Ehtesham Iqbal

Identification of Adverse Drug Events from Free Text Electronic Patient Records and Information in a Large Mental Health Case Register.

Objectives Electronic healthcare records (EHRs) are a rich source of information, with huge potential for secondary research use. The aim of this study was to develop an application to identify instances of Adverse Drug Events (ADEs) from free text psychiatric EHRs. Methods We used the GATE Natural Language Processing (NLP) software to mine instances of ADEs from free text content within the Clinical Record Interactive Search (CRIS) system, a de-identified psychiatric case register developed at the South London and Maudsley NHS Foundation Trust, UK. The tool was built around a set of four movement disorders (extrapyramidal side effects [EPSEs]) related to antipsychotic therapy and rules were then generalised such that the tool could be applied to additional ADEs. We report the frequencies of recorded EPSEs in patients diagnosed with a Severe Mental Illness (SMI) and then report performance in identifying eight other unrelated ADEs. Results The tool identified EPSEs with >0.85 precision and >0.86 recall during testing. Akathisia was found to be the most prevalent EPSE overall and occurred in the Asian ethnic group with a frequency of 8.13%. The tool performed well when applied to most of the non-EPSEs but least well when applied to rare conditions such as myocarditis, a condition that appears frequently in the text as a side effect warning to patients. Conclusions The developed tool allows us to accurately identify instances of a potential ADE from psychiatric EHRs. As such, we were able to study the prevalence of ADEs within subgroups of patients stratified by SMI diagnosis, gender, age and ethnicity. In addition we demonstrated the generalisability of the application to other ADE types by producing a high precision rate on a non-EPSE related set of ADE containing documents. Availability The application can be found at http://git.brc.iop.kcl.ac.uk/rmallah/dystoniaml.

Trajectories of dementia-related cognitive decline in a large mental health records derived patient cohort

Background Modeling trajectories of decline can help describe the variability in progression of cognitive impairment in dementia. Better characterisation of these trajectories has significant implications for understanding disease progression, trial design and care planning. Methods Patients with at least three Mini-mental State Examination (MMSE) scores recorded in the South London and Maudsley NHS Foundation Trust Electronic Health Records, UK were selected (N = 3441) to form a retrospective cohort. Trajectories of cognitive decline were identified through latent class growth analysis of longitudinal MMSE scores. Demographics, Health of Nation Outcome Scales and medications were compared across trajectories identified. Results Four of the six trajectories showed increased rate of decline with lower baseline MMSE. Two trajectories had similar initial MMSE scores but different rates of decline. In the faster declining trajectory of the two, a higher incidence of both behavioral problems and sertraline prescription were present. Conclusions We find suggestive evidence for association of behavioral problems and sertraline prescription with rate of decline. Further work is needed to determine whether trajectories replicate in other datasets.

ADEPt, a semantically-enriched pipeline for extracting adverse drug events from free-text electronic health records

Adverse drug events (ADEs) are unintended responses to medical treatment. They can greatly affect a patient’s quality of life and present a substantial burden on healthcare. Although Electronic health records (EHRs) document a wealth of information relating to ADEs, they are frequently stored in the unstructured or semi-structured free-text narrative requiring Natural Language Processing (NLP) techniques to mine the relevant information. Here we present a rule-based ADE detection and classification pipeline built and tested on a large Psychiatric corpus comprising 264k patients using the de-identified EHRs of four UK-based psychiatric hospitals. The pipeline uses characteristics specific to Psychiatric EHRs to guide the annotation process, and distinguishes: a) the temporal value associated with the ADE mention (whether it is historical or present), b) the categorical value of the ADE (whether it is assertive, hypothetical, retrospective or a general discussion) and c) the implicit contextual value where the status of the ADE is deduced from surrounding indicators, rather than explicitly stated. We manually created the rulebase in collaboration with clinicians and pharmacists by studying ADE mentions in various types of clinical notes. We evaluated the open-source Adverse Drug Event annotation Pipeline (ADEPt) using 19 ADEs specific to antipsychotics and antidepressants medication. The ADEs chosen vary in severity, regularity and persistence. The average F-measure and accuracy achieved by our tool across all tested ADEs were 0.83 and 0.83 respectively. In addition to annotation power, the ADEPT pipeline presents an improvement to the state of the art context-discerning algorithm, ConText.

