Algor-ethics: charting the ethical path for AI in critical care.

The integration of Clinical Decision Support Systems (CDSS) based on artificial intelligence (AI) in healthcare is groundbreaking evolution with enormous potential, but its development and ethical implementation, presents unique challenges, particularly in critical care, where physicians often deal with life-threating conditions requiring rapid actions and patients unable to participate in the decisional process. Moreover, development of AI-based CDSS is complex and should address different sources of bias, including data acquisition, health disparities, domain shifts during clinical use, and cognitive biases in decision-making. In this scenario algor-ethics is mandatory and emphasizes the integration of 'Human-in-the-Loop' and 'Algorithmic Stewardship' principles, and the benefits of advanced data engineering. The establishment of Clinical AI Departments (CAID) is necessary to lead AI innovation in healthcare, ensuring ethical integrity and human-centered development in this rapidly evolving field.

The Breakthrough of Large Language Models Release for Medical Applications: 1-Year Timeline and Perspectives.

Within the domain of Natural Language Processing (NLP), Large Language Models (LLMs) represent sophisticated models engineered to comprehend, generate, and manipulate text resembling human language on an extensive scale. They are transformer-based deep learning architectures, obtained through the scaling of model size, pretraining of corpora, and computational resources. The potential healthcare applications of these models primarily involve chatbots and interaction systems for clinical documentation management, and medical literature summarization (Biomedical NLP). The challenge in this field lies in the research for applications in diagnostic and clinical decision support, as well as patient triage. Therefore, LLMs can be used for multiple tasks within patient care, research, and education. Throughout 2023, there has been an escalation in the release of LLMs, some of which are applicable in the healthcare domain. This remarkable output is largely the effect of the customization of pre-trained models for applications like chatbots, virtual assistants, or any system requiring human-like conversational engagement. As healthcare professionals, we recognize the imperative to stay at the forefront of knowledge. However, keeping abreast of the rapid evolution of this technology is practically unattainable, and, above all, understanding its potential applications and limitations remains a subject of ongoing debate. Consequently, this article aims to provide a succinct overview of the recently released LLMs, emphasizing their potential use in the field of medicine. Perspectives for a more extensive range of safe and effective applications are also discussed. The upcoming evolutionary leap involves the transition from an AI-powered model primarily designed for answering medical questions to a more versatile and practical tool for healthcare providers such as generalist biomedical AI systems for multimodal-based calibrated decision-making processes. On the other hand, the development of more accurate virtual clinical partners could enhance patient engagement, offering personalized support, and improving chronic disease management.

Between human and AI: assessing the reliability of AI text detection tools.

OBJECTIVE: Large language models (LLMs) such as ChatGPT-4 have raised critical questions regarding their distinguishability from human-generated content. In this research, we evaluated the effectiveness of online detection tools in identifying ChatGPT-4 vs human-written text. METHODS: A two texts produced by ChatGPT-4 using differing prompts and one text created by a human author were analytically assessed using the following online detection tools: GPTZero, ZeroGPT, Writer ACD, and Originality. RESULTS: The findings revealed a notable variance in the detection capabilities of the employed detection tools. GPTZero and ZeroGPT exhibited inconsistent assessments regarding the AI-origin of the texts. Writer ACD predominantly identified texts as human-written, whereas Originality consistently recognized the AI-generated content in both samples from ChatGPT-4. This highlights Originality's enhanced sensitivity to patterns characteristic of AI-generated text. CONCLUSION: The study demonstrates that while automatic detection tools may discern texts generated by ChatGPT-4 significant variability exists in their accuracy. Undoubtedly, there is an urgent need for advanced detection tools to ensure the authenticity and integrity of content, especially in scientific and academic research. However, our findings underscore an urgent need for more refined detection methodologies to prevent the misdetection of human-written content as AI-generated and vice versa.

Failing categorization of severe COVID-19 ARDS into ventilatory subphenotypes studied via the clinical-histopathologic relationship.

