Primary care for cardiovascular disease (CVD) – the prevention of a CVD event in people who have not experienced one before – is an established objective in clinical practice. The premise is that certain health or lifestyle indicators associate with the risk of a future CVD event, such as a heart attack or stroke. It is hypothesised that recognising these indicators and being pro-active, where possible, will avoid, or at least defer, such events.
However, many indicators associate with CVD, including: immutable factors (age, gender, ethnicity, socioeconomic status, air pollution and other environmental factors, family history); factors that are mutable in part by behaviour (weight, alcohol, smoking, diet, physical activity, stress) and; those that may be medically treatable (cholesterol, blood pressure, inflammation, mental health, diabetes or other co-morbidities). Most of these do not act in isolation.
In order to standardise CVD risk-estimation, attempts have been made to identify those factors that are most important, and to combine them into a ‘global’ (or ‘absolute’) risk calculation yielding a single number that is believed to estimate the risk of a CVD event over a future period (5 years, 10 years or lifetime risk).
Global calculators are now at the core of Clinical Practice Guidelines (CPGs) in most countries. The thinking is that CPGs and their global risk value could help with, and to some extent standardise, clinical decision-making. The problem is that they may undermine bespoke clinical judgement if enforced rigorously.
While clinicians may treat individual factors (such as high blood-pressure), the treatment option that CPGs recommend, based on global risk, is the prescription of statins. Global calculators and their associated CPGs have become mechanisms for prescribing statins.
Given that global risk-calculators have infiltrated guidelines worldwide and are triggers for initiating statin treatment in healthy individuals, it is worth scrutinising them a little more deeply – for example, how are they derived, how predictive are they, and does their use improve clinical outcome. I will address CPGs as a separate post.
The development of global risk-calculators for CVD started 70 years ago with the Framingham Heart Study, which set out to identify the most important health and lifestyle indicators (which they termed risk-factors). These have not changed substantially in the intervening decades.
The Framingham Heart Study (FHS)
Framingham is a small, quiet township on the outskirts of Boston – mainly white and middle-class, where everyone knows nearly everyone. In 1948, around 5,000 of its citizens, 40-59 years of age, volunteered to take part in a prospective study of CVD. The study continues to this day in their offspring.
On recruitment, volunteers underwent standard tests of the day that included: a chest X-ray, electrocardiogram (ECG), urinalysis, and measurements of lung capacity, haemoglobin, haematocrit, blood glucose, uric acid, lipoproteins, total cholesterol, phospholipid, as well as a number of other measurements. The idea was to follow these people over time, identify those who developed CVD, and then determine which baseline measurements most closely associated with that outcome (assuming nothing had changed in the meantime).
The initial analysis was published in 1961, and introduced the term ‘risk-factor’ for the first time. The study identified gender, age, systolic blood pressure, total cholesterol, and an ECG abnormality (left ventricular hypertrophy, LVH – enlargement and thickening of the heart’s main pumping chamber), as the core risk-factors.
The Framingham study group went on to combine these risk-factors into a scoring system that was first described in the 1990s. It has become known as the Framingham Risk Equation (FRE), and the most recent revision was published in 2008.
The Framingham Risk Equation (FRE)
The 2008 FRE modelled the likelihood of CVD based on the primary factors of age, gender, total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), systolic blood pressure (SBP), anti-hypertensive medication use, smoking status, and diabetes status. Other factors (e.g. diastolic blood pressure, body-mass index (BMI) and triglycerides) were considered, but did not add significantly to the model in the presence of the primary factors.
A noteworthy aspect of the FRE is the absence of low-density lipoprotein cholesterol (LDL-C) as a risk-factor. LDL-C is commonly referred to as the ‘bad’ cholesterol, and it is the principal target for statin treatment in primary care settings. The authors note that “the use of LDL-C did not improve model fit or performance”. Which raises the question – if LDL-C is not associated with CVD risk, then why is lowering it such a priority in clinical practice? The story behind that is embedded in the CPGs (to come).
The current FRE offers a version in which no cholesterol values are used at all. They are replaced by BMI, which gives a score (considered adequate) that enables blood testing to be deferred.
