How three competing frameworks for measuring tumor response are reshaping the science — and the business — of cancer drug development
In the middle of a clinical trial, a radiologist looks at a scan and sees a tumor that has grown. Under the rules that have governed oncology research for decades, the conclusion would be straightforward: the patient is not responding. The drug is failing. Move on. But what if the growth is an illusion — not the cancer spreading, but the immune system flooding the site? What if stopping treatment at precisely that moment would rob the patient of a cure that was weeks away?
This dilemma sits at the heart of one of clinical oncology’s most consequential debates: how we measure whether a cancer drug is working. The answer has profound implications for which drugs reach patients, how much they cost, and how long people with cancer live.
“Small differences in progression rules can change trial results significantly — and those changes flow directly into regulatory decisions.”
Three frameworks now compete for dominance: RECIST 1.1, the longstanding standard for solid tumors; iRECIST, an adaptation built for the immunotherapy revolution; and mRECIST, a specialized tool developed for liver cancer. Each reflects a different philosophy about what “response” means, and each produces different numbers from the same underlying biology.
Understanding the differences is no longer just an academic exercise. Regulatory agencies at the FDA and the EMA evaluate efficacy claims based directly on these criteria. An incorrect framework can mislabel an effective therapy as ineffective — or fail to detect when a patient is genuinely progressing.
THE THREE FRAMEWORKS
01 — RECIST 1.1: The Foundation
Response Evaluation Criteria In Solid Tumors remains the workhorse of modern oncology trials. Built on a deceptively simple premise — measure the longest diameter of target lesions, track changes over time — RECIST 1.1 provides four response categories: Complete Response (CR), Partial Response (PR), Stable Disease (SD), and Progressive Disease (PD). Progression is defined as a ≥20% increase in the sum of target lesion sizes, with a minimum 5mm absolute increase, or the appearance of any new lesion. Its strength is its universality: applicable to chemotherapy, targeted therapies, and traditional oncology treatments across virtually all solid tumor types.
02 — iRECIST: The Immunotherapy Exception
When checkpoint inhibitors — PD-1, PD-L1, CTLA-4, CAR-T and their successors — entered mainstream oncology, they exposed a fundamental flaw in RECIST’s logic. Immunotherapy can cause pseudoprogression: the tumor mass may temporarily enlarge as immune cells infiltrate it, only to shrink dramatically weeks later. Under RECIST, these patients would be pulled from a therapy that was working.
iRECIST introduced a two-step confirmation process. Initial progression is classified as iUPD — Unconfirmed Progressive Disease. If the patient remains clinically stable, treatment continues. Only when a follow-up scan confirms growth is progression reclassified as iCPD. This seemingly small adjustment has dramatically improved the accuracy of efficacy assessment for immunotherapies and corrected a systematic bias that had been stopping effective treatments prematurely.
03 — mRECIST: The Liver Cancer Lens
Hepatocellular carcinoma (HCC) presented yet another measurement problem. Liver tumors frequently become necrotic after treatment — the tissue dies, but the physical lesion doesn’t shrink. Under standard RECIST, a successfully treated liver tumor might be scored as stable or even progressing based on its unchanged size, missing the very response that treatment produced.
Modified RECIST addresses this by measuring only the viable, arterially enhancing portion of a lesion, not its full diameter. A tumor can be large and mostly dead; mRECIST captures the biology that matters. This makes it essential for evaluating liver-directed therapies — ablation, TACE (transarterial chemoembolization), radioembolization — and anti-angiogenic treatments where tumor vascularity, not size, is the true signal.
IMPACT I — CLINICAL ENDPOINTS
How Criteria Shape the Numbers That Shape Drug Development
The choice of response criteria isn’t merely methodological housekeeping. It directly determines the clinical endpoints that define a trial’s success: Overall Response Rate (ORR), Progression-Free Survival (PFS), Duration of Response (DOR), Time to Progression (TTP), and Best Overall Response (BOR). These are the numbers that sponsors present to regulators, that regulators use to grant approval, and that physicians cite when discussing treatment options with patients.
The same patient dataset, evaluated under different criteria, can produce meaningfully different response rates, different median PFS values, and different Kaplan-Meier curves — the visual centerpiece of almost every oncology trial publication. A patient might be classified as experiencing Progressive Disease under RECIST 1.1 but Stable Disease or even Partial Response under iRECIST. That single reclassification propagates through every downstream statistical analysis.
“Before iRECIST, temporary tumor enlargement was classified as PD, and effective treatments were stopped too early. The correction wasn’t trivial — it was systematic.”
The statistical ramifications extend to hazard ratios and the p-values on which regulatory decisions turn. Trials designed under one framework and analyzed under another introduce a category of error that neither the sponsor nor the regulator may fully anticipate.
Key Oncology Endpoints Affected
- ORR — Overall Response Rate — proportion of patients with measurable tumor reduction
- PFS — Progression-Free Survival — time from treatment start to progression or death
- DOR — Duration of Response — how long a response is maintained
- TTP — Time to Progression — time to documented disease worsening
- BOR — Best Overall Response — best status recorded during treatment
IMPACT II — REGULATORY & COMMERCIAL STAKES
The Regulatory Stakes: Approval, Labels, and Market Authorization
Regulatory agencies — the FDA in the United States and the EMA in Europe — do not evaluate drugs in the abstract. They evaluate the data packages submitted by sponsors, and those packages are built on trial endpoints that are themselves built on response criteria. The criteria chosen during protocol development therefore cast a long shadow over the entire commercial lifecycle of a drug.
