Identifying System Level Risk Measurement and Risk Scores

Deepak Bhatt, MD, MPH: Mike, you’re one of the few folks I know who really plays in both the venous and arterial space. As it turns out, there are a lot of shared risk factors. Inflammation is one that’s particularly high. What are your thoughts on the overlap of risk factors for venous thromboembolic disease and arterial syndromes, things like MI [myocardial infarction] and ischemic stroke, for example?

C. Michael Gibson, MS, MD: Well, it’s interesting. We tend to study these drugs in people who are medically venous patients or arterial patients. Patients have both arteries and veins, and they’re at risk of events in both systems. You’re exactly right, Deepak. If you’re at risk of venous thrombosis, you’re probably also someone who, in terms of having bad blood, is also at risk of an arterial thrombosis. When we talk about the treatment here, we’re going to see that yes, we reduce the venous events, but we also reduce some of the arterial events, things like stroke, and heart attack. The people in the hospital often have something like a stroke, which puts them at risk of immobility, or they’re someone who had a heart attack who then went on to have heart failure. These 2 systems are highly interrelated.

Deepak Bhatt, MD, MPH: You mentioned bad blood. Valentin Fuster, MD, used to talk about this concept of vulnerable blood. Do you think there’s something to that?

C. Michael Gibson, MS, MD: Absolutely. Every meeting we come and we find out another biomarker of thrombogenecity and bad blood. We not only have to treat the risk factors in terms of immobility but also the bad blood.

Deepak Bhatt, MD, MPH: Gary, we’ve talked a bit about venous thromboembolism [VTE] and medically ill patients. They’re at heightened risk for VTE. How can we really gauge this risk? What should we be doing?

Gary Raskob, PhD: I think the most important thing is that in the health system there is a process for a routine risk assessment for VTE for every patient who’s admitted to the hospital. There are evidence-based approaches to do that. There are apps that can be put on a phone or you can work with your IT department in the hospital, you can work with your patient safety quality department in the hospital. Now every patient admitted to hospital should have a risk assessment for VTE.

Deepak Bhatt, MD, MPH: You think it’s really something that should be happening at a systems level that is pre-programmed into electronic health records, or is it up to individual physicians to sort it out on their own?

Gary Raskob, PhD: Clearly a lot of evidence indicates that you have much better outcomes and results if you do this in a system level and implementation at a system level. After all, physicians have lots to think about and worry about, and I think these kind of things can be simply automated and made routine through these information technology systems and other approaches. That’s the way to do it.

Gregory Piazza, MD, MS: Deepak, we’ve had good experience programming in things like CHADS-VASc [score for congestive heart failure, hypertension, age, diabetes, stroke—vascular disease] to identify high-risk patients with AFib [atrial fibrillation] for stroke. The electronic health records now are so elegant in their constructs that we can actually pull this information fairly easily and take the duty off of the providers, like Gary said, and make it easy for them to find the right patients.

Deepak Bhatt, MD, MPH: Well, you’ve done pioneering work in that regard. Certainly at the Brigham and Women’s Hospital you’ve been in charge of getting those systems in place. I think Gary is right, the onus does fall on the health care system. As an individual physician it’s tough to keep track of that stuff, and there are all sorts of risk scores out there like for acute coronary syndromes, atrial fibrillation, PE [pulmonary embolism], VTE. For primary care physicians, there’s been an even wider range that is focused on some cardiovascular scores. What would you actually do, Alex? What do you recommend in terms of risk scores? Which ones should we consider using, or should we just use our gestalt clinically? What’s the right way?

Alex C. Spyropoulos, MD, FACP, FCCP, FRCPC: I think, Deepak, you bring up an important point and I agree completely with Gregory in the sense of scores like the CHADS or CHADS-VASc or other scores, what they do is they not only give you individual risk factors, but they weigh these risk factors and then assign a score to them. Not all the risk factors are weighed the same. The nice thing now is that the business of risk scoring for VTE in this population is gaining more and more sophistication, and we have a lot of background data to suggest there’s a collection of very similar risk factors that put these patients at risk.

You have the acute medical illness, you have the level of immobility, which are what I like to call disease-specific risk factors. Then you have the patient-related risk factors, and a lot of these risk factors are enduring, risk factors such as advanced age, or prior history of VTE is a very strong risk factor, or history of infection, or history of thrombophilia, or active cancer or history of cancer, or admission into the ICU [intensive care unit]. These are all risk factors that are again dependent on the patient or some around the disease.

The nice thing is whether you look at the Padua Prediction Score tool or the IMPROVE tool, and the reason why I shouted out these risk models—because they’ve really undergone the most external validation, both of these scores—is that they really align nicely with those kinds of key risk factors. They give the clinician a score, and depending on what threshold score is reached, they will assign patients into at-risk or even sometimes high-VTE-risk categories. I completely agree with Gary. The key is not so much what score you use, you should use a validated score, but the fact that it’s systematized and universalized within a health system, that’s what’s critical.

Deepak Bhatt, MD, MPH: That’s good to reduce variation in care. Mike?

C. Michael Gibson, MS, MD: We now are just about to publish machine learning and artificial intelligence [AI] for VTE. Scores are great, and we worked on the IMPROVEDD score together, adding D-dimer in to all those clinical assessments. But the advantage of machine learning and AI over numeric risk scores is it handles these kinds of things as a continuous variable. Instead of using 5 or 6 things, you can use 50 or 60 things from the electronic health record to have an even more precise estimate of risk. I think all this is too much for clinicians, in many ways, that you can’t think about 50 things simultaneously. I can barely think about 2 things simultaneously. This is where the electronic medical record is going to shine is in the ability to pull all these factors together and then in an automated fashion give us these risk scores, which are better than the old 1, 2, 3, 4, 5 kind of risk scores.

Deepak Bhatt, MD, MPH: I’ll digress for a moment. Maybe you could share with the audience, I know you’re leading a lot of this type of machine learning work at Baim Institute. What exactly does machine learning and AI in this context refer to, what’s going on?

C. Michael Gibson, MS, MD: That’s a great question, Deepak. You take a lot of information, you take all the data, for instance from a trial, and then you figure out which ones of these variables are more related to the outcomes. Rather than looking at linear relationships in 1 dimension, it looks at curvy linear relationships in multiple dimensions. It’s a little bit more like a neuron firing. You put in information, and you see how do you weight the firing of all these different synapses to get the answer that fits truth, and then you tweak all those synapses. It’s a little bit of a black box, but it does outperform traditional risk scores. It needs to be learning, so the answer today may not be the same as the answer in 2020 because all the systems of practice have changed. As an adaptive system, it’s able to better adapt to all the new information than we could as clinicians.

Deepak Bhatt, MD, MPH: That’s interesting.

Transcript edited for clarity.