Belumosudil safety in chronic graft-versus-host disease is dominated by three problems that never make a headline: musculoskeletal tightness and swelling, mucosal and viral infections, and shifts in liver enzymes and immunoglobulin G. Across 1,797 primary-suspect reports covering 6,056 adverse events, no new systemic organ toxicity emerged. For a newer immunosuppressant, that is close to the best news on offer.
The numbers are what make the dataset worth reading. Muscle tightness carried a reporting odds ratio of 10.25, skin swelling 19.32, against a disease-matched background. Respiratory syncytial virus sat at 4.02, viral gastroenteritis at 4.81, hepatic enzyme elevation at 5.64 and a reported fall in blood immunoglobulin G at 5.75.
Taken together, they describe a drug behaving exactly as a ROCK2 inhibitor should in a chronically inflamed, chronically immunosuppressed patient. It stiffens connective tissue. It leaves airway and gut mucosa exposed to viruses. It nudges immunoglobulins and transaminases. None of that is a safety catastrophe. All of it is monitorable, provided somebody actually monitors.
Why belumosudil safety needed real-world data
Chronic graft-versus-host disease is a fibroinflammatory syndrome that follows allogeneic stem cell transplantation and can involve skin, mouth, eyes, lungs, liver, gut and joints. Treatment has long rested on corticosteroids plus a calcineurin inhibitor, with mycophenolate mofetil often added as a steroid-sparing third agent. Belumosudil, a selective ROCK2 inhibitor, was approved for patients who have failed at least two prior lines, a group with few options and heavy comorbidity.
Pivotal trials in that group are small by necessity. They establish efficacy in a serious disease with no good alternatives; they cannot characterize rare events, slow toxicities, or what happens when a new drug meets the dozen other agents a transplant patient already takes. Post-marketing spontaneous reporting fills that gap imperfectly but usefully, as long as indication confounding is controlled.
That control is the technical heart of the work. Patients with this disease generate reports about the disease itself, so a naive disproportionality analysis of a new drug in that population will simply rediscover graft-versus-host disease and call it toxicity. Anchoring the comparator to a disease-restricted background, rather than the whole database, is what separates a signal from a restatement.
How 1,797 reports were filtered
Four stages, and every one of them threw something away.
Stage | What happened | Result |
Case extraction | Primary-suspect reports for the drug, Q3 2021 to Q4 2025 | 1,797 reports, 6,056 events |
Background | Disease-restricted comparator cohort | 921,529 reports |
Signal detection | Four algorithms required to agree | 61 preferred terms |
Statistical validation | False discovery rate at q below 0.05 | 60 terms retained |
Clinical refinement | Administrative, indication-synonym and macro-outcome terms removed | 45 core terms |
The four-algorithm requirement deserves a word, because it is stricter than it sounds and weaker than it looks. Reporting odds ratio, proportional reporting ratio, a Bayesian confidence propagation measure and a multi-item gamma Poisson shrinker all had to cross their thresholds. All four come from the same two-by-two table, so agreement means consistency, not independent replication. The authors say so directly.
The refinement stage dropped 16 terms. Most were indication synonyms such as graft-versus-host disease of lung, eye or skin, describing the treated disease rather than a drug effect. The rest were administrative codes for scheduling or product-use problems. Removing them is uncontroversial, and it is also the step most likely to kill a real signal by accident, which is why the excluded list was published.
Composition looks like the indication. Sex was nearly balanced at 50 percent male and 44 percent female, 54 percent of patients were 45 to 74, 93 percent of reports came from a single national market, and physicians filed 54 percent. Hospitalization appeared in 40 percent of reports with a documented outcome, 551 cases, and death in 11 percent, 147 cases. Volume climbed every year and peaked at 670 reports in 2025.
Where the signals cluster at the organ-system level
Twenty-four organ-system categories carried reports. Frequency and disproportionality told different stories, and the gap is the lesson.
