Safety and efficacy of atorvastatin for rebleeding in cerebral cavernous malformations (AT CASH EPOC): a phase 1/2a, randomised placebo-controlled trial

Abstract

Background.

Cerebral cavernous malformations (CCMs) carry a high risk of rebleeding after symptomatic haemorrhage, with serious clinical sequelae. Atorvastatin was shown to prevent CCM growth and bleeding in animal models. We aimed to assess the safety and efficacy of atorvastatin on rebleeding in patients with CCM after a symptomatic haemorrhage.

Methods.

We did a phase 1/2a randomised trial at the University of Chicago’s CCM Center of Excellence. Patients aged 18–80 years with untreated CCMs who had experienced symptomatic bleeding from a CCM lesion within the previous year were eligible. We randomly allocated patients (1:1) to oral atorvastatin (80 mg daily for 2 years) or matching placebo. The primary efficacy outcome was mean percent change in lesional iron deposition per year, measured by quantitative susceptibility mapping (QSM) on MRI and averaged over 2 years; a decrease would signal potential benefit and an increase a safety concern. The primary efficacy outcome was analysed by modified intention to treat, including patients with at least one annual paired QSM assessment. Safety outcomes included rates of bleeds and serious adverse events necessitating drug discontinuation. This trial is registered at ClinicalTrials.gov, NCT02603328, and is completed.

Findings.

Between July 25, 2018, and July 22, 2022, 326 patients were assessed for eligibility, and 80 patients were allocated either atorvastatin (n=41) or placebo (n=39). 29 patients were male and 51 were female. 64 patients (33 atorvastatin and 31 placebo) had at least one annual paired QSM assessment and were included in the modified intention-to-treat analyses. The mean annual percent change in lesional QSM was 10·88 (SE 7·29) with atorvastatin versus 12·09 (7·54) with placebo (treatment effect –1·22, 95% CI –22·25 to 19·81; p=0·91). Symptomatic haemorrhage was reported in six patients assigned atorvastatin and seven patients assigned placebo (relative risk 0·81, 95% CI 0·31 to 2·13). No patients had a serious adverse event requiring drug discontinuation and no deaths were recorded.

Interpretation.

For people with symptomatic haemorrhage caused by CCM, atorvastatin did not affect the mean change in lesional iron deposition on brain MRI over 2 years when compared with placebo. Atorvastatin was well tolerated and no safety concerns were noted. The study provides a useful framework for biomarker driven drug assessment in a rare disease.

Funding.

US National Institutes of Health.

INTRODUCTION

Cerebral cavernous malformations (CCMs), also known as cavernous angiomas, are capillary-venous anomalies with dilated vascular spaces that affect about 1% of the population and are prone to repetitive bleeding.¹ CCMs manifesting symptomatic haemorrhage (ie, new bleeding on imaging studies and clinically attributable symptoms)² are rare, affecting fewer than 200 000 individuals in the USA, but they are much more likely to bleed again and to cause serious clinical sequelae than are CCMs without previous symptomatic haemorrhage. There is currently no treatment to prevent recurrent CCM bleeding, other than surgical resection or ablation, which could carry serious complications.^3,4 CCMs with symptomatic haemorrhage have, thus, been targeted for the development of novel therapies aimed at preventing rebleeding.^1,5

Central to molecular mechanisms driving CCM development and bleeding¹ is Rho kinase (ROCK) activation in endothelial cells, mediated by gene aberrations causing CCM genesis.⁶ Sporadic and familial CCMs manifest endothelial ROCK activity not present in normal endothelium, and the inhibition of ROCK has been shown experimentally to restore endothelial integrity which is disrupted in CCM.⁷ Atorvastatin at an oral daily dose of 80 mg causes ROCK inhibition in patients with atherosclerosis,⁸ and equivalent doses in mice inhibited lesion growth and bleeding in CCM models, as did specific ROCK inhibitors.^9,10 Because atorvastatin is widely used and well tolerated clinically, its repurposing for the CCM indication is attractive. Yet, patients and clinicians continue to question whether atorvastatin is safe in patients after CCM bleeding, including concerns about pleiotropic effects;^11,12 a systematic review shows there is equipoise about its clinical effectiveness (appendix pp 4, 17–19).

Designing a clinical trial in CCMs is challenging because of the number of participants that would be needed in such a rare disease to show an effect on symptomatic hemorrhage rates.⁵ Lesional iron deposition (measured by quantitative susceptibility mapping [QSM] on MRI) and vascular permeability (measured by dynamic contrast enhanced quantitative perfusion [DCEQP]) have been shown to reflect new bleeding in previously stable CCMs.¹³ A prospective multisite trial readiness study of CCMs with symptomatic haemorrhage in the previous year showed that two categorical threshold biomarker events (≥6% yearly increase in mean lesional QSM and ≥40% yearly increase in mean lesional DCEQP) were specific and more frequent than the annual recurrent symptomatic haemorrhages.¹⁴ Therefore, a drug effect on these imaging biomarkers can be observed with fewer patients than would be needed for an effect on the rate of symptomatic events.¹ Mean lesional QSM change per year has been accepted as a surrogate measure of potential drug effect on CCM bleeding by the US Food and Drug Administration (Drug Development Tool Biomarker Qualification Letter of Intent 000127).

