How to choose the optimal therapy?

-Hello, my name is Kausik Ray. I'm Professor of Public Health at Imperial College London. Now, the title of my talk is Addressing the Clinical Challenge, How to Choose the Optimal Therapy. This is one of the commonest challenges we face as clinicians. I'm going to show you some tools that we've been working on at the end, really, that might help us in the future. These are a list of my disclosures.

What are the goals of treatment? Is it to achieve a number? What is it that you want to achieve? We're really not interested in making changes in these numbers, but it's the background to those numbers that's important, whether it's blood pressure, glucose, LDL cholesterol. What we're really interested in is preventing events. We might be interested in preventing or slowing down the underlying disease process, for example, atherosclerosis, and its clinical consequences, myocardial infarction, stroke, and coronary artery disease in the form of stable vascular disease requiring revascularization. Trials give us information about relative benefits at a population level, how much, if I use that treatment, benefit return would I get? When we see patients at an individual level do we really know how much individual benefit patients will achieve?

To do that, we can look at several different things. Looking at people with established cardiovascular disease, and if you think about the European guidelines, for example, we'd probably classify all of those people as very high risk. There's a risk calculator that's been developed exactly to estimate, like primary prevention, 10-year risk of cardiovascular events, and this is called the SMART risk equation or the SMART risk calculator. That's been studied in multiple different populations, but populations studied in trials, or in cohort studies, epidemiological studies, there's a real systematic approach to data collection. In the real world, doctors are busy, measurements are imprecise. The question was, would the SMART risk prediction tool be valid if I put it into electronic health records, for example, like the UK primary care system?

We looked at the validity of this in 380,000 individuals with established cardiovascular disease, and we looked at over 45,000 cardiovascular events. In the population getting standard of care, usually statins, blood pressure lowering treatment, the residual risk over 10 years was 29%. It was high in men and in women. There's a residual risk, is the take-home message, even when we're using first-line treatments, and often costs restrict the ability to add in additional medications. If we could identify the right people at the greatest risk, not just to identify the risk, but the benefit of what it is we're trying to do, that might be something that is optimal for healthcare providers, doctors, and patients.

On the left-hand side, the first thing to show you is the distribution of risk in the patient population. Just like all populations, all of these high-risk patients, very high-risk patients don't have the same event rate. It's a skewed distribution, and that can be seen here. How well does it perform? It performs really well in men or women across most of the risk distribution, and even at different age groups as well. We can actually have confidence in simulating risks and benefits from different interventions.

Let me take you through 13 different patients in your clinic, they all have ASCVD, and by definition, they're all very high risk. Patients 1 to 4 have very different non-HDL or LDL cholesterol, but the global risk, 10-year risk using SMART is 20%. Patients 5, 6, 7, and 8 have a higher global risk. Their lipids are variable. The global risk is higher than the first four patients. What I've done in the next five patients 9, 10, 11,12, and 13, is I've kept lipids constant, non-HDL or LDL. Now, because other things vary, 10-year risk varies quite widely in patients 9, 10, 11, 12, and 13. Now, if I take a treatment, where the benefit solely depends upon global risk, we can look at what happens. For example, a treatment like rivaroxaban, which reduces risk by about one quarter would, in the first four patients, reduce risk by 4.8% in absolute terms because it's not dependent upon LDL or any other individual factor, but global risk. In the next four patients, patients 5, 6, 7, and 8, global risk is higher, therefore, absolute benefit is greater. Twice as much. In the next five patients, because global risk varies, the absolute benefit varies. If you had to choose one patient out of this group for rivaroxaban, you would probably choose that patient 9, for example. Now, if you do the same thing for LDL lowering with a treatment that lowers non-HDL by let's say 50%, you have 13 different possibilities. No doctor in the world can do that in their head because the benefit depends upon global risk, the starting level of LDL, and how much absolute lowering you achieve. That's what a tool like this could help you with.

We've developed an algorithm whereby you can take the SMART risk calculator, you can take some of the work that we've done both on LDL lowering and time of exposure. In the top line, you can see that the risk over 35 years if you do nothing, no lipid-lowering therapy, and you can see that with atorvastatin 20, for example, you can bring their risk down to about 7% over 10 years. You can look at that option of, do I add in ezetimibe, or do I add in evolocumab, for example, in this particular scenario? You can project out the amount of benefit that you're going to get because you're now looking at the residual risk with the changes in these approaches.

You can see the importance of time. If you only use these tools to look at short-term risk and benefit, on the left-hand side, you could see the number you need to treat is higher than on the right-hand side when you look at a 30-year time horizon, which is the time horizon for atherosclerosis, that we really are thinking about 30, 40, 50 years. Also, we can use this tool to present risk and benefit to patients and doctors in a way that they can utilize.

At a population level, you can model the impact. If you're a payer, for example, on the left-hand side are the number of people divided by global risk and lipid levels. That's scaled up in the middle on a log scale. Then that's flattened as a heat map. Let's say I chose to treat the population, not everybody with ASCVD, just those with a total risk above 20% and a non-HDL above 2.6, for example, with a PCSK9 inhibitor. The heat map would change. You now get those white areas, but what I've done is I've moved people to the left, i.e, and upwards into a lower LDL and into a lower-risk category. I picked those people most likely to benefit.

This is lipids, this could be applied to any therapy. You could pick, for example, the GLP-1 receptor agonist. You could take the relative benefit, and you could apply that to individual patients, and you could calculate the absolute number of events averted, and, et cetera, in that patient population.

In conclusion, choosing the most beneficial option for patient and physician is driven by quantitative information. Understanding risk is not enough. You also need to understand benefit. If you can pull the two together, we might actually reduce clinical inertia. These can be combined for all therapies, and these can be integrated into electronic health records. This might help to improve implementation of preventive strategies. Thank you for listening.