Ambient digital scribes (ADSs) are promoted to reduce clinical documentation burden, improve efficiency, and allow clinicians to focus on patients. This viewpoint examines the sociotechnical assumptions underlying these claims and argues that safe integration of ADSs requires consideration of how documentation work and clinician oversight are transformed within the wider clinical work system. Their appeal rests on 2 reassuring assumptions: that documentation work can be cleanly automated and that residual risks can be controlled because clinicians remain responsible for reviewing and correcting the output. This viewpoint argues that both assumptions can be challenged from a sociotechnical perspective. Automating the draft note does not eliminate documentation work. It transforms and redistributes it, moving clinicians from composing notes to evaluating, correcting, and editing AI-generated accounts, potentially transferring additional verification and reconciliation work to other people and later stages of care. Clinician oversight should not be treated as an automatic safety backstop. Plausible outputs may encourage cursory review, while omissions and subtle shifts in meaning can be difficult to detect. Time pressure, perceived system reliability, and organizational expectations for productivity may further weaken scrutiny. ADSs should not be designed and implemented through simple substitution of AI for documentation tasks followed by reliance on human checking. Safe integration requires sociotechnical design at the level of the wider work system, including deliberate choices about task design, organizational arrangements, and organizational learning. Participatory and context-sensitive sociotechnical design can inform explicitly what an ADS should and should not do, how the resulting documentation work should be distributed, and how impacts will be managed across the wider care system.
As the speed of adoption in health care increases, validation of AI medical devices lags behind—leaving uncertainty about whether these tools will ultimately benefit patients. In this News and Perspectives article, JMIR Correspondent Simon Spichak reports on this evidence gap and the complexities, frustrations, and risks around it.
Background Alcohol craving is associated with drinking behaviors despite negative consequences and relapse following periods of abstinence or treatment. Accurately forecasting alcohol craving is crucial for preventing alcohol consumption and relapse, thereby mitigating alcohol-related problems. Previous studies have assessed alcohol craving using diverse modalities such as psychological assessments, neurophysiological measures including electroencephalography (EEG) or heart rate variability (HRV) at a single time point, and digital phenotypes collected longitudinally via smartphone apps. However, alcohol craving is inherently heterogeneous and multidimensional across individuals and over time, making it difficult to capture using any single modality alone. In contrast, multimodal AI enables the prediction of alcohol craving by integrating complementary behavioral and neurophysiological features. Objective This study aimed to develop and evaluate AI models to forecast next-day alcohol craving in individuals with problematic alcohol use with multimodal datasets. Methods We analyzed multimodal datasets comprising smartphone-based digital phenotypes collected over 28 days, along with baseline EEG, HRV, and psychological measures in a prospective longitudinal study of 60 participants with problematic alcohol use. A tiny multimodal recurrent neural network (RNN) was implemented, in which representations of static baseline features were encoded into a latent representation and used to initialize the hidden state of the RNN. Model performance was evaluated using accuracy, weighted F 1 -score, macro F 1 -score, and mean absolute error (MAE). Results The multimodal model consistently outperformed the unimodal model based on digital phenotypes alone, achieving higher accuracy (0.56 vs 0.45), weighted F 1 -score (0.51 vs 0.38), and macro F 1 -score (0.37 vs 0.24), along with lower MAE (0.68 vs 0.90). At the individual level, 72% (43/60) of participants showed improved prediction accuracy with the multimodal model. These improvements were statistically significant (Wilcoxon signed-rank test, P <.001; paired t test, P <.001). Conclusions Integrating digital phenotypes with neurophysiological and psychological features enhances the prediction of short-term alcohol craving dynamics. This approach may enable more precise and timely personalized interventions for individuals at risk of relapse.
Mobile health apps can facilitate monitoring, communication, and administration in health care settings, but their accumulation may also create app overload for both patients and professionals. In this News and Perspectives article, JMIR Correspondent and health care digitalization researcher Vincent Peters reports on the increasing complexity of navigating digital health care.
Background Digital technologies are increasingly incorporated into long-term self-management of health conditions, yet their social impact on stroke survivors and carers remains limited. Objective This study explores how stroke survivors and carers view a prototype web-based application (“portal”) that allows patients to access and contribute up-to-date information about their health and care poststroke and how this might impact their self-management practices. Methods Through 27 semistructured interviews with 21 stroke survivors and 6 informal carers (performing unpaid care work), we explore the enablers of and barriers to the adoption and use of a portal for stroke survivors. To support discussions, participants were shown images of the portal prototype that displayed various features, including a self-living ability assessment, blood pressure monitoring, and an overview of exemplar personal data. Results Four themes were identified to articulate the anticipated effects of the portal: (1) portals as potential tools to empower through enabling survivors and carers to access stroke knowledge and enhance self-management; (2) portals as a method that could offer practical support for recovery management, therefore helping survivors manage their recovery; (3) hesitancy about using a portal, particularly regarding tracking recovery through personal data input; and (4) portals as a method to potentially share information to help keep family, friends, and wider support networks informed about their health and its management. Anticipated experiences also varied by social position. Young stroke survivors, for example, said they would value age-appropriate peer connections, while carers anticipated concerns about additional workload. Participants also felt that family members often lacked understanding of stroke impacts, suggesting that portals could help facilitate better communication about invisible and complex disabilities. Conclusions The anticipated benefits of digital tools for empowering patients need to be weighed against the complexity of people’s health and the additional responsibility digital interventions place on carers to communicate for those with severe disabilities. Our findings suggest that for a portal to be effective, it would need to address not only people’s medical needs but also their social needs, especially their use of informal care. Further research is needed, including with an implemented portal, to understand the use of portals over extended periods.
