Michael Kattan, PhD, Chair of the Department of Quantitative Health Sciences at Cleveland Clinic, and Rahul Tendulkar, MD, radiation oncologist at Cleveland Clinic Cancer Center join the Cancer Advances podcast to discuss nomograms, a calculation tool designed to allow clinicians to evaluate the collective risk of multiple factors at once. Listen as they discuss the creation of nomograms and how they can be helpful to evaluate treatment options for cancer patients.
Dale Shepard, MD, PhD: Cancer Advances, a Cleveland Clinic Podcast for medical professionals. Exploring the latest innovative research in clinical advances in the field of oncology. Thank you for joining us for another episode of Cancer Advances. I'm your host, Dr. Dale Shepard, a medical oncologist here at Cleveland Clinic, overseeing our Taussig Phase 1 and Sarcoma Programs. Today, I'm happy to be joined again by Dr. Rahul Tendulkar, an Associate Professor of Radiation Oncology, and Dr. Michael Kattan, Chair of the Department of Quantitative Health Sciences. Rahul has previously joined me on the Cancer Advances Podcast to talk about PSMA-PET scans for patients with prostate cancer and this episode is still available. Today, they're here to talk to us about nomograms for prostate and breast cancer. So welcome, Rahul, welcome, Michael.
Rahul Tendulkar, MD: Thanks, Dale.
Michael Kattan, PhD: Thanks for having us.
Dale Shepard, MD, PhD: Absolutely. Maybe just start out, Rahul, remind us your role here at Cleveland Clinic.
Rahul Tendulkar, MD: Yeah, so I'm a radiation oncologist, as mentioned, specializing in breast and GU cancers, primarily prostate cancers, where I treat with radiation therapy.
Dale Shepard, MD, PhD: Very good. Michael, what do you do here?
Michael Kattan, PhD: Chair of the Department of Quantitative Health Sciences. That's a department of about 120 bio-statisticians, health services, researchers, bio-informatics types. Mostly all the quants in research are in this department.
Dale Shepard, MD, PhD: Very good. All right, well today we're going to talk about nomograms and so maybe just start out, we have a diverse group of people. Maybe you can tell us a little bit, what is a nomogram?
Michael Kattan, PhD: So, a nomogram is a way of depicting a prediction model on a sheet of paper or visually. So a lot of times when we're working with investigators like Dr. Tim Carr, and you, we are coming up with prediction models to help you in the clinic to predict probabilities of future events that you care about to take better care of your patients. A lot of times when we make those models with you, we're making regression models. Cox regression is a very popular common tool that we'll build for you or logistic regression, same thing. Those models though, are confusing. They're not user friendly in terms of the coefficients that are behind them. But if we display that model as a nomogram, then you as a clinician can see exactly what's going on, which variables have influence, which don't, and exactly how much influence each predictor in the model has.
Dale Shepard, MD, PhD: Nomograms, most people don't immediately think of them. How did you get into nomograms?
Michael Kattan, PhD: Right. So when I was 24, I was diagnosed with Hodgkin's disease and my only symptom was night sweats, and I didn't have any other complaint and ultimately found out I was stage 4 B. And so it was confusing to me as to what my prognosis was, and that really opened my eyes to how clumsy, if you will, our ability is to predict patient outcome at the individual level. It was very crude. My oncologist, who I was treated elsewhere, was really just going purely by clinical stage and not taking into account other features about me. And that frustrated me because I was getting a degree in the business school, which was all about financial prediction, where you make money, if you predict very accurately and that's your goal.
And over here, it just felt very different. Like we didn't give enough attention to the art and science of prediction, and I really wanted to do a better job with that. And I looked around and to me, the nomogram made the best sense in terms of preserving information and enabling the clinician to take everything he or she knows about the patient and put that together in one model that is optimized for prediction. And that's really what a nomogram does.
Dale Shepard, MD, PhD: That's excellent. So Rahul, I'm guessing you could confirm that most patients come in and they want this kind of information, right?
