This study was carried out in accordance with the Declaration of Helsinki and accredited by local ethics committees. The main ethics committee renounced the need for an informed consent as this was a retrospective analysis and only entirely anonymized data was analysed. The data supporting the results of this study are available from the corresponding author upon reasonable request.
Study design
In this study, patients with non-ischaemic CS treated with or without MCS between 01 January 2010 and 31 December 2020 from 16 tertiary care centres in five countries were collected retrospectively (NCT03313687). Patients were eligible for this study if they presented with CS according to the Society for Cardiovascular Angiography & Interventions (SCAI) CS definition, as retrospectively determined by the local investigators upon reviewing the available case data. Non-ischaemic shock caused by severe decompensation in patients with known heart failure (severe acute-on-chronic heart failure) or unknown heart failure (severe de novo heart failure) was used as the main inclusion criterion.
Patients were not eligible for the study if they presented with acute myocardial infarction or had need for urgent coronary revascularization (irrespective of feasibility); had CS primarily caused by right heart failure (e.g. acute pulmonary embolism); had ECMO-assisted resuscitation; had a post-cardiotomy CS or had other disease which limits life expectancy to below 6 months.
If patients were treated with MCS, the index event was defined as the time of implantation of the first device. If patients were not treated with MCS, baseline was defined as admission to the hospital for out-patients or admission to the intensive care unit for in-patients. For variables determining shock severity, e.g. lactate and pH, the worst value within 6 h before until 6 h after this index event (e.g. a 12-h window) was recorded.
From this registry, only patients with available LVEF measurements obtained via TTE at the time of the index event according to international guidelines were included in this analysis. [
21‐
24]
Statistical analyses
Continuous variables are shown as median (25th percentile, 75th percentile) and analysed using Mann–Whitney test. For binary variables, absolute and relative frequencies are given and comparisons were made using the Fisher`s exact test. Two-level Joint Modelling Multiple Imputation was used to handle missing data. The used clusters were known/unknown history of heart failure. Parameters used for imputation were 20 imputed data sets, 5000 iterations between two successive imputations and 5000 burns in iterations (R package jomo [
25]). Table
1 indicates the variables used for the imputation. The following analyses were calculated in imputed data sets.
Table 1
Characteristics for the overall cohort and divided by LVEF > 20% vs. ≤ 20%
Demographics |
Age, years | 63.0 (51.5, 72.0) | 0 | 66.0 (55.0, 76.0) | 60.0 (49.0, 70.0) | < 0.0001 |
Male sex | 601 (74.5) | 0 | 238 (67.8) | 363 (79.6) | 0.00017 |
Previous heart failure status |
Ischaemic cardiomyopathy* | 221 (33.3) | 17.7 | 96 (35.7) | 125 (31.6 | 0.31 |
Previous heart failure hospitalizations, n | 2.0 (1.0, 3.0) | 59.1 | 1.0 (1.0, 3.0) | 2.0 (1.0, 3.0) | 0.041 |
Previous heart failure treatment |
Beta-blocker (No) | 329 (42.1) | 3.1 | 137 (40.3) | 192 (43.4) | 0.38 |
Renin–angiotensin system inhibitors (No) | 380 (48.5) | 3 | 162 (47.5) | 218 (49.3) | 0.67 |
Mineralocorticoid receptor antagonists (No) | 492 (62.8) | 2.9 | 229 (67.2) | 263 (59.4) | 0.026 |
Implantable cardioverter defibrillator | 269 (33.4) | < 1 | 89 (25.4) | 180 (39.6) | < 0.0001 |
Cardiac resynchronization therapy | 110 (13.7) | 1.8 | 33 (9.4) | 77 (16.9) | 0.0026 |
Comorbidities |
Atrial fibrillation | 353 (44.3) | 1.2 | 148 (42.4) | 205 (45.8) | 0.35 |
Diabetes mellitus | 213 (26.7) | 1 | 107 (30.5) | 106 (23.7) | 0.036 |
Arterial hypertension | 432 (54.2) | 1.2 | 220 (63.0) | 212 (47.3) | < 0.0001 |
Body mass index, kg/m2 | 26.4 (23.3, 30.5) | 3.7 | 27.0 (23.7, 31.2) | 26.1 (23.1, 29.5) | 0.051 |
Prior revascularization | 187 (24.5) | < 1 | 85 (26.0) | 102 (23.4) | 0.44 |
