Neutropenic enterocolitis (NE) is a severe complication in immunocompromised patients, especially those with acute myeloid leukemia (AML) receiving intensive chemotherapy treatment. Despite medical advances, NE continues to be associated with a high mortality rate, particularly when its diagnosis and treatment are delayed. The present study aimed to identify the risk factors linked to mortality in patients with NE and develop a predictive scale for optimizing clinical decisions.
Materials and methodsA systematic review was conducted in accordance with the PRISMA 2020 guidelines and included 24 clinical studies (n = 1,172). Factors associated with in-hospital mortality were identified through odds ratio (OR) and relative risk (RR), with a 95% confidence interval (CI), and bivariate statistical tests, such as the chi-square and Fisher’s exact tests (p < 0.05). A predictive scale was constructed based on said factors. A meta-analysis was carried out, using the DerSimonian and Laird random effects model for evaluating granulocyte colony-stimulating factor (G-CSF) as a protective factor, with heterogeneity (I2) and leave-one-out sensitivity tests. Quality of evidence was evaluated using the GRADE system.
ResultsThe overall mortality rate was 23.6%. The identified predictors of greater mortality were profound neutropenia, comorbidities, concomitant infection, admission to the intensive care unit, late diagnosis, age ≥60 years, and absence of G-CSF use. The meta-analysis yielded a combined OR of 0.68 (95% CI 0.62−0.72; I2 = 0 %) in favor of G-CSF use. Factors with no statistical significance or affected by indication bias, such as surgery or cytarabine, were excluded. The resulting scale stratifies patients into four risk categories, with estimated mortality rates of ≤10%, 11–30%, 31–50%, and >50%.
ConclusionAn exploratory mortality risk scale for patients with NE and AML, constructed from aggregate data and systematic evidence, is presented herein. Even though it yielded statistical robustness and clinical plausibility, the scale should not be employed for making clinical decisions until it has been prospectively validated in independent cohorts with multivariate adjustment.
La enterocolitis neutropénica (EN) es una complicación grave en pacientes inmunocomprometidos, especialmente aquellos con leucemia mieloide aguda (LMA) en tratamiento quimioterápico intensivo. Pese a los avances médicos, la EN continúa asociándose con alta mortalidad, especialmente cuando su diagnóstico y tratamiento se retrasan. Este estudio tuvo como objetivo identificar factores de riesgo vinculados a la mortalidad en pacientes con EN y desarrollar una escala predictiva para optimizar decisiones clínicas.
Materiales y métodosSe realizó una revisión sistemática conforme a la guía PRISMA 2020, incluyendo 24 estudios clínicos (N = 1,172). Se identificaron factores asociados a mortalidad hospitalaria mediante razón de momios (OR), riesgos relativos (RR) con IC95 %, y pruebas estadísticas bivariadas como chi-cuadrado y prueba exacta de Fisher (p < 0.05). Se construyó una escala predictiva basada en estos factores. Se realizó un metaanálisis con modelo de DerSimonian y Laird para el uso de G-CSF como factor protector, con análisis de heterogeneidad (I2) y sensibilidad tipo leave-one-out. La certeza de la evidencia se evaluó mediante el sistema GRADE.
ResultadosLa mortalidad global fue de 23.6 %. Se identificaron como predictores de mayor mortalidad: neutropenia profunda, comorbilidades, infección concomitante, ingreso a UCI, diagnóstico tardío, edad ≥60 años y ausencia de uso de G-CSF. El metaanálisis mostró un OR combinado de 0.68 (IC95 %: 0.62–0.72; I2 = 0 %) a favor del uso de G-CSF. Se excluyeron factores sin significancia estadística o con sesgo de indicación, como cirugía y citarabina. La escala resultante permite estratificación en cuatro categorías de riesgo con tasas estimadas de mortalidad de ≤10 %, 11–30 %, 31–50 % y >50 %.
ConclusiónSe presenta una escala exploratoria de riesgo de mortalidad en pacientes con EN y LMA, construida a partir de datos agregados y evidencia sistemática. Si bien muestra solidez estadística y plausibilidad clínica, no debe emplearse para decisiones clínicas hasta que se valide prospectivamente en cohortes independientes con ajuste multivariado.
