The Cost of Not Having an IEP: Three Failure Stories

A hazard ratio of 0.48, yet somehow buried in the same trial, the same year, the same drug: a HR of 2.85.[1]

One hazard ratio said the drug very nearly halved the risk of disease progression. The other said it very nearly tripled it. Same molecule, same randomised trial, opposite verdict, and for years both numbers sat inside a single pooled result, because nobody had split the population by the one variable that actually mattered.

That reversal didn't need a new discovery to explain it. The relevant biology had already been published, years before anyone ran the trial that finally proved it out. The drug is gefitinib, and it's one of three stories in this piece, because its shape, a foreseeable answer arriving years late, repeats twice more, in two entirely different failure modes.

This post argues that the cost of not having an Integrated Evidence Plan is not abstract. Three real, distinct programmes (bevacizumab in breast cancer, gefitinib in lung cancer, gemtuzumab ozogamicin in leukaemia) each lost years, and in each case the fault line that eventually forced a withdrawal or a restriction was visible, in some form, before the trial that exposed it at scale: an unconverted surrogate, a buried biomarker, an undersized safety database. Three different kinds of question. All three asked too late.

What is an Integrated Evidence Plan actually for?

An Integrated Evidence Plan exists to ask a narrower, more falsifiable question than "will this drug work", early enough that the answer costs a workshop, not a confirmatory trial: whether the surrogate we're using actually predicts the benefit regulators will require, whether we've found the subgroup that responds or are averaging it away inside the whole population, and whether the safety database is large enough to see the thing that will eventually hurt someone. We've made the broader case elsewhere for why every programme needs one; this piece is about what happens to the ones that don't.

All three cases below ran on the FDA's accelerated approval pathway, which exists specifically because the agency knows some surrogates don't convert into real benefit; the confirmatory question is already built into the pathway. What follows is what happens when that built-in question isn't pinned down tightly enough before the confirmatory trial starts.

Progression-Free Isn't Consequence-Free

In February 2008, the FDA granted bevacizumab accelerated approval, in combination with paclitaxel, for first-line metastatic HER2-negative breast cancer.[2] The entire basis was a single trial, E2100 (NCT00028990, 722 women):[3] median progression-free survival of 11.8 months on bevacizumab plus paclitaxel against 5.9 months on paclitaxel alone, a hazard ratio of 0.60 for progression or death by investigator assessment, 0.48 by the trial's independent radiology review, slightly stronger still.[4][5] Overall survival, in that same trial, did not differ between arms. The approval went ahead anyway, on the logic built into the accelerated pathway itself: the PFS effect was real and large, and two confirmatory trials would settle whether it converted into something patients actually lived longer for.

It didn't settle quickly, and it didn't settle in the drug's favour. AVADO (NCT00333775, 736 patients) and RIBBON-1 (NCT00262067, 1,237 patients), roughly 1,973 women between them, each replicated a smaller version of the same PFS gain against different chemotherapy backbones, and neither found an overall-survival or quality-of-life benefit.[6][7] A 2020 systematic review and meta-analysis, pooling 52 studies of bevacizumab in first-line metastatic breast cancer, put a number on the whole pattern: cumulative PFS hazard ratio 0.72, cumulative OS hazard ratio 0.90, a real, reproducible effect on the surrogate that never translated statistically into survival.[8] The same authors made the point that should have landed earlier: "the fact that seven clinical trials are insufficient to conclude validity (or lack thereof) for a trial-level surrogate suggests that it would be more efficient to conduct trials using the more clinically meaningful endpoints."[8]

FDA's Oncologic Drugs Advisory Committee (ODAC) recommended withdrawal in July 2010; Genentech appealed and won a second hearing; the committee voted to withdraw again in June 2011; FDA Commissioner Margaret A. Hamburg formally revoked the breast-cancer indication, not the drug itself, in November 2011 Avastin stayed approved for colorectal, lung, kidney and glioblastoma indications throughout.[9][10] Call it roughly three and a half years from accelerated approval to final word, and two large confirmatory trials' worth of patients to get there.

Here's the honest asymmetry this case demands: no dosing tweak and no population fix existed for bevacizumab. Unlike the other two cases here, the underlying effect itself was the ceiling. What a tighter evidence plan could plausibly have changed here is narrower and less dramatic than "catch the biomarker" or "catch the safety signal": pre-specify, before AVADO and RIBBON-1 even opened, exactly what magnitude of OS or quality-of-life benefit would count as confirmation. Instead, an ambiguous result, a real PFS effect sitting next to a flat OS curve, triggered a second, multi-year fight over a bar nobody had actually agreed to in advance.

