This meta-analysis was elaborated in accordance with the PRISMA guidelines.7 Two reviewers (Y. Xu and X. Liu) independently performed the online searches. Studies on CHD and cardiovascular events were acquired by searching PubMed, EMBASE, and the Cochrane Library from the inception date to January 2023. The full search strategy is provided in data 1 of the supplementary data. References from reviews, selected articles, and previous meta-analysis were also screened.

Two investigators (Y. Xu and X. Liu) reviewed the selected publications to assess their eligibility, with disagreements resolved through discussion with a senior author (X. Wang). We included studies that assessed CMD with invasive modalities and reported a risk ratio (RR) for major adverse cardiac events (MACE) and/or all-cause death. Indices of invasive coronary microvascular function measurement included CFR via Doppler or thermodilution, IMR via thermodilution, hyperemic microvascular resistance via Doppler, resistive reserve ratio, and angiography-based index. The inclusion criteria were as follows: a) evaluation of coronary microvascular function by invasive methods; b) patients with proven or suspected CHD; c) not less than 1 year of follow-up; d) clear definition of the cutoff value of CMD. The exclusion criteria were as follows: a) no follow-up regarding cardiovascular outcomes; b) studies with duplicate publication; c) RR or hazard ratio (HR) were not given in the study.

Two independent reviewers (Y. Xu and X. Liu) extracted the following data from all the eligible studies on a standardized electronic form, including baseline characteristics, risk factors such as hypertension, smoking, diabetes mellitus, dyslipidemia, and the cutoff values of abnormal CMD. We extracted the event rates, RR, and their associated 95% confidence intervals (95%CIs) from the main text and supplementary material. If studies investigated more than 2 measurement methods (eg, CFR and IMR) and provided RR for each, they were considered separate studies for the purpose of the meta-analysis. As far as possible, we chose the adjusted HR as the data source. If there was no HR in the text and supplementary material, the RR value was calculated by the original data.

Two reviewers (Y. Xu and X. Liu) independently assessed the risk of bias. Based on the categories of selection (4 items: representativeness of the exposed cohort, selection of the nonexposed cohort, ascertainment of exposure, demonstration that outcome was not present at the study start), comparability (2 items: controls for the most important factor and any additional factors), and outcome (3 items: assessment, duration, and adequacy of follow-up), a quality score was calculated for each item (table 1 of the supplementary data).

The primary endpoint was the rate of MACE. The secondary endpoint was the rate of all-cause death. There were several definitions of MACE (table 2 of the supplementary data). We adopted the definition used in each study, including all-cause or cardiac death, myocardial infarction, rehospitalization for heart failure or unstable angina, and revascularization.

We performed a random-effects meta-analysis using inverse-variance weighting (expressed as RR with 95%CI), using the natural logarithm of the RR and standard errors. The RR and 95%CI were presented as forest plots. Review Manager (RevMan) 5.4 was used for the statistical analysis. A 2-sided P<.05 was considered statistically significant. Heterogeneity was assessed using the I2 test. An I2 of 0% to 25%, 26% to 50%, and >50% was considered as low, medium, and high statistical heterogeneity, respectively. We performed subgroup analyses by different measurement modalities and CHD populations (eg, chronic coronary syndrome [CCS] or acute coronary syndrome [ACS]). In addition, studies in which most participants presented with stable CHD were also classified as the CCS group. Subgroup analyses according to the study design (retrospective vs prospective), follow-up duration ( ≤ 36 months vs >36 months), and IMR measurement methods (wire-based vs coronary angiography-based) were also conducted. In addition, we conducted meta-regression to explore whether the magnitude of risk would differ across different measurement modalities and CHD populations and the interaction P values were calculated. Sensitivity analyses were conducted to evaluate the effect of each individual study on the pooled estimates. Publication bias was examined using the funnel plot and the Egger's test. If P<.05, the trim-and-fill method was used to evaluate the impact of publication bias on the results of the study. Meta-regression, sensitivity analyses, and publication bias were all performed by Stata SE 17. An interaction P value<.017 after Bonferroni correction was considered statistically significant.

A total of 27 studies8–34 with 11 404 patients met the inclusion criteria (figure 1 of the supplementary data, figure 1). The characteristics of the included studies are summarized in table 3 of the supplementary data. Coronary microvascular function was assessed by CFR (15 studies), IMR (16 studies), resistive reserve ratio (3 studies), and hyperemic microvascular resistance (1 study). The median length of follow-up ranged from 12 to 135.6 months. The main inclusion criteria for all study populations are summarized in table 4 of the supplementary data. The average age was between 51.4 and 67 years, and there was a preponderance of men. The characteristics of the included patients are summarized in table 5 of the supplementary data.

Figure 1.

