Search

Adverse drug reactions (ADRs), which are harmful physical reactions of patients to drug treatments, are inherent to the nature of drugs; the reactions can occur with any drug and are becoming a leading cause of patient morbidity and mortality during medical procedures. ADRs can be hazardous and even fatal to patients. In traditional methods, ADRs are detected through clinical trials. To obtain a comprehensive collection of ADRs, sufficient experimental samples and time are required before a drug comes to the market, which is not a realistic possibility. Moreover, even if extensive clinical trials are performed, many undetected ADRs might still be discovered after a drug is released to the market. ADRs can lead to disastrous consequences for humanity, which obviates a dramatically increased need for precise predictions of potential ADRs as early as possible. In this paper, we propose an encoder-decoder framework based on attention mechanism and the long short-term memory (LSTM) model to predict potential ADRs. We regard the prediction of ADRs as a sequence-to-sequence problem and improve the encoder-decoder framework based on the attention mechanism to learn the interrelationships between ADRs. Unlike other classical methods utilizing molecular drug structures, our model is based solely on ADRs, which is an independent but parallel approach compared to traditional methods. We capitalize on the mask method to generate the target data and use the 5-fold cross-validation method to cyclically verify the performance of our proposed model. Based on the Top-k accuracy test results, our model outperforms the baseline models in potential ADRs predictions. [ABSTRACT FROM AUTHOR]

This paper introduces a method for mining co-occurring events from longitudinal data, and applies this method to detecting adverse drug reactions (ADRs) from patient data. Electronic health records are richer than older data sources (such as spontaneous report records) and thus are ideal for ADR mining. However, current data mining methods, such as disproportionality ratios and temporal itemset mining, ignore certain important aspects of the longitudinal data in patient records. In this paper, we highlight two specific problems with current methods, which we name temporal and contextual sensitivity, and discuss why these two properties are vital to mining patterns from longitudinal data. We also propose two sensitive longitudinal rate comparison measures, which utilize condition occurrence rates and length of drug eras, for mining ADRs from this type of data. These novel methods are then used to rank potential ADRs, along with existing state-of-the-art methods, under many simulated yet realistic datasets. In 48 out of 60 experiments, the proposed longitudinal rate comparison methods significantly outperform other methods in mining known ADRs from other drug / condition pairs. [ABSTRACT FROM AUTHOR]