Causal Inference, Predictive ML & CDSS for Precision Oncology & Chronicity (IBIMA / Hospital Costa del Sol / RICAPPS)

About Us & Institutional Background

Dr. Francisco-Javier Martínez-Barrios is an in-house Health Data Scientist, Lead Methodologist, and Postdoctoral Researcher at the Research and Innovation Unit of Hospital Universitario Costa del Sol, affiliated with the Biomedical Research Institute of Málaga (IBIMA BIONAND Platform, Group B14) and the Spanish RICAPPS network (ISCIII).

With a unique dual clinical background (Health Sciences) and advanced training in Biomedical Data Science (PhD in Health Sciences / Public Health, MIT Big Data, IE University, Samsung Innovation Campus), our group bridges the gap between complex hospital clinical data and cutting-edge causal/predictive biocomputation. We co-lead the national CARESS-CCR multicenter cohort (22 public hospitals, >2,500 patients) and have a registered patent for Clinical Decision Support Systems (CDSS).

Core Scientific & Computational Expertise Offered (What We Bring to Consortia)

1. Formal Causal Inference on Real-World Data (RWD):

   • Target Trial Emulation (TTE): Protocol-driven emulation of randomized trials using observational hospital cohorts, eliminating immortal time bias via clone-censor-weight methodologies and Marginal Structural Models (MSM / sIPCW).

   • Counterfactual Causal Mediation: Structural Causal Models (SCMs) and Pearl’s do-calculus with non-parametric bootstrap resampling to dissect direct vs. indirect effects (e.g., decomposing functional capacity as an actionable mediator of 10-year survival).

   • Confounder control via Directed Acyclic Graphs (DAGs) and stabilized propensity weighting (sIPTW).

2. Advanced Survival Machine Learning & Biocomputation:

   • Non-linear Survival Modeling: Random Survival Forests (RSF), penalized Cox regressions (Ridge/Lasso), and Fine & Gray competing risks analysis over long-term clinical horizons (8 to 10 years).

   • Agnostic Changepoint Optimization: Discovery of biological non-linear thresholds (e.g., severe hyperinflammation via Log-Rank test maximization validated with permutation tests and bootstrapping).

   • Precision Phenotyping: Unsupervised clustering for mixed data types (K-Prototypes) and topological manifold learning (UMAP) integrating clinical, metabolic (BMI), and somatic molecular markers (dMMR/MSI, KRAS, BRAF).

3. Auditable CDSS Architecture & Explainable AI (XAI):

   • Design of transparent, non-black-box Clinical Decision Support Systems.

   • Global and local feature interpretability via Shapley Additive exPlanations (SHAP matrices).

   • Evaluation of real-world clinical utility and net health benefit using Decision Curve Analysis (DCA).

4. Hospital Integration & Real Clinical Data Access:

   • Direct embedment within a tertiary hospital and regional healthcare service (SSPA), ensuring seamless extraction of Electronic Health Records (Diraya), pathology reports, and biobank linkability.

   • Robust compliance with ethics committee protocols (CEI), GDPR-compliant pseudonymization pipelines, and the EU Artificial Intelligence Act.

Consortium Partnership Needs (What We Are Looking For)

We are not seeking software development agencies or external AI consulting firms, as we internally handle the entire computational, algorithmic, and statistical pipeline. We are seeking:

• European Clinical & Hospital Oncology Consortia: Academic medical centers and hospital networks with independent international cohorts (colorectal, breast, or GI cancers) interested in external validation of predictive survival models and CDSS tools.

• Coordinated Horizon Europe Consortia: Project coordinators seeking an experienced hospital-based team to lead or co-lead the Work Package (WP) dedicated to Data Science, Causal Inference, Biostatistics, or Clinical Algorithmic Validation.

• Multi-Omics & Systems Biology Platforms: Groups generating high-throughput molecular/single-cell or spatial transcriptomic datasets in need of causal network reconstruction (DAG learning, neural SCMs) and translational mapping to clinical outcomes.