Assessment of Postdonation Outcomes in US Living Kidney Donors Using Publicly Available Data Sets

Key Points Question What are the postdonation outcomes of living kidney donors? Findings In this cohort study of 10 869 living kidney donors from the ImmPort open access data repository, 9558 individuals’ postdonation data were analyzed. Overall, 1406 living donors (14.7%) had postdonation events; the 4 most common events were hypertension, diabetes, proteinuria, and postoperative ileus, and most events that occurred more than 2 years after transplant were unrelated to surgical complications, occurring up to 40 years later. Meaning Aggregated data from publicly available clinical studies can provide insights into short-term and long-term complications affecting living donors.

(3) Finally, the data in this cohort was compiled and classified into five main categories, namely (A) demographics, (B) donor relationship to recipient, and (C) pre-transplant, (D) intraoperative, and (E) post-transplant data (eTable 5). We then standardized the data in the fields that were common across studies, in terms of units of measurement and definitions. For example, we converted all units for weight measurements to the SI unit of kilogram and centimeter for height. In terms of definitions in common fields, the same data representation in two clinical studies could have different meanings. For example, in a field that coded for the donor's relationship to the recipient in two studies, a '2' can denote 'parent-child' in one study but 'siblings' in another. Hence, when merging across studies, we converted all numeric entries to English words, as defined in the respective data dictionaries, and standardized synonymous English words. For the fields that were both unique to a study (due to the design of a clinical study), and also fell into one of the five aforementioned categories, we added new columns for these fields and tag the empty records as missing ('NA').
The consolidated data is available at http://www.immport.org/resources/TTN. The data includes all the processed clinical study files that are used in this manuscript from ImmPort for solid transplantation. We also included a Cytoscape file of the trajectory map for user visualization and download.

Demographic and relationship information from immTransplant
We extracted the gender and racial information from 10,869 LKDs and removed records with missing information. For trend comparisons, we used 10,833 LKDs with age at donation and gender, and 10,494 with age and racial information. For the relationship analysis, we used 9,526 LKDs with age at donation and the donor's relationship to the recipient. For racial group comparisons with UNOS/OPTN, we only considered the three racial groups with insufficient individuals for reasonable comparisons (an arbitrarily chosen threshold of 100 individuals), namely Caucasian, African, and Asian American. Note that the proportions in eFigure 4 were calculated based on the total number of LKDs within each racial group

Assessment of representativeness for immTransplant
In order to demonstrate that the entire immTransplant data is representative of the national data demographically for downstream analyses, we compared the demographic trends of all the LKDs in immTransplant versus the U.S. UNOS/OPTN national registry. There were 128,407 in the UNOS/OPTN dataset. Overall, there are more female than male LKDs in both datasets; such gender disparity in LKDs has been described previously 52-54 ( Figure 1A). Even as we further stratified the proportion of male and female LKDs by their ages at kidney donation, we still observed that both the female and male LKD distributions in immTransplant and the UNOS/OPTN dataset are very similar (Kolmogorov-Smirnov test of heterogeneity, D = 0.15, non-significant p = 0.36 for female; D = 0.15, p = 0.45 for male).
However, within each dataset, a higher proportion of female LKDs was not uniformly observed across all ages of LKDs. In both immTransplant and UNOS/OPTN, we noticed higher proportions of female LKDs relative to male at ages over 25 years, whereas we observed comparable proportions of female and male LKDs who donated at ages less than 25 years old ( Figure 1A and eFigure 3). Because the age of 25 is the mean age of childbearing mothers in the U.S., 55 we further investigated this gender disproportionality of donors, by examining the donors' relationships to the recipient. Specifically, we explored whether there are any trends between the gender distributions of donors in terms of their relationship to the recipients, who are less or more than 25 years of age (eFigure 2). In both immTransplant and UNOS/OPTN data, relatively more women (than men) donate kidneys after age 25 (as compared to LKDs age  25), preferentially to their spouses or offspring (eFigure 2).