A real effect. Not the effect anyone needed.

One Trial, Two Truths

Gefitinib's accelerated approval, in 2003, rested on tumour response rate alone, an even softer surrogate than progression-free survival, in an unselected population of patients who had already failed platinum-based chemotherapy and docetaxel, at a response rate researchers put in the 12-18% range depending on dose.[1] The confirmatory trial FDA required, ISEL (NCT00242801, 1,692 patients),[11][12] was designed the same way, unselected, because in 2003 there was no validated biomarker to select on.

That changed almost immediately, and inconveniently. In 2004, the same year ISEL was reading out, two groups independently published the biology gefitinib's regulatory story had been missing. Lynch and colleagues found activating EGFR tyrosine kinase domain mutations in the tumours of patients who had responded to gefitinib, and in none of the non-responders they tested.[13] Paez and colleagues found the same mutation, independently, at far higher frequency in Japanese tumours than American ones (15 of 58 tumours, against 1 of 61), a population difference that tracked, suggestively, with response.[14] The mechanism explaining who gefitinib actually worked for existed in the published literature before ISEL's result was even public.

ISEL missed its primary endpoint: overall survival hazard ratio 0.89, not statistically significant, against placebo plus best supportive care.[12] But its own pre-specified subgroups were already gesturing at the answer: never-smokers, hazard ratio 0.67; patients of Asian ethnicity, hazard ratio 0.66, both crude, clinical proxies for a biomarker nobody could yet test for at scale, visible in the data at the exact moment the trial delivered its negative headline.[12] FDA restricted gefitinib's label in 2005 to the patients already benefiting from it, then withdrew the original, unselected-population approval entirely in 2012, when the confirmatory trial for that broader population still hadn't shown an overall-survival benefit.[2]

It took IPASS (NCT00322452, 1,217 patients),[15][16] five years later, to test the biomarker directly rather than by proxy, and that's where this piece's opening numbers come from. Among mutation-positive patients, gefitinib very nearly halved the risk of progression against standard chemotherapy: hazard ratio 0.48. Among mutation-negative patients, it very nearly tripled it: hazard ratio 2.85.[1] Same drug, same trial, opposite verdict, hidden inside one pooled result for as long as nobody split it by genotype. IFUM (NCT01203917, 1,060 patients screened for EGFR mutation status across European sites)[17] then supplied the biomarker-selected safety and efficacy package behind gefitinib's re-approval, in 2015, for patients with EGFR exon 19 deletions or exon 21 L858R mutations, alongside a companion diagnostic.[18]

Twelve years, start to finish. Eleven, from the biology's publication to its regulatory operationalisation. The subgroup was never hiding especially well.

The Signal Was Already There, in Miniature

Gemtuzumab ozogamicin's accelerated approval, granted May 17, 2000, for CD33-positive relapsed AML in patients 60 and over, rested on pooled response-rate data from three small, uncontrolled Phase 2 studies (one of them, NCT00003673, enrolled 55 patients on its own). At the time of that approval decision, the interim data cut in front of the FDA covered 142 patients, at a dose of 9 mg/m² given twice, 14 days apart, with an overall response rate of around 30%; the same three studies would go on, at final completed enrolment, to treat 277 patients in total, with a 26% response rate.[19][20][27][29] The efficacy case was thinner than either of the other two cases here, but what actually matters was already sitting inside the same dataset that supported approval: 31% of patients had abnormal liver enzymes, and one patient died of liver failure.[27] That hepatotoxicity signal, at that exact dose, was visible in miniature at the moment of approval. Nobody sized it as a safety-database question. It was read, instead, as an efficacy result with a manageable side effect.

