More than eight in ten healthcare professionals working in public facilities across Lagos, Nigeria, report having used artificial intelligence technologies, yet fewer than one in five has ever received formal AI training, according to a cross-sectional survey published in BMC Digital Health.
The headline figure is striking: 86.5 percent of respondents reported prior use of AI technologies, while only 19.8 percent had received formal training through courses, workshops or seminars. Just over half of the participants, 51.3 percent, said they used AI specifically to assist with their professional or clinical duties, and 50.6 percent used it in research activities. Among those using AI for research, the most common applications were AI-based educational platforms at 72.9 percent, AI writing assistance at 69.0 percent, AI-powered literature searches at 59.5 percent and AI-driven data analysis at 55.2 percent. Meanwhile, only 25.1 percent of respondents said their current healthcare facility itself utilised AI, a gap the authors describe as clinically and institutionally significant.
That discrepancy between individual use and institutional adoption suggests that most AI-related activity in these facilities occurs outside formal organisational frameworks. The researchers draw a parallel with what the information systems literature calls shadow IT, the use of technology outside sanctioned organisational channels. In a healthcare setting, they argue, professionals using unvalidated or informal AI tools for diagnostic reasoning or clinical documentation without institutional oversight may introduce risks of clinical error, algorithmic bias and violations of patient data privacy. The authors flag this pattern as one of the most actionable and policy-relevant findings of the study, demanding urgent attention from hospital administrators and regulators.
When it came to professional practice, AI-assisted diagnosis dominated. Among the AI technologies assessed, 81.7 percent of respondents reported frequent use, meaning always or often, of AI-assisted diagnosis, far ahead of any other application. By contrast, more than 70 percent of respondents said they rarely or never used AI-assisted drug prescriptions, AI medical imaging or AI-powered wearable devices. AI-powered search engines and writing assistants showed moderate adoption, with 20.7 percent and 8.4 percent of respondents using them frequently, respectively. The professional applications reported included clinical reasoning, interpretation of laboratory and radiological findings, and medication-interaction checks, alongside the use of generative tools such as large language model chatbots for documentation and patient-education materials.
Why professionals turn to these tools is relatively clear from the data. The leading motivators were convenience, cited by 55.7 percent of respondents, efficiency in completing tasks at 44.1 percent, and improved learning at 40.5 percent, followed by personal interest at 31.3 percent, internet access at 30.6 percent and curiosity at 26.7 percent. The barriers were equally telling: data privacy concerns topped the list at 51.8 percent, followed by distrust in AI at 34.0 percent, ethical concerns at 26.7 percent and difficulty using AI at 17.3 percent. Economic and technological constraints also featured, including internet data costs at 12.5 percent and a lack of smart internet-enabled devices at 4.3 percent. The authors note that privacy concerns may stem from weak institutional data-governance structures and limited awareness of existing ethical and regulatory protections, while the opaque black-box nature of many algorithms may deepen user distrust.
Education emerged as the one factor significantly associated with professional AI use. Educational level was linked to AI use in professional duties with a p-value of 0.042, with higher usage among those holding postgraduate degrees, 66.7 percent, than among those with bachelor’s degrees, 49.1 percent. In binary logistic regression, respondents with postgraduate qualifications were twice as likely to use AI in their professional duties as those with only basic healthcare degrees, with an odds ratio of 2.07, a 95 percent confidence interval of 1.16 to 3.71, and a p-value of 0.014. Facility type did not reach statistical significance, with a p-value of 0.055, although a directional trend showed higher use in secondary facilities, 59.7 percent, than in primary, 48.9 percent, or tertiary, 45.9 percent, facilities. In the regression model, secondary-facility professionals were approximately 1.5 times more likely to use AI than those at primary facilities, with an odds ratio of 1.55 and a confidence interval of 0.99 to 2.43, which was not statistically significant.
The overall regression model was statistically significant, with a likelihood ratio chi-square of 23.47 on five degrees of freedom and a p-value below 0.001, and the Hosmer-Lemeshow goodness-of-fit test indicated adequate fit, with a chi-square of 4.82 on eight degrees of freedom and a p-value of 0.775. However, the Nagelkerke R-squared was just 0.052, meaning the model explained only about 5.2 percent of the variance in professional AI use. Age, sex, ethnicity and professional category were not significantly associated with professional AI use, although doctors and dentists reported comparatively higher use at 60.5 percent, and use was relatively more common among professionals aged 21 to 50 years.
The study’s methods carry important caveats that shape how the findings should be read. The authors acknowledge this may have introduced selection bias toward settings with greater existing exposure to digital tools.
Perhaps the most consequential limitation is definitional. These categories differ substantially in technical architecture, clinical validation and regulatory status, and the authors state that this heterogeneity limits the precision of prevalence estimates and may have inflated the reported rate of AI use. They also note possible misclassification bias, since respondents may have categorised general internet-based or digital tools as AI or held varying understandings of what constitutes AI. The study could not distinguish between validated clinical decision-support systems and generative AI tools used informally for diagnostic queries, so the high frequency of reported diagnostic-support use should be interpreted with caution.
Further limitations include the self-reported nature of the data, which introduces recall and reporting bias, and the absence of task-level disaggregation of AI use by specific job functions such as triage, administrative automation, workflow optimisation or patient communication. Non-clinical personnel, who often use AI for operational tasks, were not explicitly included, and the study focused exclusively on public-sector professionals, excluding private-sector healthcare workers, which may limit generalisability to Lagos as a whole.
Despite these constraints, the authors argue the findings carry clear implications for Nigeria’s health system and comparable urban settings across sub-Saharan Africa. They call for structured AI literacy programmes integrated into continuous professional development, robust data governance frameworks, regulatory standards addressing data privacy, algorithmic transparency and clinical accountability, and investment in digital infrastructure, particularly at the primary care level, where facilities serve as the first point of contact for most patients but are often the least equipped. They also highlight broader challenges facing African health systems, including health data poverty, the development of most AI applications outside Africa using datasets that may not represent African populations, and the resulting risks of algorithmic bias and a widening digital health divide. Future research, they suggest, should incorporate task-specific assessments, more representative sampling and longitudinal approaches to evaluate the effectiveness, clinical impact and equity implications of AI integration in healthcare.
Subject of Research: Prevalence, patterns and determinants of artificial intelligence use among healthcare professionals in public health facilities in Lagos, Nigeria
Article Title: Artificial intelligence use among healthcare professionals in healthcare facilities in Lagos, Nigeria: a cross-sectional study
Article References: Chima-Oduko, A. A., Maduafokwa, B. A., Abdulraheem, K. S., Roberts, A. A., Obua, V. C., & Okocha, A. (2026). Artificial intelligence use among healthcare professionals in healthcare facilities in Lagos, Nigeria: a cross-sectional study. BMC Digital Health, 4(1), Article 58. https://doi.org/10.1186/s44247-026-00291-8
Image Credits: AI Generated
DOI: 10.1186/s44247-026-00291-8
Keywords: artificial intelligence, healthcare professionals, Nigeria, Lagos, cross-sectional study, digital health, diagnostic support, AI training, data privacy, clinical governance, public health facilities, health workforce
News Source: Ophelia Keating. (November 5, 2026). Most healthcare workers in Lagos use AI tools, but few have any formal training. Scienmag.