Central illustration. Meta-analysis of invasive assessment of coronary microvascular dysfunction and cardiovascular outcomes. After screening 3751 articles, a meta-analysis of 27 studies with 11 404 CHD patients was performed. CMD was assessed by invasive indices. The primary endpoint was MACE, and the secondary endpoint was all-cause death. CMD was associated with a higher risk of MACE and all-cause death. Subgroup analysis showed that the strength of the impact of CMD on cardiovascular events was similar among various measurement modalities but was greater in ACS patients than in CCS patients. *Significantly different. Created using BioRender.com. 95%CI, 95% confidence interval; ACS, acute coronary syndrome; CCS, chronic coronary syndrome; CFR, coronary flow reserve; CHD, coronary heart disease; CMD, coronary microvascular dysfunction; HMR, hyperamic microvascular resistance; IMR, index of microvascular resistance; IV, inverse-variance; MACE, major adverse cardiac events; RR, risk ratio; RRR, resistive reserve ratio.

(0.32MB).

All 27 studies reported the incidence of MACE. Patients with CMD had a higher risk of MACE than those without CMD (RR, 2.18; 95%CI, 1.80-2.64; P<.01) (figure 2). There was wide heterogeneity among the included studies (I2=79%; P<.01). A total of 15 studies reported the incidence of all-cause death. The risk of all-cause death was higher in patients with CMD than in those without CMD (RR, 1.88; 95%CI, 1.55-2.27; P<.01) (figure 3). There was no evidence of heterogeneity among the included studies (I2=5%; P=.40).

Figure 2.

Forest plot for the risk of MACE. Data obtained from 27 studies using a random effects meta-analysis and expressed as risk ratios. 95%CI, 95% confidence intervals; CFR, coronary flow reserve; IMR, index of microvascular resistance; IV, inverse-variance; MACE, major adverse cardiac events; NH-IMRangio, nonhyperemic angiography-derived index of microcirculatory resistance; RRR, resistive reserve ratio; SE, standard error. References: Takahashi et al.20 Meuwissen et al.8 Fearon et al.9 van de Hoef et al.15 Ahn et al.33 Murai et al.23 de Waard et al.32 Lee JM et al.26 Nishi et al.10 Suda et al.21 Hu et al.14 Joo Myung Lee et al.27 Lee SH et al.25 Maznyczka et al.18 Abdu et al.34 Choi et al.17 Johnson et al.30 Kotronias et al.28 Scarsini et al.22 Toya et al.19 FISIOIAM.31 Kim et al.29 Seung Hun Lee et al.24 Dai et al.13 Feng et al.12 Liu et al.11 Mejia-Renteria et al.16

(0.58MB).

Figure 3.

Forest plot of the risk of all-cause death. Data obtained from 12 studies using a random effects meta-analysis and expressed as risk ratio. 95%CI, 95% confidence intervals; CFR, coronary flow reserve; IMR, index of microvascular resistance; IV, inverse-variance; RRR, resistive reserve ratio; SE, standard error. References: Takahashi et al.20 van de Hoef et al.15 Fearon et al.9 Nishi et al.10 Hu et al.14 Joo Myung Lee et al.27 Lee SH et al.25 Scarsini et al.22 Toya et al.19 Choi et al.17 Kim et al.29 Seung Hun Lee et al.24

(0.45MB).

The association of CMD with MACE was consistent among all measurement modalities (figure 4), with significant risk in patients using CFR (RR, 2.19; 95%CI, 1.65-2.90; P<.01), IMR (RR, 2.22; 95%CI, 1.63-3.02; P<.01), resistive reserve ratio (RR, 1.65; 95%CI, 1.28-2.12; P<.01), and hyperemic microvascular resistance (RR, 5.70; 95%CI, 1.94-16.74; P<.01). The heterogeneity among all groups was significant, indicating that the measurement methods were not the reason for the wide heterogeneity of the effect of CMD on MACE. There was no significant difference in the impact of CMD on MACE among different measurement modalities (interaction P value=.95). We observed similar results for the association of CMD with all-cause death. The effect of CMD on all-cause death was consistent among subgroups (figure 2 of the supplementary data). Similarly, there was no significant difference within distinct measurement modalities (interaction P value=.58). Subgroup analysis stratified by the IMR measurement method was conducted (figure 3 of the supplementary data). There was no significant difference in the impact of CMD on MACE among different IMR measurement methods (interaction P value=.16).

Figure 4.

Subgroup analysis of MACE stratified by measurement modalities. 95%CI, 95% confidence intervals; CFR, coronary flow reserve; HMR, hyperemic microvascular resistance; IMR, index of microvascular resistance; IV, inverse-variance; MACE, major adverse cardiac events; NH-IMRangio, nonhyperemic angiography-derived index of microcirculatory resistance; RRR, resistive reserve ratio; SE, standard error. References: Takahashi et al.20 Meuwissen et al.8 Fearon et al.9 van de Hoef et al.15 Ahn et al.33 Murai et al.23 de Waard et al.32 Lee JM et al.26 Nishi et al.10 Suda et al.21 Hu et al.14 Joo Myung Lee et al.27 Lee SH et al.25 Maznyczka et al.18 Abdu et al.34 Choi et al.17 Johnson et al.30 Kotronias et al.28 Scarsini et al.22 Toya et al.19 FISIOIAM.31 Kim et al.29 Seung Hun Lee et al.24 Dai et al.13 Feng et al.12 Liu et al.11 Mejia-Renteria et al.16

(0.95MB).

