Technology as an Add-On in Dizziness and Stroke Care: Reading the Evidence Together

The sources gathered here each test something added to, or substituted for, a clinician's usual work with people who have dizziness, stroke, or conditions managed in primary care and mental health services. They cover video assistance for a bedside eye examination ^[5], a large language model (LLM) offered to physicians during diagnosis ^[2], virtual reality during rehabilitation ^[1][3], and telephone or video consultations in place of face-to-face visits ^[4]. Their designs differ, and so do the certainty ratings their authors attach to the findings ^[1][2][3][5].

How strong each design is

The designs range from a randomized trial to a clinical practice guideline ^{[1][2][3][4][5]}:

• The vestibular rehabilitation guideline revises its 2016 version, drawing on a systematic review of literature published from 2015 through June 2020 across 6 databases, with 67 articles critically appraised; its recommendations are graded from strong evidence down to expert opinion ^[1].
• The LLM study is a single-blind randomized clinical trial of 50 physicians (26 attendings, 24 residents) trained in family, internal or emergency medicine, each given 60 minutes to review up to 6 clinical vignettes, analysed by intention to treat ^[2].
• The virtual reality review is a Cochrane systematic review of 190 randomised trials with 7188 adults after stroke; it used a fixed-effect meta-analysis for its critical outcome, upper limb function and activity, and rated certainty with the GRADE system ^[3].
• The teleconsultation review is a systematic review of primary studies comparing telephone or video consultations with face-to-face visits, following PRISMA guidelines; its abstract reports no pooled effect sizes ^[4].
• The HINTS review is a Cochrane diagnostic test accuracy review of 16 studies with 2024 participants, mean age 60 years, 13 of them performed in the emergency department and half by neurologists ^[5].

The number of trials in the virtual reality review did not translate into high certainty, a point the authors make themselves: only 36 (19%) of the 190 trials enrolled more than 50 participants, the largest recruited 152, risk of bias was often unclear because of poor reporting, and the evidence remained mostly low or moderate certainty ^[3]. The HINTS review rated its overall evidence as low certainty, with three studies at high and three at unclear risk of bias for participant selection ^[5]. The LLM trial is randomized and blinded, but it tested physicians on clinical vignettes rather than with patients ^[2].

Diagnosis: the tool and the clinician

Acute vestibular syndrome (AVS) is a type of dizziness that can include severe vertigo, nausea and vomiting, nystagmus (involuntary eye movements) or unsteadiness; its cause may be peripheral or central, and intervention and outcomes differ between the two ^[5]. The HINTS examination combines three bedside assessments: how the eyes move in response to a rapid head turn (head impulse), the direction of nystagmus, and whether the eyes are aligned (test of skew); HINTS Plus adds an assessment of auditory function ^[5]. The review assessed accuracy for identifying a central cause against advanced imaging or a neurologist's final diagnosis, calculating sensitivity and specificity from counts of true and false positives and negatives, and it assessed clinical (visual) and video-assisted examinations, such as video recording with goggles, separately ^[5].

The clinical HINTS examination (12 studies, 1890 participants) was 94.0% sensitive (95% CI 82.0% to 98.2%) and 86.9% specific (95% CI 75.3% to 93.6%), on low-certainty evidence ^[5]. Clinical HINTS Plus (5 studies, 451 participants) was 95.3% sensitive (95% CI 78.4% to 99.1%) and 72.9% specific (95% CI 44.4% to 90.1%), also low certainty ^[5]. Video-assisted HINTS (3 studies, 199 participants) ranged from 85.0% to 100% sensitive on low-certainty evidence and 38.9% to 100% specific on very low-certainty evidence; video-assisted HINTS Plus (2 studies, 163 participants) ranged from 85.0% to 93.8% sensitive and 28.6% to 38.9% specific, rated moderate certainty ^[5]. The authors state that data on the role of video assistance were limited ^[5].

In the LLM trial, the median diagnostic reasoning score per case was 76% with LLM access and 74% with conventional resources only, an adjusted difference of 2 percentage points (95% CI -4 to 8; P = .60) ^[2]. Median time per case was 519 seconds versus 565 seconds, a difference of -82 seconds (95% CI -195 to 31; P = .20) ^[2]. In a secondary exploratory analysis, the LLM alone scored 16 percentage points higher than the conventional resources group (95% CI 2 to 30; P = .03) ^[2].

Both diagnostic sources tie performance to the person using the method. The LLM trial authors conclude that the LLM alone outperformed both physician groups and that technology and workforce development are needed to realize physician-AI collaboration ^[2]. The HINTS review concludes that HINTS and HINTS Plus had good sensitivity and reasonable specificity for diagnosing a central cause of AVS in the emergency department when performed by trained clinicians, and it calls for research on inter-rater reliability and on accuracy across providers, specialties and experience ^[5].

Rehabilitation: virtual reality in peripheral and central conditions

The guideline addresses peripheral vestibular hypofunction, which in its included studies had to be determined by objective vestibular function tests; uncompensated vestibular hypofunction can cause dizziness, imbalance, oscillopsia, and gaze and gait instability ^[1]. On strong evidence, it recommends that clinicians offer vestibular rehabilitation to adults with unilateral or bilateral hypofunction who present with impairments, activity limitations and participation restrictions related to the vestibular deficit ^[1]. On moderate to strong evidence, it states that clinicians may offer specific exercise techniques, including virtual reality or augmented sensory feedback, to target identified activity limitations and participation restrictions, so virtual reality appears as one example within a 'may offer' recommendation ^[1].

