Dementia is the commonest known brain disorder with rates increasing as populations age: ¹ world prevalence will treble by 2050 ² and reach 1.7 million people in England & Wales by 2040. ³ Common types comprise Alzheimer’s disease (AD), vascular dementia (VaD) and mixed dementia, mostly combined AD and VaD. ¹ Most vascular cognitive impairment (VCI) and VaD, with or without AD, is caused by cerebral small vessel disease (cSVD), ⁴ a progressive neurovascular-degenerative condition that is typically diagnosed on neuroimaging. ^{5, 6} cSVD increases the future risk of stroke, dementia and functional decline 2-to-3 fold. ⁷ cSVD, including that which is covert, is extremely common and estimated to affect 10% of the world’s population, i.e. approximately 750M worldwide, more so in low-middle income countries. ⁸ cSVD contributes to 45% of dementias. ⁹ Further, cSVD causes lacunar ischaemic stroke (25% of the 17M strokes/year worldwide), ¹⁰ most intracerebral haemorrhages (ICH, 10% of strokes) and underlies many mobility, gait, falls, neurobehavioral, mood/depression and urinary incontinence disorders in older people. ^{1, 4, 9} Importantly, the biggest concern expressed by patients with cSVD is the development of VCI/VaD. ¹¹

Unfortunately, cSVD has no specific proven preventative or restorative interventions and lowering blood pressure only has a very limited effect. ^{12– 14} The only phase-3 trial was stopped early because dual antiplatelet therapy with aspirin and clopidogrel caused an increase in deaths in patients with prior lacunar ischaemic stroke. ¹⁵ This trial also found that blood pressure lowering did not reduce recurrent stroke ¹⁶ or modify cognition. ¹⁷ Recently, a phase-2 feasibility trial after lacunar stroke found that isosorbide mononitrate (ISMN), especially when given with cilostazol, improved cognition and functional outcome and reduced recurrent stroke. ¹⁸

Given the long, expensive path to new drug development, there is increasing focus on repurposing existing drugs. Repurposing may be “defined as researching new indications for already approved drugs or advancing previously studied but unapproved drugs” ¹⁹ and offers the advantages of assessing interventions which already have a wealth of preclinical and clinical data for another indication. Repurposing drugs can be faster, less expensive and risky and carry higher success rates than traditional drug development approaches primarily because researchers can bypass earlier stages of development that establish drug safety. ²⁰ As a result, repurposing reduces research waste.

Hence, an increasingly common approach to identifying interventions for testing in neurodegenerative conditions such as cSVD, Alzheimer’s disease, multiple sclerosis and motor neuron disease are structured systematic approaches that identify candidate treatments on the basis of their known mechanisms of action and existing pre-clinical and clinical studies. ^{12, 21– 24}

To update and expand on our previous mechanistic assessment ¹² with a wider scoping review of medications that might be repurposed as effective treatments for cSVD with a focus on cognitive outcomes.

We conducted a rapid multi-stage scoping review ²⁵ to identify potential interventions for preventing cognitive decline in people with cSVD with the aim of testing these in a phase-3 platform trial. Funding calls for platform trials typically specify key criteria to consider when selecting interventions for testing. We based our review on the criteria set out by the UK NIHR Health Technology Assessment programme for “large, collaborative and ambitious platform studies in areas of strategic importance” (number 23/95, call date 27 July 2023, closing date for outline applications 28 November 2023). ²⁶ Key requirements were: i) “a clear process and Independent Scientific Advisory Group (ISAG) for the identification and prioritisation of candidate interventions”; ii) “evaluation of the clinical (and cost) effectiveness of multiple interventions” (implying that interventions should not be expensive); and iii) “each technology [ intervention] must have sufficient clinical data”, ²⁶ i.e. that there will be a reasonable chance that it would be effective. Further, interventions should have already been tested in a NHS setting; be phase-3 ready; be available to be implemented rapidly if the trial is positive; be immediately useful to service users; treatment length should be commensurate with a platform design; it would be likely that the interventions would be used if supported by the results; and that there should be collective learning, hence the importance of sharing our process of identification and prioritisation of interventions for cSVD, as here. The following processes were designed to identify suitable phase-3-ready interventions that might modify cognition and could be tested in a platform trial and so fit the above call and its requirements.

Since this was performed as a rapid scoping review to be completed within two months, we did not create a prior protocol. Additionally, as a scoping review, we did not register it with PROSPERO.

The intervention selection strategy is summarised in Figure 1. Interventions of potential relevance to cSVD from preclinical and clinical searches were merged to form a long-list. From this, interventions were excluded for several reasons including that they were not available in the UK or could not be delivered feasibly. The resulting medium-list was then scored and ten interventions with the highest scores were assessed by an independent scientific advisory committee who identified three that should be taken forward into a potential multi-arm multi-stage platform trial. Searches had no date limits but were primarily limited to English due to the time constraint.

Building on our previous work, ²⁷ we (MM) searched PubMed on 14/08/23 (search terms in Figure 1), and loaded Titles and Abstracts to R, before text-mining using a series of regular expressions representing over 12000 compounds derived from DrugBank ²⁸ and from the Alzforum Therapeutic Database ( https://www.alzforum.org/therapeutics) ( Figure 1, top left). Using a similar approach, we text-mined titles and abstracts using regular expressions to identify publications using rats, mice or primates, to create a list of interventions tested in preclinical models. Our focus was primarily on medications; lifestyle, dietary manipulation and cognitive rehabilitation were not considered in scope.

To generate the medium-list, interventions were removed if they would be impossible or impractical to deliver in a phase-3 platform trial based in the UK or another similar socialised healthcare system. A full list of reasons for removal is given in Table 1 and includes drugs that were not listed in the British National Formulary (BNF ²⁹ ) in quarter 3 2023, were considered too expensive for widespread use (threshold BNF price of £30 per month) and interventions that were considered impractical to deliver long-term.

In parallel, we (EPPM, TQ) screened clinical trials assessing interventions separately for cSVD and VaD ( Figure 1, top right), again with the prime aim to improve cognition. The searches for cSVD and Vad trials were conducted between October 2021 - January 2022 and June - August 2022, respectively, and updated for both in June 2023. Completed trials were identified from Ovid MEDLINE, EMBASE and PsychInfo (for VaD only) databases for articles in any language from 2012 to the start of the search. Planned and ongoing trials were searched on clinicaltrials.gov, the European Union Clinical Trials Register (EUCTR) and International Clinical Trials Registry Platform (ICTRP) ( Figure 1). From these, we merged the identified interventions into two lists, one relevant to cSVD and the other for VaD. Interventions were classified by likely mechanism of action. Interventions were then merged with the preclinical medium-list.

The aim was to assess whether there was any evidence that individual interventions improved cognition or reduced cognitive decline in any medical indication. Published systematic reviews/meta-analyses of the effect of interventions on cognition were sought (PMB) irrespective of the outcome scale. If none were present, a rapid internal meta-analysis was performed using completed published data from randomised controlled trials or before-after studies identified from searches of the Cochrane Library and PubMed using controlled vocabulary and validated filters to identify RCTs. Data published as graphs rather than in tables were extracted on screen using mean and standard deviation (calculated from standard error if necessary) with screen values scaled for the Y-axis.

Data were entered into Cochrane Review Manager (version 5.4.1 for Mac) with the aim of identifying interventions that improved cognition, i.e. meta-analysis odds ratio for a good outcome (OR >>1 suggests benefit). For before-after studies assessing the effect of an intervention within a group of patients, the group size was split equally between the before and after results, a conservative approach. Binary data were analysed directly as an odds ratio (95% CI) using the Mantel-Haenszel method with fixed effects (i.e. assuming that the relative treatment effect was common across all included studies ³⁰ ). Continuous data were analysed as standardised mean difference (SMD) with fixed effects to allow studies using different cognition scales to be integrated. Where necessary, scales were inverted to be compatible with direction of positive cognition, i.e. to reverse the direction of effect for scales that go from high to low. SMD and 95% confidence intervals (95% CI) were converted to an odds ratio (OR, 95% CI) so results from binary and ordinal/continuous results could be compared: ^{31, 32} OR = SMD × ∏ / √ 3 = SMD × 1.814

The sample size calculation for the planned UK STEP cSVD platform of N=1,460 was designed to detect an odds ratio of 1.4 (a ‘small’ treatment effect of relevance to a large population of cSVD) with 90% power; hence, short-listed interventions would need meta-analysis evidence that this could be achieved, i.e. OR≥1.4.

We developed (PMB) a bespoke scoring system to assess the likelihood that an intervention might successfully alter cognition in people with cSVD ( Table 2). The system comprised four parts: mechanisms, meta-analysis, identification of intervention for testing in another phase-3 platform, and feasibility. The first section comprised eight mechanistic targets in cSVD that interventions might modulate in an appropriate direction (e.g. pro-endothelial, anti-inflammatory) to potentially reduce the development and progression of cSVD pathophysiology ( Table 2), as used previously ¹² (no additional mechanisms having been identified in the intervening nine years); the sectional-score ranges from 0-8. The second section comprised the results of mini rapid meta-analysis based on the odds ratio and lower limit of the 95% confidence interval, the latter to provide information on the precision of the meta-analysis ( Table 2); practically, the sectional-score based on the odds ratio ranges from above 0 to less than 100 with an odds ratio of >1 compatible with potential benefit. The third section reviewed whether any of three existing neurodegenerative platforms in the UK ACORD collaboration (MS-SMART, ²³ MND-SMART ²⁴ and OCTOPUS ^{33, 34} ) were testing the intervention or had plans to do so; the sectional-score ranges from 0-3. The reason for including this information was that it meant that an independent assessment had already identified the same class or intervention (even if this was based on over-lapping information sources). The scores for these three sections were then added to give a sub-score ranging from just above 0 to less than 100.

The fourth section identified feasibility (yes = 1, no = 0) and required each and every attribute to be feasible including drug availability in the UK (for whatever indication), acceptable side effects, limited indications (<50% of population) and contraindications, the potential for long-term administration, reasonable treatment cost and not already being tested for cSVD ( Table 2). These were multiplied together to create a second sub-score with values or 0 (not feasible) or 1 (feasible). The two sub-scores were then multiplied to give the total score ranging from zero upwards. By example, if anything was considered non-feasible the total score would be zero whatever the results of the first three sections. Otherwise, if everything is feasible, then the final score will be driven by the sum of the first three sectional scores.

We created ‘CVs’ or profiles (PMB) for each shortlisted intervention comprising information on mechanisms of effect of potential relevance to cSVD (as previously ¹² ), evidence of any clinical effect in cSVD, up to three recognised indications (or licensed indications for a drug) and meta-analysis findings (as above). Information from the BNF, ²⁹ or equivalent source if a non-drug, was presented on: dosing; route of administration; NHS cost; adverse event profile; important drug interactions; and cautions, exclusions and contraindications. Information on adherence, score (as above) and any other relevant trials testing it was then given. Finally, a summary of pros and cons resulting from the CV was given.

To compare and rank the shortlisted interventions, an independent scientific advisory committee (ISAG) was created, this comprising experts in stroke or cognition/dementia and trials/platforms; membership comprised RH (chair), JD, PM and SP (academic medics) and GC (patient-public involvement). ‘CVs’ were distributed to the ISAG who then ranked them following open discussion at a video call; PMB was present in a non-voting role to clarify any information.

A search for preclinical references to cSVD (date 14 ^th August 2023) revealed 36,948 which reduced to 19,654 when limited to those with at least 2 mentions of an intervention included in our drug dictionary ( Figure 2, top left). 17,628 of 36,948 references included mention of rats, mice or primates, and of these 10,258 had mentions of 1,757 drugs included in our drug dictionary.

Searches of registries identified 82 randomised trials of relevance to cSVD with activity status - completed 28, completed with no results 0, suspended/terminated 3, ongoing 3, recruiting 27, to start 13, unknown status 8) ( Figure 2, top right). Identified interventions included six antihypertensive drugs, two antiplatelets (when given together), five cognitive enhancers/anti-dementia drugs, three involving electromagnetic stimulation, four physiological interventions, nine vasoactive drugs and 13 others of mixed types ( Table 3).

A parallel search for interventions of relevance to VaD identified 120 randomised trials with activity status - completed 38, completed with no results 0, suspended or terminated 1, ongoing 3, recruiting 50, to start 20, unknown status 8) ( Figure 2, top right). Identified interventions included three antihypertensive drugs, no antiplatelets, ten cognitive enhancers/anti-dementia drugs, two involving electromagnetic stimulation, one physiological intervention, five vasoactive drugs and 11 others of mixed types ( Table 3).

Interventions from the preclinical and clinical searches were merged and de-duplicated to create a long-list. The long-list of potential interventions for cSVD was reduced to a medium-list by removing interventions that were impossible or impractical to study in a phase-3 platform trial running in the UK. Reasons and their explanations are given in Table 1 with the most common including drugs that are not listed in the British National Formulary and so not available to the trial (61 classes of intervention), drugs that need parenteral administration which would be impractical for widespread use (22 classes), interventions with a significant side-effect profile (15 classes), interventions without good existing preclinical or clinical evidence that the intervention might work (15 classes) and existing widespread use of the drug or class of drugs (13 classes).

The results of the mini-meta-analyses based on randomised controlled trials and before-after studies for the shortlisted interventions are shown in Table 4. Interventions varied in their potential to improve cognition for any indication; interventions with a large SMD/high OR were phosphodiesterase5-inhibitors whilst there was no evidence of efficacy for vitamin B12 (see Scoring spreadsheet, Data availability, CC-BY 4.0).

Summaries or “CVs” for 10 interventions were developed, as précised in Table 5.

The Independent Scientific Advisory Committee reviewed the summaries for 10 interventions. Three drugs were ranked: 1) metformin (biguanide, a metabolic modulator), 2) tadalafil (phosphodiesterase5-inhibitor causing vasodilation), and 3) isosorbide mononitrate (nitric oxide donor causing vasodilation). The other seven interventions were not considered to be relevant to study in a large trial for reasons that are given in Table 5, and so were not ranked: atorvastatin (hydroxymethylglutaryl-CoA reductase-inhibitor), donepezil (acetyl cholinesterase-inhibitor, AChE-i), lithium (glycogen synthase kinase 3-inhibitor, GSK3-i), minocycline (tetracycline antibiotic), remote ischaemic conditioning (RIC), semaglutide (glucagon-like peptide-1 receptor agonist, GLP-1ra) and vitamin B6/9/12.

Comparisons with recommendations for assessing interventions in other neurodegenerative conditions (AD, MND, MS) are given in Table 6. Some of the chosen interventions were not identified here for cSVD since they: are not in the BNF (e.g. fasudil, ibudilast, phenserine), ²⁹ have significant side-effects (e.g. oxcarbazepine, perfenidone), were infrequently identified in preclinical screening (e.g. trazodone - only 4 mentions), lack safety information for long-term administration (e.g. anti-herpes drugs), are in widespread use ³⁵ so would lead to a significant number of exclusions (angiotensin receptor antagonists, calcium channel blockers), have neutral effects in other neurodegenerative conditions (amiloride, fluoxetine, losartan, memantine, riluzole, trazodone), can be bought over the counter so potentially contaminating the control group (fatty acids, tadalafil, sildenafil, vitamins), or are already being tested (glucagon-like peptide-1 receptor agonists).

Following screening of pre-clinical and clinical studies, we identified more than 1700 interventions that might be candidate treatments for cSVD. These were filtered down to ten by removing duplicates and interventions that were not possible or practical to assess in a phase-3 platform trial environment or that showed inadequate evidence of efficacy. The ten interventions comprised nine drugs (atorvastatin, donepezil, ISMN, lithium, metformin, minocycline, semaglutide and combined vitamin B6/9/12) and one device (remote ischaemic conditioning) ( Table 5). These were then ranked by an independent scientific advisory committee with three prioritised: a metabolic modulator (metformin) and two vasoactive agents (tadalafil, ISMN), all of which have some evidence that they may improve cognition in other conditions although none are licensed for this purpose.

The first ranked drug, metformin, a biguanide, is licensed for the treatment of type-2 diabetes mellitus (BNF ²⁹ ) and guidelines recommend it for first-line therapy for this indication. It is also licensed for the prevention of diabetes in people with pre-diabetes. The meta-analysis showed that both the odds ratio (2.22) and lower 95% CI (1.66) were well above the target platform odds ratio of 1.40. Metformin has been identified for assessment in other neurological diseases and is being tested in the OCTOPUS platform trial in multiple sclerosis ( Table 6). ^{33, 34} However, a systematic review done for a guideline did not identify evidence supporting the use of antidiabetic drugs for treating cSVD although this recommendation was not specific to metformin. ³⁶

The second ranked drug, tadalafil, is a phosphodiesterase5 inhibitor (PDE5-i) that maintains cyclic guanosine monophosphate levels, the second messenger for nitric oxide. It is licensed for the treatment of erectile dysfunction, benign prostatic hypertrophy and pulmonary hypertension (BNF ²⁹ ) and has recently become available over the counter. The meta-analysis showed that both the odds ratio (7.91) and lower 95% CI (3.76) for PDE5-i were well above the target platform odds ratio of 1.40. However, the meta-analysis was based on a small dataset of 110 participants, mostly from before-after studies. Nevertheless, the positive findings for the third ranked drug, ISMN, a nitric oxide donor (NOD), in the LACI-2 trial ¹⁸ are supportive in view of the related mechanism of action for NOD and PDE5-i.

ISMN is licensed for the prophylaxis of angina and as an adjunct in congestive heart failure (BNF ²⁹ ). Of the three drugs, ISMN is the only one to have been tested in cSVD, specifically in patients with previous lacunar infarction. The LACI-1 and LACI-2 trials showed that ISMN was feasible to take (given with or without cilostazol) for up to a year, was safe and improved cognition and functional outcome and reduced recurrent stroke, especially when given with cilostazol. ^{18, 37– 39} Recently, NIHR HTA funded the LACI-3 trial which will test the safety and efficacy of ISMN and cilostazol in patients with previous lacunar stroke in a 2x2 partial factorial design. Hence, it would now be inappropriate to test either of these drugs in a UK platform.

Practically, the only other short-listed intervention ( Table 5) that could readily be tested in a phase-3 platform for cSVD is one of the licensed oral AD drugs, as identified for VCI. ⁴⁰ Eight short-term small-to-medium sized trials of AChE-i have been performed in VaD and they showed efficacy with an odds ratio (1.52) which exceeds the platform target odds ratio of 1.4; ⁴¹ the lower 95% CI boundary (1.36) just misses this. Donepezil at 10 mg daily would probably be the drug of choice since it appears to be more efficacious than when given at 5 mg daily ( Table 4). Memantine is also a possibility although it was neutral in MND-SMART. ⁴² A recent network metanalysis provides mixed results as to whether AChE-I improve cognition in VCI/VaD. ⁴³ However, it remains unclear whether AChE-i and memantine only modulate symptoms in AD, also have some disease modifying/neuroprotective properties or even address AD pathology which may often be present in VCI/VaD. ⁴⁴ In this respect, cSVD would be an interesting target for testing whether AChE-Is do have disease-modifying effects in view of the longer natural history for cognitive decline in cSVD.

The present study extended the methodology used previously for identifying candidate treatments for cSVD ¹² by adding preclinical and clinical screening, mini-meta-analyses to assess whether an intervention might affect cognition in other disease areas and a scoring system. These approaches are broadly similar but not identical to those used for AD, MND and MS. ^{21– 24} Some relevant pathways and hence licenced drugs may have been missed due to restrictions on search terms – a future search should include other relevant pathways in other neurological diseases even if these have not yet been tested in animal models of cSVD. In retrospect, adding OR>1.40 or even the 95% lower CI >1.40 as a multiplier in scoring would have excluded short-listed interventions with little likelihood of success such as vitamin B12 ( Table 5).

Structured systematic approaches to identifying candidate treatments are not a guarantee that any identified and highly scored intervention will definitely work. For example, five classes of compounds were identified for repurposing as treatments of AD ²¹ and yet trials of minocycline, ⁴⁵ nilvadipine ⁴⁶ and losartan ⁴⁷ were all neutral on clinical outcomes. ²² Similarly, of seven drugs identified for testing in MS, ²³ amiloride, fluoxetine and riluzole were assessed at phase-2b and found to have no effect on MRI brain volume change between baseline and 96 weeks. ⁴⁸ Recently, memantine and trazodone were found to be ineffective for MND. ⁴² Interestingly, all of these trials were much smaller than the planned UK STEP platform trial in cSVD and so underpowered for finding small but potentially useful clinical benefits at the population level providing the intervention is cost-effective. Further, some trials for neurogenerative conditions have relied on intermediate imaging endpoints which may be insensitive to changes in clinical endpoints; the incongruent relationship between neuroimaging intermediate outcomes at phase-2 and clinical functional outcomes at phase-3 has been highlighted for interventions such as citicoline for acute ischaemic stroke and blood pressure lowering for acute intracerebral haemorrhage. The framework for clinical trials in cSVD (FINESSE) recommends that clinical outcomes should be used as the main endpoint in cSVD trials although brain imaging changes may also be collected. ⁴⁹ In this respect, LACI-2 showed that whilst clinical outcomes were positive, imaging outcomes were mostly neutral, suffered from larger data losses and come with increased costs and the need for participants to visit hospitals. ⁵⁰

The main strength of this study is the triangulation of information from preclinical and clinical cSVD and VCI studies in the search for candidate treatments for cSVD, mini meta-analyses of potentially interesting interventions with scoring and then independent prioritisation and ranking. That only six of the ten shortlisted interventions were identified in clinical searches of completed, ongoing or planned trials shows the importance of also assessing the preclinical literature. Taken together, our approach offers a robust, efficient and reproducible approach to drug repurposing selection that could be used for other indications or target diseases.

However, there are also several limitations to the study. First, the intention was to identify potential interventions that could be tested at phase-3 by repurposing existing and inexpensive UK-licensed drugs. This meant we excluded drugs that are not available in the UK, could not logistically be administered long-term, were not ready for phase-3 trial assessment or were too expensive for widespread use. Hence, there will be novel interventions that could be tested at phase-2 and others that are available for repurposing outside of the UK. Equally, there will be drugs that are available in the UK but not in some other countries (such as isosorbide mononitrate). Second, the review was performed rapidly to identify interventions that could be tested in a platform trial in response to a platform trial funding call. Although the time between call and grant submission was four months, the intervention identification phase only lasted two months since the platform design and costing necessarily needed to know what interventions would be assessed initially. As a result, it is likely that we have missed some studies and interventions although many were found as duplicates in searches. We are now extending the process to perform a deeper dive into phosphodiesterase inhibitors and acetylcholinesterase inhibitors/memantine and assess their effects on cognition in non-cSVD indications. Third, the pre-clinical search focussed on models of relevance to cSVD and it is possible that this approach is too restrictive. Fourth, although the scoring system was based on one used previously, ¹² it was expanded for use here without prior validation although it appears to have face validity. Fifth, we did not, in the main, account for dose in the meta-analyses. For the platform, we planned to initiate drugs at a low-medium dose and then escalate to near maximal, as we did in LACI-1/2 ^{18, 38} and will do in LACI-3. ⁵¹ Sixth, we did not seek evidence that interventions would alter mechanistic efficacy biomarkers. And last, there are additional approaches that may be useful in identifying candidate drugs that we did not incorporate, for example Mendelian randomisation assessments ^{52– 55} and medication-wide association studies. ⁵⁶

Ethics and consent were not required.