SWOG S0106 (NCT00085709), the confirmatory Phase 3 trial FDA required, changed three things at once from every other trial in the programme: the dose (6 mg/m² rather than 9 mg/m²), the day it was given (day 4, not day 1), and the chemotherapy backbone.[21][22] It enrolled 637 patients, more than twice the size of the entire original approval dataset, and it was halted early: 17 of 295 patients on gemtuzumab ozogamicin died during induction, against 4 of 300 on standard therapy, with no improvement in remission or survival to offset the excess deaths.[21] FDA requested a voluntary withdrawal; the drug came off the US market in October 2010.[2]

The fix, when it came, was a change to the dose alone. ALFA-0701 (NCT00927498, 280 patients)[23][24] tested a fractionated regimen, 3 mg/m² on days 1, 4 and 7, a lower total dose spread further apart, and found a genuine event-free-survival benefit, hazard ratio 0.66 (95% CI 0.49-0.89), without reproducing the excess early mortality.[25] A smaller companion trial, MyloFrance-1, reported no cases of VOD, the veno-occlusive disease that had haunted the original regimen.[28] FDA re-approved gemtuzumab ozogamicin, at the corrected dose, on September 2, 2017.[26][28]

The review that reconstructs this history doesn't soften it: "This highlights the potential pitfalls when basing conditional approval of new drugs on a single trial."[19]

Ten years and five months on the market at a dose later shown to add mortality without adding benefit. A further seven years off the market while the correction was established and reviewed. Over seventeen years, start to finish, and unlike the other two cases, no company-disclosed dollar figure to attach to any of it; better to say so than to invent one. What there is, instead, is a hepatotoxicity signal that a decade of confirmatory-trial design, three separate protocol changes, and 637 randomised patients were needed to formally rediscover.

The dataset had already said so, in miniature, a decade earlier.

The Pre-Mortem Nobody Ran

In every one of these cases, the fault line that eventually forced a withdrawal or a restriction was visible, in some form, before the confirmatory or pivotal trial that exposed it at scale. Bevacizumab's accelerated approval was explicitly a bet that a real PFS effect would convert to survival, a bet the FDA's own accelerated pathway exists to allow precisely because it sometimes doesn't pay off. Gefitinib's causal biology was published the same year its confirmatory trial was reading out an unselected population. Gemtuzumab ozogamicin's hepatotoxicity signal was sitting in the same Phase 2 dataset that won its approval, at the same dose that later proved fatal in a much larger population.

The obvious objection is hindsight bias: of course these gaps look foreseeable now, because the ending is already known. Regulators and sponsors, on this view, were making reasonable calls with the evidence available at the time, and no evidence plan reads the future. It's a fair objection. That said, it doesn't survive contact with the record in two of these three cases. The EGFR-mutation biology was already public: published, peer-reviewed, the same year ISEL's unselected population was reading out its result. Gemtuzumab ozogamicin's hepatotoxicity signal was sitting in plain sight: 31% abnormal liver enzymes and a death, inside the sponsor's own approval dataset. Bevacizumab is the one case that doesn't clear this stricter bar, which is exactly why its lesson stays narrower than the other two.

Call it the pre-mortem nobody ran: ask, before the pivotal trial, why the whole programme might fail, while the answer still costs a workshop or an early biomarker assay, rather than reconstruct it years later, as an autopsy, at the cost of a decade of a drug's commercial life and, in gemtuzumab ozogamicin's case, lives.

That's the specific job an Integrated Evidence Plan exists to do: apply the same triage logic behind the minimum viable evidence framework to a single decision, not an entire programme.

I'd argue gemtuzumab ozogamicin is the sharpest of the three: the warning was sitting inside the sponsor's own approval dataset, not in some later trial nobody could have designed yet. Whether its outcome ranks as the worst of the three is beside the point.

Sometimes, though, the pre-mortem's honest answer is "we don't yet know if this converts to real benefit" bevacizumab is proof this isn't a formula that guarantees a fix. The discipline that matters there is refusing to let an ambiguous result get waved through on hope. For a biotech that can't afford to run a second multi-year confirmatory trial the way Genentech or Pfizer could, that discipline is exactly what running tight on runway makes non-negotiable. This blog has already told a version of this story once, in the specific case of aducanumab; this piece exists because three more names make the pattern harder to wave away as a one-off.

Every one of these three drugs eventually reached patients who genuinely benefited from it, at the right dose, in the right population, against the right bar. Gefitinib works, in patients with EGFR-mutant lung cancer. Gemtuzumab ozogamicin works, at 3 mg/m² on days 1, 4 and 7. Bevacizumab still works, in the tumour types where the confirmatory trials actually confirmed something. None of that had to take twelve years, or seventeen, or cost a drug its market for the better part of a decade. The question that eventually got asked was always askable earlier. Nobody asked it.

