Coenzyme Q10 to manage chronic heart failure with a reduced ejection fraction: a systematic review and economic evaluation

Population

Trials of adult patients (aged > 18 years) with a diagnosis of HFrEF were included. Paediatric trials were excluded. Mixed population trials were eligible if data from relevant individuals (adults with CHF) were available separately. As patients with a diagnosis of heart failure with a preserved ejection fraction (HFpEF) are a clinically distinct population, trials focusing on this population were excluded. One included trial33 included a small proportion of HFpEF individuals (7% of participants had LVEF ≥ 45%) and it is possible that other trials that did not report LVEF had included some patients with HFpEF.

Intervention

Trials of co-Q10 (taken singly or as part of a multi-micronutrient supplement) given as an adjunct to co-treatment (e.g. statins) or other routine care.

Comparators

Placebo given as an adjunct to co-treatment (e.g. statins) or other routine care.

Outcomes

Intended outcomes included the following:

• All-cause mortality (ACM) and cardiovascular mortality (time to event) [i.e. death from myocardial infarction (MI), stroke, heart failure or sudden cardiac death].

• Major cardiovascular events (time to first event) (i.e. non-fatal MI, non-fatal stroke and revascularisation procedures).

• Hospitalisation related to heart conditions (i.e. any, number of and duration of stays).

• Any cardiovascular event, as above, death or any hospitalisation (composite outcome).

• QoL measures using a validated instrument [e.g. the EuroQol-5 Dimensions (EQ-5D)].

• NYHA functional class (or equivalent).

• Adverse effects/side effects.

• LVEF, which is the volumetric fraction of fluid blood ejected from the left ventricle of the heart with each contraction.

• Exercise testing [e.g. change in 6-minute walk test (6MWT)] over a defined period.

• B-type natriuretic peptide (BNP)/N-terminal pro-B-type natriuretic peptide (NT-pro-BNP) level.

• Peak oxygen consumption.

  1. The NYHA classification is a simplified scale that classifies heart failure severity. It classifies patients into one of four categories based on their limitations during physical activity, including limitations to normal breathing, varying degrees of shortness of breath and/or angina pain. A higher score is associated with worse symptoms. The classification is as follows: class I – no symptoms and no limitation in ordinary physical activity (e.g. shortness of breath when walking); class II – mild symptoms (e.g. mild shortness of breath and/or angina) and slight limitation during ordinary activity; class III – marked limitation in activity due to symptoms, even during less than ordinary activity [e.g. walking short distances (20–100 m)], comfortable only at rest; and class IV – severe limitations, experiences symptoms even while at rest, mostly bedbound patients.

  2. LVEF is calculated as a ratio of the volume of blood pumped from the left ventricle per beat (stroke volume) to the volume of blood collected in the left ventricle at the end of diastolic filling (end-diastolic volume). The American College of Cardiology recommends the following classification of systolic function: normal – LVEF 50–70%; mild dysfunction – LVEF 40–49%; moderate dysfunction – LVEF 30–39%; and severe dysfunction – LVEF < 30%.

  3. In CHF, BNP and its amino-terminal NT pro-hormone BNP are released into circulation directly from the myocardium following end-diastolic stress as a result of increases in volume or pressure. Measurements of BNP or NT-pro-BNP are used in establishing diagnosis, prognosis and disease severity in CHF in current guidelines, with higher levels indicating poorer prognosis. 39,40 However, it has been suggested that BNP and NT-pro-BNP measurements may not be reliable end points because within-patient changes are not a consistent surrogate of changed prognosis either in studies or in clinical practice. 41

Study design

Only RCTs were eligible. Both parallel-group and crossover trial designs were included.

Outcomes analysed

The full set of protocol intended outcomes and their definitions are given in Table 1. Those outcomes that were reported in trial publications and able to be analysed are given in Table 4.

Effect measures

Dichotomous outcomes included in the meta-analyses were analysed by calculating the relative risk (RR) for the effect of co-Q10 compared with placebo. For continuous outcomes, mean differences (MDs) between treatment arms were analysed. Alternatively, standardised mean differences (SMDs) were used for trials that had different measurement scales for the same outcome. Analyses were performed in terms of change from baseline values and, as a sensitivity analysis, using only final outcomes, without adjusting for baseline values.

Hazard ratios were calculated for time-to-event outcomes, either as reported or using data extracted from survival curves.

Some outcomes were analysed in multiple ways. Death and cardiovascular outcomes were analysed as both dichotomous outcomes and survival outcomes (where survival data were available). NYHA data were dichotomised (classes III and IV vs. classes I and II) and also analysed in terms of improvement by one or more categories. RRs and 95% CIs were calculated and reported for all effect estimates included in the meta-analyses.

Trials supplying individual participant data

As the majority of included trial investigators either could not or declined to supply IPD, analyses were based on data reported in their trial publications. Investigators for two trials supplied IPD. For these two trials, summary results (numbers of patients and events, and means and standard deviations by treatment arm) were calculated using the IPD and these summary results were then pooled in meta-analyses alongside data extracted from the publications for which IPD were unavailable.

Crossover and parallel-group trials

The protocol-intended approach was to analyse crossover and parallel-group trials separately, as the two trial designs may not give comparable results. However, because of limited outcome reporting in the crossover trials, this was not feasible for all outcomes. For LVEF, crossover and parallel-group trials were analysed separately and subsequently in combination. For all other outcomes, crossover and parallel-group trials were combined and meta-analysed together, provided that the two trial designs produced broadly consistent results.

Standard meta-analysis

Initial analyses estimated the effect (RR, MD or HR) for each outcome reported in each trial (or provided as IPD). A map of the data was produced to identify the number of trials and participants for each outcome to identify where meta-analysis was feasible. Meta-analyses were performed where two or more trials reporting the outcome under consideration were available. When only one trial was available, its results were reported narratively.

Effect estimates were combined in inverse-variance random-effects meta-analyses using the standard DerSimonian–Laird two-stage approach. This generated forest plots, enabling results across trials to be compared visually, heterogeneity to be investigated and differences across subgroups to be visualised. Heterogeneity was quantified using the I²-statistic. Where there were sufficient trials, the potential for publication bias was assessed using contour-enhanced forest plots.

One-stage regression analyses

Although more commonly associated with IPD meta-analyses, ‘one-stage’ meta-analyses that combine data from all trials in a single regression model to estimate the overall effect, rather than estimating an effect in each trial and then pooling across trials, can be carried out using either IPD or aggregate data. 44,45 For example, for aggregate dichotomous data, this is carried out using the numbers of events and number of patients in each arm of each trial. This approach was preferred to conventional meta-analysis, as it uses an exact likelihood and so may be more robust where data are sparse. In this review, one-stage analysis of the available aggregate data was possible for the outcomes of ACM and NYHA class only. These used restricted maximum likelihood methods and regressed outcome against treatment, with correlated random intercept and random treatment effects (to account for heterogeneity). Model parameters for treatment effect were then converted into RR estimates, with associated 95% CIs.

As for two-stage analyses, meta-analyses were performed if at least two trials reported data for the specified outcome.

Potential effect modifiers (subgroups and meta-regression)

The impact of trial- and patient-level characteristics on treatment effect (i.e. treatment–covariate interactions) was examined.

For categorical covariates, the trials were divided into groups according to the characteristic and meta-analyses were performed within each subgroup. Meta-regression was used for continuous covariates (e.g. co-Q10 dose). The covariates considered were:

• trials specifically comparing co-Q10 plus statin with statin alone/other trials

• single- or multi-micronutrient supplement

• parallel or crossover design

• co-Q10 dose

• duration of treatment/trial

• percentage of patients receiving statins

• mean baseline value of outcome

• year of publication.

For dichotomous outcomes with sufficient data, one-stage meta-regression models were also fitted. To do this, the one-stage regression models (see One-stage regression analyses) were extended to include one parameter for the covariate of interest and one for the treatment–covariate interaction. To ensure model convergence, these parameters were assumed to be common to all trials (i.e. no random effects).

Network meta–analysis

A network meta-analysis (NMA) compared single- and multi-micronutrient supplements containing co-Q10 with co-Q10 alone and with co-Q10 in combination with statins or other concomitant treatments. Analyses were carried out for the main outcomes listed earlier (see Outcomes analysed) where sufficient trials reported that outcome. Two statistical models were used. The first was the Bayesian models of Lu and Ades,46 which are the most commonly used methods for NMA. The second used a frequentist ‘one-stage’ logistic or linear regression, including multiple treatment arms.

