LUMANA: A Scenario-Based Field Validation of a Culturally Adapted AI System for Mental Health Screening in Sokoto State, Nigeria
- 1eHealth Africa
Abstract
BackgroundMental health conditions represent a significant public health burden in sub-Saharan Africa, where access to specialist services remains limited. Sokoto State, Nigeria, has an estimated population exceeding 5.4 million with fewer than five psychiatrists, leaving frontline health workers to provide much of the mental health care. LUMANA is a human-centred artificial intelligence (AI) system developed to support mental health screening in resource-constrained settings.
MethodsWe conducted a field validation of LUMANA in Sokoto State, Nigeria, on 28-29 January 2026 following ethical approval (SKHREC/014/2026). Twenty stakeholders were enrolled, of whom 17 completed structured evaluations across eight standardized clinical scenarios using five assessment tools. The evaluation examined AI output safety, Hausa language performance, human-AI workflow integration, trust and cultural appropriateness, and implementation readiness.
ResultsA total of 136 AI output safety assessments were completed. Overall, 113/136 (83%) AI outputs were rated acceptable as shown, while 23/136 (17%) were considered acceptable only with human review; no AI output was rated unacceptable. Religion and culture were identified as important determinants of trust by 14/17 (82%) participants, while 17/17 (100%) agreed that disclosures of self-harm or suicide should always involve a human clinician. Mean language quality scores were 9.9/10 for Hausa transcription, 9.9/10 for Hausa-to-English translation, and 9.2/10 for English-to-Hausa summary generation. Clinicians considered 7/8 (88%) human-AI workflow scenarios comfortable for routine practice. Three operational improvements were identified before pilot implementation: clearer protocols for interpreting distress, refinement of language to minimise potentially stigmatizing expressions, and automatic escalation of suspected psychotic symptoms.
ConclusionsThis scenario-based field validation provides preliminary evidence that LUMANA is perceived as culturally appropriate, acceptable and capable of supporting human-supervised mental health screening in a low-resource setting. However, the findings are based on a small convenience sample using standardized scenarios rather than live clinical use. Accordingly, the results support progression to a carefully monitored pilot implementation following completion of the identified system improvements, with prospective evaluation of safety, workflow integration and implementation outcomes before consideration of wider deployment.
Related
- Culturally adapted home-based intervention to improve child’s development in Arab toddlers in Israel: study protocol for a randomized controlled trial
- Development and expert evaluation of a culturally adapted large language model for Malay clinical history-taking simulation: mixed methods study
- Effectiveness of Culturally Adapted Group CBT for ADHD Among Indian University Students: A Large-Scale Comparative Study from Greater Noida
- Culturally Adapted 26-Session Group CBT for Adult ADHD in Indian Universities: A Large-Scale Quasi-Experimental Study from Greater Noida
- Integrating large language model-generated culturally adapted verbal suggestions into music therapy for university students: A pilot randomized exploratory study
BibTeX
@misc{buhari2026lumana,
title = {{LUMANA: A Scenario-Based Field Validation of a Culturally Adapted AI System for Mental Health Screening in Sokoto State, Nigeria}},
author = {T. Buhari and A. Baguda and M. Aiyenuro and J. I. Ahmad and J. Galadanchi and A. Oluyori and R. M. Abdullahi and Z. Iliyasu and O. Akhigbe and A. Fawaz and R. Nkem and A. Ayuba and H. M. Salihu and A. A. Adam and N. A. Abdulmalik},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.03.26362181},
url = {https://www.medrxiv.org/content/10.64898/2026.09.03.26362181v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - LUMANA: A Scenario-Based Field Validation of a Culturally Adapted AI System for Mental Health Screening in Sokoto State, Nigeria
AU - T. Buhari
AU - A. Baguda
AU - M. Aiyenuro
AU - J. I. Ahmad
AU - J. Galadanchi
AU - A. Oluyori
AU - R. M. Abdullahi
AU - Z. Iliyasu
AU - O. Akhigbe
AU - A. Fawaz
AU - R. Nkem
AU - A. Ayuba
AU - H. M. Salihu
AU - A. A. Adam
AU - N. A. Abdulmalik
PY - 2026
DA - 2026/09/03
PB - medRxiv
DO - 10.64898/2026.09.03.26362181
UR - https://www.medrxiv.org/content/10.64898/2026.09.03.26362181v1
AB - BackgroundMental health conditions represent a significant public health burden in sub-Saharan Africa, where access to specialist services remains limited. Sokoto State, Nigeria, has an estimated population exceeding 5.4 million with fewer than five psychiatrists, leaving frontline health workers to provide much of the mental health care. LUMANA is a human-centred artificial intelligence (AI) system developed to support mental health screening in resource-constrained settings. MethodsWe conducted a field validation of LUMANA in Sokoto State, Nigeria, on 28-29 January 2026 following ethical approval (SKHREC/014/2026). Twenty stakeholders were enrolled, of whom 17 completed structured evaluations across eight standardized clinical scenarios using five assessment tools. The evaluation examined AI output safety, Hausa language performance, human-AI workflow integration, trust and cultural appropriateness, and implementation readiness. ResultsA total of 136 AI output safety assessments were completed. Overall, 113/136 (83%) AI outputs were rated acceptable as shown, while 23/136 (17%) were considered acceptable only with human review; no AI output was rated unacceptable. Religion and culture were identified as important determinants of trust by 14/17 (82%) participants, while 17/17 (100%) agreed that disclosures of self-harm or suicide should always involve a human clinician. Mean language quality scores were 9.9/10 for Hausa transcription, 9.9/10 for Hausa-to-English translation, and 9.2/10 for English-to-Hausa summary generation. Clinicians considered 7/8 (88%) human-AI workflow scenarios comfortable for routine practice. Three operational improvements were identified before pilot implementation: clearer protocols for interpreting distress, refinement of language to minimise potentially stigmatizing expressions, and automatic escalation of suspected psychotic symptoms. ConclusionsThis scenario-based field validation provides preliminary evidence that LUMANA is perceived as culturally appropriate, acceptable and capable of supporting human-supervised mental health screening in a low-resource setting. However, the findings are based on a small convenience sample using standardized scenarios rather than live clinical use. Accordingly, the results support progression to a carefully monitored pilot implementation following completion of the identified system improvements, with prospective evaluation of safety, workflow integration and implementation outcomes before consideration of wider deployment.
ER -