Leaving Stem Cells Behind in the Chronic Stroke Brain — SB623, Two-Year Results in 18 Patients, and the Reckoning That Came Afterward

Six months have passed since the stroke, and improvement from rehabilitation tends to hit a ceiling—this is the chronic phase of stroke. Into that paralysis, still unmoving, you place genetically modified mesenchymal stem cells (MSCs) harvested from another person’s bone marrow, delivering them through a small hole in the skull to the tissue immediately around the infarct. Between 2011 and 2013, this was done to 18 patients at two centers in the United States. This article covers the two-year completion report. Look only at the numbers and “the paralysis moved.” But there is no control group.

Publication Details

• Paper title: Two-year safety and clinical outcomes in chronic ischemic stroke patients after implantation of modified bone marrow–derived mesenchymal stem cells (SB623): a phase 1/2a study
• Paper type: CLINICAL ARTICLE. A two-year, open-label, single-arm phase 1/2a interventional trial in humans
• Authors and affiliations: Gary K. Steinberg (corresponding author) and colleagues, 12 in all. Stanford University, New York University (NYU), the University of Pittsburgh Medical Center (UPMC), the University of California San Francisco (UCSF), SanBio, Inc., and Biostatistical Consulting Inc.
• Journal: Journal of Neurosurgery 2019, Volume 131, Issue 5, pages 1462–1472 (the flagship journal of the American Association of Neurological Surgeons [AANS]). Published online ahead of print on November 23, 2018
• DOI / registration number: 10.3171/2018.5.JNS173147 (PMID 30497166) / NCT01287936
• Impact Factor: 3.8 (JCR 2025 edition. Not confirmable on the official pages of the publisher, AANS; this is a value from an external metrics database)
• Open access: No (“bronze OA,” with AANS retaining copyright. No full text in PMC either)
• Funding and conflicts of interest: Funded under a contract with SanBio, Inc. Bates and McGrogan are full-time SanBio employees; Case and Yankee are former employees and current shareholders; Poggio owns the company that provides statistical services to SanBio. Wechsler is a consultant to both SanBio and Athersys. The medical writing was funded by SanBio as well
• Prior report: The 12-month interim report on the same 18 patients appeared in Stroke (2016;47:1817–1824)—the journal changed for the two-year report

About corresponding author Gary K. Steinberg: A cerebrovascular surgeon at Stanford University School of Medicine. He was Chair of the university’s Department of Neurosurgery from 1995 to 2020, and is a 2017 recipient of the Smithsonian American Ingenuity Award in the life sciences category.

What Wasn’t Understood Before?

The Introduction is blunt to the point of bleakness. Stroke prevalence in the United States was an estimated 7.2 million cases in 2014, with 795,000 new or recurrent events per year. While 80% survive one year, more than 70% are left with long-term disability. Acute intravenous t-PA reaches fewer than 10% of patients and thrombectomy fewer than 1%. And then the core of it—there is no approved treatment for chronic stroke.

👦 Student: When the room for improvement runs out at six months, what exactly has finished inside the brain?

🧬 Dr. Exotaro: It is closer to “the construction crew pulling out” than to anything ending. Right after the infarct there is a reconstruction boom: surviving neurons extend branches, neighboring regions take over the work, blood vessels are rebuilt. A few months on, that boom is over and the town settles down. Treatment in the chronic phase is, among other things, the question of whether you can call the departed construction vehicles back.

The tool aimed at that recall is the SB623 cell: an allogeneic bone marrow MSC transiently transfected with a plasmid encoding the human Notch-1 intracellular domain. The plasmid is lost through passaging, so the cells do not go on producing Notch-1 in the body, but the 12-month report (Stroke 2016) states that this stimulus changes patterns of DNA methylation and protein expression. The design philosophy is that the stimulus vanishes while the change in character remains (how long it remains, and what state the cells are in inside the human body, is unknown).

The hoped-for actions run to eight items: secretion of trophic factors and extracellular matrix (ECM) proteins, anti-inflammatory action, immunosuppression, angiogenesis, and so on. But these are findings from animal experiments and cultured cells that the 12-month report listed under the phrase “In preclinical studies”—they are not actions confirmed in humans. Nowhere does it say “the implanted cells turn into neurons,” either. The survival period of “usually less than one month” likewise refers to xenografts into animals, and the report states explicitly that survival in humans is unknown.