The globalisation of global risk calculators
There are now many global risk-calculators available in different countries and often more than one calculator for a given country. They build on the core FRE risk-factors, but often use different population data or incorporate additional risk-factors. The US American College of Cardiologists and American Heart Association (ACC/AHA) recently introduced the Pooled Cohort Equation (PCE) based on populations better representing ethnic diversity than white middle-class Framingham. There is also a Reynolds Equation. The UK uses QRISK3 that incorporates ethnicity, socioeconomic status and a host of additional supposed risk-factors. The calculator stems from a university-private company partnership that provides the basic calculator for free while having for-profit add-ons (QRISK3 is trademarked). It is a form of privatisation of risk calculators, and appearing arcane by its use of multiple risk-factors is perhaps a marketing strategy. The Joint British Societies (for the prevention of cardiovascular disease) also has its own online calculator (JBS3). Australia and Canada recommend versions of the FRE. New Zealand uses a primary Prevention Equation (PPE) based on their own population. The European Union uses SCORE (countries in the EU are also stratified into high-risk, e.g. Eastern Europe, and low-risk Denmark, Finland, France etc).
Common across each of these calculators is that they all require values to be input for TC and HDL-C, but none use LDL-C in their estimation of risk. Not even QRISK3 with its daunting list of risk-factors.
Naturally, LDL-C contributes to TC, so it is indirectly involved. However, it is not considered a primary risk-factor for estimating the likelihood of CVD. Even TC and HDL-C are not independent risk factors, most calculators use their ratio.
One of the controversies surrounding risk-calculators is where to set the threshold for intervention. CPGs have been steadily lowering this threshold over time. In the US, the ACC/AHA threshold stands at a 10-year risk greater than 7.5%, whereas a different set of guidelines (US Preventative Services Task-Force, USPSTF) uses 10%. The UK recently reduced the threshold from 20% to 10% (a change that was highly contentious). The effect of threshold-creep is that, with each reduction, increasing numbers of people are medicated for primary care. There are some who consider this a good thing – more people are made ‘eligible’ for statins in their thinking. Others point to the increasing medicalisation of healthy individuals, without a confirmed health benefit.
Not surprisingly, the greatest risk-factor is age. We know this because our lifespans are not open-ended. Inevitably, the further along in a lifespan, the greater the likelihood of mortality. As CVD is a common cause of mortality, the greater the risk of CVD.
For example, using the PCE calculator, take a male who is 64 years of age. Then enter ‘healthy’ values for everything else: TC = 4.0 mmol/L (160 mg/dl); HDL-C = 1.5 mmol/L (60 mg/dl); systolic blood pressure = 120; diastolic blood pressure = 75; never smoked; non-diabetic.
The 10-year global risk for this heart-healthy individual is 8.1%, which is above the 7.5% threshold set by the ACC/AHA (but below the USPSTF). The calculator reports: ‘On the basis of your calculated risk over 7.5%, the ACC/AHA guidelines suggest you should be on a moderate to high intensity statin.’
Since this man had healthy biomarkers, it means that any male over the age of 64 following these guidelines would be advised to be on a moderate to high intensity statin, whether he has other risk-factors or not. The threshold for initiating treatment in healthy females (with the above vales) is 71 years of age.
This renders the calculators deceptive for older people. If guidelines had recommended explicitly that anyone over these ages should be on a moderate to high-intensity statin, I would hope they would not be taken seriously. However, this same advice can be hidden in a calculator, using other risk-factors as decoys.
Even without the calculator, the evidence-base for statins in older people is weak. Randomised controlled trials of statins rarely recruit anyone over 70 years of age. The association between cholesterol and CVD is weak in this age group. Other considerations include the likelihood of more pronounced statin-related side effects in the frail elderly, such as muscle weakness and fatigue (and the risk of falls), or poorer cognitive function. Consideration needs to be given to quality vs. quantity of life, with a more rounded understanding of the patient and their goals.
The calculators also do not perform well in younger people (say 30-55 yers of age). If people in this age-range are studied after their first heart attack, and the calculators applied to data available then or prior, typically the calculated risk for them is low. That is, when advancing age is taken away as a major risk-factor, the other risk-factors are not very good at predicting heart attack.
Risk of what?
