--- id: BC-820 title: "Adjuvant and post-neoadjuvant therapy" subtitle: "Why the duration, the second antibody and the conjugate for residual disease are three separate decisions resting on three separate randomisations." status: drafted updated: 2026-09-26 ---
1 · Scope, and the two distinct decisions this chapter serves
This chapter serves two decisions that share a subtype and almost nothing else.
The second belongs to a patient who received neoadjuvant therapy and has residual invasive disease at surgery. That tumour has already met HER2-directed therapy and survived it. The selection variable here is observed rather than predicted. Post-neoadjuvant intensification for residual disease carries that decision, and Unresolved questions and what reads out next sets out what is still open in it.
Merging the two decisions is the commonest error in this subtype, and it is usually made in the favourable direction. The adjuvant trials randomised patients after definitive surgery, so their populations contain nobody whose response was known. The post-neoadjuvant trials enrolled only patients whose response was known and inadequate. A hazard ratio from one population does not transfer to the other.
Read the tables this way, since each is generated from the trial registry at build time. Each table answers the question of the section that carries it, so a trial recurs where it bears on more than one. The result and overall survival cells are taken from the publication named in the reference column, and a row and its citation therefore always agree. Longer follow-up, subgroup structure and cross-trial comparison are argued in the prose.
Two of the [Chapter Stories](stories.html) run through this chapter, one for each decision. Each section that carries a clinical question opens with it, written as what was being asked, what came back and what the design could not settle. The stories carry no figures at all, so a story and a table can never disagree about a number.
2 · The trastuzumab adjuvant trials and the duration question
Early disease · HER2+ · adjuvant
Matching adjuvant HER2 treatment to baseline risk
Does adding a year of trastuzumab to adjuvant chemotherapy lower recurrence?
Rationale
HER2 amplification marked the fastest-relapsing early breast cancer of the pre-antibody era. The antibody already worked against established metastatic disease. Micrometastatic disease left behind after surgery should be the easier target, being smaller and less heterogeneous than a measurable metastasis.
Experiment
Adjuvant programmes randomised trastuzumab against no trastuzumab on a chemotherapy backbone. They differ in whether the antibody ran concurrently with the taxane or began only after all chemotherapy. They differ again in whether the backbone contained an anthracycline, and in how long the antibody was continued.
Finding
Recurrence fell in every comparison. The separation was still open at the longest follow-up any of them reported. The anthracycline-free backbone gave disease control close to that of the anthracycline backbone, with less cardiac dysfunction.
Limitation
A year of antibody was a protocol convention rather than a tested dose, so these trials establish that the antibody works without establishing how much of it is needed. The trial that compared the antibody against observation allowed its control group to cross over once the first results were public, so its long-term survival comparison is diluted by design. The two trastuzumab backbones in the anthracycline comparison were never set against each other under a non-inferiority design. The absence of a significant difference between them is therefore an absence of evidence and not a demonstration of equivalence. The short-course signal that prompted everything downstream came from a small HER2-positive substudy nested inside a randomisation of two different chemotherapies.
Next question
If a year was chosen rather than found, can less than a year hold the same disease control?
NSABP B-31/N9831HERABCIRG-006FinHer
Can adjuvant trastuzumab be shortened below a year without losing disease control?
Rationale
The year was inherited from the pivotal protocols rather than derived from them. It carries cardiac risk, a long course of infusions, and a cost that decides access in much of the world. The nested short course suggested the antibody might do most of its work early.
Experiment
Two trials randomised six months of the antibody against the standard year. Two more randomised a nine-week course against it. Each set its own non-inferiority margin, and not all on the same scale. The pivotal observation trial had already tested the question from the other end by randomising a second year against the first.
Finding
A second year bought nothing over the first. Of the shortened comparisons, only one was declared non-inferior. Both nine-week comparisons pointed against the short course, one of them with a point estimate a long way from equivalence. Cardiac events were fewer on the shorter arms wherever they were counted.
Limitation
These results do not stack. Estimates that look alike did not answer alike. The two six-month trials returned closely similar hazard ratios and reached opposite verdicts. One set its margin as a difference in percentage points of disease-free survival at a fixed time point, and the other set its margin on the hazard ratio scale, so the verdict turns on the scale as much as on the estimate. The chemotherapy partners differ across these trials, as does the point at which the antibody started relative to them. None randomised duration inside a defined risk stratum, so failing non-inferiority in an unselected population says nothing about whether a small node-negative tumour needs the full year.
Next question
If the duration cannot be cut for everyone, can the chemotherapy be cut for the patients whose baseline risk is lowest?
HERAPERSEPHONEPHAREShort-HERSOLD
The trials that established adjuvant trastuzumab.
Each fixed its duration by protocol rather than by comparison.
trastuzumab for one year or for two years vs observation
disease-free survival
at a median follow-up of 11 years, one year against observation gave a disease-free survival HR of 0.76 (0.68-0.86) and 10-year rates of 69% against 63%; two years against one year gave an HR of 1.02 (0.89-1.17)
HR for death 0.74 (0.64-0.86) for one year against observation, with 52% of the observation group crossing over to trastuzumab
doxorubicin and cyclophosphamide then docetaxel with trastuzumab, or docetaxel and carboplatin with trastuzumab, vs doxorubicin and cyclophosphamide then docetaxel alone
disease-free survival
5-year disease-free survival 84% with the anthracycline and trastuzumab regimen, 81% with docetaxel, carboplatin and trastuzumab, and 75% without trastuzumab; no significant difference between the two trastuzumab regimens
5-year overall survival 92%, 91% and 87% respectively
Nine weekly trastuzumab infusions versus none in HER2-positive disease, inside a randomisation of docetaxel versus vinorelbine followed by fluorouracil, epirubicin and cyclophosphamide.
Recurrence-free survival.
Among HER2-positive patients, 3-year recurrence-free survival was 89% with trastuzumab and 78% without it (hazard ratio 0.42, 95% CI 0.21 to 0.83, P = 0.01).
Overall survival did not differ between docetaxel and vinorelbine (P = 0.15). No overall survival result was reported for the trastuzumab comparison.
The overview of seven of these trials, in 13,864 women followed for a median of 10.7 years, prices the class5. Trastuzumab with chemotherapy reduced recurrence, with a rate ratio of 0.66, and breast cancer mortality, with a rate ratio of 0.67. Ten-year recurrence fell by 9.0 percentage points and breast cancer mortality by 6.4, and all-cause mortality by 6.5, with no increase in death without recurrence. The proportional reduction was largest in the first year.
Four trials established the class and none of them tested how much of it to give. They agree on the finding and differ on nearly everything used to produce it. Recurrence fell in every comparison, and where follow-up ran longest the curves had not reconverged.
The differences between them are design rather than drug. The antibody ran with the taxane in the American joint analysis and in BCIRG-006. It began only after all chemotherapy had finished in HERA. The backbone carried an anthracycline in the joint analysis and in FinHer, and in two of the three BCIRG-006 arms. HERA asked only that chemotherapy had been given, so its backbone was not one regimen. The antibody ran for a year in three and for nine weeks in the fourth. Each of those choices was written into a protocol rather than randomised, so the table records four conventions alongside its four comparisons.
Three of the four report an overall survival result and they do not carry equal weight. NSABP B-31/N9831 was analysed jointly against a prespecified number of deaths, on randomisations that stayed intact throughout2. Its four-year update had already shown consistent disease-free and overall survival advantages6. HERA also reported a lower hazard of death against observation, but half of its observation group crossed over to trastuzumab once the first results were public1. Those first results, at a median of one year, had shown a hazard ratio of 0.54 for an event with one year of trastuzumab7. The reduction was similar across subgroups defined by nodal and hormone receptor status, including those at lower risk8. Crossover on that scale dilutes a survival comparison by design, and no amount of further follow-up repairs it. HERA is the trial that answers the duration question. The joint analysis is the one whose survival comparison was never diluted.
BCIRG-006 was asked a second question, and its answer is weaker than the way it is usually quoted. Both trastuzumab regimens beat chemotherapy alone. Neither was set against the other under a non-inferiority design, and the trial reports no significant difference between the two3. That is an absence of evidence rather than a demonstration of equivalence. The carboplatin and taxane partnership with trastuzumab had first been shown active in metastatic disease, on weekly and three-weekly schedules9. The cardiac case for dropping the anthracycline is stronger than the efficacy case, and it is argued in Cardiac safety across the adjuvant trials.
