Longitudinal AI Models Predict Cancer Adverse Events

Abstract

Clinical interventions for cancer treatment toxicities can reduce morbidity, mortality, and healthcare cost, if high risk patients can be identified. We developed SPARC, a transformer-based foundation model, using real-world, longitudinal, routinely collected health measurements, to predict adverse event risk. Pre-training and fine-tuning SPARC on 101,920,379 measurements from 224,047 patients with cancer led to state-of-the-art performance at predicting cancer-related adverse events on (i) held-out data from the development cohort, (ii) a prospective validation cohort from the same center (Npatients=23,126), (iii) multiple publicly available external datasets (Npatients=26,012), and (iv) external, multi-hospital cohorts (Npatients=562,620, 206,840). SPARC generalized to multiple external centers and identified patients most likely to benefit from an intervention (G-CSF). Pooled multi-institution, more than single-institution, fine-tuning improved generalizability. We release SPARC as a community resource. Our results highlight the utility of AI time series models in oncologic management.

Adaptive MR-guided whole-pelvic SBRT with or without focal intraprostatic lesion boost for high-risk localized prostate Cancer: Adverse events and longitudinal quality-of-life outcomes
Darren M.C. Poon, Jing Yuan, Bin Yang +7
Clinical and Translational Radiation Oncology13 Oct 2026Peer-reviewedCC BY-NC-NDFull textOncologyDOAJEurope PMCEurope PMCPMC
An explainable artificial intelligence framework using electrocardiograms to predict mortality and major adverse cardiovascular events
Jerry Peng, Michael D. Glidden, Sai Rahul Ponnana +11
Scientific Reports5 Oct 2026Peer-reviewedCC BY-NC-ND 4.0Full textCardiovascular medicineCrossref
von Willebrand Factor and ADAMTS13 Predict Therapeutic Efficacy and Severe Adverse Events in Advanced Hepatocellular Carcinoma Treated With Atezolizumab-Bevacizumab.
Naoki Nishimura, Hiroaki Takaya, Tadashi Namisaki +10
JGH open : an open access journal of gastroenterology and hepatology27 Oct 2026Peer-reviewedCC BYFull textGeneral medicineEurope PMCPMC
PhysioFusion: A Multi-Modal Ensemble using Static Preoperative Variables and Time Series Intraoperative Data to Predict Adverse Events Following Cardiothoracic Surgery
R. Korutla, A. Hicks, M. Milosevic +9
medRxiv24 Sep 2026PreprintCC BYFull textEpidemiology & public healthmedRxiv
Dynamic near-term risk prediction of clinically significant immune-related adverse events using longitudinal electronic health records
L. Gong, A. Agyekum, N. Aswani +2
medRxiv6 Oct 2026PreprintCC BYFull textEpidemiology & public healthmedRxiv

Cite

BibTeX RIS

BibTeX

@misc{fu2026longitudinal,
  title = {{Longitudinal AI Models Predict Cancer Adverse Events}},
  author = {Chenlian Fu and Jie (Jennifer) Yu and Ruian (Ian) Shi and Michele Waters and Mitchell Pavlak and Andrea Arfe and Sohrab P. Shah and Alexis Battle and Kenneth L. Kehl and Brian M. Bot and US Cancer AI Alliance (CAIA) and Nikolaus Schultz and Quaid Morris and Justin Jee},
  howpublished = {medRxiv},
  year = {2026},
  month = oct,
  doi = {10.64898/2026.10.22.26364879},
  url = {https://www.medrxiv.org/content/10.64898/2026.10.22.26364879v1},
  note = {CC BY-NC},
}

RIS

TY  - UNPB
TI  - Longitudinal AI Models Predict Cancer Adverse Events
AU  - Chenlian Fu
AU  - Jie (Jennifer) Yu
AU  - Ruian (Ian) Shi
AU  - Michele Waters
AU  - Mitchell Pavlak
AU  - Andrea Arfe
AU  - Sohrab P. Shah
AU  - Alexis Battle
AU  - Kenneth L. Kehl
AU  - Brian M. Bot
AU  - US Cancer AI Alliance (CAIA)
AU  - Nikolaus Schultz
AU  - Quaid Morris
AU  - Justin Jee
PY  - 2026
DA  - 2026/10/22
PB  - medRxiv
DO  - 10.64898/2026.10.22.26364879
UR  - https://www.medrxiv.org/content/10.64898/2026.10.22.26364879v1
AB  - Clinical interventions for cancer treatment toxicities can reduce morbidity, mortality, and healthcare cost, if high risk patients can be identified. We developed SPARC, a transformer-based foundation model, using real-world, longitudinal, routinely collected health measurements, to predict adverse event risk. Pre-training and fine-tuning SPARC on 101,920,379 measurements from 224,047 patients with cancer led to state-of-the-art performance at predicting cancer-related adverse events on (i) held-out data from the development cohort, (ii) a prospective validation cohort from the same center (Npatients=23,126), (iii) multiple publicly available external datasets (Npatients=26,012), and (iv) external, multi-hospital cohorts (Npatients=562,620, 206,840). SPARC generalized to multiple external centers and identified patients most likely to benefit from an intervention (G-CSF). Pooled multi-institution, more than single-institution, fine-tuning improved generalizability. We release SPARC as a community resource. Our results highlight the utility of AI time series models in oncologic management.
ER  -

Abstract

Related

Cite