Treatment Options for Dementia with Lewy Bodies: A Network Meta-Analysis of Randomised Control Trials

Dementia with Lewy bodies (DLB) is the third most common type of dementia, after Alzheimer’s Disease (AD) and vascular dementia, accounting for approximately 10–15% of all dementia cases [1]. It is estimated that 1.3 million individuals in the USA and 100,000 individuals in the UK are affected by DLB [2].

DLB is a gradually progressive condition occurring as a result of intraneuronal inclusions—consisting of α-synuclein aggregates and ubiquitin (known as Lewy bodies)—in the neocortex, forebrain, brainstem and other parts of the nervous system [3]. Individuals suffer worsening cognitive, neuropsychiatric, autonomic and motor impairment [4]. The decline in function results in a high economic burden to society, with increasing direct and indirect costs, especially following admission of the individual into nursing homes [5‐7].

Treatment is targeted at specific disease manifestations or symptoms [8]. Based on the results from two randomised controlled trials (RCTs), which demonstrated improved cognitive function in patients with DLB treated with Eisai’s ARICEPT® (donepezil) [9], ARICEPT was approved by the Japanese Pharmaceuticals and Medical Devices Agency in 2014 to become the world’s first and currently only licensed treatment for DLB [10]. Worldwide, drugs licensed for the treatment of AD are widely used to treat patients with DLB, with varying degrees of success [4, 8, 11, 12]. Given the resulting limited available clinical evidence, the objective of this study was to conduct a network meta-analysis (NMA) to explore the comparative efficacy and safety of current pharmacological interventions used to treat patients with DLB, based on available evidence from RCTs.

NMAs are used widely for health technology appraisals and regulatory approval discussions where direct trials have not been conducted between treatments of interest. Increasingly, they are being performed to inform decision-making regarding the comparative efficacy and safety of alternative treatments. The NMA is an extension of traditional meta-analysis approaches that utilises all available evidence to provide an assessment of the indirect comparison of treatments, where direct evidence, due to a lack of head-to-head studies, is unavailable.

The NMA was conducted based on the findings of a previously published systematic literature review of RCTs reporting on pharmacological interventions in DLB [13]. Prior to conducting the NMA, a feasibility assessment was carried out to confirm that indirect comparisons between DLB pharmacological interventions were viable for the proposed outcome measures and that there were no new data that may impact the NMA expected to be published in the immediate future. The protocol for the analysis was developed in consultation with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis Protocols (PRISMA-P) Statement [14, 15]. This article is based on previously conducted studies and does not contain any studies with human participants or animals performed by any of the authors.

The aim of this study was to determine the comparative clinical efficacy of cognitive–behavioural treatments that are commonly used in patients with DLB. It is understood that some patients may have received l-dopa, dopamine agonists, antidepressants or benzodiazepines at baseline. However, the proportion of such medications was assumed to be the same within clinical studies, as this was unreported in most studies. Additionally, evidence has shown that that l-dopa does not compromise the cognitive or behavioural profiles of patients with DLB either acutely or over a longer period of time, meaning they will have a limited impact on the included outcome measures [16].

Within the available studies, donepezil was the only treatment that explicitly compared the outcomes for the different dose amounts. All other studies focusing on alternative therapies included a range of dose amount within a single arm due to the titration process.

Donepezil has been approved globally for the treatment of dementia of the Alzheimer’s type, with the 5 mg and 10 mg dosages being approved in the USA and Europe. In Japan specifically, donepezil dose options of 3, 5 and 10 mg have been approved for the treatment of both AD and DLB [10, 17]. On this basis, the donepezil 3 mg dose was included as a specific treatment comparator within the NMA. Due to citalopram and risperidone forming a disconnected network, these two medications were excluded from the analysis.

A unified scale to comprehensively assess the symptoms or disease progression of DLB is currently lacking. Therefore, a number of different measures were included as efficacy outcome measures in the analysis, with the most most commonly reported ones being the Mini-Mental State Examination (MMSE) [11, 12, 18‐21], variations of the Neuropsychiatric Inventory (NPI) instrument [11, 12, 18, 21‐24] and variations of the Unified Parkinson’s Disease Rating Scale (UPDRS) [25]. While these measures may not capture the full range of effects for each of the included treatment options, they were selected as the most consistently reported measures across the different studies available for patients with DLB.

A number of variations of the NPI instrument are used across different studies. However, the ten-item NPI (NPI-10) was selected for inclusion in the base–case NMA analysis, as this version was the most commonly reported instrument across the available trials. Most of the identified studies that reported the UPDRS measure only included part III of the instrument, which focuses on the motor functions of individuals, rather than using the complete UPDRS instrument.

Additionally, a range of safety measures were included to assess various aspects of the safety profiles. These measures were the overall rate of adverse events (AEs), discontinuation rate, discontinuations due to AEs and the rate of psychiatric events. A summary of all outcome measures are included in Appendix 1 in the Electronic Supplementary Material (ESM).

Variations in the time horizon evaluated, the baseline scores for outcome measures and the patient populations included within each trial were taken into consideration. These variations are likely to have introduced higher levels of heterogeneity between the studies than that expected between any studies conducted independently. The variations observed in the time horizons and the baseline scores were deemed sufficiently limited to allow the results to be meaningful. Furthermore, given the number of trials available, excluding trials with significantly differing characteristics would introduce a greater bias than the effect of including the studies in the analysis. The variation in the baseline patient populations was considered to have the potential to significantly impact treatment outcomes. Therefore, it was decided to exclude trials with non-DLB specific patient populations as part of the sensitivity analyses conducted. The NMAs were performed using Bayesian methods, following the guidance provided in the National Institute for Health and Clinical Excellence (NICE) Decision Support Unit Technical Support Document series on evidence synthesis [26]. The results of the analyses were measured as absolute and relative effect (difference in mean change from baseline for continuous variable and odds ratio [OR] for binomial variables), the probability of being the most effective or safest treatment and the rank of the treatment by outcomes. The results were deemed to be statistically significant if the 95% confidence interval (CI) did not cross the origin. All analyses were performed with WinBUGS version 1.4.

Both a fixed effect (FE) and random effect (RE) model were run for each network. The base–case model was selected according to the goodness of fit. The alternative modelling approach was included as one of the sensitivity analyses conducted. All other sensitivity analyses for all outcome measures were conducted using the modelling approach selected for the base–case model. Sensitivity analyses tested a range of different scenarios: the exclusion of mixed study populations comprising patients with other forms of dementia, including AD and Parkinson’s disease dementia; the inclusion of additional data; or the pooling of all doses of donepezil (see Appendix 2 in the ESM for further details). The pooling of the donepezil dose amounts was included as part of the sensitivity analysis to align the approaches taken within other studies, where a range of dose amounts were included within a single treatment arm.

To our knowledge, this study is the first NMA that has been developed to provide a comparative analysis across the available therapy options for DLB. Given that all of the included trials were conducted compared to placebo, NMA makes it possible to combine all of the available placebo-controlled trials so that the included existing treatment options for DLB can be indirectly compared.

The findings that donepezil and rivastigmine show the best efficacy profiles is consistent with and supports current clinical practice favouring the use of acetylcholinesterase inhibitors as first-line pharmacological treatment for patients with DLB [28]. Evidence suggesting that the cholinergic hypothesis fits better with DLB than AD provides additional rationale for the use of these medicines, which have been licensed for the treatment of AD [29]. Based on the NPI measurement scale, there was a high probability of rivastigmine being the best performing treatment in terms of efficacy alone. For donepezil, the effect is greater at the 5 mg and 10 mg doses. It may be expected that different dose amounts and formulations available for rivastigmine would produce varying results, as seen for donepezil; however, the available trial data did not differentiate between these within the available publications focusing on DLB patients.

In terms of safety, rivastigmine and quetiapine appear to rank the lowest in the safety/tolerability profile. For rivastigmine, gastrointestinal side effects, hypersalivation, lacrimation and urinary frequency may likely be the predominant reason for higher rates of discontinuation compared to placebo in its clinical trials. Such side effects, which increase patient self-care and carer burden, may indeed be more troublesome in patients with DLB; however, this notion is not borne out in the NMA, with discontinuation rates due to AEs compared to placebo being comparable—albeit with extremely wide confidence intervals (OR 1.04; 95% CI 0.04, 27.71).

For quetiapine, which was investigated specifically for managing agitation and psychosis in patients with dementia, higher odds of AEs (OR 3.73; 95% CI 0.69, 21.49) do not lead to higher odds of discontinuation or discontinuations due to AEs versus placebo (OR 0.29; 95% CI 0.02, 3.43 and OR 0.85; 95% CI 0.00, 901.8, respectively). However, regulatory bodies have highlighted caution with their use in older patients with dementia-related psychosis due to increased risk of cerebrovascular accidents and death [17]. Additionally, there are strong clinical recommendations to avoid the use of antipsychotics in older patients due to an associated greater rate of cognitive decline [30]. In relation to DLB specifically, severe sensitivity reactions, including exacerbation of parkinsonism, confusion and autonomic dysfunction. may particularly limit the usefulness of antipsychotics. Furthermore, greater cognitive decline and confusion may lead to further psychiatric events, which do appear to be increased for quetiapine versus placebo (OR 2.53; 95% CI 0.35, 20.88) and all other comparators (see section Safety Outcome Measures of the NMA).

The mixed findings reported to date for memantine are reinforced in this NMA. While in clinical trials memantine demonstrated a greater magnitude of clinical global impression of change than other agents [18], based on the present analysis, it appears to perform less well on secondary measures, such as those explored in this NMA. For the MMSE measurement scale, the outcomes even suggest that, at the mean level, memantine performed worse than placebo. This in part captures the high levels of uncertainty in the MMSE outcome, where the confidence interval crosses the point of zero difference compared to placebo. Additionally, if memantine is treating symptoms not captured within this measure, but the AEs have an impact on cognitive functions, there may appear to be a worsening compared to placebo in this one measurement scale. This possibility further highlights the need to compare the treatments based on a range of available metrics and that the decision should not be based on a single metric when selecting an appropriate treatment option for each patient. In terms of safety, overall AEs were found to be lower for memantine compared with donepezil 10 mg (OR 0.98; 95% CI 0.24, 19.36). Discontinuation rates due to AEs were higher for memantine (OR 2.95; 95% CI 0.15, 79.70), which may be due to increased psychiatric events (OR 1.11; 95% CI 0.16, 75.67). The relationship found here between discontinuations and psychiatric events is consistent with the increased risk of delusions and hallucinations that may be particularly problematic for patients with DLB.

Donepezil 3 mg had lower rates of AEs but higher rates of overall discontinuation and discontinuation due to AEs compared with the 5 mg and 10 mg doses. This could possibly be explained by greater efficacy at higher doses, potentially leading to patients and clinicians having an increased willingness to tolerate AEs. In terms of donepezil 3 mg, these findings support the current Japanese license in which this is only an initial dose, with tolerability-based up-titration to 5 mg and 10 mg recommended for the treatment of DLB [10].

Overall, while rivastigmine performed highly in terms of the efficacy, its safety outcomes were less favourable. Donepezil 5 mg ranked consistently higher on the efficacy and safety outcomes compared to other treatments. This network of evidence therefore suggests that donepezil 5 mg may have the best overall risk/benefit profile for the treatment of patients with DLB. Indeed, in real-world clinical assessments of patients, there may a potential for a better selection of those who may derive the greatest benefit from donepezil 10 mg, based on overall benefit/risk for particular individuals.

Although these trends were present in the estimated values, there were no statistically significant results observed. The lack of significance may be driven either by the limited availability of data to inform the analysis, or by heterogeneity within the study designs and populations. A high level of heterogeneity within each of the study populations may suggest that a single preferred treatment option to use across all DLB patients may not be preferable. Instead, it may suggest that the treatment option provided should be tailored to each individual, leading to the approach of personalised treatments for patients. The difference between a traditional shrewd assessment of a complex medical condition and a true personalised (or precision) medicine is the degree of reliance on data to objectively make decisions about specific treatment paths that may be more effective for each individual. Given the nature and quality of the currently available data presented here, it is difficult to achieve.

Two specific sensitivity analyses, one pooling all doses of donepezil and another excluding mixed populations, did not impact the trends seen in the base–case scenario or their significance. Additionally, the impact on heterogeneity in the sensitivity analyses was inconsistent across different outcome measures.

In the pooled donepezil dose analyses, the lack of change could be explained by the fact that the 3 mg dose is only an initiation dose and effectiveness would be best assessed when individuals are maintained for adequate time on the maximum tolerated dose of either 5 mg or 10 mg. This would also be more in keeping with actual clinical practice, where the aim is to optimise effectiveness while balancing tolerability, as well as being a better indirect comparison with other drugs for which up-titration is a standard protocol.

Due to the exclusion of studies, there were no data to include quetiapine in the analysis excluding mixed populations. For the other comparators, the lack of difference in the mixed populations exclusion analysis could be partly explained by the relatively small proportions of patients with Pervasive Developmental Disorder (PDD) or AD with parkinsonian features in the studies that formed the base–case NMA. This provides reassurance, as while there may be differences in the preponderance of specific symptoms across DLB, PDD and AD with parkinsonian features, reaching a clear diagnosis may be challenging in many patients.

As the interventions explored in this NMA have different modes of action, it is possible that standardised instruments are not sensitive to improvements in specific symptoms that may be better controlled by specific drugs and may be of particular concern for individual patients. Therefore, in the absence of a disease-modifying treatment, general management should continue to be symptomatic, with a focus on balancing benefit/risk for individual patients.

Given the low number of trials and the high degree of heterogeneity between the studies included in this NMA, caution should be exercised in making decisions based upon the results produced by the NMA. The results presented from this NMA should be used to explore the expected trends in the data, rather than as a substitute for direct evidence for the purposes of economic modelling or other analyses. It is likely that additional evidence would have a high impact on the results, particularly if head-to-head evidence was introduced to the network. This highlights the need for further research with studies comparing treatments to active comparators and utilising measures that may be more sensitive and specific to DLB.

Alongside the limitations of the available data, the lack of significant results across each of the included measures may also support the need for a more personalised treatment approach for DLB patients. Due to the heterogeneity across the range of patients who may be characterised as suffering from DLB, personalised medicine needs to be based on the different characteristics of individual patients. By combining clinical, pharmaceutical and socioeconomic information, one may observe patterns in the effectiveness of particular treatments and identify the characteristics, such as genetic, environmental and lifestyle variations, that may be correlated with success or failure for each treatment option.

In conclusion, this is the first NMA exploring the comparative efficacy and safety of pharmacological interventions used for the treatment of patients with DLB. While the results are not statistically significant, donepezil 5 mg ranked highest on the efficacy and safety outcomes; it may result in the best overall benefit/risk profile for the treatment of patients with DLB. Future trials should aim to undertake direct head-to-head treatment comparisons to improve the evidence base for a disease with a high unmet need.