Geographic Disparities in Lung Transplant Rates

Martin Kosztowski; Sheng Zhou; Errol Bush; Robert S. Higgins; Dorry L. Segev; Sommer E. Gentry


American Journal of Transplantation. 2019;19(5):1491-1497. 

In This Article


Data Source

This study used data from the Scientific Registry of Transplant Recipients (SRTR). The SRTR data system includes data on all donors, wait-listed candidates, and transplant recipients in the United States that are submitted by the members of the OPTN, and has been described elsewhere.[7] The U.S. Department of Health and Human Services Health Resources and Services Administration (HRSA), provides oversight to the activities of the OPTN and SRTR contractors.

Study Population

We looked at 7131 prevalent and incident lung transplant registrants who were on the waiting list between February 19, 2015 and March 31, 2017. We included all patients 12 years or older, since this is the age at which patients are assigned an LAS. We excluded candidates who were listed for simultaneous transplants (ie, lung/kidney).

Calculating Transplant Rates Within DSAs

We defined lung transplant rates per DSA and per LAS category as the number of transplants performed for candidates in that LAS category divided by the number of active person-years spent waiting in that LAS category. Inactive time on the waiting list was excluded because inactive candidates were not at risk for undergoing liver transplantation. If a candidate was active, then inactive, and then active again, only the active time on the waitlist was included. Because a candidate's LAS score varied, each person-day of waiting was counted in different risk sets according to that candidate's LAS category. Candidates were censored at death, waitlist removal for reasons other than LT, or administratively on March 31, 2017.

We calculated crude LT rates per person-year for each DSA. We used multilevel Poisson regression and empirical Bayes methods to estimate the adjusted LT rates per person-year for each DSA.[8–11] We adjusted for only those covariates that directly influence the rank list, as stated in section 10.4 of the OPTN lung allocation policy.[12] Covariates included blood type and LAS. The models allowed the baseline transplant rate per person-year to vary across DSAs using a random intercept, and thus we were able to estimate each DSA's adjusted transplant rate per person-year. To determine whether LT rates varied significantly among DSAs, we performed a likelihood ratio test comparing the multilevel Poisson model to the nonhierarchical Poisson model. To account for the nonlinearity of LAS in transplant rates and to calculate the expected increase in LT rate per LAS increase, LAS was categorized into groups (0–32, 32–34, 34–38, 38–42, 42–50, and 50–100). LAS scores are not rounded to integer values; for example, the score 33.8452 is in the range 32–34.

Incidence Rate Ratio (IRR) and Median Incidence Rate Ratio (MIRR)

To characterize variation in LT rates among DSAs, we derived the incidence rate ratio (IRR) of LT rate between each pair of DSAs that had active lung transplant programs using the adjusted multilevel Poisson regression results. The IRRs indicate the disparity in LT rate between each pair of DSAs. For each incidence rate ratio, the DSA with the higher LT rate was compared to the DSA with the lower LT rate, and thus all IRRs were greater than or equal to one. The median incidence rate ratio (MIRR) summarizes the between-DSA variation in LT rates and measures overall geographic disparity. The MIRR is a single quantity that is calculated from a multilevel model, so problems associated with multiple comparisons do not apply. For our study, the MIRR indicates the extent to which a candidate's DSA determines their relative rate of LT, or the expected increase in LT rate if a given candidate lists in another DSA with a higher LT rate. Ninety-five percent confidence intervals of MIRR were obtained using a bootstrap with 200 repetitions that resampled registrants in each DSA with replacement.

Annual MIRR

To determine how the MIRR has changed over time, we calculated the MIRR for each year between 2006 and 2017, using the same methods described earlier. Trend was determined using simple linear regression.

Statistical Analysis

All analyses were performed using Stata 14.2/MP for Windows (College Station, TX). For all analyses, P < .05 was considered statistically significant. MIRR was calculated using the Stata command xtmrho following a multilevel Poisson regression. All maps were generated using R 3.3.3 GUI for Mac OS.