Publications

Most lung transplantation immunosuppression regimens include tacrolimus. Single nucleotide polymorphisms (SNPs) in genes important to tacrolimus bioavailability and clearance (ABCB1, CYP3A4, and CYP3A5) are associated with differences in tacrolimus pharmacokinetics. We hypothesized that polymorphisms in these genes would impact immunosuppression-related outcomes. We categorized ABCB1, CYP3A4, and CYP3A5 SNPs for 321 lung allograft recipients. Genotype effects on time to therapeutic tacrolimus level, interactions with antifungal medications, concentration to dose (C /D), acute kidney injury, and rejection were assessed using linear models adjusted for subject characteristics and repeat measures. Compared with CYP3A poor metabolizers (PM), time to therapeutic tacrolimus trough was increased by 5.1 ± 1.6 days for CYP3A extensive metabolizers (EM, P < 0.001). In the post-operative period, CYP3A intermediate metabolizers spent 1.2 ± 0.5 days less (P = 0.01) and EM spent 2.1 ± 0.5 days less (P < 0.001) in goal tacrolimus range than CYP3A PM. Azole antifungals interacted with CYP3A genotype in predicting C /D (P < 0.001). Increased acute kidney injury rates were observed in subjects with high ABCB1 function (OR 3.0, 95% CI 1.1-8.6, P = 0.01). Lower rates of acute cellular rejection were observed in subjects with low ABCB1 function (OR 0.36, 95% CI 0.07-0.94, P = 0.02). Recipient genotyping may help inform tacrolimus dosing decisions and risk of adverse clinical outcomes.

We have three contributions in this work: 1. We explore the utility of a stacked denoising autoencoder and a paragraph vector model to learn task-independent dense patient representations directly from clinical notes. To analyze if these representations are transferable across tasks, we evaluate them in multiple supervised setups to predict patient mortality, primary diagnostic and procedural category, and gender. We compare their performance with sparse representations obtained from a bag-of-words model. We observe that the learned generalized representations significantly outperform the sparse representations when we have few positive instances to learn from, and there is an absence of strong lexical features. 2. We compare the model performance of the feature set constructed from a bag of words to that obtained from medical concepts. In the latter case, concepts represent problems, treatments, and tests. We find that concept identification does not improve the classification performance. 3. We propose novel techniques to facilitate model interpretability. To understand and interpret the representations, we explore the best encoded features within the patient representations obtained from the autoencoder model. Further, we calculate feature sensitivity across two networks to identify the most significant input features for different classification tasks when we use these pretrained representations as the supervised input. We successfully extract the most influential features for the pipeline using this technique.