Software

All analyses were conducted in the R software package (version 4.0; The R Foundation for Statistical Computing, Vienna, Austria). The meta package was used for meta-analyses, forest plots and funnel plots, and the lme4 package was used for one-stage models. For the NMA, OpenBUGS 3.2.3 and the GeMTC and BRugs packages in R were used (MRC Biostatistics Unit, University of Cambridge, Cambridge, UK; URL: www.mrc-bsu.cam.ac.uk/software/bugs/openbugs/).

All-cause mortality

Eleven trials33,48,54,55,57,59,62,67,69,70 (involving 1589 randomised patients) reported data on ACM. Three of these trials54,67,69 reported/included no deaths and one crossover trial52 reported insufficient data for inclusion in the meta-analysis. Therefore, seven trials were included. 33,48,55,57,59,62,70

The one-stage meta-analysis of ACM, comparing co-Q10 with control, had a RR of 0.68 (95% CI 0.45 to 1.03), suggesting that co-Q10 reduced the risk of mortality, but this was not quite statistically significant. There was evidence of modest heterogeneity (τ² = 0.23) across trials.

For comparison, Figure 2 shows the results of the standard DerSimonian–Laird random-effects meta-analysis for ACM. This found a substantial benefit of co-Q10, with a larger reduction in the risk of death and narrower CIs (RR 0.62, 95% CI 0.45 to 0.85). There was no evidence of any heterogeneity across trials (I² = 0). This is broadly similar to the one-stage analysis. However, the one-stage analysis identified modest heterogeneity (τ² = 0.23), which meant that the CI for the one-stage analysis results did, just, include the null value and also led to the slightly different RR estimate. The one-stage analysis was preferred because it is likely to be more robust, given the small number of deaths in several trials, and so its larger estimate of heterogeneity is more plausible.

FIGURE 2.

Meta-analysis of ACM.

A sensitivity analysis that added the Alehagen et al. trial71 to the meta-analysis slightly reduced the estimated treatment benefit (RR 0.67, 95% CI 0.51 to 0.87).

One crossover trial52 found no significant difference in deaths between treatment and placebo: four deaths occurred during the placebo and three deaths occurred during the co-Q10 treatment periods (total incidence = 8.9%). The trial52 did not report whether these events occurred before or after crossover and so was not included in the meta-analysis.

Figure 3 is a contour-enhanced funnel plot for ACM. The figure shows no indication of reporting bias or small-study effects. This was, however, based on limited numbers of trials.

FIGURE 3.

Funnel plot for ACM.

Cardiovascular mortality

Two trials33,55 (involving 472 participants) reported data on cardiovascular mortality, but the data were insufficient for meta-analysis.

The Q-SYMBIO trial33 found a substantial benefit of co-Q10, with a large reduction in cardiovascular mortality (RR 0.57, 95% CI 0.33 to 0.98). The Q-SYMBIO trial33 also reported a higher incidence of death from heart failure in the placebo arm (4.6%) than in the co-Q10 arm (0.5%) (RR 0.11, 95% CI 0.01 to 0.84). This appears to be consistent with the findings for ACM (see Figure 2).

Khatta et al. 55 recorded one death from heart failure in the placebo arm and no deaths in the co-Q10 arm.

The excluded Alehagen et al. trial71 also found a substantial benefit of co-Q10 for reducing cardiovascular mortality (RR 0.47, 95% CI 0.25 to 0.88).

Non-fatal cardiovascular events

No studies reported non-fatal cardiovascular events separately from cardiovascular disease-related deaths.

Hospitalisation (due to chronic heart failure)

Three studies33,54,62 (involving 1100 participants) reported data on hospitalisation due to CHF, but one54 recorded no events. The meta-analysis of risk of hospitalisation for the remaining two studies33,62 showed a substantial benefit of co-Q10, with a large reduction in hospitalisation (RR 0.61, 95% CI 0.48 to 0.77) (Figure 4).

FIGURE 4.

Meta-analysis of hospitalisation due to CHF.

New York Heart Association functional class

Thirteen trials33,48,49,51,54,55,57,60,61,64,65,67,69 (involving 1038 participants) reported NYHA class as an outcome, of which seven33,48,51,54,64,67,69 provided sufficient data for meta-analysis. Two trials64,69 provided full IPD.

The NYHA is a symptoms scale that classifies patients depending on their limitations and symptoms (i.e. breathing, shortness of breath or angina pain) during physical activity. Classification ranges from I (no symptoms and no limitation in ordinary physical activity) to IV (severe limitations, symptomatic even at rest).

As NYHA is a categorical outcome, it would ideally be analysed as such. However, no trial reported data to permit this (i.e. trials did not provide the numbers of patients in each NYHA class). Trials mostly reported NYHA as a continuous variable (mean value in each trial arm) or in terms of numbers in NYHA class III or IV. Meta-analyses of these outcomes are presented here, but we note that these are not ideal representations of NYHA data.

Figure 5 shows the MD between arms in the change in NYHA class from baseline. This analysis treats the four-category NYHA as if it were a continuous variable and may not, therefore, be a reliable indicator of the effect of co-Q10. The results favour co-Q10, suggesting that it lowers NYHA class by approximately half of a class on average (MD 0.50, 95% CI 0.39 to 0.62). The results were homogeneous (I² = 0).

FIGURE 5.

Meta-analysis of change from baseline in NYHA class.

Adding the excluded Alehagen et al. trial71 to this analysis gave a smaller treatment benefit (MD 0.37, 95% CI 0.07 to 0.67) and its addition resulted in substantial heterogeneity (I² = 79%).

Figure 6 shows the RR of being in NYHA class III or IV after intervention. This meta-analysis could be biased if NYHA classes were not balanced across arms at the start of each trial; however, we did not find any evidence of such bias. The results suggest a large treatment benefit (RR 0.37, 95% CI 0.19 to 0.73; I² = 0). This outcome was also analysed using a one-stage model. The RR of being in NYHA class III or IV was 0.37 (95% CI 0.17 to 0.81), which is almost identical to the two-stage results (see Figure 6), but with wider CIs.

FIGURE 6.

Meta-analysis of being in NYHA class III or IV after intervention.

The results for improvement by one or more class in NYHA, which was reported in three trials,33,64,69 are shown in Figure 7. Note that, because this is measuring improvement, a RR of > 1 suggests a benefit from co-Q10 (greater improvement). This suggests that co-Q10 confers a modest but uncertain benefit, a finding dominated by the results of the Q-SYMBIO trial33 (RR 1.19, 95% CI 0.93 to 1.52). The results for deterioration of NYHA class (see Appendix 5, Figure 22) also found a possible modest, but highly uncertain, benefit from co-Q10 (RR 0.76, 95% CI 0.15 to 4.0). These results are based on very small numbers (just 14 patients out of 463 had a deterioration in NYHA class).

FIGURE 7.

Meta-analysis of improvement in NYHA class.

Four parallel trials49,55,57,60 could not be included in the meta-analyses because of insufficient data. Davini et al. 49 found that the percentage of patients with an improved NYHA class at 4 months was greater with co-Q10 than with placebo for patients with dilated cardiomyopathy (87% vs. 43%) and valvular disease (87% vs. 29%), but smaller for patients with ischaemic heart disease (5% vs. 27%). Khatta et al. 55 reported that only 1 of 23 patients in each of the co-Q10 and placebo arms had improved NYHA class (data from publication only; no further details reported). Kumar et al. 57 reported a greater reduction in the percentage of participants with NYHA II–IV in the co-Q10 arm (from 100% at baseline to 62% after 12 weeks) than in the control arm (from 100% to 86.2%). Mareev et al. 60 reported that the change in NYHA from baseline to 24 weeks was greater in the co-Q10 group (–0.16) than in the placebo group (–0.08).

Two crossover trials61,65 reported data on NYHA class. One trial61 reported an overall reduction in the percentage of NYHA class III patients (from 25% to 5%) at 8 weeks’ follow-up from baseline, but another trial65 found no statistically significant changes in mean NYHA functional class with co-Q10 compared with placebo at 4 months’ follow-up. Further details are reported in Appendix 4, Tables 34 and 35.

Left ventricular ejection fraction

Sixteen studies29,33,47,51,52,55,56,58,60,63,65–70 (involving 1318 participants) reported/recorded data on LVEF, but four studies29,52,58,65 did not report sufficient data to be included in any meta-analysis.

Figure 8 presents the results of the meta-analysis of the change in LVEF from baseline and shows the mean difference between the co-Q10 and control arms for each study. Substantial increases in ejection fraction percentage may indicate improved condition. These results suggest a modest benefit from co-Q10 (MD 1.76%, 95% CI 0.21% to 3.31%). The trials appear to be homogeneous (I² = 0), although two trials have mean effects in the direction of harm. 55,63

FIGURE 8.

Meta-analysis of change from baseline in LVEF.