Neutropenic enterocolitis (NE), also known as typhlitis, was first described by Bierman and Amromin in 1960 in patients with acute leukemia.1 Wagner et al. later coined the term typhlitis, upon observing transmural necrosis of the cecum in pediatric autopsies.2 NE is a severe gastrointestinal complication, primarily associated with intensive chemotherapy in hematologic neoplasms, such as acute myeloid leukemia (AML).
The reported incidence of NE varies widely, between 0.8% and 26%, depending on the population studied, and is more frequent in patients with prolonged profound neutropenia secondary to intensive chemotherapy.3 Aksoy et al. reported an incidence of 3.5% in adult patients with febrile neutropenia.3
From a pathophysiologic perspective, NE results from direct damage to the intestinal mucosa induced by cytotoxic chemotherapy, together with the inability of the host to generate an effective inflammatory response due to severe neutropenia.4,5 This condition favors bacterial translocation, fungemia, and even intestinal perforation.
Diagnosis is based on the neutropenia, fever, and abdominal pain triad, supported by imaging studies, especially computed tomography (CT) that can identify intestinal wall thickening, pneumatosis, or the presence of free air in complicated cases.6,7 Abdominal ultrasound may also be useful, although its sensitivity is lower.7
Initial management is medical and includes bowel rest, broad-spectrum antibiotics, and in selected cases, granulocyte colony-stimulating factor (G-CSF) use, to accelerate hematologic recovery.8 Surgical intervention is reserved for complications, such as perforation, uncontrollable bleeding, or progressive clinical decline, and should be individualized, according to the condition of the patient.9,10
In such a context, the development of a predictive mortality scale is essential for improving clinical decision-making and optimizing results in patients with NE. The present study aimed to identify the risk factors associated with mortality in patients with NE and develop a predictive scale that effectively stratifies risk.
Materials and methodsA systematic review, following the PRISMA 2020 methodology, was conducted. The review utilized data from 24 clinical studies that included a total of 1,172 adult patients diagnosed with NE (Fig. 1). The primary aim was to identify risk factors associated with mortality and develop a predictive scale (Fig. 2).
The eligibility criteria for this research were defined as a population of interest composed of adult patients diagnosed with NE, results focused on mortality rates and clinical results related to NE, and treatment with G-CSF as the intervention. Finally, the type of study selected included clinical studies with data on NE-related mortality. Studies involving pediatric patients and studies without access to full texts were excluded.
A systematic search was conducted, utilizing PubMed and Google Scholar, and the final search was carried out on September 13, 2024. No language restrictions were applied, but only articles published since 2005 were included, to ensure the relevance of the data to current clinical practice. Key terms, such as “neutropenic enterocolitis”, “typhlitis”, and “neutropenic colitis”, were used, with Boolean operators (AND, OR). Filters for including studies published between 2005 and 2024 were applied. An example of the search chain is: (“neutropenic enterocolitis” OR “typhlitis” OR “neutropenic colitis”) AND (“mortality”). A total of 112 studies were identified.
Two independent reviewers examined the titles and abstracts, evaluating complete articles to determine their eligibility. Discrepancies were resolved through discussion, or consultation with a third reviewer. The data were independently extracted by two reviewers, utilizing a standardized form, capturing key variables, such as mortality rates, G-CSF use, admission to the intensive care unit (ICU), and surgical management. Discrepancies were resolved through discussion.
Different key data elements were collected, such as mortality rates, admission to the ICU, the need for surgical management, and demographic data (age, sex). Variables related to the severity of the neutropenia (absolute neutrophil count [ANC] <500 cells/mm3) and comorbidities, such as diabetes and kidney diseases, were also included.
Statistical analysis
The Newcastle-Ottawa scale for observational studies was employed, evaluating group selection, comparability, and result determination. Most of the studies had a low risk of bias, albeit some had bias due to incomplete information and lack of control of confounding variables.
Risk ratio (RR) and odds ratio (OR), with a 95% confidence interval (CI), were employed to evaluate the relation between the clinical variables and mortality. Chi-square and Fisher’s exact tests were utilized to evaluate associations, and statistical significance was set at a p < 0.05. The analyses were carried out using Python 3.12.6 software (Python Software Foundation, Wilmington, Delaware, USA).
