A comprehensive literature search was performed to locate all pertinent publications examining the impact of risk factors on in-hospital mortality among CHIPs. An electronic search was conducted using EMBASE, PubMed, Cochrane Library, Web of Science, Chinese Biomedical Database (CBM), China National Knowledge Infrastructure (CNKI), Wanfang databases, and China Science and Technology Journal Database (VIP) without any restrictions on publication date, sex, or ethnicity. The full search syntax used for PubMed is detailed in Supplementary Material 1. Additionally, the reference lists of all identified studies were manually reviewed to uncover any other relevant citations that the initial search may have missed.

To ensure comprehensiveness, the search strategy included not only peer-reviewed articles but also gray literature such as conference abstracts and unpublished data. This was done to minimize the risk of publication bias and to include all potentially relevant data in the analysis.

Possible publication bias was estimated by visual inspection of the funnel plots. To assess the possible impact of data from individual trials on the overall results, a sensitivity analysis was performed using a sequential leave-one-out analysis. We computed odds ratios (ORs) and confidence interval (CI) using a random-effects model when the studies had significant heterogeneity; otherwise, a fixed-effects model was selected. Results were considered statistically significant with p 0.05.

Two investigators (WJQ. and WYC.) independently assessed the quality of the included studies using the Newcastle–Ottawa Scale. Another investigator (YSZ.) resolved any differences in quality assessment. The quality of non-randomized controlled studies was assessed based on the following criteria:

(1) Representativeness of the Exposed Cohort: This criterion evaluates whether the study population is representative of the general population of interest. A score of 1 is given if the cohort is representative.

(2) Selection of the Non-Exposed Cohort: This criterion assesses the method of selecting the control group. A score of 1 is given if the control group is selected from the same population as the exposed cohort.

(3) Ascertainment of Exposure: This criterion evaluates the accuracy and reliability of the exposure measurement. A score of 1 is given if the exposure is clearly defined and measured.

(4) Outcome of Interest: This criterion assesses whether the study outcome is clearly defined and measured. A score of 1 is given if the outcome is clearly defined and measured.

(5) Comparability of Cohorts on Important Confounders: This criterion evaluates whether the study accounts for important confounders. A score of 1 is given if the study adjusts for at least one important confounder.

(6) Assessment of Outcome: This criterion assesses the method of outcome assessment. A score of 1 is given if the outcome is assessed in a valid and reliable manner.

(7) Length of Follow-Up: This criterion assesses the duration of the follow-up period. A score of 1 is given if the follow-up period is long enough to capture meaningful changes in the outcome of interest, ensuring that the study results are not biased by a short follow-up period.

(8) Adequacy of Follow-Up: This criterion evaluates whether the follow-up period is sufficient to observe the outcome of interest. A score of 1 is given if the follow-up is adequate.

Following the selection of articles for inclusion, we extracted key information including the first author, publication year, study design, data collection year, study population characteristics (such as sample size, mean age, and sex distribution), diagnostic criteria for CHIPs, risk factors being investigated, sources of comorbidity information, number of deaths, and duration of follow-up. When studies presented data on various factors affecting the time of survival, we focused solely on the risk factors associated with in-hospital mortality.

A meta-analysis was conducted to assess the nature and extent of the associations between risk factors and the outcomes under investigation. For each analysis, we employed the effect estimates for individual risk factors as documented in the original publications. Most of the included studies supplied sample sizes for these risk factors. When data were insufficient, we reached out to the authors for further details. Studies were omitted from the analysis if they could not supply the necessary data or if there was no response from the authors.

All statistical analyses were conducted utilizing the Cochrane Review Manager software (RevMan 5.4.1; The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark, 2020). Pooled odds ratios (ORs) and mean differences (MDs) were calculated, complete with 95% confidence intervals (CIs), for both categorical and continuous data sets. The estimation of mean and standard deviation (SD) followed the methodologies detailed by McGrath et al. [16]. In instances where continuous data were presented as median $\pm$ interquartile range (IQR), these estimation techniques were appropriately applied.

Heterogeneity among the studies was assessed using the I² statistic. An I² index $\ge$50% indicated significant heterogeneity among the studies. Given the significant heterogeneity observed in some analyses, a sensitivity analysis was performed to manage this heterogeneity.

Sensitivity analyses were conducted by sequentially excluding individual studies to assess their impact on the overall results. This approach helped to identify studies that may have contributed significantly to the observed heterogeneity. By excluding each study one at a time, we were able to determine the stability of the pooled estimates and identify any potential outliers that could have influenced the results. These studies may have had different methodological approaches, patient populations, or data collection methods, which could have influenced the overall results. By excluding these studies, we were able to obtain more stable and reliable estimates of the associations between the risk factors and in-hospital mortality.

