Background: In this study, we evaluate the accuracy, efficiency, and cost-effectiveness of large language models in extracting and structuring information from free-text clinical reports, particularly in identifying and classifying patient comorbidities within oncology electronic health records. We specifically compare the performance of gpt-3.5-turbo-1106 and gpt-4-1106-preview models against that of specialized human evaluators. Objective: We specifically compare the performance of gpt-3.5-turbo-1106 and gpt-4-1106-preview models against that of specialized human evaluators. Methods: We implemented a script using the OpenAI application programming interface to extract structured information in JavaScript object notation format from comorbidities reported in 250 personal history reports. These reports were manually reviewed in batches of 50 by 5 specialists in radiation oncology. We compared the results using metrics such as sensitivity, specificity, precision, accuracy, F-value, κ index, and the McNemar test, in addition to examining the common causes of errors in both humans and generative pretrained transformer (GPT) models. Results: The GPT-3.5 model exhibited slightly lower performance compared to physicians across all metrics, though the differences were not statistically significant (McNemar test, P=.79). GPT-4 demonstrated clear superiority in several key metrics (McNemar test, P<.001). Notably, it achieved a sensitivity of 96.8%, compared to 88.2% for GPT-3.5 and 88.8% for physicians. However, physicians marginally outperformed GPT-4 in precision (97.7% vs 96.8%). GPT-4 showed greater consistency, replicating the exact same results in 76% of the reports across 10 repeated analyses, compared to 59% for GPT-3.5, indicating more stable and reliable performance. Physicians were more likely to miss explicit comorbidities, while the GPT models more frequently inferred nonexplicit comorbidities, sometimes correctly, though this also resulted in more false positives. Conclusions: This study demonstrates that, with well-designed prompts, the large language models examined can match or even surpass medical specialists in extracting information from complex clinical reports. Their superior efficiency in time and costs, along with easy integration with databases, makes them a valuable tool for large-scale data mining and real-world evidence generation.

Genomic stability maintenance requires the repair of DNA breaks in the most accurate fashion. So, an exquisite regulatory network controls the choice between different repair mechanisms to maximize genome integrity. This relies mostly at the level of DNA end resection, the initial steps of the homologous recombination. On the other hand, numerous cellular activities follow a 24 h oscillation known as the circadian cycle. Thus, we explored the regulation of the choice between different DNA break repair pathways along the circadian cycle. Here we show that in human cells DNA resection shows a circadian oscillation, with a peak at early morning followed by a partial and progressive reduction until late afternoon. Such regulation depends on the circadian clock core component CRY1, which modulates the anti-resection activity of CCAR2 to limit CtIP at nightfall. Additionally, such regulation requires DNA-PK-mediated phosphorylation of CRY1. Finally, this circadian regulation impacts cancer progression and response to radiation therapy of specific tumours.

Introduction: In this study, we evaluate the accuracy, efficiency, and cost-effectiveness of Large Language Models (LLMs) in extracting and structuring information from free-text clinical reports, particularly in identifying and classifying patient comorbidities within oncology electronic health records. We specifically compare the performance of gpt-3.5-turbo-1106 and gpt-4-1106-preview models against that of specialized human evaluators. Methods: We implemented a script using the OpenAI API to extract structured information in JSON format from comorbidities reported in 250 personal history reports. These reports were manually reviewed in batches of 50 by five specialists in radiation oncology. We compared the results using metrics such as Sensitivity, Specificity, Precision, Accuracy, F-value, Kappa index, and the McNemar test, in addition to examining the common causes of errors in both humans and GPT models. Results: The GPT-3.5 model exhibited slightly lower performance compared to physicians across all metrics, though the differences were not statistically significant (McNemar’s test p = 0.79). GPT-4 demonstrated clear superiority in several key metrics (McNemar’s test p < 0.001). Notably, it achieved a sensitivity of 96.8%, compared to 88.2% for GPT-3.5 and 88.8% for physicians. However, physicians marginally outperformed GPT-4 in precision (97.7% vs. 96.8%). GPT-4 showed greater consistency, replicating the exact same results in 76% of the reports across 10 repeated analyses, compared to 59% for GPT-3.5, indicating more stable and reliable performance. Physicians were more likely to miss explicit comorbidities, while the GPT models more frequently inferred non-explicit comorbidities, sometimes correctly, though this also resulted in more false positives. Conclusion: This study demonstrates that, with well-designed prompts, the LLMs examined can match or even surpass medical specialists in extracting information from complex clinical reports. Their superior efficiency in time and costs, along with easy integration with databases, makes them a valuable tool for large-scale data mining and real-world evidence generation.

