For clinicians weighing how long a sick child will need to stay in the hospital, a handful of inexpensive inflammation scores calculated from an ordinary complete blood count have become increasingly tempting shortcuts. A new study from Vietnam suggests that, at least for children hospitalized with iron deficiency anemia, these scores add almost nothing of value. In a cohort of 1,004 pediatric patients treated at a tertiary referral center in the Mekong Delta, none of five widely used blood-count-derived inflammatory indices provided meaningful incremental information about hospitalization duration beyond routinely available clinical characteristics and C-reactive protein, the researchers report in the journal Discover Pediatrics.
The study, led by Quoc Trung Ly, Minh Phuong Nguyen, and J. Luis Espinoza, was a retrospective observational cohort conducted at Soc Trang Hospital for Women and Children. The investigators reviewed records of children aged 2 months to 15 years who were hospitalized between January 2024 and June 2025 and who had anemia together with laboratory evidence of iron deficiency. Children were not excluded because of infection or systemic inflammation, a deliberate choice meant to reflect the heterogeneous reality of a busy pediatric inpatient service rather than a narrowly defined disease population. After excluding two records with incomplete blood count data, the final analytical cohort comprised 1,004 children.
Iron deficiency anemia remains the most common nutritional disorder worldwide, affecting an estimated two billion people, with young children among the most vulnerable. The burden is particularly heavy in Southeast Asia, where nutritional factors, infectious diseases, and socioeconomic conditions combine to sustain high rates of deficiency. Because iron metabolism is tightly intertwined with immune and inflammatory pathways, hospitalized children frequently present a complicated picture in which true iron deficiency, anemia of inflammation, and active infection overlap. Ferritin, the standard marker of iron stores, is itself an acute-phase reactant that rises during systemic inflammation, further complicating interpretation in this setting.
The indices under scrutiny were calculated from admission blood counts: the neutrophil-to-lymphocyte ratio, the monocyte-to-lymphocyte ratio, the platelet-to-lymphocyte ratio, the systemic immune-inflammation index, and the systemic inflammatory response index. Each integrates counts from different hematologic lineages and has previously been reported to predict outcomes in infections, malignancies, and inflammatory diseases. The primary outcome in the new study was prolonged hospitalization, defined as a length of stay of at least seven days, a threshold that approximately corresponded to the upper quartile of the distribution. The median stay in the cohort was five days, with an interquartile range of four to six days, and 210 children, or 20.9 percent, experienced prolonged hospitalization.
The analytical design was deliberately sequential. A first model incorporated age, sex, ethnicity, admission department, and broad primary-diagnosis category. Ferritin was added next, and C-reactive protein after that. Each blood-count-derived index was then evaluated individually on top of the full model, a strategy adopted because the indices share constituent hematologic variables and showed substantial collinearity. All sequential comparisons were performed on the common set of 816 children with complete C-reactive protein data, ensuring that changes in model fit reflected the addition of predictors rather than shifts in the analyzed sample. Odds ratios for the laboratory measures were expressed per one-standard-deviation increase in their log-transformed values.
The results were unambiguous. In univariable comparisons, children with prolonged stays had significantly higher admission CRP and ferritin concentrations than those discharged within seven days, but no clinically meaningful differences emerged for any of the five derived indices. In the sequential models, admission department and underlying diagnostic category accounted for much of the variability in hospitalization duration, underscoring the weight of clinical case mix in a cohort admitted for a broad spectrum of infectious and noninfectious conditions. Ferritin was not independently associated with prolonged hospitalization after adjustment, with an odds ratio of 1.00 per standard deviation and a confidence interval spanning 0.84 to 1.18. Adding CRP significantly improved model fit, and CRP emerged as the strongest independent laboratory predictor, while its inclusion attenuated the ferritin association.
Crucially, none of the five indices provided statistically significant incremental prognostic information beyond the clinical model that already included CRP. The systemic immune-inflammation index showed a borderline improvement in model fit, but the effect did not reach conventional statistical significance. Repeated cross-validation, using five folds repeated twenty times, yielded mean area under the receiver operating characteristic curve values of only about 0.57 to 0.58 across all models, a level of discrimination only modestly better than chance. Sensitivity analyses using alternative thresholds of six, eight, and ten days, as well as continuous length of stay, generally supported the direction of the primary findings, with CRP remaining significantly associated with stays of at least six days and with continuous stay, though estimates were less precise at the higher, less frequent thresholds.
