The Decade in Oncology: What 57,603 ASCO Abstracts Reveal between 2016-2025

I used 30 AI agents to analyze every ASCO annual meeting abstracts in the past 11 years. Here are 6 top line findings

Catching up with oncology is like trying to board a bullet train that left the station two years ago. Everything moves so fast. Missing one AACR → ASCO → ESMO cycle = losing the ability to cite the top-line OS/PFS/ORR bars that keep oncology conversations relevant.

Oncology used to be my home ground. I spent my PhD in cancer biology getting to know nearly all the benchmark numbers and emerging targets. But a few years after grad school, while I could still talk oncology at a dinner party, I could no longer remember what an ORR in metastatic melanoma post Nivo+Ipi counts as impressive, without some digging.

Thus, “catching up with oncology” was at the top of my wish list during my parental leave, while the newborn was napping. Fortunately, AI makes this kind of project possible. I put together a team of ~30 AI agents to compile, fact-check, synthesize, and distill 57,603 ASCO abstracts from the past 11 years (2015–2025).

Here are some top-line impressions from this exercise.

---

Finding 1: There Are Two Oncologies, and the Gap Is Widening

A treatment-naive CLL patient on zanubrutinib has roughly an 80% chance of being progression-free at 5 years. A multiple myeloma patient on the PERSEUS regimen has an 84.3% PFS rate at 4 years, with 87.9% achieving ≥CR (complete response or better). A Hodgkin lymphoma patient on N-AVD has 92% PFS at 2 years. For many of these patients, we are talking about functional cures.

A metastatic pancreatic cancer patient receives FOLFIRINOX and lives 11.1 months. A glioblastoma patient follows the Stupp protocol and lives ~15 months on a regimen that has not changed in twenty years.

The divergence between heme and solid tumor outcomes is the defining structural feature of modern oncology, and in my opinion, it is driven by target biology that defines therapeutic index.

Blood cancers have three structural advantages.

• The dispensable lineage: you can ablate every B cell in a patient’s body without killing the patient, so CD19, CD20, BCMA, CD38, and BTK are all fair game.

• The clean surface antigen: blood cancer cells express well-characterized, accessible targets at high density. T

• he modality advantage: antibody-based modalities (mAb, TCE, ADC, Radioligand, CAR-T) benefit from target overexpression, while small molecules targeting intracellular enzymes face the opposite problem.

The data tracks this. CAR-T abstracts grew from 21 to 145. Bispecifics grew from 15 to 149, with ASCO 2025 adding the first-in-class trispecifics — ISB 2001 (BCMA × CD38 × CD3, ORR 82%) and JNJ-5322 (100% ORR at RP2D in BCMA/GPRC5D-naive myeloma). BCMA went from zero mentions in 2015 to 42 in 2023, powering two CAR-Ts, two bispecifics, and multiple ADCs. The heme toolkit compounds.

The solid tumor side is different. Pancreatic cancer ran 1,924 ASCO abstracts over the 11 years (250 in ASCO 2025 alone). Median OS is still 11.1 months. The dominant driver is KRAS, and cancer cells are behind dense stroma. Every antibody-based approach hits a wall. Daraxonrasib, the RAS(ON) multi-selective inhibitor, just entered Phase 3 (RASolute 302) as the most credible small-molecule assault in the decade, but a Phase 3 readout is still ahead.

And that’s why even within the heme oncology world, we still face the same challenge for AML and T-cell lymphoma. These lineages are not dispensable, and we still struggle to find “pristine” cell-surface targets like CD20 or BCMA.

Bonus: if you want to read more about the implications of therapeutic index in oncology, here’s one of my PhD papers published in Cancer Cell: Targeting pan-essential genes in cancer: challenges and opportunities

---

Finding 2: Within Solid Tumors, There Are Two Speeds

Some solid tumors saw survival more than double in eleven years. Others remain stuck at benchmarks set before the decade began.

