Progression-Free Survival (PFS)

Definition

"PFS is defined as the time from randomization until objective tumor progression or death, whichever occurs first." — FDA Cancer Endpoints 2018, §III.B.4 (Final guidance)

PFS is a time-to-event composite: the event is the earlier of documented objective disease progression per pre-specified criteria (e.g., RECIST 1.1) or death from any cause. Patients without a PFS event at data cutoff are censored.

TTP vs PFS: Time to progression (TTP) is defined as "the time from randomization until objective tumor progression; TTP does not include deaths." Deaths in TTP are censored — this is informative censoring. "Compared with TTP, PFS is the preferred regulatory endpoint. PFS includes deaths and thus can be a better correlate to overall survival." (FDA 2018, Final)

PFS "can reflect tumor growth and be assessed before the determination of a survival benefit. Its determination is not confounded by subsequent therapy." (FDA 2018)

Regulatory Position

Traditional (regular) approval: PFS can support traditional approval when the effect size is large, benefit-risk is favorable, and effect is durable. "A large improvement in progression-free survival… has been used to support traditional approval in select malignancies, but magnitude of effect, relief of tumor-related symptoms, and/or improvement in quality of life… are considered." (FDA 2018, Final)

Accelerated approval: PFS is an accepted surrogate for accelerated approval. Confirmatory trial verifying OS or clinical benefit required post-approval.

Indication-specific acceptability varies:

Strong regulatory precedent: NSCLC, ovarian cancer, myeloma, RCC, breast cancer (metastatic)
Weaker precedent: colorectal cancer (OS correlation weak for many regimens), pancreatic cancer (OS preferred)

FDA NSCLC guidance (2020, Final): PFS is acceptable as a primary endpoint "if a large treatment effect and acceptable risk-benefit ratio are shown." PFS alone may support traditional approval in NSCLC when HR ≤ 0.50–0.60 and OS trend is supportive, per historical FDA decisions.

Status: FDA Cancer Endpoints 2018 = Final; FDA NSCLC Endpoints guidance = Final (originally published 2015, updated 2020)

When to Use

PFS as primary endpoint — high evidence base:

NSCLC targeted therapy (1L/2L): Dominant primary endpoint (109/300 trials, 36%). Targeted agents (EGFR TKIs, ALK inhibitors) show large HR reductions (HR 0.2–0.5). Examples: osimertinib (FLAURA), alectinib (ALEX), mobocertinib (Exon 20).
Ovarian cancer maintenance: PFS is the standard (76/230, 33%). PARP inhibitors in BRCA-mutated/HRD populations show HR 0.3–0.4. FDA accepts PFS for traditional approval in maintenance with large effect.
Multiple myeloma: Most common primary endpoint (99/256, 39%). PFS dominates Phase 3 given long post-progression survival; MRD is emerging. FDA accepts PFS for traditional approval when effect size is large.
RCC: PFS + OS co-primary increasingly standard (40/112, 36%). IO combinations with nivolumab + ipilimumab, pembrolizumab + axitinib.
Breast cancer (metastatic): PFS primary in CDK4/6 inhibitor trials (palbociclib PALOMA-2, ribociclib MONALEESA).

PFS as secondary (OS primary):

Pancreatic cancer, colorectal cancer (metastatic), AML
Whenever median OS < 18 months and crossover is not planned

When PFS is insufficient alone:

When OS correlation is not established for that indication/line of therapy
When effect HR > 0.70 and OS data are immature
FDA may require OS as confirmatory evidence for traditional approval in some settings

Design Considerations

Tumor assessment schedule and progression criteria

Assessment schedule:

Standard metastatic setting: imaging every 6–9 weeks (cycles aligned to treatment schedule)
Maintenance settings: every 8–12 weeks
Asymmetric imaging schedules between arms introduce bias — must be identical or justified
Time window: data collection "should be limited to a specified short time interval around the scheduled visit" (FDA 2018)

Progression criteria:

"There are no standard regulatory criteria for defining progression." (FDA 2018). Applicants must pre-specify in protocol and SAP:

RECIST 1.1 (solid tumors): ≥20% increase in sum of target lesion diameters + absolute increase ≥5 mm; or new lesion; or unequivocal progression of non-target lesions
Lugano 2014 (lymphoma): ≥50% increase in PPD of target nodes, or new lesion
IMWG criteria (myeloma): ≥25% increase in M-protein, new bone lesions, hypercalcemia
RANO (brain tumors): MRI-based; steroid use confounds assessment
PCWG3 (prostate): PSA + imaging criteria

IRC requirements and assessment methodology

"When the primary study endpoint is based on tumor measurements (e.g., PFS or ORR), tumor assessments generally should be verified by central reviewers blinded to study treatments." (FDA 2018)

When IRC is required:

Mandatory in: Unblinded (open-label) randomized PFS trials where assessment bias is plausible
May be waived when: Trial is double-blind AND adverse event profile does not unblind treatment assignment, OR effect size is robust with pre-specified sensitivity analyses excluding investigator bias
Audit-based IRC: Acceptable alternative to full IRC if pre-specified and bias can be excluded at audit; seek FDA input before use

IRC operational standards:

Blinding: IRC must be blinded to all treatment assignments, randomization strata, investigator assessments, and interim efficacy results
Adjudication scope: Confirm progression per protocol criteria; resolve discordance between investigator and central assessment
Documentation: All IRC assessments and disagreements must be documented in SAP with sensitivity analyses reported
Training: IRC members must be trained on protocol assessment criteria; written charter required before first assessment

Pre-specified censoring rules (complete set)

Scenario	Censoring Rule	Rationale
No post-baseline tumor assessment	Censor at randomization date (Day 1)	Uninformative censoring; no disease data collected
Progression/death after ≥2 missed assessments	Censor at last adequate assessment prior to gap	Avoid imputing events; conservative approach per FDA
New anti-cancer therapy before progression	Censor at last tumor assessment before new therapy	Treatment policy strategy; confounding of PFS by subsequent therapy
Lost to follow-up	Censor at last adequate tumor assessment	Non-informative censoring assumption
Withdrawal of consent	Censor at last adequate tumor assessment	Non-informative censoring assumption
Progression documented at unscheduled visit with prior missing scheduled assessment	Censor at last scheduled assessment (conservative per FDA)	Avoids bias from unscheduled assessments triggering follow-ups
Patient still on study at data cutoff	Censor at last tumor assessment	Standard approach; no PFS event observed
Death from other causes without progression documented	Count as PFS event (composite definition)	PFS includes death from any cause

"We recommend assigning the progression date to the earliest time when any progression is observed without prior missing assessments." (FDA 2018)

Alpha allocation with PFS + OS

When PFS is primary and OS is secondary (hierarchical):

Gatekeeping: OS tested only if PFS is significant
Co-primary: alpha split (e.g., 0.025 each, or 0.04/0.01 asymmetric split)
Interim analyses for OS: at time of primary PFS analysis, use small alpha (0.001–0.005) to preserve overall Type I error
See Multiple Endpoints and Alpha Allocation for formal gatekeeping strategies

Sample Size Calculation

Schoenfeld formula for PFS

For time-to-event PFS analysis with log-rank test:

Required events: d = (z_α + z_β)² / [p₁(1-p₁) + p₂(1-p₂)] × [log(HR)]⁻²

Where:

z_α = critical value for Type I error (0.025 one-sided = 1.96)
z_β = critical value for power (80% power = 0.84)
p₁, p₂ = proportion of events in control, treatment arms
HR = hypothesized hazard ratio (< 1 favors treatment)

NSCLC Example (1L targeted therapy):

Baseline median PFS (control): 5 months
Target median PFS (treatment): 10 months (HR = 0.50)
Type I error: α = 0.025 one-sided
Power: 80% (β = 0.20)
Enrollment: 450 patients, 1:1 randomization, 12-month accrual
Median follow-up post-cutoff: 1 month

Calculation:

d = (1.96 + 0.84)² / [0.50 × 0.50] × [log(0.50)]⁻²
d = 7.84 / 0.25 × 4.73
d ≈ 147 events (for 80% power)

With expected 50–70% event rate by month 24 in this setting → ~225–280 enrolled patients needed.