The calculators look for associations between risk-factors at baseline and CVD events during a follow-up period. The more events there are, the more data there will be to strengthen those associations. To increase the number of events, a number of types of events are grouped, non-fatal or fatal. For example, the FRE events includes any of the following: “Coronary heart disease (coronary death, myocardial infarction, coronary insufficiency, and angina), cerebro-vascular events (including ischaemic stroke, hemorrhagic stroke, and transient ischaemic attack), peripheral artery disease (intermittent claudication [muscle cramp after exercise]), and heart failure.” Whereas, the PCE estimates the risk of: “Non-fatal myocardial infarction, death from coronary heart disease, or fatal or nonfatal stroke”. Therefore, the choice of calculator can matter, depending on the risk being assessed.
Furthermore, the risk-factors entered into the calculators, and the populations under study, vary by country. Taken together, this means that risk can vary greatly for an individual depending on their circumstances. Take the man with the parameters given previously. Assume he is living in the UK. Both QRISK3 (United Kingdom) and SCORE (European Union) calculators apply to him. QRISK3 has him at an 8.1% risk over 10 years, whereas the SCORE risk is 2%.
Do risk calculators achieve a clinical benefit?
The question does not have an answer because there have been no substantive trials undertaken to answer that question – i.e. no statin clinical trials that report clinical outcomes have enrolled patients based on a specific CVD risk threshold. While the concept of treatment based on CVD risk may be intuitively appealing, remarkably its effectiveness has not been validated. A recent review concluded: “…there is currently no evidence … that the prospective use of global cardiovascular risk assessment translates to reductions in CVD morbidity or mortality.“
There have been a small number of trials on the effect on biomarkers such as cholesterol and blood pressure. These find that there is some small improvement in such biomarkers, without being able to show that this translated into a clinically meaningful outcome. The biomarker results likely indicate that the risk-score frightened more people into taking medication. The influence of risk-assessments on worry, in the context of a consulting room environment, is rarely acknowledged.
Risk-estimates have little to offer an individual – accurately identifying a person’s true risk will always be imperfect. The calculators do not estimate a specific person’s risk, but rather the average risk for a person with comparable baseline values in a selected population under study. Like most risk, it’s a lottery.
At best, risk-estimates might be useful for tracking risk at a population level. This is somewhat like the body-mass index (BMI) – potentially misleading for an individual, but a useful shorthand for estimating average obesity across a population.
By definition, 10-year risk-calculators have to be based on data at least 10 years old. The PCE is based on data from the 1990s. Thus, these data come from a prior generation when cardiovascular risk profiles and pharmacological therapy were limited compared to the modern era. Whether these data can be generalised to modern patient populations is not established.
Many of the risk-calculators imply a precision – to one decimal place. This is pseudoscience. Given the vagaries of predicting anything up to 10 years in advance from a single set of baseline measures, this degree of precision is deceptive. As well, predictions contain a margin of error that is likely to be substantial. The risk-calculators never indicate a margin of error. In addition, they take themselves quite literally. For example, if the baseline parameters of our hypothetical 64-year old man are manipulated until the PCE gives a risk of 7.4%, ACC/AHA guidelines suggest that there is no need for treatment. If adjustments are made so that the risk comes out as 7.6%, then a moderate to high intensity statin (for life) is recommended. In the era of AI, these algorithms are unsophisticated and outdated.
Risk-factors don’t tell us what contributes causally to CVD. For example, left ventricular hypertrophy (LVH, a risk factor identified by the FHS), may arise from chronically high blood-pressure, which in turn may arise from obesity that itself may be a consequence of chronic insulin resistance stemming from a lifetime diet high in sugar. Treating LVH (and the other risk-factors) may well be advised, but it is treating a symptom, not a cause. In this scenario, the cause is dietary sugar.
Disclaimer: I am not a medical doctor. Nothing herein is, nor should be taken to be, medical advice.
This post is one of a series of eight exploring cholesterol, statins and heart disease. The full list and links:
The evolutionary significance of cholesterol ►
How cholesterol and other lipids are trafficked in the circulation ►
Does cholesterol even cause heart disease? ►
A case study of an RCT: the marketing science of Lipitor ►
An overview of statin RCTs ►
Cholesterol clinical guidelines ►
Statin adverse effects ►