👦 Student: Isn’t it unreasonable to implant cells that disappear within a month and then expect improvement over two years?

🧬 Dr. Exotaro: The authors’ assumption runs like this—the cells are not residents, they are the ones who light the fire. Coal thrown into a cold fireplace turns to ash overnight, but once the fire catches, the room stays warm until morning. Molecules secreted by SB623 suppress inflammation and reignite the brain’s plasticity. That is the hypothesis known as the paracrine effect.

What Did This Paper Find?

379 people were screened and 18 (4.7%) enrolled. The criteria: age 18–75; a completed non-hemorrhagic infarct in the subcortical territory of the middle cerebral artery or the lenticulostriate arteries; residual motor deficit 6–60 months after onset; NIHSS (National Institutes of Health Stroke Scale, a measure of stroke severity) > 7 and mRS (modified Rankin Scale, a measure of functional independence) 3–4. To guarantee stability, NIHSS was measured twice during the three weeks before enrollment and a change within ±1 point was required. The 18 patients averaged 61.3 years of age, 11 women and 7 men, 1 Black patient and 0 Hispanic, a mean of 22.0 months (7–36) since onset, and a mean infarct size of 42.3 cm³.

The exclusion criteria are the more eloquent part. A second or later symptomatic stroke (recurrence), intracerebral hemorrhage, infarct size over 100 cm³, age 76 or older, a history of epilepsy or use of antiepileptic drugs, poorly controlled depression (Hamilton scale > 14), spasticity treatment within three months (botulinum toxin, phenol, intrathecal baclofen, and the like), poorly controlled diabetes, hypertension, renal failure, hepatic failure, or heart failure—all excluded. Brainstem and cerebellar infarcts fell outside the eligible territory. That is what the figure of 18 out of 379 is made of, and it means the results cannot be generalized directly to older patients, recurrent cases, patients undergoing spasticity treatment, or patients carrying depression.

The procedure is meticulous. Through a single burr hole, three needle tracks were planned in the subcortical tissue around the infarct, close to the motor pathways, spaced 5–6 mm apart from one another. The cannula was inserted to the deepest point, 20 μL of cell suspension was injected at 10 μL/min, and five deposits were made at 4–5 mm intervals while withdrawing—3 tracks × 5 points = 15 sites, 300 μL in total. Dosing was a single administration in three cohorts of six patients each: 2.5×10⁶, 5.0×10⁶, and 10×10⁶ cells. There is no mention of immunosuppressive drugs anywhere, and patients were monitored with anti-HLA antibodies.

The text states explicitly that the primary endpoint was “change from baseline in the ESS (European Stroke Scale) at 6 months,” and the result was met. ESS improved by 6.5 points at 6 months (95% CI 2.6–10.4, p < 0.01), and the Discussion says outright that “the primary clinical outcome measure … was achieved.” This is a fact that cannot be blurred.

That “met,” however, is a before-and-after comparison within a single arm, not a difference against a control group. On top of that, neither the paper nor the literature supplies a minimal clinically important difference (MCID) for the ESS—the smallest difference at which a patient actually senses a change—so the clinical meaning of 6.5 points cannot be assessed. And in the ClinicalTrials.gov registration (the 2023 version with posted results), the primary outcome measure is a single item, “number of participants who experienced adverse events,” with ESS one of nine secondary outcome measures—at odds with the text of the paper.

For ESS and the Fugl-Meyer measures (F-M total / FMMS), higher is better; for NIHSS and mRS, lower is better. The 2019 paper does not state the maximum scores, but the 2016 report on the same 18 patients calls FMMS a “100-point scale,” and the ClinicalTrials.gov registration likewise lists F-M total as 0–226 and FMMS as 0–100. The abstract says the effect “plateaued at 12 months and did not decline thereafter,” yet the point estimates fall—ESS 6.9→5.7, FMMS 11.4→10.4—the p-values weaken, and the confidence intervals widen. Effect sizes in small samples are prone to overshoot (the winner’s curse), which is part of the groundwork for why this improvement signal was not reproduced in the sham-controlled trial that followed.