This study has developed a Matlab application for simulating statistical models project (SMp) probabilistic distributions that are similar to binomial and Poisson, which were created by mathematical procedures. The simulated distributions are graphically compared with these legendary distributions. The application allows to obtain many probabilistic distributions, and shows the trend (τ ) for n trials with success probability p, i.e. the maximum probability as τ=np. While the Poisson distribution PD(x;µ) is a unique probabilistic distribution, where PD=0 in x=+∞, the application simulates many SMp(x;µ,Xmax) distributions, where µ is the Poisson parameter and value of x with generally the maximum probability, and Xmax is upper limit of x with SMp(x;µ,Xmax) ≥ 0 and limit of the stochastic region of the random discrete variable X. It is shown that by simulation via, one can get many and better probabilistic distributions than by mathematical one.

The purpose of this work is to present useful recommendations for the use of [¹⁸F]FDG-PET/CT imaging in radiotherapy planning and monitoring under different versions of EARL accreditation for harmonization of PET devices. A proof-of-concept experiment designed on an anthropomorphic phantom was carried out to establish the most suitable interpolation methods of the PET images in the different steps of the planning procedure. Based on PET/CT images obtained by using these optimal interpolations for the old EARL accreditation (EARL1) and for the new one (EARL2), the treatment plannings of representative actual clinical cases were calculated, and the clinical implications of the resulting differences were analyzed. As expected, EARL2 provided smaller volumes with higher resolution than EARL1. The increase in the size of the reconstructed volumes with EARL1 accreditation caused high doses in the organs at risk and in the regions adjacent to the target volumes. EARL2 accreditation allowed an improvement in the accuracy of the PET imaging precision, allowing more personalized radiotherapy. This work provides recommendations for those centers that intend to benefit from the new accreditation, EARL2, and can help build confidence of those that must continue working under the EARL1 accreditation.

The maintenance of genomic stability is essential for cellular and organismal survival and fitness. Thus, when DNA gets damaged, is essential to repair it in the most accurate fashion. Among different DNA lesions, DNA double strand breaks are specially challenging. An exquisite regulatory network reacts to local and global cues to control the choice between different DNA double strand break repair mechanisms to maximize genomic integrity. Such regulation relies mostly at the level of DNA end resection, the initial steps of the homologous recombination repair pathway. On the other hand, most cellular and organismal activities follow a 24 h oscillation known as the circadian cycle. Such repetitive changes are controlled by an intrinsic, molecular clock built-in at the cellular level which core components are the heterodimers BMAL1-CLOCK and CRY-PER. These inherent rhythms control many different aspects of the cellular metabolism, including the fate of many different DNA transactions. Here we have explored the regulation of the choice between different DNA double strand break repair pathways along the circadian cycle. We observed that DNA end resection shows a circadian oscillation, with a peak at dawn followed by a progressive reduction until dusk. Such regulations depend on the cellular levels of the circadian clock core component CRY1. Consequently, repair by homologous recombination mirrors CRY1 expression levels. Such modulation is controlled through the circadian regulation of the anti-resection activity, but not the protein levels, of CCAR2, that limits CtIP-mediated resection preferentially at nightfall. Additionally, such regulation requires a crosstalk between the DNA damage-dependent phosphorylation of CRY1 by the kinase DNA-PK. Finally, such regulation has an impact in cancer progression and response to radiation therapy of specific tumors. One sentence summary CCAR2-dependent inhibition of DNA end resection and homologous recombination is controlled by the intrinsic cellular circadian clock