Part IX · Early-stage disease: in situ, locally advanced, residual, and the neoadjuvant platform · Chapter 52

The neoadjuvant setting as a scientific platform

Why an endpoint that predicts a patient's outcome does not predict a drug's effect, and what the setting is good for instead.

1 · Pathologic complete response as an endpoint, what it does and does not predict

The platform rests on an older result. A meta-analysis of randomised trials found neoadjuvant and adjuvant chemotherapy equivalent for survival and disease progression, with more locoregional recurrence only where radiotherapy replaced surgery1. Neoadjuvant therapy raises breast conservation. In 354,204 patients in a national database, those with tumours larger than 3 cm were more likely to keep the breast after neoadjuvant chemotherapy, with an odds ratio of 1.72. Its pathological assessment and systemic questions were set out in a consensus statement3.

Two different claims travel under the same name, and almost every argument about pathological complete response comes from confusing them.

The first is individual-level prognosis. A patient with no residual invasive disease in breast and nodes after neoadjuvant therapy has a better outcome than a patient with residual disease. That claim is established beyond reasonable dispute.

The second is trial-level surrogacy. If a new treatment raises the proportion of patients achieving a complete response, that treatment will extend event-free or overall survival. This claim is not established, and the analysis designed to establish it did not.

The CTNeoBC pooled analysis obtained individual patient data from 12 international trials and analysed 11,955 patients4. Eradication of tumour from breast and nodes together was more strongly associated with outcome than eradication from the breast alone. Using the definition of no invasive disease in breast or nodes, the hazard ratio for event-free survival among patients achieving it was 0.48, with a 95% confidence interval of 0.43 to 0.54. For overall survival it was 0.36, with an interval of 0.31 to 0.42.

The pattern had been seen in single cohorts first. Triple-negative disease had a higher complete response rate than other subtypes, 22% against 11% among 1,118 patients, and worse survival when disease remained5. A pooled analysis of Japanese neoadjuvant trials confirmed the prognostic value of complete response and nodal status, and the predictive value of subtype6.

The association was strongest in the aggressive subtypes. In triple-negative disease the hazard ratio was 0.24 for event-free survival and 0.16 for overall survival. In HER2-positive, hormone receptor-negative disease treated with trastuzumab it was 0.15 and 0.084.

Then came the trial-level analysis in the same dataset. Correlating the increase in complete response rate between treatment groups with the improvement in outcome gave an R-squared of 0.03 for event-free survival, with a 95% confidence interval of 0.00 to 0.25. For overall survival it was 0.24, with an interval of 0.00 to 0.70. The authors stated plainly that the analysis could not validate pathological complete response as a surrogate endpoint4.

A second dataset reproduces the individual-level half and declines the other half in the same terms. Across 950 patients in an adaptive platform trial spanning nine investigational combinations, 330 achieved a complete response, which is 34.7%. Their three-year event-free and distant recurrence-free survival were both 95%, with hazard ratios of 0.19 and 0.21 against patients with residual disease. The effect was consistent across molecular subtypes and across regimens7. The authors noted that whether a therapy improving the response rate can be assumed to improve long-term outcome requires further study.

The regulatory position was built with that gap visible. Pathological complete response supports accelerated approval in high-risk early breast cancer, with confirmatory outcome evidence required afterwards8. That is a defensible trade of certainty for speed. It is not a claim that the surrogate has been validated, and the two are regularly reported as though they were the same statement.

What the endpoint does not predict is worth listing separately. It does not identify which patients with residual disease will relapse, which is the subject of Residual cancer burden and continuous response measures. It carries least information in hormone receptor-positive HER2-negative disease, where complete responses are uncommon and the prognostic separation is weakest4. And it says nothing about the size of a survival effect.

Caution

The failure is not one-directional, and both directions have now occurred in the same subtype.

