Clinical Strategies for Creating Safe Online Spaces to Protect Vulnerable Patients

Dear Editor,

The short reflection by Kannai and Rice, “Telemedicine Could Reduce the Role of Family Physicians to Case Managers” (1) is an apt reminder that technology re-shapes professional roles even when its immediate aim is to widen access. As a family physician researchers, we find ourselves welcoming digital tools while noticing how easily they displace the clinical habits that once anchored our discipline. This letter, the second of three (2–4) we intend to submit on the article, turns to themes the authors only touched on: the potential of artificial-intelligence (AI) support during video consultations, the place of group therapy delivered virtually, and the safeguards required for privacy and emotional safety in remote care. Our comments draw on evidence that has accumulated since the first pandemic wave and on daily experience inside the Spanish public system, where each autonomous community now pursues its own blend of incentives for digital innovation.

Artificial-intelligence systems already participate, sometimes silently, in video visits across Europe. Most remain research prototypes, yet their relevance to family medicine is clear when one examines the literature on speech and facial-gesture analytics for early neurocognitive disorders. Automated language models can discriminate between healthy ageing and mild cognitive impairment with accuracies that rival traditional screening tools, provided that recordings are long enough and background noise is controlled (5). Huang et al. recently showed that linguistic features alone, extracted from a short free-speech task, classified cognitive decline with an area-under-the-curve of 0.84, and that the classifier could operate on data captured by an ordinary laptop microphone (6,7). Similar results have been reported with smartphone-based speech analysis in multicentre datasets collated by industry–academic consortia such as Aural Analytics, again pointing to a future in which the physician may receive a silent probability score while listening to the patient’s narrative (8). We see opportunities for carefully designed pilots in the Basque Country and Catalonia, where electronic records have pioneered some embed machine-learning modules for prescription safety alerts in research-related studies. The goal should not be to replace memory tests, but to prompt an earlier face-to-face assessment when a conversational cue suggests deterioration that neither the clinician nor the patient had recognised.

Once AI enters the consultation room, even a virtual one, familiar questions about validity, bias, and interpretability follow. That conversation is no longer hypothetical: algorithms trained on video frames can flag micro-expressions that correlate with depressive states; language-model analysis of vocal prosody can estimate Parkinsonian dysarthria; and wearable sensor streams are being parsed to infer psychomotor slowing. A JMIR Mental Health study that combined lexical, acoustic, and webcam-derived facial features reported 83 % sensitivity for major depression in cancer survivors using a model that ran in real time during short coaching sessions (9). These advances justify exploration but also require boundaries. The World Health Organization’s 2021 guidance on AI ethics underscores the need for transparency, local validation, and explicit lines of accountability before algorithmic advice is allowed to influence clinical decisions (10). Spanish legislation already demands human oversight for automated decision-making in health records, yet regional procurement contracts still vary in how they specify audit trails. We suggest that any deployment begin as an opt-in decision support tool, with performance tracked prospectively against hard outcomes such as diagnostic delay and avoidable referrals (11). Only by pairing innovation with a rigorous post-market evaluation framework can we ensure that statistical gains translate into meaningful clinical benefit and do not widen disparities for patients whose dialect, accent, or camera quality differs from the data on which the model was trained.

Technology also creates room for therapeutic formats that were previously impractical. Group videoconference interventions have moved from being an emergency substitute during lockdown to a viable adjunct for primary-care mental-health work (12). A 2021 systematic meta-analysis that pooled thirty-one trials of videoconference-delivered cognitive-behavioural therapy (VCBT) found an effect size for depression that was non-inferior to face-to-face CBT (Hedges g = 0.45) and reported retention rates above 80% across modalities (13). Acceptance-and-commitment therapy and behavioural-activation protocols delivered synchronously to small groups produced similar symptom reductions in a randomised trial of adults with chronic pain and comorbid depression, with incremental cost–utility ratios that fell below common willingness-to-pay thresholds (14). These results matter for Spain’s publicly funded system, whose prevalence of affective disorders far outstrips the capacity of on-site psychology services. Andalucían and Galician primary-care districts, where community-based group CBT was already part of the mental-health offer, could expand reach through video while preserving one in-person intake session to establish rapport. The same hybrid logic applies to eating-disorder services: Couturier et al. described how family-based treatment for adolescent anorexia shifted online during the pandemic; families appreciated the convenience yet endorsed a mix of virtual and clinic visits to secure physical monitoring and weigh-ins (15). Integrating those lessons, our proposal is that regional planners couple any remote group programme with scheduled on-site reviews whose frequency depends on clinical stability, thereby guarding against the drift towards purely transactional care that Kannai and Rice warn against.

