Elizabeth Phillips

Large language models (LLMs) have emerged as transformative technologies, revolutionising natural language understanding and generation across various domains, including medicine. In this study, we investigated the capabilities, limitations, and generalisability of Generative Pre-trained Transformer (GPT) models in analysing unstructured patient notes from large healthcare datasets to identify immune-related adverse events (irAEs) associated with the use of immune checkpoint inhibitor (ICI) therapy.
We evaluated the performance of GPT-3.5, GPT-4, and GPT-4o models on manually annotated datasets of patients receiving ICI therapy, sampled from two electronic health record (EHR) systems and seven clinical trials. A zero-shot prompt was designed to exhaustively identify irAEs at both the patient level (main analysis) and the note level (secondary analysis). The LLM-based system followed a multi-label classification approach to identify any combination of irAEs associated with individual patients or clinical notes. System evaluation was conducted for each available irAE as well as for broader categories of irAEs classified at the organ level.
Our analysis included 442 patients across three institutions. The most common irAEs manually identified in the patient datasets included pneumonitis (N = 64), colitis (N = 56), rash (N = 32), and hepatitis (N = 28). The GPT models demonstrated generalisable abilities in identifying irAEs across EHRs and clinical trial reports. Overall, the models achieved relatively high sensitivity and specificity but only moderate positive predictive values, reflecting a potential bias towards overpredicting irAE outcomes. GPT-4o achieved the highest F1 and micro-averaged F1 scores for both patient-level and note-level evaluations. Highest performance was observed in the haematological (F1 range = 1.0–1.0), gastrointestinal (F1 range = 0.81–0.85), and musculoskeletal and rheumatologic (F1 range = 0.67–1.0) irAE categories. Error analysis uncovered substantial limitations of GPT models in handling textual causation, where adverse events should not only be accurately identified in clinical text but also causally linked to immune checkpoint inhibitors.
This study demonstrated that GPT models can automate the detection of immune related adverse events in varied healthcare datasets, reducing the burden on physicians and other healthcare professionals by limiting the need for manual review. This capability will accelerate the generation of safety insights across large healthcare datasets and facilitate the characterisation of patient-level drivers of toxicities, thus enhancing safety monitoring and ultimately improving patient care.
National Institutes of Health, Roche, National Health and Medical Research Council of Australia, Stevens-Johnson Syndrome Foundation, Angela Anderson Research Fund, Larry L Hillblom Foundation and UCSF Research Allocation Program.

Immune checkpoint inhibitors are used in a wide range of cancers, offering durable responses for a substantial subset of patients. However, immune-related adverse events, the most clinically consequential checkpoint inhibitor–associated adverse reactions, pose a key challenge in practice, affecting virtually any organ system, resulting in treatment interruption, morbidity, or mortality. Patient education, early recognition, and effective management are essential to limit complications and maintain continuity of immunotherapy. Achieving this requires well-informed multidisciplinary teams who can identify, evaluate, and manage immune-related adverse events promptly. This review summarizes the most clinically significant immune-related adverse events and highlights the key principles of multidisciplinary diagnosis and management most relevant to the practicing allergist-immunologist to optimize patient outcomes.

Importance; Stevens-Johnson Syndrome and Toxic Epidermal Necrolysis (SJS/TEN) are rare, potentially fatal adverse drug reactions. As the use of immune checkpoint inhibitors (ICIs) expands, their role as direct inducers or synergistic contributors to SJS/TEN remains incompletely characterized.

Objective: To determine whether ICIs are independent risk factors for SJS/TEN, evaluate their interactions with known culprit drugs, and assess their impact on latency and mortality.

Design: Cross-sectional analysis of adverse event reports submitted to the U.S. Food and Drug Administration Adverse Event Reporting System (FDA FAERS) between January 2013 and December 2023, sanitized and de-duplicated. Logistic regression and Cox models were used to assess predictors of SJS/TEN development, mortality, and latency.

Setting: Global pharmacovigilance reports submitted to FAERS.

Participants: A total of 17,495 unique and de-identified patients reported SJS/TEN, of 13,986,839 total reports.

Exposures: Suspected causative drugs, including ICIs.

Main Outcomes and Measures: Primary outcomes were the adjusted odds of developing SJS/TEN, time-to-event (TTE) of reaction/drug latency, and all-cause mortality. Depending on the analysis, covariates included age, sex, number of concomitant drugs, cancer diagnosis, and specific drug exposures.

Results: Of 17,495 SJS/TEN cases (median age 53 years, 37.6% male), 970 (5.5%) had ICI exposure and 653 (3.7%) listed an ICI as the primary suspect. ICI exposure was associated with developing SJS/TEN (adjusted OR, 6.69; 95% CI, 6.19-7.23) while controlling for age, exposure to strong and weak culprits, number of concomitant drugs, and cancer diagnosis. ICI increases SJS-TEN risk among patients exposed to allopurinol (OR, 4.35; 95% CI, 3.12-6.06) and TMP-SMX (OR, 5.68; 95% CI, 4.05-7.95) with the same covariates. Among patients with small-molecule-induced SJS/TEN, mortality was strongly associated with ICI exposure (particularly exposure to multiple ICI, OR, 7.31; 95% CI, 3.09-17.27). Among all SJS/TEN cases, ICI exposure was associated with delayed onset, compared to cancer patients not exposed to ICI and non-cancer patients (median 20 vs 14 vs 13 days; P < .0001).

