Application of Artificial Intelligence and Iron Metabolism Markers in Predicting ICU Outcomes for Critically Ill Cancer Patients

Status

Completed

Phases

Unknown

Study type

Observational

Source

ClinicalTrials.gov

Registry ID

NCT07408661

Enrollment

1137

Registered

2026-02-13

Start date

2015-01-01

Completion date

2025-12-01

Last updated

2026-02-13

For informational purposes only — not medical advice. Sourced from public registries and may not reflect the latest updates. Terms

Conditions

Cancer

Keywords

Critical care, Cancer, Artificial Intelligence, Risk stratification, Ferritin

Brief summary

This study aimed to develop a more accurate way to predict the 30-day survival of cancer patients admitted to the intensive care unit (ICU). The researchers focused on markers of iron metabolism, as imbalances in iron are common in cancer and severe illness. The study analyzed data from 1,137 critically ill cancer patients. Using artificial intelligence (AI), specifically a model called TabPFN, the study combined these iron markers with other routine clinical data (like blood cell counts and lactate levels) to create a new prediction tool.

Detailed description

Revised Protocol Description (Study Plan): This retrospective cohort study aims to evaluate whether the integration of artificial intelligence with iron metabolism markers can improve the prediction of 30-day all-cause mortality in critically ill adult cancer patients admitted to the ICU. Data will be derived from the MIMIC-IV database. Eligible patients will be identified based on predefined inclusion and exclusion criteria. The study will assess the prognostic value of three iron metabolism markers-ferritin, serum iron, and total iron-binding capacity (TIBC)-both individually and in combination with other clinical variables. Multiple machine learning algorithms will be developed and compared. Feature selection will be performed using methods such as LASSO regression. Candidate models will include, but are not limited to, TabPFN, XGBoost, and Random Forest. Model performance will be evaluated in an independent test set using metrics including the area under the receiver operating characteristic curve (AUC), calibration plots, Brier score, and decision curve analysis. To ensure model interpretability, SHAP (SHapley Additive exPlanations) analysis will be applied to the final model to identify the most influential predictors. The study protocol has been reviewed and approved by the relevant institutional review boards, and all methods will be conducted in accordance with relevant guidelines and regulations.

Interventions

None listed

Study design

Observational model

COHORT

Time perspective

RETROSPECTIVE

Eligibility

Sex/Gender

ALL

Age

18 Years to 100 Years

Healthy volunteers

Inclusion criteria

1. Adult patients (age ≥ 18 years). 2. Diagnosis of any type of cancer, as recorded in the hospital database. 3. First ICU admission during the hospital stay (only the first ICU stay is considered for patients with multiple admissions).

Exclusion criteria

1. Length of ICU stay less than 24 hours. 2. Missing or unavailable data for the key study variables, specifically iron metabolism markers (ferritin, serum iron, total iron-binding capacity) or essential clinical parameters needed for analysis.

Design outcomes

Primary

Measure	Time frame	Description
All-cause Mortality at 30 Days	30 days from the date of ICU admission.	The primary outcome is the incidence of death from any cause within 30 days following the date of ICU admission. Mortality status will be determined by a review of the hospital discharge records and associated death records in the MIMIC-IV database.

Outcome results