The primary end point was mortality after listing for lung transplantation. We used Cox proportional hazards regression analysis to examine the relationship between serum albumin at the time of listing and transplant waiting-list mortality rate, while adjusting for important covariates. Survival time was computed starting from the date of listing for transplantation. In our primary analysis, those who did not experience an end point (transplantation, removal/delisting, or death before transplantation) were censored on August 1, 2007. Those who were transplanted were censored at the date of transplantation, and those who were removed from the list were censored at the date of delisting. To examine the nature of the relationship between serum albumin concentration and mortality (whether there was a graded or threshold relationship), we categorized serum albumin concentration by quintiles, and regrouped the three middle quintiles into equispaced categories. The resulting five categories were as follows: (1) ≤ 3.3; (2) 3.4 to 3.7; (3) 3.8 to 4.1; (4) 4.2 to 4.5; and (5) ≥ 4.6 g/dL. Because a monotonic mortality trend was found across these five categories of serum albumin, we conducted Cox analyses with serum albumin as a continuous predictor. Models were adjusted for variables that influence survival in patients with IPF (age, gender, race, percentage of predicted FVC, flow rate of supplemental oxygen required at rest [O2 Req], 6-min walk distance [6MWD], mean pulmonary artery pressure [MPAP] as determined by right-heart catheterization, and lung allocation score [LAS])23 and variables that affect serum albumin (serum creatinine, body mass index [BMI], ever-smoker status, corticosteroid use, and diabetes mellitus).10 Covariates were added to the model in three steps: (1) demographic variables; (2) measures of pulmonary status; and (3) conditions (or markers of conditions) that can affect serum albumin levels: BMI (as a marker of malnutrition), serum creatinine (as a marker of renal dysfunction/proteinuria), corticosteroid use, diabetes mellitus, and ever-smoker status. Covariates with missing data were imputed using the mean or the mode of the distribution, and for each such covariate, we included a 0/1 indicator variable that flagged those with imputed values. In sensitivity analyses, we reran the final model after excluding those with imputed values (6MWD: n = 196; ever-smoker status: n = 194; MPAP: n = 183; O2 Req: n = 98; percentage of predicted FVC: n = 26; corticosteroid use: n = 10; diabetes mellitus: n = 4). We added the LAS to the final model to test if albumin has prognostic value independent of the LAS score. Since a major reason for removal from the transplant waiting list is worsening health, we conducted a sensitivity analysis in which everyone who was removed from the list was treated as if he/she had died on the date of removal. To address the possible bias from censoring due to transplantation, we also performed sensitivity analysis in which we assumed that everyone who received transplants died on the day of the transplantation. Further, we also performed sensitivity analyses including post-transplant survival, including a time-varying covariate for the performance of lung transplantation. In supplementary analyses, we examined selected interaction terms (age × albumin and gender × albumin). The proportional hazards assumption was assessed using log-log plots. All tests were two-tailed, and p values of < 0.05 were required for statistical significance. All statistical analyses were performed using statistical software (SAS version 9.1; SAS Institute; Cary, NC; and MedCalc for Windows, version 220.127.116.11; MedCalc Software; Mariakerke, Belgium).