The preliminary search identified 157 relevant articles. We removed 15 duplicates. Following the evaluation of abstracts, the application of our inclusion and exclusion criteria, and an assessment of bias risk, we were left with ten studies [6, 7, 8, 9, 10, 11, 12, 13, 14, 15] (Fig. 1 and Table 2 (Ref. [6, 7, 8, 9, 10, 11, 12, 13, 14, 15])). A flowchart of the selection process is shown in Fig. 1. To date, no randomized trials have been conducted. All studies were cohort studies [6, 7, 8, 9, 10, 11, 12, 13, 14, 15]. A total of 306 patients (100%) were treated with ECMO.

The study characteristics and baseline patient demographics are presented in Table 2. ECMO alone was used as the intervention. The mean age was $>$60 years old. All ten studies used Veno-arterial (V/A) ECMO oxygenation. A total of 306 patients across all studies underwent PCI. Most patients were male (70.3%). Eight studies [6, 7, 8, 10, 11, 12, 13, 15] referred to BMI. Nine studies [6, 7, 8, 9, 10, 11, 12, 13, 15] reported smoking history, hypertension, and diabetes mellitus. Seven studies [7, 9, 10, 11, 12, 13, 14] reported on CS or CA to ECMO.

Ten studies [6, 7, 8, 9, 10, 11, 12, 13, 14, 15] included ECMO duration. Seven studies [6, 8, 10, 11, 12, 13, 15] reported the type of infarction: LAD, left circumflex (LCX), right coronary artery (RCA), and left main coronary artery (LMCA). Nine studies [6, 7, 8, 10, 11, 12, 13, 14, 15] mentioned lactate levels, while eight [6, 8, 10, 11, 12, 13, 14, 15] referred to LVEF. Finally, four [10, 11, 12, 13] referred to MAP, and another four studies [7, 10, 11, 12] reported heart rate (Table 2).

Many studies have offered a comprehensive examination of various factors influencing the survival rates of CHIPs, focusing primarily on statistically significant multivariate ORs. Consequently, the calculation of both univariate and multivariate pooled effect estimates for each determinant was conducted solely when these factors were explored in a minimum of three studies. The multivariate effect estimates are visually represented through forest plots.

Patients that were male (OR = 1.96, 95% CI: 1.12 to 3.42, p = 0.02, I² = 0%) or obese (mean difference, MD:1.52, 95% CI: 1.06 to 1.97, p 0.00001, I² = 47%) had higher in-hospital mortality, which was not age related. Additionally, underlying medical conditions such as smoking history, hypertension, and diabetes, were not related to in-hospital mortality (Fig. 2).

Fig. 2.

The forest plot of population characteristics as high-risk factors for in-hospital mortality in complex high-risk and indicated patients (CHIPs). (A) BMI. (B) Male. (C) Age. (D) Smoking history. (E) Hypertension. (F) Diabetes mellitus. SD, standard deviation; IV, inverse variance; M-H, mantel-haenszel method.

Our research findings indicated that patients with higher lactate levels (MD: 3.15, 95% CI: 2.37 to 3.94, p 0.00001, I² = 29%) and heart rate (MD: 19.28, 95% CI: 9.61 to 28.95, p 0.0001, I² = 0%), but lower LVEF (MD: –4.09, 95% CI: –6.17 to –2.00, p = 0.0001, I² = 0%), and MAP (MD: –24.92, 95% CI: –32.19 to –17.65, p 0.00001, I² = 0%), had a higher risk of in-hospital mortality (Fig. 5). There was considerable heterogeneity in the results among the four studies [6, 13, 14, 15] (MAP: I² = 75%, p = 0.15). For MAP, excluding the study by Pan et al. [13], resulted in 107 patients being included, decreasing heterogeneity (I² = 0%, p = 0.97).

Fig. 5.

The forest plot of biochemical and inspection indicators as high-risk factors for in-hospital mortality in CHIPs. (A) Lactate. (B) LVEF. (C) MAP. (D) Heart rate.

The main finding of this study was the identification of several ECMO-related content risk factors (CS or CA to ECMO, ECMO duration), coronary artery vascular conditions risk factors (type of infarction), population demographic risk factors (BMI, male), and biochemical and inspection indicator risk factors (lactate levels, LVEF, MAP, heart rate) that are associated with an increased risk for mortality in CHIPs. CS or CA to ECMO, lactate levels, and heart rate significantly correlated with mortality, while ECMO duration, LVEF, and MAP had negative correlations. There was a very strong correlation between CS or CA to ECMO, ECMO duration, type of infarction (LAD), and in-hospital mortality in CHIPs. Conversely, some risk factors that are typically linked to a poor prognosis in patients did not show a significant association with an increased risk of in-hospital mortality in CHIPs. These factors include the type of infarction involving the LMCA, age, smoking history, hypertension, and diabetes mellitus.