Rahul Tendulkar, MD: Absolutely. And I'm still blown away at how many people have been influenced by Dr. Kattan's own personal experience, which translated into his development of nomograms across multitude of disease sites. I think kind of the serendipity of all of those things happening has led to really, I think a huge change in how we approach patients. Because as he said, in oncology, we're so used to lumping into buckets based on stage, but we really lose out on the personalization aspect of that. And that's been a big personal interest of mine for the exact reason that you bring up, that patients come in with a new diagnosis and they want to know what are my chances, not what is the larger groups. But specific to me, based on this feature or that feature, what are my chances of success and what do I need to do to get there? And so these nomograms, we use on a regular basis in our clinics to try to help patients.
Dale Shepard, MD, PhD: Yeah. You're absolutely right. It was kind of one of these aha moments for you that really did change the field. So it's been a big deal. Give us an example, Rahul, if a patient comes in, how do you use this in a very practical way? Like what are the factors you're thinking about? Let's kind of think more of a very practical way. How do we use it?
Rahul Tendulkar, MD: So I think, and I want to ask Mike this question before I get to that. People talk about the Kattan nomogram, how many actual Kattan nomograms are there, that we're talking about?
Dale Shepard, MD, PhD: The nomogram, it's not the nomogram.
Michael Kattan, PhD: I didn't come up with my own marketing plan, it's served me quite well. Gosh, I mean, across any and all diseases, there might be a hundred nomograms I've made or something. They're probably talking about the preoperative nomogram for prostate cancer. That one, which kind of started a lot of things when they say that, but you never know. It's kind of funny.
Dale Shepard, MD, PhD: So for example, because you know, this is one where people talk about the Kattan nomogram and I say, "Well, which one?" So the ones that I've personally, found to be useful and ones that we've developed here, for example, we see a patient who's had prostate cancer, they've had their prostate removed and based on various pathology features, for example, was there extra prosthetic extension or seminal vesicle invasion? What were the surgical margin status? What's the PSA afterwards and the rate, which is changing? And so, you look at all of these variables and you can have a success rate with salvage radiation ranging from maybe 20% at the low end to 80 or 90% at the high end, depending on which features each individual patient has. And so, we use it in ways to intensify treatment for those who may have, and otherwise poorer prognosis and perhaps to de-intensify treatment, and those who are going to be successfully treated with limited treatment options. And so, it's really kind of helpful to try to personalize our care.
Rahul Tendulkar, MD: And so, Mike, you mentioned this being like a two dimensional way to assess risks. So, we live in a smartphone phone era and computers and all that. So we're not normally pulling out pieces of paper. So how does that translate into our ability to use this?
Michael Kattan, PhD: Yeah, it's kind interesting because and I'm guilty of it too. The, the nomogram term, strictly speaking applies to the visual on the piece of paper. But it gets kind of morphed into basically a kind of a comprehensive prediction model and equation, which is underneath the hood of the thing. So you could take the equation and show it on a sheet of paper. You could also make a smartphone app, you could put it directly into your electronic health record system. And so, it just automatically runs. And so we're always looking for ways to make you more efficient. I'm trying to dive into areas where you folks struggle with decisions, those are the fun ones for me. If you're in a no brainer situation where you're going to treat this patient, like so and so, and you don't really need a prediction to guide you on that.
You probably don't need a nomogram very much for that, but ones where it's controversial, I should say. And you don't know whether to do A or B and there's counterarguments and all this stuff. That's where I like to give you tools to take the prediction part out of it. So that, you're not basically fighting about predictions. You're more fighting about the outcomes that are at stake and how important they are, and how much the patient cares about them and involving the patient and that, and just get the math off of your plate. So you don't have to mess with that.
Dale Shepard, MD, PhD: I will come clean and I still fondly remember getting the piece of paper and the ruler and actually doing it old school, and I kind of miss that somehow.
