution Width and Pediatric Community-Acquired Pneumonia Disease Severity

This study used data from the Centers for Disease Control and Prevention’s Etiology of Pneumonia in the Community (EPIC) study,²⁵  which was a prospective, multicenter, population-based study of patients hospitalized with clinical and radiographic confirmation of CAP, conducted between January 2010 and June 2012. This study was approved by the institutional review board of the participating hospitals and Centers for Disease Control and Prevention.

As part of an ongoing project to supplement EPIC data with follow-up information on subsequent hospitalizations and vital status of the Tennessee sites, pediatric patients (<18 years old) admitted to Monroe Carell Jr. Children’s Hospital at Vanderbilt University Medical Center (Nashville, TN) and Le Bonheur Children’s Hospital (Memphis, TN) and enrolled in the EPIC study were included. All enrolled patients satisfied the following selection criteria for CAP: (1) evidence of acute infection (fever, chills, hypothermia, leukocytosis, leukopenia, or altered mental status); (2) evidence of acute respiratory illness (new cough or sputum, chest pain, dyspnea, tachypnea, abnormal lung exam, respiratory failure); and (3) chest radiograph changes consistent with pneumonia within 72 hours of admission. Exclusion criteria for the EPIC study included alternative diagnosis for a respiratory disorder, recent hospitalization, and immunocompromised status, as previously described.²⁵  The current study also excluded patients without an obtained RDW measurement during their hospital stay.

The earliest RDW values obtained as part of a complete blood count panel within 24 hours of hospital presentation were collected for each patient.

In-hospital pediatric CAP outcomes were classified into 2 mutually exclusive categories: (1) severe (defined as requiring ICU admission, invasive mechanical ventilation or vasopressor support, or resulting in in-hospital death); and (2) moderate (hospital admission without severe criteria). Patients were classified according to the most severe condition during their hospital stay.

Relevant covariates included the most important predictors from our previously published prognostic model demonstrating high accuracy⁵ : patient age, initial heart rate (HR), initial respiratory rate (RR), initial temperature (in Fahrenheit), initial systolic blood pressure (SBP), presence of altered mental status, presence of chest indrawing, chest radiograph pattern (single-lobar consolidation; multilobar consolidation; pleural effusion; other), PaO₂/FiO₂ ratio (P/F ratio; derived from the SpO₂/FiO₂ ratio, a proxy for hypoxemia).^26,27  The current study also included patient sex, race, and baseline home oxygen requirement.

Pearson’s χ² and Kruskal-Wallis tests were performed for descriptive analysis of patient characteristics on the basis of RDW quintiles. Multivariable logistic regression was used to assess the association between RDW and CAP outcomes, controlling for the study covariates. To acknowledge age-based variabilities in physiologic variables, HR, RR, and SBP were modeled including individual interaction terms with patient’s age.⁵  Regression analysis used RDW as exposure: (1) categorized into quintiles, and (2) as a continuous exposure variable, by using restricted cubic spline functions (5 knots) to allow for nonlinearity. Multiple imputations by chained equations (10 imputed data sets) accounted for covariates with missing values. A dominance analysis was conducted to compare the relative influences of RDW and other variables commonly associated with disease severity. Dominance analysis is an extension of multiple regression and examines all possible combinations of regression predictors to measure their relative importance on an outcome variable via pairwise comparisons.^28,29  A posthoc sensitivity analysis explored whether inclusion of hematocrit values would impact the observed associations. Moreover, a separate sensitivity analysis excluded patients with known hematologic disorders. Finally, exploratory subgroup analyses were conducted, grouping patients on the basis of age (above versus below median), sex (female versus male), race (White versus Black versus other), and hospital site (Nashville versus Memphis), to examine whether the associations between RDW quintiles and CAP outcomes differed between subgroups. All tests were 2-sided, and P values ≤ .05 were considered statistically significant. Analysis was conducted by using Stata 16 (StataCorp, College Station, TX).