“Clinically meaningful improvement”—which this paper defines with a threshold of an increase of 10 points (10%) or more on the FMMS (Fugl-Meyer Motor Score)—was reached by 13 of the 18 patients (72.2%). It took a mean of 78.4 days (SD 64.1), and 9 of the 13 got there by two months. The maximum mean change in the group that reached it was +12.4 points at two months, against +1.5 points in the 5 who did not. There was no correlation between improvement and cell dose, age, or severity.

The contents of that 72.2% call for caution, though. The definition is “reached +10 points at any single visit during the trial,” not a response rate at a fixed time point. There were nine visits in year one and one in year two. With 10 chances to measure, variability alone raises the probability of “crossing the line at least once.” In fact, the subsequent ACTIsSIMA changed this to “+10 points or more at 6 months,” and the response rate even in the SB623 arms was 7/55 (12.7%) and 9/56 (16.1%). The gap between 72.2% and 13–16% reflects not only whether the cells work but also this difference in definition.

👦 Student: Why did all four scales improve while mRS alone didn’t budge?

🧬 Dr. Exotaro: ESS, NIHSS, F-M, and FMMS are all yardsticks of impairment—they measure how far the arm can be raised. mRS, by contrast, is a yardstick of disability, of independence in daily life. In musical terms, it is the difference between “how many notes of the scale you can produce” and “whether you can play a piece.” The scale got wider; the piece still could not be played. And since neither the Barthel Index nor the FIM nor any QOL scale was measured, there is no way to verify whether the improvement translated into daily life.

All 18 patients experienced at least one treatment-emergent adverse event (TEAE). There were no withdrawals, no dose-limiting toxicity, and no deaths. The most common was surgery-related headache at 88.9% (16/18), followed by nausea at 33.3%, depression, muscle spasticity, and vomiting at 22.2% each, and pneumocephalus, subdural hematoma, seizure, and pneumonia in 2 patients each (11.1%). Not a single TEAE was judged probably or definitely related to the cell therapy, and the only “possibly” related event was muscle spasticity in one patient (5.6%)—though this adjudication has changed since the 2016 report (discussed below). By contrast, 55.6% were tied to the surgical procedure as probably related and 44.4% as definitely.

There were 9 serious adverse events (SAEs) in 7 patients. All resolved without sequelae, and their relationship to the cell therapy was unrelated or unlikely. But three of them were related to the surgical procedure—pneumonia possibly, seizure probably, subdural hematoma/hygroma definitely. Although a history of epilepsy and use of antiepileptic drugs were exclusion criteria, new-onset seizures were reported in 2 patients (11.1%). Only one SAE was explicitly tied to the procedure (probably), but this cannot be taken lightly as a risk of a procedure that passes through the cortex.

TEAEs clustered in the first year—65 events in year one against 11 in year two, though these counts cover only events occurring at 10% or higher frequency. The authors themselves append the unflattering confounder that “the difference between nine visits in year one and one visit in year two could also explain this.” The only visit in year two was the one at 24 months. The reason fewer adverse events turned up in year two may simply be that they were never captured.

The most suggestive finding is this. One to two weeks after implantation, a transient lesion on T2-weighted FLAIR appeared in the ipsilateral cortex—mainly near the premotor cortex—in the great majority of patients, and disappeared within one to two months. Lesion size correlated significantly with ESS improvement at 24 months (r = 0.619, p < 0.05) and NIHSS improvement (r = −0.735, p < 0.01). It did not correlate with F-M total or FMMS, however, and the authors state explicitly that this was post hoc; the cause is “unknown, but possibly inflammation, a graft-host response, or gliosis.”

That correlation, mind you, comes with neither a stated number of patients behind it nor a 95% confidence interval. For 12 months the paper says only 13 of 18; the number at 24 months is unknown. Changes were tested with the Wilcoxon signed-rank test, yet the correlation alone uses Pearson, which is vulnerable to outliers. And at the most fundamental level, this trial cannot distinguish whether the lesion was produced by the cells or by the needle itself.

How Will the Future Change? (The Path to the Clinic)

The Conclusions are honest. They keep the findings to “highlighting the potential of SB623,” state explicitly that “a randomized, double-blind, controlled phase 2b trial (ACTIsSIMA) is currently under way,” and go as far as disclosing the design: roughly 156 patients randomized 1:1:1 to two doses and sham (partial burr hole only), with the proportion improving 10 points or more on FMMS at 6 months as the primary endpoint. The promise of a reckoning was written into the paper itself.