The Winners

The Stuck

The RCC transformation came from three trials in rapid succession (CheckMate 214, KEYNOTE-426, CheckMate 9ER) that put nivo+ipi and IO+TKI combinations into first-line. The CheckMate 214 9-year final analysis at ASCO 2025 confirmed durable benefit (ITT mOS 53 mo, intermediate/poor-risk subset 55.7 mo). And notably, the favorable-risk OS HR finally turned positive at 0.80 (it was 1.45 at the original 2018 readout). The urothelial transformation came from EV-302, pairing an antibody-drug conjugate (enfortumab vedotin, Nectin-4) with pembrolizumab. At 2.5-year follow-up: mOS 33.8 vs 15.9 months, HR 0.51. ASCO 2025 confirmed sustained benefit at 29 months.

Pancreatic got NAPOLI-3 (NALIRIFOX) in 2023. Technically positive against gem/nab-paclitaxel. Median OS: 11.1 months. Identical to FOLFIRINOX from 2011. Fourteen years, 1,924 ASCO abstracts, still no movement on the bar, although daraxonrasib (RMC-6236) entering Phase 3 RASolute 302, and early CLDN18.2 signals (FG-M108 + gem/nab-pac, ORR 53.1% in CLDN18.2-mod/high; IBI343 ADC) are the first credible cracks.

---

Finding 3: The IO Revolution Happened Once, in 36 Months, and Then Stopped

Twelve practice-changing Phase 3 trials defined the IO era, from KEYNOTE-024 at ESMO 2016 through JAVELIN Bladder 100 at ASCO 2020. Eleven of them clustered in a tight window from 2017 to 2020. After 2020, only a handful of 1L bars moved, and none of them were IO optimizations.

The window:

• 2016 (the opener): KEYNOTE-024

• 2017: FLAURA, CheckMate 214

• 2018: KEYNOTE-189, IMpower133, KEYNOTE-048

• 2019: KEYNOTE-426, PROfound

• 2020: IMbrave150, KEYNOTE-177, JAVELIN Bladder 100, CheckMate 9ER

Before 2017, the IO revolution was a promise. After 2020, it was a memory. From 2021 through 2023, the readouts were confirmations and extensions. Six additional PD-1/PD-L1 antibodies (toripalimab, tislelizumab, camrelizumab, sintilimab, sugemalimab, serplulimab) generated 773 ASCO abstracts. All of them confirmed existing bars. None of them moved a bar.

And then there was the graveyard.

---

Finding 4: The Post-PD-1 IO Graveyard

Everyone assumed PD-1 and CTLA-4 would be the beginning, not the end. The industry launched enormous programs on IDO, TIGIT, LAG-3, TIM-3, VISTA, CD47. Almost all of them failed.

IDO was the first to fall. Epacadostat abstracts climbed from zero (2015) to 16 at peak (2018) on Phase 2 momentum. Then ECHO-301/KEYNOTE-252 reported in April 2018: no benefit from adding epacadostat to pembrolizumab in melanoma. Merck terminated the program. By 2024, epacadostat was at zero abstracts.

TIGIT was the comeback story that wasn’t. Tiragolumab produced one of the strongest Phase 2 signals post-PD-1. TIGIT abstracts climbed from zero to 32 at peak in 2023. Then SKYSCRAPER-01 (1L PD-L1-high NSCLC) missed its co-primary endpoints. SKYSCRAPER-02 (ES-SCLC) failed. Roche discontinued multiple programs.

LAG-3 has the only win, and it is narrow. Relatlimab plus nivolumab was approved in March 2022 for first-line melanoma based on RELATIVITY-047 (mPFS 10.1 vs 4.6, HR 0.75). But RELATIVITY-098, the adjuvant trial, reported HR 1.01, a complete failure. The LAG-3 category is still trying to expand beyond melanoma and failing.

ASCO 2025 added another casualty. CheckMate 901 reported its final analysis of NIVO+IPI vs gem/carbo in cisplatin-ineligible 1L mUC: median OS 19.1 vs 13.2 months, HR 0.79 (p=0.0245) — numerically positive, but it failed to meet its prespecified significance threshold. Another IO+IO combination that did not deliver in 1L solid tumor.