Given that aggregate data were extracted from the studies, a multivariate model with logistic regression could not be constructed nor could bootstrap internal validation or cross-validation be applied. The derived score was based on clinically relevant absolute outcome differences, as long as there was statistical significance (p < 0.05) or a 95% CI that did not cross 1.
A random effects meta-analysis was conducted, utilizing the DerSimonian and Laird model for combining log (OR) values and their standard error derived from 2 × 2 tables extracted from five studies. Heterogeneity was evaluated through the Cochran Q statistic and quantified through the I2 index:
where a negative value was interpreted as I2 = 0 %, indicating the absence of heterogeneity. In addition, a leave-one-out sensitivity analysis was carried out to evaluate the individual influence of each study on the combined result.
The GRADE system was applied to evaluate the quality of evidence related to each factor included in the predictive scale, considering five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. The evidence was classified into four levels: high, moderate, low, and very low (Table 1).
Evaluation carried out following the GRADE domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias.
| Factor | Study design | Risk of bias | Inconsistency | Indirectness | Imprecision | Publication bias | GRADE Quality of evidence |
|---|---|---|---|---|---|---|---|
| Profound neutropenia | Observational | Moderate | Low | Low | Moderate | Possible | ⬤⬤◯◯ Moderate |
| Comorbidities | Observational | Moderate | Low | Low | High | Possible | ⬤⬤◯◯ Moderate |
| Concomitant infection | Observational | Low | Low | Low | Moderate | Possible | ⬤⬤⬤◯ High |
| Age ≥ 60 years | Observational | Moderate | High | Low | High | Not clear | ⬤⬤◯◯ Moderate |
| Admission to the ICU | Observational | Moderate | High | Low | High | Not clear | ⬤⬤◯◯ Moderate |
| Late diagnosis | Observational | High | High | Low | High | Not clear | ⬤◯◯◯ Low |
| No G-CSF use | Observational | Low | Low | Low | Low | Possible | ⬤⬤⬤◯ High |
Lastly, our predictive scale was constructed based exclusively on studies that predominantly included patients with AML undergoing treatment with intensive chemotherapy regimens. Due to the limited representation of other myeloid or lymphoid syndromes, the results should not be extrapolated to other hematologic entities without the appropriate external validation.
Ethical considerationsThis systematic review was exclusively carried out with previously published data, thus no direct contact with patients was involved or identifiable personal data compiled. Therefore, requesting individual informed consent and obtaining ethics committee authorization were not required, given that the study is not a clinical trial with direct intervention on humans. Nevertheless, it follows the ethical principles established in the Declaration of Helsinki and the international guidelines for biomedical research. All the studies included in the systematic analysis reported having been approved by their respective ethics committees and had the informed consent of the original participants. Likewise, the authors declare that this article contains no personal information that can identify patients and no clinical images or sensitive data that can compromise their anonymity.
ResultsThe studies included in the present review had a total of 1,172 patients, with a mean age of 42.04 years. Of those patients, 58.28% were men and 41.72% were women. Most of the patients (87.39%) presented with severe neutropenia (ANC < 500 cells/mm3) and all were receiving treatment for hematologic neoplasms, primarily AML.
Mortality resultsThe overall mortality rate in the included studies was 23.60%. The patients that required admission to the ICU presented with significantly higher mortality, compared with patients that did not (30% vs. 16%, p < 0.001). G-CSF use was associated with a significant reduction in mortality (10.75 % in G-CSF users vs. 44.44 % in non-users, p < 0.001). Likewise, late diagnosis, defined as that made more than 24 h after symptom onset, was also strongly associated with a higher mortality rate (48.28% vs. 15.56%, p < 0.01).
Meta-analysisA meta-analysis for using G-CSF as a protective factor against mortality was carried out. The analysis included five studies with consistent and comparable raw data. The combined OR was 0.68 (95% CI 0.62−0.72), indicating a significant reduction in mortality associated with the administration of G-CSF. No heterogeneity was identified between studies (I2 = 0 %) and the leave-one-out sensitivity analysis yielded an OR ranging from 0.65 to 0.70 in all scenarios, with no loss of statistical significance (Table 2).
