Results
Our risk model
We’d be happy to walk you through the results and discuss what an evaluation could look like at your hospital.
Contact foundersWhat we evaluated
Our research model used prior patient records and information from the first 24 hours of a hospital stay to estimate malnutrition risk. We evaluated how well its scores distinguished admissions with and without documented malnutrition.
Study scale
The evaluation used adult non-ICU admissions from MIMIC-IV, a research dataset of de-identified hospital records. Each admission represents one hospital stay. We split the admissions into training, validation, and test sets.
We used training data to develop the model, validation data to check its performance during development, and a separate test set for the reported results.
- Total admissions
- 326,222
- Training
- 195,678
- Validation
- 64,856
- Testing
- 65,688
Methods
The research model used LightGBM, a gradient-boosted decision tree model. We identified documented malnutrition using diagnosis codes recorded at discharge.
We calculated feature influence on the validation set using mean absolute SHAP (SHapley Additive exPlanations) values on the model’s log-odds scale.
Model performance
The research model achieved 0.93 AUROC (area under the receiver operating characteristic curve) on 65,688 admissions reserved for testing. AUROC measures how well the model ranks admissions with documented malnutrition above those without it. At 0.5, the ranking is no better than random guessing. At 1.0, every admission with documented malnutrition ranks higher than every admission without it.
False alarm rate (%)
IllustrativeVisualization
- Filos
- MST
- Random
What informed the risk scores
The chart below shows the relative influence of each type of information on the research model’s scores. Each percentage is the category’s share of the total measured influence across all information categories.
- First-day laboratory results23%
- Care unit and clinical service16%
- Outpatient weight and body mass index history14%
- First-day medication orders11%
- Admission and emergency department information9%
- Laboratory results before admission9%
- Demographics and insurance coverage8%
- Prior malnutrition diagnosis codes7%
- Previous hospital admissions2%
- Other previous diagnoses2%
Limits of this evaluation
The study used MIMIC-IV records from a single medical center, Beth Israel Deaconess Medical Center. Performance at another hospital can differ.
The study reflects malnutrition documented in the available records. Incomplete documentation may cause a hospital stay with malnutrition to be counted as a stay without documented malnutrition.
Evaluating Filos at your hospital
An evaluation at your hospital would assess how well Filos identifies patients at risk of malnutrition using your hospital’s own records, how many patients it prioritizes for review, and how it fits into your nutrition team’s workflow. We agree on success criteria with your team before the evaluation begins.
The hospital product is designed to update malnutrition risk scores throughout a patient’s stay and work alongside your hospital’s electronic health record system and clinical documentation tools.
Data and citations
PhysioNet provides access to MIMIC-IV for credentialed researchers who complete training and sign a data use agreement.
The Malnutrition Screening Tool (MST) curve shown above is based on a separate study by Morris and colleagues. They compared MST results with the Subjective Global Assessment, a clinical nutrition assessment, in acute medical inpatients.
- Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, Pollard TJ, Hao S, Moody B, Gow B, Lehman LH, Celi LA, Mark RG. MIMIC-IV, a freely accessible electronic health record dataset. Scientific Data. 2023;10:1. doi:10.1038/s41597-022-01899-x
- Pollard T, Moody BE, Lehman LH, Gow B, Fernandes C, Xie C, Johnson A, Mark RG, Heldt T. PhysioNet as a global platform for biomedical research. Nature Health. 2026;1(8):792–795. doi:10.1038/s44360-026-00096-z
- Morris NF, Stewart S, Riley MD, Maguire GP. A comparison of two malnutrition screening tools in acute medical inpatients and validation of a screening tool among adult Indigenous Australian patients. Asia Pacific Journal of Clinical Nutrition. 2018;27(6):1198–1206. doi:10.6133/apjcn.201811_27(6).0005
Request the research code
We’re happy to share the code behind this evaluation with members of the research community as we work together to improve care for patients. Contact the founders to request access.
Request code access