References

[1] Armour AA, Watkins CL. (2010). "The challenge of targeting EGFR: experience with gefitinib in nonsmall cell lung cancer." European Respiratory Review;19(117):186-196. PMID: 20956191. https://pubmed.ncbi.nlm.nih.gov/20956191/

[2] Hakariya H, Ozaki A, Tanimoto T. (2025). "US FDA-accelerated approvals and subsequent withdrawals: influence on Japanese clinical oncology practice guidelines." Investigational New Drugs;43(2):311-317. PMID: 40178688. https://pubmed.ncbi.nlm.nih.gov/40178688/

[3] Eastern Cooperative Oncology Group. "A Randomized Phase III Trial Of Paclitaxel Versus Paclitaxel Plus Bevacizumab (rhuMAb VEGF) As First-Line Therapy For Locally Recurrent or Metastatic Breast Cancer" (E2100). ClinicalTrials.gov: NCT00028990. https://clinicaltrials.gov/study/NCT00028990

[4] Miller K, Wang M, Gralow J, et al. (2007). "Paclitaxel plus Bevacizumab versus Paclitaxel Alone for Metastatic Breast Cancer." New England Journal of Medicine;357:2666-2676. PMID: 18160686. https://pubmed.ncbi.nlm.nih.gov/18160686/

[5] Gray R, Bhattacharya S, Bowden C, Miller K, Comis RL. (2009). "Independent review of E2100: a phase III trial of bevacizumab plus paclitaxel versus paclitaxel in women with metastatic breast cancer." Journal of Clinical Oncology;27(30):4966-4972. PMID: 19720913. https://pubmed.ncbi.nlm.nih.gov/19720913/

[6] Hoffmann-La Roche. "AVADO" trial. ClinicalTrials.gov: NCT00333775. https://clinicaltrials.gov/study/NCT00333775

[7] Genentech, Inc. "RIBBON-1" trial. ClinicalTrials.gov: NCT00262067. https://clinicaltrials.gov/study/NCT00262067

[8] Hey SP, Gyawali B, D'Andrea E, Kanagaraj M, Franklin JM, Kesselheim AS. (2020). "A Systematic Review and Meta-Analysis of Bevacizumab in First-Line Metastatic Breast Cancer: Lessons for Research and Regulatory Enterprises." Journal of the National Cancer Institute;112(4):335-342. PMID: 31651981. https://pubmed.ncbi.nlm.nih.gov/31651981/

[9] Preusser M, Fülöp G, Berghoff AS, Heinzl H, Steger GG, Greil R, Zielinski CC, Bartsch R. (2012). "Influence of the American ODAC statement on Austrian bevacizumab prescribing practice for metastatic breast cancer." The Oncologist;17(7):e13-7. PMID: 22744818. https://pubmed.ncbi.nlm.nih.gov/22744818/

[10] FDA Commissioner Margaret A. Hamburg, announcement of the revocation of bevacizumab's metastatic breast cancer indication, November 2011, reported contemporaneously in ASCO Post, "FDA Announces Bevacizumab Decision."

[11] AstraZeneca. "A Double Blind, Placebo Controlled, Parallel Group, Multicentre, Randomised, Phase III Survival Study Comparing ZD1839 (IRESSA)... In Patients With Advanced NSCLC" (ISEL). ClinicalTrials.gov: NCT00242801. https://clinicaltrials.gov/study/NCT00242801

[12] Thatcher N, Chang A, Parikh P, et al. (2005). "Gefitinib plus best supportive care in previously treated patients with refractory advanced non-small-cell lung cancer: results from a randomised, placebo-controlled, multicentre study (Iressa Survival Evaluation in Lung Cancer)." Lancet;366(9496):1527-1537. PMID: 16257339. https://pubmed.ncbi.nlm.nih.gov/16257339/

[13] Lynch TJ, Bell DW, Sordella R, et al. (2004). "Activating mutations in the epidermal growth factor receptor underlying responsiveness of non-small-cell lung cancer to gefitinib." New England Journal of Medicine;350(21):2129-2139. PMID: 15118073. https://pubmed.ncbi.nlm.nih.gov/15118073/

[14] Paez JG, Jänne PA, Lee JC, et al. (2004). "EGFR mutations in lung cancer: correlation with clinical response to gefitinib therapy." Science;304(5676):1497-1500. PMID: 15118125. https://pubmed.ncbi.nlm.nih.gov/15118125/

