The Gulf Cardiogenic Shock (Gulf-CS) registry is a multicenter retrospective registry of CS secondary to angiographically confirmed myocardial infarction [13]. The registry included 1513 patients from 13 tertiary referral centers in 6 Gulf countries (Saudi Arabia, Qatar, Oman, the United Arab Emirates, Kuwait, and Bahrain) between January 2020 and December 2022. Data on baseline demographics, comorbidities, number of pressors or inotropes, angiographic findings, hemodynamic measurements, and mechanical circulatory devices were collected.

Patients with baseline creatinine clearance <15 mL/minute and those with chronic kidney disease requiring dialysis at presentation were excluded. For this analysis, we divided the patients into 2 groups based on their need for RRT: those who did not require RRT (n=1241) and those who required RRT (n=190). We compared the two groups’ baseline variables, angiographic findings, treatment, and in-hospital outcomes. RRT used in this study was hemodialysis (n=54), peritoneal dialysis (n=1), and continuous venovenous hemodialysis (n=135).

Hospital outcomes were compared between patients with and without RRT. The primary outcome was in-hospital all-cause mortality. The secondary outcomes included revascularization, cerebrovascular accident (CVA), the occurrence of major or minor bleeding, and the duration of hospital stay.

CS was defined according to the Standardized Definitions for Cardiogenic Shock Research and Mechanical Circulatory Support Devices as follows: (1) systolic blood pressure <90 mm Hg for ≥30 minutes or the need for vasopressors, inotropes, or mechanical circulatory support (MCS) to maintain a systolic blood pressure ≥90 mm Hg and (2) evidence of tissue hypoperfusion [14]. The time of shock onset is the time when CS is first diagnosed. CVAs include stroke (ischemic or hemorrhagic) and transient ischemic attack. Bleeding events were defined according to the Bleeding Academic Research Consortium (BARC) classification [15]. Creatinine clearance was calculated using the Cockcroft-Gault equation: CrCl=[(140 – age)×weight (kg) × (0.85 if female)]/[72×serum creatinine (mg/dL)]. We chose this method over estimated glomerular filtration rate equations because of its established use in the CS literature, its incorporation of body weight (relevant to acute volume shifts), and its direct applicability to medication dosing decisions. Right ventricular (RV) dysfunction was defined by echocardiographic criteria: tricuspid annular plane systolic excursion <17 mm, S’ wave <12 cm/s, and tricuspid annular plane systolic excursion to pulmonary artery systolic pressure ratio <0.34 [16]. Baseline data, including laboratory and echocardiographic data, were measured at the time of hospital presentation. The number of diseased vessels was defined as the total count of coronary arterial branches with angiographically significant stenosis (≥50% luminal obstruction), including major epicardial vessels and their principal branches (left anterior descending artery, diagonal branches, obtuse marginal branches, ramus intermedius, right coronary artery, posterior descending artery, and posterolateral branches), yielding a possible range of 1 to 6 in this cohort. Details of the Gulf-CS registry and variable definitions have already been published [13].

The Institutional Review Board (IRB) of King Faisal Specialist Hospital and Research Center in Jeddah approved the study on January 16, 2023 (IRB 2022‐90: Gulf-CS Registry). Approval was obtained from each center’s ethics committee for all study sites. The IRB waived the need for patient consent due to the study’s observational nature. Patients' privacy was maintained throughout the study, and deidentified data were shared. No compensation was provided to participants.

Statistical analysis was conducted with Stata 18 (StataCorp) and Python 3.11 (Python Software Foundation). Continuous variables were summarized using means (SDs) or medians (Q1-Q3) according to the normal distribution. Categorical data are reported as numbers (%). Comparisons between the two groups were made using t tests, Mann-Whitney tests, chi-square tests, or Fisher exact tests, as appropriate. Missing data were handled using complete case analysis.

The dataset was randomly partitioned into a training set (80%) and a testing set (20%), stratified by the primary outcome of in-hospital RRT. We used the LASSO logistic regression on the training set to identify the most parsimonious set of baseline predictors. The LASSO method applies a penalty that shrinks the coefficients of less important variables toward zero, effectively performing automated variable selection. The tuning parameter (lambda) was selected via 10-fold cross-validation to maximize the area under the receiver operating characteristic curve (AUC).