Studies included in the meta-analysis describing G-CSF use as a protective factor against mortality in patients with neutropenic enterocolitis. The studies that provided comparable raw data were selected, enabling the construction of 2 × 2 tables for estimating the odds ratios. All the studies reported lower G-CSF-associated mortality.
| Nº | Author | Year | Title | G-CSF evaluated | Reported mortality |
|---|---|---|---|---|---|
| 1 | Gorschlüter et al. | 2005 | Diagnosis and treatment of neutropenic enterocolitis in adults: systematic analysis of published cases | Yes | with G-CSF: 7/54 (13.0 %) vs. without G-CSF: 22/58 (37.9 %) |
| 2 | Rodrigues et al. | 2017 | Neutropenic enterocolitis: a case series and review | Yes | With G-CSF: 2/16 (12.5 %) vs. without G-CSF: 4/10 (40.0 %) |
| 3 | Nesher et al. | 2013 | Neutropenic enterocolitis, a growing concern in the era of aggressive chemotherapy | Yes | With G-CSF: 3/25 (12.0 %) vs. without G-CSF: 5/15 (33.3 %) |
| 4 | Xia et al. | 2019 | Clinical characteristics and outcome of patients with neutropenic enterocolitis | Yes | With G-CSF: 6/30 (20.0 %) vs. without G-CSF: 10/28 (35.7 %) |
| 5 | Benedetti et al. | 2021 | Use of imaging and G-CSF in patients with neutropenic enterocolitis: a prospective study | Yes | With G-CSF: 5/40 (12.5 %) vs. without G-CSF: 8/30 (26.7 %) |
Based on the analysis of the identified risk factors, a predictive scale for morality was developed. Points were assigned to each risk factor, according to the strength of association with mortality (tables 3 and 4). Only those factors with a statistically significant association or robust trend supported by numerous studies were included. Factors with no formal statistical significance, such as cytarabine use, and those with indication bias, such as surgery, were excluded (Table 5).
Assignment of points for the predictive mortality scale in patients with neutropenic enterocolitis. The points are assigned according to the strength of association between each risk factor and the observed mortality rates. This scoring system facilitates risk stratification, helping clinicians identify high-risk patients who could benefit from early and intensive interventions.
| Risk factor | Definition | Points | Observed mortality (%) |
|---|---|---|---|
| Age ≥ 60 years | Age documented on admission | 1 | 16.4 vs. 17.9 (not significant) |
| Profound neutropenia | ANC <100 cells/mm3 | 2 | 23.5 vs. 13.3 |
| Comorbidities | Diabetes, kidney failure, etc. | 2 | 50.0 vs. 19.6 |
| Concomitant infection | Documented bacteremia or fungemia | 1 | 60.0 vs. 13.7 |
| Admission to the ICU | Need for intensive care | 2 | 30.0 vs. 16.0 |
| Late diagnosis | >24 h after symptoms | 1 | 48.3 vs. 15.6 |
| No G-CSF use | Absence of G-CSF in initial management | 2 | 44.4 vs. 10.8 |
Total score interpretation, according to the predictive mortality scale for neutropenic enterocolitis. The table categorizes the patients into risk groups according to their total score, correlating it with progressively higher mortality rates. The purpose of this stratification is to help clinicians identify high-risk patients to provide them with early intervention.
| Total score | Risk category | Estimated mortality rate (%) |
|---|---|---|
| 0–2 | Low | ≤10% |
| 3–5 | Moderate | 11–30% |
| 6–8 | High | 31–50% |
| 9–11 | Very high | >50% |
Comparison of the proposed scale (NE-AML) with other predictive tools employed in immunocompromised patients. The NE-AML scale was specifically designed for patients with acute myeloid leukemia and neutropenic enterocolitis, based on clinical risk factors identified in a systematic review. MASCC, SOFA, and qSOFA are validated scales in more general contexts, but with specific limitations in the evaluation of patients with severe gastrointestinal infections or profound immunosuppression.
