Sign in Subscribe
Machine learning

Multimodal Machine Learning for Predicting Perioperative Safety Indicators in Spinal Surgery

Strategic Insights

— 2 min read

Researchers from the Albert Einstein College of Medicine have developed a novel machine learning architecture that integrates structured electronic health records (EHRs) with unstructured free-text inputs to predict perioperative safety indicators for spinal surgery patients. This preoperative multimodal model demonstrated superior performance in predicting extended length of stay, 90-day reoperation, and perioperative intensive care unit admission compared to baseline tabular EHR models. The study's findings have significant implications for improving patient outcomes and optimizing surgical care.

Key Takeaways:

  • Researchers developed a preoperative multimodal machine learning architecture that integrates structured EHRs with unstructured free-text inputs via natural language processing (NLP) to predict perioperative safety indicators in spinal surgery patients.
  • The multimodal model was trained on a cohort of 1,898 patients admitted for elective or emergency spine surgery at four large urban academic spine centers over a 5-year period from 2018 to 2023.
  • The model predicted perioperative safety indicators with area under the receiver operating characteristic curve (AUC-ROC) ranging from 0.827 to 0.903, brier scores ranging from 0.056 to 0.083, and calibration slopes ranging from 0.755 to 1.217.
  • Important tabular predictors included patient age, body mass index (BMI), hemoglobin level, white blood cell count, platelet count, and combined anterior/posterior spinal fusion approach.
  • Important free-text inputs included vertebral osteomyelitis, radiculopathy, myelopathy, and spinal metastasis.
  • The multimodal model exhibited superior performance in all outcome measures compared to the baseline tabular model.
  • Future work includes incorporating additional model dimensions, such as the history of present illness, physical exam, and spinal imaging, and clinically implementing the models into the informed consent and preoperative optimization pathway.

Statistics:

  • 1,898 patients were included in the study, with 60.7% being female.
  • The median age of the patients was 60.0 years (interquartile range (IQR): 52.0-68.0).
  • The median body mass index (BMI) was 30.3 kg/m (IQR: 26.3-34.6).
  • The 90-day reoperation rate was 10.54%, and the ICU admission rate was 7.74%.
  • The multimodal model achieved precision ranging from 0.909 to 0.933 and recall ranging from 0.979 to 0.994.

Sources:

  • Multimodal machine learning for predicting perioperative safety indicators in spinal surgery. The Spine Journal, 2025;25(11):2450-2460.
  • NewsRx. Study Results from Albert Einstein College of Medicine in the Area of Reoperation Reported (Multimodal machine learning for predicting perioperative safety indicators in spinal surgery). Medical Devices & Surgical Technology Week. November 2, 2025; p 2287.

Read more