Outcomes of Patients After Liver Transplantation
Outcomes of Patients After Liver Transplantation
All analyses used Organ Procurement and Transplant Network (OPTN)/United Network for Organ Sharing (UNOS) data from February 27, 2002 until December 14, 2012. The start date of February 27, 2002 was the inception of MELD-based allocation, and the first date waitlist candidates with HPS could receive MELD exception points. Follow-up time for waitlist candidates with HPS waitlisted before this date began on the date of the first approved MELD exception after the start date.
The HPS cohort included all adult (age 18 years or older) waitlist candidates registered for their first liver transplant who applied for an HPS exception on or after February 27, 2002, had documented HPS based on data provided in the exception narrative (Table 1), and at least 1 exception application approved. These criteria were used because <5% of exception applications included the primary data to meet strict HPS diagnostic criteria. We excluded patients with portopulmonary hypertension miscoded as HPS. Each exception narrative was reviewed by a single investigator (SB) with a random sample receiving a secondary review (DG). Waitlist candidates might have been listed before implementation of MELD-based allocation.
The non-HPS cohort included all adult waitlist candidates registered for their first transplantation on or after February 27, 2002. We excluded patients who received non-HPS exceptions to create a comparison group whose waitlist priority was based on laboratory MELD score (this included the 63 excluded portopulmonary hypertension exception patients misclassified as HPS). Secondary analyses were restricted to a more focused non-HPS comparator group whose laboratory MELD score at waitlisting (determining waitlist priority) was 21–23, as HPS exception patients initially receive 22 MELD points.
Our main outcome was patient survival. Pre-transplantation death was defined by UNOS removal code of "died" and UNOS removal code "too sick to transplant" or "other" in the setting of a confirmed Social Security Death Master File death date within 90 days of waitlist removal. Death within a short time from waitlist removal is reflective of severity of illness and viewed as equivalent to dying on the waitlist.
Pre-transplantation Oxygenation and Outcomes in HPS. We first fit competing risk Cox regression models to evaluate pre-transplantation survival, considering transplantation as a competing risk, as it influences the probability of waitlist removal for death or clinical deterioration. Death on the waitlist or within 90 days of removal was the outcome, and all other outcomes were censored (eg, condition improved). We categorized HPS patients using room-air PaO2 at the time of initial exception approval using previously defined PaO2 cut points (ie, <50 mm Hg, 50–59 mm Hg, and ≥60 mm Hg). We estimated the PaO2 of patients with room-air pulse oximetry only using formulas described previously. We analyzed the entire HPS cohort and, secondarily, the restricted cohort with confirmed PaO2 values.
Potential covariates considered were sex, race/ethnicity, age, and laboratory MELD score at exception approval, blood type, serum albumin at listing, primary diagnosis (as defined by UNOS coding), and UNOS region. We used robust standard errors to account for correlation due to patient clustering by UNOS region. We evaluated time period of exception (2002–2004, post-MELD implementation; 2005–2009, after a national consensus conference proposed more formalized MELD exception policies; and 2010–2012, simplified application process).
We evaluated post-transplantation mortality by analyzing the subset of HPS exception deceased donor transplant recipients (98% of all HPS transplant recipients) in order to adjust for key donor factors associated with post-transplantation survival. We fit Cox regression models that evaluated race/ethnicity, age, final laboratory MELD score, and serum albumin at transplantation, blood type, diagnosis, UNOS region, time period of exception, and donor-risk index.
We determined optimal PaO2 cut points to predict post-transplantation mortality to refine HPS exception policies. We evaluated the subset with documented PaO2 values using cubic splines, a statistical method to evaluate for thresholds (inflection points or knots) in the relationship between an exposure (oxygenation) and outcome (death). Cubic splines are a superior method for modeling the relationship between a continuous exposure and an outcome that does not follow a simple linear relationship, and tests for the best model fit based on inflection points in the data. We fit unadjusted cubic spline models with 2, 3, and 4 knots to identify inflection points in the data, and chose final cut points based on the best model fit. We then refit the pre- and post-transplantation survival models, as mentioned, using the new PaO2 categories, and compared model fit using the Akaike Information criterion (AIC). The AIC is a measure of the relative quality of the statistical model and can be used to identify and compare the fit of a model, with the optimal fitted model having the minimum value of AIC.
HPS vs Non-HPS Waitlist Outcomes. The demographic and clinical variables between HPS and non-HPS patients were compared using Fisher's exact tests and χ tests for categorical variables and 2-sample t tests or Wilcoxon rank-sum tests for continuous variables, depending on the distribution of the data. We fit multistate survival models to compare overall survival in HPS vs non-HPS patients. These models are considered the best approach to studying outcomes in transplantation candidates and account for transitions from pre-transplantation to post-transplantation states, with transplantation considered an intervening state rather than a censor or competing risk. We assumed proportional baseline hazards and fit Cox regression models as Markov proportional hazard models. The transition state of transplantation was fit as an interaction term to account for variable survival time in the pre- vs post-transplantation states. Survival time for HPS patients was analyzed 2 ways: time from listing or from receipt of exception points. To then determine if differences in overall survival were due to differences in pre- and/or post-transplant survival, we fit competing risk Cox (pre-transplantation) and Cox (post-transplantation) models, described here, with listing laboratory MELD score in the pre-transplantation model, and laboratory MELD score at transplantation in the post-transplantation model.
Covariates were selected for inclusion in final multivariable models if they were associated with the outcome (P > .2) or confounded the relationship between the primary exposure and the outcome by changing the hazard ratio (HR) by 10%.
Institutional Review Board Approval was obtained from the University of Pennsylvania and the University of Texas-Houston. All statistical analyses were performed using Stata 13.0 software (College Station, TX).