Network construction and analyses
We obtained 1,406 LKDs, documented with at least one post-transplant adverse outcome (LKDOs). We further removed records that had no date associated with the outcome, and those in which the initial diagnosis already occurred before transplant. Finally, we used 1,401 LKDOs with 36 documented outcomes (eTable 4).
For trajectory network construction, each node (circle) is an event, or condition. The size of a node represents the proportion of LKDOs having that event (please refer to eTable 4 for actual numbers). We connect two events with an edge when both post-donation events occur for at least one LKDO. The network is drawn using the software, Cytoscape. 42 Since the two events can be ordered according to the time it first occurs after transplant, we connect the two events by a directed edge (arrow), such that an event X precedes an event Y in at least one LKDO's post-donation timeline. We started with the time of transplantation as the leftmost black node. The thickness of the edge represents the number of LKDOs having that trajectory from event X to Y. The nodes are ordered on the horizontal axis according to the mean time of initial diagnosis, but the relative positions are not drawn to scale, i.e. the length of the edges are not necessarily scaled.

Glomerular filtration rate trends in LKDs
We imposed the following criteria on the 1,401 LKDOs used in the trajectory map and 8,152 LKDs with no conditions in RELIVE: (1) we retained an individual, if and only if, we have one pre-transplant GFR value as well as at least one post-transplant value with non-missing dates; (2) we assumed all missing dates for pre-transplant GFR measurement (those marked as 'NA') were performed on the day of transplant, hence these were set to 0; (3) we removed an individual if the pre-transplant GFR value was obtained on a date (number of days) after transplant (as denoted by a positive value in the date that the GFR was measured); (4) we removed an individual if all the posttransplant GFR values were recorded before the transplantation (denoted by a negative value in the date for GFR measurement). These filters resulted in 32 LKDOs with post-transplant hypertension and 75 LKDs with no conditions; the rest of the conditions have less than 10 individuals.

Kaplan-Meier analyses
All Kaplan-Meier analyses were performed, and Kaplan-Meier curves plotted using the R package survminer. We obtained right-censored data from the RELIVE data in immTransplant that were extracted from the National Death Index. The 'renal failure' endpoint is defined by either of the five events, whichever comes first in LKDOs with multiple endpoints: "post-operative renal failure", "post-operative dialysis", "kidney transplant", "kidney transplant waiting list" and "chronic/maintenance dialysis". For dates, a positive value indicates number of days posttransplant, while a negative value indicates number of days pre-transplant, and the day of transplant is '0'; 'days' are then converted to 'years'. 1,401 LKDOs are used to analyze overall event-free 'survival', or renal-failure-free probability, among LKDOs. For analyses in Figure 4, we stratify the immTransplant data to visualize the renalfailure-free survival of LKDOs for three scenarios: (1) the entire LKDOs sample, (2) among the LKDOs stratified by the five most frequently occurring conditions in our dataset (i.e. the five largest nodes in the trajectory network), and (3) among the LKDOs in (2) but stratified by the number of conditions they have. We also assess the survival of 831 LKDOs with only a single condition from the six most frequently occurring single conditions in the dataset (diabetes, dysrhythmia, hypertension, post-operative ileus, proteinuria, stroke); or 1,120 if we consider additionally LKDOs with at least one of the six conditions. We marked right-censored data points as individuals who died without renal failure before 40 years post-transplant.

Statistical Analyses
All statistical analyses were performed using statistical software, R, and RStudio as the integrated development environment. The Kolmogorov-Smirnov (KS) tests (two-sided) were implemented using the 'ks.test' function in the 'stats' R package. The KS test is a non-parametric test to compare the cumulative distribution functions between a sample and a reference distribution. We included the D statistic from the KS test as it measures the absolute maximum distances between the two cumulative distributions. The multinomial goodness of fit chi-squared test was implemented using the 'chisq.test' function in the 'stats' R package. We have also included the chi-square statistic (2) and the degree of freedom (df). The Fisher's exact test was implemented using the 'fisher.test' function in the 'stats' R package. We included the odds ratio as an estimate for the effect size. A p value cutoff of 0.05 (and below) is considered statistically significant in all statistical tests.