Three47,66,68 of the five crossover trials that reported LVEF supplied enough data to be incorporated into the meta-analysis. The results are shown in Figure 9. Although the crossover trials gave a slightly larger benefit from co-Q10 (MD 2.65%, 95% CI –0.73% to 6.02%), they were consistent with the parallel-group trials, with no evidence of heterogeneity. The meta-analysis combining all trials showed a modest benefit from co-Q10 (MD 1.90%, 95% CI 0.58% to 3.21%).

FIGURE 9.

Meta-analysis of change from baseline in ejection fraction percentage, including all trials.

Four studies29,52,58,65 could not be included in the meta-analysis because of insufficient data (i.e. a crossover trial with results grouped across randomised sequences)29,52,58,65 and unextractable figures. 58,65 Only one of these trials58 reported a significant improvement in LVEF from baseline that favoured co-Q10.

Of the remaining crossover trials that could not be included in the meta-analysis reported in Figure 9, two studies52,65 found no significant change from baseline to follow-up in percentage LVEF across all patients, one study52 reported a small statistically non-significant and clinically irrelevant difference (MD 0.5%, 95% CI –1% to 2%) and the other study65 did not provide any further details. One crossover trial58 found a significantly greater increase in LVEF with co-Q10 than with placebo (p < 0.0001; no further extractable details), whereas another trial29 reported improved mean percentage LVEF from baseline at 8 weeks’ follow-up across randomised sequences [from 27% (SD 11%) to 33% (SD 13%)], but did not report whether or not any differences had been observed between co-Q10 and placebo.

Figure 10 provides a contour-enhanced funnel plot for LVEF. The figure shows no indication of reporting bias or small-study effects. This was, however, based on limited numbers of trials.

FIGURE 10.

Funnel plots for LVEF change from baseline.

6-minute walk test

Six trials33,48,54,57,60,69 (involving 728 participants) reported sufficient data on 6MWT results to be included in a meta-analysis. Figure 11 presents the 6MWT results expressed as mean change from baseline in number of metres walked in 6 minutes, and meta-analyses the MD in this between co-Q10 and control arms. The results show a benefit from co-Q10 in increasing walking distance from baseline, but this is strongly influenced by the Berman et al. 48 trial, which is a substantial outlier. It is unclear why the Berman et al. 48 trial is so inconsistent with other trials. Meta-regression analyses [see Meta-regression (trial- and participant-level factors)] did not find any evidence that trial properties or baseline 6MWT values might be the cause of this inconsistency.

FIGURE 11.

Meta-analysis of 6MWT.

Removing the Berman et al. 48 trial produced a more modest benefit from co-Q10 (MD 19.89 metres, 95% CI 1.87 to 37.90 metres). However, the trials were still very heterogeneous (I² = 76%), with the largest trial (the Q-SYMBIO trial73) showing no benefit from co-Q10.

Peak oxygen consumption

Three trials47,50,57 (including one crossover trial47) reported enough data for a meta-analysis of peak oxygen consumption (Figure 12). The results are expressed as the difference between mean change from baseline (ml/kg/minute) between co-Q10 and control arms. The meta-analysis showed no significant benefit from co-Q10 in improving peak oxygen consumption (MD 0.72 ml/kg/minute, 95% CI –0.60 to 2.05 ml/kg/minute).

FIGURE 12.

Meta-analysis of peak oxygen consumption.

Removing the one crossover trial47 slightly reduced the observed effect (MD 0.54 ml/kg/minute, 95% CI –0.89 to 1.96 ml/kg/minute).

B-type natriuretic peptide and N-terminal pro-B-type natriuretic peptide

Three trials33,56,67 (involving 602 participants) reported data for NT-pro-BNP and two trials53,60 (involving 168 participants) reported data for BNP. Two of these trials53,56 could not be included in the meta-analyses because of insufficient data; one56 was a conference abstract and did not report sufficient statistical information, and the other53 reported data only as median values. The three other trials33,60,67 (involving 630 participants) were combined in a meta-analysis (Figure 13). As the analyses combined BNP and NT-pro-BNP, studies were pooled using SMDs. Earlier follow-up (4 months rather than 1 year) was used from one study33 to align it with the follow-up of the other trials included in the meta-analysis. Higher levels of BNP and NT-pro-BNP may indicate more severe CHF. Overall, there was no evidence that co-Q10 reduced BNP or NT-pro-BNP levels (SMD 0.56, 95% CI –0.5 to 1.62). The results were not statistically significant, were highly heterogeneous (I² = 97%) and were unlikely to be of clinical relevance.

FIGURE 13.

Meta-analysis of BNP and pro-BNP.

One trial,56 reported as a conference abstract, noted statistically significant changes in NT-pro-BNP levels from baseline in both co-Q10 and placebo groups. NT-pro-BNP in the co-Q10 arm was 490.7 pg/ml at baseline and 134.6 pg/ml (p < 0.05) at 3 months, although it is unclear whether these numbers were medians or means. In the placebo group, reported levels decreased significantly, from 701.3 (95% CI 271.4 to 1385.5) pg/ml at baseline to 230.8 (95% CI 178.9 to 443.4) pg/ml at follow-up (p < 0.05).

One crossover trial53 reported no significant change in median BNP levels with co-Q10 [117 pg/ml, interquartile range (IQR) 69–168 pg/ml pre co-Q10; 152 pg/ml, IQR 83–266 pg/ml post co-Q10 (p = 0.37)] or with placebo [91 pg/ml, IQR 55–165 pg/ml pre placebo; 137 pg/ml, IQR 48–291 pg/ml post placebo (p = 0.48)], and no significant difference in median BNP levels between co-Q10 and placebo.

Quality of life

Quality-of-life measures were reported in five studies50,52,60,68,69 (involving 460 participants), including three parallel-group50,60,69 and two crossover trials. 52,68 Of these trials, one parallel trial60 was reported only as conference abstract, with insufficient data for inclusion in the meta-analysis.

Different scales were used to assess QoL. Fumagalli et al. 50 used the Sickness Impact Profile, which ranges from 0% to 100%, where 0% represents a completely healthy patient and 100% represents a patient completely dependent on another person in all aspects of life. 74 Poggesi et al. 66 used the Minnesota Living With Heart Failure Questionnaire (MLWHFQ), which ranges from 0 to 105, with higher scores indicating more significant impairment in health-related quality of life (HRQoL). Witte et al. 69 applied the EuroQol Heart Failure Scale, which ranges from < 0 to 1, with higher scores indicating higher health utility. 75 Hofman-Bang et al. 52 used a tool developed by the trial centre,76 with higher scores indicating improved QoL. Finally, Mareev et al. 60 used the Scale for Heart failure to Optimise Clinical Status (SHOCS), which ranges from 0 to 20, with higher scores indicating worsening clinical condition.

As only two trials reported QoL at baseline, the meta-analysis of all four trials used QoL at the end of trial. SMDs were used to account for scales of different quality being used in each trial (Figure 14). Positive SMDs indicate improved QoL for participants receiving co-Q10, compared with control. The results suggest that there is no clear evidence of any effect of co-Q10 (SMD 0.10, 95% CI –0.12 to 0.33), with the overall effect size unlikely to be of clinical significance.

FIGURE 14.

Meta-analysis of QoL at the end of trial.

The trial by Mareev et al.,60 which was reported as a conference abstract, found that SHOCS score reduction from baseline to 24 weeks was greater in the co-Q10 group (–1.06) than in the placebo group (–0.53) (p = 0.036).

Adverse events

Twelve trials33,47,48,52–55,57,58,65,66,68 reported numbers of adverse events (of any type), of which six47,53,55,58,65,66 reported no events and two52,68 reported insufficient data for meta-analysis. Figure 15 shows the meta-analysis of the RR of any adverse event for the four remaining trials. 33,48,54,57 This figure provides no evidence of increased adverse events with co-Q10, that is the direction of effect is for reduced adverse events (RR 0.73, 95% CI 0.47 to 1.12), but with results driven largely by the Q-SYMBIO trial. 33 The most common adverse events reported in the Q-SYMBIO trial33 were gastrointestinal disturbances (2.4%), stroke (1.7%) and arrhythmia (1.7%). Cardiovascular procedures (i.e. percutaneous coronary intervention and coronary artery bypass grafting) were reported in 2.1% of participants, with no significant differences between the study arms.

FIGURE 15.

Meta-analysis of adverse events.

Three trials reported numbers of discontinuations due to adverse events and found no clear evidence of higher discontinuation with co-Q10 exposure (RR 1.54, 95% CI 0.32 to 7.47). However, this was based on just six events (see Appendix 5, Figure 23).