In KEYNOTE-522, adding pembrolizumab to neoadjuvant chemotherapy raised the complete response rate from 51.2% to 64.8% among the first 602 randomised patients, a difference of 13.6 percentage points9. Event-free survival at 36 months was 84.5% against 76.8%, with a hazard ratio of 0.63 and a 95% confidence interval of 0.48 to 0.8210. Overall survival at 60 months was 86.6% against 81.7% among the full 1174 randomised patients, with P equal to 0.00211. The survival separation is larger than the response difference alone would predict, and part of the benefit accrues to patients who did not achieve a complete response.

In GeparNuevo, adding durvalumab produced a complete response rate of 53.4% against 44.2%, a difference that did not reach significance, with P equal to 0.28712. Three-year overall survival was 95.2% against 83.5%, with a hazard ratio of 0.24 and a 95% confidence interval of 0.08 to 0.72. Durvalumab was not continued after surgery, and multivariable analysis showed the survival effect was independent of the response effect13.

Those two results rest on 1174 and 174 randomised patients respectively, and the smaller one should carry proportionate weight. Together they show that the response endpoint can understate a benefit. The opposite error, a response gain with no survival gain, is the one that accelerated approval is designed to catch, and it is the reason confirmatory evidence is required.

2 · Residual cancer burden and continuous response measures

A binary endpoint derived from a continuous measurement discards most of the measurement. Residual cancer burden keeps it.

The index is built from five measurements of the surgical specimen. They are the two largest dimensions of the residual tumour bed, the proportion of that bed that is cellular, the proportion of that cellularity that is in situ, the number of positive nodes, and the diameter of the largest nodal deposit. The in situ proportion is not optional. It is what converts overall cellularity into the invasive cellularity the index actually uses, so omitting it inflates the score. Those five combine into a single continuous score14. The same enumeration is given in Residual cancer burden and post-treatment restaging. It was developed in 382 patients across two treatment cohorts. It remained independently prognostic for distant relapse-free survival after adjustment for age, clinical stage, hormone receptor status, endocrine therapy and binary response. The hazard ratio was 2.50, with a 95% confidence interval of 1.70 to 3.69.

Two findings from that first study shaped how the classes are used. Minimal residual disease, class I, occurred in 17% of patients and carried the same prognosis as a complete response. Extensive residual disease, class III, occurred in 13% and carried a poor prognosis regardless of hormone receptor status, adjuvant endocrine therapy or the pathological stage of the residual disease.

Long-term follow-up across five cohorts confirmed the gradient within each phenotype15. Ten-year relapse-free survival in triple-negative disease was 86%, 81%, 55% and 23% across complete response and classes I, II and III. In the trastuzumab-treated cohort it was 95%, 77%, 47% and 21%. In hormone receptor-positive HER2-negative disease in the combined cohorts it was 83%, 97%, 74% and 52%.

That last row deserves a comment rather than a quiet pass. Class I outperformed complete response in that subtype, which is not biologically sensible and is what small numbers at the ends of a distribution look like. A continuous index carries more information than a binary one and it is not noise-free.

External validation used individual patient data from 5161 patients treated across 12 institutes and trials, with a median follow-up of 56 months and 1164 event-free survival events16. A one-unit increase in the score was associated with worse event-free survival in every subtype. The univariable hazard ratio ranged from 1.55 in hormone receptor-positive HER2-negative disease, with a 95% confidence interval of 1.41 to 1.71, to 2.16 in hormone receptor-negative HER2-positive disease, with an interval of 1.79 to 2.61. Adjustment for age, grade, clinical T category and nodal status barely moved those estimates.

The measurement side deserves equal attention. The index requires the tumour bed to be identified, sampled to a defined protocol, and scored for cellularity. Cellularity is an estimate made by eye, and it is the weakest link in a score that is otherwise arithmetic. A residual cancer burden result carries the reproducibility of that estimate. Reporting the number without the protocol is reporting a measurement without its method.

3 · Response-adapted and de-escalation designs

A response-adapted design uses what happened during treatment to decide what happens after it. Two conditions have to hold. The response measure must identify a population with a different outcome, and a therapy must exist that changes that outcome.

Both were satisfied twice, and those two trials are why neoadjuvant treatment is now the default in high-risk disease rather than an option about timing.

