JMIR Publications

ABSTRACT

Background:

The integration of artificial intelligence (AI)-powered conversational agents into healthcare has steadily progressed towards real-world deployment, despite limited patient-level evidence regarding acceptability in low-resource settings.

Objective:

This study evaluates patients’ perceptions, acceptance, and concerns regarding the use of an AI-powered Clinical Intelligence Companion-Hami® in outpatient clinics of a resource-constrained setting.

Methods:

A cross-sectional study was conducted at four hospitals in Karachi, Pakistan. A structured survey comprising demographic information and nine questions to assess perceptions was administered to 8,487 patients after they interacted with Hami®. Frequencies and percentages were reported for all responses, and binary logistic regression was performed to assess factors associated with patients’ acceptance and concerns. Adjusted Odds Ratios (aORs) were reported with 95% Confidence Intervals (CIs), and p-values <.05 were considered significant.

Results:

Patients were highly receptive to Hami®’s integration into healthcare settings (79·9%). Patients attending private hospitals (aOR 1·32; 95% CI: 1·14-1·53), aged 18-44 years (aOR 1·52; 95% CI: 1·29-1·78), and 45-64 years (aOR 1·32; 95% CI: 1·11-1·56) were significantly more likely to prefer Hami®’s integration into clinical care. Patients attending private hospitals also had higher odds of concerns regarding concern of medical errors (aOR 1·29; 95% CI: 1·06 – 1·56), confidentiality breaches (aOR 2·11; 95% CI: 1·57 – 2·82), reduced contact with providers (aOR 1·62; 95% CI: 1·30 – 2·01), decreased human aspects of care (aOR 2·07; 95% CI: 1·65 – 2·61) and unclear accountability (aOR 2·05; 95% CI: 1·58 – 2·65) as compared to patients attending public hospital (P <.05). Educated patients had two times higher odds of concerns regarding reduced contact with providers (aOR 1·91; 95% CI: 1·64 – 2·21) and decreased human aspects of care (aOR 2·10; 95% CI: 1·81 – 2·43) using Hami® in contrast to uneducated patients (P <.05).

Conclusions:

The integration of Hami® in a resource-constrained setting reveals differential readiness for AI-assisted care, with higher acceptance among younger and middle-aged patients and those visiting private hospitals and higher concerns expressed by educated patients and those visiting private hospitals. These findings underscore that patients’ sociodemographics mediate digital health acceptance and that implementation strategies should target educated and private-sector populations, as they are likely more aware of the risks associated with AI integration. By prioritizing transparent communication regarding data privacy and confidentiality, ensuring clinician oversight, and positioning AI as a supportive tool that preserves the essential human interaction, concerns can be mitigated.