None of the trials that we were unable to include in the meta-analysis found a significant difference in adverse events between co-Q10 and control. One crossover trial52 reported 46 adverse events throughout the 6-month study period, with no significant differences between co-Q10 and placebo. Adverse events included gastrointestinal disturbances (12.7% of patients in each period), vertigo (7.6% in co-Q10 periods and 12.7% in placebo periods) and dry skin (7.6% in the co-Q10 periods and 5.1% in the placebo periods).

Two adverse events (in 20 patients) were reported in another crossover trial. 66 Epigastric burning and slight epigastralgia were recorded; however, the study did not detail whether these occurred during the co-Q10 period or the placebo period of the study.

The Alehagen et al. 71 trial, which was excluded because it included only a small proportion of co-Q10 patients, reported no significant difference in the percentage of patients with gastrointestinal symptoms or diarrhoea between the co-Q10 (4.1%) and placebo (3.2%) arms that led to study discontinuation (p = 0.60). There is no reason to expect that adverse effects would differ between CHF patients and the general elderly patients included in this trial.

Subgroups of trials (trial-level factors)

Where there were sufficient trials, subgroup analyses by intervention type [i.e. co-Q10 only, co-Q10 plus selenium, co-Q10 plus other micronutrient(s)] were performed. Given the limited data from crossover trials, all analyses were restricted to only parallel-group trials.

As there was only one trial of co-Q10 with selenium51 and only three multinutrient trials,50,57,69 subgroup analysis was possible only for ACM (see Appendix 5, Figure 24), LVEF (see Appendix 5, Figure 25) and change from baseline in NYHA class (see Appendix 5, Figure 26). In each analysis, there was only one or no selenium trials51 and only one multinutrient trial57,69 and so it is difficult to make formal comparisons. For LVEF, effect estimates were larger for the trials of co-Q10 with selenium and multinutrients, but both trials51,69 were small and the results were not visibly different from those of the co-Q10 only trials (I² = 0). For ACM and NYHA class, the results varied across intervention types, but the data were very limited and heterogeneous.

No trials explicitly combined co-Q10 with statins and so this intended subgroup analysis could not be performed.

Meta-regression (trial- and participant-level factors)

Meta-regression was performed whenever at least five parallel-group trials reported both the required outcome and the regression factor. Factors considered were:

• intended co-Q10 dose

• intended duration of treatment

• percentage of patients receiving statins

• mean baseline value of outcome

• year of publication.

Only ACM, LVEF, NYHA change from baseline and 6MWT had sufficient data for meta-regression. Table 8 summarises the meta-regression results, presenting the estimated regression parameter and the associated p-value for the regression.

There was no clear evidence that any parameter altered the effectiveness of co-Q10 for either ACM or LVEF. Most regression parameter estimates were near zero and p-values were large. In particular, there was no evidence that the effectiveness of co-Q10 varied with the proportion of patients taking statins.

There was some inconclusive evidence (p of < 0.1 but > 0.05) that the dose of co-Q10 affected 6MWT results; however, this was clinically counterintuitive (i.e. higher doses reduced effect) and was driven largely by one outlying trial. 48 Similarly, there was a suggestion that dose affected NYHA class results, but this was also clinically counterintuitive (i.e. reduced benefit with higher dose).

For ACM, the meta-regressions in Table 8 were supplemented with ‘one-stage’ regression models, regressing outcome against treatment and dose, treatment duration and publication year. In all of these, there was no evidence that any factor influenced the effectiveness of co-Q10, with all interaction estimates being very close to the null value of 1. There were insufficient data to perform one-stage meta-regressions for any other outcome.

All-cause mortality

The results for the Bayesian NMA of ACM are shown in Tables 9 and 10. Table 9 shows the odds ratios (ORs) for comparisons between interventions and Table 10 summarises the predicted rankings of interventions.

As for the main meta-analysis, the results suggest that co-Q10 on its own reduces ACM, compared with placebo (although the CI only just includes 1). The limited data for multinutrients mean that CIs are too wide to draw any conclusions on their effectiveness. There is no evidence of any difference between interventions, although CIs are wide. The results suggest that co-Q10 alone is the highest ranked and most effective intervention, and there is almost no chance that placebo is most effective.

New York Heart Association class

The results for the Bayesian NMA of change from baseline in NYHA class are shown in Tables 11 and 12. The results suggest that co-Q10 on its own and co-Q10 with selenium or multinutrient all give a modest improvement in NYHA class compared with placebo (although the CI always includes the null). There is no conclusive evidence of any difference between interventions.

Left ventricular ejection fraction

The results of the Bayesian NMA of LVEF are shown in Tables 13 and 14. As with the main meta-analysis, the results suggest that co-Q10 on its own improves LVEF compared with placebo (although the CI just includes the null). The limited data for the other interventions mean that CIs are too wide for any conclusions on their effectiveness to be drawn. However, both have larger estimates of benefit than co-Q10 alone. There is no conclusive evidence of any difference between interventions. The results favour co-Q10, with selenium being more effective than co-Q10 alone, but CIs are too wide for any conclusions to be drawn. The results, consequently, suggest that co-Q10 with selenium is the highest ranked and most effective intervention, followed by multinutrients, but both of these results are based on a single trial. There is almost no chance that placebo is most effective.

Methods

A broad range of types of economic evaluations were eligible for inclusion in the assessment of cost-effectiveness of co-Q10, including those that considered both costs and consequences (including cost-effectiveness, cost–utility or cost–benefit analyses) and those that considered only costs. MEDLINE, EMBASE and EconLit were searched on 13 May 2020, with no restriction on dates or language.

Full details of the search strategy used are given in Appendix 6.

Two reviewers independently assessed all obtained titles and abstracts for inclusion. Any discrepancies were resolved by discussion and consultation with a third reviewer. It was expected that all studies meeting the inclusion criteria would be summarised and used as the basis for identifying major structural issues, and the assumptions and key drivers of the cost-effectiveness of co-Q10.

Results

The systematic literature search identified 81 references (69 after deduplication), none of which met the inclusion criteria for the cost-effectiveness review of co-Q10 [i.e. none of the studies undertook an assessment of the cost-effectiveness of co-Q10 against any comparator(s)].

Overview of NICE guideline NG106

A total of 26 review questions were identified in the development of the NICE guideline. 78 Six questions were related to the diagnosis of CHF (three of which included elements of cost-effectiveness), 16 were related to interventions for the treatment of CHF (13 of which included elements of cost-effectiveness) and four were prognostic and qualitative questions relating to CHF.

The evidence used to inform the review questions in NG106 was identified by conducting a broad search relating to heart failure in the NHS Economic Evaluation Database (NHS EED) and the Health Technology Assessment database with no date restrictions, and of MEDLINE and EMBASE using a health economic filter, from September 2009. The general heart failure economic search was updated by the Guideline Group on 6 December 2017.

The review excluded conference abstracts, literature reviews, posters, letters, editorials, comment articles, unpublished studies and studies not in English. Studies published before 2001, or those published after 2001 but using unit costs and resource data from before 2001, were excluded. Studies from non-Organisation for Economic Co-operation and Development countries or from the USA were excluded, as were non-comparative cost analyses, including cost-of-illness studies.

Thirteen studies relating to the 13 review questions pertaining to interventions for the treatment of CHF that included elements of cost-effectiveness were identified and included in the final review. A summary of these is presented in Appendix 7, Table 36. Of those identified, three of the cost-effectiveness studies had taken a UK perspective and modelled a lifetime time horizon for CHF, and were considered relevant to standard practice for CHF. 79–81 NICE guideline NG106 also presented new health economic analyses in selected areas. 82

Methods for updating the review

A broad range of studies were considered for inclusion in the updated review of relevant cost-effectiveness evidence for interventions in CHF, including those conducted alongside trials, modelling studies and analyses of administrative databases. Only full economic evaluations that compared two or more options and considered both costs and consequences (including cost-effectiveness, cost–utility or cost–benefit analyses) were included.

MEDLINE was searched on 13 May 2020, with no restrictions on the type of intervention. The search was restricted to journal papers, and conference abstracts were excluded. Full details of the search strategy are given in Appendix 6. Only English-language papers from 2010 onwards were considered for inclusion.

Two reviewers independently assessed all obtained titles and abstracts for inclusion. Any discrepancies were resolved by discussion and consultation with a third reviewer. All studies that met the inclusion criteria were summarised and used as the basis for identifying major structural issues, assumptions and key drivers of cost-effectiveness when modelling CHF.

Results

The literature search identified 3330 potentially relevant publications. Six studies7,81,83–86 from an NHS perspective, published since 2015, met the inclusion criteria for the cost-effectiveness review. Because more applicable and recent evidence was available to inform the decision problem, studies published before 2015 were excluded.

Appendix 8 provides a description of the cost-effectiveness evidence from these six main studies7,81,83–86 and an assessment of the relevance of the data from the perspective of the NHS. A summary of the cost-effectiveness results for standard care, against which a new decision-analytic model could be validated, is given below.

