How Can AI Make Health Insurance Fairer?

Imagine a health insurance that sets premiums transparently, based on real data instead of gut feelings. Researchers built a hybrid model that mixes artificial intelligence with game theory to achieve this. First, they used XGBoost, a machine‑learning algorithm, to predict the cost of each medical claim. With a massive dataset, the model reached an R² of 0.787, showing strong predictive power. To understand why the model makes certain decisions, they applied SHAP, an explainability tool that highlights each variable’s impact: healthcare usage, age, and contract type stand out as the main drivers. The most fascinating part is the game among insurer, insured, and employer. The team modeled these interactions as a repeated game, using the Folk Theorem to find conditions where all players can cooperate over time, achieving a stable Nash equilibrium. In practice, if parties consider the future (with an appropriate discount rate), it’s possible to keep premiums fair and incentives aligned. The outcome is a pricing system that is more just, transparent, and sustainable, balancing risk and benefit for every stakeholder.

The XGBoost model is a type of gradient boosting that builds decision trees sequentially, correcting the errors of previous ones. In this study, it was trained on over 200,000 records, including demographic variables (age, gender), clinical history (diagnoses, hospitalizations) and usage metrics (number of visits, prescribed drugs). Performance evaluation showed an RMSE of 1.64 × 10⁷ IRR and an MAE of 1.08 × 10⁶ IRR, indicating modest absolute errors compared to actual costs. SHAP analysis quantified each feature’s importance: for instance, a 10% increase in outpatient visit frequency raised the predicted value by roughly 5,000 IRR. Simultaneously, the game‑theoretic component was formalized as a three‑player game with payoffs dependent on premiums, coverage, and claim costs. The Folk Theorem, applied to an infinite horizon, demonstrates that if the discount factor δ exceeds a critical threshold (≈0.85), cooperative strategies (lower premiums in exchange for healthy behaviors) form a repeated Nash equilibrium. This means that, over the long term, all actors are incentivized to maintain virtuous behaviors, reducing high‑cost claims and enhancing system sustainability. Moreover, the framework includes feedback mechanisms: SHAP results can be communicated to insured individuals to encourage healthier choices, creating a virtuous cycle of data, decisions, and behaviors. The combination of accurate prediction, explainability, and strategic incentives paves the way for fairer insurance systems where transparency is ensured by verifiable algorithms and stakeholder relationships are stabilized by long‑term cooperation principles.