Our study contributes to the existing literature by specifically addressing gaps in the understanding of ECMO’s impact on post-PCI mortality in CHIPs. While previous meta-analyses have examined the role of ECMO in various clinical settings, none have exclusively targeted its effect on in-hospital mortality following PCI in this patient population. Our study fills this gap by identifying independent high-risk factors associated with increased in-hospital mortality among CHIPs after PCI with ECMO support.

Compared to other meta-analyses, our study provides a more focused analysis on the specific risk factors relevant to CHIPs, such as cardiogenic shock or cardiac arrest to ECMO time, ECMO duration, and type of infarction. This targeted approach allows for a deeper understanding of the factors influencing mortality in this high-risk group, offering valuable insights that can guide clinical decision-making and improve patient outcomes.

The heterogeneity observed among the studies, with an I² index $\ge$50% in some analyses, indicates significant variability in the effect sizes across studies. This heterogeneity may stem from differences in study populations, methodologies, and data collection techniques. To manage this heterogeneity, we performed sensitivity analyses by sequentially excluding individual studies to assess their impact on the overall results. This approach helped identify studies that contributed significantly to the observed heterogeneity and allowed us to obtain more stable and reliable estimates of the associations between the risk factors and in-hospital mortality.

The quality of the included studies was assessed using the Newcastle–Ottawa Scale, with studies classified as moderate or high quality based on their total scores. The quality of evidence directly impacts the reliability of the results obtained from the meta-analysis. By including only studies that met specific quality criteria, we aimed to minimize bias and ensure the robustness of our findings.

The lack of individual participant data (IPD) in our study introduces potential confounding bias, as we are unable to control for all possible confounders at the individual level. This limitation could impact the accuracy of our risk factor associations. Additionally, the small sample size of the included studies may reduce statistical power, limiting our ability to detect smaller but potentially meaningful effects.

To address these issues, future research should aim to collect and analyze IPD to allow for more granular control of confounders and improve the precision of risk factor associations. Additionally, larger sample sizes in future studies will increase statistical power and enable the detection of smaller effect sizes. Finally, a more detailed explanation and analysis of sensitivity analysis results, particularly for low-quality studies, should be provided to ensure a comprehensive understanding of the data’s robustness and reliability.

In summary, our study’s unique contributions lie in its focused approach to examining ECMO’s impact on post-PCI mortality in CHIPs and its comprehensive analysis of heterogeneity and study quality to ensure the results’ validity and reliability. Addressing these identified limitations in future research will further strengthen the field’s understanding of ECMO’s role in CHIPs.

This study had certain limitations. First, the meta-analysis was a secondary analysis; thus, defects of the included studies impact the reliability of the results of the meta-analysis. Additionally, the lack of individual participant data precluded a more detailed analysis of prognosis or the estimation of individual patient outcomes. In addition, a certain degree of heterogeneity was observed among the studies, with a possibility of bias. Therefore, it is necessary to further verify the results of this study using larger samples and greater homogeneity. Future research should focus on validating the risk assessment tools identified in this study within multicenter prospective studies. This approach will help to confirm the generalizability and reliability of these tools across different clinical settings and patient populations. By conducting multicenter studies, researchers can account for variations in patient care and patient demographics, thereby strengthening the validity of the risk assessment models. There is a need for studies that investigate the impact of dynamic lactate monitoring and targeted interventions on patient outcomes. Given the association between elevated lactate levels and mortality identified in our study, understanding how real-time lactate monitoring can guide clinical interventions is crucial. Future studies should explore how changes in lactate levels over time can inform treatment decisions and improve patient survival rates.

In addition, this study did not have a registered protocol. The study was initiated based on the urgency to address a clinical question without the foresight of protocol registration, which may impact the generalizability of our findings. Future studies in this area should aim to prospectively register their protocols to enhance the transparency and credibility of the research process.

Meanwhile, our study’s generalizability is somewhat limited by the predominance of regional data sources, which may not reflect global variations in CHIPs outcomes following PCI with ECMO support. The concentrated geographic focus could introduce biases, affecting the universality of our conclusions. Future studies should expand their scope to include a more diverse range of geographic regions to capture a broader spectrum of patient experiences and healthcare practices. This will help in developing more comprehensive and globally applicable treatment guidelines for CHIPs.

Finally, as a meta-analysis, our study relies on data from existing literature, which may limit our ability to quantitatively assess the impact of specific risk factors. The main purpose of a meta-analysis is to qualitatively evaluate the association between risk factors and in-hospital mortality, rather than to quantitatively predict the exact percentage increase in risk. Therefore, although we can confirm the significant association between risk factors and increased in-hospital mortality, we cannot precisely quantify the extent of this increased risk.