Read the Full Video Transcript

Rashid Sayyid: Hello everyone, this is Rashid Sayyid. I'm a urologic oncology fellow at the University of Toronto. And along with Zach Klaassen associate professor of the Department of Urology, we'll be discussing the recent publication from JCO looking at the external validation of a prognostic model of overall survival in men with chemotherapy naive mCRPC. This study was recently published by Dr. Susan Halabi in the Journal of Clinical Oncology.

So we know that mCRPC patients have poor prognosis. And if we look at data from randomized trials, you see that the median overall survival is about three years based on data from the abiraterone and enzalutamide trials in the first line setting. And if we look at the real world data, this doesn't... This looks even worse. At about two years opposed to three in the clinical trial setting.

We currently have numerous agents approved in the first line setting for these patients, including docetaxel. The androgen receptor signaling inhibitors, for example abiraterone and enzalutamide. We also have radium 223 for bone only mets based on data from the ALSYMPCA trial. And more recently and encouragingly for our patients, we have had the combination of olaparib plus abiraterone approved in BRCA1/2 mutated patients based on results from the PROpel trial, and the talazoparib and enzalutamide combination in the homologous recombination repair mutated patients based on results from the TALAPRO trial.

And so a lot of these differences in the prognosis of mCRPC patients aren't just only due to the therapies that patients receive. There are some disease specific factors or what we call the tumor specific factors. These can be quite variable obviously. We also have host specific factors such as your age, the race, performance status. Obviously the younger you are, the better performance status. The better you are likely going to do. And also what prior therapies have you received. And so it's important to look past just the therapies that we give our patients and also acknowledge the differences in outcomes may be due to underlying prognostic factors rather than active treatments administered.

Back in 2014, Dr. Halabi and all have developed and validated a prognostic model for overall survival in mCRPC men who were treated with first line docetaxel randomly assigned to the CALGB 90401 trial. And this prognostic model for overall survival include a lot of common clinical vares. ECOG performance status, your disease site, LDH levels, whether you used opioid analgesics or not, your albumin level, hemoglobin, PSA and alkaline phosphatase levels. And these are commonly available variables, and so they negate the need for expensive advanced genetic testing. And how can we use the clinical data that's readily available to us to better inform our patient's prognosis?

And so based on the score that was generated from this nomogram, patients were split into either low or high risk groups. And the median overall survival in the validation set was just over two years for the lowest group and just over a year for high risk group. So we see that the survival at best is only two years, which fits the data that we see in the real world setting. And this model performed well, with an area under the curve in the validation set of 0.76. And when we consider that 0.5 is essentially flipping a coin, which is the null effect, we see that 0.76 is quite good in this setting.

And so why are we doing this study again? Well, this model has not been validated in the docetaxel naive patient. And so we need to increase the external generalizability and validity of this study for these patients. And so the study objective was to validate the prognostic model risk scores and the risk groups generated by these risk scores in an external data set of doce naive patients who were subsequently treated with doce or tasquinimod or a first line potent androgen receptor inhibitor therapy. They also wanted to validate this model subgroups of patients based on your age, your race, and the class of treatments you had previously received.

They wanted to validate the two and three risk prognostic groups. So based on your score, they either employed... Cut-off to do two. To split into two groups. Or they employed to cut-off split into three risk groups. And they want to see how this model performed based on these cut-offs. They also wanted to adopt a more pragmatic approach whereby you use the number of prognostic risk groups that you have to generate the risk grouping as opposed to just having to plug in the numbers and calculate risk scores.

So you have a patient in clinic that you see. You know that there are eight risk factors that have been validated in the setting. You click one or two buttons, you see that they have one or two, and based on that you generate a risk group as opposed to having to put all of them into that calculator. So that's what they tried to also look at in this study.

And so this table summarizes the clinical trials that were included in the external validation set. And so if we look at the left hand column, you have the name of the trials. We won't go through them one by one. But also important in the right hand column. If we look at the treatment arm column, we see that these patients have received a combination of docetaxel plus or minus another agent, or they received a tasquinimod versus placebo, or they received the androgen receptor signaling inhibitors such as enzalutamide with or without abiraterone or apalutamide plus or minus abiraterone in the ACIS trial.

