P117 Machine learning tool provides new insights into risk assessment in pulmonary endarterectomy. (12th November 2019)

• Record Type: Journal Article
• Authors: Bunclark, K; Liley, J; Newnham, M; Ruggiero, A; Cannon, JE; Coghlan, G; Lordan, J; Howard, L; Jenkins, D; Johnson, M; Kiely, DG; Ng, C; Screaton, N; Sheares, K; Taboada, D; Tsui, S; Wort, SJ; Pepke-Zaba, J; Toshner, M
• Is Part Of: Thorax. Volume 74(2019)Supplement 2
• Page Start: A154
• Page End: A155
• Publication Date: 2019-11-12
• Subjects: Chest -- Diseases -- Periodicals; Thorax; Chest -- Diseases; Periodicals; Periodicals; 617.54
• DOI: 10.1136/thorax-2019-BTSabstracts2019.260
• Languages: English
• ISSNs: 0040-6376

Abstract

Abstract : Background: Chronic thromboembolic pulmonary hypertension (CTEPH) is an uncommon disorder characterised by persistent obstruction of the pulmonary arteries by thromboembolic material, usually following an acute pulmonary embolus. 1 Pulmonary endarterectomy (PEA) is the gold standard treatment for eligible patients and is potentially curative. 1 Whilst pre-operative parameters have been associated with post-operative mortality no sytematic method for predicting individualised PEA risk presently exists. Objectives: To identify pre-operative risk factors of 90 day mortality (90DM), five year mortality (5YM) and improvement in self-reported functional status (DQ) following PEA for inclusion in a clinically-implementable risk prediction tool. Methods: Consecutive patients undergoing PEA for CTEPH at Royal Papworth Hospital, UK between 2007 and 2017 were included. Potential pre-operative predictors including patient demographics, medical history and results of functional, physiological and patient self-reported measures were included in a hypothesis-free approach. Three stastical predictive models were considered (linear regression, lasso regression and random forest), each of which were calibrated, fitted and assessed using cross-validation ensuring internal consistency. Results: 1336 individuals were included in risk modelling. 96 patients (6.4%) died within 90 days of hospital discharge and 154 (11.5%) within five years of PEA. Random forest based predictions were more accurate than linear or lasso based. All post-operative outcomes were predicted well from pre-operative variables (90DM: AUROC 0.82 (95% CI 0.78, 0.87); 5YM: C-Index 0.81 (0.76, 0.85); DQ (Spearman's correlation 0.47 (0.43, 0.51)) using random forest modelling. The strongest individual pre-operative predictor of 90DM and 5YM was left atrial dilatation and of DQ, pulmonary vasodilator therapy. Post-hoc analysis confirmed not only excess mortaltiy following PEA in those with left atrial dilatation secondary to diastolic dysfunction but adverse functional, haemodynamic and patient-reported outcomes in this group. Conclusions: Outcomes from PEA can be predicted from pre-operative observations to a clinically useful degree enabling individualised risk prediction. Post-hoc analysis highlights the under-recognised adverse outcomes in those with left atrial dilatation. We present an online application to facilitate use of these tools. Further work validating our model in other centres will be necessary and aided by the open availability of our methodology. Reference: Galiè N, Humbert M, Vachiery J-L, et al . 2015 ESC/ERS Guidelines for the diagnosis and treatment of pulmonary hypertension. Eur Heart J 2016;37(1):67–119.