In view of the compelling biological rationale, clinical need, equipoise about the safety and potential efficacy of a widely available drug, and validated biomarkers with mechanistic plausibility and demonstrated feasibility, we designed an early-phase proof-of-concept trial, called Atorvastatin Treatment of Cavernous Angiomas with Symptomatic Hemorrhage Exploratory Proof of Concept (AT CASH EPOC). With this trial, we aimed to investigate whether oral atorvastatin (80 mg daily for 2 years) might produce a difference compared with placebo in lesional iron deposition (as assessed by QSM) in CCMs with a documented symptomatic haemorrhage in the preceding year.⁵ An increase in QSM change would signal a harm concern with the drug, and a decrease would signal potential benefit.

METHODS

Study design

AT CASH EPOC was an investigator-initiated, single-centre, phase 1/2a, randomised, placebo-controlled, double-blinded, two-arm parallel assignment clinical trial conducted at the University of Chicago’s CCM Center of Excellence. Ethics approval was granted by the University of Chicago Medicine Institutional Review Board (number 18–0445). Trial safety and data quality were monitored by the institutional review board, the Brain Injury Outcomes Section Clinical Trial Coordinating Center (BIOS CTCC) at Johns Hopkins Medical Institutions, and an independent medical safety monitor (appendix p 5). Data management was performed by BIOS CTCC. The trial complied with US regulations and International Council for Harmonisation Good Clinical Practice guidelines. This trial is registered with ClinicalTrials.gov, NCT02603328.

Trial participants

Adults aged 18–80 years with untreated solitary or familial CCMs, who had experienced an adjudicated symptomatic haemorrhage from a CCM lesion within 1 year of trial enrolment, were eligible for the study. The symptomatic haemorrhage lesion must have been not resected or otherwise irradiated or ablated. Exclusion criteria included prior cranial irradiation or radiosurgery and any statin use within the past year. A full list of inclusion and exclusion criteria is provided in the appendix (p 20). All participants gave written informed consent at the initial screening visit.

Enrolled patients were followed for two years from randomization per intention to treat, with clinical assessments and MRI evaluations at baseline, 365 and 730 days +/− 30 days.

Randomization and masking, allocation concealment

Randomization to active drug or placebo (1:1) was performed at BIOS CTCC using a block algorithm, with stratification by sex. Allocation concealment was achieved by the use of unlabeled indistinguishable capsules. Investigators, clinical staff, and participants were masked to the assigned treatment. The randomization algorithm, treatment assignment process, and drug discontinuation criteria have been previously published,⁵ and are detailed in the appendix (pp 4–5).

Procedures

Participants received either oral atorvastatin (80 mg/day) or matching placebo, to be taken once daily at a time of the patient’s choosing. Drug or placebo were continued for 2 years or until a symptomatic hemorrhage or another safety event requiring drug discontinuation occurred.

Clinical, laboratory, and MRI evaluations were conducted at baseline, 12 months, and 24 months after randomisation; laboratory studies were also performed 3 months after dose initiation. Functional status was assessed at each clinical visit with the Mini-Mental State Examination (MMSE), the National Institutes of Health Stroke Scale (NIHSS) score, and the modified Rankin scale (mRS) score. Quality of life was measured with Euro-QoL-5D, Euro-QoL visual analog scale (VAS), and PROMIS-29 version 2.0. MRI was performed using a 3T scanner with an eight-channel head coil. Details of imaging protocols have been published previously,^{13, 14} and are summarized in the appendix (p 6). Participants traveling more than 200 miles to the study centre for study evaluations received travel stipends.

Participants were contacted by phone or electronic mail every 3 months to monitor drug compliance, adverse events, and to assess mRS. Drug compliance was also tracked using a mobile phone application that was completed by participants and monitored by the study team.

Outcomes

The primary outcome for both safety and efficacy was the percent change in mean lesional QSM (change score) per year, as assessed on MRI. The MRI analysis was conducted on the index CCM lesion with qualifying symptomatic haemorrhage in the year before study enrolment. Secondary outcomes for potential efficacy were changes in vascular permeability, as measured by DCEQP in the index lesion and in brain white matter far from lesion, and the proportion with QSM increase of 6% or greater, and with DCEQP increase of40% or greater. Secondary outcomes for safety were rates of symptomatic haemorrhage (as per adjudicated criteria),² asymptomatic bleeding (ie, lesion expansion, defined as an increase in maximum lesion diameter on T2-weighted sequences of ≥3 mm or subclinical bleeding detected by MRI without attributable symptoms),^{14, 15} and any serious adverse events (SAEs) necessitating drug discontinuation. Other secondary outcomes were drug compliance (with a target of 90% or greater for protocol compliance, counted as number of days taking the drug per number of days in the study), changes in functional outcome measures (MMSE, mRS, and quality of life [Euro-QoL-5D, Euro-QoL VAS, and PROMIS-29 version 2.0]), and ROCK activity in peripheral blood leukocytes at each follow-up visit. For this report, we only present analyses of mRS scores and EuroQOL visual analog scales, with plans to analyze the more extensive multiple domains of functional outcomes in a subsequent publication. Exploratory safety outcomes were SAEs and other adverse events (AEs). We also analyzed non fasting cholesterol levels (a known biologic effect of atorvastatin) and vitamin D levels since these could impact CCM hemorrhage risk. We prespecified subgroup analyses of the primary outcome by sex, lesion location, and familial versus sporadic disease.

