Artificial Intelligence and Data Mining for the Pharmacovigilance of Drug-Drug Interactions.

Despite increasing mechanistic understanding, undetected and underrecognized drug-drug interactions (DDIs) persist. This elusiveness relates to an interwoven complexity of increasing polypharmacy, multiplex mechanistic pathways, and human biological individuality. This persistent elusiveness motivates development of artificial intelligence (AI)-based approaches to enhancing DDI detection and prediction capabilities. The literature is vast and roughly divided into "prediction" and "detection." The former relatively emphasizes biological and chemical knowledge bases, drug development, new drugs, and beneficial interactions, whereas the latter utilizes more traditional sources such as spontaneous reports, claims data, and electronic health records to detect novel adverse DDIs with authorized drugs. However, it is not a bright line, either nominally or in practice, and both are in scope for pharmacovigilance supporting signal detection but also signal refinement and evaluation, by providing data-based mechanistic arguments for/against DDI signals. The wide array of intricate and elegant methods has expanded the pharmacovigilance tool kit. How much they add to real prospective pharmacovigilance, reduce the public health impact of DDIs, and at what cost in terms of false alarms amplified by automation bias and its sequelae are open questions.

A Systematic Review of Artificial Intelligence and Machine Learning Applications to Inflammatory Bowel Disease, with Practical Guidelines for Interpretation.

BACKGROUND: Inflammatory bowel disease (IBD) is a gastrointestinal chronic disease with an unpredictable disease course. Computational methods such as machine learning (ML) have the potential to stratify IBD patients for the provision of individualized care. The use of ML methods for IBD was surveyed, with an additional focus on how the field has changed over time. METHODS: On May 6, 2021, a systematic review was conducted through a search of MEDLINE and Embase databases, with the search structure ("machine learning" OR "artificial intelligence") AND ("Crohn* Disease" OR "Ulcerative Colitis" OR "Inflammatory Bowel Disease"). Exclusion criteria included studies not written in English, no human patient data, publication before 2001, studies that were not peer reviewed, nonautoimmune disease comorbidity research, and record types that were not primary research. RESULTS: Seventy-eight (of 409) records met the inclusion criteria. Random forest methods were most prevalent, and there was an increase in neural networks, mainly applied to imaging data sets. The main applications of ML to clinical tasks were diagnosis (18 of 78), disease course (22 of 78), and disease severity (16 of 78). The median sample size was 263. Clinical and microbiome-related data sets were most popular. Five percent of studies used an external data set after training and testing for additional model validation. DISCUSSION: Availability of longitudinal and deep phenotyping data could lead to better modeling. Machine learning pipelines that consider imbalanced data and that feature selection only on training data will generate more generalizable models. Machine learning models are increasingly being applied to more complex clinical tasks for specific phenotypes, indicating progress towards personalized medicine for IBD.

Incidence of venous and arterial thromboembolic events reported in the tofacitinib rheumatoid arthritis, psoriasis and psoriatic arthritis development programmes and from real-world data.

