What’s the Problem?

Colorectal cancer is one of the most common cancers worldwide. When it spreads (becomes metastatic), treatment becomes more complex. Doctors often use combinations of chemotherapy and targeted drugs like panitumumab, which blocks a protein called EGFR that helps cancer cells grow.

But here’s the catch: not all patients respond to panitumumab, even if they have the right genetic profile (KRAS wild-type). Giving it to everyone means some people get side effects without any benefit, and that’s not ideal.

 The Search for a Better Way

To solve that problem, in Topazium, we have asked: can we predict who will actually benefit from panitumumab before treatment starts?

To do this, we have analyzed data from nearly 1,000 patients across two large clinical trials. Using machine learning (a type of artificial intelligence), we have generated virtual patients or digital twins and looked for patterns in blood tests and clinical features that could separate responders from non-responders.

Meet PROCC: A Smart, Simple Score

The result is PROCC — short for Panitumumab Response Optimizer for Colorectal Cancer. It’s a score based on just four common blood tests:

• CEA (Carcinoembryonic Antigen)
• LDH (Lactate Dehydrogenase)
• ALP (Alkaline Phosphatase)
• Platelet count

These are tests any colorectal cancer patient would typically get before starting treatment anyway.

 How It Works

Each test result is given a score based on how high it is. The total score ranges from 0 to 18. If your score is 8.5 or higher, you’re likely to benefit from panitumumab. If it’s lower, the drug might not help.

 What Did the Study Find?

We have divided patients into two groups based on their PROCC scores and compared survival outcomes.

Patients with High PROCC Score (≥8.5)

• Better survival when treated with panitumumab + chemotherapy
• Lower risk of disease progression
• Longer overall survival

Patients with Low PROCC Score (<8.5)

• No significant benefit from adding panitumumab
• Chemotherapy alone worked just as well

Why These Four Markers?

These blood markers are linked to how aggressive the cancer is:

• CEA is often elevated in colorectal cancer and reflects tumor burden.
• LDH and ALP are enzymes that can indicate tissue damage or cancer spread.
• Platelets play a role in cancer progression and inflammation.

How AI Helped

We have used feature encoders and embedding systems to generate virtual representations or digital twins of patients based on 36 clinical features. An unsupervised machine learning analysis has revealed two subpopulations:

• SPA: Patients who benefited from panitumumab
• SPB: Patients who didn’t

Feature contribution analysis identified 15 differential features between both subpopulations. Then, we have used traditional biostatistics to build the PROCC score, making it easy for doctors to use. This approach let us use a complex AI system to spot clinical and lab features linked to poor health outcomes. Then, we translated those findings into a clear and familiar scoring method that healthcare professionals are used to working with, adding explainability to the model.

 Why This Matters

PROCC could help:

• Personalize treatment for colorectal cancer patients based on a true survival benefit
• Avoid unnecessary side effects
• Save costs by targeting therapies more effectively

And because it uses routine blood tests, it’s accessible and easy to implement in real-world clinics.

What’s Next?

While PROCC looks promising, it needs to be tested in more patient groups and clinical settings. We are working to adapt it for other drugs and cancer types too.

Final Thoughts

This study is a great example of how AI, data science and medicine can work together to improve patient care. By turning complex data into a simple score, PROCC empowers doctors to make smarter decisions, and gives patients a better chance at effective treatment. A true collective intelligence, for wiser medical decisions.