Any intervention for peripheral artery disease | 48 (6.0) | 25 | 22 (6.4) | 26 (5.8) | 0.77 |
Clinical presentation |
Systolic blood pressure, mmHg (worst value within 6 h)* | 82.0 (71.0, 91.0) | 1.5 | 85.0 (72.0, 95.0) | 80.0 (70, 90) | 0.0093 |
Diastolic blood pressure, mmHg (worst value within 6 h)* | 50.0 (40.0, 59.8) | 2.1 | 50.0 (40.0, 59.5) | 50.0 (42.0, 59.5) | 0.12 |
Vasopressor use | 697 (86.5) | < 1 | 301 (86.0) | 396 (86.8) | 0.76 |
Heart rate, bpm (worst value within 6 h) | 100.0 (78.0, 120.0) | 1.5 | 90.0 (72.0, 114.5) | 102 (80, 128) | < 0.0001 |
Lactate, mmol/l (worst value within 6 h)* | 5.1 (2.7, 8.6) | 6.8 | 4.7 (2.7, 8.4) | 5.3 (2.8, 8.9) | 0.18 |
pH (worst value within 6 h)* | 7.3 (7.2, 7.4) | 3.6 | 7.3 (7.2, 7.4) | 7.3 (7.2, 7.4) | 0.13 |
Prior cardiac arrest* | 277 (34.5) | < 1 | 137 (39.3) | 140 (30.9) | 0.016 |
Duration of cardiac arrest, min | 10.0 (0, 25.0) | 56.3 | 10.0 (1.5, 30.0) | 7.0 (0, 20.0) | 0.0081 |
Mechanical ventilation* | 507(64.4) | 2.5 | 233 (67.9) | 274 (61.7) | 0.072 |
Horowitz index (worst value within 6 h) | 201.5 (109.5, 297.0) | 27.5 | 180.0 (95.0, 293.0) | 219.8 (121.3, 300.0) | 0.0088 |
Creatinine, mg/dl (worst value within 6 h) | 1.7 (1.3, 2.6) | 1.5 | 1.7 (1.2, 2.6) | 1.8 (1.3, 2.7) | 0.16 |
eGFR (ml/min) | 37.5 (22.6, 60.1) | 1.5 | 37.5 (22.6, 60.1) | 37.6 (23.4, 57.4 | 0.76 |
SCAI cardiogenic shock class* |
B | 114 (14.7) | | 62 (18.4) | 52 (11.8) | 0.011 |
C | 294 (37.8) | | 134 (39.8) | 160 (36.4) | 0.37 |
D | 204 (26.3) | | 85 (25.2) | 119 (27.0) | 0.62 |
E | 165 (21.2) | | 56 (16.6) | 109 (24.8) | 0.0061 |
Use of mechanical circulatory support |
VA-ECMO | 144 (17.8) | 0 | 41 (11.7) | 103 (22.6) | < 0.0001 |
Impella | 133 (16.5) | 0 | 67 (19.1) | 66 (14.5) | 0.085 |
Impella + VA-ECMO | 83 (10.3) | 0 | 26 (7.4) | 57 (12.5) | 0.019 |
Multivariable mixed effects logistic regression models with centre as a random intercept were fitted in order to investigate patient characteristics (demographics, clinical characteristics, comorbidities, heart failure treatments, index event parameter) independently associated with LVEF dichotomized by median (20), adjusted for age, sex, SCAI class, lactate, prior cardiopulmonary resuscitation, mechanical ventilation and pH.
Survival curves were produced using the Kaplan–Meier method. The number of individuals at risk was given and groups were compared using log-rank test. None of the relevant variables for estimation of mortality rate were imputed, so crude mortality rates for 30-day mortality were estimated by the reverse Kaplan–Meier estimator in original data.
To assess the association between LVEF (as a continuous logarithmic variable and as a binary variable dichotomized by LVEF ≤ 20% vs. > 20%) and 30-day mortality, cohort stratified Cox proportional hazard regression models were fitted, adjusted for age, sex, SCAI class, lactate, prior cardiopulmonary resuscitation, mechanical ventilation and pH. To allow possible non-linearities in the association of LEVF with time-to-event, the previous models were modified modelling LVEF using natural cubic splines. Plots were produced to examine the shape of the association. As a sensitivity analysis, given the heterogeneity of the LVEF measurements between cohorts, the respective LVEF median value per centre was calculated, and a Cox regression model for 30-day mortality was fitted using LVEF dichotomized by this centre-specific median, also adjusted for the above described variables.
Linear mixed models with centre as a random intercept were used to identify the most important predictors for continuous LVEF, adjusted for the above described variables.
To evaluate the impact of LVEF on the association between MCS use and all-cause mortality (e.g. to assess whether MCS use would be associated with mortality only in patients with or without lower LVEF), Cox regression models with an interaction term for MCS use and LVEF were used, adjusted for the same variables described above.
All analyses were performed with R statistical software version 4.1.2. A p value below 0.05 was considered as statistically significant.