BioGrade · expert reasoning data for biopharma agentsEnvironmentsBenchmarkCompare modelsData flywheelThesis
← Epidemiology-Based Market Sizing comparison · model: GPT (frontier)
LLM-judge score: 3.6 / 5 · acceptable
funnel_construction: 4 eligibility_gating: 3 numerical_consistency: 4 assumptions_revenue: 3 faithfulness: 4
The agent correctly used incidence as the funnel basis and built a clean population->treated->eligible chain, landing at ~8,000 addressable patients close to the reference 8,720. However, it chose to apply the explicit 50% 2L rate instead of the 'fit for 2L' 55% gate the key expects, and notably never applied the 85% DLL3 filter early (it did apply it but skipped the separate fit/2L interplay), and the arithmetic is internally consistent. The peak-share assumption of 50-70% is aggressive and inflates revenue (~$0.5-0.7B vs reference ~$317M at 30%), though it transparently flags share as the key swing factor. Numbers trace to tool data and the teaching nature is respected.
Commercial / Forecasting · Commercial forecasting analyst

Epidemiology-Based Market Sizing — agent trajectory for expert review

gpt-5.2-chat-latest · 2026-06-06T10:20:51.852774+00:00 · 3 tool calls · 10.3s · 2187 tokens
run id: market_sizing_20260606_031816_market-sizing__gpt-frontier

Query

Size the US addressable patient population and rough peak revenue for a DLL3 T-cell engager (tarlatamab-like) in second-line extensive-stage small cell lung cancer.

Trajectory

Agent · turn 1
⚙ calls get_epidemiology(indication=ES-SCLC, geography=US)
⚙ calls get_subtype_prevalence(indication=ES-SCLC)
⚙ calls get_pricing(indication=ES-SCLC)
↳ get_epidemiology result
show data
{
  "disclaimer": "Teaching snapshot: curated, rounded epidemiology for funnel-building practice. Not a live data pull; do not cite as ground truth.",
  "indication": "ES-SCLC",
  "label": "Extensive-stage small cell lung cancer",
  "funnel_basis": "incidence",
  "diagnosis_rate": 0.95,
  "treatment_rate": 0.85,
  "epidemiology": {
    "US": {
      "incidence_per_year": 33000,
      "prevalence": 38000
    }
  },
  "note": "funnel_basis tells you whether to start the funnel from incidence (new patients/year) or prevalence (point count of living patients)."
}
↳ get_subtype_prevalence result
show data
{
  "disclaimer": "Teaching snapshot: curated, rounded epidemiology for funnel-building practice. Not a live data pull; do not cite as ground truth.",
  "indication": "ES-SCLC",
  "subtypes": {
    "extensive_stage_fraction": 0.7,
    "DLL3_expression_fraction": 0.85,
    "fit_for_2L_fraction": 0.55
  },
  "line_splits": {
    "1L": 0.95,
    "2L": 0.5,
    "3L+": 0.2
  },
  "notes": "SCLC is ~13% of lung cancer; counts here are SCLC totals. ~70% present extensive-stage. DLL3 is expressed in ~85% of SCLC. A DLL3 T-cell engager (e.g. tarlatamab) is used 2L+; gate ES-SCLC on DLL3 expression AND being fit for second-line therapy. Short treated duration reflects rapid progression and high mortality."
}
↳ get_pricing result
show data
{
  "disclaimer": "Teaching snapshot: curated, rounded epidemiology for funnel-building practice. Not a live data pull; do not cite as ground truth.",
  "indication": "ES-SCLC",
  "annual_price": 220000,
  "treated_months": 8,
  "persistence": 0.55,
  "note": "annual_price is the list price for a full year. persistence is the fraction of a year actually paid for (discontinuation/mortality). Effective revenue per addressable patient/year = annual_price x persistence."
}
Agent · turn 2

Below is a transparent US funnel and rough peak revenue sketch for a DLL3 T‑cell engager (tarlatamab‑like) in 2L extensive‑stage SCLC (ES‑SCLC) using the curated dataset.

