hacktheuterus

low risk classification example
intermediate risk example
high risk example
digital calculator features part 1: age, bmi, ASA score
digital calculator features part 2: histology type, grade, myometrial invasion
digital calculator features part 3: distant metastasis
The treatment recommendations are based on the 2023 Oncoguía SEGO guidelines

Project Story — NSMP Endometrial Cancer Risk Calculator

1. The Problem

Endometrial cancer patients with a Non-Specific Molecular Profile (NSMP) represent a large and heterogeneous group. At diagnosis, clinicians must decide how aggressive treatment should be, but current risk stratification is largely post-operative and does not fully leverage all available preoperative clinical information.

This creates two major challenges:

Potential overtreatment of low-risk patients
Delayed intensification for patients at high risk of recurrence

2. Our Goal

Our goal was to develop a transparent, data-driven risk stratification tool that:

Uses information available at the time of diagnosis
Predicts recurrence risk and recurrence-free survival
Translates model outputs into clinically interpretable risk groups
Provides guideline-aligned treatment recommendations

3. The Data

We worked with a real-world clinical dataset of NSMP endometrial cancer patients including:

Patient factors (age, BMI, ASA)
Preoperative pathology (histology, grade, myometrial invasion)
Disease spread indicators (distant metastasis)
Longitudinal outcomes (recurrence, death, follow-up time)

We carefully:

Cleaned and harmonized variables
Handled missing values
Translated clinical fields into model-ready features
Restricted features to those available preoperatively

4. Modeling Strategy

We implemented a two-model approach:

🔹 Classification model (Random Forest)

Outcome: Recurrence / recurrence-related death
Output: Individual recurrence probability
Purpose: Assign patients to Low / Intermediate / High risk

🔹 Survival model (Cox proportional hazards)

Outcome: Recurrence-free survival
Output: Personalized survival curves
Purpose: Estimate when recurrence risk accumulates over time

This combination allows both risk ranking and time-to-event interpretation, which is crucial in oncology.

5. Risk Stratification

Patients are grouped into:

Low risk – very low observed recurrence
Intermediate risk – moderate recurrence risk
High risk – substantially higher recurrence rates

We validated the stratification by showing that:

Observed recurrence rates increase monotonically across groups
High-risk patients truly experience more events

6. Clinical Interpretability

To ensure clinical trust:

We examined hazard ratios from the Cox model
Key drivers of recurrence included:
- High tumor grade
- Deep myometrial invasion
- Non-endometrioid histology

These align closely with known clinical risk factors and international guidelines.