Safety assessment were pre-specified to take place at the end of the first year, after 30 participants have completed 12 months of follow-up and again after 60 subjects have completed 12 months of follow-up (appendix pp 5–6). A protocol provision also called to preemptively suspend trial enrollment for a full safety review if at any time, >40% of subjects in the treatment group suffered a symptomatic brain haemorrhage from the index CCM lesion or another source.

Statistical analysis

The trial was powered to detect a 20% relative difference in the percent change of mean lesional QSM per year (two-tailed, power 0·9, alpha 0·05), with a sample size of 50 (which was expanded to 80 to account for missing QSM data or patient attrition). The 20% effect size was proposed as a minimum to be clinically meaningful and mechanistically plausible (appendix pp 6–7). The initial sample size was calculated based on a pilot study.⁵ A futility analysis and sample size recalculation was conducted when 50 paired biomarker assessments were successfully completed (ie, half the 100 assessments that were initially projected to be needed for testing the primary hypothesis). Futility criteria were not met and the initial sample size was endorsed (appendix pp 8–9).

The primary outcome analysis was performed as a time-averaged difference between two arms, using a repeated measures analysis implemented as a sex-adjusted linear mixed model, averaged over 2 years (appendix p 6).¹⁶ Secondary analyses were performed on QSM change during each year of follow-up. The same approach was used for analysis of lesional DCEQP change in the two arms of the study. Both the QSM change (primary outcome) and lesional vascular permeability (DCEQP) change were analyzed in the modified intention-to-treat (ITT) sample of patients who had paired biomarker assessments at the beginning and end of at least one annual epoch of follow-up. Safety outcomes (ie, bleeding rates and AEs) and other secondary outcomes were evaluated in the overall trial cohort by ITT. The time course of symptomatic haemorrhage was evaluated using Kaplan-Meier survival analysis, censoring cases who were lost to follow-up. An adjusted Poisson linear mixed model controlling for sex was used to evaluate the difference in AE rates between the two treatment arms.

We compared post-hoc the features of participants who contributed at least one QSM paired annual data (the modified ITT cohort) to those who did not. Because a few people with symptomatic haemorrhage did not contribute paired QSM assessments (ie, they either underwent surgery or declined follow-up), and these patients typically have a greater QSM change than patients without a symptomatic haemorrhage, we implemented a secondary analysis with imputation of missing data for mean annual QSM change. The imputation approach was proposed as biologically and clinically relevant, after trial readiness analyses of the QSM biomarker in a similar CCM cohort.¹⁴ Details of imputation method and analyses are reported in the appendix (pp 9 and 25).

Point estimates and 95% CIs are reported for each analysis, with p values for the primary outcome. Continuous endpoints are presented by treatment arms as mean (SD) or median (Q1–3 range). Categorical endpoints are summarized by frequency in each category. Continuous variables are analyzed using either the Student’s t-test for normally distributed data or the Mann-Whitney U test for non-normally distributed data. Categorical variables are assessed using the Chi-square test or Fisher’s exact test, when appropriate. For the primary outcome, biased effects in prespecified subgroups by sex, lesion location, and solitary versus familial disease are queried using Chi-square test. There was no allowance for multiplicity of variables queried. Statistical analyses were done with STATA-SE version·18.0.

Role of the funding source

The funding source had no involvement in the study design; collection, analysis, or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication.

RESULTS

Randomization and follow-up

Between July 25, 2018, and July 22, 2022, 326 patients were assessed for eligibility, and 80 patients were randomized to either atorvastatin (n=41) or placebo (n=39) and were included in intention-to-treat analyses (figure 1). In year one of the study, five patients had a symptomatic haemorrhage, including one that occurred after randomization but before starting the study (the patient was allowed to remain in the study). Further, one patient became pregnant and stopped the study drug, one patient participated in remote clinical follow-up without blood draws or imaging, and 12 patients withdrew from the study or contributed no follow-up. In year two of the study, eight subjects had a symptomatic haemorrhage, one became pregnant, one withdrew from the study and underwent lesion resection, and four participated in remote clinical follow-up without blood draws or imaging. All these patients were excluded from the modified ITT analyses because they did not have paired QSM measurements. Therefore, 64 patients were included in modified ITT analyses, 33 who were assigned atorvastatin and 31 who were assigned placebo (figure 1).

Figure 1. Screening and follow-up diagram of study subjects.