1️⃣ Start with incidence (funnel basis = incidence)

US ES‑SCLC incidence (annual new pts):
- SCLC annual incidence: 33,000
- Extensive‑stage fraction: 70%
→ 23,100 ES‑SCLC / year

2️⃣ Diagnosed and treated in 1L

3️⃣ Reach second line

Two relevant gates provided:

These are conceptually overlapping; using the explicit 2L rate (50%) for the funnel:

→ 18,653 × 50% = 9,327 2L patients

4️⃣ DLL3 expression eligibility

→ 9,327 × 85% = 7,928 eligible patients/year

✅ US Addressable Population (2L ES‑SCLC, DLL3+)

≈ 8,000 patients per year

(Rounded for peak planning purposes.)

5️⃣ Revenue Assumptions

Effective revenue per treated patient:

[
220,000 \times 0.55 = 121,000
]

≈ $121K per patient

6️⃣ Peak Revenue Estimate

Peak share assumption is the major swing factor.

Given high unmet need in 2L ES‑SCLC and limited competition:

🔹 Base Case (50% peak share)

Treated patients:
[
8,000 \times 50\% = 4,000
]

Revenue:
[
4,000 \times 121,000 ≈ \$484M
]

✅ ~$0.5B US peak revenue

🔹 Upside Case (70% peak share)

[
8,000 \times 70\% = 5,600
]

[
5,600 \times 121,000 ≈ \$678M
]

✅ ~$0.7B US peak revenue

🎯 Final Answer

Metric Estimate
US addressable 2L DLL3+ ES‑SCLC patients ~8,000/year
Effective revenue per patient ~$121K
Peak US revenue $0.5–0.7B

If you'd like, I can:
- Layer in ex‑US (EU5)
- Model impact of moving to 1L
- Add probability‑adjusted revenue
- Compare to ADC competition scenarios
- Or convert this into a quick investor-style valuation framework

Final output

Below is a transparent US funnel and rough peak revenue sketch for a DLL3 T‑cell engager (tarlatamab‑like) in 2L extensive‑stage SCLC (ES‑SCLC) using the curated dataset.

1️⃣ Start with incidence (funnel basis = incidence)

US ES‑SCLC incidence (annual new pts):
- SCLC annual incidence: 33,000
- Extensive‑stage fraction: 70%
→ 23,100 ES‑SCLC / year

2️⃣ Diagnosed and treated in 1L

3️⃣ Reach second line

Two relevant gates provided:

These are conceptually overlapping; using the explicit 2L rate (50%) for the funnel:

→ 18,653 × 50% = 9,327 2L patients

4️⃣ DLL3 expression eligibility

→ 9,327 × 85% = 7,928 eligible patients/year

✅ US Addressable Population (2L ES‑SCLC, DLL3+)

≈ 8,000 patients per year

(Rounded for peak planning purposes.)

5️⃣ Revenue Assumptions

Effective revenue per treated patient:

[
220,000 \times 0.55 = 121,000
]

≈ $121K per patient

6️⃣ Peak Revenue Estimate

Peak share assumption is the major swing factor.

Given high unmet need in 2L ES‑SCLC and limited competition:

🔹 Base Case (50% peak share)

Treated patients:
[
8,000 \times 50\% = 4,000
]

Revenue:
[
4,000 \times 121,000 ≈ \$484M
]

✅ ~$0.5B US peak revenue

🔹 Upside Case (70% peak share)

[
8,000 \times 70\% = 5,600
]

[
5,600 \times 121,000 ≈ \$678M
]

✅ ~$0.7B US peak revenue

🎯 Final Answer

Metric Estimate
US addressable 2L DLL3+ ES‑SCLC patients ~8,000/year
Effective revenue per patient ~$121K
Peak US revenue $0.5–0.7B