The answer arrived just two months later. On January 29, 2019, SanBio and Sumitomo Dainippon Pharma announced that ACTIsSIMA (NCT02448641, 163 patients randomized and treated) did not meet its primary endpoint. The number of patients improving 10 points or more on FMMS at 6 months was 7/55 in the 2.5×10⁶ arm, 9/56 in the 5.0×10⁶ arm, and 7/52 in the sham arm, p = 0.6743 (results posted on ClinicalTrials.gov). The 72.2% from the single-arm trial in 18 patients was not reproduced as a between-group difference the moment a sham operation was placed as the control.

👦 Student: What does a sham operation actually involve?

🧬 Dr. Exotaro: The sham in ACTIsSIMA was a “partial burr hole”—the skull is drilled partway through, but no needle enters the brain parenchyma. You are put under anesthesia, your head hurts afterward, and you feel you have “received treatment”; everything is the same up to that point, and only the cells and the needle track differ. Since no difference emerged, it has not been demonstrated that the improvement seen in the single-arm trial involved an effect specific to the SB623 cells. The effect of the sham surgery, expectation and assessor bias, measurement variability, natural history—this result does not tell you which of them contributed how much.

In September 2020, SanBio released a post hoc analysis restricted to the 77 patients with smaller infarct volumes (47% of enrollment)—on a composite FMMS endpoint at 24 weeks, 49% of 51 in the SB623 group versus 19% of 26 controls, p = 0.02. But this was an after-the-fact analysis: not prespecified, narrowed to less than half the participants, with the endpoint definition swapped out as well. As of a PubMed search on August 4, 2026, no peer-reviewed primary report of ACTIsSIMA can be confirmed; the only primary sources are ClinicalTrials.gov and company press releases.

The SB623 cell, however, did not die. Let us get the chronology right, though. STEMTRA (NCT02416492), which targets chronic traumatic brain injury (TBI), was not a trial launched in response to the failure of ACTIsSIMA. Its registration was posted in April 2015 and it began in July 2016, running in parallel with ACTIsSIMA (posted May 2015, began March 2016). Double-blind and sham-controlled, STEMTRA met its primary endpoint on change in FMMS at 24 weeks (Kawabori et al., Neurology 2021;96(8):e1202–e1214). In the approval dossier, the pooled SB623 group (46 patients) showed 8.3 ± 10.6 and the sham group (15 patients) 2.3 ± 4.7, p = 0.0401 (mean ± SD). Those 46 + 15 = 61 patients are the analysis set left after excluding, from 63 enrolled, the 2 in whom a safe needle trajectory could not be secured. Even in that same comparison of pooled SB623 versus sham, however, the significant difference disappears at 48 weeks.

In Japan, on July 31, 2024, “AKUUGO® Suspension for Intracranial Implantation” obtained conditional and time-limited marketing approval with the indication “improvement of chronic motor paralysis associated with traumatic brain injury.” It was added to the national price list on May 20, 2026 (72,716,528 yen per administration) and launched on May 21. “Conditional and time-limited” is a scheme that permits marketing for a fixed period at the stage where efficacy has been “inferred”; it does not mean efficacy has been confirmed. The dosage and administration (5×10⁶ viable cells, 300 μL, three tracks from a single small hole with five deposits each) are all but identical to the procedure in this paper—and of the two indications sharing the same cells and the same procedure, it was the TBI side that reached approval. As of August 2026, there is no approval of SB623 for ischemic stroke in Japan or in the United States. HeartSheet (Terumo), approved under the same scheme in 2015, failed to show efficacy in post-marketing evaluation and had its approval withdrawn in July 2024. The verification of AKUUGO is still ahead of us.

Other companies have hit the same wall. ReNeuron’s CTX0E03 had its primary endpoint met in exactly one patient in the single-arm PISCES-2 (23 patients), and the confirmatory PISCES-III was halted at 15 patients. Athersys’s MultiStem missed in MASTERS (126 patients), with an odds ratio of 1.08 (95% CI 0.55–2.09), p = 0.83 at Day 90, and Japan’s TREASURE also returned P = .90. The phase 3 MASTERS-2 had enrollment suspended after a 2023 interim analysis judged that 300 patients provided insufficient power, and no formal results have been published. Athersys filed for Chapter 11 in January 2024, and ReNeuron entered administration in 2024, surviving since then as a private company.

