Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
4949280 | Computational Statistics & Data Analysis | 2017 | 23 Pages |
Abstract
Medical imaging techniques are being rapidly developed and used for diagnosis and prognostic predictions. To validate a prognostic predictive utility of a new imaging marker, a temporal association needs to be established to show an association between its baseline value with a subsequent chance of having the relevant clinical outcome. Validation of such techniques has several difficulties. First, different from techniques based on blood or tissue specimen, imaging techniques often have no historical samples to study and require new studies to collect data. For rare events, it can be costly. Second, the rapid technology evolution requires such validation studies to be short in order to keep the evaluation relevant. A new statistical design is proposed that extends traditional prospective cohort study by adding cases with known time of events and including a short-term follow-up to estimate the prospective odds ratio for the clinical endpoint of interest within a reasonably short duration of time. Under a Markov model, this new design can deliver a consistent estimate of the odds ratio and a formula for asymptotic variance. Numerical studies suggest that the new design induces a smaller variance than the corresponding prospective cohort study within three follow-ups. An application to Alzheimer's disease data demonstrates that the proposed design has a potential to be useful to rapidly establish a prognostic validity of a new imaging marker within a reasonable time, with a small sample size.
Keywords
Related Topics
Physical Sciences and Engineering
Computer Science
Computational Theory and Mathematics
Authors
Joong-Ho Won, Xiao Wu, Sang Han Lee, Ying Lu,