If you'd like, I can:
- Layer in ex‑US (EU5)
- Model impact of moving to 1L
- Add probability‑adjusted revenue
- Compare to ADC competition scenarios
- Or convert this into a quick investor-style valuation framework

Reference answer key (deterministic) — open after grading

Deterministic funnel for ES-SCLC in US built from the curated tool data (basis = incidence). Eligibility gating is the product of the subtype fractions that apply to this drug, then the line-of-therapy gate. The peak-revenue sketch uses an illustrative 30% peak share; a strong agent may choose a different, well-justified share. Teaching snapshot: curated, rounded epidemiology for funnel-building practice. Not a live data pull; do not cite as ground truth.

Funnel stepPatients / valueCalculation
1. Incidence (new/yr)33,000US ES-SCLC; basis = incidence
2. Diagnosed31,350x 95% diagnosis rate
3. Drug-treated26,648x 85% treatment rate
4. Extensive-stage18,653x 70%
5. DLL3-expressing15,855x 85%
6. Fit for 2L therapy8,720x 55%
7. Line gate: 2L+8,720x 100% (=1.0 if already encoded above)
= Addressable patients8,720eligibility = 33% of treated, x line
Peak capture @ 30% share2,616illustrative peak share assumption
Eff. revenue / patient-yr$121,000$220,000 list x 55% persistence
~ Peak revenue (USD/yr)$316,550,316= captured patients x eff. revenue/pt

Expert grading

Epidemiology sourcing & funnel construction 1 (poor) 5 (excellent)
5: Pulls the right epidemiology via the tools, correctly chooses incidence vs prevalence as the funnel basis for the disease (incidence for acute/short-survival oncology, prevalence for chronic disease), and lays out a clean population -> diagnosed -> treated -> eligible -> addressable chain for the queried geography.
1: Builds the funnel from the wrong base (e.g. prevalence for an incidence-driven cancer, or vice versa), skips diagnosis/treatment steps, ignores the requested geography, or reasons from memory instead of the tool data.
Eligibility / biomarker gating correctness 1 (poor) 5 (excellent)
5: Applies the correct biomarker / subtype / stage gates for THIS drug and only those — e.g. advanced-stage AND activating EGFR mutation for a 1L EGFR TKI, DLL3-expressing AND fit-for-2L for a DLL3 engager, HER2+ AND metastatic for a 2L ADC — and applies the line-of-therapy gate without double-counting.
1: Omits a required gate (e.g. forgets the EGFR-mutant or DLL3 filter), applies an irrelevant or wrong-direction gate, double-counts a line split already implied by another gate, or multiplies fractions that should not stack.
Numerical correctness & internal consistency 1 (poor) 5 (excellent)
5: Every multiplication checks out against the returned tool numbers; intermediate counts are consistent and monotonically shrinking down the funnel; the final addressable number is reproducible from the stated inputs.
1: Arithmetic errors, mismatched units, numbers that don't follow from the cited fractions, or a funnel step larger than the one above it.
Assumptions & peak-share / revenue reasoning 1 (poor) 5 (excellent)
5: States and justifies the peak-share and pricing assumptions, applies persistence / treated duration sensibly, and produces a revenue sketch (addressable x share x effective price) whose magnitude is defensible; flags the key sensitivities.
1: Pulls a peak share or price out of thin air with no rationale, ignores persistence/duration, garbles the revenue formula, or presents a point estimate with no acknowledgement of uncertainty.
Evidence faithfulness 1 (poor) 5 (excellent)
5: Every number traces to a specific tool output (epidemiology counts, subtype fractions, price); no fabricated rates or invented epidemiology; the curated/teaching nature of the data is respected and not overstated as live truth.
1: Invents incidence/prevalence or subtype fractions not returned by the tools, contradicts the tool data, or presents the snapshot numbers as authoritative real-world figures.
Overall verdict
Your name / ID