The short-course exception is worth naming, because everything downstream came out of it. FinHer gave nine weeks of trastuzumab rather than a year, and it found the same kind of recurrence benefit4. Its final analysis found the brief course concomitant with docetaxel safe and effective, and worth further evaluation10. In PACS 04, the HER2-positive subgroup given trastuzumab after chemotherapy showed the direction of the larger trials without the power to confirm it11. The FinHer result came from a 232-patient HER2-positive substudy nested inside a randomisation of two different chemotherapies. A signal from a nested substudy is a reason to run a duration trial. It is not a duration result.
Longer is now closed. HERA is the only trial that randomised a second year, and the second year bought nothing on disease-free survival. It cost cardiac function, and that trade is in Cardiac safety across the adjuvant trials.
Shorter is not closed. Four trials tested it and the conclusion in each turns on the margin rather than on the estimate.
The four trials of less than a year.
Read the endpoint column before the result column, because the margin is where these trials differ most.
six months vs twelve months of adjuvant trastuzumab
disease-free survival, non-inferiority against a hazard margin of 1.15
adjusted disease-free survival HR 1.08 (0.93-1.25) at a median 7.5 years; the interval contained the margin, so six months was not shown to be non-inferior
nine weeks vs one year of adjuvant trastuzumab with sequential anthracycline and taxane chemotherapy
disease-free survival, non-inferiority against a hazard margin of 1.29
5-year disease-free survival 85% with nine weeks and 88% with one year, HR 1.13 (90% CI 0.89-1.42); the upper limit crossed the margin, so non-inferiority was not shown
5-year overall survival 95.0% with nine weeks and 95.2% with one year, HR 1.07 (90% CI 0.74-1.56)
Designing de-escalation trials and the non-inferiority margin problem names four things to establish before accepting any non-inferiority claim. The margin and its justification. Whether a per-protocol analysis accompanied the intention-to-treat one. The observed control event rate against the rate the power calculation assumed. And whether the margin was set on an absolute or a relative scale. Three of the four decide this section.
Start with the two six-month trials, because they are the pair that makes the point. PHARE observed a hazard ratio of 1.08 and did not declare non-inferiority. That was its planned final analysis, of 3,380 patients with 704 events at a median of 7.5 years, and the 95% interval included the margin13. PERSEPHONE observed 1.07 and did. Two trials of one comparison, in broadly similar populations, reached opposite verdicts on estimates that differ by one hundredth.
The margin separated them, and so did the scale the margin was set on. PHARE set a hazard margin of 1.15. PERSEPHONE set three percentage points of four-year disease-free survival. A fixed allowance in percentage points becomes more permissive the fewer events the control arm has, and events in this population are few. Converted to the scale PHARE used, the allowance PERSEPHONE worked to was the wider of the two by a long way.
Put the two allowances side by side and the verdicts stop contradicting each other. PERSEPHONE established that six months is not worse by more than its own generous allowance. PHARE established that six months has not been shown to be less than 15% worse on the hazard scale. Both statements are true of closely similar data. The conclusion that reached practice was written by whoever set the wider margin.
That is the lesson of Designing de-escalation trials and the non-inferiority margin problem arriving inside a decision made weekly. A margin expressed in percentage points behaves differently from one expressed as a ratio, and the gap between them widens as the control event rate falls. Neither trial did anything improper, and neither result should be described as a mistake. Only one of them published an allowance that a patient would have been asked to accept in advance.
The four shorter trials also differ structurally, and the differences sit upstream of any margin. PHARE randomised patients who were already receiving adjuvant trastuzumab, between the third and sixth month, so its population had already tolerated the first part of the course. PERSEPHONE randomised some patients before trastuzumab began and others after it had started, with the antibody given alongside or after chemotherapy. Short-HER used sequential anthracycline and taxane. SOLD gave docetaxel with trastuzumab up front and then fluorouracil, epirubicin and cyclophosphamide. SOLD also reported that patients given the lower docetaxel dose had inferior disease-free survival14. Part of what reads as a duration effect in that trial may be a chemotherapy dose effect, and its design cannot separate the two.
The two nine-week trials are not a matched pair either, and they should not be read as one finding. Short-HER's interval crossed its own margin and also crossed equality, so it is inconclusive in both directions. SOLD's interval excluded equality at a lower bound of 1.12, so its estimate points to harm rather than to absent proof. Failing to demonstrate non-inferiority and demonstrating inferiority are different findings. Only SOLD produced the second, and that distinction should reach the patient being offered the short course.
The third of the four checks is the one these trials make hardest to apply. A margin is justified against an assumed control event rate, so it becomes more or less demanding as the observed rate departs from that assumption. Recurrence in this subtype has fallen steadily as the rest of treatment improved, which makes every margin set a decade ago more permissive than it looked when it was written.
Two things the table cannot show constrain all of this. None of the four randomised duration inside a defined risk stratum. A failed non-inferiority claim in an unselected population therefore says nothing about the small node-negative tumour, which is the commonest candidate for a short course. And none of them gave dual blockade, so no shortened course has ever been tested under two antibodies.
Four trials therefore left three states of knowledge rather than one answer. Six months is acceptable under a generous margin and unproven under a strict one. Nine weeks is unproven in one trial and probably worse in the other.
In practice
One year remains the default. It was fixed by protocol convention rather than chosen by comparison, and four attempts to shorten it have not displaced it.
Six months is defensible where cardiac risk, frailty, distance or cost make twelve unattractive. Say what the supporting trial allowed. Its margin was an absolute difference in disease-free survival at a fixed time point, which in a population with few events is a wide allowance12.
Nine weeks is for a patient who has already had a cardiac event on treatment, or whose baseline risk is very low. One trial of it was inconclusive and the other pointed towards harm15,14.
None of the four duration trials used dual blockade and most predate routine pertuzumab. A duration result obtained under trastuzumab alone does not license a shortened course under two antibodies.
3 · Dual blockade in the adjuvant setting
Matching adjuvant HER2 treatment to baseline risk, continued
Does escalating beyond chemotherapy and a year of trastuzumab lower recurrence, and if so for whom?
Rationale
A second antibody binds a different epitope and blocks the dimerisation that trastuzumab leaves intact. Neoadjuvant trials had already shown that the pair raises pathological complete response above the single antibody. A kinase inhibitor offers the same second block from inside the cell, and a conjugate offers to carry the cytotoxic to the target rather than through the patient. The adjuvant question is whether any of that converts into fewer recurrences years later.
Experiment
One trial randomised pertuzumab against placebo on top of chemotherapy and a year of trastuzumab, in node-positive and high-risk node-negative disease. A second tested a different route to a dual block, randomising a kinase inhibitor given with trastuzumab, or sequenced after it, against trastuzumab alone. A third kept pertuzumab running through both arms and replaced the taxane and trastuzumab portion of the regimen with the conjugate.
Finding
The second antibody lowered invasive recurrence, with the gain sitting in the node-positive stratum and nothing to show in node-negative disease. The kinase inhibitor given concurrently fell short of its significance threshold. Sequencing it after trastuzumab added nothing at all. The conjugate substitution did not improve on the taxane arm on either of its co-primary populations.
Limitation
The nodal split is a subgroup gradient inside one randomisation rather than a randomised comparison of two populations. Overall survival in that trial has not separated. The kinase inhibitor's concurrent arm was judged against a threshold tightened after a third arm closed for futility, so the same estimate reads as a narrow miss or as a small real benefit depending on which threshold is applied to it. The conjugate's null result is a failure to improve on an active control and not evidence of an inactive drug. Far fewer patients completed treatment on the conjugate arm, so it compares strategies as delivered rather than drugs as prescribed.
Next question
If a second antibody pays only at higher nodal risk, does a drug with a different mechanism help the patients still at risk once the year of blockade has finished?
APHINITYALTTOKAITLIN
Three ways of adding a second HER2-directed agent after surgery.