Organ-system category | Share of events | Signal strength |
Injury, poisoning and procedural complications | 18.0 percent (1,093) | Reporting odds ratio 3.03 |
General disorders and administration site conditions | 15.9 percent (965) | Not a leading signal |
Gastrointestinal disorders | 10.2 percent (617) | Not a leading signal |
Infections and infestations | 8.9 percent (538) | Not a leading signal |
Investigations | 7.8 percent (470) | Not a leading signal |
Social circumstances | Not a leading frequency | Reporting odds ratio 5.69 |
Surgical and medical procedures | Not a leading frequency | Reporting odds ratio 2.84 |
Eye disorders | Not a leading frequency | Reporting odds ratio 2.24 |
The biggest categories are not the strongest signals, which is the standard pattern in spontaneous reporting and the reason frequency tables mislead on their own. Infections and infestations produced 538 events, a large absolute number, but did not stand out against a background population that is also chronically infected.
Eye disorders merit a note. That category reached a reporting odds ratio of 2.24 with agreement across all four algorithms and a Bayesian lower bound of 1.88, just below the usual shrinkage cutoff. Chronic graft-versus-host disease causes dry eye, sclerotic conjunctival disease and lacrimal dysfunction, so eye findings are part disease and part drug. Sensible conclusion: keep ocular surveillance in follow-up, and do not attribute every dry eye to the ROCK2 mechanism.
The three-tier evidence framework
Instead of dumping 45 signals into one list, the analysis sorted them by what each can support. That stratification is the most transferable idea in the paper.
Tier | Definition | Count | Use |
Tier 1 | Signals already documented in product labeling or pivotal trials | 6 | Confirms the pipeline detects what is known |
Tier 2 | Unlisted events with a plausible mechanistic link to ROCK2 inhibition | 6 | Hypotheses worth prospective follow-up |
Tier 3 | Clinical confounders reflecting progressive disease and debilitation | 33 | Reported for transparency, not attribution |
A belumosudil safety framework of this kind blocks the two standard failures of pharmacovigilance writing. The first is treating every disproportionality as a drug effect. The second is discarding everything unlisted as noise. Splitting confirmation from hypothesis from confounder lets a reader decide which part applies to the patient in front of them.
The ratio also says something. Thirty-three of 45 core terms landed in the confounder tier, which means most of what a spontaneous reporting system flags in this population is the disease talking. That is not an indictment of the drug. It is a description of how hard attribution is in a fibrosing, multi-organ, polypharmacy condition.
Musculoskeletal and mucosal signals dominate
The pharmacological core sits in two clusters, and both map onto known ROCK2 biology.
Signal | Reporting odds ratio | Cluster | Clinical meaning |
Skin swelling | 19.32 | Musculoskeletal and soft tissue | Fluid and tissue change rather than allergy |
Muscle tightness | 10.25 | Musculoskeletal and soft tissue | Connective tissue stiffness on a fibrotic background |
Blood immunoglobulin G decreased | 5.75 | Laboratory | Humoral immune suppression |
Hepatic enzyme increased | 5.64 | Laboratory | Usually transient transaminitis |
Viral gastroenteritis | 4.81 | Mucosal and viral infection | Enteric viral exposure without barrier defense |
Respiratory syncytial virus infection | 4.02 | Mucosal and viral infection | Lower respiratory risk in a lung-involved population |
ROCK2 inhibition shifts the balance between T helper 17 and regulatory T cell programs and modulates actin cytoskeleton dynamics in fibroblasts and endothelial cells. Both effects plausibly produce tissue stiffness, impaired barrier function and weaker humoral responses. The signal pattern is therefore mechanistically coherent rather than random, which raises confidence in the cluster instead of lowering it.
The immunoglobulin finding needs its own paragraph because it is easy to miss and easy to act on. A reported fall in blood immunoglobulin G at 5.75 is not a diagnosis of hypogammaglobulinemia, and spontaneous reports cannot supply the quantitative levels that would define one. But in a patient on a ROCK2 inhibitor plus corticosteroids plus mycophenolate mofetil, recurrent sinopulmonary infection and a low IgG together justify measuring rather than assuming.
