Introduction

Sirtuins, dubbed as cellular ‘watchmen’ and ‘stress sensors’, control cell function by determining cell fate, maintaining energy supply, and preventing DNA damage to maintain genomic integrity. The mammalian sirtuin family of enzymes, composed of sirtuins 1–7 (SIRT1-SIRT7), have received considerable interest due to their role in regulating responses aligned to physiological stress and thus, health and longevity. Sirtuins function as histone deacetylase (HDAC), removing acetyl groups from target proteins and effectively activating or inhibiting these proteins depending on the specific cellular context1. For sirtuins to work efficiently, they use nicotinamide adenine dinucleotide (NAD+), through a reduction and oxidation mechanism which fuels the synthesis of ATP2. NAD+ and its redox couples, NADH and NADP(H), not only control metabolism and sirtuins, but also regulate several redox-sensitive pathways3. Sirtuins, in turn, regulate these redox pathways directly through deacetylation and indirectly by maintaining the NAD+ pool.

SIRT1 is known to deacetylate a range of salient transcription factors and proteins, including: AMP-activated protein kinase (AMPK), a central regulator of energy metabolism that maintains cell ATP concentration; peroxisome proliferator-activated receptor alpha (PPAR-α) involved in lipid metabolism; peroxisome proliferator-activated receptor gamma coactivator 1-alpha (PGC-1α) involved in mitochondrial metabolism; nuclear factor erythroid 2–related factor 2 (Nrf2), an antioxidant transcription factor; nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB), the regulator of innate immunity; and Ku70, p53, forkhead box transcription factors (FOXO) that are involved in DNA repair and cell survival in ageing and cancer1,4. In cancer for example, the balance between inhibitory SIRT1 deacetylation of p53 and SIRT1 recruitment of p53 acetylation, can determine whether damaged cells survive or undergo apoptosis5. SIRT1 also enhances cell survival in the heart-brain axis by upregulating brain-derived neurotrophic factor (BDNF) that is crucial for neuronal growth, synaptic plasticity, and vascular endothelial growth factor (VEGF) signalling6.

In addition to transcription factors, SIRT1 deacetylates histones—structures around which DNA wraps—facilitating chromatin compaction and silencing of genes involved in disease. For example, SIRT1 deacetylates histones at H4K16, decreasing its binding with the pro-inflammatory cytokine TNF-α promoter and alleviating inflammation7. Histone deacetylation is also involved in SIRT1’s regulation of circadian genes that control the production of hormones and enzymes, including nicotinamide phosphoribosyltransferase (NAMPT), the rate-limiting enzyme in NAD synthesis8,9. Hence, loss of SIRT1 results in reduced NAD, impaired circadian rhythm, and increased expression of ageing-related genes10.

Ageing is characterised by decreased SIRT1, NAMPT, NAD+, and changes in circadian rhythm9. This regulatory network forms the basis of developing SIRT1 activators and NAD+ precursors for ageing-related disorders. In old mice, supplementation with NAD+ precursors can increase SIRT1 activity, stem cell regeneration, mitochondrial and physical function, and lifespan11. Similarly, SIRT1 overexpression can delay ageing, restore physical fitness, and extend lifespan in old mice12. Exercise-induced SIRT1 upregulation has been shown to decrease inflammation, apoptosis, and metabolic dysfunction in mice13,14,15. Similarly, in humans, aerobic exercise-induced SIRT1 increases antioxidant capacity (catalase and superoxide dismutase mainly) while decreasing cell senescence in heart failure16,17, and improves overall metabolic profile in type 2 diabetes18.

ATP demand during high-intensity exercise increases the AMP:ATP ratio that is sensed by AMPK, which responds by stimulating cell glucose uptake and fatty acid oxidation to generate more ATP and NAD+, the latter serving as a fuel for SIRT119. Exercise also generates reactive oxygen species (ROS) such as H2O2, a known signalling molecule that activates AMPK as a secondary consequence of mitochondrial ATP production20. However, excessive ROS can also repress SIRT1 activity by post-translational oxidative modification or by modifying intracellular NAD+ levels21. Indeed, high-intensity exercise increases DNA damage, which can recruit other NAD+-consuming repair enzymes such as poly ADP ribose polymerases (PARPs) that compete with and inhibit SIRT121,22. As the SIRT1 response to exercise is complex, it is essential to ascertain how exercise intensity, type, or duration can affect SIRT1 levels, both acutely (single bout) and after repeated training. Therefore, the aim of this exploratory review is to summarise and systematically assess published exercise interventions quantifying SIRT1 (protein content, gene expression, enzyme levels, and enzyme activity) in apparently healthy participants. Specifically, our primary objective is to determine if exercise can increase SIRT1, with a further objective to ascertain the type of exercise that may cause any modification to SIRT1.

