ABSTRACT
Unlabelled: Large language model (LLM)-powered services are gaining popularity in various applications due to their exceptional performance in many tasks, such as sentiment analysis and answering questions. Recently, research has been exploring their potential use in digital health contexts, particularly in the mental health domain. However, implementing LLM-enhanced conversational artificial intelligence (CAI) presents significant ethical, technical, and clinical challenges. In this viewpoint paper, we discuss 2 challenges that affect the use of LLM-enhanced CAI for individuals with mental health issues, focusing on the use case of patients with depression: the tendency to humanize LLM-enhanced CAI and their lack of contextualized robustness. Our approach is interdisciplinary, relying on considerations from philosophy, psychology, and computer science. We argue that the humanization of LLM-enhanced CAI hinges on the reflection of what it means to simulate "human-like" features with LLMs and what role these systems should play in interactions with humans. Further, ensuring the contextualization of the robustness of LLMs requires considering the specificities of language production in individuals with depression, as well as its evolution over time. Finally, we provide a series of recommendations to foster the responsible design and deployment of LLM-enhanced CAI for the therapeutic support of individuals with depression.
Subject(s)
Artificial Intelligence , Depression , Humans , Depression/psychology , Depression/therapy , Language , Communication , HumanismABSTRACT
Large language models (LLMs) have now entered the realm of medical ethics. In a recent study, Balas et al examined the performance of GPT-4, a commercially available LLM, assessing its performance in generating responses to diverse medical ethics cases. Their findings reveal that GPT-4 demonstrates an ability to identify and articulate complex medical ethical issues, although its proficiency in encoding the depth of real-world ethical dilemmas remains an avenue for improvement. Investigating the integration of LLMs into medical ethics decision-making appears to be an interesting avenue of research. However, despite the promising trajectory of LLM technology in medicine, it is crucial to exercise caution and refrain from attributing their expertise to medical ethics. Our thesis follows an examination of the nature of expertise and the epistemic limitations that affect LLM technology. As a result, we propose two more fitting applications of LLMs in medical ethics: first, as tools for mining electronic health records or scientific literature, thereby supplementing evidence for resolving medical ethics cases, and second, as educational platforms to foster ethical reflection and critical thinking skills among students and residents. The integration of LLMs in medical ethics, while promising, requires careful consideration of their epistemic limitations. Consequently, a well-considered definition of their role in ethically sensitive decision-making is crucial.
Subject(s)
Ethics, Medical , Humans , Decision Making/ethics , LanguageABSTRACT
Reminiscence is the act of recalling or telling others about relevant personal past experiences. It is an important activity for all individuals, young and old alike. In fact, reminiscence can serve different functions that can support or be detrimental to one's well-being. Although previous studies have extensively investigated older adults' recalling of autobiographical memories, the evidence for young adults remains scarce. Therefore, in this work, we analyze young adults' production of reminiscence and their functions with a naturalistic observation method. Furthermore, we demonstrate that natural language processing and machine learning can automatically detect reminiscence and its negative functions in young adults' everyday conversations. We interpret machine learning model results using Shapley explanations. Our results indicate that young adults reminisce in everyday life mostly to connect with others through conversation, to compensate for a lack of stimulation or to recall difficult past experiences. Moreover, our models improve existing benchmarks from the literature on the automated detection of older adults' reminiscence in everyday life. Finally, our results may support the development of digital health intervention programs that detect reminiscence and its functions in young adults to support their well-being.
ABSTRACT
In the intensive care unit, it can be challenging to determine which interventions align with the patients' preferences since patients are often incapacitated and other sources, such as advance directives and surrogate input, are integral. Managing treatment decisions in this context requires a process of shared decision-making and a keen awareness of the preference-sensitive instances over the course of treatment. The present paper examines the need for the development of preference-sensitive decision timelines, and, taking aneurysmal subarachnoid hemorrhage as a use case, proposes a model of one such timeline to illustrate their potential form and value. First, the paper draws on an overview of relevant literature to demonstrate the need for better guidance to (a) aid clinicians in determining when to elicit patient preference, (b) support the drafting of advance directives, and (c) prepare surrogates for their role representing the will of an incapacitated patient in clinical decision-making. This first section emphasizes that highlighting when patient (or surrogate) input is necessary can contribute valuably to shared decision-making, especially in the context of intensive care, and can support advance care planning. As an illustration, the paper offers a model preference-sensitive decision timeline-whose generation was informed by existing guidelines and a series of interviews with patients, surrogates, and neuro-intensive care clinicians-for a use case of aneurysmal subarachnoid hemorrhage. In the last section, the paper offers reflections on how such timelines could be integrated into digital tools to aid shared decision-making.
