Date: Sunday, June 2, 2019
Session Time: 4:30pm-6:00pm
Presentation Time: 5:18pm-5:30pm
Location: Room 312
*Purpose: Identification of hard-to-place deceased donor livers is important due to its potential role in expedited placement. Determining hard-to-place must be based on data available at allocation (generation of the match run or time of offer).
*Methods: We investigated the utility of two sources of donor information: DonorNet and the Deceased Donor Registration (DDR). DonorNet has comprehensive information but may be impractical due to optional data fields. DDR information is required and more complete, but may be unavailable at match run. An Expedited Placement work group of transplant professionals provided guidance on data likely to be known by organ procurement organizations at the time of match run or offer. For each data source, a predictive model estimated the probability of a local or regional liver transplant from donors with any recovered organ. LASSO was used to improve predictions and create parsimonious models. Models were estimated with donors recovered January 1, 2016-December 31, 2016. The predictive performance of simple decision rules based on each model was investigated and compared to a decision rule that classified hard-to-place livers as only from donation after circulatory death (DCD) donors. Predictive performance was evaluated with donors recovered January 1, 2017-December 31, 2017.
*Results: Rules based on the DonorNet model had similar, although marginally better, correct classification rates than rules based on the DDR model (Table 1). Rules based on the predictive models performed better than the DCD-only rule by better classifying donors without a locally or regionally transplanted liver as hard-to-place, i.e., better specificity.
*Conclusions: Despite the similar or marginally better correct classification rate, the DonorNet model had 41 non-zero covariate effects, including 10 non-zero effects for missingness of certain covariates. The DDR model had only 20 non-zero covariate effects. Thus, a decision rule based on DDR information may provide a parsimonious and practical approach for the a priori identification of hard-to-place livers.
Table 1. Predictive performance of decision rules based on predicted probability of local or regional placement from the DonorNet and DDR models, and on donor DCD status.
To cite this abstract in AMA style:Wey A, Hadley N, Rosendale J, Hunter R, Prinz J, Audette K, Salkowski N, Zaun D, Snyder J. A Priori Identification of Hard-to-Place Livers [abstract]. Am J Transplant. 2019; 19 (suppl 3). https://atcmeetingabstracts.com/abstract/a-priori-identification-of-hard-to-place-livers/. Accessed December 14, 2019.
« Back to 2019 American Transplant Congress