Knowledge, Attitude and Factors Influencing the Implementation of Artificial Intelligence in Clinical Decision-Making Among Health Professionals at Buea Regional Hospital

Why this study

The global conversation asks whether health workers will accept AI. In this region that question was being answered by assumption rather than measurement, and mostly by people who had never worked a shift here.

Nobody had a profession-by-profession baseline for a Cameroonian regional referral hospital. Without one, any AI rollout is guesswork, and the failure gets blamed on staff resistance afterwards.

Design

A descriptive cross-sectional study at Buea Regional Hospital, covering six professional cadres: nurses, medical doctors, midwives, pharmacists, laboratory scientists, and other allied staff.

419 self-administered questionnaires were distributed. 387 returned complete with valid consent, a response rate of 92.4%, comfortably above the 377 minimum required by the sample size calculation. Ethical clearance came from the Institutional Review Board of Gracious Higher Institute of Excellence. Analysis ran in SPSS with significance set at p < 0.05.

Respondents were young, mean age 33.4 years, and the sample was 59.2% women and 55.3% nurses, which matches the staffing profile of a regional referral hospital. That matters for reading the results: the workforce that will actually meet AI at the bedside here is a young nursing one, and readiness has to be judged on their terms.

The instrument

I built the KAP questionnaire rather than adapting an existing one, because existing tools ask about AI in computer science vocabulary. That measures whether a respondent recognises jargon, not whether they understand what the technology does in a ward.

The knowledge section tested twelve concrete, awareness-level statements. The difference is not cosmetic. Studies that rely on self-rating produce results like 60% of students rating themselves high while 92% score objectively low. Testing beats asking.

Findings

Knowledge sits at moderate, and the gaps are specific

Mean score was 7.35 out of 12, or 61.3%. On Bloom’s scale, 41.6% reached good knowledge, 30.0% moderate, and 28.4% poor.

The shape of the gap matters more than the average. Respondents scored highest on tangible clinical uses, AI reading medical images at 75.2% and catching drug errors at 71.1%. They scored lowest on abstract and regulatory items: the absence of AI legislation in Cameroon at 41.3% and the limits of AI autonomy at 50.9%.

Staff here understand what AI does considerably better than they understand what it is or how it is governed. The uncertainty about Cameroonian AI law is not really ignorance. It is an accurate reflection of a policy vacuum that genuinely exists.

Attitudes are cautiously optimistic, and trust lags willingness

The composite attitude score was 3.50 out of 5. 30.7% held a positive attitude, 66.9% moderate, and only 2.3% negative. Outright rejection was almost absent.

Two items tell the real story. The strongest agreement in the whole section was the demand for training before AI is introduced, at 76.5%. Willingness to use AI if it were made available followed at 71.8%. The weakest item was trust that AI recommendations are accurate and evidence-based.

Staff want the help and withhold full trust. In a setting where you cannot easily check an AI output against a second opinion, that is not resistance. That is good clinical judgment.

Fears existed but were moderate, centred on de-skilling rather than replacement. Misconceptions were rare: few believed AI has human empathy or outperforms a clinician outright.

The barriers are structural, and staff name them precisely

Respondents rated eighteen factors on how much each would influence their own use of AI. Every single one scored above the midpoint, and institutional factors clearly outranked personal ones.

The top four were reliable electricity, rated influential by 94.8%, then the cost of AI systems, reliable internet, and access to computers or tablets.

Fear of losing their job ranked last of all eighteen.

The digital access data explains why. Smartphone ownership was near universal at 95.6%, yet only 33.6% had reliable internet at work and 28.2% had a departmental computer. Only 12.9% had ever had formal AI training. Staff are personally connected and institutionally under-equipped.

Asked directly about barriers, they named lack of training at 48.1%, poor infrastructure at 42.6%, and high cost at 39.8%. Job loss came last again at 13.7%.

Attitude is downstream of knowledge

This was the clearest single relationship in the data.

Among respondents with good knowledge, 57.1% held a positive attitude. Among those with moderate knowledge, 21.6%. Among those with poor knowledge, 1.8%. Every respondent with a negative attitude came from the poor-knowledge group.

Knowledge itself tracked exposure rather than demographics. Digital comfort showed the steepest gradient, from 12.0% good knowledge in the lowest comfort group to 87.0% in the highest. Prior AI use and formal training were both highly significant. Sex, years of experience, and internet access at work were not.

Attitude here is not a fixed trait. It is a downstream effect of what people know and have touched, which means it can be moved.

Conclusion

The people at Buea Regional Hospital are not resistant to AI. They are under-trained and under-equipped, and they know it.

That reframes the local problem from persuasion to provision. Much of the global literature is still debating whether health workers will accept AI. In this setting acceptance is largely already there, and the real work is training, infrastructure, and governance.

The constraints staff named are modifiable ones. That is the encouraging part: these are problems investment and training can actually relieve, unlike an entrenched attitude problem.

Research interests

AI in healthcare, digital health, clinical decision support systems, public health innovation, healthcare informatics, health systems strengthening, global health, and medical technology.

The thread running through all of it is the same as the thread in this study: the gap between what a technology can do and what a specific ward can actually run.