Subgroup analysis

Heterogeneity in the cost-effectiveness of co-Q10 was investigated in three subgroup populations with different levels of disease severity:

1. heart failure patients with a LVEF of < 45

2. heart failure patients with a LVEF of < 45 and NYHA classes II–IV

3. heart failure patients with a LVEF of < 45 and NYHA classes III and IV.

Baseline event rates for mortality

Most of the RCTs in the systematic review of clinical effectiveness were conducted outside the UK, and patient characteristics, treatment patterns and resource use in the UK can be expected to differ from those in the centres involved in those trials. One implication is that the baseline event rates observed in the control groups of the trials are unlikely to provide reliable estimates for UK practice.

The Q-SYMBIO trial provided Kaplan–Meier survival curves for ACM for the full analysis set (FAS) and for a European subpopulation up to a period of 106 weeks. 33,73 Long-term survival was estimated by fitting a parametric exponential distribution to the control arm in the FAS and in the European subset to predict survival over a patient’s lifetime, as shown in Figure 18.

FIGURE 18.

Projected and observed survival in the Q-SYMBIO trial. 33,73 KM, Kaplan–Meier.

In the FAS, the exponential model predicted that 34.2% of patients would be alive after 10 years and 11.7% would be alive after 20 years. For a trial that reported that 88% of its patients were in NYHA class III, these projections appeared implausibly high, based on clinical opinion. Furthermore, this group of trial patients, who had a mean age of 62 years, was also considerably younger than those in the ‘West Yorkshire’ data set,89–91 whose mean age was 69.6 years and 50% had CHF of NYHA class II. Therefore, the cohort in the Q-SYMBIO trial33,73 was not considered representative of the patients typically seen in UK practice who would potentially receive co-Q10.

Further consideration was given to calibrating the projected survival estimates from the Q-SYMBIO trial33,73 to the long-term survival estimates from UK observational data sets of patients with CHF to make an adjustment to the hazard of mortality predicted by the Q-SYMBIO trial. 33,73 Three potential sources were identified for this purpose: (1) The Health Improvement Network (THIN) data set (1998–2012),85 (2) the Clinical Practice Research Datalink (CPRD) linked to inpatient Hospital Episode Statistics (HES) and Office for National Statistics (ONS) mortality data (i.e. CPRD–HES–ONS) (2000–17)92 and (3) the EchoCardiographic Heart Of England Screening (ECHOES) study. 93 THIN is one of the largest databases of general practitioner (GP) records, and includes data from 587 practices in the UK for patients with a code of CHF. The survival rates of those with CHF (i.e. patients with a first diagnostic label) in THIN varied considerably by age. 85 For those aged 55–64 years, the survival estimates were 91.5% at 1 year, 75.0% at 5 years and 58.4% at 10 years, whereas the corresponding survival estimates for those aged 65–74 years were 87.6%, 64.5% and 40.4%. CPRD–HES–ONS92 provides more recent trends in survival after a diagnosis of CHF in the UK. For those aged 55–64 years, the 1-, 5- and 10-year survival rates were 87.9%, 70.6% and 52.8%, respectively, whereas for those aged 65–74 years they were 83.5%, 59.1% and 35.4%, respectively. The ECHOES study93 screened a total of 6162 patients from 16 randomly selected primary care practices in England into four prespecified cohorts (i.e. the general population, diuretic users, those with a prior clinical label of CHF and a population with risk factors for CHF) to identify the prevalence and prognosis of CHF and LVSD. The 5-year survival rate was 53% in patients with CHF and LVSD (mean age of 70.5 years), and survival improved significantly with increasing ejection fraction. However, these estimates were based on a much smaller subset of the full population (only 219 patients with CHF and LVSD).

The THIN, CPRD–HES–ONS and ECHOES data sets85,92,93 are helpful for providing estimates of long-term mortality in patients with a diagnosis of CHF, but their generalisability to a HFrEF population is limited. The CPRD–HES–ONS data set,92 which provides the most recent contemporary evidence, does not specify the type of heart failure (reduced or preserved ejection fraction). There are also additional concerns regarding the reliability of GP coding and misclassification of CHF diagnosis in the THIN and CPRD–HES–ONS data sets,85,92 and the very small sample size of the population with CHF and LVSD in the ECHOES study. 93

Given these concerns, and the fact that the baseline data from the RCTs included in the meta-analysis were not considered representative of patients typically seen in UK practice, extrapolation of the Q-SYMBIO control arm (or calibration to THIN, CPRD–HES–ONS or ECHOES study survival estimates85,92,93) was not considered appropriate for informing long-term baseline survival. However, the impact of estimating ACM from the Q-SYMBIO trial was explored in a scenario analysis. 33,73

Baseline event rates specific to UK practice were therefore obtained from the ‘West Yorkshire’ data set,89–91 which prospectively recorded ACM, hospitalisations and subsequent deaths. Among the sample of patients who were followed until death, for 38.9% the cause of death was cardiovascular (of which 79.4% were related to CHF) and for 57.8% the cause was non-cardiovascular or their death was sudden. Access to this data set permitted the analysis of IPD, allowing the development of risk equations for different patient subgroups. It also permitted the estimation of survival over a longer time period than was available from the RCTs. It may also provide a better estimation of the ‘real-world’ effectiveness of standard care, as used outside a RCT setting. However, this assumes that the treatment effect of co-Q10, derived from the RCTs included in our meta-analysis on which the relative effect used in the model is based, can be generalised to the patients with CHF in the ‘West Yorkshire’ data set. 89–91

Multivariable parametric survival analysis using the ‘West Yorkshire’ data set was used to model ACM over time using baseline characteristics. Backwards, stepwise elimination was undertaken to select covariates for the final model. The model was applied both during and beyond the duration of the follow-up of the data set. The baseline risk for ACM was assumed to follow an exponential distribution, selected from potential distributions, on the basis of statistical fit and clinical plausibility, and from consultation with clinical experts, following review of projected life expectancies and discussion of the nature of the mortality hazard over time. Survival predictions for patients in each subgroup are presented in Figure 19. In scenario analyses, the Weibull distribution was used to evaluate the robustness of the results.

FIGURE 19.

Predicted baseline rate of survival.

Baseline event rates of hospitalisation

The baseline event rates of hospitalisation were obtained from a subset of the ‘West Yorkshire’ data set. 89–91 Hospitalisations in the first 1091 of 1802 patients recruited to the data set were recorded up to 1 year. In the remaining 711 patients, hospitalisations were recorded over a longer time period, with a median follow-up of 2.86 years. 90 Heart failure hospitalisation was defined as a new onset or worsening of signs and symptoms of heart failure, with evidence of fluid overload requiring at least 24-hour hospitalisation and the use of intravenous diuretics.

In the subset of patients with long-term hospitalisation data, the 1-year rate of CHF-related hospitalisation was 6.47% and non-CHF cardiovascular-related hospitalisation was 11.11%. The mean time to first CHF-related hospitalisation was 603 days and the mean time to non-CHF cardiovascular-related hospitalisation was 527 days. In a previous analysis of 1-year hospitalisation in the cohort, predictors of CHF-related hospitalisation included furosemide-equivalent dose, the presence of type 2 diabetes, hospitalisation relating to acute heart failure syndrome within the previous year and pulmonary congestion on chest radiograph. A summary of hospitalisation events in the study is given in Table 19.

To predict the rate of hospitalisation beyond the follow-up period of the study, a multivariate regression model was fitted to the hospitalisation data set. A large proportion of patients were found not to experience any CHF-related hospitalisations or other cardiovascular hospitalisation events. Among the 711 patients in the cohort for whom there were long-term hospitalisation data, 125 (17.6%) experienced a CHF-related hospitalisation and 162 (22.8%) experienced a non-CHF cardiovascular-related hospitalisation (see Table 19).

Previous economic analyses have applied a negative binomial model for all-cause hospitalisation; however, this model fits less well here, as it does not account for excess zeros. To reflect the high proportion of patients with zero events, a hurdle model was selected to predict hospitalisation events. This has two components: (1) a right-censored hurdle process, which models zero versus larger counts (i.e. whether or not a patient has any hospitalisation episodes at all) and (2) a left-truncated count process for positive counts, which predicts the number of hospitalisation episodes conditional on the patient experiencing at least one event. Modelling the two processes separately has the advantage that the fit of the count and the hurdle component can be optimised separately:94

The hurdle process was modelled using a binomial distribution. Based on discussions with clinical experts, and from plotting histograms of follow-up times and hospitalisation events, the proportion of patients who were never hospitalised for CHF or for other cardiovascular reasons was assumed not to vary over time, as it is expected that there would be a certain group of patients who would never require cardiovascular hospitalisations throughout their lifetime, regardless of their survival time. Predictions of the hurdle process (see Table 24) corresponded well to the observed data.

