518P - Artificial intelligence real-world applications in pediatric neuro-oncology: The AICCELERATE project

Background

Artificial Intelligence (AI) holds the promise to transform medicine during diagnosis, management and therapy of diseases. The AICCELERATE project will introduce an operational approach to develop AI-enabled digital solutions to improve medical processes by creating a digital twin. In our study, we performed the first pilot application of the AICCELERATE project at Bambino Gesù Children’s Hospital in Rome (OPBG). The aim of our study is to design a supervised machine learning (ML) model to predict time to surgery and time to diagnosis based (outputs) on time from admission to first consultation and imaging (inputs) in newly diagnosed pediatric patients.

Methods

We included pediatric patients with new diagnoses of brain tumors, treated in OPBG from December 2017 to December 2022. We collected from EHR the following data: age and sex, start and end date of the first admission, department of hospitalization, consultations and radiological imaging date, diagnosis, and surgery date.

Results

We included 48 patients with a pediatric brain tumor, with a mean age of 8.6 years. We used a multiple linear regression model, which does not require large datasets such as more complex models. Our analysis showed that delayed consulting and imaging led to a delayed diagnosis, with the consulting presenting a higher importance than the imaging. This result was validated by the model’s estimated regressors coefficients, with the consulting factor being assigned a larger weight β_C=1.81 than the imaging factor β_I=-0.33. Similarly, we observed that delayed consulting and imaging led to a delayed surgery, estimating a higher coefficient for the consulting β_C=0.71 than the one estimated for the imaging β_I=-0.06. While the ML model achieved a reasonable performance in both frameworks, the time to surgery was predicted with a lower MAE 4.45±0.87 compared to the diagnosis MAE of 6.74±0.95.

Conclusions

This preliminary study conducted in a small number of patients, suggests that decreasing time to consultation can shorten time to surgery in children with brain cancer. As the AICCELERATE is ongoing, more accurate models will be developed and will help to better understand how to optimize the journey of children with brain cancer.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.