Reading Tumor Response on a Scan: What RECIST Can and Cannot Tell You

The short answer#

When an oncology report says a tumor is responding, progressing, or stable, that verdict almost always comes from a rulebook rather than a clinical impression. The dominant rulebook, the Response Evaluation Criteria in Solid Tumors (RECIST 1.1), asks a radiologist to pick a few representative lesions, measure them the same way on every scan, add the measurements into a single number, and compare that number to baseline. A drop of about 30 percent in the summed diameters is a partial response, a rise of about 20 percent is progression, and everything else is stable disease. The system is clean and reproducible for chemotherapy, but immunotherapy can make a tumor swell before it shrinks, which is why a companion ruleset, iRECIST, now asks for a confirmatory scan before a treatment is abandoned.

Key points#

• RECIST 1.1 converts serial imaging into four categories: complete response, partial response, stable disease, progressive disease.
• It tracks up to five target lesions (no more than two per organ), measured the same way each time, summed into one reference number.
• Roughly a 30 percent shrinkage signals partial response; roughly a 20 percent growth (plus at least 5 mm) signals progression.
• Immunotherapy can cause early growth that later reverses (pseudoprogression), which strict RECIST can misread as failure.
• iRECIST keeps the same measurements but requires a follow-up scan 4 to 8 weeks later before confirming progression.

Why a rulebook exists at all#

The instinct to eyeball a scan and declare a tumor "better" or "worse" is exactly what response criteria were built to discipline. If two radiologists at two hospitals can look at the same patient and reach opposite conclusions, then trial results become impossible to compare and regulators cannot trust an endpoint. RECIST 1.1, set out by Eisenhauer and colleagues in 2009, solves that by removing discretion at the moment it matters most: the reader must choose which lesions to follow before knowing how they will behave, then follow those same lesions to the end.

Think of it as pre-registering your measurements. By committing in advance to specific target lesions, the criteria block the very human temptation to notice whichever lesion happens to support the conclusion you were hoping for. That restraint, not any special accuracy, is what makes RECIST the standard reference for clinical trials.

The arithmetic, step by step#

At the baseline scan, the reader selects the target lesions: up to five in total, and no more than two in any single organ, favoring the largest lesions that can be measured reliably. Each solid lesion is measured along its longest diameter. Enlarged lymph nodes are handled differently and measured along their short axis. These figures are added into one sum of diameters, and that sum becomes the yardstick for every later scan.

From there the categories follow directly:

• Complete response: all target lesions vanish, and any pathological nodes fall below 10 mm on the short axis.
• Partial response: the sum of diameters drops by at least 30 percent.
• Progressive disease: the sum climbs by at least 20 percent above the smallest value ever recorded, and by at least 5 mm in absolute terms so that measurement noise alone cannot trip the alarm.
• Stable disease: anything between those two thresholds.

Lesions that are watched but not measured, and any brand-new lesion, are scored separately. One unequivocal new lesion counts as progression by itself, whatever the target lesions are doing.

Where a length measurement runs out of road#

The same discipline that makes RECIST trustworthy also flattens a complicated disease into a one-dimensional length. Size is only a proxy for how much living cancer is present, and the proxy slips in several predictable ways.

A lesion can enlarge from inflammation, bleeding, or scarring without holding any more viable tumor. A cancer can turn biologically nastier without ever crossing the 20 percent line. And because the criteria sum everything into a single number, a patient whose disease is genuinely mixed, one lesion melting away while another advances, can be averaged into a falsely reassuring "stable." RECIST was designed around drugs that kill dividing cells, so it simply assumes that shrinking means working and growing means failing. That assumption is where immunotherapy parts ways with the rulebook.

The immunotherapy problem#

Immune checkpoint inhibitors do not poison the tumor directly. They lift the restraints on a patient's own T cells, which then pour into the tumor. That flood of immune cells can briefly swell a lesion, and can even make tiny deposits visible for the first time, before the cancer starts to regress. Under strict RECIST 1.1, that early swelling reads as a 20 percent increase, earns the label progressive disease, and can prompt a clinician to stop a therapy that was about to work.

This pattern is called pseudoprogression: an apparent early rise in tumor burden followed by real shrinkage. It is genuine but not common. As reviewed in the Journal of Thoracic Disease, pseudoprogression with checkpoint inhibitors generally sits at or below roughly 10 percent, and is lower in many tumor types. The cost of getting it wrong is lopsided, though, because pulling an effective drug from that minority of patients does real harm.

The mirror-image danger is hyperprogression, a paradoxical speeding-up of tumor growth after immunotherapy begins. Reported rates span a wide band across studies, from about 4 percent to 29 percent, largely because the definitions differ. One approach, described by Champiat and colleagues, defines it as at least a doubling of the tumor growth rate versus the pre-treatment trajectory, which means you need a scan from before therapy started to know the earlier slope. RECIST 1.1 compares only against the on-treatment baseline, so it was never built to catch a change in growth rate at all.

What iRECIST changes, and what it does not#

To bring order to these atypical patterns, the RECIST working group published iRECIST in Lancet Oncology in 2017, led by Seymour and colleagues. It leaves the measurement machinery untouched and rewrites only how progression is confirmed. The first scan meeting the RECIST 1.1 threshold for progression is labeled immune unconfirmed progressive disease (iUPD) rather than a definitive failure. A confirmatory scan then follows no sooner than 4 weeks and no later than 8 weeks.

If that second scan shows more growth, the reading becomes immune confirmed progressive disease (iCPD), which needs supporting evidence such as a further 5 mm rise in the target sum, worsening non-target disease, or enlargement of the new lesions. If instead the tumor has shrunk or held steady, the progression clock resets and treatment can continue, capturing the delayed responders that strict RECIST would have thrown away. New lesions, which end the assessment outright under RECIST 1.1, do not confirm progression on their own under iRECIST; they trigger iUPD and must worsen at the next scan to count.

iRECIST is narrower than it sounds. It was designed mainly for consistent data collection in immunotherapy trials, and the working group still recommends RECIST 1.1 for primary endpoints in phase 3 studies, with iRECIST layered on as a structured exploratory tool. It lowers the chance of prematurely quitting a working drug, but it does nothing for hyperprogression, which needs the pre-treatment imaging that response criteria cannot supply. As the World Journal of Clinical Oncology review puts it, none of these frameworks measures living tumor directly; each infers biology from geometry.

The reading habit worth keeping#

If you appraise cancer evidence, do not memorize the thresholds. Interrogate them. Ask which criteria a trial applied, how progression was confirmed, and whether the observed pattern matches the drug's mechanism. A single label like "progressive disease" can carry very different meaning depending on the rulebook behind it, and knowing that difference is part of reading the evidence honestly.