[15] Mok TS, Wu YL, Thongprasert S, et al. (2009). "Gefitinib or Carboplatin-Paclitaxel in Pulmonary Adenocarcinoma." New England Journal of Medicine;361:947-957. PMID: 19692680. https://pubmed.ncbi.nlm.nih.gov/19692680/

[16] AstraZeneca. "IPASS" trial. ClinicalTrials.gov: NCT00322452. https://clinicaltrials.gov/study/NCT00322452

[17] AstraZeneca. "IFUM" trial. ClinicalTrials.gov: NCT01203917. https://clinicaltrials.gov/study/NCT01203917

[18] AstraZeneca, press release, "Iressa approved by FDA for the 1st-line treatment of NSCLC patients whose tumours have EGFR mutations," 13 July 2015. https://www.astrazeneca.com/media-centre/press-releases/2015/iressa-fda-approved-non-small-cell-lung-cancer-treatment-13072015.html

[19] Egan PC, Reagan JL. (2018). "The return of gemtuzumab ozogamicin: a humanized anti-CD33 monoclonal antibody-drug conjugate for the treatment of newly diagnosed acute myeloid leukemia." OncoTargets and Therapy;11:8265-8272. PMID: 30538495. https://pubmed.ncbi.nlm.nih.gov/30538495/

[20] Pfizer/Wyeth. "A Study of the Safety of CMA-676 in Treatment of Elderly Patients With Acute Myeloid Leukemia (AML) in First Relapse." ClinicalTrials.gov: NCT00003673. https://clinicaltrials.gov/study/NCT00003673

[21] Petersdorf SH, Kopecky KJ, Slovak M, et al. (2013). "A phase 3 study of gemtuzumab ozogamicin during induction and postconsolidation therapy in younger patients with acute myeloid leukemia." Blood;121(24):4854-4860. PMID: 23591789. https://pubmed.ncbi.nlm.nih.gov/23591789/

[22] SWOG Cancer Research Network. "SWOG S0106" trial. ClinicalTrials.gov: NCT00085709. https://clinicaltrials.gov/study/NCT00085709

[23] Acute Leukemia French Association. "ALFA-0701" trial. ClinicalTrials.gov: NCT00927498. https://clinicaltrials.gov/study/NCT00927498

[24] Castaigne S, Pautas C, Terré C, et al. (2012). "Effect of gemtuzumab ozogamicin on survival of adult patients with de-novo acute myeloid leukaemia (ALFA-0701): a randomised, open-label, phase 3 study." Lancet;379(9825):1508-1516. PMID: 22482940. https://pubmed.ncbi.nlm.nih.gov/22482940/

[25] Lambert J, Pautas C, Terré C, et al. (2019). "Gemtuzumab ozogamicin for de novo acute myeloid leukemia: final efficacy and safety updates from the open-label, phase III ALFA-0701 trial." Haematologica;104(1):113-119. PMID: 30076173. https://pubmed.ncbi.nlm.nih.gov/30076173/

[26] Pfizer, press release, "Pfizer Receives FDA Approval For Mylotarg (gemtuzumab ozogamicin)," September 2017. https://www.pfizer.com/news/press-release/press-release-detail/pfizer_receives_fda_approval_for_mylotarg_gemtuzumab_ozogamicin

[27] Bross PF, Beitz J, Chen G, Chen XH, Duffy E, Kieffer L, Roy S, Sridhara R, Rahman A, Williams G, Pazdur R. (2001). "Approval summary: gemtuzumab ozogamicin in relapsed acute myeloid leukemia." Clinical Cancer Research;7(6):1490-1496. PMID: 11410481. https://pubmed.ncbi.nlm.nih.gov/11410481/

[28] Norsworthy KJ, Ko CW, Lee JE, Liu J, John CS, Przepiorka D, Farrell AT, Pazdur R. (2018). "FDA Approval Summary: Mylotarg for Treatment of Patients with Relapsed or Refractory CD33-Positive Acute Myeloid Leukemia." The Oncologist;23(9):1103-1108. PMID: 29650683. https://pubmed.ncbi.nlm.nih.gov/29650683/

[29] Larson RA, Sievers EL, Stadtmauer EA, et al. (2005). "Final report of the efficacy and safety of gemtuzumab ozogamicin (Mylotarg) in patients with CD33-positive acute myeloid leukemia in first recurrence." Cancer;104(7):1442-1452. PMID: 16116598. https://pubmed.ncbi.nlm.nih.gov/16116598/