Results

Literature search

A database search retrieved a total of 3,971 non-duplicate articles, from which 34 were included in the meta-analysis (13 acute response and 21 training interventions). The search and selection process is summarised in Fig. 1.

Figure 1
figure 1

PRISMA flow diagram showing the database search and selection process.

Full size image

Study characteristics

Participants

Participant age ranged from 20–66 years old. Six studies involved seniors, six studies tested overweight/obese participants, and the remaining studies involved young to middle-aged, normal weight participants. Only 10 studies involved women.

Quality assessment of individual studies

Studies scored 6 or higher, which is within the threshold for separating high-quality from low-quality studies based on a validity study of the original 11-item Cochrane Back Group Risk of Bias Tool23.

Biomarkers/analytical techniques

Studies were classified according to tissue and biomarker: SIRT1 gene expression in skeletal muscle (measured via qPCR or RT-PCR), SIRT1 protein content in skeletal muscle (measured via Western blot), SIRT1 enzyme levels in blood (measured via ELISA), and SIRT1 enzyme activity in skeletal muscle (measured via fluorometric or sirtuin activity assay).

Exercise classification

Studies were also grouped based on exercise type, intensity, duration, and feeding status of participants (fasted or fed). “Fasted” was used to accurately describe the exercise protocols that implemented an overnight fast, however, we cannot determine the effects of fasting on exercise per se since the said studies did not include a non-fasted control group. Intermediate and advanced yoga and Pilates interventions were included and classified as resistance training, while meditative yoga interventions consisting of breathing exercises alone were excluded. Table 1 summarises all exercise studies used in the meta-analysis and their abbreviations, while the exact protocols used are summarised in Tables 2 and 3.

Table 1 Exercise classification table based on type, intensity, duration, tissue sample, biomarker, and feeding status of participants.
Full size table
Table 2 Effects of a single bout of exercise on SIRT1.
Full size table
Table 3 Effects of repeated exercise training on SIRT1.
Full size table

Acute SIRT1 response to exercise

Skeletal muscle SIRT1 gene expression (measured via qPCR or RT-PCR) increased after a single bout of high-intensity exercise or following fasted exercise. Study characteristics are summarized in Figs. 2 and 3.

Figure 2
figure 2

Forest plot quantifying skeletal muscle SIRT1 gene expression (measured via qPCR or RT-PCR) following a single bout of high-intensity exercise. Adjusted standardised mean difference (Hedges’ g), relative weight of each acute study response, confidence interval (diamond), and prediction interval (blue line) are also shown. HIA high-intensity aerobic, HIIT high-intensity interval training, OM old men, M men, SIT sprint interval training, W women, YM young men. Percentages denote proportion of exercise intensity. n = sample size.

Full size image
Figure 3
figure 3

Forest plot quantifying skeletal muscle SIRT1 gene expression (measured through qPCR or RT-PCR) following a single bout of fasted exercise (overnight fast). Adjusted standardised mean difference (Hedges’ g), relative weight of each acute study response, confidence interval (diamond), and prediction interval (blue line) are also shown. EnA-C endurance aerobic cycling, EnA-T endurance aerobic treadmill, HIA high-intensity aerobic, OM old men, M men, SIT sprint interval training. n = sample size.

Full size image

High-intensity exercise

The analysis is based on 12 studies and utilised a random-effects model. The mean effect size adjusted with Hedges’ g is 0.723 with a 95% confidence interval of 0.311–1.134. To test the null hypothesis that the mean effect size is zero, we used the Z-value which is 3.444 with p = 0.001, and using a criterion alpha of 0.050, we reject the null hypothesis and conclude that in the universe of populations comparable to those in the analysis, the mean effect size is not precisely zero. To test the null hypothesis that all studies in the analysis share a common effect size, we used the Q-value which is 35.016 with 11 degrees of freedom and p < 0.001, and using a criterion alpha of 0.100, we reject the null hypothesis that the true effect size is the same in all these studies. The I-squared statistic is 69%, which suggests that some 69% of the variance in observed effects reflects variance in true effects rather than sampling error. Assuming that the true effects are normally distributed, we can estimate that the prediction interval is − 0.674 to 2.119. The true effect size in 95% of all comparable populations falls between this interval.

