Cancer Risk Assessment & Insurance Coverage

To manage rising medical costs, an insurance company is using a computer AI model to predict which customers are most likely to develop cancer. The goal is to find high-risk patients early using their medical records and insurance data.

Using this data, the computer automatically sorts patients into two groups: standard risk (no extra testing) or high risk (requires expensive cancer screenings and higher monthly insurance prices).

Ground Truth

In our sample data, 30% of people are high risk a (they will develop cancer within two years). The remaining 70% are low risk / healthy b.

Model Predictions

In a perfect world, the computer would only flag patients who are truly at high risk for cancer, and leave patients who are healthy marked as standard risk.

Model Mistakes

But computer models aren't perfect.

Mistake 1: The computer misses a patient who actually has high cancer risk c.

Mistake 2: The computer flags a healthy patient as high risk f, causing unnecessary fear, costly medical tests, and unfair price hikes.

Never Miss a High-Risk Patient...

One approach is to set the computer to flag people very cautiously. Any borderline case is marked "high risk" so the computer rarely misses anyone who might develop cancer.

We can check how well this works by counting how many patients are flagged correctly vs. by mistake:

... Or Avoid False Alarms & High Bills?

On the other hand, flagging too many people wastes money on unnecessary tests, forces healthy people to pay higher insurance bills, and causes huge emotional stress.

Finding the right balance between catching cancer early and avoiding false alarms is a major challenge when setting up the AI model.

Improve Prediction Power

Adding more personal data can help the computer make better predictions.

But does every piece of data actually help? And is it fair or private to collect such personal details?

Previous oncology and medical records
Family's medical history
Patient's lifestyle and habits
Environmental exposure data
Past insurance claims history
Occupational hazard history
Socioeconomic indicators
Genomic and DNA testing results
Patient AI Chatbot Conversations

Putting it all Together

Now it's your turn! Choose (1) which personal data sources the computer can use, and (2) how cautious the computer should be when flagging someone as high risk.

Previous oncology and medical records
Family's medical history
Patient's lifestyle and habits
Environmental exposure data
Past insurance claims history
Occupational hazard history
Socioeconomic indicators
Genomic and DNA testing results
Patient AI Chatbot Conversations

Try adjusting how cautious or aggressive the AI model is when flagging patients for cancer risk