How to Critically Read This Research (Limitations and Paths to Improve It)

In fairness, start with the strengths. They ran a trial that dosed 18 patients all the way through two years and published a completion report that documents even the two withdrawals and losses to follow-up (24-month data cover 16 patients). Nor do they hide unflattering findings—mRS not moving, no correlation between dose and improvement, the 24-month FLAIR correlation being non-significant for F-M/FMMS. They chose a population past the recovery plateau, and according to the 12-month report the participants were not given rehabilitation during the trial—a control of confounding that comparable trials lacked.

With that said, the limitations. First, the consequences of a single-arm, open-label design with n = 18. The paper justifies the single arm by saying “the baseline values provide the control,” but a baseline is not a concurrent control, and it cannot rule out any of the following: (1) spontaneous recovery, (2) the effect of the surgery itself along with placebo and expectation effects, (3) unblinded assessor bias, (4) practice effects, or (5) regression to the mean. The authors’ counterargument that “because the deficits are chronic and stable, a placebo effect is unlikely” begs the question. “Stable without treatment” and “unresponsive to the non-specific components of an intervention” are different propositions, and the intervention here is a stereotactic intracerebral injection through a burr hole, with every participant knowing they received an implant. In the same paragraph, however, the authors themselves also wrote that “a placebo effect cannot be completely excluded. This point is being tested in the ongoing ACTIsSIMA.”

Second, NIHSS was the only measure whose stability was confirmed. Baseline stability is not documented for F-M total (+19.4) or FMMS (+10.4), the measures that showed the largest improvement. Third, multiple comparisons. Figure 1 alone runs about 32 Wilcoxon tests—4 scales × roughly 8 time points—with α left at 0.05, and since ESS and NIHSS overlap in content and FMMS is a subcomponent of F-M total, “all four scales improved” is not independent corroboration.

Fourth, the number of patients actually analyzed. The abstract says “in the 16 patients who completed 24 months,” while the legend of Figure 1 says “n = 18.” Yet the per-time-point n printed below the axis of Figure 1 gives 15 for ESS/NIHSS and 14 for F-M/FMMS at 24 months—three different values of n coexisting in a single figure. The abstract’s “F-M total score, 19.4” is in fact a result from 14 patients. There is no description of how missing data were handled; it is effectively a complete-case analysis, with no sensitivity analysis.

Fifth, the distance between statistical significance and clinical meaning. The authors write that “because mean FMMS increased by 10.4 points, patients on average experienced a clinically meaningful improvement,” but MCID is an individual-level concept—the smallest change a single patient perceives—and applying it to a group mean is a category error. The distribution is bimodal, and the lower bound of the 95% CI for FMMS was 4.0, less than half the threshold of 10.

Sixth, zero events is not 0%. For an event observed zero times in n = 18, the upper 95% confidence limit on the true rate is roughly 16.7% (the rule of three). “No deaths” and “no tumor formation” deny nothing about events occurring at a rate below 17%.

👦 Student: The 12-month report and the two-year report cover the same 18 patients, don’t they? Do the numbers match?

🧬 Dr. Exotaro: Line them up and they diverge in several places. Baseline NIHSS is 9.44 (SD 1.89) in the 2016 version and 9.3 (1.7) in the 2019 version. The footnote to the table in the 2019 version says only “One value has been modified,” with no explanation of which value was corrected or how. The 12-month change also shifts from −2.00 (−2.7 to −1.3) to −1.9 (−2.6 to −1.1), and the p-value for mRS changes from “P = 1.0000” to “p < 0.99.”

👦 Student: Do numbers really move after the fact like that?

🧬 Dr. Exotaro: The adverse events are more troubling still. Headache probably or definitely related to the procedure is 77.8% (14/18) in the 2016 version and 66.7% (12/18) in the 2019 version. TEAEs judged possibly related to the cells also fell, from 4 events in the 2016 version (22.2%: 2 muscle spasticity, 1 gait disturbance, 1 procedural headache) to a single patient with muscle spasticity (5.6%) in the 2019 version. Such changes could arise from a re-adjudication of causality or a change in coding, but the reason is written nowhere in the paper. The capper is that three disclosures present in the 2016 version—“multiplicity was not corrected for,” “participants were not given rehabilitation,” and “the neurologists were not blinded”—are not carried over into the 2019 version.