BACKGROUND: Categorization of severe COVID-19 related acute respiratory distress syndrome (CARDS) into subphenotypes does not consider the trajectories of respiratory mechanoelastic features and histopathologic patterns. This study aimed to assess the correlation between mechanoelastic ventilatory features and lung histopathologic findings in critically ill patients who died because of CARDS. METHODS: Mechanically ventilated patients with severe CARDS who had daily ventilatory data were considered. The histopathologic assessment was performed through full autopsy of deceased patients. Patients were categorized into two groups according to the median worst respiratory system compliance during ICU stay (CrsICU). RESULTS: Eighty-seven patients admitted to ICU had daily ventilatory data. Fifty-one (58.6%) died in ICU, 41 (80.4%) underwent full autopsy and were considered for the clinical-histopathological correlation analysis. Respiratory system compliance at ICU admission and its trajectory were not different in survivors and non-survivors. Median CrsICU in the deceased patients was 22.9 ml/cmH2O. An inverse correlation was found between the CrsICU and late-proliferative diffuse alveolar damage (DAD) (r = -0.381, p = 0.026). Late proliferative DAD was more extensive (p = 0.042), and the probability of stay in ICU was higher (p = 0.004) in the "low" compared to the "high" CrsICU group. Cluster analysis further endorsed these findings. CONCLUSIONS: In critically ill mechanically ventilated patients, worsening of the respiratory system compliance correlated pathologically with the transition from early damage to late fibroproliferative patterns in non-survivors of CARDS. Categorization of CARDS into ventilatory subphenotypes by mechanoelastic properties at ICU admission does not account for the complexity of the histopathologic features.

Long-term outcome of COVID-19 patients treated with helmet noninvasive ventilation vs. high-flow nasal oxygen: a randomized trial.

BACKGROUND: Long-term outcomes of patients treated with helmet noninvasive ventilation (NIV) are unknown: safety concerns regarding the risk of patient self-inflicted lung injury and delayed intubation exist when NIV is applied in hypoxemic patients. We assessed the 6-month outcome of patients who received helmet NIV or high-flow nasal oxygen for COVID-19 hypoxemic respiratory failure. METHODS: In this prespecified analysis of a randomized trial of helmet NIV versus high-flow nasal oxygen (HENIVOT), clinical status, physical performance (6-min-walking-test and 30-s chair stand test), respiratory function and quality of life (EuroQoL five dimensions five levels questionnaire, EuroQoL VAS, SF36 and Post-Traumatic Stress Disorder Checklist for the DSM) were evaluated 6 months after the enrollment. RESULTS: Among 80 patients who were alive, 71 (89%) completed the follow-up: 35 had received helmet NIV, 36 high-flow oxygen. There was no inter-group difference in any item concerning vital signs (N = 4), physical performance (N = 18), respiratory function (N = 27), quality of life (N = 21) and laboratory tests (N = 15). Arthralgia was significantly lower in the helmet group (16% vs. 55%, p = 0.002). Fifty-two percent of patients in helmet group vs. 63% of patients in high-flow group had diffusing capacity of the lungs for carbon monoxide < 80% of predicted (p = 0.44); 13% vs. 22% had forced vital capacity < 80% of predicted (p = 0.51). Both groups reported similar degree of pain (p = 0.81) and anxiety (p = 0.81) at the EQ-5D-5L test; the EQ-VAS score was similar in the two groups (p = 0.27). Compared to patients who successfully avoided invasive mechanical ventilation (54/71, 76%), intubated patients (17/71, 24%) had significantly worse pulmonary function (median diffusing capacity of the lungs for carbon monoxide 66% [Interquartile range: 47-77] of predicted vs. 80% [71-88], p = 0.005) and decreased quality of life (EQ-VAS: 70 [53-70] vs. 80 [70-83], p = 0.01). CONCLUSIONS: In patients with COVID-19 hypoxemic respiratory failure, treatment with helmet NIV or high-flow oxygen yielded similar quality of life and functional outcome at 6 months. The need for invasive mechanical ventilation was associated with worse outcomes. These data indicate that helmet NIV, as applied in the HENIVOT trial, can be safely used in hypoxemic patients. Trial registration Registered on clinicaltrials.gov NCT04502576 on August 6, 2020.

From Big Data's 5Vs to clinical practice's 5Ws: enhancing data-driven decision making in healthcare.

The use of AI-based algorithms is rapidly growing in healthcare, but there is still an ongoing debate about how to manage and ensure accountability for their clinical use. While most of the studies focus on demonstrating a good algorithm performance it is important to acknowledge that several additional steps are needed for reaching an effective implementation of AI-based models in daily clinical practice, with implementation being one of the main key factors. We propose a model characterized by five questions that can guide in this process. Additionally, we believe that a hybrid intelligence, human and artificial respectively, is the new clinical paradigm that offer the most benefits for developing clinical decision support systems for bedside use.