OBJECTIVES: Tofacitinib is a Janus kinase inhibitor for the treatment of rheumatoid arthritis (RA), psoriatic arthritis (PsA) and ulcerative colitis, and has been investigated in psoriasis (PsO). Routine pharmacovigilance of an ongoing, open-label, blinded-endpoint, tofacitinib RA trial (Study A3921133; NCT02092467) in patients aged ≥50 years and with ≥1 cardiovascular risk factor identified a higher frequency of pulmonary embolism (PE) and all-cause mortality for patients receiving tofacitinib 10 mg twice daily versus those receiving tumour necrosis factor inhibitors and resulted in identification of a safety signal for tofacitinib. Here, we report the incidence of deep vein thrombosis (DVT), PE, venous thromboembolism (VTE; DVT or PE) and arterial thromboembolism (ATE) from the tofacitinib RA (excluding Study A3921133), PsA and PsO development programmes and observational studies. Data from an ad hoc safety analysis of Study A3921133 are reported separately within. METHODS: This post-hoc analysis used data from separate tofacitinib RA, PsO and PsA programmes. Incidence rates (IRs; patients with events per 100 patient-years' exposure) were calculated for DVT, PE, VTE and ATE, including for populations stratified by defined baseline cardiovascular or VTE risk factors. Observational data from the US Corrona registries (including cardiovascular risk factor stratification), IBM MarketScan research database and the US FDA Adverse Event Reporting System (FAERS) database were analysed. RESULTS: 12 410 tofacitinib-treated patients from the development programmes (RA: n=7964; PsO: n=3663; PsA: n=783) were included. IRs (95% CI) of thromboembolic events among the all tofacitinib cohorts' average tofacitinib 5 mg and 10 mg twice daily treated patients for RA, respectively, were: DVT (0.17 (0.09-0.27) and 0.15 (0.09-0.22)); PE (0.12 (0.06-0.22) and 0.13 (0.08-0.21)); ATE (0.32 (0.22-0.46) and 0.38 (0.28-0.49)). Among PsO patients, IRs were: DVT (0.06 (0.00-0.36) and 0.06 (0.02-0.15)); PE (0.13 (0.02-0.47) and 0.09 (0.04-0.19)); ATE (0.52 (0.22-1.02) and 0.22 (0.13-0.35)). Among PsA patients, IRs were: DVT (0.00 (0.00-0.28) and 0.13 (0.00-0.70)); PE (0.08 (0.00-0.43) and 0.00 (0.00-0.46)); ATE (0.31 (0.08-0.79) and 0.38 (0.08-1.11)). IRs were similar between tofacitinib doses and generally higher in patients with baseline cardiovascular or VTE risk factors. IRs from the overall Corrona populations and in Corrona RA patients (including tofacitinib-naïve/biologic disease-modifying antirheumatic drug-treated and tofacitinib-treated) with baseline cardiovascular risk factors were similar to IRs observed among the corresponding patients in the tofacitinib development programme. No signals of disproportionate reporting of DVT, PE or ATE with tofacitinib were identified in the FAERS database. CONCLUSIONS: DVT, PE and ATE IRs in the tofacitinib RA, PsO and PsA programmes were similar across tofacitinib doses, and generally consistent with observational data and published IRs of other treatments. As expected, IRs of thromboembolic events were elevated in patients with versus without baseline cardiovascular or VTE risk factors, and were broadly consistent with those observed in the Study A3921133 ad hoc safety analysis data, although the IR (95% CI) for PE was greater in patients treated with tofacitinib 10 mg twice daily in Study A3921133 (0.54 (0.32-0.87)), versus patients with baseline cardiovascular risk factors treated with tofacitinib 10 mg twice daily in the RA programme (0.24 (0.13-0.41)).

The impact of database restriction on pharmacovigilance signal detection of selected cancer therapies.

BACKGROUND: The aim of this study was to investigate whether database restriction can improve oncology drug pharmacovigilance signal detection performance. METHODS: We used spontaneous adverse event (AE) reports in the United States (US) Food and Drug Administration (FDA) Adverse Event Reporting System (FAERS) database. Positive control (PC) drug medical concept (DMC) pairs were selected from safety information not included in the product's first label but subsequently added as label changes. These medical concepts (MCs) were mapped to the Medical Dictionary for Regulatory Activities (MedDRA) preferred terms (PTs) used in FAERS to code AEs. Negative controls (NC) were MCs with circumscribed PTs not included in the corresponding US package insert (USPI). We calculated shrinkage-adjusted observed-to-expected (O/E) reporting frequencies for the aforementioned drug-PT pairs. We also formulated an adjudication framework to calculate performance at the MC level. Performance metrics [sensitivity, specificity, positive and negative predictive value (PPV, NPV), signal/noise (S/N), F and Matthews correlation coefficient (MCC)] were calculated for each analysis and compared. RESULTS: The PC reference set consisted of 11 drugs, 487 PTs, 27 MCs, 37 drug-MC combinations and 638 drug-event combinations (DECs). The NC reference set consisted of 11 drugs, 9 PTs, 5 MCs, 40 drug-MC combinations and 67 DECs. Most drug-event pairs were not highlighted by either analysis. A small percentage of signals of disproportionate reporting were lost, more noise than signal, with no gains. Specificity and PPV improved whereas sensitivity, NPV, F and MCC decreased, but all changes were small relative to the decrease in sensitivity. The overall S/N improved. CONCLUSION: This oncology drug restricted analysis improved the S/N ratio, removing proportionately more noise than signal, but with significant credible signal loss. Without broader experience and a calculus of costs and utilities of correct versus incorrect classifications in oncology pharmacovigilance such restricted analyses should be optional rather than a default analysis.