Mycophenolate mofetil and the co-medication problem in chronic GVHD
Almost nobody on belumosudil is on belumosudil alone. Standard therapy pairs corticosteroids with a calcineurin inhibitor, and mycophenolate mofetil is widely used as the third agent for steroid sparing in exactly the population that progresses to second-line treatment.
That shared exposure creates three interpretive problems. First, mycophenolate mofetil suppresses lymphocyte proliferation through inosine monophosphate dehydrogenase inhibition, so it contributes independently to the viral infection signals this analysis attributes to the ROCK2 inhibitor. Second, mycophenolate has no routine therapeutic drug monitoring, so its exposure is unknown and unmeasured in every report here. Third, when viral gastroenteritis or a falling IgG appears, the bedside question is which immunosuppressant to reduce, and reporting data cannot answer it because they contain no dose for any co-medication.
The analysts handled part of this with a monotherapy sensitivity analysis. Restricting to reports naming no other suspect, concomitant or interacting drug left 1,255 reports, and baseline covariates balanced against the main cohort at standardized mean differences below 0.10 for nearly every variable. Signals survived that restriction, which is real evidence the cluster is not purely a co-medication artifact.
For a manufacturer of mycophenolate mofetil the implication runs against the usual marketing line. In a combination where one partner is monitored and the other is not, product consistency is the only control the unmonitored partner ever gets. Assay content, dissolution and batch uniformity become the substitute for a drug level, particularly in a population whose intestinal absorption is already compromised by graft-versus-host disease of the gut.
Who is vulnerable: fatal cases, age and sex
Stratification turned a flat list into three recognizable risk profiles.
Stratum | Enriched signals | Statistical note |
Fatal cases | Obliterative bronchiolitis and muscle atrophy | Descriptive enrichment rather than a disproportionality estimate |
Age 65 and older | Joint stiffness and lip pain | P below 0.05 on exact testing |
Female sex | Transaminase elevation, skin atrophy and oral mucosal blistering | P below 0.05 on exact testing |
The fatal-case pattern is the one that should change practice. Obliterative bronchiolitis is the pulmonary manifestation of chronic graft-versus-host disease and the leading cause of late mortality in this population, so its enrichment among fatal reports is partly disease progression. Muscle atrophy sitting beside it, though, points at a debility pathway that includes prolonged corticosteroid exposure and possibly the drug itself.
Age and sex differences were tested with exact methods and reached conventional significance. They also rest on modest subgroup cells inside an already selected reporting population. Direction is plausible: older patients accumulate joint disease, and females report more mucosal and cutaneous immune reactions across drug classes. Magnitudes are not reliable enough to justify a monitoring schedule set by demographic alone.
The counterintuitive finding: no new organ toxicity, and that is the result
The most important sentence in any belumosudil safety analysis is a negative one. After filtering, there was no novel systemic organ toxicity: no cardiac, renal, hepatic or neurological cluster strong enough to stand out from a disease-matched background.
For a new immunosuppressant entering a population with a dozen competing causes of organ dysfunction, that absence is the finding. The profile extends the mechanism rather than surprising it: tissue stiffness, barrier compromise, humoral suppression. The monitoring burden therefore falls where the mechanism predicts, not across every organ system.
The second surprise is subtler. The strongest single signal, skin swelling at 19.32, is not a classic adverse drug reaction. It is a soft-tissue change that could reflect fluid shifts, corticosteroid effects, sclerotic skin disease or the ROCK2 mechanism. A very high disproportionality on a nonspecific term is a warning about term definition as much as about the drug. Reading the number without the term is how pharmacovigilance produces headlines that later evaporate.
What to monitor, and at what interval
The signals translate into a belumosudil safety monitoring plan both specific enough to act on and modest enough to survive a busy clinic.