Event count targets by indication

NSCLC 1L targeted therapy: 140–200 events (HR 0.50, large effect)
NSCLC IO ± chemotherapy: 280–350 events (HR 0.70–0.75, moderate effect; longer censoring tail)
Ovarian maintenance: 180–250 events; longer median PFS (12–18 months)
Myeloma: 300–400 events; very long PFS (36–48 months) requires large enrollment

Note: These assume standard accrual timelines. Slow enrollment or high dropout extends follow-up; sensitivity analyses recommended.

Intercurrent Events

1. Subsequent anticancer therapy before progression

Applicability: Common in trials where crossover or sequential therapy is allowed.

ICH E9(R1) strategy: Treatment Policy — PFS ignores subsequent therapy; censoring rules handle it.

Statistical consequence: Patients censored at last pre-therapy assessment; subsequent progression/OS not included in PFS.

SAP language: "Patients who initiate new anti-cancer therapy prior to documented progression or death will have their PFS event censored at the date of the last adequate tumor assessment prior to initiation of new therapy. Tumor assessments after initiation of new therapy will not be included in PFS analysis."

Sensitivity analysis: Report results censoring vs. not censoring at new therapy initiation; assess robustness to crossover impact.

2. Treatment discontinuation without progression

Applicability: Most common IE in targeted therapy trials. Patient stops drug due to toxicity but tumor has not progressed.

ICH E9(R1) strategy: Treatment Policy — continue assessments per schedule; PFS event dated to actual progression.

Statistical consequence: Continued tumor assessments post-discontinuation are included; censoring does not occur at discontinuation.

SAP language: "Tumor assessments will continue per protocol schedule regardless of treatment discontinuation status. Patients who discontinue study treatment without documented progression will remain in the study for PFS assessments until progression, death, withdrawal of consent, or study closure. Discontinuation of treatment does not constitute a PFS event."

Critical requirement: Protocol must mandate on-study tumor assessments post-discontinuation. If assessments stop at discontinuation, informative censoring bias occurs and PFS HR estimates will be biased toward null.

3. Death without prior progression

Applicability: Applies to all trials; death is included in PFS definition.

ICH E9(R1) strategy: Composite — death counts as PFS event regardless of cause.

Statistical consequence: No censoring; death from any cause is counted as PFS event on date of death.

SAP language: "Death from any cause without prior documented disease progression will be counted as a progression-free survival event on the date of death."

Applicability: When patients withdraw from study before progression is documented.

ICH E9(R1) strategy: Treatment Policy — censor at last adequate assessment.

Statistical consequence: Last known tumor status defines censoring time; subsequent disease status unknown.

SAP language: "Patients who withdraw consent for further follow-up will be censored at the date of the last adequate tumor assessment. No post-withdrawal data will be included in PFS analysis."

Statistical Analysis and IRC Sensitivity

IRC vs. Investigator Assessment Concordance

Key regulatory expectation: "Sensitivity analyses exploring the impact of discordance between investigator and IRC assessments should be pre-specified and reported."

Analysis plan:

Primary (IRC-based): PFS HR, 95% CI, p-value using IRC progression dates
Sensitivity (Investigator-based): PFS HR using investigator progression dates; compare magnitude and statistical significance
Discordance table: Show proportion of cases where IRC and investigator agree/disagree on progression timing; categorize by magnitude of delay (≤30 days, 31–90 days, >90 days)

When discordance is substantial (>10% of progressions):

FDA may view IRC-driven HR as primary estimate
Investigator-based analysis is supporting evidence of robustness
Large discordance raises concerns about assessment bias or criteria clarity

R implementation:

# Compare PFS curves by adjudicator
irc_pfs <- survfit(Surv(time_irc, event_irc) ~ treatment, data=trial_data)
inv_pfs <- survfit(Surv(time_inv, event_inv) ~ treatment, data=trial_data)

# Log-rank test and HR comparison
coxph(Surv(time_irc, event_irc) ~ treatment, data=trial_data)  # IRC primary
coxph(Surv(time_inv, event_inv) ~ treatment, data=trial_data)  # Investigator

Interim analysis timing and alpha spending

Typical PFS interim schedule:

Interim 1: At 50% of planned PFS events → O'Brien-Fleming spending (α_I = 0.00025, α_F = 0.0248)
Interim 2: At 75% of planned events → Lan-DeMets spending
Final: At 100% of planned events

Conditional power at interim: If interim PFS HR is close to null (0.85–1.05), DSMC may recommend futility stopping using conditional power ≤ 10% at planned sample size.

Regulatory Precedent

NCT#	Trial	Drug	Indication	PFS HR	Outcome
NCT04129502	EXCLAIM-2	TAK-788 (mobocertinib)	1L NSCLC EGFR Exon 20	Not yet reported	Active; PFS by IRC primary
NCT01828112	ASCEND-5	Ceritinib	2L+ ALK+ NSCLC	HR ~0.49	PFS supported traditional approval
NCT02864251	CheckMate-722	Nivolumab + ipilimumab	EGFR-mut NSCLC	—	PFS co-primary with OS

Note: NSCLC CTG dataset contains 109 trials with PFS as primary endpoint. Median enrollment ~450 patients; ~36% used PFS exclusively vs OS.

Limitations and Pitfalls

1. Surrogate validity is indication-dependent:

PFS predicts OS well in: ovarian cancer (maintenance), colorectal cancer (oxaliplatin-based), myeloma (conventional agents)
PFS correlation with OS is weak in: NSCLC immunotherapy (immune-related delayed responses), breast cancer metastatic HER2+, some targeted therapies with post-progression activity
Mitigation: FDA may require OS as primary or OS trend as supportive evidence when PFS-OS correlation is not established; pre-specify surrogate validation plan

2. Assessment frequency bias:

More frequent imaging in one arm (or per-protocol vs. off-protocol) inflates PFS event rate asymmetrically
Mitigation: Symmetric assessment schedules are critical; protocol deviations in imaging frequency must be analyzed as sensitivity

3. Informative censoring:

Patients censored for new therapy or withdrawal may have worse tumors than those remaining — violates non-informative censoring assumption
Mitigation: Sensitivity analyses with alternative censoring rules (e.g., censor vs. not censor at new therapy) required; DSMC oversight

4. Unblinding through adverse events:

In open-label trials, investigators aware of grade 3–4 AEs may over-read borderline scans as progression
Mitigation: IRC is the primary remedy; DSMC oversight of assessment imbalances; sensitivity analysis with investigator vs. IRC assessments

5. PFS benefit without OS benefit:

Multiple failed OS confirmatory trials (bevacizumab in breast cancer, sunitinib in GIST 2L) demonstrate that PFS improvement does not guarantee OS improvement
Mitigation: FDA may withdraw accelerated approval if PFS-based surrogate fails OS confirmation; plan confirmatory OS trial pre-approval

6. Landmark PFS endpoints (PFS-12, PFS-24):

Rate at fixed timepoint (e.g., 12-month PFS rate) is an alternative to HR-based analysis in non-proportional hazards settings
Mitigation: Used in lymphoma (EFS-12) and immunotherapy trials where HR may not be constant; pre-specify both HR and landmark rate in SAP

Backlinks

Source: FDA Cancer Endpoints 2018 (Final); FDA NSCLC Endpoints Guidance (Final, 2015/2020) Status: Final guidance Compiled from retrieved FDA chunks + ClinicalTrials.gov records (300 NSCLC Phase 3 trials, 230 ovarian Phase 3 trials, 256 myeloma Phase 3 trials)