Pneumothorax as an adverse drug event: an exploratory aggregate analysis of the US FDA AERS database including a confounding by indication analysis inspired by Cornfield's condition.

INTRODUCTION: Pneumothorax is either primary or secondary. Secondary pneumothorax is usually due to trauma, including various non-pharmacologic iatrogenic triggers. Although not normally thought of as an adverse drug event (ADE) secondary pneumothorax is associated with numerous drugs, though it is often difficult to determine whether this association is causal in nature, or reflects an epiphenomenon of efficacy or inefficacy, or confounding by indication (CBI). Herein we explore this association in a large health authority drug safety surveillance database. METHODS: A quantitative pharmacovigilance (PhV) methodology known as disproportionality analysis was applied to the United States Food and Drug Administration (US FDA) Adverse Event Reporting System (AERS) database to explore the quantitative reporting dependencies between drugs and the adverse event pneumothorax as well the corresponding reporting dependencies between drugs and reported indications that may be risk factors for pneumothorax themselves in order to explore the potential contribution of CBI. RESULTS: We found 1. Multiple drugs are associated with pneumothorax; 2. Surfactants and oncology drugs account for most statistically distinctive associations with pneumothorax; 3. Pulmonary surfactants, pentamidine and nitric oxide have the largest statistical reporting associations 4. CBI may play a prominent role in reports of drug-associated pneumothorax. CONCLUSIONS: Disproportionality analysis (DA) can provide insights into the spontaneous reporting dependencies between drugs and pneumothorax. CBI assessment based on DA and Cornfield's inequality presents an additional novel option for the initial exploration of potential safety signals in PhV.

Paradoxical and bidirectional drug effects.

A paradoxical drug reaction constitutes an outcome that is opposite from the outcome that would be expected from the drug's known actions. There are three types: 1. A paradoxical response in a condition for which the drug is being explicitly prescribed. 2. Paradoxical precipitation of a condition for which the drug is indicated, when the drug is being used for an alternative indication. 3. Effects that are paradoxical in relation to an aspect of the pharmacology of the drug but unrelated to the usual indication. In bidirectional drug reactions, a drug may produce opposite effects, either in the same or different individuals, the effects usually being different from the expected beneficial effect. Paradoxical and bidirectional drug effects can sometimes be harnessed for benefit; some may be adverse. Such reactions arise in a wide variety of drug classes. Some are common; others are reported in single case reports. Paradoxical effects are often adverse, since they are opposite the direction of the expected effect. They may complicate the assessment of adverse drug reactions, pharmacovigilance, and clinical management. Bidirectional effects may be clinically useful or adverse. From a clinical toxicological perspective, altered pharmacokinetics or pharmacodynamics in overdose may exacerbate paradoxical and bidirectional effects. Certain antidotes have paradoxical attributes, complicating management. Apparent clinical paradoxical or bidirectional effects and reactions ensue when conflicts arise at different levels in self-regulating biological systems, as complexity increases from subcellular components, such as receptors, to cells, tissues, organs, and the whole individual. These may be incompletely understood. Mechanisms of such effects include different actions at the same receptor, owing to changes with time and downstream effects; stereochemical effects; multiple receptor targets with or without associated temporal effects; antibody-mediated reactions; three-dimensional architectural constraints; pharmacokinetic competing compartment effects; disruption and non-linear effects in oscillating systems, systemic overcompensation, and other higher-level feedback mechanisms and feedback response loops at multiple levels. Here we review and provide a compendium of multiple class effects and individual reactions, relevant mechanisms, and specific clinical toxicological considerations of antibiotics, immune modulators, antineoplastic drugs, and cardiovascular, CNS, dermal, endocrine, musculoskeletal, gastrointestinal, haematological, respiratory, and psychotropic agents.

Modelling the time to onset of adverse reactions with parametric survival distributions: a potential approach to signal detection and evaluation.