Materials and Methods
All analyses used Organ Procurement and Transplant Network (OPTN)/United Network for Organ Sharing (UNOS) data from February 27, 2002 until December 14, 2012. The start date of February 27, 2002 was the inception of MELD-based allocation, and the first date waitlist candidates with HPS could receive MELD exception points. Follow-up time for waitlist candidates with HPS waitlisted before this date began on the date of the first approved MELD exception after the start date.
Study Sample
The HPS cohort included all adult (age 18 years or older) waitlist candidates registered for their first liver transplant who applied for an HPS exception on or after February 27, 2002, had documented HPS based on data provided in the exception narrative (Table 1), and at least 1 exception application approved. These criteria were used because <5% of exception applications included the primary data to meet strict HPS diagnostic criteria. We excluded patients with portopulmonary hypertension miscoded as HPS. Each exception narrative was reviewed by a single investigator (SB) with a random sample receiving a secondary review (DG). Waitlist candidates might have been listed before implementation of MELD-based allocation.
The non-HPS cohort included all adult waitlist candidates registered for their first transplantation on or after February 27, 2002. We excluded patients who received non-HPS exceptions to create a comparison group whose waitlist priority was based on laboratory MELD score (this included the 63 excluded portopulmonary hypertension exception patients misclassified as HPS). Secondary analyses were restricted to a more focused non-HPS comparator group whose laboratory MELD score at waitlisting (determining waitlist priority) was 21–23, as HPS exception patients initially receive 22 MELD points.
Outcomes
Our main outcome was patient survival. Pre-transplantation death was defined by UNOS removal code of "died" and UNOS removal code "too sick to transplant" or "other" in the setting of a confirmed Social Security Death Master File death date within 90 days of waitlist removal. Death within a short time from waitlist removal is reflective of severity of illness and viewed as equivalent to dying on the waitlist.
Statistical Analysis
Pre-transplantation Oxygenation and Outcomes in HPS. We first fit competing risk Cox regression models to evaluate pre-transplantation survival, considering transplantation as a competing risk, as it influences the probability of waitlist removal for death or clinical deterioration. Death on the waitlist or within 90 days of removal was the outcome, and all other outcomes were censored (eg, condition improved). We categorized HPS patients using room-air PaO2 at the time of initial exception approval using previously defined PaO2 cut points (ie, <50 mm Hg, 50–59 mm Hg, and ≥60 mm Hg). We estimated the PaO2 of patients with room-air pulse oximetry only using formulas described previously. We analyzed the entire HPS cohort and, secondarily, the restricted cohort with confirmed PaO2 values.
Potential covariates considered were sex, race/ethnicity, age, and laboratory MELD score at exception approval, blood type, serum albumin at listing, primary diagnosis (as defined by UNOS coding), and UNOS region. We used robust standard errors to account for correlation due to patient clustering by UNOS region. We evaluated time period of exception (2002–2004, post-MELD implementation; 2005–2009, after a national consensus conference proposed more formalized MELD exception policies; and 2010–2012, simplified application process).
We evaluated post-transplantation mortality by analyzing the subset of HPS exception deceased donor transplant recipients (98% of all HPS transplant recipients) in order to adjust for key donor factors associated with post-transplantation survival. We fit Cox regression models that evaluated race/ethnicity, age, final laboratory MELD score, and serum albumin at transplantation, blood type, diagnosis, UNOS region, time period of exception, and donor-risk index.
We determined optimal PaO2 cut points to predict post-transplantation mortality to refine HPS exception policies. We evaluated the subset with documented PaO2 values using cubic splines, a statistical method to evaluate for thresholds (inflection points or knots) in the relationship between an exposure (oxygenation) and outcome (death). Cubic splines are a superior method for modeling the relationship between a continuous exposure and an outcome that does not follow a simple linear relationship, and tests for the best model fit based on inflection points in the data. We fit unadjusted cubic spline models with 2, 3, and 4 knots to identify inflection points in the data, and chose final cut points based on the best model fit. We then refit the pre- and post-transplantation survival models, as mentioned, using the new PaO2 categories, and compared model fit using the Akaike Information criterion (AIC). The AIC is a measure of the relative quality of the statistical model and can be used to identify and compare the fit of a model, with the optimal fitted model having the minimum value of AIC.
HPS vs Non-HPS Waitlist Outcomes. The demographic and clinical variables between HPS and non-HPS patients were compared using Fisher's exact tests and χ tests for categorical variables and 2-sample t tests or Wilcoxon rank-sum tests for continuous variables, depending on the distribution of the data. We fit multistate survival models to compare overall survival in HPS vs non-HPS patients. These models are considered the best approach to studying outcomes in transplantation candidates and account for transitions from pre-transplantation to post-transplantation states, with transplantation considered an intervening state rather than a censor or competing risk. We assumed proportional baseline hazards and fit Cox regression models as Markov proportional hazard models. The transition state of transplantation was fit as an interaction term to account for variable survival time in the pre- vs post-transplantation states. Survival time for HPS patients was analyzed 2 ways: time from listing or from receipt of exception points. To then determine if differences in overall survival were due to differences in pre- and/or post-transplant survival, we fit competing risk Cox (pre-transplantation) and Cox (post-transplantation) models, described here, with listing laboratory MELD score in the pre-transplantation model, and laboratory MELD score at transplantation in the post-transplantation model.
Covariates were selected for inclusion in final multivariable models if they were associated with the outcome (P > .2) or confounded the relationship between the primary exposure and the outcome by changing the hazard ratio (HR) by 10%.
Institutional Review Board Approval was obtained from the University of Pennsylvania and the University of Texas-Houston. All statistical analyses were performed using Stata 13.0 software (College Station, TX).