The Phase 3 failure rate for new checkpoint targets in the post-PD-1 space has been near-universal, with the narrow exception of 1L melanoma. Adding more checkpoints to the same backbone does not produce more benefit. The IO biology that works may represent a specific and unusual configuration. The narrow exceptions emerging at ASCO 2025 are equally telling: cobolimab (TIM-3) + dostarlimab in 1L HCC delivered ORR 42.9% in a small Phase 2 — meaningful, but a single-tumor signal in a small cohort, not a generalizable pattern.

The 1L bars that moved after 2020 used new biology, not IO optimization. FLAURA2 (2023) added platinum/pemetrexed to osimertinib and moved 1L EGFRm NSCLC PFS from 16.7 to 25.5 months. MARIPOSA (2024) paired amivantamab (EGFR/MET bispecific) with lazertinib (next-gen TKI) — the first bispecific-based win in EGFR-mutant NSCLC. EV-302 (2024) paired enfortumab vedotin (Nectin-4 ADC) with pembrolizumab in 1L urothelial. mOS 33.8 vs 15.9 months. IMforte (ASCO 2025) added lurbinectedin maintenance to atezo+chemo in 1L ES-SCLC, OS HR 0.73.

The pattern is clear: ADCs, bispecifics, novel chemo+IO scheduling, next-generation TKIs. Not more checkpoints.

The 2021–2025 window was a hinge: the IO era ended, and the ADC and bispecific era began.

---

Finding 5: The Ladder Is the Best Predictive Signal

Every practice-changing drug of the decade climbed a ladder from last-line therapy to first-line therapy. The speed of that climb predicted how important the drug would become.

Pembrolizumab in NSCLC (≈2 years, the fastest): 2L entry 2015 (KEYNOTE-010) → 1L PD-L1 high 2016 (KEYNOTE-024). Nivolumab reached the same 2L rung in 2015 but never captured the 1L market, because its 1L strategy (TMB biomarker, ipi combo) was slower and more complex. Pembro abstracts: 51 → 378. Nivo abstracts: 33 → 247. Same mechanism class, different climb speed, different outcome.

Enfortumab vedotin in urothelial (≈5 years, the most dramatic): 3L+ 2019 (EV-201, ORR 44%) → 2L 2021 (EV-301, mOS 12.9 vs 8.9) → 1L 2024 (EV-302, mOS 33.8 vs 15.9). The climb destroyed the 2L landscape behind it. Post-EV+pembro 2L options are sacituzumab govitecan and erdafitinib, and 2L mOS dropped back to 6–8 months. The biggest unmet need in bladder cancer exists because EV was good enough to climb.

T-DXd in HER2+ breast (≈5–6 years, still climbing): 3L+ 2019–2020 (DESTINY-Breast01, ORR 60.9% after a median of 6 prior regimens) → 2L 2022 (DESTINY-Breast03, mPFS 28.8 vs 6.8, HR 0.28 at primary analysis; mOS 52.6 vs 42.7 at long-term follow-up) → 1L positioning (DESTINY-Breast07 Phase 1b/2: ORR 82.6% in the T-DXd monotherapy arm; DESTINY-Breast09 Phase 3 advancing). The 4× PFS improvement over T-DM1 is one of the largest single bar movements in HER2+ breast cancer history.

T-DXd abstract count is the steepest single-agent growth curve in the dataset: 0 in 2015, 11 in 2020, 38 in 2023, 69 in 2024. The drug has gone from non-existent to the most-discussed agent at ASCO in under a decade.

Lurbinectedin’s late climb (the surprise): Until ASCO 2025, lurbinectedin had been stuck on 2L SCLC for 5 years on the strength of a 35% single-arm ORR. The expectation was that it would never climb. Then IMforte read out at ASCO 2025: lurb + atezolizumab maintenance after 1L IO+chemo, OS HR 0.73. The drug climbed from 2L to 1L maintenance in a single Phase 3. The lesson: even drugs assumed to be stuck can climb late if the right combination context is found.