Fasted exercise

The analysis is based on seven studies and utilised a random-effects model. The mean effect size adjusted with Hedges’ g is 0.809 with a 95% confidence interval of 0.246–1.372. The Z-value is 2.815 with p = 0.005, hence we reject the null hypothesis and conclude that in the universe of populations comparable to those in the analysis, the mean effect size is not precisely zero. The Q-value is 17.807 with 6 degrees of freedom and p = 0.007, thus we reject the null hypothesis that the true effect size is the same in all these studies. The I-squared statistic is 66%, which suggests that some 66% of the variance in observed effects reflects variance in true effects rather than sampling error. Assuming that the true effects are normally distributed, we can estimate that the prediction interval is -0.930 to 2.548. The small number of studies may limit the reliability of the analysis.

Adaptive SIRT1 response to exercise

A limited number of studies showed an increase in SIRT1 levels in blood (measured via ELISA) and increased SIRT1 activity in muscle (measured via fluorometric or sirtuin activity assay) after exercise training. The small number of studies may limit the reliability of the meta-analysis. Meanwhile, SIRT1 protein content in skeletal muscle (measured via Western blot) did not reach statistical significance. Figures 4 and 5 summarise the studies.

Figure 4
figure 4

Forest plot quantifying SIRT1 levels in blood (measured via ELISA) after exercise training. Adjusted standardised mean difference (Hedges’ g), relative weight of each acute study response, confidence interval (diamond), and prediction interval (blue line) are also shown. n = sample size.

Full size image
Figure 5
figure 5

Forest plot quantifying SIRT1 protein content (Western blot) in skeletal muscle after exercise training. Adjusted standardised mean difference (Hedges’ g), relative weight of each acute response study, confidence interval (diamond), and prediction interval (blue line) are also shown. C cycling, SIT sprint interval training. Percentages denote exercise intensity. n = sample size.

Full size image

Blood SIRT1 levels after training

The analysis is based on seven studies (n = 162 participants) and utilised a random-effects model. The mean effect size adjusted with Hedges’ g is 1.226 with a 95% confidence interval of 0.557 to 1.895. The Z-value is 3.592 with p < 0.001, while the Q-value is 39.873 with 6 degrees of freedom and p < 0.001. The I-squared statistic is 85%, which suggests that some 85% of the variance in observed effects reflects variance in true effects rather than sampling error. Assuming that the true effects are normally distributed, we can estimate that the prediction interval is − 0.894 to 3.345.

With analysing resistance training exercise studies alone (4 studies, n = 131 participants), the mean effect size adjusted with Hedges’ g is 0.987 with a 95% confidence interval of 0.001 to 1.973. The Z-value is 1.962 with p = 0.05, while the Q-value is 29.474 with 3 degrees of freedom and p < 0.001. The I-squared statistic is 90%, and the prediction interval is − 3.221 to 5.195.

Muscle SIRT1 activity after training

There are only three studies (N = 32 participants) quantifying SIRT1 activity in muscle (via fluorometric or a sirtuin activity assay) after exercise training. The mean effect size adjusted with Hedges’ g is 1.476 with a 95% confidence interval of 0.464–2.487. The Z-value is 2.860 with p = 0.004, while the Q-value is 7.262 with 2 degrees of freedom and p = 0.026. The I-squared statistic is 72%, and the prediction interval is -9.990 to 12.941. The small number of studies may exaggerate the range of prediction interval.

Muscle SIRT1 content after training

The analysis is based on ten studies and utilised a random-effects model. The mean effect size adjusted with Hedges’ g is 0.642 with a 95% confidence interval of − 0.128 to 1.412. The Z-value is 1.634 with p = 0.102, hence we accept the null hypothesis. Nine out of 10 studies used high-intensity aerobic training, while one used resistance training. Analysing high-intensity aerobic training studies alone (9 studies) does not reach statistical threshold.