While group screens can democratise therapy, remote communication can also obscure signals of distress that patients feel unsafe to voice (16). Intimate-partner violence (IPV) is a salient example. Simon’s clinical update in JAMA (17) outlined practical steps for recognising IPV during telehealth encounters, explicitly recommending pre-visit electronic questionnaires and the use of coded yes/no questions if privacy seems compromised. A systematic review of ICT-based interventions for IPV (18) confirms that digital tools can facilitate disclosure and safety planning, but only when confidentiality is assured and the patient controls the timing and channel of contact. Building on these principles, several Spanish health centres have asked to modify their corporate patient portals with an unobtrusive “exit” button that hides the browser instantly and insert a silent alert icon within the video interface that the patient can click to request offline help. These design features should be complemented by professional training. Garber and colleagues’ telehealth-etiquette framework (19) emphasises simple but effective manoeuvres, confirming who else is in the room, lowering one’s voice when posing sensitive questions, and offering follow-up by a secure text channel for topics the patient could not address aloud. We recommend that such protocols become mandatory content in continuing-education credits tied to the performance-based incentives many autonomous communities now employ.

Confidentiality is not limited to safety from violence; it also guards the emotional candour on which therapeutic relationships depend. Clinicians notice that discussions of grief, sexuality, or existential doubt often pause when the patient senses the virtual space might be recorded or overheard. One solution, adopted in several Catalan primary-care teams, is to let the patient select the channel—telephone, video, or in-person—at each step of an episode of care and to make switching frictionless. Another is to establish “camera-off minutes,” during which either party can turn off video to restore a sense of privacy while still sharing voice. Early qualitative feedback suggests that these small concessions reduce screen fatigue and encourage spontaneous disclosure, though formal evaluation is pending. The point is consistent with Kannai and Rice’s insistence on maintaining relational depth: technology should expand the palette of communication, not restrict it to the default configuration chosen by the clinic’s scheduling software.

The introduction of AI, expanded group therapy, and new privacy practices will require a recalibration of professional training. Spanish medical schools and residency programs still devote limited curricular time to telehealth. The post-graduate family-medicine curriculum, updated in 2024, now lists “digital clinical competence” as a transversal skill, but implementation varies by teaching unit. Part of the highlighted ideas in this eLetter could be taught online, yet simulated patient sessions remain valuable for practising the non-verbal calibration that escapes algorithmic capture. By foregrounding these concrete skills, training programmes may help new physicians work “through” technology rather than merely “with” it, preserving the clinician’s role as an active decision-maker, even as the screen becomes the main space for care delivery.

In our opinion, Kannai and Rice remind us that making care easier to access is not the same as making it better. Artificial-intelligence augmentation, virtual group therapy, and enhanced confidentiality tools show promise, but they amplify, rather than solve, the ethical and professional questions raised in the original article. Their safe adoption hinges on maintaining the relational orientation that distinguishes family medicine. Our subsequent letter (4) will examine in more detail the organisational incentives that influence technology use and the research agenda needed to evaluate long-term outcomes. For now, we thank the authors for stimulating a dialogue that is as relevant in Spain as it is in Israel.

REFERENCES:

1. Kannai R, Rice (Alon) A. Telemedicine Could Reduce the Role of Family Physicians to Case Managers. Ann Fam Med. 2024 Jan 1;22(1):63–4.

2. Tenajas R, Miraut D. Learning to Listen Again Preserving Clinical Presence in a Digital World. Ann Fam Med. 2025 May 4;22(1):eLetter.

