Erschienen in:
21.07.2017 | Original Article
Impact of Age on the Risk of Advanced Colorectal Neoplasia in a Young Population: An Analysis Using the Predicted Probability Model
verfasst von:
Yoon Suk Jung, Chan Hyuk Park, Nam Hee Kim, Mi Yeon Lee, Dong Il Park
Erschienen in:
Digestive Diseases and Sciences
|
Ausgabe 9/2017
Einloggen, um Zugang zu erhalten
Abstract
Background
The incidence of colorectal cancer is decreasing in adults aged ≥50 years and increasing in those aged <50 years.
Aims
We aimed to establish risk stratification model for advanced colorectal neoplasia (ACRN) in persons aged <50 years.
Methods
We reviewed the records of participants who had undergone a colonoscopy as part of a health examination at two large medical examination centers in Korea. By using logistic regression analysis, we developed predicted probability models for ACRN in a population aged 30–49 years.
Results
Of 96,235 participants, 57,635 and 38,600 were included in the derivation and validation cohorts, respectively. The predicted probability model considered age, sex, body mass index, family history of colorectal cancer, and smoking habits, as follows: Y
ACRN = −8.755 + 0.080·X
age − 0.055·X
male + 0.041·X
BMI + 0.200·X
family_history_of_CRC + 0.218·X
former_smoker + 0.644·X
current_smoker. The optimal cutoff value for the predicted probability of ACRN by Youden index was 1.14%. The area under the receiver-operating characteristic curve (AUROC) values of our model for ACRN were higher than those of the previously established Asia–Pacific Colorectal Screening (APCS), Korean Colorectal Screening (KCS), and Kaminski’s scoring models [AUROC (95% confidence interval): model in the current study, 0.673 (0.648–0.697); vs. APCS, 0.588 (0.564–0.611), P < 0.001; vs. KCS, 0.602 (0.576–0.627), P < 0.001; and vs. Kaminski’s model, 0.586 (0.560–0.612), P < 0.001].
Conclusion
In a young population, a predicted probability model can assess the risk of ACRN more accurately than existing models, including the APCS, KCS, and Kaminski’s scoring models.