One worked in one stratum and two did not work at all.
anthracycline followed by trastuzumab emtansine with pertuzumab vs anthracycline followed by a taxane with trastuzumab and pertuzumab
invasive disease-free survival in the node-positive and overall populations
node-positive HR 0.97 (0.71-1.32) and overall HR 0.98 (0.72-1.32); 3-year invasive disease-free survival 93.1% with the conjugate and 94.2% with the taxane
trastuzumab with lapatinib, or trastuzumab followed by lapatinib, or lapatinib alone, vs one year of trastuzumab alone
disease-free survival
at a median 4.5 years, the combination against trastuzumab gave a disease-free survival HR of 0.84 (97.5% CI 0.70-1.02, p=0.048 against a 0.025 threshold); the sequence gave an HR of 0.96 (0.80-1.15)
not reported at the protocol-specified analysis
Three trials asked whether more HER2 blockade after surgery lowers recurrence, and each was set against a control arm of chemotherapy with a year of HER2 antibody. Only one moved its endpoint. Reading them as three verdicts on dual blockade misreads them, because they differ in what they added, in what they replaced and in the threshold each was judged against.
APHINITY is the one that worked, and its benefit should be described honestly. The gain sits in the node-positive stratum, and the node-negative stratum shows nothing17,19. That split is the whole clinical content of the trial.
Nodal status is not a subgroup here in the usual weak sense. It was a stratification factor rather than a post-hoc slice, so the strata are balanced on everything randomisation balances. Balance is not the issue. A difference in effect between two strata is not itself a randomised comparison, so it carries less weight than the overall result does.
ALTTO is the instructive failure, and what makes it instructive is arithmetic rather than biology. Its lapatinib-alone arm closed for futility during the trial, and the protocol was then amended to require a p value at or below 0.025 for the two remaining comparisons. The combination against trastuzumab returned 0.04818. The same estimate reads as a narrow miss or as a small real effect depending on which threshold is applied to it. The sequential arm is the cleaner result and it is rarely quoted. Giving lapatinib after trastuzumab rather than with it added nothing, which separates the question of a second block from the question of which second block.
KAITLIN tested substitution rather than addition, which is a different question with a different failure mode. Pertuzumab ran in both arms, so KAITLIN is not a test of dual blockade at all. It replaced the taxane and trastuzumab portion of the regimen with a conjugate, and invasive disease-free survival was unchanged in both coprimary populations. Completion of the planned 18 cycles fell from 88.4% to 65.0%, largely through discontinuation for laboratory abnormality16. That is a comparison of strategies as delivered rather than of drugs as prescribed. A regimen that a large share of patients never finish is a different regimen from the one that was designed.
The decision rests on APHINITY alone, because it is the only one of the three that randomised an addition against what the clinic would otherwise give. What the table cannot show is the rest of that decision. Nobody randomised the duration of pertuzumab, which was fixed at a year by the same convention that fixed trastuzumab. Nobody powered a trial on node-negative disease alone, so the absence of benefit there rests on a stratum rather than on a trial. And overall survival has not separated across three interim analyses20.
4 · De-escalated adjuvant therapy for small node-negative tumours
Matching adjuvant HER2 treatment to baseline risk, continued
For a small node-negative HER2-positive tumour, how little chemotherapy is enough?
Rationale
The pivotal trials were built on node-positive and high-risk disease and prescribed an anthracycline followed by a taxane. Small node-negative tumours were largely excluded from them. Those tumours still recur, so they are still treated. They are asked to carry a toxicity that was justified against a much higher baseline risk.
Experiment
One trial gave a single-group adjuvant course of a weekly taxane with trastuzumab in node-negative disease below a size cut-off. A second randomised the antibody-drug conjugate against that same taxane-and-antibody pair, with a protocol-defined composite of clinically relevant toxicity as a co-primary endpoint alongside invasive disease-free survival. From the neoadjuvant side, a third tested whether patients responding early to dual blockade could be spared the taxane altogether.
Finding
Almost everyone given the short taxane-and-antibody course was free of invasive disease at the primary analysis, and the great majority still were at the long-term report. The conjugate arm reached a very low rate of invasive disease of its own. It did not lower the composite toxicity endpoint, which was the comparison that trial was built to make. Dropping the taxane from dual blockade lost pathological response, including among the early responders it was designed to protect.
Limitation
The single-group trial has no control arm, so its outcome measures the population selected into it as much as the regimen given to it. The substitution trial was powered for toxicity rather than for a disease-free survival comparison, and it allocated most of its patients to the conjugate while keeping only a small taxane comparator. Its outcome arms therefore cannot establish equivalence. Its null toxicity result is also a statement about a composite. The totals matched, which is not the same as the two regimens costing a patient the same things. The taxane-omission question was settled on pathological response rather than on recurrence, in hormone receptor-negative disease only, with very few patients in each response group.
Next question
If the floor can be lowered this far for low-risk disease, does the ceiling need raising for node-positive disease?
APTATEMPTWSG-ADAPT HER2+/HR−
The two de-escalated adjuvant regimens for stage I disease.
Neither row is a randomised comparison against a full regimen.
stage I HER2-positive; randomised 3:1, with the analysis population the 497 who began protocol therapy
trastuzumab emtansine vs paclitaxel with trastuzumab, randomised 3:1
clinically relevant toxicity, with 3-year invasive disease-free survival as a co-primary endpoint
clinically relevant toxicity 46% with the conjugate against 47% with the taxane, p=0.83; 3-year invasive disease-free survival with the conjugate 97.8% (96.3-99.3)
weekly paclitaxel with trastuzumab for 12 weeks then trastuzumab to one year, single group
survival free from invasive disease
3-year survival free from invasive disease 98.7% (97.6-99.8) at a median 4.0 years, with 12 recurrences of which 2 were distant
Patients with small node-negative tumours were largely excluded from the registration trials23. Their untreated risk was measured in cohorts. HER2 positivity carried a worse outcome in node-negative disease in a large tissue microarray series24. In 965 T1a and T1b node-negative tumours, HER2 positivity raised the risk of recurrence, and the authors recommended adjuvant HER2-directed therapy25. A second cohort found a low absolute risk at five years, with HER2 overexpression associated with worse disease-free survival in receptor-positive disease26. The regimen used for them was therefore built rather than inherited, and the way it was built constrains what can be claimed for it. Both rows agree that outcomes in this population are excellent. Neither row is a comparison against a full regimen, and the reason differs between them.
APT has no comparator at all, which fixes in advance the ceiling on what it can claim. A single-group trial in a population with a very low event rate can show only that outcomes are excellent. It cannot show that a shorter regimen is equivalent to a longer one, because the longer one is absent from the dataset. No margin was stated and none could have been tested. Ten-year follow-up has since been reported and it does not change that structure27.
ATEMPT was randomised, and only one of its two co-primary endpoints was a comparison. Clinically relevant toxicity was compared between the arms. Invasive disease-free survival was evaluated on the conjugate arm alone, so it was never set against the taxane under a margin. The 3:1 allocation left 114 patients in the comparator arm. An efficacy non-inferiority claim from that arm is unavailable at any margin a clinician would accept Designing de-escalation trials and the non-inferiority margin problem. At a median follow-up of 5.8 years, five-year invasive disease-free survival was 97.0% on the conjugate28. The comparator arm reached 91.1%, and the report states that the trial was not powered for that comparison.
Its null toxicity result also needs reading as a statement about a composite. The totals matched. That is not the same as the two regimens costing a patient the same things. Neuropathy and alopecia sit on one side of the trade and hepatic and haematological abnormality on the other21.
De-escalation has been tried once from the other end, in the neoadjuvant setting rather than here. WSG-ADAPT HER2+/HR− asked whether patients responding early to dual blockade could be spared the taxane altogether, and dropping it lost pathological response, including among the early responders it was meant to protect. That result was read on response rather than on recurrence, in hormone receptor-negative disease, with very few patients in each response group. It narrows the question without settling it.
What a real de-escalation trial here would need is worth stating, because it explains why none exists. A margin narrow enough to matter, against a control event rate of a few per cent, demands a sample size that no sponsor will fund for a population this favourable. The gap is therefore structural rather than an oversight.
The honest position is narrow. Twelve weeks of paclitaxel with trastuzumab produces excellent outcomes in node-negative tumours of 3 cm or less, and that is the whole of what has been shown. Trastuzumab emtansine produces excellent outcomes in stage I disease with a different toxicity profile rather than a smaller one. Neither statement is a demonstration of non-inferiority against a full regimen.
Caution
An impressive event-free rate in a low-risk population is not evidence of equivalence. It is mostly evidence that the population was low risk. The rate would also have been high on a longer regimen, and nothing in either trial measures the gap between the two. Treat the phrase "non-inferior" as unavailable here, because no margin was ever set.