So what was the patient population? So you had to have mCRPC and not received prior chemotherapy. Well you may ask, "Well, did they... Was it allowed to receive it in the MHSPC setting?" It's important to highlight that obviously with the CHAARTED trial and the STAMPEDE trial becoming available in the setting, this was later on. 2015, 2016. Whereas a lot of these patients we were accruing prior to that. So it was very rare for patients to have received docetaxel in the metastatic hormone sensitive setting. And prior abiraterone and enzalutamide use was also similarly not available due to similar rationale. Patients had a good performance status for the most part. ECOG performance status zero two. And they had to have adequate hematologic, hepatic, renal, and cardiac functions.

The primary endpoint for this analysis was overall survival and this was defined from the date of random assignment to death from any cause. And this analysis included eight predictors. And this was based on the adaptive LASSO trial, which helps you pick out which ones are the ones that affect the overall survival the most. And so these were opioid analgesic use, ECOG performance status, your albumin level, the disease site. So whether it was lymphatic only metastases versus bone mets with no visceral mets or visceral mets. Whether you had an lactate dehydrogenase LDH level greater than the upper limit of normal based on the local lab, the hemoglobin level, PSA, and alkaline phosphatase. So variables that are routinely available for clinicians in clinical practice. So this is what makes this a very attractive model, is that it's very easy to adopt in most settings.

It's important to note that with this analysis, if you have a missing value, then a score can't be calculated. So patients with any missing values were excluded from this analysis.

So how do they validate this model? So they estimated the parameters from the prognostic model were applied to each variable. And so you come up with a risk calculator. You plug in each value. And based on that, you generate a risk score for each patient in each data set. And then the predictive performance based on the score you generate was calculated using a time dependent area under the receiver operating characteristic curve, which is again very standard in this setting. And specifically evaluated the performance of the prognostic model in the following subgroups. So based on your age. Your race. Whether you were Asian, black or white. And the study drug mechanism of action. So docetaxel, tasquinimod, and first line AR therapy. And that's again to try to control the subgroup analysis for the different effects of different drugs. And just from a technical standpoint, they use the bootstrap approach to calculate the 95% confidence intervals for this time dependent area under the AUC curve.

So first of all, they validated the prognostic risk groups on the basis of the previously determined cut-off points for the risk score. So again, they calculated the risk score and based on the cut-off. And this is... They generate that trying to split the cohort into two equally sized groups. So if your risk score was less than -0.18, you're in the low risk for a worse survival outcome. And then if your score was greater than -0.18, you were in the higher risk group. So you're at higher risk for adverse events. So the lower risk group in theory should do better than the higher risk group. And using the same rationale with the different cut-offs, they also split the core into three risk groups. Low, intermediate, and high risk.

Next. As we discussed, they constructed prognostic risk groups based on the number of adverse factors that each patient had. And they did it in a smart way where they tried to approximate the risk group classification based on your risk score. So trying to mirror those two risk groupings.

And so they tried this in two different ways. So first they used the two risk groupings. So if you had less than three adverse factors. So zero, one, two, you're a low risk. If you have three or more, you're a high risk. And then kind of a similar rationale for the three risk grouping. Zero, one, or less than two for low risk. And then two to three for intermediate. And then greater than three for high risk. They employed survival analysis techniques for the Kaplan-Meier approach to estimate the overall survival by trial and the prognostic risk group. And comparisons were performed using the log-rank test, which is very standard in this setting. And they also very importantly stratified the analysis by the AR status trials. And that's because of the shift that we've seen in overall survival in the first line AR trials. Again, to account for the fact that differences in your prognosis may be related to the treatments that you received.

And at this point, I'm going to turn it over to Zach to go over the results and discussion for this study.

Zach Klaassen: Thanks so much, Rashid. So we'll go over the baseline characteristics for this study. And typically we see these in tabular form, but this is in a figure form. So we'll walk through the next three slides going over the baseline characteristics.