The account of governance is thin as well. The statistical analysis is listed as a single name, “Statistical analysis: Poggio”—one person carrying a conflict of interest did the work, and no independent analysis or reanalysis is reported in the paper. There is no mention of a data safety monitoring board (DSMB) in the paper either. This deserves to be judged separately, though: the current ClinicalTrials.gov record marks Data Monitoring as Yes, and the protocol, the informed consent form, and the statistical analysis plan (SAP) were all released when the results were posted in 2022–2023. The problem is not that these things did not exist but that readers had no material for verification at the time the paper was published. In the Author Contributions, the one person out of 12 who “reviewed the submitted version of the manuscript” is Steinberg.

So how could the quality have been raised? (1) Randomizing the waiting period (a delayed-start design)—if everyone is to be dosed and the wait is assigned to 0, 3, or 6 months, you get a concurrent control period without denying anyone the treatment, and you can estimate natural history, practice effects, and regression to the mean. (2) A vehicle-only arm, or an arm in which only the cannula is passed, would have separated whether the FLAIR lesion came from the cells or from the needle track. (3) Video-recording the examinations, randomizing the order of the time points, and having blinded assessors outside the institution score them makes “blinding of the time point” achievable even in a single-arm trial. (4) Abandoning 32 separate tests in favor of modeling everything at once with a mixed model for repeated measures (MMRM), and defining responders at a fixed time point (which is exactly what ACTIsSIMA changed to).

Dr. Exotaro’s Perspective

I work on applying MSCs and extracellular vesicles (EVs) to spinal cord injury (SCI) myself, so I cannot take my eyes off this one sentence: “sustained neurological recovery may result not from engraftment and long-term survival of the cells but from paracrine effects of proteins and molecules secreted by the implanted cells.” If what works is not the cells but the message the cells leave behind, then it should be enough to deliver the message. And in that case, is it really necessary to open a burr hole and pass three needles? That question was not somebody else’s problem.

Reading this paper alongside ACTIsSIMA changes how the question should be framed. If the paracrine effect were truly large, the cell arms should have pulled away from a sham that never reaches the brain parenchyma. They did not. At least part of the “+10.4 points” from the single-arm trial may have been a number produced neither by the message nor by the needle track but by the act of measuring itself—expectation, motivation, practice, unblinded scorers. When an EV effect shows up in an animal model, the first thing to do is to design, in advance, the answer to “what control would make this effect disappear?”

One finding still keeps pulling at me: that transient T2-FLAIR lesion. A trace of the brain responding, captured on imaging, and tied to recovery two years later—it is the prototype of the biomarker that anyone trying to move the brain with EVs would give an arm for. Yet as of August 2026 no follow-up study verifying it can be found, and for ACTIsSIMA, PISCES-III, and MASTERS-2 alike, a PubMed search turns up no peer-reviewed primary report of the main results. The more negative a result is, the less likely it is to survive in the literature—and that reality robs the next researcher walking the same road of a map.

Finally, to patients and their families. As of August 2026, there is no cell therapy for the chronic sequelae of ischemic stroke that has been approved under Japan’s Pharmaceuticals and Medical Devices Act and covered by public insurance. AKUUGO’s indication is traumatic brain injury only. The “acceptance of a Class 2 regenerative medicine provision plan” under the Act on the Safety of Regenerative Medicine is likewise not a statement that the Ministry of Health, Labour and Welfare has approved efficacy; it is a separate scheme from marketing approval under the Pharmaceuticals and Medical Devices Act. Please use the numbers in this article—72.2% reached the mark in 18 patients under the loose definition of “+10 points at any single visit,” and p = 0.6743 once a sham control was placed in 163 patients and the definition was rewritten as “+10 points at 6 months”—as your yardstick when you read a brochure from a self-pay clinic.

And what I do not want to forget is the fact that these 18 people actually had a hole drilled in the skull and three needles passed into the brain. What was left in exchange for that burden is not small—two years of safety data with no deaths and no dose-limiting toxicity, a primary endpoint met within a single arm, and the clue of the FLAIR lesion. What was not confirmed is efficacy specific to the cells. It is precisely because this data exists that we can say clearly today that a single-arm before-and-after comparison must not be read as evidence of efficacy. Without cynicism and without exaggeration, that is something I want written down.