Table 1 Research contexts of studies that evaluated Artificial Intelligence implementations in low- and middle-income countries.
Study | Study focus | Study design | Study outcome | Study methodology | Analytical approach | Study sample |
|---|---|---|---|---|---|---|
Love et al.17 | Training of non-radiologist healthcare workers in a Mexican hospital to use AI CADx system to triage palpable breast lumps for further examination by oncologists | Cross-sectional study | The concordance between AI’s scoring of breast lumps, when used by non-radiologist health workers, and a radiologist’s BI-RADS of the breast lumps | Quantitative | Predictive analysis | 32 palpable breast lumps, examined by three non-radiologist health workers using AI tool |
Zhou et al.20 | The performance of IBM Watson for Oncology when producing treatment plans for cervical cancer patients in a Chinese university hospital | Cross-sectional study | Concordance between AI recommended and ‘for consideration’ treatments, and treatments implemented by physicians | Quantitative | Inferential analysis | 362 retrospective cancer patients; single case-study patient |
Kisling et al.21 | The ability of MD Anderson’s ‘Radiation Planning Assistant’ to automate the production of safe and effective radiotherapy treatment plans for cervical cancer patients in two South African hospitals | Cross-sectional study | The acceptability and accuracy of AI generated treatment plans by specialists in gynaecologic radiation oncology, in addition to AI run-time for generating plans | Quantitative | Descriptive analysis | 14 cervical cancer patients |
Ugarte-Gil et al.15 | The utility of the ‘eRx’ CADx system for TB diagnostics in primary care clinics in Peru, and the challenges of implementing such a system from the user’s perspective | Cross-sectional study | The experiences of nurses and doctors with the technology, including barriers and complications that arose | Mixed methods | Descriptive and content analysis of reported provider experiences | Seven nurses and five doctors working at primary care hospital or health centres |
Garzon-Chavez et al.18 | The implementation of an AI-assisted CT screening tool for COVID-19 patient triage in the workflows of an Ecuadorian hospital | Cross-sectional study | The sensitivity and specificity of AI-assisted CT screening to correctly identify likely COVID-19 positives as confirmed by RT-PCR test | Quantitative | Predictive analysis | 75 chest CTs for patients with laboratory confirmed SARS-CoV-2 diagnosis |
Fan et al.23 | The real-world use of an AI health chatbot for primary care self-diagnosis in China, including issues and barriers in their usage, and user experiences | Cross-sectional study | The characteristics of users, length and frequency of chatbot sessions, health concerns presented, and user feedback | Mixed methods | Descriptive and inferential analysis of chatbot sessions; content analysis of user feedback | 47,684 consultation sessions initiated by 16,519 users |
Ganju et al.16 | The use of an Indian child health and nutrition education mHealth app’s usage data to predict user churn and target interventions to improve user engagement | Cross-sectional study | The engagement and retention of users with mHealth app | Quantitative | Predictive analysis of user churn | 45,000 mHealth app users |
Wang et al.22 | The experiences of physicians using a clinical decision support system for diagnostic assistance and treatment suggestions in rural primary care clinics in China | Cross-sectional study | The perspectives of physician using the AI tool, including the perceived challenges, limitations, trustworthiness, and usefulness of the tool | Qualitative | Content analysis | 22 clinicians from rural primary care clinics |
Wang et al.24 | The anonymous implementation of a social support chatbot for online pregnancy healthcare in a community in China | Cross-sectional study | The AI’s response rate and response time to community members, and emotional valence of responses, compared with responses from other community members | Mixed methods | Descriptive analysis, content analysis | 3445 users of YouBaoBao online pregnancy healthcare community |
MacPherson et al.19 | The improvement in patient outcomes and cost-effectiveness of implementing an X-ray CADx screening tool for TB in a primary clinic in Malawi, as part of an existing HIV-TB screening programme | Randomised controlled trial | The time in days, up to 56 days, to TB treatment initiation compared with standard-of-care and HIV treatment arm, and ICER of TB screening treatment arm | Quantitative | Economic analysis | 1462 resident adults attending health centre reporting TB symptoms with no history of TB |
