Faculty Bibliography

UNet(SM) , the UNOS data collection and electronic organ allocation system, allows centers to specify organ offer acceptance criteria for patients on their kidney waiting list. We hypothesized that the system might not be fully utilized and that the criteria specified by most transplant centers would be much broader than the characteristics of organs actually transplanted by those centers. We analyzed the distribution of criteria values among waitlist patients (N = 304 385) between January 2000 and February 2009, mean criteria values among listed candidates on February 19, 2009 and differences between a center's specified criteria and the organs it accepted for transplant between July 2005 and April 2009. We found wide variation in use of criteria variables, with some variables mostly or entirely unused. Most centers specified very broad criteria, with little within-center variation by patient. An offer of a kidney with parameters more extreme than the maximum actually transplanted at that center was designated a 'surplus offer' and indicated a potentially avoidable delay in distribution. We found 7373 surplus offers (7.1% of all offers), concentrated among a small number of centers. The organ acceptance criteria system is currently underutilized, leading to possibly avoidable inefficiencies in organ distribution.

We recently showed that DonorNet 2007 has reduced the efficiency of kidney distribution in the United States, particularly for those with prolonged cold ischemia time (CIT), by requiring systematic allocation of all kidneys regardless of quality. Reliable early identification of those most likely to be discarded or significantly delayed would enable assigning them to alternate, more efficient distribution strategies. Based on 39 035 adult kidneys recovered for possible transplantation between 2005 and 2008, we created a regression model that reliably (AUC 0.83) quantified the probability that a given kidney was either discarded or delayed beyond 36 h of CIT (Probability of Discard/Delay, PODD). We then analyzed two PODD cutoffs: a permissive cutoff that successfully flagged over half of those kidneys that were discarded/delayed, while only flagging 7% of kidneys that were not eventually discarded/delayed, and a more stringent cutoff that erroneously flagged only 3% but also correctly identified only 34%. Kidney transplants with high PODD were clustered in a minority of centers. Modifications of the kidney distribution system to more efficiently direct organs with high PODD to the centers that actually use them may result in reduced CIT and fewer discards.

In 2007, UNOS released DonorNet 2007 (DN07) in hope of improving allocation equity and efficiency. We hypothesized that hard-to-place organs might be less efficiently handled through this regimented process. We analyzed associations between DN07 and center-level equity, number of refusals per organ and cold ischemia time (CIT). A total of 8244 kidney transplants between 1/2006 and 12/2006 (pre-DN07) were compared with 6029 transplants between 5/2007 and 2/2008 (post-DN07). Distribution equity was assessed by the Gini coefficient, changes in the number of refusals and CIT by negative binomial regression and discard rates by logistic regression. We estimated quantile-specific differences in CIT by bootstrapping. We found no significant change in center-level distribution equity after DN07. Number of refusals per organ increased by 20% (adjusted rate ratio (1.12)1.20(1.28), p < 0.001) at the patient level and 11% (ARR (1.07)1.11(1.16), p < 0.001) at the center level. Regression models of CIT showed no global change in CIT associated with DN07, but those kidneys with the longest CIT pre-DN07 had statistically significantly longer CIT post-DN07. The discard rate also increased significantly (ARR (1.06)1.11(1.17), p < 0.001). DN07 has not improved equity or efficiency in allocation of deceased donor kidneys, and may be harming the allocation of hard-to-place kidneys.

This paper describes cortical analysis of 19 high resolution MRI subvolumes of medial prefrontal cortex (MPFC), a region that has been implicated in major depressive disorder. An automated Bayesian segmentation is used to delineate the MRI subvolumes into cerebrospinal fluid (CSF), gray matter (GM), white matter (WM), and partial volumes of either CSF/GM or GM/WM. The intensity value at which there is equal probability of GM and GM/WM partial volume is used to reconstruct MPFC cortical surfaces based on a 3-D isocontouring algorithm. The segmented data and the generated surfaces are validated by comparison with hand segmented data and semiautomated contours, respectively. The L(1) distances between Bayesian and hand segmented data are 0.05-0.10 (n = 5). Fifty percent of the voxels of the reconstructed surface lie within 0.12-0.28 mm (n = 14) from the semiautomated contours. Cortical thickness metrics are generated in the form of frequency of occurrence histograms for GM and WM labelled voxels as a function of their position from the cortical surface. An algorithm to compute the surface area of the GM/WM interface of the MPFC subvolume is described. These methods represent a novel approach to morphometric chacterization of regional cortex features which may be important in the study of psychiatric disorders such as major depression.

This paper describes the construction of cortical metrics quantifying the probabilistic occurrence of gray matter, white matter, and cerebrospinal fluid compartments in their correlation to the geometry of the neocortex as measured in 0.5-1.0 mm magnetic resonance imagery. These cortical profiles represent the density of the tissue types as a function of distance to the cortical surface. These metrics are consistent when generated across multiple brains indicating a fundamental property of the neocortex. Methods are proposed for incorporating such metrics into automated Bayes segmentation.