Michael Kattan, PhD: Yeah. It's fun.
Rahul Tendulkar, MD: I really appreciate that all the risk calculators are on online, right?
Michael Kattan, PhD: Yeah. So we have a free website. It's RCALC, that's rcalc.ccf.org. All the risk calculators that I've ever worked on are up there and they're free. They're not always pretty, they're meant for a clinician audience, so they don't hold your hand. They're meant for you, who knows exactly what you're trying to find and because they're not as patient friendly as the ones I had made before. But they're up there, they're free.
Rahul Tendulkar, MD: And I've been using these for years in the clinic, specifically for these exact situations where, there's controversy about how to treat somebody and we say, "okay, look, if we just treat you with radiation, for example, this is the odds of success." If we add hormonal therapy to the mix, that might increase by this percent, and it's not a 100%. With all due respect, it's not a 100% accurate, but it's going to give us in the right ballpark, right? In terms of orders of magnitude, in terms of how much difference a treatment intensification can make.
Dale Shepard, MD, PhD: Patients truly do appreciate that sort of more granular approach to risk.
Michael Kattan, PhD: Yeah. I was desperate for it and I couldn't get that when I had to make my decision. You can't say, "we'll come back and we'll do more research." It's like, "well, no, I think I have to get treated now if I'm going to do this." And so, we need more of these. And that's what I want to know is, what's missing from the website. What tools would you guys like to have out there that we don't have yet? And I rely every one of these is coauthored with some clinician here or elsewhere and who's driving the thing I would, "Mike, it would really help me if you could predict such and such type of an outcome in a patient with so and so treated with such and such." I need to know what those are to help you.
Dale Shepard, MD, PhD: Got you. Now, just from a very practical standpoint. When you go to build the nomograms, oftentimes you're trying to answer questions that we really don't know what the standard is, and we're trying to, that's maybe one of the questions. Where do you draw upon the data?
Michael Kattan, PhD: Yeah. So a lot of the data we poll from here, it's a high volume institution and I can get my hands on it very quickly. And so, that's pretty low hanging fruit if we have the patient volume here. But otherwise, if you have a common goal in mind where you're trying to predict an important outcome and I don't have an axe to grind here, because I don't treat patients any particular way. I don't really care what the answer is. And people are pretty generous in terms of sending data sets and things like that and collaborating and it's surprisingly easy. And that's one of the things that led me to here was the low friction to collaborate with Cleveland Clinic, it was just striking.
Dale Shepard, MD, PhD: I'm going to guess, part of what Mike was just talking about sort of these unanswered questions. Are there things where, well, you've seen clinical utility nomogram and are there still areas you're like, if only I knew and patients ask me. Are there examples you can think of where a nomogram might, or setting mic off to get busy here?
Rahul Tendulkar, MD: Oh, absolutely.
Dale Shepard, MD, PhD: What are some of unmet needs in nomograms in prostate cancer?
Rahul Tendulkar, MD: So, in my personal areas of prostate cancer and breast cancer, these are two that are very common in the community, in our population. You know, 1 out of every, 6 to 8 men or women will get one of these diagnoses in their lifetime. And so, we've really had a number of nomograms in the prostate cancer world, which have been really helpful over the years and how I think Mike, became famous and was recruited here to help establish our QHS program. In the breast cancer world, we recently published a nomogram. Sarah Sittenfeld, one of our former residents was the first author. Emily Zabor, one of the statisticians here, did a terrific job in women with node-positive breast cancer after mastectomy, which patients need radiation afterwards. And what's the magnitude of benefit because there's such again, a wide variability in the prognosis of these patients.
And so again, which patients need more aggressive treatment in which don't. So we poll data from five institutions across North America. And it's the largest data set published., I think in the literature to date on this subject. Specifically, this project was designed to answer that question that you pose, which is, how do we better estimate which patients are going to benefit from more aggressive treatments? So this just got published last month and the calculator's up and running. And so I hope people use it and find it useful.