ABSTRACT
BACKGROUND: Work stress places a heavy economic and disease burden on society. Recent technological advances include digital health interventions for helping employees prevent and manage their stress at work effectively. Although such digital solutions come with an array of ethical risks, especially if they involve biomedical big data, the incorporation of employees' values in their design and deployment has been widely overlooked. OBJECTIVE: To bridge this gap, we used the value sensitive design (VSD) framework to identify relevant values concerning a digital stress management intervention (dSMI) at the workplace, assess how users comprehend these values, and derive specific requirements for an ethics-informed design of dSMIs. VSD is a theoretically grounded framework that front-loads ethics by accounting for values throughout the design process of a technology. METHODS: We conducted a literature search to identify relevant values of dSMIs at the workplace. To understand how potential users comprehend these values and derive design requirements, we conducted a web-based study that contained closed and open questions with employees of a Swiss company, allowing both quantitative and qualitative analyses. RESULTS: The values health and well-being, privacy, autonomy, accountability, and identity were identified through our literature search. Statistical analysis of 170 responses from the web-based study revealed that the intention to use and perceived usefulness of a dSMI were moderate to high. Employees' moderate to high health and well-being concerns included worries that a dSMI would not be effective or would even amplify their stress levels. Privacy concerns were also rated on the higher end of the score range, whereas concerns regarding autonomy, accountability, and identity were rated lower. Moreover, a personalized dSMI with a monitoring system involving a machine learning-based analysis of data led to significantly higher privacy (P=.009) and accountability concerns (P=.04) than a dSMI without a monitoring system. In addition, integrability, user-friendliness, and digital independence emerged as novel values from the qualitative analysis of 85 text responses. CONCLUSIONS: Although most surveyed employees were willing to use a dSMI at the workplace, there were considerable health and well-being concerns with regard to effectiveness and problem perpetuation. For a minority of employees who value digital independence, a nondigital offer might be more suitable. In terms of the type of dSMI, privacy and accountability concerns must be particularly well addressed if a machine learning-based monitoring component is included. To help mitigate these concerns, we propose specific requirements to support the VSD of a dSMI at the workplace. The results of this work and our research protocol will inform future research on VSD-based interventions and further advance the integration of ethics in digital health.
Subject(s)
Occupational Stress , Workplace , Humans , Occupational Stress/prevention & control , Digital Technology , Machine Learning , Cell PhoneSubject(s)
Patient Preference , Personal Autonomy , Humans , Paternalism , Decision Making , Artificial IntelligenceABSTRACT
BACKGROUND AND OBJECTIVE: Work-related stress affects a large part of today's workforce and is known to have detrimental effects on physical and mental health. Continuous and unobtrusive stress detection may help prevent and reduce stress by providing personalised feedback and allowing for the development of just-in-time adaptive health interventions for stress management. Previous studies on stress detection in work environments have often struggled to adequately reflect real-world conditions in controlled laboratory experiments. To close this gap, in this paper, we present a machine learning methodology for stress detection based on multimodal data collected from unobtrusive sources in an experiment simulating a realistic group office environment (N=90). METHODS: We derive mouse, keyboard and heart rate variability features to detect three levels of perceived stress, valence and arousal with support vector machines, random forests and gradient boosting models using 10-fold cross-validation. We interpret the contributions of features to the model predictions with SHapley Additive exPlanations (SHAP) value plots. RESULTS: The gradient boosting models based on mouse and keyboard features obtained the highest average F1 scores of 0.625, 0.631 and 0.775 for the multiclass prediction of perceived stress, arousal and valence, respectively. Our results indicate that the combination of mouse and keyboard features may be better suited to detect stress in office environments than heart rate variability, despite physiological signal-based stress detection being more established in theory and research. The analysis of SHAP value plots shows that specific mouse movement and typing behaviours may characterise different levels of stress. CONCLUSIONS: Our study fills different methodological gaps in the research on the automated detection of stress in office environments, such as approximating real-life conditions in a laboratory and combining physiological and behavioural data sources. Implications for field studies on personalised, interpretable ML-based systems for the real-time detection of stress in real office environments are also discussed.