An incorrect response assessment framework can undermine an efficacy signal that actually exists — leading to rejection of drugs that would benefit patients — or, in the opposite direction, fail to detect that a patient is progressing when they genuinely are, compromising both trial integrity and patient safety.
The operational consequences compound the regulatory ones. Trials using iRECIST require additional confirmation scans, creating more complex imaging schedules, additional radiologist review cycles, more intensive query management, and more demanding clinical data collection workflows. These are not merely administrative inconveniences; they translate into material cost differences across large multi-site global trials.
What Incorrect Assessment Can Affect
- Drug Approval — FDA and EMA base efficacy evaluations directly on response criteria outcomes
- Label Claims — Approved indications and labeling language tied to trial-defined endpoints
- Market Authorization — Commercial viability contingent on demonstrated efficacy
- Reimbursement — Payers assess value based on published trial data
- Post-Market Studies — Ongoing obligations shaped by initial approval framework
QUICK COMPARISON
The table below summarizes how the three frameworks differ across key dimensions:
| FEATURE | RECIST 1.1 | iRECIST | mRECIST |
| Primary Use | Standard oncology — chemotherapy, targeted therapies, traditional treatments | Immunotherapy — checkpoint inhibitors, CAR-T, immune-based regimens | Liver cancer — HCC, liver-directed and anti-angiogenic therapies |
| Measurement Focus | Total tumor size (longest diameter of target lesions) | Immune response patterns, accounting for pseudoprogression | Viable, arterially enhancing tumor tissue only |
| Handles Pseudoprogression? | No — enlargement is scored as PD regardless of cause | Yes — iUPD allows confirmation before classifying as iCPD | Limited — not designed for immunotherapy contexts |
| Progression Confirmation Required? | No — single scan sufficient to declare PD | Yes — iUPD must be confirmed on subsequent scan (iUPD → iCPD) | Sometimes — depends on protocol design |
| Response Categories | CR, PR, SD, PD | CR, PR, SD, iUPD, iCPD | CR, PR, SD, PD (viable tissue basis) |
| Key Clinical Benefit | Standardized, universal applicability across solid tumors | Prevents premature discontinuation of effective immunotherapy | Accurately captures liver cancer treatment response |
| Common Cancer Types | All solid tumors | Immuno-oncology cancers (melanoma, NSCLC, renal cell, etc.) | Hepatocellular carcinoma (HCC) |
IMPACT III — DATA SCIENCE & PROGRAMMING
The Hidden Complexity: Programming Oncology Trial Data
Behind every clean line on a Kaplan-Meier curve is a clinical SAS programmer who has translated ambiguous clinical events into unambiguous data derivations. The choice of response criteria has made that job significantly more difficult.
RECIST 1.1 already demands careful handling of confirmation windows, new lesion tracking, and the distinction between target and non-target lesions. iRECIST layers on top of that the management of iUPD/iCPD handling, multiple progression states, and additional derivation rules that interact with standard ADRS datasets, BOR derivations, PFS calculations, and response confirmation logic.
These aren’t theoretical complications. They manifest as longer programming timelines, more complex validation requirements, higher rates of data queries, and — if not handled carefully — systematic errors in the derivation of primary efficacy endpoints. For global Phase III trials where a single endpoint determination drives a regulatory filing, the cost of ambiguity is measured in years and hundreds of millions of dollars.
SAS Programming Complexity Added by iRECIST
- iUPD / iCPD logic — Multiple progression state management across visit sequences
- Confirmation windows — Scan timing tolerances that interact with PFS derivation
- New lesion handling — New lesions trigger iUPD, not immediate PD
- BOR derivation — Best response must account for unconfirmed states
- ADRS datasets — Increased record volume and flag complexity
THE NEXT CHAPTER —Artificial Intelligence and Response Evaluation
All three frameworks share a foundational assumption that is now under pressure: that human radiologists, reading two-dimensional measurements from CT or MRI scans, can reliably capture tumor biology. Artificial intelligence is beginning to challenge that assumption in productive ways.
Future oncology trials will increasingly combine RECIST, iRECIST, and mRECIST with AI-based imaging biomarkers and real-world evidence — not replacing the criteria but augmenting them with richer data than any radiologist can manually extract from a scan.
Emerging AI Capabilities
- Automated Imaging — AI systems that automatically segment and measure target lesions across longitudinal scans, reducing inter-reader variability
- Tumor Segmentation — Three-dimensional volumetric analysis capturing tumor heterogeneity that linear diameter measurements miss
- Radiomics — Extraction of hundreds of quantitative features from imaging data that correlate with treatment response and survival
- Early Detection — AI-driven prediction of progression weeks before it becomes measurable by conventional RECIST criteria
“These criteria are foundational to modern oncology trials and directly influence how cancer drugs are developed, evaluated, and approved.”
OVERALL IMPACT — RECIST · iRECIST · mRECIST IN ONCOLOGY
The debate over how to measure tumor response is not a dispute among statisticians about decimal places. It is a debate about whether patients receive drugs that could save their lives, whether sponsors can demonstrate the value of treatments they have spent billions developing, and whether regulators can trust the evidence placed before them. Getting the framework right is not optional. It is foundational.