The count process was modelled using a Poisson distribution, which assumes a constant rate of events per unit of time. Although an increase in the number of hospitalisation events is often observed among CHF patients in the last year of life, the evidence suggests that this is attributable to non-cardiovascular causes and that the rate of hospitalisation for CHF remains constant up until the end of life. 95 Examination of the rate of hospitalisation by time did not reveal any consistent trends, with peaks in both those with a low censoring time and those with a high censoring time. This suggests that those censored earlier experienced a larger number of hospitalisations (the reasons for this are unclear, but it could reasonably be considered to be because they die earlier and, therefore, have more severe disease). Likewise, those who had been followed up for a longer amount of time experienced more episodes, most likely because they simply had a longer at-risk period. The monthly CHF-related and other cardiovascular hospitalisation rate (in those patients who experienced at least one event) for the base-case population was 2.15% and 2.38%, respectively. The monthly rate increased for subgroups who had more severe symptoms (Table 20).

Impact of coenzyme Q10

All-cause mortality

In the base-case analysis, the impact of co-Q10 on ACM was estimated using the one-stage meta-analysis estimate of effect (see Chapter 4 and Meta-analysis results), which included data from seven trials. 33,48,55,57,59,62,70 The RR was estimated as 0.68 (95% CI 0.45 to 1.03), suggesting that co-Q10 reduced the risk of death, but this finding was associated with a degree of uncertainty.

The seven trials33,48,55,57,59,62,70 included in the meta-analysis recruited populations with differing disease severity, with inclusion criteria ranging from patients in NYHA class I60,65 to patients with end-stage CHF awaiting heart transplant. 48 None of the trials reported that co-Q10 had a negative effect on ACM. There was no apparent relationship between the disease severity of the populations recruited to the trial and the estimated treatment benefit (shown in the two-stage analysis presented in Figure 20). Therefore, the same treatment effect estimate (i.e. RR 0.68) was assumed in the base case and in the subgroup analyses.

FIGURE 20.

Relative impact of co-Q10 on ACM. SoC, standard of care.

Alternative estimates of the treatment effect were considered in scenario analyses: (1) from the two-stage meta-analysis (RR 0.62, 95% CI 0.45 to 0.85), (2) from the Bayesian NMA (OR 0.54, 95% CI 0.23 to 1.10) and (3) from the one-stage meta-analysis, excluding the Q-SYMBIO trial33,73 (RR 0.72, 95% CI 0.43 to 1.21). The RR for ACM was varied in the probabilistic analysis using a log-normal distribution.

Long-term follow-up was not available in most RCTs. However, our clinical experts judged it plausible that the treatment effect would endure while patients remained on treatment, which is in line with other interventions for CHF, and the base-case analysis applied the duration of treatment effect for the patient lifetime. 96 In a scenario analysis, the treatment effect was limited to 4 and 10 years. Figure 20 presents the relative impact of co-Q10 on ACM when the treatment effect is applied (1) over the patient’s remaining lifetime and (2) for 4 years.

Identification of quality-of-life evidence

None of the trials identified in our meta-analysis of HRQoL was considered appropriate for the economic model because they did not present baseline HRQoL values, they did not report on a generic preference measure or they provided only mean values over time, which could not be used to inform health state utility values.

A targeted review of utility scores was undertaken to identify health state utility values and appropriate utility values for major cardiovascular events.. From the review of cost-effectiveness studies, HRQoL values from two trials, the Systolic Heart failure treatment with the IF inhibitor ivabradine Trial (SHIFT)97,98 and the Eplerenone Post-AMI Heart Failure Efficacy and Survival Study (EPHESUS),99 were considered relevant to our economic model. Utility values from SHIFT were used in the base-case analysis, whereas the utility values from EPHESUS, consisting of post-MI patients whose CHF may be more severe than the modelled population, were used in a scenario analysis.

Impact of coenzyme Q10 on quality of life

The results of our meta-analysis of the data available from four trials,50,52,68,69 suggested no clear evidence of any effect of co-Q10 on QoL (see Figure 14). Therefore, the same health state utility values were applied to co-Q10 and to standard therapy alone.

Health state utility values

The Systolic Heart failure treatment with the IF inhibitor ivabradine Trial97,98 evaluated the effect of ivabradine (Procoralan, Servier Laboratories, Suresnes, France) compared with placebo added to guidelines-driven background therapy in 6558 adult patients with CHF of NYHA classes II–IV, a LVEF of ≤ 35% and a resting heart rate of ≥ 70 beats per minute. The study collected HRQoL data using the EQ-5D questionnaire, which was administered to patients (n = 5313) in countries in which the questionnaire has been validated. UK tariff values were used in the study, regardless of the country of origin of the HRQoL data. HRQoL values from SHIFT97,98 were analysed and applied in two separate economic evaluations of ivabradine. 97,100

In one cost-effectiveness study,100 EQ-5D data were analysed using multilevel modelling, which takes into account correlation between measurements from the same individual to increase precision and avoid bias. The regression equation included coefficients for treatment allocation, baseline characteristics, NYHA class and hospitalisation episode. The baseline utility values by NYHA class for the SHIFT97,98 population were 0.82 for NYHA class I, 0.74 for NYHA class II, 0.64 for NYHA class III and 0.46 for NYHA class IV. Table 21 presents the predicted baseline utility values for the four patient subgroups of interest in the cost-effectiveness analysis, estimated using the mean utility value for each NYHA class, and weighted by the proportion of patients in each NYHA class in the ‘West Yorkshire’ data set (see Table 21). 89–91 This study also estimated the impact on HRQoL of all-cause hospitalisation; however, we did not apply this value to our economic analysis.

In the second cost-effectiveness study,97 the authors did not distinguish utility values by NYHA class. However, the authors considered the impact of cardiovascular hospitalisations rather than all-cause hospitalisations, and further split the impact into CHF-related and non-CHF cardiovascular-related hospitalisation, which were considered more appropriate for use in the decision model. Two regression equations were developed to estimate change in EQ-5D score from baseline, using treatment, beta-blocker use and number of hospitalisations as independent variables. The same utility decrements were applied to each hospitalisation event type regardless of patient subgroup.

A study by Göhler et al. 99 used primary data from EPHESUS99 to estimate the utility values for patients with CHF according to their NYHA classification and number of cardiovascular rehospitalisations. EPHESUS99 recruited CHF patients from a post-MI population who were considered to have greater ill-health and worse HRQoL than patients who would generally be eligible for co-Q10. The baseline utility values by NYHA class for the EPHESUS99 population were slightly lower than those estimated from the SHIFT97,98 population, reflecting the greater debilitation of the population: 0.79 for NYHA class I, 0.71 for NYHA class II, 0.62 for NYHA class III and 0.48 for NYHA class IV. The decrement was –0.024 for one rehospitalisation, –0.031 for two rehospitalisations and –0.055 for three or more rehospitalisations. These utility values were used in a scenario analysis in our model, with the mean rehospitalisation utility decrement applied for hospitalisation events.

Quality of life over time

Health-related quality of life was adjusted to reflect its decreases associated with ageing. Age- and sex-adjusted norms for the UK were adjusted downwards by approximately 13% to reflect the presence of CHF. The adjustment factor was estimated by comparing the baseline utility in SHIFT97,98 with the average utility of a person of the equivalent mean age (60 years) in the UK, derived from a nationally representative UK sample using the EQ-5D. 101

Coenzyme Q10

Although many patients purchase co-Q10 themselves, this analysis explores the impact of the cost of co-Q10 being borne by the NHS. The unit costs of co-Q10 were obtained from the British National Formulary (BNF) [provided as ubidecarenone (Bio-Quinone^®, Pharma Nord)]. 102 The mean daily cost was estimated assuming a dosing regimen of 100 mg three times a day, as per the Q-SYMBIO trial. 33 It was assumed that compliance would be 100% if co-Q10 were to be adopted. The mean daily cost was £0.94, equating to £28.61 per month (Table 22).

Standard care

Standard care was assumed to comprise background medication and resource use for CHF, including accident and emergency referrals, outpatient contacts and GP visits. Standard care for CHF comprised a number of pharmacological agents, including ACE inhibitors, beta-blockers, mineralocorticoid receptor antagonists and diuretics. The proportion of patients receiving each type of medication and the mean dose of each medication were informed by the ‘West Yorkshire’ data set (see Table 18). 89–91 The costs of background therapies were based on average doses and utilisation reported in the ‘West Yorkshire’ data set at baseline (Table 23). Unit costs were obtained using the electronic market information tool (eMIT). 103 This provides information on the average price that the NHS pays for pharmaceuticals, which can differ from the prices listed in the BNF, and it is a more accurate and up-to-date indicator of the costs incurred by the NHS. The total monthly cost for standard therapy was estimated to be £1.94.