Subject(s)
Machine Learning , Mental Health , Heart Rate , Movement , Random ForestABSTRACT
Developing spinal circuits generate patterned motor outputs while many neurons with high membrane resistances are still maturing. In the spinal cord of hatchling frog tadpoles of unknown sex, we found that the firing reliability in swimming of inhibitory interneurons with commissural and ipsilateral ascending axons was negatively correlated with their cellular membrane resistance. Further analyses showed that neurons with higher resistances had outward rectifying properties, low firing thresholds, and little delay in firing evoked by current injections. Input synaptic currents these neurons received during swimming, either compound, unitary current amplitudes, or unitary synaptic current numbers, were scaled with their membrane resistances, but their own synaptic outputs were correlated with membrane resistances of their postsynaptic partners. Analyses of neuronal dendritic and axonal lengths and their activities in swimming and cellular input resistances did not reveal a clear correlation pattern. Incorporating these electrical and synaptic properties into a computer swimming model produced robust swimming rhythms, whereas randomizing input synaptic strengths led to the breakdown of swimming rhythms, coupled with less synchronized spiking in the inhibitory interneurons. We conclude that the recruitment of these developing interneurons in swimming can be predicted by cellular input resistances, but the order is opposite to the motor-strength-based recruitment scheme depicted by Henneman's size principle. This form of recruitment/integration order in development before the emergence of refined motor control is progressive potentially with neuronal acquisition of mature electrical and synaptic properties, among which the scaling of input synaptic strengths with cellular input resistance plays a critical role.SIGNIFICANCE STATEMENT The mechanisms on how interneurons are recruited to participate in circuit function in developing neuronal systems are rarely investigated. In 2-d-old frog tadpole spinal cord, we found the recruitment of inhibitory interneurons in swimming is inversely correlated with cellular input resistances, opposite to the motor-strength-based recruitment order depicted by Henneman's size principle. Further analyses showed the amplitude of synaptic inputs that neurons received during swimming was inversely correlated with cellular input resistances. Randomizing/reversing the relation between input synaptic strengths and membrane resistances in modeling broke down swimming rhythms. Therefore, the recruitment or integration of these interneurons is conditional on the acquisition of several electrical and synaptic properties including the scaling of input synaptic strengths with cellular input resistances.
Subject(s)
Interneurons , Swimming , Animals , Swimming/physiology , Xenopus laevis/physiology , Larva/physiology , Reproducibility of Results , Interneurons/physiology , Spinal Cord/physiologyABSTRACT
Relying on shared tasks and stimuli to conduct research can enhance the replicability of findings and allow a community of researchers to collect large data sets across multiple experiments. This approach is particularly relevant for experiments in spatial navigation, which often require the development of unfamiliar large-scale virtual environments to test participants. One challenge with shared platforms is that undetected technical errors, rather than being restricted to individual studies, become pervasive across many studies. Here, we discuss the discovery of a software bug in a virtual environment platform used to investigate individual differences in spatial navigation: Virtual Silcton. The bug, which was difficult to detect for several reasons, resulted in storing the absolute value of a direction in a pointing task rather than the signed direction and rendered the original sign of the direction unrecoverable. To assess the impact of the bug on published findings, we collected a new data set for comparison. Results revealed that although the bug caused suppression in pointing errors and had different effects across people (less accurate navigators had more suppression), the effect of the bug on published data is small, partially explaining the difficulty in detecting the bug. We also used the new data set to develop a tool that allows researchers who have previously used Virtual Silcton to evaluate the impact of the bug on their findings. We summarize the ways that shared open materials, shared data, and collaboration can pave the way for better science to prevent errors in the future. (PsycInfo Database Record (c) 2023 APA, all rights reserved).
Subject(s)
Spatial Navigation , Humans , Spatial Behavior , IndividualityABSTRACT
Artificial intelligence (AI) systems are quickly gaining ground in healthcare and clinical decision-making. However, it is still unclear in what way AI can or should support decision-making that is based on incapacitated patients' values and goals of care, which often requires input from clinicians and loved ones. Although the use of algorithms to predict patients' most likely preferred treatment has been discussed in the medical ethics literature, no example has been realised in clinical practice. This is due, arguably, to the lack of a structured approach to the epistemological, ethical and pragmatic challenges arising from the design and use of such algorithms. The present paper offers a new perspective on the problem by suggesting that preference predicting AIs be viewed as sociotechnical systems with distinctive life-cycles. We explore how both known and novel challenges map onto the different stages of development, highlighting interdisciplinary strategies for their resolution.
