aceClinical Research on Microecological Landscape for Infection Risk Stratification in Newly Diagnosed Patients with Hematological Conditions
Abstract: Introduction Infection is a common and potentially fatal complication during the treatment of hematological diseases, particularly in the context of chemotherapy-induced immunosuppression. The nonselective use of antibiotic prophylaxis in patients with neutropenia in China has persistently accelerated antimicrobial resistance. Early identification of patients at high risk for infection before clinical symptom onset could enable targeted preventive strategies; however, reliable and biologically informed screening…
Doctor Grays Hospital at Elgin - geograph.org.uk - 765088 by Ann Harrison. CC BY-SA 2.0 · https://creativecommons.org/licenses/by-sa/2.0
In a prospective study registered as ChiCTR2100042992, investigators collected plasma for metagenomic next-generation sequencing from 230 newly diagnosed hematological patients before and after chemotherapy and used machine learning to map a complex microecological landscape linked to neutropenia and subsequent infection.
The resulting microorganism-based random forest model demonstrated high predictive accuracy for infection risk by the August 2026 publication date, suggesting potential to target preventive strategies and reduce nonselective antibiotic prophylaxis that accelerates antimicrobial resistance, though real-world clinical deployment and impact on antibiotic use and outcomes remain to be validated.
- Prospective cohort of 230 newly diagnosed hematological patients: 116 prechemotherapy non-neutropenic (cohort A) and 114 postchemotherapy neutropenic (cohort B) with plasma metagenomic next-generation sequencing.
- Random forest classifier distinguished non-neutropenia from neutropenia with AUC 0.8324 based on distinct microbial features associated with neutropenia.
- Infection-risk model achieved AUC 0.942 and 99.1% sensitivity for subsequent infections, rising to AUC 0.953 after integrating microbial features with clinical metrics.
A microorganism-based random forest model built from plasma metagenomic profiles predicted subsequent infection in newly diagnosed hematological patients with AUC 0.942, identifying 99.1% of those who later developed infections, and improved to AUC 0.953 when combined with clinical metrics, supporting targeted prophyl-
The rundown
Researchers performed plasma metagenomic next-generation sequencing in 230 newly diagnosed hematological patients split into prechemotherapy non-neutropenic and postchemotherapy neutropenic cohorts, then analyzed microbial community profiles to find neutropenia-associated features.
Machine learning classifiers were trained on those profiles; the neutropenia classifier reached AUC 0.8324, while the infection-risk classifier reached AUC 0.942 with nested cross-validation showing 99.1% identification of patients who subsequently developed infections and 72.7% of those who remained infection-free, improving to AUC 0.953 with clinical metrics.
Sources
- Peer-reviewedInfectious Diseases and Therapy2026-08-06
How should this claim be treated?
The debate