When Numbers Start Making Decisions · Season Three, “The People the Average Does Not See” · Article 8
1. What does a personal-looking percentage actually mean?
A cardiovascular risk calculator combines information such as age, sex, smoking, blood pressure, cholesterol, diabetes and other factors, then reports the estimated probability of a cardiovascular event within five years. A result of 10 per cent can move a person into Australia’s high-risk category and change the treatment conversation.
The 2023 Australian guideline classifies five-year risk as low below 5 per cent, intermediate from 5 to under 10 per cent, and high at 10 per cent or above. For people assessed as high risk, preventive pharmacotherapy is recommended after a discussion of benefits and harms unless contraindicated or clinically inappropriate. Australian CVD Risk Calculator: Overview
The percentage looks individual because personal data entered the model. It is still a group-based estimate. It means that among people sufficiently similar in the development and validation evidence, roughly a stated proportion would be expected to experience the defined outcome over the period. It does not identify which individual will.
The threshold converts uncertainty into an action structure. The central question is how to use that structure without allowing 9.9 and 10.0 to impersonate radically different biological realities.
2. Risk prediction is not diagnosis
A person can feel well and still have elevated future risk. That is why prediction matters: prevention acts before an event. Unlike a diagnostic test for an existing disease, a risk model estimates a possible outcome from patterns observed across populations.
The estimate depends on the outcome definition, time horizon, variables, equation and population. It can be well calibrated overall while over- or under-estimating risk in a subgroup. It also cannot include every determinant: family history, social circumstances, treatment adherence, changing behaviour and unmeasured biology may affect the individual’s future.
Uncertainty does not make the result meaningless. Clinicians routinely make decisions under uncertainty. A validated model usually improves consistency over unaided impressions and ensures major risk factors are considered together rather than one at a time.
Its authority should nevertheless remain predictive. “Your five-year risk is estimated at 10 per cent” is accurate; “you have a 10 per cent condition” is not. The estimate can support a choice without becoming a fact about which future will occur.
3. Why a 10 per cent line is useful
Preventive medicine must balance expected benefit, adverse effects, burden and cost. The higher the untreated baseline risk, the greater the absolute number of events a treatment can prevent if relative benefit is similar. A threshold helps clinicians identify the range where medication is more likely to offer worthwhile absolute benefit.
The guideline was designed to improve assessment and management for Australian adults and replaces older approaches with updated prediction. Its published account stresses clinical judgement and shared decision-making rather than automatic prescribing. Medical Journal of Australia: 2023 Australian guideline
The line also supports system consistency. Without categories, similar patients may receive very different advice depending on which risk factor catches a clinician’s attention. A common model can reduce arbitrary variation.
But the threshold is a decision convention applied to a continuous risk curve. A person at 9.8 per cent can have almost the same expected benefit as one at 10.1. The closer the result is to the line, the less defensible a purely categorical response becomes.
4. Input uncertainty travels into the output
Blood pressure varies with technique, stress, time and repeated measurement. Cholesterol changes over time. Smoking status may be simplified. Diagnoses and medicines can be incomplete in a record. Entering inaccurate or non-current data produces a precise-looking but unreliable estimate.
Even correct inputs have measurement error. A risk calculator often reports a single percentage without showing how much the result might change under plausible input variation. Near 10 per cent, repeating blood pressure or correcting a record can change the category.
Good practice verifies inputs, uses recommended measurement methods and saves the assumptions underlying the result. The calculator should be rerun when relevant circumstances change. A risk estimate is time-stamped evidence, not a permanent score.
Some people are already considered at high risk on clinical grounds or fall outside the population for whom the equation is intended. The tool’s scope and guideline exceptions must be checked before its number is allowed to lead.
5. Treatment is a decision made with the person
A high-risk classification changes the expected balance of preventive medication; it does not remove patient preference or clinical judgement. The discussion should cover absolute benefit, possible adverse effects, interactions, cost, monitoring and the person’s values.
Absolute risk is especially important. If a medicine reduces relative risk by a similar proportion, people with higher baseline risk generally gain more absolute benefit. Communicating only relative reduction can exaggerate the apparent effect. Natural frequencies—events expected among 100 similar people with and without treatment—can make the choice clearer.
Lifestyle support is not a consolation for people below the threshold, nor should medication replace it above the threshold. Smoking cessation, nutrition, activity, blood-pressure management and other interventions remain relevant across categories.
Shared decision-making does not mean abandoning evidence to preference. It means presenting the evidence in a form that allows the person to judge benefit and burden, while the clinician identifies contraindications and takes responsibility for a recommendation.
6. Fairness requires calibration across populations
Risk models can reproduce gaps in their source data. If a population was underrepresented or experienced different access to diagnosis and care, observed outcomes may not transfer cleanly. Social disadvantage can affect risk through mechanisms not fully captured by conventional variables.
Validation should examine calibration and discrimination across sex, age, First Nations status and other relevant groups where evidence permits. It should also examine the consequences of false reassurance and overtreatment. A model that is accurate on average may still distribute errors unfairly.
The guideline and calculator include specific advice about populations and clinical conditions; users must not strip the percentage from that context. Australian CVD Risk Calculator
Monitoring after implementation matters too. Are high-risk people receiving a discussion and effective prevention? Are intermediate-risk people with additional evidence being reviewed? Do event rates correspond to predicted risks? Outcomes should be able to revise the model and its threshold.
7. A practical framework for risk thresholds
When a predicted risk changes treatment, ask:
- Is the model intended for this person and outcome?
- Are the inputs accurate, current and measured appropriately?
- How close is the estimate to the threshold?
- Which important risk factors are absent or handled separately?
- What is the absolute expected benefit and harm of treatment?
- Can the result be explained in natural frequencies?
- What do clinical judgement and patient preference add?
- Has the model been validated and monitored across relevant groups?
The framework turns the calculator from an answer machine into a structured source of evidence.
Conclusion: risk should inform the decision without claiming to know the future
The five-year CVD estimate makes prevention more systematic. It combines risk factors, directs attention before symptoms occur and helps allocate treatment to those likely to gain more absolute benefit. The 10 per cent line is useful because a health system needs an actionable rule.
My judgement is that crossing 10 per cent should strengthen and structure the recommendation, not mechanically decide it. Results near the line need verified inputs and recognition of continuity; every result needs applicability, absolute benefit, clinical context and a real discussion with the person.
A predicted group frequency can guide an individual choice. It cannot tell the individual which member of the future group they will become.