In practice
For a node-negative tumour of 3 cm or less, weekly paclitaxel with trastuzumab is the default22. Describe the evidence as a single-arm trial with an excellent rate.
Choose trastuzumab emtansine where neuropathy or alopecia dominates the conversation rather than where total toxicity does, since clinically relevant toxicity was not reduced21.
Neither trial carried an untreated arm, so nothing here says whether a T1a or T1b node-negative tumour needs HER2-directed treatment at all. That decision is extrapolation, and the consultation should say so.
5 · Post-neoadjuvant intensification for residual disease
Early disease · HER2+ · post-neoadjuvant
Escalating treatment on residual disease after HER2 blockade
For residual invasive disease at surgery, does switching to an antibody-drug conjugate lower recurrence more than completing the planned trastuzumab?
Rationale
Residual disease is observed resistance rather than inferred risk. The tumour met trastuzumab and a taxane and was still present at operation. Completing the same antibody asks a drug to finish work it has already been shown not to do in that patient. A conjugate keeps the target and changes the mechanism, by delivering a cytotoxic to a cell the antibody can still find. The selection rule itself was not new when it reached this subtype. Escalating chemotherapy on residual disease had already lowered recurrence in HER2-negative disease.
Experiment
KATHERINE randomised patients with residual invasive disease in the breast or axilla after neoadjuvant taxane-based chemotherapy with trastuzumab to trastuzumab emtansine or to completing trastuzumab, with invasive disease-free survival as the primary endpoint. CREATE-X had tested the same selection rule in HER2-negative residual disease, with capecitabine as the escalation. Beside both sits KAITLIN, which gave the same conjugate with pertuzumab after upfront surgery in place of the taxane. Its patients were selected by nodal status and tumour size rather than by residual disease. It is the same drug asked to work without the selection rule.
Finding
In selected residual disease the conjugate lowered invasive recurrence against continuing the antibody. Follow-up ran long enough for that recurrence benefit to become an overall survival benefit, which is uncommon in this setting. The selection rule had already held outside the subtype, where post-neoadjuvant capecitabine lowered recurrence and lengthened survival in HER2-negative residual disease. Where the rule was absent the conjugate did nothing. In the adjuvant population it matched the taxane with dual antibody on invasive disease-free survival. Fewer patients completed the planned course.
Limitation
The control arm was the drug the tumour had already grown through. What the trial establishes is that a conjugate beats continuing an agent known to have failed in that patient. It does not establish that the conjugate beats any other escalation that might have been offered. The entry rule was also loose about what the tumour had been exposed to. An anthracycline was optional. Most patients were treated before pertuzumab was routine. Residual disease in the trial therefore followed a weaker regimen than residual disease in clinic today. The capecitabine trial supplies the principle but not the subtype, because it enrolled HER2-negative disease only.
Next question
If the class is right, does a conjugate with a different payload beat the one that established the setting?
KATHERINECREATE-XKAITLIN
Does a bystander-capable conjugate lower recurrence further than the conjugate that established this setting?
Rationale
The first conjugate carries a payload that stays inside the cell it enters. It therefore kills only cells that bind and internalise enough antibody. Residual disease after HER2 blockade is the compartment most likely to carry reduced or patchy antigen. A cleavable payload that crosses into neighbouring cells should reach what the first conjugate cannot. A single-arm neoadjuvant study had already reported no responses at all to the non-bystander conjugate where HER2 was spatially heterogeneous.
Experiment
DESTINY-Breast05 randomised trastuzumab deruxtecan against trastuzumab emtansine in residual invasive disease, with entry restricted to node-positive disease at surgery or inoperable disease at diagnosis. The comparator was the established conjugate rather than the antibody. This is the first trial in this position whose control arm was already known to work. The antigen argument behind it comes from the heterogeneity study, which gave the older conjugate with pertuzumab before surgery and compared prospectively defined heterogeneous and non-heterogeneous tumours.
Finding
The newer conjugate lowered invasive recurrence against an active comparator. Drug-related interstitial lung disease was far more frequent than on the older conjugate. Patients on the newer conjugate died of it. In the heterogeneity study no tumour with spatially heterogeneous HER2 reached a pathological complete response, against a substantial response rate in the tumours called homogeneous.
Limitation
Recurrence was the endpoint that reported. Overall survival was immature at the primary analysis. A pulmonary toxicity that killed patients is therefore being weighed against a recurrence endpoint whose survival counterpart has not yet read out. The enrolled population was also narrower than that of the trial it displaced, because residual disease with negative nodes at surgery was not admitted. The antigen reasoning rests on a single-arm study whose heterogeneous and non-heterogeneous groups were prospectively defined but never randomised, with very few heterogeneous tumours among them. Heterogeneity there was read off two biopsies per tumour against a fixed cut-off for the proportion of amplified cells. The exposure variable there is a sampling scheme and a scoring convention before it is a property of the disease.
Next question
Does the benefit reach the residual-disease patients the trial did not enrol?
DESTINY-Breast05KATHERINENeoadjuvant T-DM1-pertuzumab HER2-heterogeneity study
Does low-volume residual disease with negative nodes at surgery need the same escalation as residual nodal disease?
Rationale
Residual disease is a category rather than a quantity. A small deposit left in the breast with clear nodes carries a different hazard from disease still in the nodes after treatment. Trials have written that gradient into their entry criteria instead of testing it, which converts a biological question into an eligibility question.
Experiment
Read the entry criteria against each other rather than the results. KATHERINE admitted any residual invasive disease in the breast or axilla. DESTINY-Breast05 required node-positive disease at surgery or inoperable disease at diagnosis. The two randomised additions still running narrow entry further. CompassHER2 RD admits oestrogen receptor-negative or node-positive residual disease. ASTEFANIA admits residual disease called high-risk.
Finding
The randomised record is dense where risk is highest and thin where it is lowest. For residual disease with negative nodes, the only randomised evidence is still the older comparison against the antibody. The newer conjugate is untested in exactly the group whose absolute risk is smallest.
Limitation
Nodal status after neoadjuvant therapy is a staging convention as much as a biological variable. It depends on how the axilla was sampled and on how many nodes were examined. A patient called node-negative after a sentinel procedure has not had the same test as one after a dissection. No trial randomised the threshold itself. The line between escalating and not escalating is an eligibility rule that was chosen once and then inherited by the trials that followed.
Next question
For a patient outside the newer trial's entry criteria, is there anything worth adding to the established conjugate instead?
KATHERINEDESTINY-Breast05CompassHER2 RDASTEFANIA
The two reported trials in this position.
The population column decides which of them a patient belongs to.
HER2-positive with residual invasive disease in breast or axilla after neoadjuvant taxane, with or without anthracycline, and trastuzumab
adjuvant trastuzumab emtansine for 14 cycles vs trastuzumab
invasive disease-free survival
3-year freedom from invasive disease 88.3% with the conjugate against 77.0% with trastuzumab, HR 0.50 (0.39-0.64)
7-year overall survival 89.1% vs 84.4%, HR 0.66, 95% CI 0.51 to 0.87, P = 0.003
Residual invasive disease after adequate neoadjuvant HER2-directed therapy is the strongest selection rule available in early breast cancer. It identifies a tumour that has met the drug and survived it, in this patient rather than in a population resembling them. That is a measured property of this tumour rather than a prediction about tumours like it. The rule was not invented here. CREATE-X had already shown that escalating chemotherapy on residual disease lowers recurrence in HER2-negative disease, which supplies the principle without supplying the subtype.
Both trials in the table agree that a conjugate given on that selection rule lowers invasive recurrence. They differ in the one thing that decides what each can claim, which is the control arm. KATHERINE's control was the antibody the tumour had already grown through, so the trial establishes that a conjugate beats continuing an agent known to have failed in that patient30. It does not establish that a conjugate beats any other escalation that might have been offered. DESTINY-Breast05 randomised against a comparator already known to work, which makes it the harder comparison and the more informative one29.
Maturity is the second difference and it is follow-up rather than drug. KATHERINE has now run long enough for an early recurrence benefit to become an overall survival benefit31. That conversion is uncommon in this book, and it is why the conjugate rather than the antibody became the comparator for everything that came after. DESTINY-Breast05 reported recurrence at a median of about 30 months with overall survival immature. Nothing in the contrast between the two survival cells is a statement about the newer drug.
