cT1N0 breast cancer. RNI?

[scarbrtj]

finally, I’ve addressed the lack of a dmfs benefit on long term follow up of eortc multiple times on this thread yet you continue to bring it up. More non breast cancer events. That’s why breast cancer mortality was still lower in the rni group on long term follow up. The study was powered at the initial publication date. It wasn’t powered to see a difference beyond that time point. It’s important to power a study at a particular time point because that drives the number of patients you enroll. Beyond that time point you don’t take into account the event frequency and the important competing risks!

Almost exactly what I'm saying. Suppose the following situation. Let's assume you powered a study to show a 10% risk reduction for a particular outcome at 5 years. You assumed the risk of that otucome without intervention was 50%. So you want to show that it improves to 45%. And for ease lets say that's a 200 patient study (each arm). Now you conduct th4e study and at 5 years you find that in the control arm the risk of the outcome was 80% without intervention and 82 with intervention. No way is a 2 patient differnece going to be statistically significant but the 10% risk reduction is there. This is probably why CALGB was postiive for paclitaxel and NSABP was not. This is also why if you enroll a low risk population on study or survival improves for other reasons (such as lead time bias), an intervention may be successful at the rate you expect, but the study will still be negative.

Suppose the second situation. Let's take a patient population of 75 year old Females with some disease. We want to see if an intervention reduces a particular outcome X or death at 5 years. Let's take the same numbers 50% risk of death or X and 100 patients each arm. The 5 year study shows that there is a benefit. However in 10 years 80% of the 75 year olds are deadin both arms. There will be no benefit at 10 years.

Powering a study has to be taken into understanding the outcomes of a trial.

You can ignore the graph, but all these points are shown in the graph...

1. Re: power. With 15y vs 10y median folowup, there are more patients at risk and thus (presumably) more power at the 10 year and 15 year time points. For example, at the 10 year time point, with 10y median f/u, there were 890 patients at risk in the control group. With 15y median f/u at the 10y time point, there were many more at risk thus evaluable: 1298. So a 10y DMFS estimate is "more robust" (and would always have "tighter" 95% CI's) w/ 15y median f/u vs 10y median f/u...
2. EORTC w/ 10y followup showed a 10y DMFS benefit. Now w/ 15y fu, the 10y timepoints are actually slightly higher and the deltas are exactly the same: 3% at 10y. This is data maturity. To reiterate #1: we can now make better assessment of the 10y DMFS with 15y median followup than we could at 10y median followup.
3. So why did EORTC at 15y f/u lose significance vs DMFS result at 10y f/u? "...I’ve addressed the lack of a dmfs benefit on long term follow up of eortc... More non breast cancer events." It may or may not be because of excess non-breast cancer deaths (the MA.20 only looked at mets & breast cancer related deaths in its DMFS analysis FWIW), but this is not hinted at in the results section (nor in the curves themselves, see #4). And you'd have to see a "speeding up" of the hazard function in one curve vs a "slowing down" in the other. It depends on how the hazard functions behave over time...
4. Here's making the specific point. With ~10y median f/u, the 15y DMFS Kaplan-Meier estimate was 68.8% vs 62.7% (see the green text). This is a delta of 6.1%, and even though there's only 10y median followup this large delta in the curve contributes to the p=0.02 when looking at the OVERALL DMFS difference. Then, with 15y median f/u, the 15y DMFS estimate was better in both arms: 68.8%→70.0% and 62.7%→68.2%. So even with the "risk" of excess non-breast cancer deaths as you say with more f/u, the DMFS rates GOT BETTER AT 10 AND 15 YEARS WITH MORE FOLLOWUP. (In fact 15 vs 10y f/u meant better DMFSs at every single time point... 1y, 2y, etc.) But the 15y DMFS delta decreased from 5.9 to 1.2%, now the p=0.18 with more data maturity. It certainly appears to be there is more of a "curve cross-over" later in time... w/ better DMFS w/o RNI, in the 15y median f/u data. As you know, survival analysis "values the whole curve and not isolated points." This is why e.g. in EORTC 2015 the 10y DMFS was 78.0%(76.1-79.8) vs 75.0%(73-77); there is substantial 95% CI overlap here AT THIS TIME POINT but the p=0.02 values the whole curve. The curve evolves with longer followup times (or crossing censors over into events with better data collection; ~10% of the EORTC patients were censored/lost to followup).

EORTC 22922 shows no significant DMFS (defined as mets or any death) benefit w/ RNI. Full statistical stop. The median time of f/u of the sample can't be blamed because of the difference not being statistically significant.

---

Members don't see this ad.