The rule of thumb: if a drug’s rung-1 data does not make oncologists say “this obviously needs to be tested earlier,” it probably never will be. Regorafenib has been stuck on 3L+ colorectal for 12 years. Lurbinectedin was the rare exception — but its climb still took five years and required a different setting (maintenance after IO+chemo) rather than a head-to-head against the SOC.

---

Finding 6: The ADC Wave Is Replacing Chemo

ADC abstracts grew from ~50 in 2015 to ~290 in 2024, with ~330+ projected at ASCO 2025. ~6.6× growth, with +44% year-over-year in each of 2023 and 2024. In 2024 alone, more ADC abstracts were presented than in the entire 2015–2018 window combined.

Three dynasties:

Dynasty 1: The Pioneers (2015–2019). T-DM1 and brentuximab vedotin dominated, together accounting for roughly half of ADC abstract mentions. T-DXd, sacituzumab govitecan, and enfortumab vedotin were all in early clinical development with minimal ASCO footprints.

Dynasty 2: The Daiichi Inflection (2020–2022). T-DXd appeared in 11 abstracts in 2020, driven by DESTINY-Breast01. Sacituzumab govitecan grew on ASCENT. Enfortumab vedotin built on EV-201. Total ADC abstracts jumped from ~67 (2019) to ~110 (2020), a 64% single-year increase.

Dynasty 3: The ADC Tsunami (2023–2025). T-DXd reached 69 mentions in 2024. Sacituzumab govitecan hit 42. Enfortumab vedotin reached 38 on EV-302’s 1L win. Disitamab vedotin (RC48, China) surged from 5 to 21. HER2 target mentions in ADC abstracts grew from 13 (2015) to 142 (2024), a 10.9× increase. Trop-2 from 3 to 33. Nectin-4 from 0 to 21. ASCO 2025 expanded the ADC universe further: biparatopic HER2 ADCs (JSKN003 ORR 67.9%, IBI354, TQB2102 with chemo-free neoadjuvant pCR 76.9%), CLDN18.2 ADCs (ASKB589, IBI343), bispecific ADCs (Iza-bren EGFR×HER3 with cPR 85.7% in EGFR ex20ins NSCLC), and a new generation of payloads (SuperTopoi 5–10× more potent than Dxd, with MHB088C reporting cORR 43.3% in relapsed SCLC).

The headline is that ADCs are starting to replace chemotherapy in specific settings.

DESTINY-Breast03: T-DXd destroyed T-DM1 in 2L HER2+ breast. mPFS 28.8 vs 6.8 months. mOS 52.6 vs 42.7 at long-term follow-up. The ~10-month OS advantage is one of the largest bar movements in HER2+ breast cancer history.

DESTINY-Breast04: T-DXd extended into HER2-low, a population that did not exist as a therapeutic category before the trial. Roughly 60% of traditional HER2-negative breast cancers became HER2-low and targetable. The standard of care for these patients now involves an ADC, not chemotherapy.

EV-302: The cleanest example of ADC+IO replacing platinum-chemo. The chemotherapy backbone that anchored 1L urothelial since the 1990s was displaced in a single Phase 3.

ASCENT: Sacituzumab govitecan in metastatic TNBC delivered mOS 12.1 vs 6.7 months vs chemotherapy.

ZL-1310 (DLL3 ADC) at ASCO 2025: The first ADC to enter the SCLC space with serious efficacy — ORR 68% in Phase 1, including 80% response rate in patients with brain metastases and activity even after tarlatamab (DLL3 BiTE) failure. SCLC, which had no actionable surface antigen for ADCs a year ago, now has one.

The target landscape shows what is next. HER2 is dominant and growing (10.9×). Trop-2 is the next (sacituzumab govitecan, datopotamab deruxtecan). Nectin-4 surging. Claudin 18.2 emerging. HER3, B7-H3, CEACAM5, DLL3 all in active development. ADC+IO combinations surpassed ADC+chemo combinations in 2024 abstract counts, and ASCO 2025 reinforced the trend, suggesting the next tactical question is whether cytotoxic payload synergy with checkpoint activation extends beyond urothelial.