As we can see at the top, the most common race included in these studies was white. We look at the performance status. The majority of these patients were excellent performance status of zero or one. With regards to the most common metastatic site, this was bone only followed secondly by bone plus lymph node, followed third by visceral metastases.

Looking at opioid analgesic use, most patients did not need analgesics. And when we look at LDH greater than upper limit of normal, we see that the majority of patients had normal or lower than... Greater than one upper limit of normal for LDH.

When we look at these figures, this looks at age, LDH, PSA, alk phos, hemoglobin, and albumin. We see the median age is roughly 70 years of age. And the combine. In all of these trials combined, the median LDH is slightly over 200. The median alkaline phosphatase was just over 100. The median PSA was roughly 40 to 50. The median hemoglobin was slightly under 13. And the median albumin was slightly under 4.25 at baseline for these patients.

This is the Kaplan-Meier OS curves for the clinical trials, and we can see there's quite a range of overall survival among these trials. So this ranges from 18.3 months for the MAINSAIL trial, 22 months for the SYNERGY and TASQ trials, as well as 34.7 months for contemporary ARI therapy trials.

This is the forest plot of time dependent area under the curve. At the very bottom is the overall time dependent area under the curve, which for this study was 0.75. And if we look at these subgroups based on age, race, class of trial, risk groups and individual studies, the majority of these were all pretty close to 0.75, as you can see under the time dependent area of the curve numbers to the right of the figure.

This is the Kaplan-Meier overall survival curve by the two risk prognostic groups on the basis of number of adverse factors. This figure is specific to the first line non androgen receptor trials. We can see that low risk patients had a median overall survival of 27.6 months. High risk patients, 13.9 months. And the hazard ratio for death of high versus low risk patients was 2.8 with a 95% confidence interval of 2.6 to 3.0.

Similar looking figure for the two risk prognostic groups for first line androgen receptor trials for low risk patients. Median overall survival 38.8 months compared to high risk 18.9 months. Hazard ratio for death of high versus low risk patients was 2.6 with a 95% confidence interval of 2.3 to 3.0.

Again, looking at moving on to the three risk prognostic groups. This is for first-line non-AR trials. Low risk patients had a median OS of 32.5 months. Intermediate risk, 20.3 months. And high risk 11.6 months. So we see even more granularity in the three risk prognostic groups compared to the two risk prognostic groups. And the high risk. Or excuse me, the hazard ratio for death for high versus low risk patients was 4.7. And the hazard ratio for death for intermediate versus low risk patients was 2.2, both of which were statistically significant.

Looking at the three risk groups for the first line androgen receptor trials. Low risk, median overall survival of 43.3 months. Intermediate risk, 27.7 months. And high risk 15.4 months. The hazard ratio for death for high versus low risk patients was 4.3 and for intermediate versus low risk patients was 1.9. So again, we see granularity in these three risk prognostic groups, both for first-line non-AR as well as first line AR trials.

So this study externally validated a prognostic model of overall survival using data from seven phase three randomized trials in men with mCRPC. This is the first time that an OS model has been externally validated in patients who receive first line potent androgen receptor therapies. And the overall predictive measure was very close to what has previously been reported with a time dependent area under the curve of 0.75.

This study also validated two and three risk groups on the basis of the quantiles of the risk score and number of adverse prognostic factors. And despite longer overall survival in first line AR therapies, there was still a clear separation of the overall survival by prognostic risk groups. As such, prognostic risk groups should be used to select men with mCRPC for inclusion in future trials on the basis of number of adverse factors whereas random assignment can be stratified using either the two or three risk grouping. Furthermore, in the future hopefully combined clinical and genetic models of overall survival may further select precision medicine approaches for these patients.

So in conclusion, this prognostic model of overall survival has been extensively validated and is sufficiently robust to be used in patients with docetaxel naive mCRPC. Future models need to incorporate important clinical and genomic factors to better prognosticate and ideally predict the outcomes of specific therapies such as androgen receptor inhibitors or docetaxel chemotherapy.

We thank you very much for your attention, and we hope you enjoyed this UroToday Journal Club discussion.