 

[gmsquid]

The curves do overlap but that’s also misleading. One of the assumptions of a Kaplan Meier log rank analysis is that the rate of events is constant through the entire analysis (wiki it if you don’t believe me). The rates of death
increase at the end of the curve (this is common sense). So your using a power analysis designed for 10 years and an event rate at 10 years to design patient numbers and then making conclusions on a 15 year curve. It’s going to be inadequately powered by the known fact that more people die over time. You would need more patients to make a conclusion at 15 years.

From the initial study:
Distant disease-free survival was calculated from the date of randomization to the first date of distant disease or death from any cause, whichever oc- curred first.

From the abstract of the long term follow up:
Conclusions: The 15-years results show a significant reduction of breast cancer mortality and breast cancer recurrence by internal mammary and medial supraclavicular lymph node irradiation in stage I-III breast cancer. However, this is not converted in improved overall survival without a clear explanation for this. Subgroup analyses and continued follow-up will be performed to better define patients that may benefit from this treatment and define the causes of death.”

So RNI reduces recurrence and breast cancer mortality but it does not turn into an OS benefit is their conclusion. So there are more non breast cancer events, but you can’t make a conclusion about overal survival or dmfs at 15 years because of inadequate power to do so. You don’t have a power analysis driven by the risk of late events.

 

[gmsquid]

Also from the paper:

Although overall disease-free survival was not significantly longer with IM-MS irradiation, breast cancer recurrence at 15 years was significantly lower. The fact that this does not translate into longer overall survival might be explained by non-breast cancer-related deaths or those from an unknown cause, which together constituted 38% of all deaths, an imbalance in missing data on the cause of death between treatment groups, and salvage treatment after recurrence. Finally, the overall outcomes were better than expected, resulting in smaller absolute differences than expected based on historical results, which can be explained by a high proportion of node-negative patients who were eligible on the basis of a medially located primary tumour only.

 

[scarbrtj]

but you can’t make a conclusion about overal survival or dmfs at 15 years because of inadequate power to do so.

Then your problem is not with "me" per se. You should definitely should write a letter to the editor of Lancet Oncology that the EORTC's DMFS assessments (because the "at 15 years" has nothing to do with the p=0.18; the p=0.18 references all time points from randomization) are inadequately powered. Because in the meantime there's the risk that statistical dummies such as myself will keep quoting/referencing their results. And in 2025 they will publish 20y results.

To provide further provocation (triggering?)...
1) Past 15y, the OS of non-RNI patients is (insignificantly) better in EORTC 22922 (this was also shown in the NEJM 2015 EORTC study past 10y).
2) Prediction: with more follow-up in MA.20 (if they ever provide updated results), the OS curves will cross favoring non-RNI.
3) You don't have to be a statistician to know that if curves cross, even if it's on the ass-end of the curve, it's never a good thing.

 

[gmsquid]

Then your problem is not with "me" per se. You should definitely should write a letter to the editor of Lancet Oncology that the EORTC's DMFS assessments (because the "at 15 years" has nothing to do with the p=0.18; the p=0.18 references all time points from randomization) are inadequately powered. Because in the meantime there's the risk that statistical dummies such as myself will keep quoting/referencing their results.

To provide further provocation (triggering?)...
1) Past 15y, the OS of non-RNI patients is (insignificantly) better in EORTC 22922 (this was also shown in the NEJM 2015 EORTC study past 10y).
2) Prediction: with more follow-up in MA.20 (if they ever provide updated results), the OS curves will cross favoring non-RNI.
3) You don't have to be a statistician to know that if curves cross, even if it's on the ass-end of the curve, it's never a good thing.

I would but The authors clearly state that they were underpowered: “The main remaining question is how the 15-year analysis of the EORTC trial can help to clarify the uncertainty around the topic of regional lymph node irradiation. The primary endpoint, overall survival, was not met, probably because our trial was underpowered to detect small absolute differences, as outcomes were much better than anticipated following inclusion of a relatively favourable patient group compared with the other trials. However, the significant reduction of breast cancer mortality and breast cancer recurrence demonstrates that a benefit exists with IM-MS irradiation for at least a subgroup of patients, which might be, however, at least partially offset by an increase in late non-breast cancer- related mortality. “

And of course curves frequently cross at the tail end as patients have inadequate follow up or the number at risk is low. Only a few dozen patients in each arm were evaluated at 20 years. It only takes a few patients to make curves cross like that. The final numbers may be different. What do you think happened in Michigan and Georgia this year?

