Multiple Listing in Lung Transplant Candidates

A Cohort Study

Joshua J. Mooney; Lingyao Yang; Haley Hedlin; Paul Mohabir; Gundeep S. Dhillon


American Journal of Transplantation. 2019;19(4):1098-1108. 

In This Article



This study used data from the Scientific Registry of Transplant Recipients (SRTR). The SRTR data system includes data on all donors, waitlisted candidates, and transplant recipients in the United States, submitted by the members of the Organ Procurement and Transplantation Network. The Health Resources and Services Administration (HRSA), US Department of Health and Human Services, provides oversight to the activities of the OPTN and SRTR contractors. Using SRTR standard analysis files, we identified all lung transplant candidates on the waitlist during the post-LAS period of May 4, 2005 and March 5, 2015. To be included in our analysis, candidates had to be at least 18 years old, be awaiting their initial lung transplant, and have a nonzero LAS. We removed candidates who had a diagnosis of "other." Candidates who were listed simultaneously by at least two different centers were classified as multiple listed (ML) candidates. The date of active listing at the second center was used as the date of multiple listing. Candidates listed only at a single center or whose active listings at different centers did not overlap were classified as single listed (SL) candidates. This study received an exemption from the Stanford University Institutional Review Board as it uses previously collected de-identified data.

Statistical Analysis

Descriptive variables were generated from the existing SRTR data fields and compared between SL and ML candidates. Karnofsky performance scale scores were categorized as needed into three functional status categories according to previously described thresholds.[12] Categorical variables were displayed as number (%) and continuous variables were displayed as median (interquartile range, IQR). To understand whether candidates who multiple list do so at a transplant center, organ procurement organization (OPO), or OPTN region with a higher transplant rate, the transplant rates for all transplant centers, OPOs, and OPTN regions were calculated and then the difference between the multiple listing transplant rate and initial listing transplant rate was calculated. Transplant rate was determined by calculating the number of transplants per 100 waitlist years during the study period. To determine whether candidates who multiple list do so at specific transplant centers or in a specific OPO or OPTN region, the proportion of all multiple listed candidates by individual transplant center, OPO, and OPTN region was calculated. To understand the association between sociodemographic and clinical characteristics with multiple listing, we fit a multivariable logistic regression model with multiple listing status as the outcome to all candidates who met the inclusion criteria for this study. We fit an adjusted, stratified Cox proportional hazards regression model to estimate the association between multiple listing status and the outcome of transplant or death in the matched dataset, ie, the matched ML candidates and the SL candidates matched to them.

Matching. In order to balance the covariates observed in the SL and ML groups, we a priori selected the following variables to be used to match SL patients to ML patients in a 4:1 ratio: lung diagnosis group, time on the waitlist, initial LAS, year of first listing, initial OPTN region, height, antibody crossmatch requirement, blood type, single/bilateral transplant preference, gender, age at listing, race, and median household income of the candidate's county. Given the number of potential confounders selected a priori, we employed a matching method that allowed us to match SL and ML candidates. We matched candidates using the following criteria: in the same diagnosis group, same year of first listing, and within a caliper distance equal to 10% of the median on initial LAS and height, and a propensity score.[13] Pairs of SL and ML candidates with distances outside the caliper were not matched. The propensity score was created by fitting a logistic regression with an outcome of multiple listing status and covariates including DSA, antibody crossmatch requirement, blood type, single/bilateral transplant preference, gender, age at listing, race, and household income. To ensure that the SL candidates were still at risk of transplant or death at the time that a ML candidate was multiple listed, we matched a SL candidate to a ML candidate only if the SL candidate's time on the waitlist was longer than the time until multiple listing for a ML candidate. We checked the matching by comparing the frequency distribution of the categorical variables between the ML and SL candidates and by comparing the standardized mean difference, a measure of the difference in means between two groups expressed in units of standard deviation, between single and multiple listed candidates in the unmatched and matched datasets.[14]

Modeling. A logistic regression model was fit in the full dataset including all candidates eligible for inclusion in our study to assess the relationship between multiple listing status and the following variables: age at listing, diagnosis group, gender, race (white vs nonwhite), initial listing LAS, height, blood type, insurance type (public vs private), highest education level attained (high school education or below, college education, postcollege graduate degree), antibody crossmatch requirement, and county median household income.

To assess whether multiple listed candidates have a decreased likelihood of waitlist death or increased likelihood of transplant, two Cox proportional hazards models stratified by the matched one ML: four SL sets were fit in the matched dataset, one for the waitlist outcome of death, and another for the waitlist outcome of transplant. The primary exposure of interest was whether the candidate was multiple listed. Waitlist outcomes were determined by the SRTR reported removal code. Candidates removed from the waitlist for being too sick for transplant were considered to have the same outcome as those removed for death. Time 0 for both the SL and ML candidates was set as the time of multiple listing for the ML patient in the matched set; this study design approach was chosen to reduce bias as comparable candidates would be evaluated during the follow-up period when candidates are multiple listed. Candidates who were removed from the list in error, not in contact with their center, listed by a program that was inactive for at least 2 years, transferred to another center, removed from the list for an unspecified reason, or remained alive on the waitlist were right censored. Candidates were followed until the event of interest or until right censoring. SL candidates who were not matched to a ML candidate were not included in this analysis. Models were adjusted for status of multiple listing, age at listing, gender, race, initial listing LAS, height, blood type, antibody crossmatch requirement, single/bilateral lung preference, OPTN region, insurance type, and median household income. The proportional hazards assumption was assessed by visual examination of Schoenfeld residual plots and Kaplan-Meier plots.

In exploratory secondary analyses, we fit our primary models to a subset of the matched candidates where the multiple listing candidates had higher medical urgency as indicated by a listing LAS of at least 40. In sensitivity analyses, we additionally adjusted the multivariable logistic regression by initial listing OPTN region and functional status and the primary and secondary waitlist outcome analyses by functional status.

All statistical analyses were performed with R 3.1.3.[15] To address missing data, imputations were applied to all models. Multiple imputations were used in the Cox proportional hazards model and the logistic regression models where we calculated confidence intervals (CI). A single imputation was used in the model used to create the propensity scores. Matching was performed using the R package "Matching" and multiple imputations were performed with the "mice" package.[16,17] All tests were two sided and conducted at the 0.05 significance level and all model estimates are shown with 95% CI.