Linear regression analysis of censored medical costs

Key Points

• To develop valid linear regression techniques for estimating medical costs when subject follow-up is incomplete and total costs are right-censored.
• Modified the classical normal equations for least-squares regression to correct for bias induced by censored follow-up time.
• Derived estimators for settings with only total costs recorded and formulated more efficient estimators when longitudinal cost data are available across multiple discrete time intervals.
• Demonstrated the regression methodology using empirical healthcare cost data from an ovarian cancer patient cohort.
• The modified least-squares estimators are proven to be statistically consistent and asymptotically normal.
• The associated variance-covariance matrices are readily estimable to facilitate standard statistical inference and hypothesis testing.
• Incorporating interval-specific cost information provides substantial efficiency gains over analyzing aggregate total costs alone.