Discussion

The main purpose of this exploratory review was to summarise and systematically assess published exercise interventions measuring SIRT1 (protein content, gene expression, enzyme levels, enzyme activity) in apparently healthy participants. Specifically, we wanted to know if and what type of exercise can increase SIRT1. A single bout of high-intensity or fasted exercise was shown to increase skeletal muscle SIRT1 gene expression as measured by qPCR or RT-PCR, while repeated exercise training enhances blood SIRT1 levels as measured by ELISA. A limited number of studies (3, N = 32 participants) also observed an increase in muscle SIRT1 activity following exercise training. Overall, we determine that exercise acutely upregulates muscle SIRT1 gene expression and chronically increases blood enzyme concentration. To our knowledge, this is the first systematic review and meta-analysis on SIRT1 response to exercise. Based on data from 34 studies (13 acute response, 21 training interventions), we highlight several novel outcomes.

Firstly, high-intensity or fasted exercise immediately upregulates SIRT1 gene expression in human skeletal muscle. The said studies measured SIRT1 expression using qPCR or RT-PCR from 0 to 3 h after a single bout of exercise. Immediate SIRT1 upregulation with PGC-1α deacetylation through protein kinase A (PKA) has been shown in C2C12 myotubes following adrenergic administration57. Although not in an exercise context, this demonstrates that hormonal signals may directly and rapidly modulate SIRT1 deacetylase activity, which can subsequently be sustained through an elevation in AMPK-dependent NAD+58. Cardiac muscle contraction and catecholamines, which are released during intense exercise, can activate PKA, perhaps explaining the immediate SIRT1 activation in high-intensity exercise59. In addition, PKA is activated when plasma glucose concentration falls below 65–70 mg/dL, which may explain the immediate increase in SIRT1 while in the fasted exercise state60.

SIRT1 levels increase with exercise intensity26 but only up to a certain point. One study found that supramaximal exercise (133% peak work rate) resulted in less expression of the SIRT1 target PGC-1α but a larger expression in early growth response 1 (EGR1), which is also known to induce SIRT1 as a response to mechanical stretch in skeletal muscle28. This stretch-induced activation of SIRT1 leads to deacetylation of FOXO4 and upregulation of superoxide dismutase which scavenges excess ROS61. EGR1 and SIRT1 form a negative-feedback loop, which may explain the decreased SIRT1 expression at supramaximal exercise61. Supramaximal exercise also increases DNA damage, which recruits PARPs that use NAD and compete with SIRT122.

High-intensity exercise increases muscle fibre recruitment and ATP turnover, which activates AMPK and SIRT1 activity; the latter by increasing cellular NAD+61. In support, a study in mice observed AMPK-dependent increases in NAD+ and PGC-1α deacetylation 3 h after resistance running, however, SIRT1 was not directly measured62. In young and fit individuals, but not in old and unfit, a single bout of cycling (20 min at 70% VO2max) can release extracellular NAMPT (eNAMPT) into the circulation, which increases NAD and SIRT1 within 1 h when administered to skeletal muscle cells63. Interestingly, injecting eNAMPT into muscle cells from young mice can prolong health and lifespan in old mice64.

Two studies have examined blood SIRT1 levels (as measured via ELISA) after a single bout of exercise (moderate intensity and high-intensity) and both observed an increase following exercise26,29. An additional investigation demonstrated that blood SIRT1 activity (fluorometric assay) is also enhanced following a single bout of exhaustive cycling35. With regards to repetitive exercise training, resting blood SIRT1 levels (ELISA) has also been shown to increase; albeit based on a moderate number of studies (7, n = 162 participants). The aforementioned studies involved resistance training and high-intensity aerobic training, which can both improve body composition, a factor that has been associated with increased circulating SIRT1 in larger long-term studies18,65. Circulating SIRT1 is negatively associated with fat mass, leptin, and insulin resistance, and positively associated with adiponectin, an “anti-obesity” hormone66,67. Interestingly, leptin and adiponectin are regulated partially by SIRT1 through PPAR-α and the circadian cycle8,68.

Low circulating SIRT1 is also associated with ageing-related disorders and has been proposed as a biomarker for Alzheimer’s and Parkinson’s disease69,70. In senior participants, two studies measured blood SIRT1 levels after resistance training, while another measured blood SIRT1 activity after resistance training; all demonstrating an increase post training47,55,56. Acutely, high-intensity aerobic exercise can also increase skeletal muscle SIRT1 expression in seniors36.

