Objective Estimates Improve Risk Stratification for Primary Graft Dysfunction after Lung Transplantation

Abstract

Primary graft dysfunction (PGD) is a major cause of early mortality after lung transplant. We aimed to define objective estimates of PGD risk based on readily available clinical variables, using a prospective study of 11 centers in the Lung Transplant Outcomes Group (LTOG). Derivation included 1255 subjects from 2002 to 2010; with separate validation in 382 subjects accrued from 2011 to 2012. We used logistic regression to identify predictors of grade 3 PGD at 48/72 h, and decision curve methods to assess impact on clinical decisions. 211/1255 subjects in the derivation and 56/382 subjects in the validation developed PGD. We developed three prediction models, where low-risk recipients had a normal BMI (18.5-25 kg/m(2) ), chronic obstructive pulmonary disease/cystic fibrosis, and absent or mild pulmonary hypertension (mPAP<40 mmHg). All others were considered higher-risk. Low-risk recipients had a predicted PGD risk of 4-7%, and high-risk a predicted PGD risk of 15-18%. Adding a donor-smoking lung to a higher-risk recipient significantly increased PGD risk, although risk did not change in low-risk recipients. Validation demonstrated that probability estimates were generally accurate and that models worked best at baseline PGD incidences between 5% and 25%. We conclude that valid estimates of PGD risk can be produced using readily available clinical variables.

Keywords: clinical research / practice; lung (allograft) function / dysfunction; lung failure / injury; lung transplantation / pulmonology.

Figure 1

Decision curve analysis of alternative strategies for prognostic models for primary graft dysfunction after lung transplantation. The plot compares three models against the alternatives of (a) considering everyone to be higher-risk (downward sloping dotted gray line) or (b) foregoing any prognostic modeling and treating no one as being at high risk (solid horizontal line). The y axis is net beneft (the tradeoff between true positive and false positive classifications for high risk status), and the x axis represents the threshold probability for classifying a patient as being at high risk for PGD. Net benefit will vary depending on the threshold risk because that threshold probability reflects the relative loss of missing high risk patients (fralse negatives) and of considering too many low risk patients (false positives). Because the threshold might vary with individual patients and their centers, the decision curves plot net benefit against a range of thresholds. In this setting, typical threshold probabilities might range from 5% to 20%. In that range of thresholds, the proposed prognostic models are not only simple to implement but also superior in net benefit in the classification of higher-risk PGD patients than treating all patients as being at high risk.