3. Tenajas R, Miraut D. Clinical Strategies for Creating Safe Online Spaces to Protect Vulnerable Patients. Ann Fam Med. 2025 May 4;22(1):eLetter.

4. Tenajas R, Miraut D. From Flat Screens to Deep Care, Designing Telemedicine for Human Connection. Ann Fam Med. 2025 May 4;22(1):eLetter.

5. Tenajas R, Miraut D. Enhancing Conversations on Cognitive Decline Through Patient-Centered Tools. Ann Fam Med. 2025 Apr 18;23(2):eLetter.

6. Huang L, Yang H, Che Y, Yang J. Automatic speech analysis for detecting cognitive decline of older adults. Front Public Health [Internet]. 2024 Aug 8 [cited 2025 May 4];12. Available from: https://www.frontiersin.orghttps://www.frontiersin.org/journals/public-h...

7. Tenajas R, Miraut D. The Role of Voice Analysis in Early Detection of Mental and Neurodegenerative Disorders. Ann Fam Med. 2025 Feb 28;23(1):eLetter.

8. Li R, Huang G, Wang X, Lawler K, Goldberg LR, Roccati E, et al. Smartphone automated motor and speech analysis for early detection of Alzheimer’s disease and Parkinson’s disease: Validation of TapTalk across 20 different devices. Alzheimers Dement Diagn Assess Dis Monit. 2024;16(4):e70025.

9. Smrke U, Mlakar I, Lin S, Musil B, Plohl N. Language, Speech, and Facial Expression Features for Artificial Intelligence–Based Detection of Cancer Survivors’ Depression: Scoping Meta-Review. JMIR Ment Health. 2021 Dec 6;8(12):e30439.

10. World Health Organization. Ethics and governance of artificial intelligence for health: large multi-modal models. WHO guidance [Internet]. World Health Organization; 2024. Available from: https://www.who.int/publications/i/item/9789240029200

11. Tenajas R, Miraut D. The 24 Big Challenges of Artificial Inteligence Adoption in Healthcare: Review Article. Acta Medica Ruha. 2023 Sep 20;1(3):432–67.

12. Tenajas R, Miraut D. The strength of connection: The virtual adaptation of Twelve-Step programs’ meetings. Front Health Inform. 2023;2023(12):1–3.

13. Matsumoto K, Hamatani S, Shimizu E. Effectiveness of Videoconference-Delivered Cognitive Behavioral Therapy for Adults With Psychiatric Disorders: Systematic and Meta-Analytic Review. J Med Internet Res. 2021 Dec 13;23(12):e31293.

14. Kysely A, Bishop B, Kane RT, McDevitt M, De Palma M, Rooney R. Couples Therapy Delivered Through Videoconferencing: Effects on Relationship Outcomes, Mental Health and the Therapeutic Alliance. Front Psychol [Internet]. 2022 Feb 4 [cited 2025 May 4];12. Available from: https://www.frontiersin.orghttps://www.frontiersin.org/journals/psycholo...

15. Couturier J, Pellegrini D, Grennan L, Nicula M, Miller C, Agar P, et al. A qualitative evaluation of team and family perceptions of family-based treatment delivered by videoconferencing (FBT-V) for adolescent Anorexia Nervosa during the COVID-19 pandemic. J Eat Disord. 2022 Jul 26;10(1):111.

16. Tenajas R, Miraut D. From Stigma to Support: Improving Alcohol Use Screening in Family Medicine. Ann Fam Med. 2025 Mar 7;23(1):eLetter.

17. Simon MA. Responding to Intimate Partner Violence During Telehealth Clinical Encounters. JAMA. 2021 Jun 8;325(22):2307–8.

18. El Morr C, Layal M. Effectiveness of ICT-based intimate partner violence interventions: a systematic review. BMC Public Health. 2020 Sep 7;20(1):1372.

19. Kelli Garber D, Tina Gustin D. Put PEP into telehealth: an etiquette framework for successful encounters. Online J Issues Nurs. 2023;28(2):1–10.