Exploiting complex medical data with interpretable deep learning for adverse drug event prediction.

A variety of deep learning architectures have been developed for the goal of predictive modelling and knowledge extraction from medical records. Several models have placed strong emphasis on temporal attention mechanisms and decay factors as a means to include highly temporally relevant information regarding the recency of medical event occurrence while facilitating medical code-level interpretability. In this study we utilise such models with a large Electronic Patient Record (EPR) data set consisting of diagnoses, medication, and clinical text data for the purpose of adverse drug event (ADE) prediction. The first contribution of this work is an empirical evaluation of two state-of-the-art medical-code based models in terms of objective performance metrics for ADE prediction on diagnosis and medication data. Secondly, as an extension of previous work, we augment an interpretable deep learning architecture to permit numerical risk and clinical text features and demonstrate how this approach yields improved predictive performance compared to the other baselines. Finally, we assess the importance of attention mechanisms in regards to their usefulness for medical code-level and text-level interpretability, which may facilitate novel insights pertaining to the nature of ADE occurrence within the health care domain.

A classification framework for exploiting sparse multi-variate temporal features with application to adverse drug event detection in medical records.

BACKGROUND: Adverse drug events (ADEs) as well as other preventable adverse events in the hospital setting incur a yearly monetary cost of approximately $3.5 billion, in the United States alone. Therefore, it is of paramount importance to reduce the impact and prevalence of ADEs within the healthcare sector, not only since it will result in reducing human suffering, but also as a means to substantially reduce economical strains on the healthcare system. One approach to mitigate this problem is to employ predictive models. While existing methods have been focusing on the exploitation of static features, limited attention has been given to temporal features. METHODS: In this paper, we present a novel classification framework for detecting ADEs in complex Electronic health records (EHRs) by exploiting the temporality and sparsity of the underlying features. The proposed framework consists of three phases for transforming sparse and multi-variate time series features into a single-valued feature representation, which can then be used by any classifier. Moreover, we propose and evaluate three different strategies for leveraging feature sparsity by incorporating it into the new representation. RESULTS: A large-scale evaluation on 15 ADE datasets extracted from a real-world EHR system shows that the proposed framework achieves significantly improved predictive performance compared to state-of-the-art. Moreover, our framework can reveal features that are clinically consistent with medical findings on ADE detection. CONCLUSIONS: Our study and experimental findings demonstrate that temporal multi-variate features of variable length and with high sparsity can be effectively utilized to predict ADEs from EHRs. Two key advantages of our framework are that it is method agnostic, i.e., versatile, and of low computational cost, i.e., fast; hence providing an important building block for future exploitation within the domain of machine learning from EHRs.

Ca2+ removal by the plasma membrane Ca2+-ATPase influences the contribution of mitochondria to activity-dependent Ca2+ dynamics in Aplysia neuroendocrine cells.

After Ca(2+) influx, mitochondria can sequester Ca(2+) and subsequently release it back into the cytosol. This form of Ca(2+)-induced Ca(2+) release (CICR) prolongs Ca(2+) signaling and can potentially mediate activity-dependent plasticity. As Ca(2+) is required for its subsequent release, Ca(2+) removal systems, like the plasma membrane Ca(2+)-ATPase (PMCA), could impact CICR. Here we examine such a role for the PMCA in the bag cell neurons of Aplysia californica CICR is triggered in these neurons during an afterdischarge and is implicated in sustaining membrane excitability and peptide secretion. Somatic Ca(2+) was measured from fura-PE3-loaded cultured bag cell neurons recorded under whole cell voltage clamp. Voltage-gated Ca(2+) influx was elicited with a 5-Hz, 1-min train, which mimics the fast phase of the afterdischarge. PMCA inhibition with carboxyeosin or extracellular alkalization augmented the effectiveness of Ca(2+) influx in eliciting mitochondrial CICR. A Ca(2+) compartment model recapitulated these findings and indicated that disrupting PMCA-dependent Ca(2+) removal increases CICR by enhancing mitochondrial Ca(2+) loading. Indeed, carboxyeosin augmented train-evoked mitochondrial Ca(2+) uptake. Consistent with their role on Ca(2+) dynamics, cell labeling revealed that the PMCA and mitochondria overlap with Ca(2+) entry sites. Finally, PMCA-dependent Ca(2+) extrusion did not impact endoplasmic reticulum-dependent Ca(2+) removal or release, despite the organelle residing near Ca(2+) entry sites. Our results demonstrate that Ca(2+) removal by the PMCA influences the propensity for stimulus-evoked CICR by adjusting the amount of Ca(2+) available for mitochondrial Ca(2+) uptake. This study highlights a mechanism by which the PMCA could impact activity-dependent plasticity in the bag cell neurons.

Separate Ca2+ sources are buffered by distinct Ca2+ handling systems in aplysia neuroendocrine cells.

Although the contribution of Ca(2+) buffering systems can vary between neuronal types and cellular compartments, it is unknown whether distinct Ca(2+) sources within a neuron have different buffers. As individual Ca(2+) sources can have separate functions, we propose that each is handled by unique systems. Using Aplysia californica bag cell neurons, which initiate reproduction through an afterdischarge involving multiple Ca(2+)-dependent processes, we investigated the role of endoplasmic reticulum (ER) and mitochondrial sequestration, as well as extrusion via the plasma membrane Ca(2+)-ATPase (PMCA) and Na(+)/Ca(2+) exchanger, to the clearance of voltage-gated Ca(2+) influx, Ca(2+)-induced Ca(2+)-release (CICR), and store-operated Ca(2+) influx. Cultured bag cell neurons were filled with the Ca(2+) indicator, fura-PE3, to image Ca(2+) under whole-cell voltage clamp. A 5 Hz, 1 min train of depolarizing voltage steps elicited voltage-gated Ca(2+) influx followed by EGTA-sensitive CICR from the mitochondria. A compartment model of Ca(2+) indicated the effect of EGTA on CICR was due to buffering of released mitochondrial Ca(2+) rather than uptake competition. Removal of voltage-gated Ca(2+) influx was dominated by the mitochondria and PMCA, with no contribution from the Na(+)/Ca(2+) exchanger or sarcoplasmic/endoplasmic Ca(2+)-ATPase (SERCA). In contrast, CICR recovery was slowed by eliminating the Na(+)/Ca(2+) exchanger and PMCA. Last, store-operated influx, evoked by ER depletion, was removed by the SERCA and depended on the mitochondrial membrane potential. Our results demonstrate that distinct buffering systems are dedicated to particular Ca(2+) sources. In general, this may represent a means to differentially regulate Ca(2+)-dependent processes, and for Aplysia, influence how reproductive behavior is triggered.