How to read a clinical study on stem cell therapy

Read the design line first; it bounds everything else. Then find the six load-bearing sentences in the abstract (design, sample, endpoint, result, claim, hedge) and ignore the rest until you've understood them.

Five words tell you the tier. Everything else in the abstract is context for these five. The rest of this guide is how to read them.

A study's design line, decoded
  1. Phase 2,Phase Where on the evidence pyramid.
  2. randomized,Assignment How patients got the arm.
  3. double-blind,Bias control Who knew what during follow-up.
  4. placebo-controlledComparator What the effect is measured against.
  5. trial of UC-MSCs in
  6. 82Sample How much noise swamps the signal.
  7. RA patients refractory to methotrexate.

A study is not a study is not a study.

If you spend even an hour reading "the research" on cell therapy, you'll find a study supporting almost any claim. And another one contradicting it. That's not because the field is incoherent. It's because "study" is a wildly elastic word and most of the writing patients encounter uses it without distinction. A 12-patient case series and a 1,200-patient randomized trial both get called "a study." They carry vastly different weight.

Most patients can't tell a registry write-up from an RCT, and the difference matters more than almost anything else on the page.

A 12-patient case seriesA 1,200-patient randomized trial

Levels of evidence: not all studies are equally heavy.

Clinicians read papers through a near-automatic mental filter that weighs design before content. A claim from a meta-analysis lands differently than the same sentence from a single-arm case series, even when the words are identical. The pyramid below is the version of that filter you can hold in your head. A claim's weight is bounded by the strongest source supporting it.

  1. 1Meta-analysis / systematic review Pooled data across multiple RCTs, with formal heterogeneity testing.Heaviest
  2. 2Randomized controlled trial (RCT) Prospective; randomized assignment; ideally double-blinded with placebo.Strong
  3. 3Prospective cohort Group followed forward in time; no randomization. Confounders not neutralized by design.Moderate
  4. 4Case series / registry "We treated N patients; here's what we observed." No comparator. Selection bias unbounded.Weak
  5. 5Expert opinion / single case report Anecdote with credentials. Useful for hypothesis generation, not for treatment decisions.Weakest

The thing to know is that most of the cell-therapy literature for non-orthopedic indications is at tier 3 or 4. Some published autism papers from named clinics are registry write-ups. Some MS papers are small prospective cohorts. RCTs exist for graft-versus-host, ARDS, and a handful of others, but they are the exception. When a clinic says "we have published research," ask which tier.

Sample size, to scale: what N= looks like.

N = 18

Pilot / case series

What it can find: only enormous effects. A study this small cannot detect a 20% improvement; the noise is louder than the signal. Useful for safety screening and hypothesis generation. Not useful for "this treatment works" claims.

N = 200

Phase 2

What it can find: moderate effects with reasonable certainty. A well-designed N=200 RCT can detect a response-rate difference of about 20 percentage points. Confidence intervals still wide enough that the result needs phase 3 confirmation.

N = 2,000

Phase 3 / registration

What it can find: small effects with high certainty and rare adverse events. The N at which regulators approve drugs. The cell-therapy literature has almost none of this for non-orthopedic indications.

How to read an abstract: six sentences do all the work.

A typical abstract is 250 words and looks dense. It isn't: about 80% of it is context. Six sentences carry the actual weight. If you can find these six and read them carefully, you have read the paper for almost all decision purposes.

Six sentences worth slowing down on, highlighted in a sample abstract.

Background. Mesenchymal stromal cells (MSCs) have shown immunomodulatory promise across autoimmune indications, but evidence in moderate-to-severe rheumatoid arthritis (RA) remains limited. We conducted a phase-2 randomized, double-blind, placebo-controlled trial of allogeneic UC-MSCs in RA patients refractory to methotrexate. Methods. Eighty-two adults (ages 35–65) with DAS28 >5.1 were randomized 1:1 to a single IV infusion of 1×10⁸ UC-MSCs or saline placebo. The primary endpoint was DAS28 change from baseline at 24 weeks; secondary endpoints included ACR20/50/70 response, serum CRP, and patient-reported outcomes. Results. At 24 weeks, mean DAS28 decreased by 1.8 (SD 0.9) in the UC-MSC arm versus 0.7 (SD 1.1) in placebo (between-group difference 1.1; 95% CI 0.6–1.6; p<0.001). ACR20 was met by 64% vs. 28%. No grade ≥3 treatment-related adverse events. UC-MSCs were well tolerated and associated with significant disease-activity reduction; larger confirmatory trials are warranted.