A total of 326 subjects with CCMs were screened for eligibility of which 202 did not have a symptomatic haemorrhage in the previous year from the enrollment date and 44 met other exclusion criteria. Of the 80 randomized subjects, 41 received active trial drug (atorvastatin) and 39 placebo, and these comprised the ITT population. Sixty four patients contributed at least one paired annual QSM assessment, and comprised the modified ITT population (all met ≥90% active drug compliance). Safety endpoints, remote follow-ups, and patients lost to follow-up or withdrawn from the study are listed per epoch year in each assigned treatment. *One subject had a symptomatic haemorrhage before starting the study drug (placebo). †One subject had a symptomatic haemorrhage in year two after stopping the drug (atorvastatin) following a prior symptomatic hemorrhage in year one. SH=symptomatic haemorrhage. QSM=quantitative susceptibility mapping. CCM=cerebral cavernous malformation. mRS=modified Rankin score.

Of the 80 patients enrolled in the trial, 54 (68%) were referred from further than 200 miles away from the study site. The median age of participants was 41 years (IQR 34·25 to 51·50), 29 (64%) were female, and 21 (26%) were Hispanic or African American. In 44 (55%) participants, the symptomatic haemorrhage lesion was in a brainstem location; 51 (64%) had sporadic/solitary lesion and 29 (36%) had familial/multifocal CCM disease. Baseline demographic and clinical characteristics were well balanced between arms (table 1 and appendix p 23–24), except for mRS score, which appeared to be higher in patients assigned atorvastatin compared with placebo. Use of vitamin D, sex hormones, and propranolol at baseline did not differ between groups (appendix p 23).

Table 1.

Baseline demographic and clinical characteristics of trial patients per ITT assignment

Variable	Total N=80	Atorvastatin N=41	Placebo N=39
Age – year*	41 (34·25–51·50)	39 (34–54)	41 (34–49)
Sex – number of cases (%)
Male	29 (36%)	15 (37%)	14 (36%)
Female	51 (64%)	26 (63%)	25 (64%)
Ethnicity – number of cases (%)†
Hispanic or Latino	13 (16%)	6 (15%)	7 (18%)
Not Hispanic or Latino	65 (81%)	34 (83%)	31 (79%)
Unknown	2 (3%)	1 (2%)	1 (3%)
Race – number of cases (%)†
Asian	2 (3%)	1 (2%)	1 (3%)
Black or African American	8 (10%)	6 (15%)	2 (5%)
White	63 (79%)	32 (78%)	31 (79%)
Other	5 (6%)	2 (5%)	3 (8%)
Unknown	2 (3%)	0 (0%)	2 (5%)
Genotype – number of cases (%)
Sporadic/Solitary	51 (64%)	23 (56%)	28 (72%)
Familial/Multifocal	29 (36%)	18 (44%)	11 (28%)
MRI characteristics *
number of lesions on SWI in familial cases	1 (1–9)	1 (1–13)	1 (1–6)
number of lesions on T2 ≥ 4mm in familial cases	1 (1–3)	1 (1–5)	1 (1–1)
Size on T2 (mm)	14·80 (9·75–19·80)	14·70 (9·65–18)	16 (10–23·70)
Location of index CCM lesion with SH – number of cases (%)
Brainstem	44 (55%)	23 (56%)	21 (54%)
Cerebellum	4 (5%)	2 (5%)	2 (5%)
Frontal lobe	5 (6%)	3 (7%)	2 (5%)
Occipital lobe	4 (5%)	1 (2%)	3 (8%)
Parietal lobe	3 (4%)	1 (2%)	2 (5%)
Temporal lobe	8 (10%)	2 (5%)	6 (15%)
Thalamus	7 (9%)	5 (12%)	2 (5%)
Other location	5 (6%)	4 (10%)	1 (3%)
Time from most recent SH to enrollment (days)*	104 (57–150)	103 (51–171)	104 (69–137)
Number of SH prior to enrollment *	1 (1–2)	1 (1–2·50)	1 (1–2)
mRS score – number of patients (%)
0	14 (18%)	6 (15%)	8 (21%)
1	43 (54%)	18 (44%)	25 (64%)
2–3	23 (29%)	17 (42%)	6 (15%)
4	0 (0%)	0 (0%)	0 (0%)
5–6	0 (0%)	0 (0%)	0 (0%)
Euro-QoL Visual Analog Scale	76·4 ± 14·7	74·59 ± 16·29	78·28 ± 12·67

^*Median (interquartile range).

^†Self-designated

SWI=susceptibility weighted imaging. SH=symptomatic haemorrhage. mRS=Modified Rankin Scale. Euro-QoL= European quality of life index

Median follow-up of participants from trial enrollment until safety endpoint, withdrawal from the study or final follow-up visit was 723 days (IQR 695 to 746). The last follow-up visit was logged on July 10, 2024. Overall, 51 patients completed imaging biomarker acquisitions during two annual epochs (greater than the targeted number per sample size calculations to test the primary hypothesis), and 13 contributed a single annual epoch of paired biomarker acquisitions. These 64 patients were included in the modified ITT analysis. All but four trial participants had greater than 90% drug compliance. Four patients who did not meet this compliance threshold did not contribute to the biomarker assessments; one had symptomatic haemorrhage and three withdrew from the study in the first year (appendix p 10). One of these four patients had requested a dose reduction to 40 mg/day or placebo (while maintaining blinded treatment assignment) before withdrawing from the study. All other patients in the trial, including the cases who contributed paired QSM assessments, continued per their assigned treatment and original dose with >90% compliance. Hence, there was no need for a separate analysis of the primary outcome per treatment received. No demographic or clinical differences were noted between patients who contributed paired QSM assessments and those who did not, except for a greater likelihood that males and active smokers would not attend follow-up (appendix pp 23–24).