Background resource use (Table 24) was informed by an analysis of the CPRD reported by McMurray et al. 83 The CPRD data set covers the English NHS and provides bespoke data tables that consider resource use, excluding hospitalisation and pharmacological therapies. Unit costs were from published national sources. 104,105

The annual cost of background resource use was estimated to be £925.27 (equivalent to a monthly cost of £77.11).

Hospitalisation

The analysis includes the costs of hospitalisations by admission type (i.e. CHF-related or non-CHF cardiovascular-related) estimated from NHS Reference Costs. 105

The reference costs are the average costs to the NHS of providing a defined service or resource in a given financial year. The costs are categorised into groups [Healthcare Resource Groups (HRGs)] according to episodes that are clinically coherent and consume similar resources. NHS reference costs provide the unit costs of a hospitalisation event and not a cost per day. This method is considered to be aligned with the process through which care is reimbursed in England and Wales. The costs of each type of hospitalisation event are based on a weighted average of short- and long-stay visits, as well as the numbers of patients who incur complications. The unit costs of hospitalisations were assumed to be the same in both arms of the model.

The unit cost of a CHF-related hospitalisation event was estimated as £1935 from HRG codes EB03A-E (heart failure or shock). A non-CHF cardiovascular-related event was estimated as £1272 from HRG codes AA22C-G (cerebrovascular accident, nervous system infections or encephalopathy), AA35A-F (stroke), EB07A-E (arrhythmia or conduction disorders) and EB10A-E (actual or suspected myocardial infarction).

Overview

The cost-effectiveness of co-Q10 as an adjunct to standard therapy in patients with CHF and reduced ejection fraction was evaluated by comparing the total expected lifetime costs and QALYs with those of standard therapy alone. The mean costs and QALYs for co-Q10 as an adjunct therapy and for standard care alone are presented, and their cost-effectiveness is compared using conventional cost-effectiveness decision rules, estimating the incremental cost-effectiveness ratio (ICER) as appropriate. The ICER uses the additional costs that one strategy incurs over another and compares this with the additional benefits.

The cost-effectiveness results are presented for the base-case population and separately for each of the scenario and subgroup populations. Scenario analyses were used to test the robustness of the cost-effectiveness results to changes in the structural assumptions of the model.

Probabilistic sensitivity analysis

Uncertainty in the data used to populate the economic model was characterised using a probabilistic analytic approach, with each input entered as an uncertain parameter with an assigned probability distribution representing its uncertainty.

The uncertainty in the probability and utility parameters was represented using beta distributions, as these values are typically bounded at 0 and 1. Log-normal distributions were used to estimate uncertainty in HRs, RRs and ORs. Gamma distributions were used to represent the uncertainty in the cost parameters, as these values are constrained to be non-negative, but often are right-skewed distributions. To account for uncertainty around the parametric models fitted to ACM and hospitalisation, outcomes were sampled using their associated variance–covariance matrices of covariates. When the variation around the mean input was not available, the standard error was assumed to be equivalent to 20% of the width of the 95% CI (normally distributed) around the point estimate.

Probabilistic sensitivity analysis was used to estimate 95% CIs around the cost-effectiveness results. The mean probabilistic estimate of costs and QALYs was estimated from 10,000 iterations of the model to estimate the uncertainty surrounding the total costs and QALYs calculated from the model. Cost-effectiveness acceptability curves (CEACs) are presented. The CEACs show the probability that co-Q10 is cost-effective at a range of threshold values that NHS decision-makers attach to an additional QALY (e.g. £20,000–30,000 per additional QALY, as used by NICE).

The base-case parameters, associated assumptions and sources are given in Appendix 9, Table 37.

Value of information

Value-of-information analysis was used to quantify the expected benefits of further research by estimating the value of reducing decision uncertainty related to the cost-effectiveness of adjunct co-Q10 compared with standard care alone in HFrEF. The maximum amount that the NHS should be willing to invest to reduce decision uncertainty can be informed by the expected value of perfect information (EVPI). 106 The EVPI evaluates the expected consequences of decision uncertainty, in terms of costs incurred and health benefits lost, should it later transpire (with further research) that the decision based on the evidence currently available is not correct. The EVPI takes into account both the probability that the decision based on existing evidence is wrong, and the magnitude of the consequences of making an incorrect decision. The consequences of making an incorrect decision because of uncertainty can be compared with the costs of conducting new research (e.g. a clinical trial) to establish the potential value of new research.

The EVPI provides an estimated value of resolving all uncertainty parameterised within the model through the provision of perfect information and provides a measure of the expected maximum return from further research. Therefore, EVPI represents an expected upper bound to the amount that a decision-maker should be willing to pay for additional evidence to support the current decision. If the EVPI exceeds the expected costs of additional research, then it is potentially cost-effective to acquire more information by undertaking this research.

Information generated in research is used to inform decisions for the population of patients who could potentially benefit from the information. This depends on the size of the benefiting population, whose decision choice will be informed by the additional research and the time horizon over which the information generated by research is useful. This means that the population-level EVPI is estimated by scaling up the individual (per-patient) EVPI by the number of people who would be affected by the information over the anticipated lifetime of the technology. This can be expressed as:

where I is the incidence in the period, t is the period, T is the total number of periods for which information from research would be useful and r is the discount rate.

The British Heart Foundation estimates that the prevalence of CHF in the UK is 658,944 (the number of people in the UK on their GP’s heart failure register),107,108 although as 920,000 people may be living with heart failure. 109 The annual incidence of CHF is estimated to be 89,986 cases, based on an age-standardised rate of 164.1 per 100,000 of the population (209.6 for men and 127.6 for women per 100,000 population). 110 Assuming a 10-year time horizon for the lifetime of the technology gives an effective CHF population of 1,433,512 (prevalent population plus incident population per annum discounted at 3.5% over the lifetime of the technology). It is assumed that 50% of this population has a reduced ejection fraction, giving an estimate of 716,756 for the population EVPI calculations. 5

The EVPI analysis quantifies the decision uncertainty predicted by the model and provides the maximum value that can be placed on additional research. However, it does not address the structural or methodological uncertainty inherent in the set of model assumptions. To explore the value of reducing decision uncertainty related to the assumptions on the key effectiveness parameter for co-Q10 on ACM, the EVPI calculations were conducted for two scenarios using different assumptions about the treatment effect on ACM and its associated uncertainty. The first scenario used the estimate of treatment effect as applied in the base-case analysis from the one-stage meta-analysis (RR 0.68, 95% CI 0.45 to 1.03). With this estimate of treatment effect, the CI around the RR indicates that there is very little uncertainty about whether or not co-Q10 has a positive impact on ACM. However, there are concerns that this estimate, which is largely influenced by the effect observed in the Q-SYMBIO trial33 (the largest trial with the greatest weight in the meta-analysis), underestimates the true uncertainty about the effect of co-Q10 on ACM. In the second scenario, the treatment effect of co-Q10 on ACM was estimated from the one-stage meta-analysis by excluding the Q-SYMBIO trial. 33 Excluding this study leads to greater uncertainty in the treatment effect of co-Q10 on ACM, with the CI estimating a higher likelihood that co-Q10 is not associated with a positive impact (RR 0.72, 95% CI 0.43 to 1.21).

Scenario analyses

Several alternative scenarios were considered, in which the assumptions used as part of the base-case results were varied. These analyses were undertaken to assess the robustness of the base-case results to variation in the sources of data used to populate the model and alternative assumptions. The alternative scenarios are summarised in Table 25.

Expected value of perfect information results

The value of the uncertainty surrounding a decision based on the expected ICER estimates is expressed using both individual (per patient) estimates of EVPI and total population EVPI, based on a 10-year time horizon. Table 27 presents the EVPI estimates for two scenarios that vary the assumptions about the treatment effect of co-Q10 on ACM.

Under the base-case assumptions, the individual EVPI ranges from £162 to £292 across the range of cost-effectiveness thresholds of £10,000 to £30,000 per additional QALY. The decision uncertainty is relatively low because co-Q10 is increasingly cost-effective as the threshold is increased (i.e. the probability of decision error is low at thresholds above £10,000, as seen in Figure 21, with a chance of only approximately 5% that co-Q10 is not a cost-effective option), but the consequences of error are valued more highly as the threshold is increased. The implication is that the EVPI increases with higher values of the threshold because the opportunity losses associated with making an incorrect decision are increasing. The estimates of population EVPI ranged from £116M to £209M across the different thresholds for the base-case assumptions. These estimates demonstrate that there appears to be value in undertaking additional research to reduce the existing decision uncertainty.

