What is a key ethics risk when using cost as a label for predicting health need?

What is a key ethics risk when using cost as a label for predicting health need?

• A. Proxy label bias: cost can reflect unequal access and spending rather than true need.
• B. Privacy concerns about patient data.
• C. Overfitting to historical billing patterns.
• D. Data leakage from training to test.

Answer: (A)

Using cost as the label for predicting health need introduces a bias because cost acts as a stand-in for factors beyond true medical necessity. The amount someone spends on healthcare is heavily influenced by access to care, insurance coverage, provider networks, and local spending patterns. This means that two people with the same underlying health needs could have very different costs simply because one has good access and the other does not. When cost drives predictions, the model may flag high-need patients who happen to be expensive to treat and miss those whose needs are real but not reflected in their spending, perpetuating or worsening inequities in care. Privacy concerns, while important in general, address data protection rather than the fairness of using cost as a signal. Overfitting to historical billing patterns and data leakage from training to test are issues about model performance and evaluation, not the ethical risk introduced by labeling health need with cost.

Using cost as the label for predicting health need introduces a bias because cost acts as a stand-in for factors beyond true medical necessity. The amount someone spends on healthcare is heavily influenced by access to care, insurance coverage, provider networks, and local spending patterns. This means that two people with the same underlying health needs could have very different costs simply because one has good access and the other does not. When cost drives predictions, the model may flag high-need patients who happen to be expensive to treat and miss those whose needs are real but not reflected in their spending, perpetuating or worsening inequities in care.

Privacy concerns, while important in general, address data protection rather than the fairness of using cost as a signal. Overfitting to historical billing patterns and data leakage from training to test are issues about model performance and evaluation, not the ethical risk introduced by labeling health need with cost.