In CMD vs no-CMD populations, the RR of MACE was consistent among different CHD populations (figure 5), with greater risk observed in ACS patients (RR, 2.84; 95%CI, 2.26-3.57; P<.01) compared with CCS patients (RR, 1.77; 95%CI, 1.44-2.18; P<.01). Heterogeneity within CCS subgroups was still significant, suggesting that the included population was not the source of the wide heterogeneity of CMD in MACE. The strength of the association of CMD on MACE differed among ACS and CCS subgroups (interaction P value<.01). The RR of all-cause death applied to CMD vs no-CMD was consistent among different CHD populations (figure 4 of the supplementary data), with no significant differences among subgroups (interaction P value=.32).

Figure 5.

Subgroup analysis of MACE stratified by CHD populations. 95%CI, 95% confidence intervals; ACS, acute coronary syndrome; CCS, chronic coronary syndrome; CFR, coronary flow reserve; CHD, coronary heart disease; IMR, index of microvascular resistance; IV, inverse-variance; MACE, major adverse cardiac events; NH-IMRangio, nonhyperemic angiography-derived index of microcirculatory resistance; RRR, resistive reserve ratio; SE, standard error. References: Takahashi et al.20 Meuwissen et al.8 Fearon et al.9 van de Hoef et al.15 Ahn et al.33 Murai et al.23 de Waard et al.32 Lee JM et al.26 Nishi et al.10 Suda et al.21 Hu et al.14 Joo Myung Lee et al.27 Lee SH et al.25 Maznyczka et al.18 Abdu et al.34 Choi et al.17 Johnson et al.30 Kotronias et al.28 Scarsini et al.22 Toya et al.19 FISIOIAM.31 Kim et al.29 Seung Hun Lee et al.24 Dai et al.13 Feng et al.12 Liu et al.11 Mejia-Renteria et al.16

(1.09MB).

We conducted subgroup analysis of different measurement modalities in ACS and CCS populations (figures 5-8 of the supplementary data). There was no significant difference in the impact of CMD on MACE among different measurement modalities in either the ACS (interaction P value=.56) and CCS populations (interaction P value=.72). Similarly, there was no significant difference in the impact of CMD on all-cause death in the ACS (interaction P value=.25) or CCS population (interaction P value=.78).

We also performed subgroup analyses with respect to MACE and all-cause death according to study design and length of follow-up (table 1 and table 2; figure 9-12 of the supplementary data). Additionally, we conducted a subgroup analysis based on the diverse definitions of study outcomes. Our findings revealed that the definition of MACE did not account for the heterogeneity observed in the impact of CMD on MACE, with consistent results noted among different subgroups (figure 13 of the supplementary data). Moreover, a subgroup analysis was performed for both the original data source and the HR directly extracted from the original text. Among these, 15 studies calculated RR values based on the original data of the article, while 20 studies directly extracted HR values from the article. Interestingly, we found that the data source did not influence the heterogeneity, with consistent outcomes noted among subgroups (figure 14 of the supplementary data). A summary of the study results is shown in the figure 1.

We observed that the funnel plot for MACE was asymmetrical, suggesting the possible existence of publication bias (figure 15 of the supplementary data). Egger's test proved the possibility of publication bias (P < .01). Considering that publication bias might lead to instability or even reversal of the results, we used the trim-and-fill method to observe the effect of publication bias on this study. Thirteen additional studies were supplemented to control publication bias, the results of which showed that patients with CMD had a higher risk of MACE (RR, 1.69; 95%CI, 1.43-1.99; P<.01). This suggested that publication bias did not have a significant impact on this study. Using the same method, we found that the impact of publication bias of all-cause death was not significant.

Sensitivity analysis showed that the exclusion of any single study from the analysis for MACE did not alter the overall findings (figure 16 of the supplementary data). The sensitivity analysis of all-cause death showed similar results (figure 17 of the supplementary data).

There was heterogeneity in the assessment of the impact of CMD on MACE. Thus, we performed subgroup analyses based on measurement modalities, study population, study design, length of follow-up, outcome definition, and data sources, etc., and found the results were consistent across different subgroups. Furthermore, publication bias was unavoidable. We used the trim-and-fill method to observe the effect of publication bias and found that the impact of publication bias on this study was not significant. After Bonferroni correction, the risk of invasive CMD assessment in predicting MACE varied among different CHD populations. There is insufficient evidence to demonstrate invasive CMD assessment is associated with all-cause death in the CHD population. The limited number of studies assessing all-cause death compared with MACE, imbalances in the number of patients in the ACS and CCS groups, and the lower overall incidence of all-cause death may contribute to the lack of evidence.