Abstract:
|
New US national screening management guidelines for cervical cancer will be based on risk. We estimated risk to guide management for nearly all decision points that occur in cervical cancer screening by using electronic health records from 1.5 million patients undergoing routine screening from 2003 to 2017. To estimate risks, we used prevalence-incidence mixture models, which combines a logistic regression model for partially latent prevalent disease (precancer/cancer) and a proportional hazards model for interval-censored incidence disease. Risk thresholds for different management options were determined by a consensus group of experts representing 19 organizations after consideration of trade-offs in benefit, harms, and efficiency. Uncertainty of the determined management was defined as the probability of the estimated risk distribution falling into defined thresholds. We assessed calibration of the estimated risks and associated management using data from the CDC (Centers for Disease Control and Prevention) and industry partners.
|