Conclusions and Relevance: ICIs are associated with increased SJS/TEN risk, both independently and in combination with known culprit drugs, and may delay disease onset. These findings support increased vigilance in prescribing known culprits alongside ICI.

Co-trimoxazole is a leading global cause of severe cutaneous adverse drug reactions (SCAR) including Stevens-Johnson syndrome/toxic epidermal necrolysis (SJS/TEN) and drug reaction with eosinophilia and systemic symptoms (DRESS). Co-trimoxazole-induced SCAR are associated with HLA class I alleles including HLA-B*13:01 and HLA-B*38:02 in Southeast Asian (SEA) populations. However, the global generalizability of these associations is unknown but critical for population-appropriate risk stratification and diagnosis.
To determine HLA risk factors associated with co-trimoxazole-induced SJS/TEN and DRESS in populations from the United States (US) and South Africa (SA).
We performed high-resolution HLA typing on dermatologist-adjudicated co-trimoxazole-induced SCAR patients in the US (n=63) and SA (n=26) compared to population controls. Peptide binding and docking analyses were performed using MHCcluster2.0 and CB-Dock2.
In a multiple logistic regression model, HLA-B*44:03 (Pc<0.001, OR: 4.08), HLA-B*38:01 (Pc<0.001, OR: 5.66), and HLA-C*04:01 (Pc=0.003, OR: 2.50) were independently associated with co-trimoxazole-induced SJS/TEN in the US. HLA-B*44:03 was also associated with co-trimoxazole-induced DRESS in SA (Pc=0.019, OR: 10.69). Distinct HLA-B variants with shared peptide binding specificities (SPBS) and HLA-C*04:01 identified 94% and 78% of co-trimoxazole-induced SJS/TEN and DRESS in the US, respectively. The SEA risk allele HLA-B*13:01, with SPBS to HLA-B*44:03, was identified in just 1/63 US SCAR patients.
HLA alleles with SPBS to SEA-related risk alleles including HLA-B*44:03 (SPBS with HLA-B*13:01) and HLA-B*38:01 (SPBS with HLA-B*38:02) but also HLA-C*04:01 predisposed to co-trimoxazole-induced SCAR in the US and SA. These findings provide biological plausibility and strategies for global risk prediction and diagnosis of co-trimoxazole-induced SCAR.
HLA alleles including HLA-B*13:01 and HLA-B*38:02 are risk factors for co-trimoxazole-induced SCAR in Asian populations. However, the generalizability of these associations to other global populations is unknown but critical for population-appropriate risk stratification and diagnosis.
HLA alleles with shared peptide binding specificities (SPBS) to Asian-related risk alleles including HLA-B*44:03 (SPBS with HLA-B*13:01) and HLA-B*38:01 (SPBS with HLA-B*38:02) but also HLA-C*04:01 predisposed to co-trimoxazole-induced SCAR in the US and South Africa.
HLA alleles previously associated with co-trimoxazole-induced SCAR do not identify risk across populations. However, HLA alleles with SPBS provide biological plausibility and strategies for global and population-appropriate clinical risk stratification and diagnosis of cotrimoxazole-induced SCAR.

Adverse drug reactions (ADR) are a significant concern in medicine due to their potential to cause substantial morbidity and mortality. Among the most serious of ADRs are severe cutaneous adverse reactions (SCARs), including Stevens-Johnson Syndrome (SJS)/toxic epidermal necrolysis (TEN), drug reaction with eosinophilia and systemic symptoms (DRESS), acute generalized exanthematous pustulosis (AGEP), and generalized bullous fixed drug eruption (GBFDE). These conditions differ in phenotype, causative drugs, demographics, and latency (time between administration and reaction). We used the FDA Adverse Event Reporting System (FAERS), a comprehensive database with millions of reports submitted by providers, patients, and manufacturers, applying disproportionality measures and machine learning to analyze these rare reactions at scale. After sanitization and deduplication, FAERS was queried from 2004 to 2003 for SCAR cases. A total of 56,683 cases were reported, with median age 53 years (interquartile range [IQR] 32-68), with significant differences in age between phenotypes. Over 200 drugs had positive disproportionality signals. SCAR reporting has increased over time, particularly to biologics and checkpoint inhibitors. Using random forest classifiers, we showed causative drug is the most influential variable on latency, followed by number of concomitant drugs, and mortality is most strongly tied to age and number of concomitant drugs, regardless of SCAR phenotype. This largest retrospective study of SCAR to date shows the variety of phenotypes, causative agents, demographic variables, latencies, and mortality in SCAR patients. Continued mining of these databases, retrospective analyses of electronic health records, and prospective data can expand upon these results, better characterize variations, and improve recognition and care for patients with SCAR.

The liver is continuously exposed to a large number of substances, including pathogens, xenobiotics, tumor cells and harmless dietary antigens. Infectious agents from systemic circulation need to be efficiently removed, whereas tolerance needs to be developed against the large number of antigens derived from the gastrointestinal tract. To respond to these challenges, the adaptive immune responses in the liver favor tolerance rather than immunity. However, when regulation of the immune system goes awry, the delicate balance of immunity and tolerance in the liver is compromised, which can result in immune-mediated liver injury. In this article, we review the current literature in which the pathogenesis of liver injury is advanced by adaptive immunity, specifically in cases of primary biliary cirrhosis, viral hepatitis, alcoholic liver disease, and nonalcoholic fatty liver disease.