2022 World Conference on Lung Cancer (ePosters)-EP01.05-008. Pre-Trial Validation of a Risk Prediction Model for Pulmonary Nodules Trained on a Nested Case-Control Study with No External Cohort

EP01.05-008. Pre-Trial Validation of a Risk Prediction Model for Pulmonary Nodules Trained on a Nested Case-Control Study with No External Cohort

Back to course

Pdf Summary

The document discusses the validation and calibration of a risk prediction model for pulmonary nodules. The model was developed using a nested case-control study design and assessed using metrics such as discrimination (measured by the area under the ROC curve) and calibration (measured by the accuracy of risk estimates).<br /><br />To validate the model, a combination of bootstrap sampling and repeated k-fold cross-validation was used. The model was recalibrated to reflect the prevalence of the outcome in the target population using a Bayes factor adjustment. The estimated mean and confidence intervals for the recalibrated intercept and calibration slope were also determined. The AUC of the model was reported to be 0.758.<br /><br />The document acknowledges the challenges of assessing the calibration of a model when an external representative population is not available. The proposed approach aims to provide an uncertainty measure about the performance metrics and is applicable to small datasets and situations where no external validation is available.<br /><br />The document concludes that this approach can be used to expedite the implementation of biomarker models for early detection, given the limited availability of validation cohorts. It emphasizes the importance of appropriate calibration and suggests pursuing other methods if the calibration is not appropriate.<br /><br />The document includes references to previous work on assessing risk prediction models and validation. It also provides visual representations of the calibration plot and ROC curve for the model.<br /><br />Overall, the document presents a principled approach to validate and calibrate a risk prediction model for pulmonary nodules, providing insights into the model's performance and generalizability.

Asset Subtitle

Michael Nolan Kammer

Meta Tag

Speaker Michael Nolan Kammer

Topic Early Detection and Screening - Pulmonary Nodule

Keywords

risk prediction model

pulmonary nodules

validation

calibration

nested case-control study

discrimination

area under the ROC curve

bootstrap sampling

repeated k-fold cross-validation

Bayes factor adjustment