University of Campania Luigi Vanvitelli

Chatbots have been increasingly used as a source of patient education. This study aimed to compare the answers of ChatGPT-4 and Google Gemini to common questions on benign anal conditions in terms of appropriateness, comprehensiveness, and language level.

Each chatbot was asked a set of 30 questions on hemorrhoidal disease, anal fissures, and anal fistulas. The responses were assessed for appropriateness, comprehensiveness, and reference provision. The assessments were made by three subject experts who were unaware of the name of the chatbots. The language level of the chatbot answers was assessed using the Flesch-Kincaid Reading Ease score and grade level.

Overall, the answers provided by both models were appropriate and comprehensive. The answers of Google Gemini were more appropriate, comprehensive, and supported by references compared with the answers of ChatGPT. In addition, the agreement among the assessors on the appropriateness of Google Gemini answers was higher, attesting to a higher consistency. ChatGPT had a significantly higher Flesh-Kincaid grade level than Google Gemini (12.3 versus 10.6, p = 0.015), but a similar median Flesh-Kincaid Ease score.

The answers of Google Gemini to questions on common benign anal conditions were more appropriate and comprehensive, and more often supported with references, than the answers of ChatGPT. The answers of both chatbots were at grade levels higher than the 6th grade level, which may be difficult for nonmedical individuals to comprehend.

Boron Neutron Capture Therapy (BNCT) is attracting renewed attention due to advancements in compact proton accelerators for the production of neutron beams, and new BNCT facilities are being planned all around the world. A key aspect in BNCT treatments will be patient dosimetry, particularly given the complex radiation field created by neutron interactions with biological tissues. This study aimed at developing a prototype of BNCT dosimetry workflow based on Monte Carlo (MC) simulations for the GATE toolkit. Investigating the feasibility of performing voxel-level dosimetry through full MC transport in terms of simulation time and statistical uncertainties, the ICRP110 male and female adult voxelized phantoms were used to model the human body, adding the possibility to set in their organs user-defined concentrations of ¹⁰B. Irradiation simulations of the head district with two monoenergetic neutron beams and with a realistic clinical neutron spectrum were carried out. The absorbed dose matrices for each simulation, assuming both no ¹⁰B and then a systemic distribution of 15 ppm, were scored separating the contributions from ⁷Li, alpha particles, protons and photons. Results showed the expected increase, in presence of ¹⁰B distribution, of the ⁷Li and alpha average dose components in organs of interest of the head, such as brain, reaching in it about 1.3 and 2.3 fGy/evt, respectively, in presence of 15 ppm of ¹⁰B. The present prototype of dosimetric workflow, whose macros and files are freely shared for interested users and developers, will serve as a basis for future studies aiming at simulating similar BNCT scenarios with larger statistics, for example by exploiting high computing resources, to verify the obtained results with lower statistical uncertainties and possibly optimize the workflow to reduce simulation times while ensuring suitable dosimetric accuracy.

Negative symptoms in schizophrenia spectrum are associated with minimal treatment responses. The search for effective treatments is potentially hampered by heterogenous study-designs and sample characteristics depending on the intervention category. This PRISMA-compliant systematic review/synthesis aims to describe the literature on negative symptoms interventions for schizophrenia spectrum disorders by comparing 12 study design, demographical and clinical variables in different intervention categories: antipsychotics (AP), other pharmacological agents (OPA), brain stimulation (BS), psychological/psychosocial (PSI), lifestyle (LS), mixed interventions. Kruskal-Wallis and Chi-square tests measured differences between intervention-groups. Out of 19,935 articles, 489 (AP=149/OPA=187/BS=49/PSI=79/LS=19/mixed=6) were selected for data extraction. Concerning study designs, AP had the largest average arm size (mean ± SD=91.1 ± 122.8participants), OPA the highest double/triple-blinding (97.9 %) rates, PSI the longest follow-up (26.7 ± 21.8weeks). Age/gender demographical differences were significant but of negligible magnitude. OPA illness duration (14.8 ± 9.0years) was longer compared to AP (11.4 ± 6.7years). Positive and Negative Syndrome Scale (PANSS) negative scores were milder in PSI (18.6 ± 6.9) compared to AP/OPA/BS (23.8 ± 6.4/23.4 ± 4.9/24.2 ± 9.2). PANSS total scores were worse in AP (83.6 ± 18.2) than in OPA/BS/PSI (77.1 ± 20.5/75.5 ± 14.7/67.0 ± 23.3). The same was true for dropout rates (AP=25.5 %, OPA/BS/PSI=14.3/9.7/14.5 %). Prevalent treatment as usual was "none" for AP (36.7 %) and "antipsychotic" for other categories (42.3-82.8 %). Implementing cross-over, factorial or multi-arm designs may increase the comparability between studies investigating different intervention categories. Concerning clinical differences, reporting individual treatments at baseline and clinical severity, evaluating cognitive profiles and considering patients' perspectives will allow to better understand the efficacy of the available treatments and develop tailored interventions.