A new open-source web tool may increase the interpretability and accessibility of machine learning (ML) models for predicting sepsis based on complete blood count (CBC) data (J Appl Lab Med 2025; doi:10.1093/jalm/jfaf091).
The tool, called SBC-SHAP, visualizes the sepsis risks and individual interpretations of several ML classifiers.
Early detection of sepsis allows for a more timely start on appropriate therapy. CBC data with information about white blood cells, platelets, hemoglobin, red blood cells, and mean corpuscular volume could serve as early indicators of this potentially fatal condition. However, clinicians have had difficulty interpreting previous lab approaches for detecting sepsis because they do not explain how a specific value, such as white blood cell count, contributes to risk predictions. Previously, proposed approaches also required programming expertise that many clinicians lack, according to research.
In response, the study authors developed a graph-based approach for training ML algorithms to incorporate time-series information for prediction based on CBC data. They studied the effect of integrating different ratios into a healthy reference measurement to improve the performance of a previously published ML model. To increase accessibility, they also developed a web application.
The researchers report that their approach increased the sensitivity at 80% specificity across all ML models from 78.2% to 82.9% on an internal dataset and from 65.4% to 73.4% on an external dataset from an independent tertiary care center for prospective time-series analysis.
Example use cases detailed in the paper show how the tool breaks out patient age, sex, and hemoglobin values and how white blood cells, red blood cells, and mean corpuscular volume indicate patient risk. Another use case demonstrates how clinicians might use the tool to correct prior values for a more accurate estimate of sepsis risk.
Users can investigate sepsis risks for individual CBC measurements and investigate how specific feature values contribute to predicted sepsis risks. Users can also filter data with and without results for diverse use cases, the researchers said.
SBC-SHAP is available for use at mdoa-tools.bi.denbi.de/sbc-shap.
MOUTH MICROBES LINKED TO INCREASED PANCREATIC CANCER RISK
Oral microbiota, including fungi, hold promise as biomarkers for high risk of pancreatic cancer, according to a recent study (
JAMA Oncol 2025; doi:10.1001/jamaoncol.2025.3377).
The study collectively linked 27 out of the hundreds of species of bacteria and fungi living in people’s mouths with a 3.5 times greater risk of developing pancreatic cancer.
Smoking, obesity, pancreatitis, and genetics are known to be pancreatic cancer risk factors, but they explain less than 30% of all pancreatic cancer cases. Research on the oral microbiome and pancreatic cancer has focused on the bacterial microbiome, but the relationship between oral microbes and pancreatic cancer risk has remained largely unexplained.
In response, the researchers conducted a prospective study nested in a cohort from two larger American studies encompassing more than 339,000 participants in their 50s, 60s, and 70s. The studies tracked many factors involved in cancer and provided saliva samples that preserved the numbers and species of microbes for testing. The researchers analyzed bacterial and fungal DNA from the saliva samples and followed participants for roughly 9 years on average to record tumor presence.
They identified 445 patients who were diagnosed with pancreatic cancer and compared the DNA of their microbes with DNA from another 445 randomly selected study subjects who had remained cancer-free. The team accounted for pancreatic cancer risk factors including age, race, and how often subjects smoked cigarettes.
In addition to 24 species of bacteria and fungi that individually either raised or lowered pancreatic cancer risk, the researchers also found three kinds of bacteria tied to the cancer that already were known to contribute to periodontal disease. Additionally, the researchers prospectively demonstrated that increased oral
Candida abundance is associated with a higher risk of pancreatic cancer and further confirmed the presence of
Candida in both tumor and normal tissue biospecimens in patients with cancer.
The oral fungal and bacterial microbiotas may serve as readily accessible, noninvasive biomarkers to identify individuals at high risk of pancreatic cancer, the researchers wrote.
TONGUE SWAB TEST MAY SIMPLIFY TUBERCULOSIS TESTING
Preliminary evidence shows that a refined CRISPR-based assay that relies on a tongue swab may transform tuberculosis (TB) testing in low-resource communities (
Nat Commun 2025; doi:10.1038/s41467-025-63094-x).
Current TB tests rely on sputum. Although rich in TB bacteria required for testing, sputum is difficult to collect, making it inefficient for large-scale community testing. Sputum testing is also unfeasible in about 25% of symptomatic cases and nearly 90% of asymptomatic cases, a gap that contributes to an estimated 4 million TB cases going undiagnosed annually, researchers note.
Meanwhile, microbiologic tests take weeks to yield useful results, and some rapid, sensitive, and expensive molecular tests that use other sample types require infrastructure often unavailable in regions with a high prevalence of TB.
In response, the researchers refined an existing CRISPR-based assay to better detect TB in samples with very low levels of bacteria, such as stool, spinal fluid, and tongue swabs. They developed a streamlined “one-pot” asymmetric CRISPR TB assay that eliminates the need for lab or trained medical staff. The assay attenuates amplicon degradation to achieve 5 copies/μL sensitivity within 60 minutes and detect positive patient samples within 15 minutes.
In a cohort of 603 samples, the assay exhibited 93%, 83%, and 93% sensitivity on adult respiratory, pediatric stool, and adult cerebral spinal fluid specimens, respectively. The assay detected 64% of clinically diagnosed tuberculous meningitis cases.
The assay’s overall sensitivity of 74% is greater than the most sensitive reference test with prospectively collected tongue swabs. The assay also showed similar performance when adapted to a lateral flow assay format and employed to analyze self-collected tongue swabs, according to the paper.
The test might be used across diverse specimen types, including those suitable for use in remote and resource-limited settings, to improve access to molecular diagnostics. Future studies need to validate these findings in larger prospective cohorts, the researchers wrote.