The stroke review defines virtual reality as an advanced form of human-computer interface that allows the user to interact with a computer-generated environment in a naturalistic fashion, and it reports effects as standardised mean differences (SMD) with 95% CIs ^[3]. Against alternative therapy, where control groups usually received the same amount of a different therapy, virtual reality may slightly improve upper limb function and activity (SMD 0.20, 95% CI 0.12 to 0.28; 67 studies, 2830 participants; low certainty) and balance (SMD 0.26, 95% CI 0.12 to 0.40; 24 studies, 871 participants; low certainty) ^[3]. It probably reduces activity limitation (SMD 0.21, 95% CI 0.11 to 0.32; 33 studies, 1495 participants; moderate certainty) ^[3]. For gait speed the evidence was very uncertain (10 studies, 304 participants), and for participation and quality of life there may be little or no effect (SMD 0.11, 95% CI -0.02 to 0.24; 16 studies, 963 participants; low certainty) ^[3].

When virtual reality was added to usual care, which increased time spent in therapy for the intervention group, upper limb function and activity probably improved (SMD 0.42, 95% CI 0.26 to 0.58; 21 studies, 689 participants; moderate certainty), balance may have improved (SMD 0.68, 95% CI 0.46 to 0.91; 12 studies, 321 participants; low certainty), and activity limitation probably improved (SMD 0.22, 95% CI 0.04 to 0.41; 15 studies, 513 participants; moderate certainty) ^[3]. The authors note that greater benefits for upper limb function were seen in this add-on comparison, where virtual reality was used to increase overall therapy time ^[3].

The guideline also gives dose suggestions, all as 'may' statements ^[1]. On moderate to weak evidence, these include weekly clinic visits plus home gaze stabilization exercises totalling at least 12 minutes daily for acute or subacute unilateral hypofunction, at least 20 minutes daily for 4 to 6 weeks for chronic unilateral hypofunction, and 20 to 40 minutes daily for approximately 5 to 7 weeks for bilateral hypofunction ^[1]. Static and dynamic balance exercises may be prescribed for at least 20 minutes daily for at least 4 to 6 weeks in chronic unilateral hypofunction on moderate evidence, and for at least 6 to 9 weeks in bilateral hypofunction on expert opinion ^[1].

Across 59 stroke trials that monitored adverse events, few were reported and those reported were relatively mild ^[3]. The guideline identifies harm in a specific exercise choice: on strong evidence and a preponderance of harm over benefit, clinicians should not use voluntary saccadic or smooth-pursuit eye movements in isolation, without head movement, to promote gaze stability ^[1].

Supervision and distance

On strong evidence and in consideration of patient preference, the guideline recommends that clinicians offer supervised vestibular rehabilitation ^[1]. The teleconsultation review found telephone and video consultations as effective as face-to-face visits for improving clinical outcomes in adults with mental health conditions and those attending primary care, with high patient satisfaction and therapeutic alliance ^[4]. It also reported high discontinuation rates among patients receiving teleconsultations, indicating the modality may not suit all patients, and described training of healthcare professionals and management of technical issues as essential ^[4].

Where the sources agree

• When compared with an active alternative, the tested technologies showed small, non-significant or comparable effects: an SMD of 0.20 for virtual reality versus alternative therapy, a non-significant 2-point difference with LLM access, and teleconsultations as effective as, rather than better than, face-to-face care ^[2][3][4].
• Clinician skill and training recur as conditions for benefit, whether in performing HINTS, working with an LLM, or delivering remote care ^[2][4][5].
• Both rehabilitation sources attend to the amount of therapy: the stroke review reports greater upper limb benefit when virtual reality increased overall therapy time, and the guideline offers dose suggestions graded from moderate to weak evidence or expert opinion ^[1][3].

Where they conflict or do not connect

The guideline states that its recommendations may not apply to central vestibular disorders, the stroke review included only adults after stroke, and the HINTS review exists because peripheral and central causes of AVS call for different interventions ^[1][3][5]. Video also appears in two different roles: the teleconsultation review evaluated it as a mode of consultation, while the HINTS review evaluated video assistance as part of the examination itself, with limited data ^[4][5]. Certainty language differs too, with strong evidence cited for vestibular physical therapy and mostly low- to moderate-certainty evidence for virtual reality after stroke ^[1][3].

What remains unknown

• Whether HINTS changes clinical care: the review calls for direct comparison with no HINTS or with MRI, and notes limited data on video assistance ^[5].
• Whether virtual reality affects participation and quality of life after stroke, where evidence was insufficient, and gait speed, where it was mixed ^[3].
• Whether LLM access changes decisions with real patients, since the trial used clinical vignettes ^[2].
• Which patients are unsuited to teleconsultation, since the review flags high discontinuation without, in its abstract, identifying who discontinues ^[4].
• Whether supervised vestibular rehabilitation can be delivered remotely with the same benefit, a question neither abstract addresses directly ^[1][4].

Limitations

Each source is represented here by its published abstract, and full texts may qualify the figures or interpretations reported ^{[1][2][3][4][5]}. The sources concern different conditions and settings, so this cross-reading is interpretive rather than a pooled analysis ^{[1][2][3][4][5]}. Search and study dates differ: literature through June 2020 for the guideline, September 2022 for HINTS, September 2023 for virtual reality, and November to December 2023 for the LLM trial, so newer studies and newer models are not reflected ^[1][2][3][5]. The teleconsultation review was framed by the COVID-19 pandemic and reports no effect sizes in its abstract, so the size of any differences between consultation modes cannot be given here ^[4].