In total, 1775 children were recruited in Nashville or Memphis, TN with radiographic confirmation of CAP as part of the EPIC study. After excluding 316 patients without measured RDW values, 1459 pediatric patients (82.2%) were ultimately included in this study. The median age of these patients was 29 months (interquartile range [IQR]: 12–73); 639 patients (44.3%) were female, and 731 patients (50.1%) were Black. The most common comorbidity was asthma (542 patients; 37.1%). Median length of hospital stay was 3 days (IQR: 2–4), and a severe CAP outcome was experienced by 289 patients (19.8%). In comparison, in the group of 316 patients excluded for not having RDW measurements, which were a younger group of children (median age 22 months; IQR: 11–59), there was a lower percentage of severe outcomes (14.6%; P = .031) and a shorter median hospital stay (2 days; IQR, 1–3; P < .001).

The median RDW value for the study population was 13.3% (IQR: 12.5–14.3), which is within the normal range of RDW commonly described for children of 11.3% to 15.0% (median, 13.2, SD, 0.9).¹⁶  In unadjusted analyses, higher RDW quintiles were associated with younger age (P < .001) and public insurance status (P = .037). RDW values also varied between race groups (P < .001) and treatment sites (P < .001; Table 1). There were no differences in the time interval from disease onset to hospitalization among RDW quintile groups.

In total, 8 study covariates in our data set had missing values that required accounting for by multiple imputations. These variables and their associated missing value were systolic BP (109 missing), white blood cell count (66), altered mental status (29), P/F ratio (18), RR (7), chest indrawing (5), HR (1), and temperature (1). Using the remaining complete covariates, 10 multiple imputed data sets were generated and used in subsequent multivariable regression analyses.

The proportion of patients with severe CAP outcomes varied between RDW quintile groups (Q1–Q5): the first quintile had 13.4% of patients experience severe CAP, whereas the fourth quintile had 26.2% (Table 1). In multivariable analyses, RDW quintiles were significantly associated with pneumonia severity outcomes (P = .023). In comparison with the lowest RDW quintile (Q1), the odds ratio (95% CI) for severe CAP in subsequent quintiles were Q2: 1.20 (0.72–1.99); Q3: 1.28 (0.76–2.14); Q4: 1.69 (1.01–2.82); Q5: 1.25 (0.73–2.13), with a proportion test for trend P = .008 (Table 2). In the sensitivity analysis that additionally accounted for hematocrit values, the resulting estimates were largely consistent with our primary analysis: Q2: 1.43 (0.86–2.37); Q3: 1.64 (1.00–2.72); Q4: 2.26 (1.39–3.66); Q5: 1.66 (0.99–2.76). In a separate sensitivity analysis, patients with evidence of hematologic disorders (n = 76) were excluded. Findings from this analysis were consistent with the findings from the primary analysis: Q2 = 1.44 (0.87–2.37); Q3 = 1.54 (0.92–2.55); Q4 = 2.27 (1.40–3.69); Q5 = 1.80 (1.07–3.04); test for trend P = .003.

In a separate analysis, RDW values assessed as a continuous variable were modeled with restricted cubic splines and demonstrated a significant association with CAP severity via multivariable analysis (P = .027). The sensitivity analysis accounting for hematocrit values was also consistent (P = .024). The association between continuous RDW and CAP severity was nonlinear: the risk of severe CAP increased with higher RDW values up to 15% and then appeared to plateau at greater RDW values, although the precision of estimates for those greater values was limited (Fig 1).

The relative strengths of influence between RDW and the other 12 study covariates on CAP severity were compared. RDW placed near the middle of the list of predictors, demonstrating a stronger predictive association than patient age, race, sex, home oxygen requirement, RR, SBP, or chest radiograph pattern. RDW had a lower predictive association than P/F ratio, chest indrawing, altered mental status, HR, and temperature (Table 3).

Subgroup analyses did not demonstrate statistically significant differences in the association between RDW quintiles and CAP severity by age (P = .20), sex (P = .79), race (P = .81), or hospital site (P = .20) groups(Fig 2).

Without a widely accepted, validated, reliable, and practical pneumonia risk assessment tool for children, pediatric clinicians often rely on clinical judgement for CAP management.³⁰  Patient biomarkers could provide objective data to help stratify the risk of pneumonia severity and guide clinical evaluations.⁶  This study demonstrated an association between presenting RDW values and subsequent pediatric CAP severity. Higher RDW values were associated with an increased risk of requiring ICU admission, mechanical ventilation, vasopressor support, or death. Routinely collected and widely available, RDW is a convenient biomarker that can provide objective information for pediatric CAP evaluations.

