Performance assessment of different machine learning approaches in predicting diabetic ketoacidosis in adults with type 1 diabetes using electronic health records data.

PURPOSE: To assess the performance of different machine learning (ML) approaches in identifying risk factors for diabetic ketoacidosis (DKA) and predicting DKA. METHODS: This study applied flexible ML (XGBoost, distributed random forest [DRF] and feedforward network) and conventional ML approaches (logistic regression and least absolute shrinkage and selection operator [LASSO]) to 3400 DKA cases and 11 780 controls nested in adults with type 1 diabetes identified from Optum® de-identified Electronic Health Record dataset (2007-2018). Area under the curve (AUC), accuracy, sensitivity and specificity were computed using fivefold cross validation, and their 95% confidence intervals (CI) were established using 1000 bootstrap samples. The importance of predictors was compared across these models. RESULTS: In the training set, XGBoost and feedforward network yielded higher AUC values (0.89 and 0.86, respectively) than logistic regression (0.83), LASSO (0.83) and DRF (0.81). However, the AUC values were similar (0.82) among these approaches in the test set (95% CI range, 0.80-0.84). While the accuracy values >0.8 and the specificity values >0.9 for all models, the sensitivity values were only 0.4. The differences in these metrics across these models were minimal in the test set. All approaches selected some known risk factors for DKA as the top 10 features. XGBoost and DRF included more laboratory measurements or vital signs compared with conventional ML approaches, while feedforward network included more social demographics. CONCLUSIONS: In our empirical study, all ML approaches demonstrated similar performance, and identified overlapping, but different, top 10 predictors. The difference in selected top predictors needs further research.

A kinetic model unifying presynaptic short-term facilitation and depression.

Short-term facilitation and depression refer to the increase and decrease of synaptic strength under repetitive stimuli within a timescale of milliseconds to seconds. This phenomenon has been attributed to primarily presynaptic mechanisms such as calcium-dependent transmitter release and presynaptic vesicle depletion. Previous modeling studies that aimed to integrate the complex short-term facilitation and short-term depression data derived from varying synapses have relied on computer simulation or abstract mathematical approaches. Here, we propose a unified theory of synaptic short-term plasticity based on realistic yet tractable and testable model descriptions of the underlying intracellular biochemical processes. Analysis of the model equations leads to a closed-form solution of the resonance frequency, a function of several critical biophysical parameters, as the single key indicator of the propensity for synaptic facilitation or depression under repetitive stimuli. This integrative model is supported by a broad range of transient and frequency response experimental data including those from facilitating, depressing or mixed-mode synapses. Specifically, the theory predicts that high calcium initial concentration and large gain of calcium action result in low resonance frequency and hence depressing behavior. In contrast, for synapses that are less sensitive to calcium or have higher recovery rate, resonance frequency becomes higher and thus facilitation prevails. The notion of resonance frequency therefore allows valuable quantitative parametric assessment of the contributions of various presynaptic mechanisms to the directionality of synaptic short-term plasticity. Thus, the model provides the reasons behind the switching behavior between facilitation and depression observed in experiments. New experiments are also suggested to control the short-term synaptic signal processing through adjusting the resonance frequency and bandwidth.

Integrating Clinical Knowledge and Real-World Evidence for Type 2 Diabetes Treatment.

Clinical decision support system (CDSS) plays a significant role nowadays and it assists physicians in making decisions for treatment. Generally based on clinical guideline, the principles of the recommendation are provided and may suggest several candidate medications for similar patient group with certain clinical conditions. However, it is challenging to prioritize these candidates and even refine the guideline to a finer level for patient-specific recommendation. Here we propose a method and system to integrate the clinical knowledge and real-world evidence (RWE) for type 2 diabetes treatment, to enable both standardized and personalized medication recommendation. The RWE is generated by medication effectiveness analysis and subgroup analysis. The knowledge model has been verified by clinical experts from the advanced hospitals. The data verification results show that the medications that are consistent with the method recommendation can lead to better clinical outcome in terms of glycemic control, compared to those inconsistent.

Osmolyte controlled fibrillation kinetics of insulin: New insight into fibrillation using the preferential exclusion principle.

Amyloid proteins are converted from their native-fold to long beta-sheet-rich fibrils in a typical sigmoidal time-dependent protein aggregation curve. This reaction process from monomer or dimer to oligomer to nuclei and then to fibrils is the subject of intense study. The main results of this work are based on the use of a well-studied model amyloid protein, insulin, which has been used in vitro by others. Nine osmolyte molecules, added during the protein aggregation process for the production of amyloid fibrils, slow-down or speed up the process depending on the molecular structure of each osmolyte. Of these, all stabilizing osmolytes (sugars) slow down the aggregation process in the following order: tri > di > monosaccharides, whereas destabilizing osmolytes (urea, guanidium hydrochloride) speed up the aggregation process in a predictable way that fits the trend of all osmolytes. With respect to kinetics, we illustrate, by adapting our earlier reaction model to the insulin system, that the intermediates (trimers, tetramers, pentamers, etc.) are at very low concentrations and that nucleation is orders of magnitude slower than fibril growth. The results are then collated into a cogent explanation using the preferential exclusion and accumulation of osmolytes away from and at the protein surface during nucleation, respectively. Both the heat of solution and the neutral molecular surface area of the osmolytes correlate linearly with two fitting parameters of the kinetic rate model, that is, the lag time and the nucleation rate prior to fibril formation. These kinetic and thermodynamic results support the preferential exclusion model and the existence of oligomers including nuclei and larger structures that could induce toxicity.

A three-stage kinetic model of amyloid fibrillation.

Amyloid fibrillation has been intensively studied because of its association with various neurological disorders. While extensive time-dependent fibrillation experimental data are available and appear similar, few mechanistic models have been developed to unify those results. The aim of this work was to interpret these experimental results via a rigorous mathematical model that incorporates the physical chemistry of nucleation and fibril growth dynamics. A three-stage mechanism consisting of protein misfolding, nucleation, and fibril elongation is proposed and supported by the features of homogeneous fibrillation responses. Estimated by nonlinear least-squares algorithms, the rate constants for nucleation were approximately 10,000,000 times smaller than those for fibril growth. These results, coupled with the positive feedback characteristics of the elongation process, account for the typical sigmoidal behavior during fibrillation. In addition, experiments with different proteins, various initial concentrations, seeding versus nonseeding, and several agitation rates were analyzed with respect to fibrillation using our new model. The wide applicability of the model confirms that fibrillation kinetics may be fairly similar among amyloid proteins and for different environmental factors. Recommendations on further experiments and on the possible use of molecular simulations to determine the desired properties of potential fibrillation inhibitors are offered.