  1. Design

    "Phase-2 randomized, double-blind, placebo-controlled." This is the line that tells you which tier of the pyramid you're on. Three words to look for: randomized, blinded, controlled. Missing any one demotes the study by a tier. Missing all three (e.g. "single-arm open-label") puts you at tier 4.

  2. Sample

    "Eighty-two adults (ages 35–65) with DAS28 >5.1 were randomized 1:1." Sample size and randomization ratio. Under 30 patients per arm is exploratory. The ratio tells you whether the comparator group is the same size, important for the stats.

Read moreShow less

  1. Primary endpoint

    "DAS28 change from baseline at 24 weeks." The one outcome the study was built to measure. Secondary endpoints are interesting; only the primary one is binding for the study's claim. If the primary endpoint isn't met but a secondary one is, that's a much weaker result than the abstract may make it sound.

  2. Effect size + p-value

    "Between-group difference 1.1; 95% CI 0.6–1.6; p<0.001." Three things in that sentence: the size of the effect (1.1 DAS28 points), the confidence interval (0.6–1.6, wider intervals = less certainty), and the p-value (<0.001 here is strong; >0.05 is "not statistically significant"). A small p-value with a tiny effect size is still a tiny effect size.

  3. The authors' claim

    "UC-MSCs were well tolerated and associated with significant disease-activity reduction." Note "associated with": the careful version of the claim. The press release will say "shown to reduce" or "proven to treat." That escalation usually happens between the abstract and the marketing.

  4. The honest hedge

    "Larger confirmatory trials are warranted." This sentence is almost always in the abstract of any honest phase-2 result. It's the authors saying: this is real, but it isn't yet definitive. A paper that doesn't include some version of this sentence is doing something rhetorically odd.

How to read a confidence interval

The width of the bar is the certainty.

Tight CI, real effect 1.1 [0.6 – 1.6]

Strong. CI sits entirely above zero; effect is real and reasonably precise.

Wide CI, same point estimate 1.1 [0.1 – 2.1]

Weaker. Same effect on paper, but the true value could be anywhere from "barely real" to "twice as big." Usually smaller N.

CI crosses zero 0.8 [−0.3 – 1.9]

Not significant. The CI includes zero: no effect is among the plausible truths. Press release may still call this "trending toward benefit." It isn't.

What RCT, phase 2, ITT and CI mean

What follows isn't a complete list (there are hundreds), but these are the ones you'll see again and again in cell-therapy abstracts. The "what it tells you" column is the practical use: what the abbreviation means about a paper's strength when you see it in the design line.

RCT
Randomized Controlled Trial The gold standard.
Strong evidence If "double-blind, placebo-controlled" follows, even stronger.
DBRCT
Double-Blind RCT Neither patient nor evaluator knows assignment.
Strongest single-trial design Bias from expectation and assessment both controlled.
Open-label
Both parties know who got what. Often paired with "single-arm."
Demotes the study Placebo-effect and assessment bias both uncontrolled.
Read moreShow less
Phase 1
Safety / dose-finding Typically 20 to 100 people.
Not an efficacy claim If "phase 1" results are pitched as efficacy, that's overreach.
Phase 2
Efficacy signal-seeking Up to several hundred patients.
Hypothesis-generating Real signal possible; needs phase 3 to confirm.
Phase 3
Confirmatory Typically 300 to 3,000 patients.
Regulatory-quality evidence The phase used for drug approval.
ITT
Intent-To-Treat Analysis includes all randomized patients.
Conservative analysis The honest way to handle dropouts. "Per-protocol" excludes them, inflates apparent effect.
CI
Confidence Interval The plausible range of the true effect.
Tighter = better If the CI crosses zero, the effect is not statistically distinguishable from no effect.
p < 0.05
Conventional statistical significance threshold. 5% chance the result is noise.
Necessary, not sufficient A tiny effect with a small p-value is still a tiny effect.
SAE
Serious Adverse Event Hospitalization, death, life-threatening event.
Safety signal The SAE count in the treatment arm vs. control arm is the safety read.
N=
Sample size "N=12" = twelve patients.
The single most undersold number An N<20 study cannot find anything but enormous effects.
f/u
Follow-up duration How long patients were tracked.
Short f/u limits claims Six-month f/u cannot speak to durability at 24 months.