RDW has demonstrated prognostic potential for some diseases in adults and children, but the precise biological mechanism remains unestablished.^18–24  One potential explanation for higher RDW values in more severe pediatric CAP is related to an enhanced inflammatory response with worsening pneumonia. Both hypoxemia and oxidative stress seen in severe pneumonia can lead to erythrocyte size variations and subsequently higher RDW values.²¹  RDW is also highly correlated with other acute phase reactants, including procalcitonin, which has been reported in association with CAP severity and bacterial infections.¹⁵  RDW increases in response to inflammatory cytokines occur even without any detectable changes in hemoglobin levels.²⁰  In the context of pediatric pneumonia, an elevated RDW could be a marker of chronic health conditions (eg, anemia, nutritional deficiency) or an underlying pathophysiological process (eg, hypoxemia, impaired erythropoiesis), which would predispose the patient to more severe infections.

Our study offers the potential for RDW to be included as an objective measurement in clinical evaluations of children with CAP to assist in determining the appropriate site of treatment. Most CAP guidelines strongly suggest incorporating at least 1 objective data point into clinical judgements, to improve the accuracy of severity risk estimates.^30–33  Validated CAP assessment tools in adults that use objective parameters, such as pneumonia severity index and CURB-65, have been effective in identifying low-risk patients and reducing the rate of unnecessary hospitalizations.^32–37  Pediatric providers, in contrast, do not yet have a similar level of guidance. Even chest radiograph changes that typically serve as the diagnostic gold standard in adult CAP are less reliable in children.³⁸  There is a need for additional, reliable objective measurements for pediatric CAP evaluations, and RDW can help.

There are over 160 000 pediatric CAP admissions each year resulting in over 1 billion dollars of health care spending.^39,40  Strategies to better assess the risk of severe outcomes and inform the clinical management of children with CAP are urgently needed.^7–9  Previous studies have reported that RDW values were associated with severe outcomes among critically ill children.^16,20,21  Our results indicate that RDW values combined with clinical risk factors could contribute information to the pediatric CAP management decision-making process as an inexpensive, rapid, and highly reproducible biomarker. Another algorithm, the modified PIRO score⁴¹  (predisposition, insult [hypoxia, hypotension, bacteremia], response [eg, multilobar, complicated], and organ dysfunction) has been reported to have high predictive ability for pediatric CAP. However, the original study describing the predisposition, insult, response, and organ dysfunction algorithm was retrospective, and further prospective or systematic validations have not yet been reported. Our study findings suggest that available RDW values in combination with other established risk factors could provide useful information for assessments of the risk of severe disease among children hospitalized with pneumonia. This and other previously developed prognostic models are undergoing additional evaluation and validation.⁵ 

Our study findings should be interpreted with consideration of several limitations. First, this study only examined in-hospital CAP outcomes, which might skew toward more severe patients, despites the reportedly high percentage of low-risk patients routinely admitted.³²  This is further underscored by our examination of participants excluded for lacking RDW measurements, which seemed to have had milder disease and of a younger age than those with RDW measurements available. Secondly, our in-hospital outcomes were only reported in 2 categories, which may oversimplify the spectrum of experiences in children hospitalized with pneumonia. Nevertheless, our study used strict CAP definitions and was based on a well-characterized, prospective, multicenter cohort of patients, which decreases the concern for potential misclassifications. Since RDW was collected retrospectively at the same time as other outcomes, only an association can be reported here, whereas future work is ongoing to evaluate the predictive nature of RDW. Although our estimates demonstrated a clear nonlinear increase in the risk of severe disease with higher RDW values, there were relatively few observations of very high RDW values, which limits the precision of our estimates for higher RDW values. Moreover, although our selection of regression covariates was informed by existing literature and our sensitivity analysis accounted for hematocrit, the possibility of residual confounding cannot be definitively ruled out. Our findings, however, are consistent with previous reports, mainly conducted in adults, showing an association between higher RDW values and negative pneumonia outcomes.^22–24  Future studies could examine RDW in children with CAP receiving ambulatory care, and further studies in another pediatric population will be helpful to further assess our findings.

RDW is independently and positively associated with an increased risk of severe outcomes in children hospitalized with community-acquired pneumonia. RDW can serve as an objective data point for assessing pneumonia risk severity in children.