KATHERINE enrolled 1486 patients with HER2-positive disease and residual invasive disease after neoadjuvant taxane and trastuzumab, randomising them to adjuvant trastuzumab emtansine or to trastuzumab17. Invasive disease or death occurred in 91 of 743 patients on the conjugate, which is 12.2%, against 165 of 743, which is 22.2%. Three-year invasive disease-free survival was 88.3% against 77.0%, with a hazard ratio of 0.50 and a 95% confidence interval of 0.39 to 0.64.

CREATE-X randomised 910 patients with HER2-negative residual invasive disease after neoadjuvant chemotherapy to standard treatment after surgery with or without capecitabine18. Five-year disease-free survival was 74.1% against 67.6%, with a hazard ratio of 0.70 and a 95% confidence interval of 0.53 to 0.92. Five-year overall survival was 89.2% against 83.6%, with a hazard ratio of 0.59. In the triple-negative subgroup disease-free survival was 69.8% against 56.1%. Hand-foot syndrome occurred in 73.4% of patients receiving capecitabine.

What made both of those work is worth naming, because it is not that the response measure was validated. It is that residual disease identified a population with enough events to prevent, and that a non-cross-resistant agent was available to give them.

De-escalation runs the other way and is much harder. Escalating on poor response risks over-treating a group that was high risk to begin with. De-escalating on good response risks under-treating a group whose good response was partly luck. The statistical asymmetry compounds the clinical one. An escalation trial waits for events to accumulate. A de-escalation trial has to demonstrate that events did not accumulate, which takes more patients and more time, and it turns on a non-inferiority margin that somebody has to choose Designing de-escalation trials and the non-inferiority margin problem.

Adaptive platform designs sit on top of all this. I-SPY 2 randomises patients to investigational regimens within molecular subtypes, updates allocation as responses accrue, and graduates a regimen on its predicted probability of succeeding in a later confirmatory trial7. The platform is efficient and its currency is the response endpoint, so the surrogate problem is inherited by the design rather than solved by it.

Caution

A successful response-adapted trial is regularly misread as validating the response measure. It does not. KATHERINE shows that a population defined by residual disease benefits from a specific drug17. It says nothing about whether a regimen producing more complete responses will produce longer survival. Those are different questions, and only the first has been answered.

4 · Window-of-opportunity studies and short-term biologic endpoints

A window-of-opportunity study gives a drug for a short defined period between diagnostic biopsy and definitive treatment, with tissue taken at both ends. The design buys a within-patient, before-and-after measurement of what a drug does to a human tumour. It costs weeks rather than years.

What it cannot buy is an outcome. A change in a proliferation index is a pharmacodynamic effect. Whether that effect translates into fewer recurrences is a separate question, and the window does not answer it.

POETIC demonstrates both halves in a single trial, which is why it is the example worth knowing. 4480 postmenopausal women with hormone receptor-positive operable breast cancer were randomised in a 2 to 1 ratio to 14 days of perioperative aromatase inhibitor or to none19. The intervention did not work. Breast cancer recurrence occurred in 434 of the 4480 women, and the hazard ratio was 0.92 with a 95% confidence interval of 0.75 to 1.12, P equal to 0.40. Freedom from recurrence at five years was 91.0% against 90.4%.

The measurement did work. Among treated HER2-negative patients, the five-year recurrence risk fell into three bands19. It was 4.3% when both the baseline and the two-week Ki67 were low, 8.4% when the baseline was high and the two-week value had fallen, and 21.5% when both were high. A near fivefold separation in risk was produced by a two-week exposure to a drug that changed nothing about outcome when given that way.

That is the clearest available statement of what these studies deliver. The treatment was useless and the information was valuable. A design that conflates the two reads the result backwards.

The constraint on all of this is the reproducibility of the endpoint. Ki67 scoring varies substantially between laboratories and observers even under a common protocol20, which is why successive working group statements have concentrated on standardising the method rather than on the cut-point21,22. A two-week change is the difference between two measurements that each carry that variation.