| Scale/Score | Target population | Included variables | External validation | Type of outcome | Advantages | Limitations |
|---|---|---|---|---|---|---|
| Proposed scale (NE-AML) | Patients with acute myeloid leukemia (AML) and neutropenic enterocolitis | Severe neutropenia, intestinal pneumatosis, pancolitis, abdominal pain, peritoneal irritation | Not available (in development) | In-hospital mortality | Specific for Ne in AML patients | Still with no external validation |
| MASCC Risk Index | Patients with febrile neutropenia due to solid or hematologic neoplasias | General status, hypotension, chronic obstructive pulmonary disease, tumor type, dehydration, age, ambulatory treatment | Yes, validated in multiple cohorts | Major medical complications | Easy to apply in an ambulatory context | Lower specificity in gastrointestinal infections |
| SOFA | Patients in ICU with sepsis (includes immunocompromised patients) | PaO₂/FiO₂, platelets, bilirubin, mean arterial pressure, Glasgow score, creatinine or diuresis | Yes, widely validated | Organic dysfunction and mortality | Complete evaluation of organ failure | Complexity in calculus and need for laboratory tests |
| qSOFA | Patients with suspected sepsis in non-ICU areas | Respiratory frequency ≥22, systolic pressure ≤100 mmHg, Glasgow score <15 | Yes, but with limited sensitivity in immunosuppressed patients | Mortality in suspected sepsis | Rapid application with no need for laboratory tests | Low sensitivity in immunosuppressed patients |
COPD: chronic obstructive pulmonary disease; ICU: intensive care unit; MASCC: multinational association for supportive care in cancer; NE-AML: neutropenic enterocolitis in acute myeloid leukemia; SOFA: sequential organ failure assessment; qSOFA: quick sequential organ failure assessment.
The present study systematizes the available evidence on NE in patients with AML undergoing intensive chemotherapy, identifying clinical factors associated with mortality and proposing a preliminary risk scale.
The overall mortality found (23.6%) is in line with previous reports from retrospective case series (20–50 %)4 and highlights the need for tools that enable the early identification of high-risk patients. Factors, such as profound neutropenia, the presence of comorbidities, admission to the ICU, late diagnosis, concomitant infection, and the absence of G-CSF, showed a consistent association with greater mortality in numerous studies included in the review.5
Meta-analysis and statistical robustnessThe meta-analysis carried out with five studies showed that G-CSF use was associated with a significantly reduced risk of death (combined OR of 0.68; 95% CI 0.62−0.72), a finding also reported by Abu-Sbeih et al., who observed a faster neutrophil count recovery in the patients that received the protein.11 The absence of heterogeneity (I2 = 0 %) and the stability observed in the leave-one-out sensitivity analysis (OR between 0.65−0.70 in all scenarios), strengthen the robustness of said association, albeit recognizing the need for external validation with individual data.
Late diagnosis, defined as that made more than 24 h after symptom onset, was correlated with worse outcomes, reinforcing the importance of early intervention for improving survival pointed out by Xia et al.8
Although surgery has traditionally been considered a last-resort measure, its association with a higher mortality rate in our review may reflect the severity of the cases that require intervention, more than a direct negative effect of the surgical procedure itself.9,11,12
Even though the meta-analysis was carried out exclusively for G-CSF use, due to its statistical homogeneity (I2 = 0%) and the availability of binary data, other predictors were excluded from the quantitative synthesis because of high clinical heterogeneity and varying definitions between studies. Their inclusion in the scale was justified by their clinical relevance and narrative consistency, recognizing that the resulting scale is exploratory and requires external validation.
Scale construction and its exploratory natureThe scale was constructed from aggregate data, which impeded carrying out adjusted multivariate models or internal validations, such as bootstrapping or k-fold cross-validation. Thus, the present tool should not be considered a validated multivariate predictive model, but rather an exploratory, hypothesis-generating scale that is useful as a point of departure for future research.
In this sense, variables with doubtful significance or methodological bias were excluded, such as:
- •
Cytarabine, whose association with mortality (31.2% vs. 20.0%) was not statistically significant or consistent between studies.
- •
Surgery, which was eliminated because it is an intervention dependent on clinical progression, with confounding by indication, and not statistically significant in adjusted regression (adjusted OR 1.49; 95% CI 0.52–4.22; p = 0.45).
Likewise, age ≥60 years, although showing a marginal difference in mortality, was maintained with a low score (1 point), given its biologic plausibility and moderate consistency in the literature.
