A Chinese study found that routine inflammatory and nutritional markers, combined with machine-learning models, could identify inpatients who later developed diabetic foot ulcers, although larger prospective studies are needed before clinical use.
Hospital patients with type 2 diabetes can develop diabetic foot ulcers suddenly, even during an admission for an unrelated problem, and a new study from China suggests that routine blood tests taken on arrival may help identify who is most at risk. Researchers at the Sixth Affiliated Hospital of Xinjiang Medical University in Urumqi reported in Frontiers in Medicine that a set of inexpensive inflammatory and nutritional markers, when combined with machine-learning models, showed strong ability to distinguish patients who later developed an ulcer during their stay from those who did not.
That matters because diabetic foot ulcers are a major and often underestimated complication of diabetes. Reviews from the National Library of Medicine and JAMA note that they can lead to infection, osteomyelitis, gangrene and amputation, while long-term mortality remains high. A large review in the Journal of the American Medical Association reported that roughly 50% to 60% of ulcers become infected, about 20% of moderate to severe infections end in lower-extremity amputation, and five-year mortality is around 30%, rising above 70% after major amputation. Other reviews estimate that diabetic foot ulcers affect about 18.6 million people worldwide each year.
The Chinese team followed 550 inpatients with type 2 diabetes who had no active foot ulcer on admission and who had blood results available within 48 hours. Instead of using a random split, which can make a model look better than it really is, the researchers trained their models on patients admitted in 2022 and 2023 and tested them on a separate group admitted in 2024. During the study period, 66 patients developed a new ulcer in hospital, and the event rate was 12% in both the development and validation groups.
The predictors went beyond standard diabetes measures. The researchers examined age, diabetes duration, blood pressure, smoking and drinking history, plus laboratory values including neutrophils, lymphocytes, platelets, haemoglobin, albumin, C-reactive protein, high-density lipoprotein cholesterol, glycated haemoglobin, creatinine and the ankle-brachial index, a bedside test of blood flow in the legs. They also calculated several composite indices that combine inflammation and nutritional status, including the neutrophil percentage-to-albumin ratio, the albumin-to-neutrophil-to-lymphocyte ratio, the haemoglobin-albumin-lymphocyte-platelet score and the systemic immune-inflammation index. In the final models, age, diabetes duration, HbA1c and the albumin-to-neutrophil-to-lymphocyte ratio remained especially important.
Among the four models tested, logistic regression performed best on discrimination in the validation cohort, with an area under the receiver operating characteristic curve of 0.976 and high sensitivity at the pre-set threshold. Random forest and XGBoost also performed strongly, while LightGBM produced the lowest Brier score but missed most ulcer cases at its chosen threshold. The authors cautioned that no model was uniformly best across discrimination, calibration and sensitivity, and they stressed that the findings are not ready for bedside use. The study was single-centre, retrospective and based on only 18 ulcer events in the validation group, so larger prospective studies will be needed before hospitals can rely on such tools in routine care. Broader reviews of diabetic foot disease also underline that established clinical risks such as prior ulcers, neuropathy, foot deformity and peripheral artery disease remain central to prevention and should not be replaced by laboratory modelling alone.
Disclaimer: This content is for informational purposes only and is not intended to be a substitute for professional medical judgment, advice, diagnosis, or treatment.
