“We were actually quite pleased with how well our model performed, particularly in the independent cohort that was separate from our training cohort,” says Eric Li, MD.
In this video, Eric Li, MD, and Rashid Siddiqui, MD, discuss notable findings from the recent Prostate Cancer and Prostatic Diseases paper, “Optimizing detection of clinically significant prostate cancer through nomograms incorporating MRI, clinical features, and advanced serum biomarkers in biopsy naïve men.” Siddiqui and Li are urology residents at Northwestern University Feinberg School of Medicine in Chicago, Illinois.
Li: First, we can talk about how we developed the nomograms. As I said, we focused on the biopsy-naive setting, and we focused on people who had initial PSAs of 2 [ng/mL] to 20 [ng/mL]. Our training cohort was about 1494 patients, and we had a separate independent validation cohort of 366 patients. The main findings in generating our models were age, Black race, PSA density, MRI, and percent free for prostate health index. And so we also developed different versions of nomograms, depending on what serum biomarker was available. As we said, we primarily use prostate health index at Northwestern. But not everyone has access to that. So we also made different versions of the nomogram with percent-free PSA, which is another commonly used biomarker, as well as just total PSA. And our outcomes were also looking at clinically significant prostate cancer, which is defined as grade group 2 or higher, and then also higher degree cancer, which is the defined as grade group 3 or higher. And we did this mainly just because providers and patients have various biopsy thresholds. And so, by stratifying the outcomes more, we're giving them more information to decide whether they want to proceed with biopsy. So we had a total of 6 nomograms, with the previously mentioned variables. And if you look at the performance of our various models, in both the training and validation cohort, you see our area under the curve ranged from .885 to .923. And it's fairly consistent across all the models and also in the different cohorts. To look at the clinical relevance of this, we looked at our various models, looking at different biopsy thresholds. For example, if you look at the grade group 2 model with prostate health index, in our training cohort, you see that you apply a biopsy threshold 20%. So, in other words, if the output of our model is 20% or higher, you biopsy, we save 49% of biopsies, while missing 9.7% of clinically significant prostate cancer in the training cohort. One of the surprising findings was that when we actually looked at this in the validation cohort, again, with a grade group 2 model with prostate health index, at a biopsy threshold of 20%, we save 39% of biopsies, but we really missed only 1 case of clinically significant prostate cancer. So I think we were actually quite pleased with how well our model performed, particularly in the independent cohort that was separate from our training cohort.
Siddiqui: Just backtracking a little bit, kudos to our radiologists, our pathologists, and our urologists, basically all of them who are doing the biopsies. I would say for this nomogram to work well, not just at Northwestern, but beyond, if you look at a nomogram, the PI-RADS score gives you the most number on the nomogram. And, you get a good PI-RADS score by having a good radiologist who is GU trainedlooking at the MRI. I would say our nomogram, unfortunately, will fall apart if you don't have a good MRI read. When we actually started our project, we just wanted to look at how we are doing in our system over the past several years in terms of, we get these MRIs, we get these biopsies, our biopsies as good as what's been reported in clinical trials and it turns out actually, our numbers are highly concordant with what's been reported in the literature. Our PI-RADS 5 is over 80% grade group 2 or higher disease on biopsies. Our PI-RADS 4 is about 50%, which is what's been reported in big trials. Our PI-RADS 3 is about 20% of all PI-RADS 3 that get biopsied have clinically significant prostate cancer. So I think it all starts there. We did 2 different methods of using our training cohort of 1494 patients to help us be reassured. But then we were really pleasantly surprised when we looked at our independent cohort. Our training cohort went from March of 2018 through June of 2021. And our independent cohort is a prospective group of patients who were seen at Northwestern medicine from July of 2021 through February 2022. So we were really pleased that, when we looked at the performance of our nomogram, it really performed well and we only missed 1 patient with cancer, at a threshold of about 20%, which means we can avoid up to 40% of biopsies and sleep well at night.
This transcription was edited for clarity.