Abstract 1898: The Liverpool Lung Project risk model: An independent validation and clinical utility in primary care

• Published in: Cancer research (Chicago, Ill.) Ill.), 2011-04, Vol.71 (8_Supplement), p.1898-1898
• Main Authors: Field, John K., Raji, Olaide Y., Cassidy, Adrian, Duffy, Stephen W., Baker, Stuart G., Christiani, David C.
• Format: Article
• Language: English
• ISSN: 0008-5472; 1538-7445
• DOI: 10.1158/1538-7445.AM2011-1898

Background: The results from a recent CT screening trial for lung cancer showed evidence of mortality reduction in the screened arm of the study. The long term success of future National Screening Programme as an early diagnosis tool may be dependent upon identifying populations at sufficient risk of developing the disease such that the benefit:harm ratio of the intervention can be maximised. Risk prediction models can play an important role in risk stratification by providing an estimate of individual's risk of developing a disease at a future time, but require validation in independent populations before they can be successfully generalised. Within the Liverpool Lung Project (LLP), we have developed a risk prediction model for estimating individual's 5-year absolute risk of lung cancer. The model was based on five epidemiological risk factors namely smoking duration, prior diagnosis of pneumonia, family history of lung cancer, occupational exposure to asbestos and prior diagnosis of other cancer. We present here an independent validation and clinical utility of the model using data from two case-control studies and a cohort population from Europe and North America. Method: The 5-year absolute risk of lung cancer was estimated for subjects in the Harvard and European Early Lung Cancer (EUELC) case-control and LLP prospective cohort studies. The model's performance was assessed through its predictive accuracy (discrimination and calibration) and clinical utility. The area under the receiver-operator characteristic curve (AUC) measures the model's discriminatory power while calibration was assessed by comparing the observed and the expected lung cancer cases in the cohort population. The clinical relevance of the model was examined using the decision and relative utility curves. Results: There was an evidence of good discriminations in Harvard (AUC = 0.76) and LLP cohort (c-index = 0.77, 0.82) with no significant difference in discrimination by age, gender and smoking status. In general, the model calibration indicated an underestimation of absolute risk; this showed improvement with high absolute risks. The application of the model was associated with reasonably good clinical utility across the three datasets as demonstrated by its superior ‘net benefit’ against any other alternative strategy. Conclusion: The LLP risk model demonstrates good performance and evidence of clinical usefulness in three independent settings and can serve as an adjunct tool for c