Model-Based versus Clinical Prediction of the Spirometric Response to Lung Volume Reduction Surgery

Abstract

Background: Lung volume reduction surgery (LVRS) improves symptoms and lung function in selected patients with severe emphysema. Objectives: We investigated whether models based on physiologic and radiologic predictors discriminated patients with a favorable from those with a poor spirometric response to LVRS. Methods: Data of a derivation cohort of 70 patients who had previously undergone LVRS served to develop two types of prediction models, lookup functions and logistic regression equations. Presence or absence of improvement in forced expiratory volume in 1 s (FEV₁) ≧300 ml and forced vital capacity (FVC) ≧500 ml represented dichotomous outcomes. The residual volume/total lung capacity ratio, CT-radiological emphysema heterogeneity scores and diffusing capacity, a marker of emphysema severity, were the predictors. Models were used to predict spirometric outcomes for a validation cohort of 60 emphysema patients referred for LVRS. Furthermore, the surgeon preoperatively estimated outcomes based on all available clinical data but blinded to model predictions. Spirometric changes within 6 months following surgery were compared to predictions. Results: Median FEV₁ in the validation cohort increased from 0.69 to 1.00 liters (+41%), and FVC from 2.07 to 2.78 liters (+29%; p < 0.05 for changes). Lookup functions and logistic regression equations identified patients experiencing major increases in FEV₁ ≧300 ml and FVC ≧500 ml with an accuracy quantified by areas under the receiver-operating characteristic curves of 0.72 to 0.76 (all areas >0.5, p < 0.05). Predictions by the surgeon had an accuracy of 0.71 to 0.78 (p = NS vs. models). Conclusions: The accuracy of models based on three predictors was fair and similar to assessment by an experienced surgeon based on all available clinical information. Prediction models may contribute to the consistent assessment of LVRS candidates.