The same drug without the selection rule did nothing, which is the cleanest available check on whether the rule is doing the work. KAITLIN gave trastuzumab emtansine with pertuzumab after upfront surgery, selected on nodal status and tumour size, and matched the taxane arm rather than beating it Dual blockade in the adjuvant setting. Observed resistance and predicted risk are not interchangeable selection variables.
The eligible population of DESTINY-Breast05 is narrower than KATHERINE's, and this is routinely missed. Entry required residual invasive disease together with node-positive disease at surgery, or inoperable disease at diagnosis29. A patient with small-volume node-negative residual disease was studied in KATHERINE and was not studied in DESTINY-Breast05.
That boundary is an eligibility rule rather than a finding, and nobody randomised the threshold itself. It was chosen once for one trial and then inherited by the trials that followed. Nodal status after neoadjuvant therapy is also a staging convention before it is a biological variable. It depends on how the axilla was sampled and on how many nodes were examined, and the conventions are in Pathological assessment after neoadjuvant treatment. A patient called node-negative after a sentinel procedure has not had the same test as one after a dissection.
The biological argument for the newer payload rests on a study that no table here carries. Neoadjuvant T-DM1-pertuzumab HER2-heterogeneity study gave the older conjugate with pertuzumab before surgery and compared prospectively defined heterogeneous and non-heterogeneous tumours. No tumour called heterogeneous reached a pathological complete response. Heterogeneity there was read from two biopsies per tumour against a fixed cut-off for the proportion of amplified cells. The exposure variable is therefore a sampling scheme and a scoring convention before it is a property of the disease. The groups were never randomised and the heterogeneous group was very small. It is a reason to expect a bystander-capable payload to do better and not a demonstration that it does.
One further limitation belongs to KATHERINE and is easy to miss. Its entry rule asked only for residual disease after a taxane and trastuzumab, with an anthracycline optional and pertuzumab not yet routine. Residual disease in that trial therefore followed a weaker regimen than residual disease follows in clinic today. The population that survives current neoadjuvant treatment is a harder one, and the older comparison is still being applied to it.
Caution
The recurrence benefit in DESTINY-Breast05 is large and the price is respiratory. Adjudicated drug-related interstitial lung disease occurred in 9.6% of patients on trastuzumab deruxtecan against 1.6% on trastuzumab emtansine, and two patients with it died29. Overall survival is immature, so the current position is a measured recurrence benefit, an unmeasured survival benefit, and two treatment-related deaths from a toxicity that is detectable early. The surveillance that separates grade 1 from grade 5 is in Toxicity mechanisms, interstitial lung disease, ocular, and mucosal and belongs in place before the first dose.
In practice
Ask one question before choosing the agent. Was the disease node-positive at surgery, or inoperable at diagnosis? If yes, the patient matches the DESTINY-Breast05 population. If no, the patient matches KATHERINE, and the randomised evidence for the newer conjugate does not cover them.
Adequate neoadjuvant therapy is part of the selection rule. Residual disease after an inadequate regimen is a different finding, because the tumour was never properly tested.
Matching adjuvant HER2 treatment to baseline risk, continued
Does extending HER2-directed treatment beyond the first year add anything?
Rationale
Recurrences keep arriving after the antibody stops, and in hormone receptor-positive disease they arrive late. Whether more time on treatment helps at all had already been put to the antibody itself. An irreversible pan-HER kinase inhibitor acts inside the cell and covers signalling the antibody leaves intact, including the crosstalk with the oestrogen receptor that is one proposed route of escape.
Experiment
The pivotal observation trial had randomised a second year of the antibody against one. The extension trial randomised a further year of the kinase inhibitor against placebo in patients who had completed trastuzumab-based adjuvant therapy, stratified by hormone receptor status, with invasive disease-free survival read at a fixed early landmark as the primary endpoint.
Finding
A second year of the antibody itself added nothing over the first. The kinase inhibitor lowered invasive recurrence inside that early window. Severe diarrhoea was far more common than on placebo. By the final analysis the two arms had the same overall survival.
Limitation
An early landmark is a short window for the very group the drug is now aimed at, whose recurrences come late, so the primary result was read before the population it selects for had time to relapse. Hormone receptor status was a stratification factor and not a randomised comparison of the strata. The trial enrolled before adjuvant pertuzumab and before conjugate treatment for residual disease, so almost nobody in it had received the treatment a candidate would receive today. No antidiarrhoeal strategy was part of the randomisation, so the tolerability on record is the tolerability of the drug given without one.
Next question
Every escalation and every de-escalation here is allocated on nodal status and tumour size measured before a drop of treatment is given. Can the tumour's own response to treatment allocate it better?
ExteNETHERA
Escalating treatment on residual disease after HER2 blockade, continued
After the post-neoadjuvant conjugate is completed, is there anything more to offer a patient whose risk continues?
Rationale
Recurrence in this population goes on after the planned cycles of conjugate end. The escalation therefore stops while the hazard is still running. Extended oral pan-HER inhibition after the antibody is finished is the only continuation that has been randomised in this subtype.
Experiment
ExteNET randomised extended neratinib against placebo after completion of trastuzumab-based adjuvant therapy, with randomisation stratified by hormone receptor status. The population was defined by having finished trastuzumab rather than by residual disease. Patients with residual disease sit inside the trial as a fraction of it.
Finding
Extension lowered invasive recurrence against placebo over the early years of follow-up. The effect was concentrated in hormone receptor-positive disease. Long-term overall survival was no different between the arms. Severe diarrhoea was common.
Limitation
The trial predates routine pertuzumab. It predates post-neoadjuvant conjugates entirely. Nobody in it had received the treatment that now defines this position. The diarrhoea on record was collected without mandatory antidiarrhoeal prophylaxis, which is a statement about how the trial was run rather than about the drug as it is given now. The hormone receptor restriction that governs current use is a comparison between two randomisation strata rather than a randomised comparison, however well balanced those strata were. Eligibility was also narrowed partway through accrual. The population enrolled is not the one the protocol opened with.
Next question
If everything available after surgery is constrained by what was given before it, does a better preoperative regimen dissolve this question or only move it?
ExteNETKATHERINEDESTINY-Breast05
The only trial of extended oral HER2 blockade after a completed year of trastuzumab.
stage 1-3 HER2-positive who had completed trastuzumab-based adjuvant therapy; randomisation stratified by hormone receptor status
12 months of neratinib vs placebo after completion of trastuzumab-based adjuvant therapy
invasive disease-free survival at 2 years
2-year invasive disease-free survival 93.9% with neratinib against 91.6% with placebo, stratified HR 0.67 (0.50-0.91), p=0.0091
final analysis at a median 8.1 years, 8-year overall survival 90.1% with neratinib and 90.2% with placebo, stratified HR 0.95 (0.75-1.21), p=0.69
One row, and it is the only randomised test of extending HER2-directed treatment with a different agent after the antibody is finished. The other way of asking the question was put to the antibody itself, and HERA answered it in The trastuzumab adjuvant trials and the duration question. A second year of trastuzumab added nothing over the first. More time on treatment is therefore not a class effect, and ExteNET is a claim about a kinase inhibitor rather than about duration.
State the trade plainly, because the two sides of it are measured to very different standards. ExteNET lowered invasive recurrence against placebo inside its primary window. Grade 3 diarrhoea occurred in 40% of patients against 2% on placebo, in a trial run without antidiarrhoeal prophylaxis32,33. The tolerability on record is therefore a statement about how the trial was run rather than about the molecule.
The hormone receptor-positive subgroup is where the drug is now used, and the claim needs handling. Hormone receptor status was a randomisation stratification factor, so the subgroups are balanced. Balance is not what fails here. A comparison between two subgroups is not itself a randomised comparison, and a difference in effect across strata carries far less weight than the overall result. Treat the restriction as a plausible reading of a stratified analysis rather than as a demonstrated interaction34.
The endpoint window constrains the claim further, and in the direction that matters most. The primary analysis read invasive disease-free survival at an early landmark. The group the drug is now aimed at is the group whose recurrences arrive late, so the result was declared before that group had time to relapse. Overall survival did not separate at the final analysis35.
A second feature narrows the population again. Eligibility was amended during recruitment from stage 1 to 3 down to stage 2 to 3, and the interval allowed since trastuzumab was narrowed at the same time33. The enrolled population therefore changed partway through accrual, and a reader applying the result should be clear which population they mean.
