Current safeguards, risk mitigation, and transparency measures of large language models against the generation of health disinformation: repeated cross sectional analysis.

OBJECTIVES: To evaluate the effectiveness of safeguards to prevent large language models (LLMs) from being misused to generate health disinformation, and to evaluate the transparency of artificial intelligence (AI) developers regarding their risk mitigation processes against observed vulnerabilities. DESIGN: Repeated cross sectional analysis. SETTING: Publicly accessible LLMs. METHODS: In a repeated cross sectional analysis, four LLMs (via chatbots/assistant interfaces) were evaluated: OpenAI's GPT-4 (via ChatGPT and Microsoft's Copilot), Google's PaLM 2 and newly released Gemini Pro (via Bard), Anthropic's Claude 2 (via Poe), and Meta's Llama 2 (via HuggingChat). In September 2023, these LLMs were prompted to generate health disinformation on two topics: sunscreen as a cause of skin cancer and the alkaline diet as a cancer cure. Jailbreaking techniques (ie, attempts to bypass safeguards) were evaluated if required. For LLMs with observed safeguarding vulnerabilities, the processes for reporting outputs of concern were audited. 12 weeks after initial investigations, the disinformation generation capabilities of the LLMs were re-evaluated to assess any subsequent improvements in safeguards. MAIN OUTCOME MEASURES: The main outcome measures were whether safeguards prevented the generation of health disinformation, and the transparency of risk mitigation processes against health disinformation. RESULTS: Claude 2 (via Poe) declined 130 prompts submitted across the two study timepoints requesting the generation of content claiming that sunscreen causes skin cancer or that the alkaline diet is a cure for cancer, even with jailbreaking attempts. GPT-4 (via Copilot) initially refused to generate health disinformation, even with jailbreaking attempts-although this was not the case at 12 weeks. In contrast, GPT-4 (via ChatGPT), PaLM 2/Gemini Pro (via Bard), and Llama 2 (via HuggingChat) consistently generated health disinformation blogs. In September 2023 evaluations, these LLMs facilitated the generation of 113 unique cancer disinformation blogs, totalling more than 40 000 words, without requiring jailbreaking attempts. The refusal rate across the evaluation timepoints for these LLMs was only 5% (7 of 150), and as prompted the LLM generated blogs incorporated attention grabbing titles, authentic looking (fake or fictional) references, fabricated testimonials from patients and clinicians, and they targeted diverse demographic groups. Although each LLM evaluated had mechanisms to report observed outputs of concern, the developers did not respond when observations of vulnerabilities were reported. CONCLUSIONS: This study found that although effective safeguards are feasible to prevent LLMs from being misused to generate health disinformation, they were inconsistently implemented. Furthermore, effective processes for reporting safeguard problems were lacking. Enhanced regulation, transparency, and routine auditing are required to help prevent LLMs from contributing to the mass generation of health disinformation.

The Role of the Pharmacist in Optimizing Cancer Immunotherapy: A Retrospective Study of Nivolumab Adverse Events.

BACKGROUND: Immune checkpoint inhibitors (ICIs) are an emerging treatment in cancer therapy for prolonging life, minimizing symptoms, and selectively targeting cancer. Program death 1 (PD-1) inhibitors, such as nivolumab, fall within this class, enabling the patient's immune system to detect and destroy cancer. The introduction of ICIs is changing cancer therapy, with new drugs and new toxicities-an evolving area encountered by pharmacists. OBJECTIVE: This study aims to compare the pattern of nivolumab-induced adverse events observed in practice, when compared with clinical trial and literature data. The secondary aim of the study is to identify the presentation and treatment modalities initiated in practice. METHODS: We performed a retrospective case note review across 2 South Australian hospitals to identify the common toxicities and symptomatic treatments experienced by patients receiving nivolumab. Results were compared with clinical trial data from product innovator Bristol-Myer Squib and other published literature. RESULTS: Seventy patients were included in the study; of these, 60 (86%) experienced any grade adverse event(s). A total of 59 (84%) of 70 experienced mild to moderate grade 1 to grade 2 adverse events and 10 (14%) of 70 patients experienced severe grade 3 to grade 4 adverse events, displaying some consistencies with clinical trial and published literature data. Together, the prevalence of adverse events with details on presentation and treatments illustrates possible pharmacy practice strategies and areas for intervention. CONCLUSIONS: The listed prevalence of adverse events and practice strategies identified throughout this study highlights how pharmacists may assist in the identification of predictable ICI toxicities associated with gastrointestinal, endocrine, dermatological toxicities, and fatigue.

Barriers and opportunities for the clinical implementation of therapeutic drug monitoring in oncology.

There are few fields of medicine in which the individualisation of medicines is more important than in the area of oncology. Under-dosing can have significant ramifications due to the potential for therapeutic failure and cancer progression; by contrast, over-dosing may lead to severe treatment-limiting side effects, such as agranulocytosis and neutropenia. Both circumstances lead to poor patient prognosis and contribute to the high mortality rates still seen in oncology. The concept of dose individualisation tailors dosing for each individual patient to ensure optimal drug exposure and best clinical outcomes. While the value of this strategy is well recognised, it has seen little translation to clinical application. However, it is important to recognise that the clinical setting of oncology is unlike that for which therapeutic drug monitoring (TDM) is currently the cornerstone of therapy (e.g. antimicrobials). Whilst there is much to learn from these established TDM settings, the challenges presented in the treatment of cancer must be considered to ensure the implementation of TDM in clinical practice. Recent advancements in a range of scientific disciplines have the capacity to address the current system limitations and significantly enhance the use of anticancer medicines to improve patient health. This review examines opportunities presented by these innovative scientific methodologies, specifically sampling strategies, bioanalytics and dosing decision support, to enable optimal practice and facilitate the clinical implementation of TDM in oncology.

Nuances to precision dosing strategies of targeted cancer medicines.

Selecting the dose of a targeted cancer medicine that is most appropriate for a specific individual is a rational approach to maximize therapeutic outcomes and minimize toxicity. There are many different options for optimizing the dose of targeted cancer medicines and the purpose of this review is to provide a comprehensive comparison of the main options explored in prospective studies. Precision initial dose selection of targeted cancer therapies has been minimally explored to date; however, concentration, toxicity, and therapeutic outcome markers are used to guide on-therapy dose adaption of targeted cancer therapies across several medicines and cancers. While a specific concentration, toxicity, or therapeutic outcome marker commonly dominates an investigated precision on-therapy dose adaption strategy, greater attention to simultaneously account for exposure, toxicity, therapeutic outcomes, disease status, time since treatment initiation and patient preferences are required for optimal patient outcomes. To enable successful implementation of precision dosing strategies for targeted cancer medicines into clinical practice, future prospective studies aiming to develop strategies should consider these elements in their design.