Dale Shepard, MD, PhD: That's great. Oftentimes when people aren't using nomograms are trying to estimate based on a clinical stage, a pathologic stage, a risk group, something like that. Are there areas where those are more effective than nomograms? What kind of limitations are there with nomograms?
Michael Kattan, PhD: Well, I'll go first, because this is not going to be a popular answer, but I've done several comparisons of clinicians against tools that I've either made or selected and happy to say, I'm undefeated. Because that's the ultimate way to sell the tool is, because you'll get pushback because anyone can look at a tool and criticize it. Oh, it's missing the such and such predictor, which we all know is the best predictor out there and you don't have it in your tool. Or I see the nomogram, it gets this many points for PSA, that can't be right. This thing is fundamentally flawed. And so, but if you do a head-to-head comparison with clinical judgment on a data sheet of patients and I get your predictions and I go against the tool, any tool will uniformly, unfortunately, or fortunately win in terms of a predictive accuracy setting in the studies that I've seen.
Now, some of them been pretty close, but because it's a human being problem, it's just very hard for a human to take all these pieces of information and perfectly weigh them in our heads. And we have biases and it happens with attorneys and accountants and whoever it is in a situation where you need to predict like that. It's just very hard to do, so we fall on these very easy user friendly things to do like counting risk factors. How many, oh, the patient's got three risk factors. So that's worse than a patient with two risk factors, but is that necessarily high risk? You know, these are some problems with some other approaches that we have always fallen back on. And if you really want a predicted probability, then a model is really the only way to go.
Rahul Tendulkar, MD: Yeah. And you know, I think as oncologists, we're so used to staging patients and their cancers into certain bins based on these factors. And that's, I think useful when, for example, enrolling patients on a clinical trial, right. They should have a stage 3 to be eligible for this trial. And so, I think useful in that regard to kind of been similar patients together for those purposes. But staging is otherwise, not as helpful for individual prognosis and estimations and things like that. And interestingly, the latest version of the breast cancer staging manual is like four pages for breast cancer staging. It's no longer something that you could memorize based on their TNM status and ER/PR and HER-2 receptor status. And so, it's almost like they've created a four page nomogram, so like board recertification easier.
Michael Kattan, PhD: Right, exactly.
Dale Shepard, MD, PhD: So, I think that's a great example of where, I think a nomogram would probably be a more useful tool, than trying to memorize a four page staging section.
Michael Kattan, PhD: It's the staging system that really just pushed me over the edge because when I was a 4 B, I remember vividly in the waiting room at MD Anderson and I was next to this older lady who I was sitting in the waiting room with, and she was also a stage 4 B Hodgkin's disease patient. Now she was every bit of 90, confined to a wheelchair on an oxygen tank and I'm playing basketball every day. I just sweat at night, that's all I got going on. And we had the same oncologist. So I talked to him about it and I said, "I was out there with," we'll call her Betty. "And you told me to look at the staging at the Merck Manual for my prognosis last time. And so you tell Betty the same thing you tell me, do we have the same chances of five year survival? Is that what I'm led to understand?" Then there's lots of back peddling and hand gesturing. And that's when I had heard enough.
Dale Shepard, MD, PhD: Interesting. Well, clearly nomograms are helpful, I mean, your track record speaks for itself. Has revolutionized how we treat prostate cancer, breast cancer, many other diseases, what's next? Is there a nomogram plus, is there another kind of tool we can do? Other things we can incorporate we're not currently doing?
Michael Kattan, PhD: So you like the AJCC is getting more friendly with allowing prediction tools in the staging manual and there's this way to get kind of certification and has to meet certain characteristics. And I thought that was a big step to being able to open the door to these models that could creep into more directly, I think into patient care. Because I think if you step back and say, "well, what is the goal of a staging system in the first place?" And I actually don't know the answer to this, but I feel like it might be to predict, I don't know about that. But if it is to predict then, why don't we predict as accurately as we can, if that's really the goal. Is that how you should be doing? I know we've always done it that way.