 

[scarbrtj]

I would but The authors clearly state that they were underpowered: “The main remaining question is how the 15-year analysis of the EORTC trial can help to clarify the uncertainty around the topic of regional lymph node irradiation. The primary endpoint, overall survival, was not met, probably because our trial was underpowered to detect small absolute differences, as outcomes were much better than anticipated following inclusion of a relatively favourable patient group compared with the other trials. However, the significant reduction of breast cancer mortality and breast cancer recurrence demonstrates that a benefit exists with IM-MS irradiation for at least a subgroup of patients, which might be, however, at least partially offset by an increase in late non-breast cancer- related mortality. “

And of course curves frequently cross at the tail end as patients have inadequate follow up or the number at risk is low. Only a few dozen patients in each arm were evaluated at 20 years. It only takes a few patients to make curves cross like that. The final numbers may be different. What do you think happened in Michigan and Georgia this year?

Any time someone doesn't get a statistical result they want/expect , I see ~100% of the time in the discussions "our trial was probably underpowered." Well, yeah, that or a true difference doesn't exist. In this 2020 paper, they say in 2025 they will waste everyone's time again with updated 20 year results. We won't get Michigan and Georgia recounts from 2020 in 2025, but the EORTC is going to give us a recount then. I'm calling all the results valid... Michigan, Georgia, and the EORTC 2020's! We can reconvene in 5 years to re-argue.

 

[gmsquid]

Any time someone doesn't get a statistical result they want/expect , I see ~100% of the time in the discussions "our trial was probably underpowered." Well, yeah, that or a true difference doesn't exist. In this 2020 paper, they say in 2025 they will waste everyone's time again with updated 20 year results. We won't get Michigan and Georgia recounts from 2020 in 2025, but the EORTC is going to give us a recount then. I'm calling all the results valid... Michigan, Georgia, and the EORTC 2020's! We can reconvene in 5 years to re-argue.

Well the risk reduction was there, but if your initial assumption was wrong on your baseline survival rates you will be underpowered.

 

[scarbrtj]

Well the risk reduction was there, but if your initial assumption was wrong on your baseline survival rates you will be underpowered.

Whaddaya mean by "your" lol. Assumptions are bad, and bad assumptions are even worse. RNI is the Donald Trump of breast cancer. There REALLY IS an OS difference there; our assumptions have just been rigged and the statistical tests stole all our low p-values.

 

[gmsquid]

Whaddaya mean by "your" lol. Assumptions are bad, and bad assumptions are even worse. RNI is the Donald Trump of breast cancer. There REALLY IS an OS difference there; our assumptions have just been rigged and the statistical tests stole all our low p-values.

Do you know how to power a survival analysis? One has to assume a baseline survival of the enrolled population and the proportional improvement with an intervention. Then there will be an X% (usually 80 is chosen) to correctly identify the difference in survival outcomes when it actually exists. If the actual enrolled baseline survival is higher than anticipated, it is less than X% likely that there will be enough patients to demonstrate the proportional improvement with the intervention. . Depending how much larger that baseline survival is the less likely to detect the assumed difference. In EORTC baseline survival was assumed 75%. It was 5% higher than that. Now that may not seem like much but when the desired improvement with intervention is 4% that’s a lot.

 

[gmsquid]

I’m not saying rni isn’t toxic, particularly for left sided tumors and especially with old techniques. But this data is compelling too: https://ascopubs.org/doi/10.1200/JC...=ori:rid:crossref.org&rfr_dat=cr_pub 0pubmed

 

[communitydoc13]

Technically/on paper, this is not prevailingly true in breast cancer. Forty years ago Fisher wrote "Breast cancer is a systemic disease at diagnosis. The metastatic phenotype is either present or absent but is not acquired over time." I realize this notion doesn't "sit well" with rad oncs. Were it really to have obtained purchase in rad oncs' noggins, trials like MA.20 and EORTC 22922, and Jagsi et al's analysis of Z0011, wouldn't have happened.

Just witnessed the Socratic method par excellence here with very good points made by both Scarb and Squid. I'm definitely on Squid's side here. Both MA.20 and EORTC 22922 were XRT vs XRT trials and I was impressed that there was a measurable change in efficacy demonstrated by these trials. The degradation of impact on survival and DFS at later updates is typical of survival analysis and is mostly a function of competing risk and decreased evaluable patients with time IMO.

The power assumptions regarding survival were widely off (as is typical of XRT trials, the ongoing prostate nodal XRT trial being a prime example). For MA.20 they powered for a 5% (absolute survival) benefit in a population that demonstrated an ~10% breast cancer mortality at 10 years. This was never going to be positive for OS long term.

Taken together with the Danish naturally selected population based study, I have very little doubt that by targeting regional nodes, you are improving oncologic outcomes as a whole for many node positive and a small subgroup of node negative patients. (I wish I had better data to justify RNI in T2N0 triple negative patients; I'm convinced it's the right thing to do, and it's a shame that the Chinese early TNBC trial was retracted.) There is a cost to RNI of course. I guess Scarb and I just view the relative cost/benefit differently.

