← Caroline Kim
IMMUNO-ONCOLOGY PIPELINE

Where is cancer immuno-oncology clinical development concentrating — which therapy classes, phases, and sponsors are crowding the pipeline, and how has that shifted over the past decade?

10,971 Interventional Immunotherapy Trials · Six Therapy Classes · ClinicalTrials.gov · 2010–2025

Checkpoint inhibitors still dominate the immuno-oncology pipeline, but the fastest growth now comes from CAR-T / cell therapies and bispecifics. Most of that newer activity is early-phase and increasingly driven by non-industry (academic and hospital) sponsors, while overall trial volume, after a decade of steep growth, has plateaued near its peak. The work has globalized — the US and China lead, with China concentrated in cell therapy.

Trials Analyzed
10,971
between 2010 and 2025
Checkpoint Share
64%
of the pipeline
Top-2 Class Share
75%
how crowded the pipeline is
Fastest-Growing Class
+32% per yr
Bispecific
Best Approval Yield
10 per 1,000
Bispecific
Where the Pipeline Is Heading
Total Trials per Year — Observed, with a Poisson Trend Projection and 95% Prediction Interval
Volume rose steeply through ~2021, then held near its peak; the fitted trend still points up (dashed) but recent years have flattened.
How Activity Grew Across the Decade
Trials Started per Year, Stacked by Therapy Class
Checkpoint inhibitors dominate; the CAR-T / cell-therapy band widens fastest.
How the Class Mix Flipped over the Decade
Each Class's Share of Trials: 2010–2012 vs 2023–2025
The field flipped: cancer vaccines (41%) and cytokine immunomodulators (37%) led in the early 2010s, but by 2023–2025 checkpoint inhibitors had taken over (16% → 63%) as those two collapsed to ~5% each.
The Mix Shift Tracks FDA Approvals
Checkpoint vs Cancer-Vaccine Share of Trials, with FDA Approval Events
Checkpoint's share nearly doubles the year after the pivotal 2014 approvals (pembrolizumab, nivolumab), while cancer vaccines — 925 trials for one approval — fade. The pipeline chases clinical validation (a temporal association, not proof of cause).
Which Classes Convert Trials Into Approvals
FDA Approvals per 1,000 Trials, by Class
Bispecifics convert most efficiently (~10 per 1,000); cancer vaccines crowd the pipeline but have one approval — the modalities that convert are the ones trials chase.
How Fast Is Each Therapy Class Growing
Poisson Trend, 2015–2024, with 95% Confidence Intervals
Bispecifics grow fastest — significantly faster than CAR-T; the profile of the next breakout.
Where Each Therapy Class Sits in Development
Column Width ∝ Number of Trials · Color = Trial Phase
CAR-T / cell therapy is early-phase-heavy; checkpoints reach Phase 3.
Who Is Driving Each Therapy Class
Trial Counts by Lead-Sponsor Type
The registry's non-industry 'Other' class is the largest in every class.
Where Trials Start, and What Fills Each Pipeline
Top Countries by Trials Started, Animated by Year — Press Play
Activity is global; the US and China lead, though China is under-captured on this US registry.
Two Countries, Two Bets
Therapy-Class Mix of US vs China Trials
China skews to CAR-T / cell therapy; the US to cancer vaccines and cytokines.
Methods & Data

Data. Interventional oncology-immunotherapy trials pulled from the ClinicalTrials.gov v2 API, plus FDA approvals from openFDA (Drugs@FDA with a curated table of CBER cell / gene / oncolytic / vaccine approvals). Analysis window: start years 2010–2025.

Classification. Each trial is tagged into one or more of six therapy classes by a controlled-vocabulary classifier (normalize → resolve against a curated lexicon → classify by mechanism), with audit-driven screens (prophylactic-vaccine, dendritic-cell, cytokine-induced-killer, neoantigen-directed T cell, supportive-care GM-CSF). A leakage-free NCI-Thesaurus resolution layer recovers company code names. A trial may belong to more than one class.

Validation. Three independent labelers over a 580-trial stratified sample give a majority-vote consensus; per-class precision and recall carry Wilson 95% confidence intervals, and inter-annotator agreement is Fleiss' κ ≈ 0.97. A separate re-audit of the dropped trials estimates end-to-end recall at ≈92% (true population ≈11,900).

Two-method reconciliation. An independent full-corpus LLM re-tag is compared against the vocabulary tagger at mean Cohen's κ ≈ 0.79 (substantial). With the code-name layer the vocabulary counts now track the LLM closely rather than trailing it — the two independent methods converge.

Inferential statistics. Growth: a Poisson log-linear GLM per class (CAGR with 95% CI) plus a total-volume forecast with a 95% prediction interval. Association: a chi-square test of class × sponsor independence with Cramér's V. Groups: two-proportion z-tests of US-vs-China share with a Holm correction. An effect size accompanies every p-value.

Second source. FDA approvals are joined to the trials on the same six classes to compute in-decade approvals per 1,000 trials — which crowded classes are actually converting to approved products.

Caveats. Exploratory analysis of observational registry data: findings are associations, not causal claims. ClinicalTrials.gov under-captures trials registered only in national registries (e.g. China's ChiCTR), so non-US activity is a floor.