Digital light processing (DLP) bioprinting, which provides predominant speed, resolution, and adaptability for fabricating complex cell-laden three-dimensional (3D) structures, requires a combination of photoinitiator (PI) and UV absorber (UA) that plays critical roles during the photo-polymerization of bioinks. However, the PI and UA combination has not been highlighted for cell-based DLP bioprinting. In this study, the most used PIs and UAs in cell-based bioprinting were compared to optimize a combination that can ensure the maximum DLP printability, while maintaining the cellular activities during the process. The crosslinking time and printability of PIs were assessed, which are critical in minimizing the cell damage by the UV exposure during the fabrication process. On the other hand, the UAs were evaluated based on their ability to prevent the over-curing of layers beyond the focal layer and the scattering of light, which are required for the desirable crosslinking of a hydrogel and high resolution (25–50 µms) to create a complex 3D cell-laden construct. Lastly, the cytotoxicity of PIs and UAs was assessed by measuring the cellular activity of 2D cultured and 3D bioprinted cells. The optimized PI and UA combination provided high initial cell viability (>90%) for up to 14 days in culture and could fabricate complex 3D structures like a perfusable heart-shaped construct with open vesicles and atriums. This combination can provide a potential starting condition when preparing the bioink for the cell-based DLP bioprinting in tissue engineering applications.

INTRODUCTION AND OBJECTIVE: Delayed graft function (DGF), often defined as need for post-operative dialysis during the first week after kidney transplantation, contributes to both patient outcomes and economic costs. Several attempts have been made to predict DGF in the adult population. However, there remains limited information on predicting DGF in children. Therefore, we aim to create a machine learning model that predicts DGF outcomes in children who undergo kidney transplantation.METHODS: An institutional database of children and adolescents who underwent kidney transplantation was examined. Python 3.9.13 was used for model development. The model was built using an 80:20 train-test split. To avoid model bias for lack of DGF outcome, Synthetic Minority Oversampling Technique for Nominal and Continuous (SMOTE-NC) was employed, effectively increasing the number of cases of DGF in the dataset in a balanced manner. Extreme Gradient Boosting (XG Boost) was ultimately employed to build our model. Gridsearch was performed to optimize model parameters.RESULTS: A total of 409 patients were included in our analysis. Forty patients had DGF (9.7%). Following SMOTE-NC, the training data generated 295 patients in each group. After training with XG Boost model with Gridsearch parameters, the model had a 5-fold cross-validation accuracy of 93.1% and a receiver operating characteristics (ROC) area under the curve of 91.7%. On evaluation of the confusion matrix, the model had an excellent specificity of 96.0% (71/74) but a rather modest sensitivity of 37.5% (3/8). The deployed model can be found in: https://kimjk4-dgfprediction-app-omaz4b.streamlitapp.com.CONCLUSIONS: The generated model had very high negative predictive value, allowing us to identify patients with high risk of DGF, providing an opportunity for closer monitoring. This novel model is the first attempt at predicting DGF in children undergoing kidney transplantation and holds promise for further development and improvement with additional variables and patient numbers.