Hggpredictor

hgg.predictor

Predictor Model for the Presence of High Grade Glioma Cells

Based of prospective intraoperative data, the investigators developed the present model for predicting the possibility of encountering high grade glioma cells in a certain region of the surgical cavity given two variables: 5-aminolevulinic acid (5-ALA) and intraoperative ultrasound (ioUS).

Solid tumor (>= 20% of cells in 10X microscope, H&E stain) and infiltration (< 20% of cells in 10X microscope, H&E stain) were defined. The presence of any kind of tumor referred to the combination of both.

The model is significantly suited for predicting the probability of encounter the presence of solid tumor, and infiltration (logistic regression models, p < 0.0001 in both cases). The measure of the effect was different, however, for each outcome. Here are presently those statistically significant:

For the presence of solid tumor, the “positive” 5-ALA result showed an OR of 4.05 [1.6 – 10.3, p=0.002), the “positive” ioUS result showed and OR of 3 [1.2 – 7.4, p=0.02), and the “negative” ioUS result showed an OR of 0.41 [0.18 – 0.93, p=0.03). This data was based on a logistic regression model.

For the presence of infiltration, only the “positive” 5-ALA result showed an OR of 5 [1.1 – 20] (p=0.05). This data was based on a logistic regression model.

Finally, the model would predict the percentage of tumor cells in a certain biopsy of the surgical cavity using a lineal regression model. The mean absolute error of this model is reported to by 26%.

Prediction model

This model predicts the probability of encounter
high grade glioma cells in the surgical cavity. The
results are expressed in three ways.

Prognostic Results:

Probability for any kind of tumor (infiltration and solid tumor):

of HGG cells in the biopsy.

of probability of encountering solid tumor.

of probability to encounter infiltration.

2024

Legal Statement

The HGGPredictor can be used to predict, in real time during HGG surgery, the probability of encountering tumor cells in the surgical cavity. The results from the present algorithm may contribute as complementary data during the surgery, but they should be interpreted cautiously (since the tool is still under investigation), and they should never replace clinical judgement and surgical experience.

Given the infiltrative nature of high grade gliomas, the prediction of the presence of infiltration can be very challenging, justifying some of the contradictory results evoked by the present model (i.e, 0% probability of infiltration and 49,8% probability of solid tumor). For that reason, the authors also provided the different OR for each input variable.