Limitations of p-values and R-squared for stepwise regression building: A fairness demonstration in health policy risk adjustment