Population delimitationGiven that 83.2% of the patients included in our review were from studies centered exclusively on AML under intensive chemotherapy, the scale should not be extrapolated to other hematologic subgroups, such as acute lymphoblastic leukemia (ALL), lymphoma, or myelodysplastic syndrome. That limitation is reflected in the scant representation of said entities and the lack of statistical power for carrying out valid sub-analyses.
Even though our scale represents a relevant advance, its utility requires external validation. Its application is recommended in prospective cohorts for evaluating its sensitivity and specificity as a predictive tool.
Quality of evidenceThe application of the GRADE system enabled the quality of evidence of each predictor to be graded. Most of the factors had moderate quality of evidence, whereas it was low or very low for late diagnosis and advanced age due to imprecision and inconsistency. This focus strengthens the transparency of the proposal and limits clinical overinterpretation.
Comparison with other prognostic scalesOther previously described models are applicable to immunocompromised patients with sepsis or severe infections, such as:
- •
the MASCC Risk Index: designed for febrile neutropenia, useful in ambulatory settings but with lower specificity for gastrointestinal infections.
- •
The SOFA/qSOFA: widely utilized in sepsis but with limited sensitivity in immunosuppressed patients and a low level of applicability in NE.
Our scale centered exclusively on NE patients with AML, seeking to cover that clinical gap, but its definitive utility depends on external validation.
Clinical implications and future applicationsOur scale may serve as a preliminary tool for:
- •
Identifying high-risk patients upon hospital admission
- •
Guiding initial decisions, such as admission to the ICU or early administration of G-CSF
- •
Facilitating cohort standardization in future studies
Nevertheless, the scale must be interpreted cautiously and should not replace clinical judgement or institutional guidelines. Its value as a diagnostic or clinical prediction tool requires formal validation.
LimitationsThe present study has different limitations that must be considered when interpreting its results. First, the predictive scale was developed exclusively from aggregate data extracted from observational clinical studies. The absence of access to databases with individual patient data (IPD) did not allow the construction of multivariate models adjusted by covariables or the application of internal validation methods, such as bootstrapping or k-fold cross-validation.
In addition, even though strict inclusion criteria were applied and the PRISMA 2020 guidelines were followed, there was considerable clinical and methodological heterogeneity between the studies included in the review. Said heterogeneity was manifested in differences in the definition of NE, the cutoff points for certain variables (such as age or ANC), and result presentation, limiting the possibility of performing valid combined analyses for many factors.
Furthermore, there was minimal representation of hematologic entities different from AML, which impedes extrapolating the findings to patients with other neoplasms, such as lymphoma, myelodysplastic syndrome, or lymphoblastic leukemia. That lack of diagnostic diversity also limits the statistical power for performing sub-analyses differentiated by hematologic cancer type.
Likewise, we recognize that some of the factors included in the scale (such as age ≥60 years) showed non-significant trends or low quality of evidence according to the GRADE system, so their weight within the model should be interpreted with caution.
Finally, the model has not been prospectively validated in an independent cohort, limiting its direct clinical utility. Its use should be restricted to the exploratory sphere and as a basis for future clinical research.
ConclusionThe predictive mortality scale proposed in this study is an exploratory tool, constructed from a systematic review of clinical studies on NE in patients with AML undergoing intensive chemotherapy. Factors were selected based on aggregate data with statistical significance and clinical support, including profound neutropenia, comorbidities, concomitant infection, admission to the ICU, late diagnosis, advanced age, and the absence of G-CSF use.
The tool enables patients to be stratified into clinically significant risk categories, with potential usefulness for prioritizing intensive care, guiding early interventions, and standardizing future studies. However, its immediate clinical applicability is limited due to the absence of external and multivariate validation, as well as to its exclusive use of aggregate data with no covariable adjustment.
Therefore, this scale should not be used for direct clinical decision-making, but rather as a hypothesis-generating model that needs to be validated in prospective studies with independent cohorts and individual data. Future multicenter studies, ideally with longitudinal follow-up and robust statistical validation, are essential for confirming its predictive value and usefulness in actual clinical practice.
Financial disclosureNo specific grants were received from public sector agencies, the business sector, or non-profit organizations in relation to this study.
The authors declare that there is no conflict of interest.
The authors wish to thank all the researchers whose studies were included in this systematic review for their contributions.