BACKGROUND: It has been postulated that the time to onset of adverse drug reactions is connected to the underlying pharmacological (or toxic) mechanism of adverse drug reactions whether the reaction is time dependent or not. OBJECTIVE: We have conducted a preliminary study using the parametric modelling of the time to onset of adverse reactions as an approach to signal detection on spontaneous reporting system databases. METHODS: We performed a parametric modelling of the reported time to onset of adverse drug reactions for which the underlying toxic mechanism is characterized. For the purpose of our study, we have used the reported liver injuries associated with bosentan, and the infections associated with the use of the tumour necrosis factor (TNF) inhibitors, adalimumab, etanercept and infliximab, which are used in Crohn's disease and rheumatoid arthritis, reported to EudraVigilance between December 2001 and September 2006. RESULTS: The main results reflect the fact that the reported time to onset is a surrogate of the true time to onset of the reaction and combines three hazards (occurrence, diagnosis and reporting) that cannot be disentangled. Consequently, the modelling of the time to onset of reactions reported with TNF inhibitors showed differences that could reflect different pharmacological activities, indications, monitoring of the patients or different reporting patterns. These variations could also limit the interpretation of the parametric modelling. CONCLUSIONS: Some consistency that was found between the occurrences of the infections with the TNF inhibitors suggests a causal association. There are statistical issues that are important to keep in mind when interpreting the results (the impact of the data quality on the fit of the distributions and the absence of a test of hypothesis linked to the absence of a relevant comparator). The study suggests that the modelling of the reported time to onset of adverse reactions could be a useful adjunct to other signal detection methods.

Endotoxin-like reactions with intravenous gentamicin: results from pharmacovigilance tools under investigation.

OBJECTIVE: To apply two data mining algorithms (DMAs) to Food and Drug Administration (FDA) Adverse Event Reporting System (AERS) reports that involved endotoxin-like reactions with intravenous gentamicin to determine whether a signal of disproportionate reporting of these events would have been generated concurrently with surveillance based on clinical observation. DESIGN: Multi-item gamma-Poisson shrinker (MGPS) and proportional reporting ratios (PRRs) were used. Data used for data mining consisted of an extract of the FDA AERS database. Previously published details of clusters of endotoxin-like reactions to intravenous gentamicin were used to select adverse events for data mining. RESULTS: The first signal of disproportionate reporting with any relevant event occurred in 1998, the year in which the outbreak was identified and evaluated by the Centers for Disease Control and Prevention and the FDA. In 1997, there were only 6 reports of rigors in the AERS; this jumped to 68 in 1998. In 1998, a signal was generated for endotoxic shock with PRRs but not with MGPS, based on one case. CONCLUSIONS: The two DMAs generated signals concurrently with the influx of reports. It would have been difficult for safety reviewers to ignore an increase in rigors by traditional methods of safety surveillance; therefore, DMAs might not have had a great deal to offer in this instance. If data mining were considered as a second-line defense to diligent clinical observations under similar circumstances, simple disproportionality methods such as PRRs might be more useful than DMAs such as MGPS when commonly cited thresholds are used.

Postmarketing surveillance of potentially fatal reactions to oncology drugs: potential utility of two signal-detection algorithms.

PURPOSE: Several data mining algorithms (DMAs) are being studied in hopes of enhancing screening of large post-marketing safety databases for signals of novel adverse events (AEs). The objective of this study was to apply two DMAs to the United States FDA Adverse Event Reporting System (AERS) database to see whether signals of potentially fatal AEs with cancer drugs might have been identified earlier than with traditional methods. METHODS: Screening algorithms used for analysis were the multi-item gamma Poisson shrinker (MGPS) and proportional reporting ratios (PRRs). Data mining was performed on data from the FDA AERS database. When a signal was identified, it was compared with that in the year in which the event was added to package insert and/or the year a "case series" was published. A recent publication summarizing the time of dissemination of information on potentially fatal AEs to cancer drugs provided the data set for analysis. RESULTS: The peer-reviewed published analysis contained 21 drugs and 26 drug-event combinations (DECs) that were considered sufficiently specific for data mining. Twenty-four of the DECs generated a signal of disproportionate reporting with PRRs (6 at 1 year and 16 from 2 years to 18 years prior to either a published "case series" or a package insert change) and 20 with MGPS (3 at 1 year and 11 from 2 years to 16 years prior to either a published "case series" or a package insert change). Two DECs did not signal with either DMA. CONCLUSION: At least one commonly cited DMA generated a signal of disproportionate reporting for 24 of 26 DECs for selected cancer drugs. For 16 DECs, one could conclude that a signal was generated well in advance (> or =2 years) of standard techniques in use with at least one DMA. DMAs might be useful in supplementing traditional surveillance strategies with oncology drugs and other drugs with similar features. (i.e., drugs that may be approved on an accelerated basis, are known to have serious toxicity, are administered to patients with substantial and complicated comorbid illness, are not available to the general medical community, and may have a high frequency of "off-label" use).