When the EVPI calculations were based on more uncertain estimates of the treatment effect of co-Q10 on ACM, the individual EVPI ranged from £635 to £1487 across the range of cost-effectiveness thresholds of £10,000 to £30,000 per additional QALY. The corresponding estimates of population EVPI ranged from £455M to £1066M. The EVPI results are higher under this scenario than with the base-case assumptions, because the decision uncertainty is increased (i.e. the probability of error increases from 5% in the base case to 12% in this scenario, while the consequences of error are valued highly as the threshold is increased).

Search strategy

1. heart failure.mp. (784)

2. ((heart or cardiac or myocardial) adj4 failure$).ti,ab,sh. (789)

3. cardiomyopath$.ti,ab,sh. (86)

4. 1 or 2 or 3 (879)

5. exp Trace elements/(555)

6. micronutrient$.ti,ab,sh. (116)

7. Ubiquinon$.ti,ab,sh. (26)

8. ubiquinol.ti,ab,sh. (7)

9. ubidecarenone.ti,ab,sh. (0)

10. quinone.ti,ab,sh. (83)

11. neuquinon$.ti,ab,sh. (0)

12. (“bio-quinone Q10” or “bioquinone Q10”).ti,ab,sh. (0)

13. (“co-enzyme Q$” or “coenzyme Q$”).ti,ab,sh. (100)

14. (COQ10 or C0Q10 or “COQ 10” or “C0 Q10”).ti,ab,sh. (27)

15. (Q10 or “Q 10” or Q-10).ti,ab,sh. (99)

16. 5 or 6 or 7 or 8 or 9 or 10 or 11 or 12 or 13 or 14 or 15 (852)

17. 4 and 16 (15).

Search strategy

1. #1 MeSH descriptor: [Heart Failure] explode all trees

2. #2 ((heart or cardiac or myocardial) NEAR/4 failure*):ti,ab,kw (Word variations have been searched)

3. #3 MeSH descriptor: [Cardiomyopathies] explode all trees

4. #4 MeSH descriptor: [Cardiomyopathy, Dilated] explode all trees

5. #5 (cardiomyopath*):ti,ab,kw (Word variations have been searched)

6. #6 #1 OR #2 OR #3 OR #4 OR #5

7. #7 MeSH descriptor: [Micronutrients] explode all trees

8. #8 MeSH descriptor: [Ubiquinone] explode all trees

9. #9 (ubiquinon*):ti,ab,kw OR (ubiquinol):ti,ab,kw OR (ubidecarenone):ti,ab,kw OR (quinone):ti,ab,kw OR (neuquinon*):ti,ab,kw (Word variations have been searched)

10. #10 (“bio-quinone Q10”):ti,ab,kw (Word variations have been searched)

11. #11 (“co-enzyme Q*”):ti,ab,kw (Word variations have been searched)

12. #12 (“coenzyme Q*”):ti,ab,kw (Word variations have been searched)

13. #13 (COQ10):ti,ab,kw OR (C0Q10):ti,ab,kw (Word variations have been searched)

14. #14 (Q10):ti,ab,kw OR (“Q 10”):ti,ab,kw OR (Q-10):ti,ab,kw (Word variations have been searched)

15. #15 #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #13 OR #14

16. #16 #6 AND #15.

Search strategy

1. exp Heart Failure/ (433,070)

2. (heart adj4 failure$).mp. (365,011)

3. (cardiac adj4 failure$).mp. (31,005)

4. (myocardial adj4 failure$).mp. (12,338)

5. exp cardiomyopathy/ (121,733)

6. cardiomyopath$.mp. (136,317)

7. 1 or 2 or 3 or 4 or 5 or 6 (568,017)

8. exp Trace Elements/ (35,322)

9. micronutrient$.mp. (17,727)

10. ubiquinone/ (7221)

11. ubiquinon$.mp. (19,834)

12. ubiquinol.mp. (6323)

13. ubidecarenone.mp. (7836)

14. quinone.mp. (26,940)

15. neuquinon$.mp. (12)

16. bioquinone Q10.mp. (0)

17. bio-quinone Q10.mp. (4)

18. co-enzyme Q$.mp. (217)

19. coenzyme Q$.mp. (6347)

20. COQ10.mp. (2444)

21. COQ 10.mp. (224)

22. Q10.mp. (8216)

23. (“Q 10” or Q-10).mp. (1572)

24. C0Q10.mp. (9)

25. C0Q 10.mp. (0)

26. 8 or 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23 or 24 or 25 (101,092)

27. 7 and 26 (2341)

28. exp animal/or exp nonhuman/or exp animal experiment/or exp animal model/ (24,850,091)

29. exp human/ (18,840,975)

30. 28 not 29 (6,009,116)

31. 27 not 30 (1945).

Search strategy

Search strategy

1. exp Heart Failure/ (109,307)

2. (heart adj4 failure$).mp. (185,558)

3. (cardiac adj4 failure$).mp. (20,579)

4. (myocardial adj4 failure$).mp. (8237)

5. Cardiomyopathies/ (25,814)

6. Cardiomyopathy, Dilated/ (14,800)

7. cardiomyopath$.mp. (88,142)

8. 1 or 2 or 3 or 4 or 5 or 6 or 7 (262,456)

9. Micronutrients/ (5085)

10. micronutrient$.mp. (14,686)

11. Ubiquinone/ (8516)

12. ubiquinon$.mp. (12,111)

13. ubiquinol.mp. (1796)

14. ubidecarenone.mp. (65)

15. quinone.mp. (20,439)

16. neuquinon$.mp. (0)

17. bio-quinone Q10.mp. (2)

18. co-enzyme Q$.mp. (127)

19. coenzyme Q$.mp. (5528)

20. (COQ10 or C0Q10).mp. (1538)

21. (COQ 10 or C0 Q10).mp. (332)

22. Q10.mp. (6430)

23. Q 10.mp. (2374)

24. Q-10.mp. (2374)

25. 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23 or 24 (49,941)

26. 8 and 25 (620)

27. exp animals/not humans/ (4,502,274)

28. 26 not 27 (545).

Search strategy

ClinicalTrials.gov via internet

Search date: 12 October 2018.

Records identified: 93.

Series of searches carried out.

ISRCTN registry

URL: www.isrctn.com/

Search date: 12 October 2018.

A series of individual searches using the following terms were carried out and three relevant records were identified.

OpenTrials.net

URL: https://opentrials.net/

A series of individual searches using the following terms were carried out and six relevant records were identified.

World Health Organization International Clinical Trials Registry Platform

URL: http://apps.who.int/trialsearch/

Search date: 12 October 2018.

Five unique records identified.

Search strategy

1. exp Heart Failure/ (118,987)

2. (heart adj4 failure$).mp. (204,092)

3. (cardiac adj4 failure$).mp. (21,965)

4. (myocardial adj4 failure$).mp. (8932)

5. Cardiomyopathies/ (27,624)

6. Cardiomyopathy, Dilated/ (15,496)

7. cardiomyopath$.mp. (95,608)

8. 1 or 2 or 3 or 4 or 5 or 6 or 7 (286,729)

9. Micronutrients/ (5692)

10. micronutrient$.mp. (16,959)

11. Ubiquinone/ (9153)

12. ubiquinon$.mp. (12,971)

13. ubiquinol.mp. (1881)

14. ubidecarenone.mp. (66)

15. quinone.mp. (22,404)

16. neuquinon$.mp. (0)

17. bio-quinone Q10.mp. (2)

18. co-enzyme Q$.mp. (150)

19. coenzyme Q$.mp. (6117)

20. (COQ10 or C0Q10).mp. (1773)

21. (COQ 10 or C0 Q10).mp. (338)

22. Q10.mp. (7120)

23. Q 10.mp. (2580)

24. Q-10.mp. (2580)

25. 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 21 or 22 or 23 or 24 (55,185)

26. 8 and 25 (683)

27. exp animals/not humans/ (4,679,006)

28. 26 not 27 (602).

Ovid MEDLINE(R) ALL via Ovid

Database: Ovid MEDLINE(R) ALL via Ovid.

Dates searched: 1946 to 12 May 2020.

Search date: 14 May 2020.

Records identified: 13.

EMBASE via Ovid

Database: EMBASE via Ovid.

Dates searched: 1974 to 2020 week 19.

Search date: 14 May 2020.

Records identified: 68.

EconLit

Database: EconLit via Ovid.

Dates searched: 1886 to 30 April 2020.

Search date: 14 May 2020.

Records identified: 0.

Ovid MEDLINE(R) ALL

Database: Ovid MEDLINE(R) ALL.

Dates searched: 1946 to 12 May 2020.

Search date: 13 May 2020.