Short exposures also change the target being measured. Brief neoadjuvant endocrine therapy frequently increases HER2 protein expression in hormone receptor-positive, HER2-negative tumours23. The biomarker measured after a window is therefore not interchangeable with the biomarker measured before it. In an era where HER2 expression level determines eligibility Expression and alteration-defined treatment across subtypes, that is a practical point rather than a theoretical one.

5 · Paired sampling and biomarker discovery under treatment pressure

The neoadjuvant setting is the only routine circumstance in which a human tumour is sampled before a known exposure and then removed whole afterwards. Nothing else in solid tumour oncology provides that.

Paired material answers questions that single samples cannot. Which features predict response before treatment starts. Which features change under treatment. And whether the change itself carries information beyond the baseline value, which is the POETIC result in Window-of-opportunity studies and short-term biologic endpoints.

Immune features are the best-worked example. Stromal tumour-infiltrating lymphocytes measured at baseline predict pathological complete response and prognosis, across 3771 patients treated with neoadjuvant therapy in a pooled analysis24, and the prognostic relationship holds in early triple-negative disease25. The scoring method is standardised sufficiently to be compared across studies26. Chemotherapy then changes both the infiltrate and PD-L1 expression27, so a post-treatment immune measurement is not a delayed version of the pre-treatment one.

Genomic features behave the same way. PIK3CA mutation is associated with a lower complete response rate in HER2-positive disease28. Homologous recombination deficiency score predicts response to platinum-containing neoadjuvant chemotherapy in triple-negative disease29. HER2 heterogeneity assessed at baseline separated a group in which no patient achieved a complete response from a group in which 55% did30.

Two sampling caveats govern how far any of this generalises. The baseline core samples one region of a tumour that may be spatially structured Spatial heterogeneity. The surgical specimen is the surviving population rather than the original one, so a feature enriched in residual disease may have been selected rather than induced Temporal heterogeneity and clonal evolution.

In practice

Collect and bank both specimens, including the residual tumour, in every neoadjuvant trial and wherever local practice allows it outside one.

Record the interval between the last dose and surgery. Biological changes measured in a specimen removed two weeks after the last cycle are not the same as those measured at four.

Report the sampling scheme alongside the biomarker result. A percentage from a single core and a percentage from a whole section are different quantities.

A trial that discards the residual specimen has destroyed the most informative tissue in the disease, and no amount of later analysis recovers it.

6 · Neoadjuvant therapy as an observable experiment in tumor evolution

Everything in Temporal heterogeneity and clonal evolution about therapy as selection is inferred indirectly. The evidence comes from metastatic samples taken at different times, from different lesions, under exposures that overlap. In the neoadjuvant setting the same process runs on a known schedule against a tumour that is then removed intact.

That converts several inferences into measurements.

Which subclones survive becomes observable, because the surviving population is the specimen on the pathologist's bench. The composition of residual disease is a direct read-out of what the regimen failed to kill rather than a reconstruction from a later biopsy.

Whether survival was genetic or a reversible state change becomes approachable, because both the starting and the surviving populations are available for sequencing and profiling Genetic evolution and reversible cell state change as different problems. The residual population can be compared against that patient's own baseline rather than against a cohort average.

The rate of change becomes measurable, because circulating tumour DNA fills in the time axis between the two tissue points. Clearance during treatment and persistence after surgery both carry prognostic information, and high-sensitivity structural-variant assays detect residual disease in the post-neoadjuvant setting31. What plasma can and cannot see is set out in Liquid biopsy and longitudinal monitoring.

The limit of the experiment is where it stops. The tumour is resected at the end of the window, so the neoadjuvant setting can show what treatment selects and cannot show what the selected population would have done over the following decade. That population has been removed.

Which is precisely why the surrogate question in Pathologic complete response as an endpoint, what it does and does not predict cannot be settled from inside this setting. The question is about what happens after the specimen leaves the patient. Answering it requires long follow-up in randomised populations, and an explicit account of what would count as validation. That is the subject of Endpoints, surrogates, and the overall survival question and the thread resolves in Clinical trials and evidence interpretation.