Subject(s)
Artificial Intelligence , Goals , Humans , Patient Preference , Ethics, Medical , Advance DirectivesABSTRACT
Despite the presumed value of advance directives, research to demonstrate impact has shown mixed results. For advance directives to serve their role promoting patient autonomy, it is important that patients be informed decision makers. The capacity to make decisions depends upon understanding, appreciation, reasoning, and communication. Advance directives are in part faulty because these elements are often limited. The present paper explores how the application of digital technology could be organized around a framework promoting these four elements. Given the state of digital advancements, there is great potential for advance directives to be meaningfully enhanced. The beneficial effects of incorporating digital technology would be maximized if they were organized around the aim of making advance directives not only documents for declaring preferences but also ethics-driven tools with decision aid functionality. Such advance directives would aid users in making decisions that involve complex factors with potentially far-reaching impact and would also elucidate the users' thought processes to aid those tasked with interpreting and implementing decisions based on an advance directive. Such advance directives might have embedded interactive features for learning; access to content that furthers one's ability to project oneself into possible, future scenarios; review of the logical consistency of stated preferences; and modes for effective electronic sharing. Important considerations include mitigating the introduction of bias depending on the presentation of information; optimizing interfacing with surrogate decision makers and treating clinicians; and prioritizing essential components to respect time constraints.
Subject(s)
Decision Making , Digital Technology , Advance Directives , Communication , Delivery of Health Care , HumansABSTRACT
Undetectable measurable residual disease (uMRD) is achievable in patients with chronic lymphocytic leukemia (CLL) with the BCL2-inhibitor venetoclax alone or combined with the Bruton's tyrosine kinase inhibitor ibrutinib. This phase 2, multicenter, MRD-driven study was designed to discontinue treatment upon reaching uMRD4 (<10-4) in patients with relapsed/refractory CLL receiving venetoclax monotherapy or after the addition of ibrutinib. Primary end point of the study was proportion of uMRD4 with venetoclax ± ibrutinib. Secondary end points were overall response rate, partial response, complete response, progression-free survival, duration of response, overall survival, and safety of venetoclax ± ibrutinib. Patients with uMRD4 at Cycle 12 Day 1 discontinued venetoclax. MRD+ patients added ibrutinib and continued both drugs up to Cycle 24 Day 28/uMRD4/progression/toxicity. After Cycle 24 Day 28, MRD+ patients continued ibrutinib. Thirty-eight patients (29% with TP53 aberrations; 79% with unmutated IGHV) started venetoclax. Overall response rate with venetoclax was 36 (95%) of 38 patients (20 complete; 16 partial response). Seventeen patients (45%) with uMRD4 at Cycle 12 Day 1 discontinued venetoclax. Nineteen (55%) MRD+ subjects added ibrutinib. After a median of 7 months (range, 3-10 months) of combined treatment, 16 (84%) of 19 achieved uMRD4, thus stopping both drugs. Two MRD+ patients at Cycle 24 Day 28 continued ibrutinib until progression/toxicity. After a median follow-up of 36.5 months, median progression-free survival was not reached; 10 patients progressed (4 restarted venetoclax, 3 without treatment need, 2 developed Richter transformation, and 1 dropped out). Seven (22%) of 32 patients remain uMRD4 after 3 years of follow-up. Neutropenia was the most frequent grade 3 to 4 adverse event; no grade 5 events occurred on study. This sequential MRD-guided approach led to uMRD4 in 33 (87%) of 38 patients, with venetoclax monotherapy or combined with ibrutinib, delivering treatment combination only in a fraction, and ultimately identifying the few patients benefiting from continuous therapy. This trial was registered at www.clinicaltrials.gov as # NCT04754035.
Subject(s)
Leukemia, Lymphocytic, Chronic, B-Cell , Humans , Neoplasm, Residual/drug therapy , Pyrimidines/therapeutic use , Pyrazoles/adverse effects , Antineoplastic Combined Chemotherapy Protocols/adverse effects , Bridged Bicyclo Compounds, HeterocyclicABSTRACT
BACKGROUND: Language use and social interactions have demonstrated a close relationship with cognitive measures. It is important to improve the understanding of language use and behavioral indicators from social context to study the early prediction of cognitive decline among healthy populations of older adults. OBJECTIVE: This study aimed at predicting an important cognitive ability, working memory, of 98 healthy older adults participating in a 4-day-long naturalistic observation study. We used linguistic measures, part-of-speech (POS) tags, and social context information extracted from 7450 real-life audio recordings of their everyday conversations. METHODS: The methods in this study comprise (1) the generation of linguistic measures, representing idea density, vocabulary richness, and grammatical complexity, as well as POS tags with natural language processing (NLP) from the transcripts of real-life conversations and (2) the training of machine learning models to predict working memory using linguistic measures, POS tags, and social context information. We measured working memory using (1) the Keep Track test, (2) the Consonant Updating test, and (3) a composite score based on the Keep Track and Consonant Updating tests. We trained machine learning models using random forest, extreme gradient boosting, and light gradient boosting machine algorithms, implementing repeated cross-validation with different numbers of folds and repeats and recursive feature elimination to avoid overfitting. RESULTS: For all three prediction routines, models comprising linguistic measures, POS tags, and social context information improved the baseline performance on the validation folds. The best model for the Keep Track prediction routine comprised linguistic measures, POS tags, and social context variables. The best models for prediction of the Consonant Updating score and the composite working memory score comprised POS tags only. CONCLUSIONS: The results suggest that machine learning and NLP may support the prediction of working memory using, in particular, linguistic measures and social context information extracted from the everyday conversations of healthy older adults. Our findings may support the design of an early warning system to be used in longitudinal studies that collects cognitive ability scores and records real-life conversations unobtrusively. This system may support the timely detection of early cognitive decline. In particular, the use of a privacy-sensitive passive monitoring technology would allow for the design of a program of interventions to enable strategies and treatments to decrease or avoid early cognitive decline.
