1 / 10

Conceptual Issues in Risk Prediction

Conceptual Issues in Risk Prediction. Colin B. Begg Memorial Sloan-Kettering Cancer Center. Conceptual Issues in Risk Prediction. What do we mean by the term “risk”? Model A – prevention – inherent risk germline genetic susceptibility environmental exposures

lyndon
Download Presentation

Conceptual Issues in Risk Prediction

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Conceptual Issues in RiskPrediction Colin B. Begg Memorial Sloan-Kettering Cancer Center

  2. Conceptual Issues in Risk Prediction • What do we mean by the term “risk”? • Model A – prevention – inherent risk • germline genetic susceptibility • environmental exposures • Model B – early detection – probability that a cancer is in the process of development • factors reflecting causal events that have occurred: somatic changes, tumor markers, # breast biopsies

  3. Predictability of a Cancer • How predictable is cancer occurrence? • How much does random variation limit prediction? • Model A – The stochastic (unpredictable) aspects of future cancer events is inherent to the biology of the disease • Model B – As more evidence of a latent cancer is identified its existence is increasingly predictable

  4. Why Does Predictability Matter? • Less predictability  broad population-based prevention strategies are appropriate • More predictability  greater rationale for focussing prevention strategies on high risk individuals.

  5. 39% of cancers 25% of population 25% of population 15% of cancers Distribution of Gail Model PredictorsNurses Health Study

  6. To What Extent are Future Cancers Inherently Predictable • Inferences on maximal attainable predictability are possible by studying incidence rates of second primary cancers (with some assumptions) • Genetic predictability can be inferred from studies of familial aggregation (with some assumptions)

  7. Maximum Attainable Predictability • Each member of population has unique risk, r, density p(r), mean  • Density of risks among incident cases = q(r) = rp(r)/ • Incidence rate of second primaries = • Standardized incidence ratio = • Implication: standardized risk variation can be inferred from the standardized incidence ratio, an estimable quantity

  8. 69% of cancers 25% of population 25% of population 3% of cancers Maximum Attainable PredictabilityBreast Cancer

  9. Data Sources for Risk Prediction Models • In practice, models should use known risk factors • Validation is a pivotal concern • Issues • Absolute risks – generalizability in the context of geographic variations in incidence • Power/precision – studies with large number of “events” are ideal • Implications • Aggregates of population-based case-control studies may be the ideal • In practice, datasets have been used opportunistically • BCDDP (volunteers) for the Gail model • CARET study (RCT) for lung prediction model (Bach) • BCLC (high risk families) for BRCA penetrance

  10. Models for Enhancing Cancer Screening • Tumor markers/ proteomics -- Can these be used to enhance screening strategies? • Goal – identify patients who should be triaged to radiology (mammography) to localize/ identify a tumor. • Predictability is not constrained by fundamental stochastic limitations. • Just as for mammography, because of:- -- uncertain clinical impact -- lead time and length bias these strategies ultimately need to be evaluated in randomized trials with mortality as the endpoint.

More Related