Eight phrases that should slow you down

Eight constructions that almost always mean something weaker than they sound. They aren't lies. They're rhetorical hedges: phrasings that let the writer claim something without saying it. Every one of them can be perfectly accurate; what they almost never are is what the casual reader takes them to mean. If you see one of these, look for the method.

  • Patients reported improvement

    Translates to: we asked them how they felt. No objective measure, no control group. This is a sentence about self-report, not a sentence about effect.

  • Many of our patients have seen

    Translates to: we have not counted. "Many" is unfalsifiable. The honest version is "in N=X patients, Y showed improvement" with both numbers.

  • Shown to be effective

    Translates to: in some study somewhere. Ask: shown by whom, in what design, with how many patients, against what comparator?

  • Promising preliminary results

    Translates to: phase 1 or single-arm phase 2 data. "Preliminary" is the giveaway. Real efficacy claims don't need the word.

Read moreShow less
  • In our experience

    Translates to: we have not published this. Clinical experience is real and useful, but it isn't evidence at the tier the abstract is pretending to occupy.

  • Statistically significant

    Translates to: the result is unlikely to be random noise. Says nothing about effect size. A 0.2-point change on a 10-point scale can be statistically significant and clinically meaningless.

  • Comparable to

    Translates to: we did not run a head-to-head. "Comparable to" almost always means "in a different study with a different population, the numbers look similar." That isn't equivalence.

  • Well-tolerated

    Translates to: there was no acute disaster. Says nothing about long-term safety or about adverse events below the SAE threshold. Look for the explicit SAE count.

How one finding escalates from abstract to ad copy.

The same result, three rooms.

  1. The published abstract

    Peer-reviewed

    "UC-MSCs were well tolerated and associated with significant disease-activity reduction at 24 weeks; larger confirmatory trials are warranted."

    Accurate

  2. The university press release

    PR office, not authors

    "New trial shows stem cell therapy reduces rheumatoid arthritis activity, a promising step toward a new treatment option."

    Drifting

  3. The clinic marketing page

    Unaffiliated with the study

    "Stem cells have been proven to treat rheumatoid arthritis. Our patients report life-changing results."

    Overreach

Apply this to us too.

We don't have peer-reviewed publications yet. Stem Cell Institute Panama has multiple peer-reviewed papers; we have none to date. By the hierarchy on this page, that puts us at tier 5 for any clinical-outcome claim. We try to compensate by being precise about manufacturing (which has objective release tests) and by publishing no outcome statistics of our own (we have none, and we say so).

Where our site says "some patients report" or "many patients report," read it by the standard above: it is self-report, not a measured effect. If you read a success rate, a "proven to," or any outcome number presented as our own, that wording is broken on our end and we want to hear about it. The standard we're inviting you to hold the rest of the field to is the one we're trying to meet here first. Write to [email protected] if you find one.

Stem Cell Institute vs Celva

Four questions to ask any cell-therapy outcome claim.

  • What was the study design? RCT, cohort, case series, or registry. The answer should be one word. If the answer is "many studies show," ask for a citation.

  • What was the sample size, and what was the control? "N=18, no control" is a different claim than "N=200 randomized vs. placebo."

  • What was the primary endpoint, and was it met? If the primary endpoint failed but a secondary one passed, the headline result is "negative trial with a positive secondary."

  • What was the effect size (not the p-value, the effect size)? A small p-value can hide a clinically meaningless effect. Ask for the magnitude, not the significance.

Read on