Interplay

The neoadjuvant setting is where the heterogeneity argument of Part VI becomes an experiment rather than a description. A tumour with a spatially or clonally structured antigen is predicted to respond incompletely, and that prediction was tested in tissue rather than inferred from imaging30. The mean expression determined eligibility and the variance determined what survived. Read this way, residual cancer burden is a measure of how much of the tumour was outside the reach of the regimen, expressed as a number14. Every other setting in this book measures response through a shadow. This one measures it in the thing itself, once, and then the experiment ends.

See Integrative biological interplay

References

All 31 references
  1. Mauri D, Pavlidis N, Ioannidis JP. Neoadjuvant versus adjuvant systemic treatment in breast cancer: a meta-analysis. J Natl Cancer Inst 2005 97:188-94. PMID 15687361
  2. Killelea BK, Yang VQ, Mougalian S, et al. Neoadjuvant chemotherapy for breast cancer increases the rate of breast conservation: results from the National Cancer Database. J Am Coll Surg 2015 220:1063-9. PMID 25868410
  3. Gralow JR, Burstein HJ, Wood W, et al. Preoperative therapy in invasive breast cancer: pathologic assessment and systemic therapy issues in operable disease. J Clin Oncol 2008 26:814-9. PMID 18258991
  4. Cortazar P, Zhang L, Untch M, et al. Pathological complete response and long-term clinical benefit in breast cancer: the CTNeoBC pooled analysis. Lancet 2014 384:164-172. PMID 24529560
  5. Liedtke C, Mazouni C, Hess KR, et al. Response to neoadjuvant therapy and long-term survival in patients with triple-negative breast cancer. J Clin Oncol 2008 26:1275-81. PMID 18250347
  6. Kuroi K, Toi M, Ohno S, et al. Prognostic significance of subtype and pathologic response in operable breast cancer; a pooled analysis of prospective neoadjuvant studies of JBCRG. Breast Cancer 2013 22:486-95. PMID 24338638
  7. I-SPY2 Trial Consortium, Yee D, DeMichele AM, et al. Association of event-free and distant recurrence-free survival with individual-level pathologic complete response in neoadjuvant treatment of stages 2 and 3 breast cancer: three-year follow-up analysis for the I-SPY2 adaptively randomized clinical trial. JAMA Oncol 2020 6:1355-1362. PMID 32701140
  8. Prowell TM, Pazdur R. Pathological complete response and accelerated drug approval in early breast cancer. N Engl J Med 2012 366:2438-2441. PMID 22646508
  9. Schmid P, Cortes J, Pusztai L, et al. Pembrolizumab for early triple-negative breast cancer. N Engl J Med 2020 382:810-821. PMID 32101663
  10. Schmid P, Cortes J, Dent R, et al. Event-free survival with pembrolizumab in early triple-negative breast cancer. N Engl J Med 2022 386:556-567. PMID 35139274
  11. Schmid P, Cortes J, Dent R, et al. Overall survival with pembrolizumab in early-stage triple-negative breast cancer. N Engl J Med 2024 391:1981-1991. PMID 39282906
  12. Loibl S, Untch M, Burchardi N, et al. A randomised phase II study investigating durvalumab in addition to an anthracycline taxane-based neoadjuvant therapy in early triple-negative breast cancer: clinical results and biomarker analysis of GeparNuevo study. Ann Oncol 2019 30:1279-1288. PMID 31095287
  13. Loibl S, Schneeweiss A, Huober J, et al. Neoadjuvant durvalumab improves survival in early triple-negative breast cancer independent of pathological complete response. Ann Oncol 2022 33:1149-1158. PMID 35961599
  14. Symmans WF, Peintinger F, Hatzis C, et al. Measurement of residual breast cancer burden to predict survival after neoadjuvant chemotherapy. J Clin Oncol 2007 25:4414-4422. PMID 17785706