Domain | What to check | Trigger for action |
Muscle and joint | Tightness, range of motion, new contracture, joint stiffness | New functional limitation rather than a subjective ache |
Skin and mucosa | Swelling, sclerosis, oral blisters, atrophy | Rapid change or mucosal ulceration |
Liver | Transaminases at baseline and during dose changes | Rise above three times the upper limit of normal |
Immunoglobulin | Quantitative IgG if recurrent sinopulmonary infection | Recurrent infection plus low IgG |
Airway | Spirometry with forced expiratory volume in one second, especially with pulmonary involvement | Declining trend rather than a single value |
Infection | Respiratory syncytial virus and enteric viral testing during symptoms | Any new hypoxemia in a lung-involved patient |
The airway row saves lives. Obliterative bronchiolitis progresses silently, and the enrichment of pulmonary disease among fatal reports argues for scheduled spirometry rather than symptom-driven testing. A falling forced expiratory volume in one second beats any single reported adverse event.
The infection rows matter for another reason. On triple immunosuppression, viral gastroenteritis and respiratory syncytial virus are not incidental. They are the visible edge of a barrier and humoral defense problem, and they are exactly the events where cutting mycophenolate mofetil or adjusting corticosteroids often works better than adding another drug.
Limitations worth stating plainly
Any belumosudil safety conclusion drawn from spontaneous reporting is provisional, because reports are voluntary, unverified and unvalidated, and none in this dataset was adjudicated for causality. Reporting rates cannot be converted to incidence, so nothing here says how often these events occur in treated patients. Under-reporting is uneven: dramatic events get filed, gradual joint stiffening does not.
Disease restriction cuts indication confounding without eliminating it. The comparator shares the indication but not the line of therapy, and a patient on a third-line ROCK2 inhibitor is by definition further along than the average patient in the background. Residual confounding by disease severity cannot be removed.
Co-medication exposure, dose, duration and calendar time are all absent. The monotherapy sensitivity analysis is a partial remedy, not a replacement for a registry with structured drug data. And because reporting concentrated in one national market over one five-year window, the profile reflects a single health system's prescribing habits.
Bottom line
Belumosudil safety in chronic graft-versus-host disease looks like its mechanism: musculoskeletal tightness and swelling, mucosal and viral infections, transaminase and immunoglobulin shifts, and no new systemic organ toxicity. The highest reporting odds ratios were skin swelling at 19.32, muscle tightness at 10.25, immunoglobulin G decrease at 5.75 and hepatic enzyme rise at 5.64. Monitoring should follow the mechanism, with scheduled spirometry, quantitative immunoglobulin when infection recurs, transaminases during dose changes, and a lower threshold for reducing co-immunosuppression such as mycophenolate mofetil when viral events appear.
FAQ
What are the most common safety signals for belumosudil?
Musculoskeletal and soft tissue events lead, with skin swelling at a reporting odds ratio of 19.32 and muscle tightness at 10.25 against a disease-matched background. Laboratory signals follow, including decreased blood immunoglobulin G at 5.75 and raised hepatic enzymes at 5.64. Mucosal and viral infections form the third cluster, with viral gastroenteritis at 4.81 and respiratory syncytial virus at 4.02. No novel cardiac, renal or neurological toxicity signal emerged after clinical refinement of the term list.
How reliable is spontaneous reporting for a new immunosuppressant?
It is reliable for generating hypotheses and unreliable for quantifying risk. Reports are voluntary, unverified and rarely adjudicated for causality, and reporting rates cannot be converted into incidence because the denominator of exposed patients is unknown. The strength of this design is speed and coverage of rare events that trials miss. The weakness is that a high reporting odds ratio can reflect a memorable event, a marketing-driven reporting wave or a loosely defined term rather than a pharmacological effect.
Why does the analysis restrict its comparison to a disease-matched background?
Because patients with chronic graft-versus-host disease generate a large volume of reports describing their own disease. Comparing a new drug used in that population against the whole database would rediscover the indication and label it as toxicity. Restricting the comparator to reports from the same disease and transplant context removes most of that confounding, which is why terms such as graft-versus-host disease of the lung, eye or skin could then be excluded as indication synonyms rather than counted as signals.
What does a three-tier evidence framework add?