Caution
ExteNET predates routine adjuvant pertuzumab and predates post-neoadjuvant conjugates entirely. A patient who has completed trastuzumab emtansine or trastuzumab deruxtecan for residual disease was not represented in it. Extending with neratinib after a conjugate is extrapolation, and the toxicity is not.
7 · Cardiac safety across the adjuvant trials
Two cardiac toxicities run through this chapter and they behave differently. Conflating them produces the wrong surveillance schedule.
Anthracycline injury is dose-dependent and largely irreversible. In a prospective cohort of 2,625 patients receiving anthracycline-containing therapy with serial ejection fraction measurement, cardiotoxicity occurred in 9%. The median interval from the end of chemotherapy to the event was 3.5 months, and 98% of events occurred within the first year. With heart failure therapy, 11% of affected patients recovered fully and 71% recovered partially. Two variables predicted the event independently. Ejection fraction at the end of chemotherapy carried a hazard ratio of 1.37 for each one-point decrement, and cumulative doxorubicin dose a hazard ratio of 1.09 for each 50 mg per square metre36.
HER2-directed injury is not dose-dependent in the same way and is usually reversible on stopping the drug. In the randomised adjuvant setting the rates are low, and primary cardiac events were uncommon in both arms of APHINITY19.
The duration trials are the cleanest cardiac dataset in breast oncology, because they randomised exposure time rather than observing it. HERA randomised a second year, and secondary cardiac endpoints occurred in 7.3% of the two-year group, 4.4% of the one-year group and 0.9% of the observation group1. PERSEPHONE randomised half the exposure and early discontinuation for cardiotoxicity fell from 8% to 3%12. The first American trials measured the same thing. In NSABP B-31, trastuzumab with paclitaxel after doxorubicin and cyclophosphamide increased heart failure, and 4% stopped the antibody for symptomatic cardiotoxicity38. At seven years, late heart failure after trastuzumab was uncommon39. In N9831, the three-year cardiac event rate was higher with trastuzumab by less than 4%, and older age, lower baseline ejection fraction and antihypertensive use predicted it40. Short-HER randomised nine weeks and reported a cardiac event risk ratio of 0.33 for the short course15. SOLD reported better maintenance of ejection fraction with nine weeks14.
Read those four results together. Trastuzumab cardiotoxicity is usually described as idiosyncratic and reversible, in contrast to the dose-dependent anthracycline injury. The duration trials show that exposure time behaves as a dose. That is a randomised dose-response relationship, and it is the reason a shortened course is a genuine cardiac intervention rather than a concession.
It also changes how the margin argument in The trastuzumab adjuvant trials and the duration question should be used, and it is the one place in this chapter where a de-escalation can be defended on positive evidence. A patient with real cardiac risk is not choosing between equal options under an uncertain margin. They are trading an unproven loss of disease control against a randomised reduction in cardiac events. The two sides of that trade are known to different standards, and the consultation should say which is which.
The anthracycline is the other lever and it was tested directly. In BCIRG-006, congestive heart failure and cardiac dysfunction were significantly more frequent with the anthracycline and trastuzumab regimen than with docetaxel, carboplatin and trastuzumab. Seven of the eight reported leukaemias arose in anthracycline-containing groups3. TRAIN-2 priced the same trade under dual blockade in the neoadjuvant setting. In that trial, an ejection fraction fall of 10 points or more to below 50% occurred in 7.7% of patients given an anthracycline and 3.2% of those who were not41. The strongest cardiac intervention available is therefore not surveillance. It is omitting the anthracycline where the evidence permits, which is priced in Chemotherapy backbones with trastuzumab.
In practice
Treat planned trastuzumab duration as a modifiable cardiac exposure in a patient with cardiac risk factors. The evidence that shorter means safer is randomised in four trials.
Treat anthracycline omission as the larger cardiac decision. BCIRG-006 found no significant difference in disease-free survival between its two trastuzumab regimens, and clearly less cardiac toxicity without the anthracycline3. The efficacy side of that trade is an absence of difference rather than a demonstration of equivalence.
Measure ejection fraction before the first anthracycline dose and again at the end of chemotherapy. The end-of-chemotherapy value was an independent predictor of subsequent cardiotoxicity, alongside cumulative anthracycline dose36.
Concentrate surveillance in the first year after treatment, which is when 98% of anthracycline events occurred36.
Start heart failure therapy when a decline is detected rather than when symptoms appear. Most affected patients treated on that basis recovered fully or partly36.
A fall on HER2-directed therapy warrants holding the drug and reassessing rather than abandoning it. A fall after an anthracycline is treated rather than observed.
In a patient with cardiac risk factors, the anthracycline decision is the cardiology decision. The evidence that it can be omitted under dual blockade in the neoadjuvant setting is randomised41.
8 · Unresolved questions and what reads out next
Escalating treatment on residual disease after HER2 blockade, continued
Does adding a second agent to the established post-neoadjuvant conjugate lower recurrence further?
Rationale
Once systemic control improves in this subtype, a large share of first relapses arrive in the brain. A conjugate is a poor drug for that site. The obvious partner is therefore the small molecule that has intracranial activity in metastatic disease. Residual disease has also survived in an intact immune context, which is the separate argument for adding a checkpoint antibody.
Experiment
Two randomised trials add an agent to the established conjugate in residual disease. CompassHER2 RD adds tucatinib against placebo, with entry restricted to oestrogen receptor-negative or node-positive disease. ASTEFANIA adds atezolizumab against placebo in high-risk residual disease. The case for each partner was made elsewhere. HER2CLIMB tested tucatinib in metastatic disease already treated with the antibodies and the conjugate. It enrolled patients with brain metastases. IMpassion050 tested a checkpoint antibody added to preoperative chemotherapy and dual blockade.
Finding
Neither addition has reported. The only publication on the record for the immunotherapy trial is its design protocol. The enrolment figure attached to that record is a plan rather than a count of patients randomised. The borrowed evidence points in two directions. Tucatinib lengthened progression-free and overall survival in previously treated metastatic disease. It also delayed progression in the patients with brain metastases. The checkpoint antibody did not raise pathological complete response before surgery in this subtype. It did not do so in the tumours selected for PD-L1 expression either. Its final analysis, with the antibody continued after surgery, found no significant gain in event-free or disease-free survival.
Limitation
Both trials were designed when the established conjugate was the only post-neoadjuvant option. A positive result will therefore show that an addition improves on a backbone that a newer conjugate has already beaten in node-positive residual disease. No trial compares those two strategies against each other. The tucatinib evidence also comes from disease that had already progressed on the conjugate in the metastatic setting, which is not the same biology as disease left behind at surgery. The checkpoint trial randomised before surgery, and its survival results are secondary endpoints. It constrains the immunotherapy question without settling it.
Next question
When the conjugate course is finished and the risk is not, is there anything to give next?
CompassHER2 RDASTEFANIAHER2CLIMBIMpassion050
Does a stronger preoperative regimen remove the population this whole strategy was built for?
Rationale
Every question in this position presupposes residual disease at surgery. The preoperative regimen decides how many patients arrive with it and which ones they are. It also decides which drugs remain unused, because a conjugate spent before surgery is not available after it.
Experiment
Three trials move that boundary in different directions. DESTINY-Breast11 randomised a conjugate given alone, or followed by the taxane with dual antibody, against dose-dense anthracycline chemotherapy followed by the same triplet. CompassHER2 pCR gave a chemotherapy-light preoperative regimen in a single arm and omitted further chemotherapy after a pathological complete response. KRISTINE substituted the older conjugate with pertuzumab for chemotherapy altogether.
Finding
The conjugate sequence raised pathological complete response against the anthracycline standard, which shrinks the residual-disease population. The chemotherapy-light regimen left most patients with residual disease. Those patients were predominantly oestrogen receptor-positive. Substituting the conjugate for chemotherapy lowered pathological complete response and enlarged the residual-disease population. Acute toxicity on that arm was much lower. Recurrence on follow-up was nonetheless more frequent than on chemotherapy. The substitution cost more than it saved.