Author Correction: Knowledge graph prediction of unknown adverse drug reactions and validation in electronic health records

A correction to this article has been published and is linked from the HTML and PDF versions of this paper. The error has been fixed in the paper.

S221. QUANTITATIVE SYSTEMS PHARMACOLOGY AS AN ALTERNATIVE TO CHLORPROMAZINE EQUIVALENTS: PREDICTIVE VALIDATION FROM A CRIS DATABASE EXPERIMENT

Abstract Background Polypharmacy is common in real clinical practice and in pharma-sponsored clinical trials. Chlorpromazine equivalents do not take into account pharmacodynamic interactions of drug combinations. If there is a sufficiently deep calibration set available, bio-informatics approaches can build classifiers for clinical phenotypes. However, this is not always the case which severely limits the generalizability of the predictions. Methods We applied a mechanism-based computer model of a cortico-striatal-thalamocortical loop of the dorsal motor circuit that has been calibrated with clinical data on the prevalence of extrapyramidal symptoms after antipsychotic treatment in schizophrenia patients and therapeutic interventions in Parkinson’s patients[1]. The Quantitative Systems Pharmacology (QSP) model is based on the appropriate connections between basal ganglia regions and consists of 220 neurons (8 different cell types), 3500 synapses and implementations of 32 CNS active targets, based on their unique locations and coupling with intracellular pathways. Modulation of the various CNS targets were calculated on simulating the competition between the endogenous neurotransmitter and the two drugs at their appropriate concentrations and affinity. The model was challenged to blindly predict the extrapyramidal symptoms liability of 1,124 patients prescribed two antipsychotics for six or more months (772 unique combinations). Anonymized data were derived from South London and Maudsley NHS Foundation Trust (SLAM) electronic health records (EHR). Extrapyramidal side effects were captured and identified using a combination of Natural Language Processing and a bespoke algorithm [2]. Only names and doses of the two drugs were made available without any calibration set. Results Blind prediction of the outcomes using a Receiver Operating Characteristic curve with the QSP model resulted in an Area-Under-the Curve of 0.64 (p<0.01), compared to an AUC of 0.52 for the sum of the chlorpromazine equivalents, 0.53 for the sum of affinity constants or the sum of D2R occupancies of the individual antipsychotics (AUC=0.52). Discussion QSP is a powerful approach to predict PD-PD interactions in the absence of any calibration set or with limited and unique data and is superior to chlorpromazine equivalents for predicting EPS liability. A major application is the simulation of pharmacodynamic interactions of comedications in clinical trials with novel compounds leading to possible better balance between the different treatment arms

Predicting parkinsonism side effects of antipsychotic polypharmacy prescribed in secondary mental healthcare

Background: Computer-modelling approaches have the potential to predict the interactions between different antipsychotics and provide guidance for polypharmacy. Aims: To evaluate the accuracy of the quantitative systems pharmacology platform to predict parkinsonism side-effects in patients prescribed antipsychotic polypharmacy. Methods: Using anonymized data from South London and Maudsley NHS Foundation Trust electronic health records we applied quantitative systems pharmacology, a neurophysiology-based computer model of humanized neuronal circuits, to predict the risk for parkinsonism symptoms in patients with schizophrenia prescribed two concomitant antipsychotics. The performance of the quantitative systems pharmacology model was compared with the performance of simple parameters such as: combination of affinity constants (1/Ksum); sum of D2R occupancies (D2R) and chlorpromazine equivalent dose. Results: We identified 832 patients with schizophrenia who were receiving two antipsychotics for six or more months, between 1 January 2007 and 31 December 2014. The area under the receiver operating characteristic (AUROC) curve for the quantitative systems pharmacology model was 0.66 (p = 0.01), while AUROCs for D2R, 1/Ksum and chlorpromazine equivalent dose were 0.52 (p = 0.350), 0.53 (p = 0.347) and 0.52 (p = 0.330) respectively. Conclusion: Our results indicate that quantitative systems pharmacology has the potential to predict the risk of parkinsonism associated with antipsychotic polypharmacy from minimal source information, and thus might have potential decision-support applicability in clinical settings.

