Risk stratification in lung resection surgery

Abstract

THESIS ABSTRACT Objectives Lung cancer is the leading cause of cancer death worldwide. For patients with anatomically resectable disease and sufficient physiological reserve, surgical resection is recognised as the best treatment option. However, less than 20% of all patients with lung cancer in the United Kingdom receive this treatment. It is therefore crucial to ensure that patient selection for surgery is effective and accurate. The initial aim of this study was to identify existing clinical prediction models designed or utilised to predict short-term mortality after lung resection and statistically validate their performance using contemporary patient data. The subsequent aim was to use large regional and national databases to develop and externally validate a new clinical risk prediction model designed to predict 90-day mortality after lung resection. Methods A systematic review of the literature was performed to identify existing models and a regional clinical database comprised of 6600 patients from two centres was used to externally validate six of these models. Logistic regression was then used to develop a new model (RESECT-90) designed to predict 90-day mortality after lung resection. The model was internally validated using bootstrapping to adjust for in-sample optimism. A large national database of 12241 patients from 12 centres across the UK was then used to externally validate the RESECT-90 model. Results A total of 22 models were identified from the systematic review and six were externally validated. None demonstrated sufficient discrimination, calibration and clinical validity to be recommended for use in contemporary practice. The newly developed RESECT-90 model was comprised of 12 variables (age, sex, Performance Status score, percentage predicted diffusion capacity of the lung for carbon monoxide, pre-operative serum creatinine, body mass index, pre-operative anaemia, pre-operative arrhythmia, laterality of resection, surgical approach, number of resected bronchopulmonary segments, confirmed or suspected malignant disease). Internal validation demonstrated good model discriminatory and calibratory performance. External validation also demonstrated good model discrimination. However, calibration varied between centres with some evidence of overall model miscalibration. After recalibration, all measures of calibration indicated good overall performance. Conclusions None of the existing models used to predict short-term mortality after lung resection published prior to the start of this research were found to be entirely suitable for use in contemporary practice. Therefore, the RESECT-90 model was developed. It comprises 12 clinically relevant variables and was found to have good predictive ability on internal validation. A subsequent large-scale national external validation and recalibration demonstrated good model performance. Whilst ongoing model evaluation and updating will be required, these results demonstrate that the RESECT-90 model is suitable for use in contemporary thoracic surgery practice in the UK as part of the pre-operative risk stratification process. If used routinely throughout the country, the greater accuracy of the RESECT-90 model in comparison to existing models has the potential to improve outcomes for patients with lung cancer on a national level.

Date of Award	1 Aug 2024
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Stuart Grant (Supervisor) & Glen Martin (Supervisor)