P-CARDIAC finds a place in Hong Kong primary care, but awareness still trails clinical need
A machine-learning cardiovascular risk model designed for Chinese patients in Hong Kong is moving from prediction to practical use, and the early signal is encouraging. In a BMJ-nutrition, prevention & health study, P-CARDIAC was examined from two angles at once: whether clinicians and members of the public would accept it, and whether it could function inside pharmacist-led primary healthcare services.
The premise is straightforward. Cardiovascular risk tools are only useful if people understand them, clinicians trust them, and the workflow is simple enough to sustain in everyday practice. P-CARDIAC was developed for the local population and had already shown stronger predictive performance than existing risk tools, making it a natural candidate for implementation work in Hong Kong’s primary care setting.
The study did not test hard cardiovascular outcomes. It asked a more immediate question: can a locally tailored risk model be used in a way that patients and providers actually value?
Public interest was higher than public awareness
The cross-sectional portion surveyed 113 members of the general public and 17 clinicians. Public awareness of cardiovascular risk prediction tools was low, yet health-seeking behavior around such tools was high. That combination matters at the bedside. People may not know the names of risk calculators or prediction platforms, but many are open to tools that explain their personal cardiovascular risk in a concrete way, especially when those tools are linked to practical next steps.
For prevention medicine, this is a familiar gap. Risk prediction only becomes actionable when a patient can connect numbers on a screen to daily decisions about blood pressure control, lipid management, medication adherence, diet, exercise, and follow-up. A model that feels abstract will struggle; one that translates risk into an understandable care plan can support behavior change.
Clinicians brought a different perspective. Experience with risk prediction tools was limited, particularly among cardiologists, and that limitation was tied to unavailability and limited user-friendliness. In real-world primary care, the interface often determines whether a model survives beyond a pilot. Even a well-performing algorithm can fail if it adds friction to a busy clinic session or demands extra time that no one has.
The pilot study suggests pharmacist-led delivery is workable
The implementation arm was small, with 15 participants, but it offered a practical proof of concept. P-CARDIAC was incorporated into pharmacist-led services at local primary healthcare centres, and participants responded positively to the experience. The service format itself appears to have been part of the value proposition: risk assessment paired with medication review and counseling gave the tool clinical context instead of leaving it as a standalone score.
Participants felt the model could support better medication adherence and disease management. In cardiovascular prevention, that belief is clinically plausible. A personalized risk estimate can sharpen the rationale for antihypertensive treatment, lipid-lowering therapy, smoking cessation, and other preventive measures. When risk is individualized, the conversation becomes less theoretical and more immediately relevant to the patient sitting in front of the pharmacist or clinician.
The pilot also speaks to workflow. Pharmacist-led services are well positioned for this kind of intervention because they already center on medication use, adherence barriers, patient education, and longitudinal support. Embedding a risk tool in that setting may help close the common divide between risk calculation and treatment action.
Machine learning is useful only if the clinic can use it
The appeal of a machine-learning-driven model lies in tailoring prediction to local population data, which may improve calibration and relevance compared with imported risk equations. For Hong Kong, and particularly for Chinese patients, a locally developed tool can feel more clinically credible than a generic model built elsewhere. Precision, however, is only part of the equation. Usability, familiarity, and workflow integration decide whether a model changes care.
This study makes that point plainly. Clinicians had limited experience with such tools, while the public had limited awareness. Both groups still responded positively once the model was presented within a care pathway. The lesson is not that risk tools are unnecessary; it is that adoption depends on education, usability, and a service structure that makes the output actionable.
A risk calculator that cannot fit into a consultation is just another screen. A calculator that supports medication conversations, adherence counseling, and disease management can become part of prevention care.
What the pilot can and cannot claim
The study used descriptive statistics, univariate regression modeling, and reliability assessments of validated measurement scales. Those methods are appropriate for an early acceptability and feasibility project, but they do not establish whether P-CARDIAC improves cardiovascular outcomes, reduces events, or changes long-term prescribing behavior. The pilot size was small, and the clinician sample was limited as well.
Even so, the practical signal is hard to ignore. Participants received the service well, and the model appears to have matched a real need in primary care: a locally relevant way to talk about cardiovascular risk without losing the patient in technical complexity.
The next step is clear from the design itself. Larger implementation studies will need to test how the tool performs across broader primary care settings, how often it is used, whether it changes medication adherence or follow-up behavior, and whether clinicians continue to find it practical after the novelty fades. Those questions will determine whether P-CARDIAC remains a promising pilot artifact or becomes part of routine cardiovascular prevention in Hong Kong.