---

What’s Rising Next?

(Just some quick examples, as I’m still distilling the long list of the reminders!)

Bispecific antibodies, including T-cell engagers (TCEs) are the fastest-growing modality at ASCO over the decade. 9.9× growth, from 15 abstracts in 2015 to 149 in 2024. Faster than IO (5.3×), ADC (5.8×), and CAR-T (6.9×). Teclistamab and talquetamab (BCMA, GPRC5D) for multiple myeloma; epcoritamab, glofitamab, mosunetuzumab (CD20) for lymphoma; amivantamab (EGFR/MET) in NSCLC; tarlatamab (DLL3) in SCLC. ASCO 2025 added trispecifics: ISB 2001 (BCMA × CD38 × CD3, ORR 82%), JNJ-5322 (BCMA/GPRC5D × CD3, 100% ORR at RP2D), and ZG006 (CD3/DLL3/DLL3 trispecific in SCLC, ORR 78.6% at 30 mg).

KRAS inhibitors crossed a threshold. Sotorasib and adagrasib (KRAS G12C) were the first wave, modestly positive. ASCO 2025’s KRYSTAL-7 (1L adagrasib + pembrolizumab in KRAS G12C NSCLC) reported ORR 44.3%, mPFS 11.0 mo, mOS 18.3 mo — the first survival data for a 1L KRAS inhibitor + IO combination, with a notable 26.3-month median DoR. The next wave is KRAS G12D inhibitors (GFH375, TSN1611, MRTX113) and pan-KRAS inhibitors like daraxonrasib (RMC-6236, now in Phase 3 RASolute 302). This is the first credible small-molecule assault on the one target that has blocked pancreatic cancer progress for a generation.

CAR-T in solid tumors showed its first real signals in 2024 (76% YoY growth in solid tumor CAR-T abstracts), and ASCO 2025 added more: B7H3 CAR-T (intracranial in GBM), CARv3-TEAM-E (EGFRvIII CAR-T with bystander killing in GBM), CEA CAR-T in NSCLC (ORR 47% with 13-month persistence), DLL3 CAR-T in SCLC (LB2102, AFNT-211). Targets include GD2, claudin 18.2, mesothelin, GPC3, B7H3, EGFRvIII. Most remain early-stage, but the inflection is the first sign that years of engineering work might be starting to pay off.

The waves stack, they do not replace each other. IO abstracts did not decline when ADCs rose. They both grew. The oncology arsenal is diversifying, not rotating.

---

Some final thoughts

Progress in oncology is discontinuous, not continuous. Flat, jump, flat again. Each step function compressed a decade of expected gradual improvement into a single readout, then held for five-plus years.

The IO revolution was real and over. Pembrolizumab was the most important drug of the decade. It also has diminishing returns in combination, and the post-PD-1 checkpoints that were supposed to succeed it have largely failed. The industry learned this at a cost of tens of billions of dollars. ASCO 2025’s CheckMate 901 miss in 1L cisplatin-ineligible mUC was the most recent reminder.

The ADC era has arrived and is still accelerating. If I had to bet on the modality that defines the next five years, it is ADCs plus their combinations with checkpoint inhibitors, bispecifics, and next-generation payloads. ASCO 2025 added biparatopic HER2 ADCs, CLDN18.2 ADCs, the first DLL3 ADC, bispecific ADCs, and payloads 5–10× more potent than Dxd. The ceiling has not been reached.

The speed of the ladder climb is the best predictive signal — but the lurbinectedin exception is humbling. Three years from 2L to 1L means standard of care. Twelve years with no climb usually means the drug will not move. But IMforte showed that even drugs written off as “stuck” can climb late if the right combination context appears. Don’t write a drug off entirely until the entire combinatorial space has been explored.