So, where's this going to go? I think ideally, what I would love for you folks to have is just built into your EMR tools that run behind the scenes without any extra work from you. You may or may not use the predictions for every patient or whatever, but if you need the prediction, you just look over there and you say, "okay, well, that's good to know." And I have these things automated. There's not the extra work, and we're always giving you the most accurate predictions we can for what we think is going to happen as a consequence of what you select for the patient. And so, I would really like to see us get further along the lines. I think to date as many of these paper based nomograms as I've worked on, I have a three that are running an epic right now, and it's a lot of work. But it's technically possible, but it's not a trivial matter. So it'd be nice if the barrier to entry could come down over time.
Dale Shepard, MD, PhD: And I guess, just from a logistic standpoint, I like Rahul, you use these, you treat patients that these are important. So much, like you said, you can't necessarily, in your mind, think of all these factors, but you're familiar enough with the nomograms. You don't have to, because you've seen enough patients and you are familiar with the nomogram predictions that pretty accurate. You don't necessarily need to open the nomogram for every patient, because you've kind of built in kind of what it's going to be telling you for most patients.
Rahul Tendulkar, MD: Yeah. I mean, I think that every patient has sort of different philosophies and how much information they want to know. And for those who want really detailed individualized estimates, you got to pull up the nomogram, go to the website and it's very helpful to do that. And so, I think integrating them more into the EMR can be really helpful. So you don't have to pull it up separately or something like that. But, I think the way that the website is designed right now is really helpful. And I use it all the time.
Michael Kattan, PhD: We've started to add a few more features in where with some kind of emerging novel tests that you might not order on every patient, or they might, whatever reason you can't get them to show you what the prediction might be, if you get that test. So without the test, here's the predictive probability of such and such outcome for this patient. If you go and get this test, it might change it to this or to that, and what's the range. And then you can look at that and say, well, if it's really just bumping in a little bit and it's within the same area, then I don't really need to know whether it's 40 or 51. But if I order the test and it drops it to 5 or raises it to 92, that's interesting information. And that's the model based, that's not the only... That's after factoring into the model, and that's a whopping thing. You might want to go chase that down and see if we can't get that.
Dale Shepard, MD, PhD: And I guess the way we think about a lot of tumors is changed dramatically. And I guess that might lead to a need to think about nomograms as well. So you think about lung cancer, for instance, it's no longer non-small cell lung cancer. Now, it's histology and genomic based. And so, it's really like a collection of rare diseases, almost.
Michael Kattan, PhD: Interesting.
Dale Shepard, MD, PhD: And I'm guessing that some of those things could be worked into nomograms for better predicted power.
Rahul Tendulkar, MD: I think as we learn more about genomics, we're going to keep this guy busy over the years because heretofore, all of the nomograms we have are based largely on clinical or pathologic factors. Then, if you can plug in a genomic classifier score, for example, in addition to the clinical features that we already know about, that's just going to improve our ability to prognosticate, I think.
Dale Shepard, MD, PhD: Well, great insights on nomograms and how we can use them clinically guys, appreciate you being with us.
Michael Kattan, PhD: Thanks.
Rahul Tendulkar, MD: Thanks for having us.
Dale Shepard, MD, PhD: To make a direct online referral to our Taussig Cancer Institute, complete our online cancer patient referral form by visiting clevelandclinic.org/cancerpatientreferrals. You will receive confirmation once the appointment is scheduled. This concludes this episode of Cancer Advances. You will find additional podcast episodes on our website, clevelandclinic.org/canceradvancespodcast. Subscribe to the podcast on iTunes, Google play, Spotify, SoundCloud, or wherever you listen to podcasts. And don't forget, you can access real-time updates from Cleveland Clinic's cancer center experts on our Consult QD website at consultqd.clevelandclinic.org/cancer. Thank you for listening, please join us again soon.