Limitation
All three moved a surrogate first. Pathological complete response is the endpoint that reports early. Event-free survival in DESTINY-Breast11 was immature at the primary report. Enrolment in its conjugate-alone arm also closed early. What survives from it is a comparison of two sequences rather than of a conjugate against chemotherapy. The response-adapted trial has not reported its recurrence endpoint at all. Nothing has randomised what to do about residual disease that persisted through a preoperative conjugate. Every post-neoadjuvant trial in this position enrolled tumours that survived an antibody and a taxane. A tumour that survives trastuzumab deruxtecan before surgery is a different population with no evidence behind it.
Next question
Should the escalation rule be written in terms of what the tumour was exposed to rather than what is left in the specimen?
high-risk HER2-positive residual disease after neoadjuvant chemotherapy and HER2-directed treatment, defined as oestrogen receptor-negative or node-positive or both
adjuvant trastuzumab emtansine plus tucatinib vs trastuzumab emtansine plus placebo
invasive disease-free survival
Two trials are running in this position and they are built the same way. Each adds a second agent to trastuzumab emtansine in residual disease, each randomises against placebo, and each admits a narrower population than KATHERINE did. CompassHER2 RD requires oestrogen receptor-negative or node-positive residual disease and adds tucatinib43. ASTEFANIA requires residual disease called high risk and adds atezolizumab42. Neither has reported anything, so the status column is the only cell in that table with a result in it.
Read the other cells with care. ASTEFANIA's enrolment figure comes from its design paper and is a plan rather than a count of patients randomised. CompassHER2 RD carries no enrolment figure at all. Nothing in either row is an outcome, and a reader who sees an endpoint named should read it as an intention.
What divides the two trials is the strength of the evidence behind each partner rather than anything in their designs. In HER2CLIMB tucatinib lengthened progression-free and overall survival in metastatic disease already treated with the antibodies and the conjugate, and it delayed progression in the patients with brain metastases. In IMpassion050 atezolizumab did not raise pathological complete response when added to preoperative chemotherapy and dual blockade, and it did not do so in the tumours selected for PD-L1 expression either. Borrowed evidence is weaker still for being borrowed across settings. The tucatinib result comes from disease that had already progressed on the conjugate, which is not the biology of disease left behind at surgery. The checkpoint trial kept atezolizumab going after surgery, and patients with residual disease could switch to the conjugate with atezolizumab or placebo. Its final analysis found no significant gain in event-free or disease-free survival, both secondary endpoints44. Randomisation came before surgery, so the trial does not isolate the residual-disease question.
Both trials share a problem neither created. DESTINY-Breast05 has displaced their comparator in the node-positive population. A positive result from either will therefore show that an addition improves on a backbone already beaten by a different drug. No trial compares the two strategies, and none is planned.
The preoperative regimen is moving underneath all of this, which is the second half of the problem. Those regimens belong to Neoadjuvant therapy and are not restated here, but their consequence belongs here, and it is a denominator that shrinks while its composition changes. DESTINY-Breast11 raised pathological complete response against the anthracycline standard, which removes patients from this position. CompassHER2 pCR gave a chemotherapy-light preoperative regimen and left most patients with residual disease, predominantly oestrogen receptor-positive. KRISTINE substituted a conjugate for chemotherapy, lowered pathological complete response and enlarged the residual-disease population. Event-free survival was worse on the conjugate than on chemotherapy, owing to locoregional progression before surgery. Invasive disease-free survival after surgery was similar between the arms. All three moved a surrogate first, and the recurrence endpoints are immature or unreported. A tumour that survives a conjugate before surgery has not been studied by anything in this chapter.
Four questions have no trial addressing them at all.
Whether the DESTINY-Breast05 benefit extends to node-negative residual disease. The trial excluded that group and there is no plan to study it separately.
Whether trastuzumab deruxtecan, or trastuzumab emtansine with tucatinib, is the better choice for a patient at risk of brain metastasis. No randomised trial exists or is planned to make that comparison.
Whether a shortened trastuzumab course is acceptable under dual blockade. Every duration trial predates routine pertuzumab, and the cardiac argument for shortening is strongest in exactly the patients who now receive two antibodies.
One older question remains open on its own terms. Overall survival in APHINITY has not crossed its boundary19,20. The trial continues, and a node-positive patient consenting to pertuzumab today is consenting on a recurrence benefit rather than a survival one.
Interplay
Both decisions in this chapter are attempts to act on a tumour that is no longer in the body. The adjuvant decision infers risk from a resected specimen. The post-neoadjuvant decision infers it from what a specimen looked like after treatment. Part VI argues that a single specimen misrepresents a heterogeneous tumour, and residual disease is where that argument bites hardest. The surviving compartment after HER2 blockade need not carry the antigen level the diagnostic biopsy reported HER2 loss as an acquired resistance mechanism. That is one reason a bystander-capable conjugate outperforms a non-bystander one in exactly this population, and the mechanism is in Bystander killing and why it matters in heterogeneous tumors rather than here.
Cameron D, Piccart-Gebhart MJ, Gelber RD, et al. 11 years' follow-up of trastuzumab after adjuvant chemotherapy in HER2-positive early breast cancer: final analysis of the HERceptin Adjuvant (HERA) trial. Lancet 2017 389:1195-1205. PMID 28215665
Perez EA, Romond EH, Suman VJ, et al. Trastuzumab plus adjuvant chemotherapy for human epidermal growth factor receptor 2-positive breast cancer: planned joint analysis of overall survival from NSABP B-31 and NCCTG N9831. J Clin Oncol 2014 32:3744-3752. PMID 25332249
Slamon D, Eiermann W, Robert N, et al. Adjuvant trastuzumab in HER2-positive breast cancer. N Engl J Med 2011 365:1273-1283. PMID 21991949
Joensuu H, et al. Adjuvant docetaxel or vinorelbine with or without trastuzumab for breast cancer. N Engl J Med 2006 354:809-20. PMID 16495393
Early Breast Cancer Trialists' Collaborative Group (EBCTCG). Trastuzumab for early-stage, HER2-positive breast cancer: a meta-analysis of 13 864 women in seven randomised trials. Lancet Oncol 2021 22:1139-1150. PMID 34339645
Perez EA, Romond EH, Suman VJ, et al. Four-year follow-up of trastuzumab plus adjuvant chemotherapy for operable human epidermal growth factor receptor 2-positive breast cancer: joint analysis of data from NCCTG N9831 and NSABP B-31. J Clin Oncol 2011 29:3366-73. PMID 21768458
Piccart-Gebhart MJ, Procter M, Leyland-Jones B, et al. Trastuzumab after adjuvant chemotherapy in HER2-positive breast cancer. N Engl J Med 2005 353:1659-1672. PMID 16236737
Untch M, Gelber RD, Jackisch C, et al. Estimating the magnitude of trastuzumab effects within patient subgroups in the HERA trial. Ann Oncol 2008 19:1090-6. PMID 18296421
Perez EA, Suman VJ, Rowland KM, et al. Two concurrent phase II trials of paclitaxel/carboplatin/trastuzumab (weekly or every-3-week schedule) as first-line therapy in women with HER2-overexpressing metastatic breast cancer: NCCTG study 983252. Clin Breast Cancer 2005 6:425-32. PMID 16381626
Joensuu H, Bono P, Kataja V, et al. Fluorouracil, epirubicin, and cyclophosphamide with either docetaxel or vinorelbine, with or without trastuzumab, as adjuvant treatments of breast cancer: final results of the FinHer Trial. J Clin Oncol 2009 27:5685-92. PMID 19884557