SIRT1 levels in skeletal muscle, however, may not always increase after exercise training. It is also possible that SIRT1 muscle protein content does not increase until after several months of training, and several of the observed studies implemented less than 1 month of training. Several post-translational protein modifications may also regulate skeletal muscle SIRT1 adaptation to exercise. One of the studies found increased NAD, NAMPT, global sirtuin activity, and mitochondrial density, but no increase in skeletal muscle SIRT1 protein content after 10 weeks of resistance training, suggesting that it may be NAD or NAMPT levels, rather than SIRT1 protein content, that contributes to increased SIRT1 activity49.

There are only three published studies that quantified skeletal muscle SIRT1 activity through a fluorometric or sirtuin activity assay-based approach after exercise training. Although the meta-analysis reached statistical significance, it is limited by the small sample (n = 32). Only one study measured SIRT1 activity in blood, with significant effects after resistance training observed56. Muscle SIRT1 activity, rather than content, has been associated with mitochondrial biogenesis during exercise46. Meanwhile, blood SIRT1 activity has been positively associated with basal metabolic rate71. Thus, SIRT1 activity may be more responsive to exercise training. It is therefore suggested that any future studies in this exciting domain of investigation should consider measuring SIRT1 activity as well as expression.

Most studies documented in our analysis utilized high-intensity aerobic exercise protocols and young participants, with several using resistance training in aged and overweight/obese participants. A genuine attempt was made for this review to be exhaustive and group all eligible studies according to biomarker, exercise protocol, and participant characteristics. However, there were limited studies on old and overweight/obese groups to reach statistical significance. Hence, all age and weight groups (young, old, overweight, obese, normal) as well as training levels (trained, untrained, athletes, sedentary) were combined. We are aware that ageing and overweight participants may have heightened inflammatory profiles that may affect the exercise response. Moreover, athletes may have a blunted response to exercise due to their enhanced fitness. It is also conceivable that differences due to sex in the exercise-induced SIRT1 response may also exist, particularly aligned to muscle fiber type composition, given that SIRT1 expression is higher in female Type I muscle fibers.

One major limitation of this study is the unclear risk of bias in exercise related studies due to failure of reporting randomisation, concealment, and blinding protocols72. We used the original 11-item Cochrane Back Group Risk of Bias Tool that was partially utilized in a meta-analysis on exercise and DNA damage22. A subsequent validity analysis of the original tool determined that a summary score of 6 is the threshold for the separation of high-quality from low-quality studies23. Hence, and in line with a previous study, lower quality is classified with a score of 0–5, while the 6–11 range confirms higher quality studies24. Studies scoring 6 or higher are thus within the quality threshold. It is however, important to note, that a number of items in the current Cochrane tools repertoire are nearly impossible to implement in exercise focused studies, such as blinding and randomisation (e.g., in a repeated measures design, acute response studies where pre and post-exercise cannot be interchanged). As such, this can be interpretated that in fact all exercise studies are inherently at risk for performance bias. Comprehensive and recent work by Bonafiglia et al. suggests that given the apparent and perceived methodological issues when conducting a meta-analysis, future studies are encouraged to implement bias-reducing methodologies, such as the one used in the current work, and indeed the authors report on the various approaches that should be considered when attempting to mitigate performance bias in exercise studies72.

Another limitation of this study is that most data were extracted from figures rather than raw values using a graph digitizer software. However, to minimise error, highly accurate software was used based on other published work73.

We conclude that high-intensity exercise has a relatively small effect in acutely increasing SIRT1 gene expression in skeletal muscle, while the effect observed in fasted exercise appears to be larger. Exercise training in general, or resistance training alone, has a large effect on resting blood SIRT1 levels, while a limited number of studies provide an evidence base for an increase in resting SIRT1 activity in skeletal muscle following exercise training. Taken together, these results reiterate the potential of exercise in prolonging health, partly due to upregulating SIRT1 acutely and chronically. Training variables such as intensity, adding resistance, or feeding status can be adjusted to maximise the health benefits of exercise. More studies examining SIRT1 activity following exercise and incorporating aged and overweight/obese populations are warranted.

