Unrequested imaging findings on routine chest CT : results from the PROVIDI study

Abstract

Prognostic research is of growing importance. However, rarely are results from imaging techniques considered for medical prognostication, whilst prognostically promising unrequested imaging findings are increasingly being detected in daily routine care. Therefore this thesis aimed to contribute to the general knowledge of prognostic research and specifically to determine the prognostic relevance of unrequested imaging findings, detected on routine chest CT, in relation to cardiovascular disease (CVD). Regarding the first aim we began addressing the general methodology of prognostic research using data from the SMART study. First, the pro’s, con’s and recommendations concerning the reporting of composite-endpoints were discussed, followed by the presentation of an adaptation method, designed to accurately adjust absolute risks for a composite-endpoint to risks for the individual component outcomes. Secondly, we presented an ‘efficient’ method of composing a temporal validation cohort. This method replenishes a traditional temporal validation cohort with the left-truncated follow-up time of event-free patients censored at the end of a derivation-study. The major advantages were increased power and increased length of follow-up time, available for validation. Through both works we strove to improve upon current methodological practices aiming to improve the quality of future research. The bulk of this thesis addresses the second aim presenting results from the PROgnostic-Value-of-unrequested-Information-in-Diagnostic-Imaging study. We started with discussing the rationale that led to the initiation of this study. 10.410 subjects (>40 years without cardiovascular CT-indications) were followed for 4 years, in which 515 CVD events were ascertained. We found that vascular calcification (severe coronary calcification HR3.7, 95%CI 2.7-5.2; severe aortic calcification HR2.7, 95%CI 2.0-3.7) and valve calcification (severe Aortic valve-, Mitral valve- or Mitral annular calcification respectively HR’s 2.0, 95%CI 1.48-2.78; 2.1, 95%CI 1.04-4.19 and 1.5, 95%CI 1.13–2.08) were strong predictors of incident CVD events. The heart (HR1.04, 95%CI 1.03-1.06) and ascending thoracic diameter (HR1.002 , 95%CI 1.001-1.004) showed an exponential prognostic effect beyond respectively 11cm and 30mm; the descending aortic diameter (HR1.04 , 95%CI 1.01 – 1.13) and cardiothoracic diameter (HR1.06, 95%CI 1.04 -1.08) showed linear prognostic effects beyond respectively 25mm and a ratio of 0.45. Finally, we developed two different prediction models to identify patients who might benefit from preventative measures. A simple prediction model incorporating aortic calcifications, plus age, gender and CT-indication performed well in an external dataset (C-index 0.71, sensitivity 46%, specificity 76%). A more sophisticated CVD prediction rule (Triple-R score) was designed to establish the optimal combination of imaging findings for predicting CVD, and also to establish which findings may be omitted. The backward-specified Triple-R score incorporated age, gender, CT-indication, LAD calcification, Mitral valve calcification, descending aortic calcification and the heart- and ascending aortic diameter. At external validation the C-index was 0.72 and sensitivity and specificity was 46% was 79%. Compared to the simple risk score, the Triple-R score classified 7.8% of the patients into more accurate risk categories. In conclusion, we showed that unrequested imaging findings can be used for risk stratification purposes. Impact studies are needed to build upon and corroborate on our findings, before applying these results to daily practice.