Rehabilitation of the Case-Control Design in Prediction Research

• Activity Number: 371
• Type: Topic Contributed
• Date/Time: Wednesday, August 1, 2007 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Karel G. Moons*+ and Cornelis Biesheuvel and Yvonne Vergouwe and Ruud Oudega and Arno W. Hoes and Diederick E. Grobbee
• Companies: Julius Center UMC Utrecht and Julius Center UMC Utrecht and Julius Center UMC Utrecht and Julius Center UMC Utrecht and Julius Center UMC Utrecht and Julius Center UMC Utrecht
• Keywords: Prediction research ; Absolute disease probabilties ; Nested case control ; Diagnostic research ; Prognostic research ; Valid and efficient

Abstract

Case-control studies are widely disapproved for (diagnostic and prognostic) prediction research; potential for selection bias and cannot obtain absolute disease probabilities. Both can be solved with nested case-control design. Using data from diagnostic cohort study, we drew nested case-control samples including all cases. Controls were randomly sampled with ratios 1:1, 1:2, 1:3 and 1:4. Per ratio, 100 samples. We weighted controls by corresponding 1/sample fraction. Predictive accuracy of index tests, including post test probabilities, in nested case-control samples were same as in full cohort. Width of 95% CIs decreased with increasing control number. Nested case-control is a valid and efficient design for prediction studies. It should be (re)appraised and is notably advantageous if disease is rare, index tests are costly/invasive, or when stored data or human material is reanalyzed.