  15. Symmans WF, Wei C, Gould R, et al. Long-term prognostic risk after neoadjuvant chemotherapy associated with residual cancer burden and breast cancer subtype. J Clin Oncol 2017 35:1049-1060. PMID 28135148
  16. Yau C, Osdoit M, van der Noordaa M, et al. Residual cancer burden after neoadjuvant chemotherapy and long-term survival outcomes in breast cancer: a multicentre pooled analysis of 5161 patients. Lancet Oncol 2022 23:149-160. PMID 34902335
  17. 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
  18. Masuda N, Lee SJ, Ohtani S, et al. Adjuvant capecitabine for breast cancer after preoperative chemotherapy. N Engl J Med 2017 376:2147-2159. PMID 28564564
  19. Smith I, Robertson J, Kilburn L, et al. Long-term outcome and prognostic value of Ki67 after perioperative endocrine therapy in postmenopausal women with hormone-sensitive early breast cancer (POETIC): an open-label, multicentre, parallel-group, randomised, phase 3 trial. Lancet Oncol 2020 21:1443-1454. PMID 33152284
  20. Polley MY, Leung SC, McShane LM, et al. An international Ki67 reproducibility study. J Natl Cancer Inst 2013 105:1897-1906. PMID 24203987
  21. Dowsett M, Nielsen TO, A'Hern R, et al. Assessment of Ki67 in breast cancer: recommendations from the International Ki67 in Breast Cancer working group. J Natl Cancer Inst 2011 103:1656-1664. PMID 21960707
  22. Nielsen TO, Leung SCY, Rimm DL, et al. Assessment of Ki67 in breast cancer: updated recommendations from the International Ki67 in Breast Cancer Working Group. J Natl Cancer Inst 2021 113:808-819. PMID 33369635
  23. Chaudhary LN, et al. Frequent upregulation of HER2 protein in hormone-receptor-positive HER2-negative breast cancer after short-term neoadjuvant endocrine therapy. Breast Cancer Res Treat 2023. PMID 37460683
  24. Denkert C, von Minckwitz G, Darb-Esfahani S, et al. Tumour-infiltrating lymphocytes and prognosis in different subtypes of breast cancer: a pooled analysis of 3771 patients treated with neoadjuvant therapy. Lancet Oncol 2018 19:40-50. PMID 29233559
  25. Loi S, Drubay D, Adams S, et al. Tumor-infiltrating lymphocytes and prognosis: a pooled individual patient analysis of early-stage triple-negative breast cancers. J Clin Oncol 2019 37:559-569. PMID 30650045
  26. Salgado R, Denkert C, Demaria S, et al. The evaluation of tumor-infiltrating lymphocytes (TILs) in breast cancer: recommendations by an International TILs Working Group 2014. Ann Oncol 2015 26:259-271. PMID 25214542
  27. Pelekanou V, Carvajal-Hausdorf DE, Altan M, et al. Effect of neoadjuvant chemotherapy on tumor-infiltrating lymphocytes and PD-L1 expression in breast cancer and its clinical significance. Breast Cancer Res 2017 19:91. PMID 28784153
  28. Zagami P, et al. Association of PIK3CA mutation with pathologic complete response and outcome by hormone receptor status and intrinsic subtype in early-stage HER2-positive breast cancer (CALGB 40601). JAMA Netw Open 2023. PMID 38117494
  29. Telli ML, Timms KM, Reid J, et al. Homologous recombination deficiency (HRD) score predicts response to platinum-containing neoadjuvant chemotherapy in patients with triple-negative breast cancer. Clin Cancer Res 2016 22:3764-3773. PMID 26957554
  30. Metzger Filho O, Viale G, Stein S, et al. Impact of HER2 heterogeneity on treatment response of early-stage HER2-positive breast cancer: phase II neoadjuvant clinical trial of T-DM1 combined with pertuzumab. Cancer Discov 2021 11:2474-2487. PMID 33941592
  31. Blaye C, Bidard FC, Hue L, et al. Tracking structural variants in early breast cancer using a high-sensitivity ctDNA assay predicts relapse in the post-neoadjuvant setting: the multicenter ALIENOR trial. ESMO Open 2026 11:107766. PMID 42242004