D'Hondt V, Canon JL, Roca L, et al. UCBG 2-04: Long-term results of the PACS 04 trial evaluating adjuvant epirubicin plus docetaxel in node-positive breast cancer and trastuzumab in the human epidermal growth factor receptor 2-positive subgroup. Eur J Cancer 2019 122:91-100. PMID 31634648
Earl HM, Hiller L, Vallier AL, et al. 6 versus 12 months of adjuvant trastuzumab for HER2-positive early breast cancer (PERSEPHONE): 4-year disease-free survival results of a randomised phase 3 non-inferiority trial. Lancet 2019 393:2599-2612. PMID 31178152
Pivot X, Romieu G, Debled M, et al. 6 months versus 12 months of adjuvant trastuzumab in early breast cancer (PHARE): final analysis of a multicentre, open-label, phase 3 randomised trial. Lancet 2019 393:2591-2598. PMID 31178155
Joensuu H, Fraser J, Wildiers H, et al. Effect of adjuvant trastuzumab for a duration of 9 weeks vs 1 year with concomitant chemotherapy for early human epidermal growth factor receptor 2-positive breast cancer: the SOLD randomized clinical trial. JAMA Oncol 2018 4:1199-1206. PMID 29852043
Conte P, Frassoldati A, Bisagni G, et al. Nine weeks versus 1 year adjuvant trastuzumab in combination with chemotherapy: final results of the phase III randomized Short-HER study. Ann Oncol 2018 29:2328-2333. PMID 30219886
Krop IE, Im SA, Barrios C, et al. Trastuzumab emtansine plus pertuzumab versus taxane plus trastuzumab plus pertuzumab after anthracycline for high-risk human epidermal growth factor receptor 2-positive early breast cancer: the phase III KAITLIN study. J Clin Oncol 2022 40:438-448. PMID 34890214
von Minckwitz G, Procter M, de Azambuja E, et al. Adjuvant pertuzumab and trastuzumab in early HER2-positive breast cancer. N Engl J Med 2017 377:122-131. PMID 28581356
Piccart-Gebhart M, Holmes E, Baselga J, et al. Adjuvant lapatinib and trastuzumab for early human epidermal growth factor receptor 2-positive breast cancer: results from the randomized phase III Adjuvant Lapatinib and/or Trastuzumab Treatment Optimization trial. J Clin Oncol 2016 34:1034-1042. PMID 26598744
Piccart M, Procter M, Fumagalli D, et al. Adjuvant pertuzumab and trastuzumab in early HER2-positive breast cancer in the APHINITY trial: 6 years' follow-up. J Clin Oncol 2021 39:1448-1457. PMID 33539215
Loibl S, et al. Adjuvant Pertuzumab and Trastuzumab in Early Human Epidermal Growth Factor Receptor 2-Positive Breast Cancer in the APHINITY Trial: Third Interim Overall Survival Analysis With Efficacy Update. J Clin Oncol 2024 42:3643-3651. PMID 39259927
Tolaney SM, Tayob N, Dang C, et al. Adjuvant trastuzumab emtansine versus paclitaxel in combination with trastuzumab for stage I HER2-positive breast cancer (ATEMPT): a randomized clinical trial. J Clin Oncol 2021 39:2375-2385. PMID 34077270
Tolaney SM, Barry WT, Dang CT, et al. Adjuvant paclitaxel and trastuzumab for node-negative, HER2-positive breast cancer. N Engl J Med 2015 372:134-141. PMID 25564897
Romond EH, Perez EA, Bryant J, et al. Trastuzumab plus adjuvant chemotherapy for operable HER2-positive breast cancer. N Engl J Med 2005 353:1673-1684. PMID 16236738
Chia S, Norris B, Speers C, et al. Human epidermal growth factor receptor 2 overexpression as a prognostic factor in a large tissue microarray series of node-negative breast cancers. J Clin Oncol 2008 26:5697-704. PMID 19001334
Gonzalez-Angulo AM, Litton JK, Broglio KR, et al. High risk of recurrence for patients with breast cancer who have human epidermal growth factor receptor 2-positive, node-negative tumors 1 cm or smaller. J Clin Oncol 2009 27:5700-6. PMID 19884543
Curigliano G, Viale G, Bagnardi V, et al. Clinical relevance of HER2 overexpression/amplification in patients with small tumor size and node-negative breast cancer. J Clin Oncol 2009 27:5693-9. PMID 19884553
Tolaney SM, et al. Adjuvant paclitaxel and trastuzumab for node-negative, HER2-positive breast cancer: final 10-year analysis of the open-label, single-arm, phase 2 APT trial. Lancet Oncol 2023 24:273-285. PMID 36858723
Tarantino P, et al. Adjuvant Trastuzumab Emtansine Versus Paclitaxel Plus Trastuzumab for Stage I Human Epidermal Growth Factor Receptor 2-Positive Breast Cancer: 5-Year Results and Correlative Analyses From ATEMPT. J Clin Oncol 2024 42:3652-3665. PMID 38935923
Loibl S, Park YH, Shao Z, et al. Trastuzumab deruxtecan in residual HER2-positive early breast cancer. N Engl J Med 2026 394:845-857. PMID 41370739
von Minckwitz G, Huang CS, Mano MS, et al. Trastuzumab emtansine for residual invasive HER2-positive breast cancer. N Engl J Med 2019 380:617-628. PMID 30516102
Geyer CE, Untch M, Huang CS, et al. Survival with trastuzumab emtansine in residual HER2-positive breast cancer. N Engl J Med 2025 392:249-257. PMID 39813643
Chan A, Delaloge S, Holmes FA, et al. Neratinib after trastuzumab-based adjuvant therapy in patients with HER2-positive breast cancer (ExteNET): a multicentre, randomised, double-blind, placebo-controlled, phase 3 trial. Lancet Oncol 2016 17:367-377. PMID 26874901
Martin M, Holmes FA, Ejlertsen B, et al. Neratinib after trastuzumab-based adjuvant therapy in HER2-positive breast cancer (ExteNET): 5-year analysis of a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet Oncol 2017 18:1688-1700. PMID 29146401
Chan A, et al. Final Efficacy Results of Neratinib in HER2-positive Hormone Receptor-positive Early-stage Breast Cancer From the Phase III ExteNET Trial. Clin Breast Cancer 2020 21:80-91.e7. PMID 33183970
Holmes FA, et al. Overall survival with neratinib after trastuzumab-based adjuvant therapy in HER2-positive breast cancer (ExteNET): A randomised, double-blind, placebo-controlled, phase 3 trial. Eur J Cancer 2023 184:48-59. PMID 36898233
Cardinale D, Colombo A, Bacchiani G, et al. Early detection of anthracycline cardiotoxicity and improvement with heart failure therapy. Circulation 2015 131:1981-1988. PMID 25948538
Lyon AR, López-Fernández T, Couch LS, et al. 2022 ESC guidelines on cardio-oncology developed in collaboration with the European Hematology Association, the European Society for Therapeutic Radiology and Oncology and the International Cardio-Oncology Society. Eur Heart J 2022 23:e333-e465. PMID 36017568
Tan-Chiu E, Yothers G, Romond E, et al. Assessment of cardiac dysfunction in a randomized trial comparing doxorubicin and cyclophosphamide followed by paclitaxel, with or without trastuzumab as adjuvant therapy in node-positive, human epidermal growth factor receptor 2-overexpressing breast cancer: NSABP B-31. J Clin Oncol 2005 23:7811-9. PMID 16258083
Romond EH, Jeong JH, Rastogi P, et al. Seven-year follow-up assessment of cardiac function in NSABP B-31, a randomized trial comparing doxorubicin and cyclophosphamide followed by paclitaxel (ACP) with ACP plus trastuzumab as adjuvant therapy for patients with node-positive, human epidermal growth factor receptor 2-positive breast cancer. J Clin Oncol 2012 30:3792-9. PMID 22987084
Perez EA, Suman VJ, Davidson NE, et al. Cardiac safety analysis of doxorubicin and cyclophosphamide followed by paclitaxel with or without trastuzumab in the North Central Cancer Treatment Group N9831 adjuvant breast cancer trial. J Clin Oncol 2008 26:1231-8. PMID 18250349
van der Voort A, van Ramshorst MS, van Werkhoven ED, et al. Three-year follow-up of neoadjuvant chemotherapy with or without anthracyclines in the presence of dual ERBB2 blockade in patients with ERBB2-positive breast cancer: a secondary analysis of the TRAIN-2 randomized, phase 3 trial. JAMA Oncol 2021 7:978-984. PMID 34014249
Hurvitz SA, et al. ASTEFANIA: adjuvant ado-trastuzumab emtansine and atezolizumab for high-risk, HER2-positive breast cancer. Future Oncol 2022 18:3563-3572. PMID 36382554
O'Sullivan CC, Ballman KV, McCall L, et al. Alliance A011801 (compassHER2 RD): postneoadjuvant T-DM1 + tucatinib/placebo in patients with residual HER2-positive invasive breast cancer. Future Oncol 2021 17:4665-4676. PMID 34636255
Huober J, Barrios CH, Niikura N, et al. Atezolizumab plus pertuzumab, trastuzumab, and chemotherapy in HER2-positive early breast cancer: final results of the phase III IMpassion050 trial. ESMO Open 2026 11:108541. PMID 42785163