Christof Manuel Schönenberger

Introduction: Tobacco smoking is an insufficiently addressed health burden among people living with HIV. The Reduce tobacco use in people living with HIV in Switzerland (RETUNE) trial tests the effectiveness of offering a menu of nicotine substitute products, including e-cigarettes, nicotine...

Background: Among people living with HIV, there has been a shift of focus from HIV-related health issues to cardiovascular diseases and cancer. For both, tobacco smoking is a major but insufficiently addressed etiological factor. Evidence from randomized trials suggests that nicotine substitute products such as e-cigarettes and nicotine patches can reduce tobacco smoking and its associated health burden. However, most previous smoking cessation trials primarily included people who are motivated to quit smoking and focused on testing a single nicotine substitute product. The effectiveness of offering a menu of nicotine substitute products to tobacco smokers regardless of their willingness to quit smoking (“opt-out” approach) is unknown. Methods: Reduce tobacco use in people living with HIV in Switzerland (RETUNE, NCT06789692) is a pragmatic, 1:1 randomized, multicenter, superiority clinical trial using the Trials within Cohorts (TwiCs) design within the Swiss HIV Cohort Study. RETUNE assesses the effectiveness of offering a menu of different nicotine substitute products, namely electronic cigarettes, nicotine pouches, and nicotine patches, versus usual care. Cohort participants are eligible if they smoke more than one tobacco cigarette per day, do not use any of the substitute products, and have signed the randomization consent following the TwiCs design. Participants randomized to the intervention may choose any of the offered substitutes to be used free of charge for 6 months or decline the offer. Overall, we plan to recruit 972 participants. The primary outcome is tobacco abstinence at 6 months measured as participant-reported past 7-day prevalence abstinence. The primary outcome will be assessed in the intention-to-treat set using a logistic regression model adjusted for region, men having sex with men, current drug users, and number of cigarettes per day at baseline. Secondary outcomes are long-term smoking cessation rates and tobacco-associated health outcomes. Discussion: RETUNE started recruitment in February 2025 and is currently ongoing. RETUNE using the TwiCs design will clarify the effectiveness of a preference-based opt-out smoking cessation intervention among people living with HIV. Trial registration: Clinicaltrials.gov NCT06789692. Registered on January 17th, 2025. The manuscript is aligned with the registry. https://clinicaltrials.gov/study/NCT06789692?cond=NCT06789692&rank=1

Background: Children and adolescents with HIV have lower treatment success than adults. Suboptimal adherence and resistance to antiretroviral therapy (ART) are known aetiological factors. This preplanned analysis in the GIVE MOVE trial (NCT04233242) describes drug resistance patterns in children and adolescents in Lesotho and Tanzania. Materials and Methods: GIVE MOVE randomized children and adolescents (6 months to below 19 years) with recent viraemia whilst taking ART to genotypic resistance testing (GRT)-informed care or usual care. Here, we conducted additional post-hoc GRT on stored samples from both groups and included participants with at least one successful resistance test. We assessed the number of drugs predicted to be active in participants’ three-drug ART regimens and resistance-associated mutations. Results: Amongst 137 participants, the majority were female (58%) and lived in Lesotho (77%). At their initial GRT, 69/137 (50%) were receiving protease inhibitor-based, 59/137 (43%) dolutegravir-based and 9/137 (7%) efavirenz-based ART. At that time, 80/137 (58%) participants had three, whilst 8/137 (6%) had two, 36/137 (26%) had one and 13/137 (9%) had no drugs predicted to be active in their regimens. Seventeen (12%) participants had resistance against their ART core agent, including one with high-level dolutegravir resistance. Across 312 detected resistance-associated mutations (222 major, 90 accessory), 146 conferred resistance to non-nucleoside reverse transcriptase inhibitors, 127 to nucleoside reverse transcriptase inhibitors, 28 to protease inhibitors and 11 to integrase strand transfer inhibitors. Conclusion: Given that more than half had an ART regimen predicted to be fully active, most viraemia in children and adolescents could not be explained by resistance. Registration: The GIVE MOVE trial was registered on Clinicaltrials.gov NCT04233242

Importance: Manual data extraction from clinical text is resource intensive. Locally hosted large language models (LLMs) may offer a privacy-preserving solution, but their performance on non-English data remains unclear. Objective: To investigate whether the classification accuracy of locally hosted LLMs is non-inferior to human accuracy when determining metastasis status and treatment response from German radiology reports. Design: In this retrospective comparative accuracy study, five locally hosted LLMs (llama3.3:70b, mistral-small:24b, qwq:32b, qwen3:32b, and gpt-oss:120b) were compared against humans. To calculate accuracy, a ground truth was established via duplicate human extraction and adjudication of discrepancies by a senior oncologist. Both initial human extraction and LLM outputs were compared against this ground truth. Setting: The study was conducted at a tertiary referral hospital in Switzerland; data processing and analyses took place inside the hospital network. Participants: 400 randomly sampled radiology reports from adult cancer patients (CT, MRI, PET) generated between January 2023 and May 2025. Exposures: Automated classification of metastasis status and treatment response by LLMs using a standardized prompt pipeline compared to manual human review. Main Outcomes and Measures: Primary outcomes were non-inferiority (5 percentage points [pp] margin) of LLM classification accuracy compared with human accuracy for metastasis status (presence/absence by anatomical site) and treatment response categories. Secondary outcomes included accuracy for primary tumor diagnosis, radiological absence of tumor, and extraction time per report. Results: The analysis included 400 reports from 317 patients (mean age 63 years, 32% women). On the test set (n=300), human accuracy for metastasis status was 98.4% (95% CI 98.0%–98.8%). All LLMs were non-inferior; gpt-oss:120b performed best (97.6% accuracy; difference:xs −0.8pp [90% CI, −1.3 to −0.3 pp]). For response to treatment, human accuracy was 86.0% (95% CI 83.2%–88.8%). All LLMs were inferior; the most accurate model, gpt-oss:120b, achieved 78.3% (difference −7.7 pp [90% CI, −11.6 to −3.8 pp]). Mean human time per report was 120 seconds vs 11–63 seconds for LLMs. Conclusion and Relevance: In this study, LLMs were non-inferior to human accuracy for classification of metastasis status but were inferior for response to treatment assessment. gpt-oss:120b was the most accurate among tested LLMs. Study Registration: OSF: 45PVQ Key Points: Question: Can locally hosted large language models (LLMs) match human performance when extracting sites of metastases and response to treatment from radiology reports of cancer patients? Findings: In this preregistered, single center study of 300 German radiology reports, all evaluated LLMs were non-inferior to humans in extracting the presence or absence of metastasis by organ site, but LLMs were inferior to humans in classification of response to treatment. Meaning: LLMs can be suitable for classification of metastasis status, whereas more caution is warranted for more complex tasks where additional clinical reasoning may be required.