No difference was recorded between atorvastatin and placebo in the primary outcome of annual percent change in lesional QSM (10·88 [SE 7·29] vs 12·09 [7·54]; treatment effect –1·22, 95% CI –22·25 to 19·81; p=0·91; table 2). Further, there was no difference in mean lesional QSM change in years 1 or 2 (figure 2), or when analyses considered relative or absolute changes (table 2). Prespecified analyses of the QSM outcome in which missing QSM data were imputed also revealed no difference between atorvastatin and placebo (appendix p 25). Further, no differences in the primary outcome were noted in prespecified subgroups (appendix pp 30–31).

Table 2.

Mean annual QSM change in cerebral cavernous malformation lesion with symptomatic haemorrhage per assigned treatment to atorvastatin and placebo (modified ITT cohort)

	Atorvastatin N=33 Point estimate [S.E.]	Placebo N=31 Point estimate [S.E.]	Treatment effect (Atorvastatin minus Placebo) [95% C.I.]
QSM % change averaged over both epochs	10·88 [7·29]	12·09 [7·54]	−1·22 [−22·25, 19·81] −10%; p=0·91*
QSM % change by assigned treatment; year one minus baseline	7·35 [9·77]	0·24 [10·08]	7·12 [−20·38, 34·62]
QSM % change by assigned treatment; year two minus year one	15·18 [10·80]	26·80 [11·22]	−11·62 [−42·14, 18·91]
Absolute QSM change by assigned treatment; year one minus baseline	0·00 [0·04]	−0·06 [0·04]	0·06 [−0·05, 0·18]
Absolute QSM change by assigned treatment; year two minus year one	0·06 [0·04]	0·08 [0·05]	−0·02 [−0·15, 0·11]

All analyses are based on mixed model adjusted for sex

QSM=quantitative susceptibility mapping. CI=confidence interval.

The treatment effect represents the difference in mean lesional QSM percent or absolute change in reference to placebo (negative reflects smaller change than placebo, and positive reflects greater change than placebo). None of the differences were statistically significant.

^*Relative effect and p value are presented for the pre-specified primary outcome.

Figure 2. Mean lesional QSM percent change during the first and second follow-up years per assigned treatment in the modified ITT cohort.

Five out of six symptomatic haemorrhages with paired QSM imaging had a QSM percent change larger than 6%. One statistical outlier value (> 2 SD) was identified in the placebo arm in each epoch, the first of which had an asymptomatic change identified at year one clinical MRI imaging, and subsequently suffered an symptomatic haemorrhage in year two. Solid and hollow dots represent cases under atorvastatin and placebo arms, respectively. Cases with symptomatic haemorrhages and asymptomatic changes are colored in red and blue, respectively. QSM=quantitative susceptibility mapping.

No difference was recorded between atorvastatin and placebo in the secondary efficacy outcome of annual percent change in lesional DCEQP (108·87 [SE 40·64] vs 59·09 [42·19]; treatment effect 49·77, 95% CI –71·41 to 170·96; appendix p 26). Further, no difference was recorded in mean lesional DCEQP change in years one or two (appendix p 11), or when the analyses considered relative or absolute changes (appendix p 26). Vascular permeability of brain white matter far from lesions did not differ between groups (appendix pp 15–16). No differences in functional status (mRS score) or quality of life (Euro-QoL VAS) were noted at year one or two between atorvastatin and placebo (appendix p 27).

Prescribed safety reviews were performed as planned (appendix p 5), and raised no concerns during the course of the trial. Total bleed rates did not reach greater than 40% at any time which would have suspended the trial, and no deaths were reported. Six patients assigned atorvastatin and seven assigned placebo had a symptomatic haemorrhage event, all of which occurred in the index CCM lesion (table 3). The time to symptomatic haemorrhage did not differ between patients allocated atorvastatin or placebo (figure 3). Further, no differences were recorded between groups in rates of subclinical bleeds, assessed as asymptomatic change on MRI, mean lesional QSM change ≥6% or mean lesional DCEQP ≥40% (table 3). No SAEs necessitated discontinuation of treatment. Two SAEs were reported in patients receiving atorvastatin. One patient was an 18-year-old male who had elevated creatine kinase on laboratory tests, without overt symptoms. Per study protocol, he was given the option of stopping the drug temporarily then resuming a lower dose, but he decided to discontinue the drug. Another patient, a 59-year-old female, was hospitalized after a fall, adjudicated as having a questionable relationship to the study drug. She continued taking the drug without further events and completed the trial. AEs were reported by 27 patients on atorvastatin and 21 on placebo, which were significantly more common in atorvastatin-treated patients by Poisson regression analysis. AEs per organ system class are reported in the appendix (pp 28–29).

Table 3:

Clinical and subclinical bleeds and rate of biomarker changes during the study, according to assigned treatment (intention-to-treat cohort)

	Atorvastatin (n=41)	Placebo (n=39)	Absolute risk difference (95% CI)	Relative risk (95% CI)
Symptomatic haemorrhage, year 1	1(2%)	4 (10%)*	0.08 (−0.07, 0.22)	0.24 (0.04, 1.50)
Symptomatic haemorrhage, year 2	5 (12%)†	3 (8%)	−0.05 (−0.20, 0.12)	1.58 (0.45, 5.72)
Symptomatic haemorrhage total	6 (15%)	7 (18%)	0.03 (−0.15.0.21)	0.81 (0.31 to 2.13)
Asymptomatic change, year 1	3 (7%)	4 (10%)	0.03 (−0.13 to 048)	0.71 (0.19 to 2.69)
Asymptomatic change, year 2	5 (12%)	3 (8%)	−0.05 (−0.20 to 0.12)	1.59 (0.45 to 5.73)
QSM ≥6%, year 1‡	15 (37%)	11(28%)	−0.08 (−0.29 to 014)	1.30 (0.69 to 2.48)
QSM ≥6%, year 2‡	17 (41%)	14 (36%)	−0.06 (−0.27 to 0.17)	1.15 (0.67 to 2.02)
DCEQP ≥40%, year 1‡	15 (37%)	13(33%)	−003(−025 to 019)	1.10 (0.61 to 2.00)
DCEQP ≥40%, year 2‡	10 (24%)	7 (18%)	−0.06 (−0.25 to 0.13)	1.36 (0.59 to 3.17)
Symptomatic haemorrhage, asymptomatic change, QSM ≥6% or DCEQP ≥40%, year 1	24 (59%)	22 (56%)	−0.02 (−0.24 to 0.20)	1.04 (0.71 to 1.53)
Symptomatic haemorrhage, asymptomatic change, QSM ≥6% or DCEQP ≥40%, year 2	24 (59%)	20 (51%)	−0.07 (−0.29 to 0.16)	1.14 (0.77 to 1.73)

Data are n (%), unless otherwise stated. QSM used to measure mean lesional iron content. DCEQP used to measure mean lesional vascular permeability. QSM=quantitative susceptibility mapping. DCEQP=dynamic contrast quantitative perfusion.

^*One patient had symptomatic haemorrhage after being randomly assigned to the placebo group but before starting the drug.

^†One patient in the atorvastatin group who stopped the drug due to a symptomatic haemorrhage in year 1 and had a recurrent symptomatic haemorrhage from the same index lesion in year 2.

^‡Cases with QSM or DCEQP change in index symptomatic lesion.

Figure 3. Kaplan-Meier curve survival estimates showing time to symptomatic haemorrhage event in cases assigned to atorvastatin and placebo (whole trial ITT cohort).

A total of six symptomatic hemorrhage events in five patients occurred in patients assigned to the atorvastatin treatment vs seven symptomatic hemorrhage events in seven patients assigned to the placebo group. One subject in the placebo group had a symptomatic haemorrhage after enrollment but before starting to take the drug, with the bleed counted as day zero per intention-to-treat. Another subject in the atorvastatin group had a symptomatic haemorrhage identified at year one clinic visit and stopped the drug; he later suffered a recurrent symptomatic haemorrhage from the same index lesion two months later and this was counted in the atorvastatin group per intention treat. There were seven censored observations due to patients lost to follow-up, noted as vertical dash, with the censored date as the last date of logged follow-up. Years one and two follow-up visits were planned at 360 and 730 days +/− 30 days, respectively, from the date of enrollment (subjects completed the study as early as 700 days post-enrollment, hence fewer numbers of subjects at risk after 730 days). The log-rank test revealed non-significant differences between treatment arms in the rate and time to symptomatic haemorrhage.

Total cholesterol levels were significantly lower in patients receiving atorvastatin at years 1 and 2 of the study (appendix p 12). Vitamin D levels were not different in atorvastatin and placebo patients at years 1 and 2 (appendix p 22). Peripheral leukocyte ROCK activity and its change from baseline in individual subjects in years 1 and 2 did not differ between groups (appendix pp 13–14).

DISCUSSION

The trial provides new evidence about the safety of atorvastatin in CCM lesions with recent symptomatic haemorrhage. There was no signal in biomarker effects, symptomatic haemorrhage rates, subclinical bleeding rates, or any other suggestion of increased bleeding risk or SAEs requiring drug discontinuation with atorvastatin as compared to placebo. This is a relevant clinical result addressing a knowledge gap about the risk of atorvastatin in CCM patients, even after a recent bleed, and allays concerns previously raised in this regard.^11,12 AEs did not affect a significantly greater fraction of participants taking atorvastatin than placebo, but the AEs per patient were significantly more common with the drug. These were mostly mild or non-specific symptoms. Two SAEs not requiring permanent drug discontinuation per protocol occurred in the atorvastatin group, and one patient withdrew from the trial after this event. The rate of AEs in this study was greater than in clinical trials of atherosclerotic patients, where these occurred almost equally with atorvastatin and placebo.¹⁸ The more prevalent symptoms may represent a greater sensitivity of younger CCM subjects to the drug. As with other large clinical trials of atherosclerotic patients, AEs in our study had no lasting morbidity.

The trial did not endorse a hypothesized benefit in lesional bleeding, as had been demonstrated in animal studies. Studies in animal models showed decreased CCM lesion development as well as bleeding with atorvastatin,^{9, 10} but did not assess recurrent bleeding after a prior haemorrhage per se, and hence may not have simulated recurrent bleeding in humans. Our trial did not examine a time course long enough to query lesion development as in animal models. The lower rate of symptomatic haemorrhage in patients receiving atorvastatin in year one of our study (a fourth the rate in patients receiving atorvastatin than placebo) would have suggested a benefit based on the 80% confidence interval method (80% CI 0·06–0·87)¹⁹ advocated in another exploratory CCM trial of propranolol.¹⁷ Yet we had cautioned about the credulity of over interpreting such results in the absence of sufficient statistical power.²⁰ With small event rates, a single symptomatic hemorrhage or its slightly different timing would eliminate the apparent difference, and such differences are virtually never replicated in studies with larger sample sizes. Analysis of the symptomatic haemorrhage-free survival curves, effect on biomarkers, and rates of subclinical bleeding did not endorse such benefit. It is possible that a bleeding benefit was merely too weak (a point estimate about half of the 20% postulated effect), but the clinical benefits of such weak effect would need further investigation. We also cannot exclude a potential synergistic benefit of atorvastatin in combination with other therapies. There was a non-significant reduction in the primary outcome of males but not females, symptomatic haemorrhage lesions at brainstem location, and sporadic/solitary lesions. These observations are merely hypothesis generating and we note that the trial was not powered to detect significance in each of these subgroups. Future trials might consider stratifying for these subgroups and adaptive designs to drop subgroups without benefit.

The lack of postulated benefit on recurrent bleeding is consistent with the evidence from prior cohort studies, suggesting a neutral effect (appendix pp 17–19). There may be several reasons for this lack of benefit of atorvastatin. It has been suggested that the benefit of inhibition of ROCK activity by statins may be countered by pleiotropic depletion of other prenylation-dependent cellular processes that could increase bleeding.¹² In other research, lower cholesterol levels, as achieved with our patients on atorvastatin, were associated with greater clinical aggressiveness of CCM, including greater hemorrhagic events.²¹ More targeted ROCK inhibitors under current development may achieve a greater benefit in the absence of such pleiotropic effects.²²

The non-significant decrease in the point estimate of the primary outcome with atorvastatin in males, but not females, is consistent with observations in other studies which implicated sex-related differences in endothelial inflammation with the drug and other postulated mechanisms.^{23, 24} With more females enrolled in the study than males, it is possible that the fewer males who enrolled in the trial had more aggressive lesions, and the more aggressive lesions could have been more responsive to the drug. A similar pre-selection bias of more aggressive lesions could have explained the primary outcome point estimate decrease in lesions at the brainstem but no other locations, as patients with CCM lesions with symptomatic haemorrhage at non-brainstem locations may have more likely opted for resection of their lesion rather than enroll in the trial. This would have pre-selected for surgery (and out of the trial) lesions at more operable non-brainstem locations, while more aggressive lesions at the brainstem (where surgery is more prohibitive) were more likely to enroll. The decrease in primary outcome point estimate in sporadic, but not familial cases, could also have been biased by the greater prevalence of males among these cases.

Most of our trial subjects took Vitamin D supplements preventing low systemic levels, and avoided sex hormone therapy. These measures have been suggested to potentially prevent CCM bleeding.^{21, 25, 26} Lower rates of symptomatic haemorrhage in our trial than reported in prior studies¹⁵ could imply that the medical care of our trial subjects may have diluted potential added benefits of atorvastatin. If true, this would imply that simpler medical measures per good current clinical practice may by themselves lessen CCM rebleeding, and novel experimental therapies would need to do better.

The lack of effect on lesional vascular permeability by DCEQP is not unexpected, in view of its predicted poor performance in recent trial simulations,¹⁴ and we observed even greater variances of these measurements in this trial. Despite excellent compliance and demonstrated lower cholesterol levels, atorvastatin did not inhibit peripheral leukocyte ROCK activity as had been demonstrated in atherosclerotic patients,⁸ even as we used the same ROCK activity assay in our trial subjects, conducted by the same principal investigator of those earlier studies. It is possible that older atherosclerotic patients have a higher baseline ROCK activity in peripheral leukocytes, or are more prone to its inhibition by atorvastatin, than mostly non-atherosclerotic and typically younger CCM subjects.

We could not replicate higher baseline permeability in the brain white matter of familial CCM patients with systemic haploinsufficiency of genes expected to increase vascular permeability, and previously observed in cohort studies.²⁷ However, a statistically non-significant reduction of individual patient change in brain white matter vascular permeability from baseline in years one and two of the study was demonstrated in atorvastatin subjects as compared to placebo. This is consistent with mild ROCK inhibition by the drug in brain vasculature. ROCK activity within resected lesions receiving placebo or atorvastatin would have been interesting to explore, but there were too few resected lesions in study subjects to test this hypothesis.

Our study was based on extensive mechanistic and preclinical investigations and benefited from prior characterization of the clinical features of trial subjects, event rate estimations, and biomarker validations in a trial readiness project.^14,15 Notwithstanding the negative results with this particular drug, the study provides a conceptual framework for deploying a surrogate biomarker to increase the sensitivity of detecting therapeutic effects in a rare disease. The absence of effect on biomarkers also adds confidence endorsing the lack of difference in clinical event rates. This approach can also be applied in comparing different doses and drug effects in platform trials.

Limitations

This proof-of-concept trial enrolled patients with symptomatic haemorrhage in the prior year. A significant treatment effect would not likely have been observed in cases with longer intervals since the most recent symptomatic haemorrhage or with non-symptomatic haemorrhage symptoms, as these patients would have had likely lower rebleed rates. Our study was conducted at a single site, in the absence of multisite validations of imaging biomarkers at the launch of the trial. We instituted a nationwide outreach through patient support organizations, and provided a stipend to more than two thirds of patients traveling for enrollment. More recently, the biomarkers deployed have been validated as reflecting CCM bleeding in multisite studies,¹⁴ so future trials using them can be performed at multiple sites. It is unclear if the results would have been different in a multisite trial and different patient referral patterns. The sample size was small, limiting statistical power of many of the secondary and exploratory outcomes. We assessed treatment effect over two years, and the persistent bleed risks, in the second year of our study and in recent trial readiness results, justifies testing over that period.^14,15 We cannot speculate if a one-year study with near double the number of patients would have yielded different results.

We also note that 64% of enrolled patients were female, and they were more likely to contribute complete biomarker data in the study, but we cannot speculate about any implications of this limitation.

We do note that QSM had never been shown to be affected by a drug, and this will need to be explored with other pharmacotherapies. Novel plasma biomarkers may also reflect CCM hemorrhage.^{28, 29} These may enhance the assessment of drug effects in future trials, as dual-criteria in conjunction with imaging biomarkers.³⁰

Conclusion

In all, our results endorse no safety concerns with atorvastatin in CCM patients after recent symptomatic haemorrhage. However, we cannot recommend its use with the aim of preventing CCM rebleeding.

Table 4:

Serious adverse events and adverse events, according to assigned treatment (intention-to-treat cohort)

	Atorvastatin (n=41)	Placebo (n=39)	Absolute risk difference (95% CI)	Relative risk (95% CI)
Deaths	0	0	0	0
Serious adverse events requiring discontinuation of drug	0	0	0	0
Patients with other serious adverse events	2	0	−0.05 (−0.18 to 0.07)	Not calculable
Patients with adverse events	27	21	−0.12 (−0.33 to 0.11)	1.22 (0.85 to 1.80)
Patient adverse events by Poisson regression*	27	21	0.79 (0.30 to 1.28)†	0.53 (0.35 to 0.80)‡

Data are n, unless otherwise stated. NA-not applicable.

^*Poisson regression considered the number of adverse events as well as the number of patients reporting them.

^†Absolute risk difference was obtained using a generalised linear model with a Poisson distribution and an identity link.

^‡Placebo vs atorvastatin.

Panel.

RESEARCH IN CONTEXT

Evidence before this study

Mechanistic and animal studies have suggested Rho kinase (ROCK) inhibition could be a therapeutic target for cerebral cavernous malformations (CCMs), and atorvastatin at doses achieving pleotropic ROCK inhibition has been shown to decrease lesion growth and bleeding in mouse models of CCM. We searched PubMed on Dec 2, 2024, with the terms (“statin” OR “atorvastatin”) AND (“cavernous malformation” OR “CCM” OR “cavernoma” OR “cavernous angioma”). Human studies indicated clinical equipoise about whether this therapy impacts CCM bleeding rate. And there is concern about potentially increased brain bleeding with statins in general.

Added value of this study

To the best of our knowledge, our study is the first phase 1/2a randomised trial to assess ROCK inhibition with atorvastatin in people with CCM and previous symptomatic haemorrhage. Our study incorporated a validated imaging biomarker of CCM haemorrhage, reflected by a change in lesional iron deposition per year, measured by quantitative susceptibility mapping (QSM) on MRI. A decrease in QSM change by a drug would signal potential benefit, and an increase would signal a safety concern.

Implications of all the available evidence

Our trial demonstrated no effect of atorvastatin compared with placebo on rebleeding in patients with CCM who had suffered a symptomatic haemorrhage in the previous year. No safety concerns were noted. The results do not justify the use of atorvastatin with the aim of preventing CCM rebleeding. Drugs with stronger and more specific ROCK inhibition properties than atorvastatin might be needed to achieve a meaningful benefit.

Data sharing

As required by the National Institute for Neurological Disorders and Stroke for Phase II studies, a complete de-identified dataset containing all variables collected in the trial and a data dictionary will be submitted to https://www.ninds.nih.gov/current-research/research-funded-ninds/clinical-research/archived-clinical-research-datasets for data sharing within a timeframe to be agreed upon with National Institute for Neurological Disorders and Stroke Program Officer, ideally within one year of publication of the primary outcome paper.