Evaluating the Feasibility of ChatGPT in Healthcare: An Analysis of Multiple Clinical and Research Scenarios.

This paper aims to highlight the potential applications and limits of a large language model (LLM) in healthcare. ChatGPT is a recently developed LLM that was trained on a massive dataset of text for dialogue with users. Although AI-based language models like ChatGPT have demonstrated impressive capabilities, it is uncertain how well they will perform in real-world scenarios, particularly in fields such as medicine where high-level and complex thinking is necessary. Furthermore, while the use of ChatGPT in writing scientific articles and other scientific outputs may have potential benefits, important ethical concerns must also be addressed. Consequently, we investigated the feasibility of ChatGPT in clinical and research scenarios: (1) support of the clinical practice, (2) scientific production, (3) misuse in medicine and research, and (4) reasoning about public health topics. Results indicated that it is important to recognize and promote education on the appropriate use and potential pitfalls of AI-based LLMs in medicine.

Artificial intelligence in intensive care: moving towards clinical decision support systems.

The high complexity of care in the Intensive Care Unit environment has led, in the last decades, to a big effort in term of the improvement of patient's monitoring devices, increase of diagnostic and therapeutic opportunities, and development of electronic health records. Such advancements have enabled an increasing availability of large amounts of data that were supposed to provide more insight and understanding regarding pathophysiological processes and patient's prognosis providing useful tools able to support physicians in the clinical decision-making process. On the contrary, the interpolation, analysis, and interpretation of a such big amount of data has soon proven to be much more complicated than expected, opening the way for the development of tools based on machine learning (ML) algorithms. However, at the present, most of the AI-based algorithms developed in intensive care do not reach beyond the prototyping and development environment and are still far from being able to assist physicians at the bedside in the clinical decisions to improve quality and efficiency of care. The present review aimed to provide an overview of the status of ML-based algorithms in intensive care, to explore the concept of digital transformation, and to highlight possible next steps necessary to move towards a routine use of ML-based clinical decision support systems at the bedside. Finally, we described our attempt to apply the pillars of digital transformation in the field of microcirculation monitoring with the creation of the Microcirculation Network Research Group (MNRG).

Microcirculatory tissue perfusion during general anaesthesia and noncardiac surgery: An observational study using incident dark field imaging with automated video analysis.

BACKGROUND: Handheld vital microscopy allows direct observation of red blood cells within the sublingual microcirculation. Automated analysis allows quantifying microcirculatory tissue perfusion variables - including tissue red blood cell perfusion (tRBCp), a functional variable integrating microcirculatory convection and diffusion capacities. OBJECTIVE: We aimed to describe baseline microcirculatory tissue perfusion in patients presenting for elective noncardiac surgery and test that microcirculatory tissue perfusion is preserved during elective general anaesthesia for noncardiac surgery. DESIGN: Prospective observational study. SETTING: University Medical Center Hamburg-Eppendorf, Hamburg, Germany. PATIENTS: 120 elective noncardiac surgery patients (major abdominal, orthopaedic or trauma and minor urologic surgery) and 40 young healthy volunteers. MAIN OUTCOME MEASURES: We measured sublingual microcirculation using incident dark field imaging with automated analysis at baseline before induction of general anaesthesia, under general anaesthesia before surgical incision and every 30 min during surgery. We used incident the dark field imaging technology with a validated automated analysis software. RESULTS: A total of 3687 microcirculation video sequences were analysed. Microcirculatory tissue perfusion variables varied substantially between individuals - but ranges were similar between patients and volunteers. Under general anaesthesia before surgical incision, there were no important changes in tRBCp, functional capillary density and capillary haematocrit compared with preinduction baseline. However, total vessel density was higher and red blood cell velocity and the proportion of perfused vessels were lower under general anaesthesia. There were no important changes in any microcirculatory tissue perfusion variables during surgery. CONCLUSION: In patients presenting for elective noncardiac surgery, baseline microcirculatory tissue perfusion variables vary substantially between individuals - but ranges are similar to those in young healthy volunteers. Microcirculatory tissue perfusion is preserved during general anaesthesia and noncardiac surgery - when macrocirculatory haemodynamics are maintained.