Improving RNN with Attention and Embedding for Adverse Drug Reactions

Electronic Health Records (EHR) narratives are a rich source of information, embedding high-resolution information of value to secondary research use. However, because the EHRs are mostly in natural language free-text and highly ambiguity-ridden, many natural language processing algorithms have been devised around them to extract meaningful structured information about clinical entities. The performance of the algorithms however, largely varies depending on the training dataset as well as the effectiveness of the use of background knowledge to steer the learning process. In this paper we study the impact of initializing the training of a neural network natural language processing algorithm with pre-defined clinical word embeddings to improve feature extraction and relationship classification between entities. We add our embedding framework to a bi-directional long short-term memory (Bi-LSTM) neural network, and further study the effect of using attention weights in neural networks for sequence labelling tasks to extract knowledge of Adverse Drug Reactions (ADRs). We incorporate unsupervised word embeddings using Word2Vec and GloVe from widely available medical resources such as Multiparameter Intelligent Monitoring in Intensive Care (MIMIC) II corpora, Unified Medical Language System (UMLS) as well as embed pharmaco lexicon from available EHRs. Our algorithm, implemented using two datasets, shows that our architecture outperforms baseline Bi-LSTM or Bi-LSTM networks using linear chain and Skip-Chain conditional random fields (CRF).

Knowledge graph prediction of unknown adverse drug reactions and validation in electronic health records

Unknown adverse reactions to drugs available on the market present a significant health risk and limit accurate judgement of the cost/benefit trade-off for medications. Machine learning has the potential to predict unknown adverse reactions from current knowledge. We constructed a knowledge graph containing four types of node: drugs, protein targets, indications and adverse reactions. Using this graph, we developed a machine learning algorithm based on a simple enrichment test and first demonstrated this method performs extremely well at classifying known causes of adverse reactions (AUC 0.92). A cross validation scheme in which 10% of drug-adverse reaction edges were systematically deleted per fold showed that the method correctly predicts 68% of the deleted edges on average. Next, a subset of adverse reactions that could be reliably detected in anonymised electronic health records from South London and Maudsley NHS Foundation Trust were used to validate predictions from the model that are not currently known in public databases. High-confidence predictions were validated in electronic records significantly more frequently than random models, and outperformed standard methods (logistic regression, decision trees and support vector machines). This approach has the potential to improve patient safety by predicting adverse reactions that were not observed during randomised trials.

DEMENTIA SEVERITY AND PROGRESSION: IDENTIFYING PATIENTS AT RISK FOR FASTER COGNITIVE DECLINE

outcomes. Results: In ADNI, higher hippocampal atrophy was associated with greater baseline systolic ((r, p value); 0.2, p1⁄40.004) and pulse pressure (0.2, p1⁄40.007) in MCIs; no associations were found in controls. In NACC, higher baseline systolic BP was associated with lower baseline MMSE in MCIs (-0.23, <0.001); falling MMSE was predicted by falling systolic (0.35, p<0.001) and pulse pressure (0.34, p1⁄40.001). Conversely, in AD patients, lower MMSEs were associated with lower systolic (0.10, p<0.001), diastolic (0.06, p1⁄40.05), and pulse pressure (0.08, p1⁄40.01). Conclusions: These results suggest whilst high BP may be associated with AD development, falling BP may be important in the mid-late AD stages; MCIs with falling BP were more likely to worsen on MMSE, and AD subjects with low systolic BP more likely to have lower baseline MMSEs. Further work is required to understand whether this could inform AD management and prevention strategies.

Encoding Medication Episodes for Adverse Drug Event Prediction

Understanding the interplay among the multiple factors leading to Adverse Drug Reactions (ADRs) is crucial to increasing drug effectiveness, individualising drug therapy and reducing incurred cost. In this paper, we propose a flexible encoding mechanism that can effectively capture the dynamics of multiple medication episodes of a patient at any given time. We enrich the encoding with a drug ontology and patient demographics data and use it as a base for an ADR prediction model. We evaluate the resulting predictive approach under different settings using real anonymised patient data obtained from the EHR of the South London and Maudsley (SLaM), the largest mental health provider in Europe. Using the profiles of 38,000 mental health patients, we identified 240,000 affirmative mentions of dry mouth, constipation and enuresis and 44,000 negative ones. Our approach achieved 93 % prediction accuracy and 93 % F-Measure.