Dataset Information

Explainable detection of adverse drug reaction with imbalanced data distribution.

ABSTRACT: Analysis of health-related texts can be used to detect adverse drug reactions (ADR). The greatest challenge for ADR detection lies in imbalanced data distributions where words related to ADR symptoms are often minority classes. As a result, trained models tend to converge to a point that strongly biases towards the majority class and then ignores the minority class. Since the most used cross-entropy criteria is an approximation to accuracy, the model focuses more readily on the majority class to achieve high accuracy. To address this issue, existing methods apply either oversampling or down-sampling strategies to balance the data distribution and exploit the most difficult samples of the minority class. However, increasing or reducing the number of individual tokens alone in sequence labeling tasks will result in the loss of the syntactic relations of the sentence. This paper proposes a weighted variant of conditional random field (CRF) for data-imbalanced sequence labeling tasks. Such a weighting strategy can alleviate data distribution imbalances between majority and minority classes. Instead of using softmax in the output layer, the CRF can capture the relationship of labels between tokens. The locally interpretable model-agnostic explanations (LIME) algorithm was applied to investigate performance differences between models with and without the weighted loss function. Experimental results on two different ADR tasks show that the proposed model outperforms previously proposed sequence labeling methods.

SUBMITTER: Wang J

PROVIDER: S-EPMC9239481 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:Introduction:Developing countries can improve their pharmacovigilance systems by analysing their own medication safety data. Objective:The aims of this study were to characterize Uganda's reported adverse drug reaction (ADR) onsets in 2012-2015 that were registered on VigiBase® by 31 December 2017, to document delays in international visibility and the influence of covariates on this delay from ADR onsets in 2013 + 2014, to examine data quality, and to illustrate analytical approaches for safety data, particularly for patients receiving antiretroviral therapy (ART). Methods:International delay was defined as elapsed time from complete ADR onset date to entry date on VigiBase®, with covariates examined using Cox proportional hazards regression. Simple random sampling was used to locate the paper-based ADR forms for data quality assurance. Disproportionality for signal detection focused on serious singleton ADR onsets in patients receiving ART. Results:Uganda's VigiBase® had 1018 patient entries with complete ADR onset dates: 260 in 2012, 293 in 2013, 305 in 2014 and 160 in 2015. Only 16% (154/953) of ADR onsets in 2012-2015 were in patients aged?<?20 years for whom randomly sampled ADR forms were less fully completed; 87% (889/1018) comprised a singleton sign/symptom; half were serious. Median delay from ADR onset to international visibility was 11 months for ADR onsets in 2013?+?2014, and longest for healthcare professionals other than pharmacists and physicians. Disproportionality for serious ADR onsets in patients receiving ART included anaemia with zidovudine, renal impairment with tenofovir, Stevens-Johnson syndrome with nevirapine and skin rash with efavirenz. Conclusions:Barely one ADR onset per day was registered on VigiBase® from those submitted to Uganda's National Pharmacovigilance Centre during 2012-2014; only one in six was from patients aged?<?20 years. Paediatric pharmacovigilance requires more emphasis in Uganda. Delays from reported ADR onset to international visibility on VigiBase® need to reduce dramatically. Quality assurance revealed rectifiable data entry deficits. Signal detection performed well for patients receiving ART.

Project description:BackgroundBased on data of clinical trials, new agents are receiving approval to the pharmaceutical market, for which information concerning safety issues under real-life conditions is not yet available.ObjectivesThe aim was to evaluate the tolerability of newer antiepileptic drugs (AEDs), such as topiramate, levetiracetam, zonisamide, pregabalin, extended-release oxcarbazepine, lacosamide and eslicarbazepine, under real-life conditions by means of an assessment of routine clinical data of inpatients.MethodOver 2.75 years data of all inpatients receiving one of the newer AEDs were documented. Occurring adverse drug reactions (ADRs) were classified according to the WHO-UMC Causality Assessment concerning their likely relationship to the prescribed AEDs. For each AED, the total number of patients without and with ADRs, assessed as at least possibly related to the particular drug, was calculated and corresponding incidences compared with reference data provided in the Summary of Product Characteristics (SmPC). For statistical evaluation Spearman correlation (rs), estimated relative risk and logistic regression analysis were used.ResultsIn total, the data of 562 patients were assessed, of which 90 % received up to six different AEDs. The proportion of off-label use with regard to dosage varied between 6.4 and 64.7 %. Levetiracetam and oxcarbazepine as an extended-release formulation were most commonly used, and levetiracetam showed the best tolerance. By using logistic regression, the occurrence of ADRs was significantly associated with the number of AEDs (p < 0.001) as well as the defined daily doses (p = 0.003). In total, ADRs of AEDs were documented for 318 patients (56.6 %). The most common referred to electrolyte imbalance, e.g., low sodium (n = 79, 14.1 %) and potassium (n = 25, 4.4 %) levels, the central nervous system, including dizziness (n = 61, 10.9 %), disturbed vision (n = 47, 8.4 %), fatigue (n = 40, 7.1 %), nystagmus (n = 36, 6.4 %) and ataxia (n = 29, 5.2 %), or cognitive deficits, especially disturbance of speech (n = 37, 6.6 %), memory impairment (n = 36, 6.4 %) and mental slowing (n = 32, 5.7 %). By comparing the assessed ADR incidences with specification data, for some ADRs, a probable underestimation by the SmPC of respective risk could be assumed.ConclusionDuring inpatient treatment, valuable data are generated, which are currently rarely utilized for pharmacoepidemiologic or pharmacovigilance purposes. A systematic evaluation of these data can increase the probability of detecting ADRs and can promote real-life-related drug surveillance.

Project description:BackgroundAdministrative health data are increasingly used to detect adverse drug events (ADEs). However, the few studies evaluating diagnostic codes for ADE detection demonstrated low sensitivity, likely due to narrow code sets, physician under-recognition of ADEs, and underreporting in administrative data. The objective of this study was to determine if combining an expanded ICD code set in administrative data with e-prescribing data improves ADE detection.MethodsWe conducted a prospective cohort study among patients newly prescribed antidepressant or antihypertensive medication in primary care and followed for 2 months. Gold standard ADEs were defined as patient-reported symptoms adjudicated as medication-related by a clinical expert. Potential ADEs in administrative data were defined as physician, ED, or hospital visits during follow-up for known adverse effects of the study medication, as identified by ICD codes. Potential ADEs in e-prescribing data were defined as study drug discontinuations or dose changes made during follow-up for safety or effectiveness reasons.ResultsOf 688 study participants, 445 (64.7%) were female and mean age was 64.2 (SD 13.9). The study drug for 386 (56.1%) patients was an antihypertensive, and for 302 (43.9%) an antidepressant. Using the gold standard definition, 114 (16.6%) patients experienced an ADE, with 40 (10.4%) among antihypertensive users and 74 (24.5%) among antidepressant users. The sensitivity of the expanded ICD code set was 7.0%, of e-prescribing data 9.7%, and of the two combined 14.0%. Specificities were high (86.0-95.0%). The sensitivity of the combined approach increased to 25.8% when analysis was restricted to the 27% of patients who indicated having reported symptoms to a physician.ConclusionCombining an expanded diagnostic code set with e-prescribing data improves ADE detection. As few patients report symptoms to their physician, higher detection rates may be achieved by collecting patient-reported outcomes via emerging digital technologies such as patient portals and mHealth applications.

Dataset Information

Explainable detection of adverse drug reaction with imbalanced data distribution.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets