We searched MEDLINE, Embase and ClinicalTrials.gov for studies that fulfilled all criteria: (1) study populations defined as patients with any of Alzheimer’s disease, mild cognitive impairment (MCI), Parkinson’s disease (PD), Lewy Body dementia, frontotemporal dementia (FTD) or progressive supranuclear palsy; (2) prospective clinical trials that compared drugs that increase the level of NA or act on one of the three receptor classes, versus placebo and (3) studies that reported cognitive, neuropsychiatric or behavioural outcomes (table 1). Flow diagram of search is shown in figure 1. Full search terms and inclusion/exclusion criteria are listed in table 2. Ongoing clinical trials of noradrenergic agents were not included.

We searched MEDLINE and Embase from 1980 to 22 December 2021 using a combination of controlled vocabulary (eg, Medical Subject Headings) and free-text terms. We manually searched further information sources, including published studies or reviews, conference abstracts and supplementary notes. We searched ClinicalTrials.gov to identify trials with published results.

Four reviewers (MCBD, MdG, BG and IO) independently screened titles and abstracts. Potentially eligible studies were then discussed between the reviewers before deciding on inclusion. Arbitration was conducted by RH and PAM. Data were extracted from full texts by two reviewers (MCBD and MdG) and additional data were requested from authors where required.

Outcome measures for each domain were selected based on a hierarchy determined by their frequency of use across included studies. For example, for the global cognition outcome measure, Mini-Mental State Examination (MMSE) was used in seven Alzheimer’s disease studies and therefore included in the analysis for these studies. Two Alzheimer’s disease studies reporting measures of global cognition employed the Alzheimer’s Disease Assessment Scale—Cognitive Subscale (ADAS-Cog), and so the results from this were included in analysis. For apathy, four Alzheimer’s disease studies used the Apathy Evaluation Scale (AES), and this was extracted where available. The next most common index of apathy was the Neuropsychiatry Inventory—Apathy (NPI-A), while one study used the Frontal Systems Behaviour Scale—Apathy (FrSBe-A). For selection of outcome measures for the cognitive subdomains the most frequently used measures were prioritised, after which, those measures deemed most similar were chosen from the remaining studies. See table 1 for details on the included outcome measures.

Drugs were grouped by principal mechanism of action (table 1). Analyses were run for all studies investigating specific classes of drugs, where there were enough (>2) to do so (see online supplemental material).

Two reviewers (BG and IO) assessed the quality of the studies using the National Heart, Lung, and Blood Institute Quality Assessment of Systematic Reviews and Meta Analyses.35 This comprises 14 questions on blinding, randomisation, equality between study arms, drop-out rates, outcome measures, power and analyses. Raters score each question as 1 if the methodology is suboptimal, and rank studies as ‘good’, ‘fair’ or ‘poor’ if they scored 0, 1 or >1 respectively across the 14 questions.

To evaluate treatment efficacy, our meta-analysis used Review Manager V.5.4.36 For each outcome measure, we calculated the change in group means for drug and placebo groups in each study, from baseline to the final timepoint. Where the relevant information was specified in the studies, we chose to use datasets that had excluded the participants who dropped out before the final time point. For outcome measures in which a negative change in score indicated an improvement, change scores were multiplied by −1; and are shown as such in the figures. As all outcomes were continuous, we present the calculated standardised mean differences (SMDs) with 95% CIs using an inverse variance random effects model. If a study used multiple intervention groups (different doses), we treated each study arm as a separate trial and compared them against the same control group.23 37 38 Where SDs for the change in mean were not reported, SDs were imputed using formulas for continuous outcomes in the Cochrane Handbook.39 SMDs were used because studies used different measures for the same outcome, except for digit span forwards and backwards where the mean difference was calculated. On the assumption that clinical and methodological heterogeneity was present and could influence outcomes, we used random effects meta-analysis models to estimate SMDs.

Heterogeneity was measured using the I² statistic. Funnel plots were created using JASP40 to graphically represent effect sizes and identify asymmetry resulting from publication bias. This was quantified using Egger’s tests, with the caveat that tests for asymmetry with <10 studies are likely to be underpowered.41 Studies that differed significantly from the pooled effect in that their 95% CIs did not overlap with the CIs of the pooled effect, were considered outliers. The effect of their exclusion on the pooled effect size and study heterogeneity was explored.

Post hoc meta-regression analyses were conducted to define whether age, gender, duration of treatment and year of publication had any effect on the results obtained for global cognition and apathy. Meta-regressions were conducted for each covariate separately. For each meta-regression, the number of studies included in the model, the covariate estimate (β), the p value and the proportion of variance explained (R²) were reported (online supplemental tables S4 and S5) .

Table 1 shows the baseline characteristics of included trials. We focus on the 19 trials of Alzheimer’s disease and MCI, as the search revealed only four eligible PD trials, one in FTD, and none in Lewy Body dementia or Progressive Supranuclear Palsy. Details of the non-Alzheimer’s studies are shown in online supplemental table S1.

The Alzheimer’s disease studies were prospective randomised controlled trials, with treatment duration between two and 52 weeks. Study participant number ranged from 5 to 346 and the mean participant age ranged from 60 to 85 years. The most common drugs were norepinephrine reuptake inhibitors (NRIs; nine studies), followed by α₁-AR antagonists (four studies), α₂-AR agonists (three studies), α₂-AR antagonists (two studies) and β-AR antagonists/blockers (one study).

Six studies were of ‘good’ quality, with seven ‘fair’ and six ‘poor’ (online supplemental table S2); all were included in the analysis. Methods of randomisation were adequate in all but one study (unable to determine in five). Four studies had a high drop-out rate (>20%) from the treatment arm and were considered ‘poor’ overall. Only five included studies reported a sufficiently large sample size to detect a significant difference in primary outcome measure with 80% power; four of these were considered ‘good’ overall, with one ‘fair’ (the remaining two ‘good’ papers did not report on their power).

Meta-regression analyses showed that none of mean age, sex, duration of treatment and year of publication were significantly associated with effect size differences in either the global cognition or apathy meta-analyses (p>0.05) (online supplemental tables S4 and S5 and figure S6). Results for global cognition were limited by insufficient power and one study could not be included as age and sex were not reported,37 as was one study for the investigation of the influence of gender on the apathy results.46

This meta-analysis suggests that drug repurposing with established noradrenergic treatments, such as atomoxetine, methylphenidate and guanfacine, may benefit people with Alzheimer’s disease, particularly given existing evidence of their relative safety in clinical practice, and pharmacological target engagement.54 However, several factors may explain the variability in results across studies. These factors are important to bear in mind when designing future clinical trials.

First, cognitive performance is proposed to be optimal at an intermediate level of noradrenergic tone, in accordance with the Yerkes–Dodson arousal curve.55 Therefore, it is likely that noradrenergic treatments are maximally effective only at a specific dose; or only in patients with a sufficient degree of LC-NA dysfunction so as to not to induce a state of over-activity of the LC-NA system. Such overactivity might lead to agitation and anxiety, or worsening of cognition.27 56 For example, atomoxetine does not merely produce a general increase in synaptic NA levels,54 but increases the likelihood that patients will be at an ‘engaged’, intermediate state of noradrenergic tone. In promoting this state, LC neurons are rendered less tonically active with heightened neural gain. This can increase responsiveness to stimuli, increase network integration, and improve cognitive performance.57 However, in clinical psychopharmacological studies, there is baseline-dependency such that the effect of atomoxetine on connectivity and cognition depends on severity of disease and integrity of the LC. Such baseline-dependency may apply in Alzheimer’s disease too, calling for stratification tools in clinical trials and practice.27 58

Second, we analysed the effects of both AR agonists and antagonists. While it might seem counterintuitive, there is evidence for effectiveness of different treatment strategies. Recall that some pre-synaptic receptors are inhibitory on NA release, such that antagonism can paradoxically increase NA neurotransmission. In addition, receptor subtypes can have opposing actions in different locations within the brain. With the inverted-U shape response to noradrenergic stimulation, stimulation and inhibition may be helpful at different stages of illness and for different cognitive or behavioural domains. Further evidence is required to differentiate the effects of individual drugs. At present, neither our meta-analysis of global cognition nor apathy, both of which showed positive effects of treatment, included trials evaluating agents that had opposing effects on the same receptor subtype.

Third, optimal symptomatic treatment of AD may not come from targeting a single neurotransmitter system. Yu and Dayan’s theoretical account of synergistic cholinergic and noradrenergic systems in attentional processes suggests that noradrenergic therapy may prove most effective when used in tandem with cholinergic approaches.59 This approach was taken with the trial by Mohs and colleagues included here.22 There are ongoing studies looking at combination therapies with noradrenergic and cholinergic agents based on this principle, including a Phase 3 trial of guanfacine as an 'add-on' to cholinergic treatment (NCT03116126).60 A number of included studies involved drugs with a combination of noradrenergic and dopaminergic action, such as methylphenidate, and it is possible that dual stimulation leads to greater clinical efficacy, particularly for apathy, than more targeted approaches. Direct comparison between selective noradrenergic and dopaminergic agents and a less targeted treatment such as methylphenidate is required to clarify this.

Fourth, it is important to consider the potential benefits of noradrenergic treatments beyond cognitive and behavioural effects. Several studies have shown that with dysfunction in the LC-NA system in Alzheimer’s disease, there is a loss of endogenous anti-inflammatory effects and of the capacity for amyloid beta peptide (Aβ) degradation and clearance1 61 62 that is normally promoted by NA. Noradrenergic deficits may thereby lead to an increase in Aβ production. In the context of reduced NA, the toxic inflammatory effect of Aβ and tau are heightened, exacerbating neurodegeneration.63 64 NA also regulates microglial surveillance and synaptotoxicity,6 where microglia activation is predictive of faster cognitive decline of people with Alzheimer’s disease and MCI.65 Therefore, noradrenergic treatments might have both symptomatic and disease-modifying effects. It is important to note that these effects can be positive or negative. Brain activity leading to the release of NA can lead to the aggregation of Aβ and tau, and thus there is potential for both deleterious and disease-modifying effects.62 Future trials aiming to measure effects on pathological progression in addition to symptomatic improvement will need sufficient treatment duration to detect any such effect.

Fifth, the benefits of noradrenergic treatment need to be weighed against potential adverse effects, including cardiac risks, especially in people with multimorbidities. Noradrenergic reuptake inhibition and agonism have the potential to increase heart rate, blood pressure and cardiac risk. Even in older adults, the actual changes in rate and pressure seem minimal where reported in trials, but screening and risk-stratification may be required.

The strengths of our study include the following: the comprehensive approach to the search that discovered 19 Alzheimer’s disease studies, including 1811 patients, that investigated the effect of noradrenergic drugs on cognition and behaviour; pre-planned subgroup analyses to explore the effect of treatments on a range of outcome measures; and the use of established methods to assess effect of outliers and quality of evidence.

There are nonetheless limitations to our study. The inclusion criteria used a threshold of drugs’ noradrenergic action, which was not based on fixed pharmacokinetic or pharmacodynamic metrics. Rather, we considered the relative action of drugs on noradrenergic and non-noradrenergic systems at the doses used in current licenced applications of the drugs. We acknowledge that other drugs have some noradrenergic activity, even if not enough to be included here. The treatments included medications that have varying mechanisms within the noradrenergic system, including both AR agonists and antagonists, as well as some non-noradrenergic activity. It is uncommon for drugs used in clinical practice or human trials to have exact specificity of pharmacological action. Critically for the conclusions drawn here, the commonality of effect of all the compounds included here lies in their shared noradrenergic action. Different measures (for the same outcome) were used across studies. The change in mean score for each outcome measure was reported at the group level in most studies, whereas a change at the individual level may have yielded more accurate results and allowed estimation of state-dependency in relation to disease severity. We did not model baseline performance or symptom severity, which may account for some of the variation in treatment response. Regarding study quality, only 6 of the 19 being rated as ‘good’. We propose that further phase 3 clinical trials are warranted in Alzheimer’s disease, with optimal design and outcomes, with consideration of pragmatic baseline stratification with a view to regulatory benchmarking. In view of the impact of non-Alzheimer dementias on the LC-NA system, clinical trials of other diagnostic groups are also indicated, with a strong rationale for expected treatment potential.