That RNI would impact distant metastases free survival is not this: https://www.apa.org/pubs/journals/features/psp-a0021524.pdf. While I agree that the Halstedian hypothesis is antiquated, a simplistic view of the "alternative" Fisherian hypothesis is just as damaging and can become itself a non-helpful dogma.

To my knowledge, radoncs have not led the way in refining the model of metastatic progression. I believe that our access to molecular imaging and small animal radiation models should make this a real push in terms of radonc translational research. (Maybe it is? If so, please let me know about it.)

I remember a cancer evolution talk from around 2010 where the speaker noted that they could model much of cancer behavior applying simple Darwinian models of selection to environmental pressure with the exception of metastatic behavior. In principle, the cancer cell that leaves the home is more vulnerable to the environment (immune system) and less likely to propagate.

Origins of Metastatic Traits

How cancer cells acquire the competence to colonize distant organs remains a central question in cancer biology. Tumors can release large numbers of cancer cells into the circulation, but only a small proportion of these cells survive on infiltrating ...

www.ncbi.nlm.nih.gov

Around the same time tumor self-seeding by metastatic cell lines Tumor self-seeding by circulating cancer cells was being demonstrated. (Same group, MSKCC). This group seems to me to be at the lead in terms of studying metastases, including cool stuff like the impacts of trophism on mets and the genetic basis of latency competent metastatic cells (you know those cells in the bone marrow (among others) that may manifest years later as clinically evident metastatic disease).

It is highly conjectural, but not crazy, to imagine that viable regional (or distant) metastatic niches (e.g. microscopically involved lymph nodes) might reduce the evolutionary pressure on circulating tumor cells and facilitate metastatic progression to other sites in the future. (Cells have a home to come back to, can be adapted for transit without being viable distant colonizers and can evolve in a more stepwise fashion to true metastatic monsters than if no home base were present.)

 

[scarbrtj]

Do you know how to power a survival analysis

Yes! Initiate a multinational trial with four thousand patients and follow them a median of a decade or more.

Well darn. See you after the meta-analysis. They’d just better get to it quick because Honey I Shrunk The Survival Advantages.

I’m incorrigible.

 

[drewdog1973]

Yes! Initiate a multinational trial with four thousand patients and follow them a median of a decade or more.

Well darn. See you after the meta-analysis. They’d just better get to it quick because Honey I Shrunk The Survival Advantages.

I’m incorrigible.

Ha! I read that study and think ... “man, if they did the radiation just a wee bit better with respect to the heart, they’d have had a survival benefit. Good enough for me!”

 

[scarbrtj]

Ha! I read that study and think ... “man, if they did the radiation just a wee bit better with respect to the heart, they’d have had a survival benefit. Good enough for me!”

Hope is the last thing that dies in man; and though it be exceedingly deceitful it is of this good use to us that while we are traveling through life it conducts us in an easier and more pleasant way to our journey's end.​

François de la Rochefoucauld

 

[scarbrtj]

Haven't seen any mention of (or honestly really thought much about) thyroid in the breast RNI arena...

 

[Palex80]

Haven't seen any mention of (or honestly really thought much about) thyroid in the breast RNI arena...

This used to be an issue in the old days. When I started training, breast cancer patients were planned without a CT and the cranial border of an RNI field for the SC-fossa was at the hyoid bone!
Modern CTVs for breast cancer RNI end substantially lower and even with a large PTV-margin, the thyroid is pretty much kept out.

On the other hand: YOU DONT NEED TO WORRY ABOUT HYPOTHYROIDISM IF YOUR LETHAL BREAST CANCER IS BACK!
The excessive risk pointed out at that graph is pretty much the same as the DFS improvement with RNI.
Just trolling, sorry...

 

[Ray D. Ayshun]

This used to be an issue in the old days. When I started training, breast cancer patients were planned without a CT and the cranial border of an RNI field for the SC-fossa was at the hyoid bone!
Modern CTVs for breast cancer RNI end substantially lower and even with a large PTV-margin, the thyroid is pretty much kept out.

On the other hand: YOU DONT NEED TO WORRY ABOUT HYPOTHYROIDISM IF YOUR LETHAL BREAST CANCER IS BACK!
The excessive risk pointed out at that graph is pretty much the same as the DFS improvement with RNI.
Just trolling, sorry...

It's been my experience that another reason to not worry about hypothyroidism is synthroid.

 

[scarbrtj]

It's been my experience that another reason to not worry about hypothyroidism is synthroid.

Well, I alluded to hypothyrodism being associated w/ incr mortality. And Hillary Clinton takes armour thyroid!

 

[Ray D. Ayshun]

Well, I alluded to hypothyrodism being associated w/ incr mortality. And Hillary Clinton takes armour thyroid!

You and your reading. There's a doc I used to know of that basically diagnosed everyone with hypothyroidism and lyme disease. He gave armour. People loved him.