It separates what is known from what is hypothesized from what is confounded. Six signals matched events already documented in labeling or pivotal trials, which confirms the pipeline works. Six unlisted events had plausible mechanistic links to ROCK2 inhibition and deserve prospective follow-up. Thirty-three reflected progressive disease and debilitation. Without that split, a reader has no way to weigh a term like obliterative bronchiolitis against a term like muscle tightness.
Are the viral infection signals caused by belumosudil?
Partly, and the data cannot apportion the shares. ROCK2 inhibition reduces humoral responses and the cohort is on multiple immunosuppressants, including corticosteroids and in many cases mycophenolate mofetil. A monotherapy sensitivity analysis restricted to 1,255 reports with no other suspect or concomitant drug preserved the signal pattern, which argues that co-medication alone does not explain it. What remains unresolved is how much belongs to the drug and how much to the underlying immune dysregulation.
Why does immunoglobulin G appear as a safety signal?
Because humoral suppression is a plausible consequence of the mechanism and because recurrent infection in this population prompts immunoglobulin testing. A reported decrease at 5.75 is not a diagnosis of hypogammaglobulinemia, and spontaneous reports carry no quantitative levels. The practical use is as a prompt: a patient with recurrent sinopulmonary infection on a ROCK2 inhibitor plus corticosteroids plus an antimetabolite should have quantitative immunoglobulin measured rather than assumed normal.
Which patients appear most vulnerable in the subgroup analysis?
Three profiles emerged. Fatal reports were enriched for obliterative bronchiolitis and muscle atrophy. Patients aged 65 and older showed more joint stiffness and lip pain. Females showed more transaminase elevation, skin atrophy and oral mucosal blistering. The age and sex differences reached conventional significance on exact testing but rest on modest subgroup sizes, so they should guide attention rather than determine a different monitoring schedule.
How should muscle tightness be distinguished from disease progression?
By trajectory and distribution. Chronic graft-versus-host disease causes sclerotic skin and fascial involvement that restricts movement, and a ROCK2 inhibitor may add connective tissue stiffness on top of it. New tightness that appears after a dose change, spreads beyond established disease territories, or produces measurable loss of joint range supports a drug contribution. Stable sclerosis in the same distribution as before treatment is more likely disease. Serial photography and range-of-motion measurement settle what memory cannot.
Does the absence of new organ toxicity mean the drug is safe?
It means no new organ-level signal emerged from this dataset after refinement, which is a narrower and more defensible claim. Spontaneous reporting detects what clinicians notice and choose to report, so a slow or asymptomatic toxicity can remain invisible for years. The correct reading is that the observed profile is consistent with the mechanism and that monitoring should follow the mechanism, not that surveillance can be relaxed.
How does mycophenolate mofetil complicate attribution?
It contributes independently to the viral and hematologic events that this analysis attributes to the ROCK2 inhibitor, and its exposure is unmeasured because mycophenolate has no routine therapeutic drug monitoring. When viral gastroenteritis or falling immunoglobulin appears in a patient on both drugs, there is no dataset here that says which one to reduce. Clinical judgment, infection severity and graft-versus-host disease activity have to settle it, and the reporting data can only flag that the question exists.
What monitoring schedule follows from these signals?
Baseline and periodic transaminases, with a check after any dose change. Spirometry at scheduled intervals for anyone with pulmonary involvement, because obliterative bronchiolitis progresses without symptoms. Quantitative immunoglobulin when sinopulmonary infection recurs. Viral testing during new gastrointestinal or respiratory symptoms. Joint range of motion and skin assessment at every visit, documented rather than remembered. None of it is exotic, and all of it is easy to skip when a patient is doing reasonably well.
Is the safety profile different in monotherapy?
The signal pattern persisted when the analysis was restricted to 1,255 reports naming no other suspect, concomitant or interacting medication, with baseline covariates balanced at standardized mean differences below 0.10 for nearly every variable. That is reassuring for attribution but not decisive, because reports describe what was recorded rather than what was taken, and patients on monotherapy may differ in disease severity from those on triple therapy in ways the analysis cannot fully adjust for.