Methods

Search strategy

Following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines, a comprehensive article search was conducted using the keywords “exercise AND SIRT1” from March 1–31, 2022, using five databases: Scopus, PubMed, Embase, MEDLINE, and Web of Science. Search results were filtered to include only human trials, in peer-reviewed journals, and in English language. New studies published after the said dates were also added. Our search protocol was registered and published on PROSPERO (CRD42023427141).

Inclusion/exclusion criteria

Articles were checked for the following inclusion criteria: (1) full report published in a peer-reviewed journal, (2) involving healthy adults, (3) controlled trial, and (4) with measures of SIRT1 (gene expression, protein content, enzyme levels, enzyme activity) in blood or skeletal muscle before and after exercise in acute response studies, or before and after training in intervention studies.

Studies with independent variables other than exercise (e.g., supplementation, diet, etc.) were included providing they consisted of a control group that received exercise alone, and only data from exercise-only groups were included in the meta-analysis.

“Training intervention studies” were those that implemented repeated exercise for a minimum of 2 weeks, which is the shortest intervention among the included studies. Meanwhile, “acute response studies” were those that conducted a single bout of exercise. “Fed state” suggests normal diet with the recommended macronutrient ratio (45–65% carbohydrates, 20–35% fats, 10–35% protein), while “fasted state” suggests 8–12 h overnight fasting which is a normal practice in exercise related studies. As such, data from participants doing more than 12 h fasting and special diets (e.g., vegan diet, high-carbohydrate diet, etc.) were excluded. Retracted articles were also excluded. The inclusion/exclusion criteria are shown in Table 4.

Table 4 Inclusion/exclusion criteria.
Full size table

Data extraction

Articles that met the inclusion criteria were printed and summarised in a table following PICO (Patient, Intervention, Comparison, Outcome) guidelines and were grouped according to tissue sample (muscle vs blood), biomarkers of SIRT1 (gene expression, protein content, enzyme levels, enzyme activity), participants’ nutritional status (fed vs fasted), exercise type and intensity (see Table 1). In case of missing data (e.g., nutritional status of participants, average intensity of exercise protocol), authors were contacted via email to clarify information. Means, standard deviations (SD), and standard errors (SE) were extracted from full texts and figures using GetData Graph Digitizer software73.

Quality assessment

Primary outcomes were defined as pre- and post-exercise SIRT1 measures (gene expression, protein content, enzyme levels, or enzyme activity). Risk of bias was assessed using the original 11-item Cochrane Back Group Risk of Bias Tool, namely: randomisation, concealment, baseline differences, patient blinding, care provider blinding, outcome blinding, co-intervention, compliance, dropouts, timing, and intention to treat (i.e., whether all participants were included in the analysis regardless of compliance). This approach was based on a related meta-analysis on exercise-induced DNA damage in healthy participants22. Studies were scored from 1 to 11, with a score of 6 distinguishing low from high quality studies; aligned to a validity study by Van Tulder et al.23. In practice, Paige et al. considered a score of 0–5 as “lower quality”, while a score of 6–11 was considered “higher quality”24. Table 5 summarises the quality assessment for risk of bias.

Table 5 Quality assessment for risk of bias using the Cochrane back group risk of bias tool.
Full size table

Statistical analysis

Assessment of effect size

The Comprehensive Meta-Analysis (Version 4, NJ: USA: Biostat, Inc.) software was used to calculate random effects from means, standard deviations, and standard errors extracted from each article. Standardized mean difference (SMD) adjusted with Hedges’ g at 95% CI was calculated as the difference in means before and after exercise divided by the pooled standard deviation. SMD was used to express effect size, which was assessed using Cohen’s categories: 0.2–0.5 = small, 0.5–0.8 = medium, and > 0.8 = large. The overall effect size was assessed using Z-values with a significance level of p < 0.05.

Assessment of heterogeneity

The Q-statistic was used to test the null hypothesis that the true effect size is the same in all these studies, where p value ≤ 0.10 was considered significant heterogeneity. The I-squared (I2) statistic was used to express the proportion of the variance in observed effects that reflects variance in true effects rather than sampling error74.

Publication bias

Publication bias was assessed by visually analysing funnel plots, with a caveat that funnel plots may not be appropriate for small studies75. Both observed values and imputed values were plotted whereby imputed studies are perceived to be primarily negative in origin, while actual studies are regarded as being more positive (see Supplementary Materials S1).