Breathlessness, Diabete Mellitus, Fever, Hypertension
Conditions
Keywords
Artificial Intelligence, Delivery of Health Care, Health Personnel, Frontline Workers, Resource-Limited Settings
Brief summary
The goal of this clinical trial is to learn whether AI-enabled, nurse-led treatment planning can improve the quality of clinical reasoning and management compared with standard physician-led care in adult primary care patients (≥18 years) presenting with hypertension, diabetes mellitus, fever, breathlessness, or musculoskeletal pain in rural and semi-urban India. The main questions it aims to answer are: * Does a nurse + large language model (LLM) consultation achieve non-inferior clinical quality scores compared with a standard doctor consultation? * Is AI-assisted nurse-led care acceptable and satisfactory to patients in primary healthcare settings? Researchers will compare nurse + LLM-led consultations with physician-led standard-of-care consultations within the same participant to see if the AI-enabled nurse model delivers comparable or improved clinical reasoning and treatment planning. Participants will: * Receive two sequential consultations for the same visit (one with a nurse using an AI tool and one with a physician, order randomized). * Have both consultations audio recorded for blinded clinical quality assessment. * Complete a brief exit survey on communication, trust, and satisfaction after the AI-assisted nurse consultation.
Interventions
OTHERAI-enabled clinical decision support tool (software) used by nurses
A nurse-led primary care consultation supported by a large language model-based clinical decision support tool. The nurse uses the AI tool during the patient encounter to support clinical reasoning, differential diagnosis, and evidence-based treatment and follow-up planning.
Participants receive a routine physician-led primary care consultation conducted according to existing clinical practice. The physician independently performs history taking, clinical assessment, diagnosis, and treatment planning without use of the AI tool.
Sponsors
Sarah Nabia
Liver Foundation, West Bengal
Endless Health
Study design
Allocation
RANDOMIZED
Intervention model
CROSSOVER
Primary purpose
TREATMENT
Masking
SINGLE (Outcomes Assessor)
Intervention model description
This study uses a randomized, within-participant crossover interventional design. Each enrolled patient participates in two sequential clinical consultations during a single visit: (1) an AI-assisted, nurse-led consultation (intervention) and (2) a standard physician-led consultation (control). The order of consultations is randomized to minimize order effects. Both consultations address the same clinical condition or symptom and result in independent treatment plans. Because each participant serves as their own control, this design reduces between-subject variability and improves statistical efficiency. Clinical interactions are audio recorded and de-identified, and resulting treatment plans are independently scored by blinded physicians using standardized rubrics to assess clinical reasoning and management quality. In addition, patient experience is assessed via a post-consultation exit survey, and nurse experiences are explored through qualitative interviews.
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
1. Adults aged ≥18 years 2. Presenting to participating primary care facilities in study sites 3. Meeting criteria for at least one of the following conditions or symptoms: * Hypertension: Known diagnosis * Diabetes mellitus: Known diagnosis or laboratory evidence (HbA1c ≥6.5%, fasting blood glucose ≥126 mg/dL, or post-prandial glucose ≥200 mg/dL) * Fever: Presenting as chief complaint * Breathlessness: Presenting as chief complaint, without evidence of fever * Musculoskeletal pain: Presenting as chief complaint, without evidence of fever 4. Able and willing to provide written informed consent 5. Willing to participate in two sequential consultations and complete an exit survey
Exclusion criteria
1. Inability to provide informed consent due to cognitive impairment (e.g., dementia or intellectual disability) 2. Medical instability or condition requiring immediate emergency referral 3. Prior participation in the study during an earlier visit
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Clinical Quality of Consultation (Clinical Management and Clinical Reasoning Score) | Day 1 (same study visit, immediately after completion of both consultations) | Clinical quality of the consultation, scored by two blinded physician graders using a domain-based rubric (Annexure 1). Each domain is scored 0 (inadequate), 1 (suboptimal), or 2 (optimal). Disease (clinical management: hypertension, diabetes) cases are scored on four domains - quality of history, accuracy of next steps, safety, and comprehensiveness - for a total of 0-8. Symptom (clinical reasoning: fever, breathlessness, musculoskeletal pain) cases are scored on all six domains, adding quality of differential and accuracy of provisional diagnosis, for a total of 0-12. The primary outcome is the absolute total score; results are also reported normalised to 0-100% for concordance with the original registration. The two study arms (nurse+LLM vs. physician standard of care) are compared within each patient. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Patient Experience on Exit Survey | Day 1 (immediately after completion of the nurse + LLM consultation during the study visit) | Patient-reported experience of the nurse+LLM consultation, measured by a brief exit survey covering three domains: communication and understanding, trust and comfort with AI use, and respect and satisfaction (one item per domain; three-point response scale). Responses are summarised descriptively (frequencies/proportions per item and domain). |
| Nurse-Reported Acceptability and Feasibility Themes from Semi-Structured Interviews | Through study completion (after nurses complete a minimum of 10 AI-assisted consultations; up to 9 months) | Qualitative assessment of nurse-reported usability, trust in AI recommendations, workflow impact, barriers, facilitators, and willingness to continue use. Interviews are audio recorded and thematically analyzed. Outcomes will be reported as identified themes with representative quotations and frequency of theme occurrence across participants. |
Countries
India
Outcome results
None listed