ABSTRACT
Ambiguous sensory information can lead to spontaneous alternations between perceptual states, recently shown to extend to tactile perception. The authors recently proposed a simplified form of tactile rivalry which evokes two competing percepts for a fixed difference in input amplitudes across antiphase, pulsatile stimulation of the left and right fingers. This study addresses the need for a tactile rivalry model that captures the dynamics of perceptual alternations and that incorporates the structure of the somatosensory system. The model features hierarchical processing with two stages. The first and the second stages of model could be located at the secondary somatosensory cortex (area S2), or in higher areas driven by S2. The model captures dynamical features specific to the tactile rivalry percepts and produces general characteristics of perceptual rivalry: input strength dependence of dominance times (Levelt's proposition II), short-tailed skewness of dominance time distributions and the ratio of distribution moments. The presented modelling work leads to experimentally testable predictions. The same hierarchical model could generalise to account for percept formation, competition and alternations for bistable stimuli that involve pulsatile inputs from the visual and auditory domains.
Subject(s)
Vision, Binocular , Visual Perception , Visual Perception/physiology , Vision, Binocular/physiology , Models, Neurological , Photic StimulationABSTRACT
In their article 'Who is afraid of black box algorithms? On the epistemological and ethical basis of trust in medical AI', Durán and Jongsma discuss the epistemic and ethical challenges raised by black box algorithms in medical practice. The opacity of black box algorithms is an obstacle to the trustworthiness of their outcomes. Moreover, the use of opaque algorithms is not normatively justified in medical practice. The authors introduce a formalism, called computational reliabilism, which allows generating justified beliefs on the algorithm reliability and trustworthy outcomes of artificial intelligence (AI) systems by means of epistemic warrants, called reliability indicators. However, they remark the need for reliability indicators specific to black box algorithms and that justified knowledge is not sufficient to justify normatively the actions of the physicians using medical AI systems. Therefore, Durán and Jongsma advocate for a more transparent design and implementation of black box algorithms, providing a series of recommendations to mitigate the epistemic and ethical challenges behind their use in medical practice. In this response, I argue that a peculiar form of black box algorithm transparency, called design publicity, may efficiently implement these recommendations. Design publicity encodes epistemic, that is, reliability indicators, and ethical recommendations for black box algorithms by means of four subtypes of transparency. These target the values and goals, their translation into design requirements, the performance and consistency of the algorithm altogether. I discuss design publicity applying it to a use case focused on the automated classification of skin lesions from medical images.
Subject(s)
Algorithms , Artificial Intelligence , Humans , Knowledge , Morals , Reproducibility of ResultsABSTRACT
Artificial intelligence (AI) systems are increasingly being used in healthcare, thanks to the high level of performance that these systems have proven to deliver. So far, clinical applications have focused on diagnosis and on prediction of outcomes. It is less clear in what way AI can or should support complex clinical decisions that crucially depend on patient preferences. In this paper, we focus on the ethical questions arising from the design, development and deployment of AI systems to support decision-making around cardiopulmonary resuscitation and the determination of a patient's Do Not Attempt to Resuscitate status (also known as code status). The COVID-19 pandemic has made us keenly aware of the difficulties physicians encounter when they have to act quickly in stressful situations without knowing what their patient would have wanted. We discuss the results of an interview study conducted with healthcare professionals in a university hospital aimed at understanding the status quo of resuscitation decision processes while exploring a potential role for AI systems in decision-making around code status. Our data suggest that (1) current practices are fraught with challenges such as insufficient knowledge regarding patient preferences, time pressure and personal bias guiding care considerations and (2) there is considerable openness among clinicians to consider the use of AI-based decision support. We suggest a model for how AI can contribute to improve decision-making around resuscitation and propose a set of ethically relevant preconditions-conceptual, methodological and procedural-that need to be considered in further development and implementation efforts.