The investigators also pursued two exploratory approaches. Machine-learning classifiers, including penalized logistic regression, random forest, and gradient boosting, were trained on the 816 complete cases and evaluated with internal holdout testing and stratified cross-validation. Although the random forest and gradient boosting models achieved excellent apparent performance on training data, their cross-validated discrimination was only marginally greater than chance, a pattern the authors highlight as a cautionary lesson about overfitting when models are judged solely on training performance. Separately, unsupervised k-means clustering of the five indices identified three inflammatory phenotypes with progressively higher index values, but CRP concentrations did not differ significantly across clusters, median length of stay was five days in all three, and the frequency of prolonged hospitalization was statistically similar, leaving the phenotypes without prognostic relevance.
The authors offer several explanations for why these indices, so often reported as useful elsewhere, fell short here. The cohort was highly heterogeneous, and disease-specific factors may simply dominate hospitalization duration over baseline inflammatory ratios, an interpretation supported by the strong associations with admitting department and primary diagnosis. Moreover, iron deficiency itself can distort leukocyte and platelet profiles: reactive thrombocytosis is common in iron deficiency anemia and may inflate indices such as the platelet-to-lymphocyte ratio and the systemic immune-inflammation index without reflecting clinically meaningful inflammation. CRP, by contrast, may more accurately capture the severity of the underlying illness driving admission, and the attenuation of ferritin’s association after adjustment for CRP suggests ferritin functioned primarily as an inflammatory marker in this cohort rather than as an indicator of iron status alone.
The study carries important caveats. It was conducted at a single tertiary-care center, which may limit generalizability, and the wide range of underlying diseases means residual confounding by disease severity is likely. Detailed information on treatment exposure and timing, including iron supplementation, transfusion, antimicrobial therapy, and anti-inflammatory treatment, was not consistently available and could not be incorporated into the models, nor were standardized measures of illness severity, fever duration, nutritional status, or detailed comorbidity burden. Because the exact timing of blood sampling relative to therapeutic interventions could not be verified for every patient in the retrospective design, some misclassification is possible. The authors also note that no independent external validation cohort was available for the machine-learning analyses.
Nevertheless, the practical message is clear. Even though CRP remained independently associated with prolonged hospitalization, the cross-validated discrimination of the models stayed near 0.57 to 0.58, indicating that admission inflammatory biomarkers capture only a small component of the many factors that determine how long a child remains hospitalized. The authors conclude that blood-count-derived inflammatory indices provide little incremental prognostic information beyond clinical characteristics and CRP in this population and should not be used as standalone tools for risk stratification. Length of stay, they emphasize, is inherently multifactorial, shaped by diagnosis, severity, treatment requirements, and institutional factors, and no inexpensive admission biomarker is likely to replace careful clinical judgment in anticipating which children with iron deficiency anemia will face an extended stay.
Subject of Research: Prognostic value of CBC-derived inflammatory indices for hospitalization duration in children with iron deficiency anemia
Article Title: Inflammatory biomarkers and hospitalization duration in children with iron deficiency anemia
Article References: Ly, Q. T., Nguyen, M. P., & Espinoza, J. L. (2026). Inflammatory biomarkers and hospitalization duration in children with iron deficiency anemia. Discover Pediatrics, 1(1), Article 8. https://doi.org/10.1007/s44571-026-00010-8
Image Credits: AI Generated
DOI: 10.1007/s44571-026-00010-8
Keywords: iron deficiency anemia, pediatrics, hospitalization duration, C-reactive protein, ferritin, neutrophil-to-lymphocyte ratio, systemic immune-inflammation index, inflammatory biomarkers, retrospective cohort study, Vietnam, length of stay, risk stratification
News Source: Harold Sullivan. (October 28, 2026). Common Blood Inflammation Scores Fail to Predict Long Hospital Stays in Children with Iron Deficiency Anemia. Scienmag.
