Virtual Library
Start Your Search
M. Markaki
Author of
-
+
P2.13 - Radiology/Staging/Screening (ID 714)
- Event: WCLC 2017
- Type: Poster Session with Presenters Present
- Track: Radiology/Staging/Screening
- Presentations: 1
- Moderators:
- Coordinates: 10/17/2017, 09:30 - 16:00, Exhibit Hall (Hall B + C)
-
+
P2.13-026a - A Validated Clinical Lung Cancer Risk-Prediction Model for Light-, Heavy- and Ex-Smokers: the Lung-HUNT Model (ID 10328)
09:30 - 09:30 | Author(s): M. Markaki
- Abstract
Background:
Lung cancer screening will become an important way of reducing lung cancer mortality. Identifying high-risk population based purely on age and pack years may leave out 3/4 of high-risk individuals. There is an urgent need for validated, accurate risk-prediction models for all ages and types of smokers.
Method:
In the prospective cohort of 65 237 people aged 20-100 years participating in the HUNT2 study in Norway in 1995-97 (70% of the regional adult population), median follow-up time of 15·2 years (800 845 person-years), 583 incident lung cancer cases were diagnosed (cumulative incidence 0·9%). Thirty-six candidate risk variables for lung cancer were examined using univariate and multivariate analyses and backwards feature selection using multiple imputation. The model was validated in ten comparable Norwegian population studies of 44 600 ever-smokers (CONOR), with a median follow-up time of 11·6 years and 675 incident lung cancer events.
Result:
In the total HUNT2 cohort at base-line, the smokers were light smokers (median 10·3 pack-years). Among the lung cancer cases 94·7% were ever-smokers (median 22·5 pack-years) and 70% of lung cancer cases had reported smoking <30 pack-years at base-line. There were only seven risk variables selected in the final model; age, pack-years, smoking intensity (number of cigarettes daily), years since quitting, body mass index, daily cough and hours of daily exposure to cigarette smoke. The model for ever-smokers had a concordance index of 0·869 (interquartile range 0·868-0·870). A nomogram was made to calculate the personal 5, 10, and 15-year risk of lung cancer. External validation of the model in CONOR on 44600 ever-smokers showed a similar concordance index of 0·867 ([0·854, 0·880]95% CI). Selecting a threshold of median risk one would need to screen only 22·7% of ever-smokers to identify 78% of all lung cancers.
Conclusion:
The resulting Lung-HUNT model is simple, robust and accurate, and identify lung cancer risk individuals of all ages and smoking patterns. This model is useful for prospective screening studies for lung cancer and can motivate smokers to quit smoking.
-
+
P3.02 - Biology/Pathology (ID 620)
- Event: WCLC 2017
- Type: Poster Session with Presenters Present
- Track: Biology/Pathology
- Presentations: 1
- Moderators:
- Coordinates: 10/18/2017, 09:30 - 16:00, Exhibit Hall (Hall B + C)
-
+
P3.02-097a - Metabolic Biomarkers in Serum for the Early Diagnosis of Lung Cancer: First Results from the Cancer-Biomarkers in HUNT Initiative (ID 9792)
09:30 - 09:30 | Author(s): M. Markaki
- Abstract
Background:
To date there are no clinical biomarkers for the early diagnosis lung cancer. The Cancer-Biomarkers in HUNT initiative analyses serum samples collected two months to five years before diagnosis (prospective HUNT study, Trondheim, Norway) for identifying metabolomics signatures for the early detection of lung cancers.
Method:
Thirty-six serum samples of individuals that subsequently developed adenocarcinoma (n=12), squamous cell carcinoma (n=12) and small-cell lung cancer (n=12) were profiled with LC-MS untargeted (Amide-) metabolites (n = 1042), along with 36 sera from individuals that were cancer-free 5 years after blood sampling matched for smoking status, gender and age. Each cancer subtype as well as adeno plus squamous (non-small cell lung cancer) was contrasted against its respective controls as well as . For each contrast, the moderated t-test implemented in the R package limma was used for performing univariate analysis, while multivariate analyses were performed using the Just Add Data software (Gnosis Data Analysis), which implements a data-analysis pipeline comprehensive of feature selection, non-linear modelers (e.g., Random Forests) and cross-validation with bootstrapping for optimizing algorithms and providing unbiased performance estimation.
Result:
Two non-overlapping signatures, each containing four metabolites were identified by the non-linear data analysis pipeline, the first discriminating adeno patients (AUC 0.71, CI = [0.52, 0.9]) (Figure 1) and the second discriminating adeno and squamous cases from their respective controls (AUC = 0.643, CI = [0.452, 0.803]). No association between metabolites and cancer was identified by the univariate analyses at FDR level 0.1.
Conclusion:
The results suggest that metabolic information in serum may help in detecting lung cancer two months to five years prior to clinical lung cancer diagnosis. This is the first large-scale untargeted metabolomics screening of pre-diagnostic serum of future lung cancer patients. Further studies are in progress for validation of these findings Figure 1