The model’s discrimination was assessed on the held-out test set using the AUC. We performed two forms of internal validation:

Model calibration was assessed visually by plotting the observed versus predicted probabilities on the testing set.

Model calibration was assessed visually by plotting the observed versus predicted probabilities on the testing set. To improve calibration, we applied Platt scaling (also known as sigmoid calibration) to the LASSO model predictions. Platt scaling fits a logistic regression model to the predicted probabilities from the LASSO model, effectively recalibrating the predictions to better match the observed event rates. This post hoc calibration was performed using 5-fold cross-validation on the training set, and the calibrated model was then applied to the testing set. Calibration performance was evaluated both before and after Platt scaling using calibration plots.

A clinical nomogram was constructed to provide individualized, point-based estimates of the probability of requiring RRT. The nomogram was derived from the LASSO logistic regression model described above, using the penalized coefficients of all 15 candidate variables retained at the optimal regularization parameter. To facilitate clinical application, the standardized LASSO coefficients were backtransformed to their original measurement scales by dividing each coefficient by the corresponding variable’s SD, with continuous variables expressed in their natural clinical units. For each predictor, a point score proportional to its backtransformed coefficient and clinical range was assigned relative to the variable with the largest absolute contribution; the sum of individual point scores maps to a predicted probability of RRT via the logistic function. The nomogram is intended as a bedside decision-support tool to stratify patients with CS-AMI by their risk of acute kidney injury requiring RRT using variables available at or shortly after admission.

To account for potential clustering of outcomes by institution, we fitted a final logistic regression model using the LASSO-selected variables with cluster-robust SEs for each center. Continuous variables were entered into the final model as unstandardized values to yield clinically interpretable odds ratios (ORs). The Society for Cardiovascular Angiography and Interventions (SCAI) shock stage was treated as a categorical variable with stage B as the reference category, as this approach more accurately captures the nonlinear risk transitions between stages. To demonstrate the clinical utility of our model, we compared its performance on the test set against a parsimonious model that included only age and baseline creatinine clearance.

We assessed multicollinearity among the predictors in the final model using the variance inflation factor (VIF). A VIF >5 was considered indicative of significant multicollinearity. All included features had VIF <1.5.

The baseline characteristics of the patients in our study revealed several demographic and clinical differences between those who required RRT and those who did not. The cohort’s mean age was 60.34 (SD 12.14) years, with patients requiring RRT significantly older (mean 64.17, SD 12.14 y) than those who did not (mean 59.75, SD 11.77 y; P<.001). Males comprised most of the cohort (1162/1431, 81.20%), with no significant difference between the 2 groups (P=.46). Patients who required RRT had a greater prevalence of comorbid conditions, including diabetes mellitus (72.11% vs 61.89%; P=.008), peripheral arterial disease (PAD; 11.58% vs 3.71%; P<.001), a history of CVA (11.05% vs 5.72%; P=.005), and congestive heart failure (15.79% vs 10.64%; P=.04). Other comorbidities, such as dyslipidemia, hypertension, previous myocardial infarction, and history of coronary interventions (percutaneous coronary intervention [PCI] and coronary artery bypass grafting [CABG]), were similar between the groups. Patients requiring RRT presented with more severe disease, as evidenced by higher rates of anterior ST-elevation myocardial infarction (STEMI) and cardiac arrest at presentation. Additionally, RV dysfunction was more prevalent among patients requiring renal support (23.28% vs 15.76%; P=.01), further highlighting the increased severity of their clinical presentation.

Laboratory investigations revealed lower hemoglobin, creatinine clearance, sodium bicarbonate levels, and higher potassium levels in patients requiring RRT. Patients who required RRT had a higher SYNTAX score, a greater number of vessels with significant coronary disease, and an advanced SCAI stage than patients who did not require RRT (Table 1).

Inotropic support, MCS, and mechanical ventilation were used more frequently in patients who needed renal support. CABG was performed more frequently in patients requiring RRT, and PCI was more common in patients without RRT (Table 2).

Hospital mortality (76% vs 39%; P<.001) and CVA (11% vs 5%, P=.001) were significantly greater in patients with RRT. Hospitalization duration was prolonged in patients with RRT (10 vs 6 d; P<.001). No differences were found in revascularizations or minor bleeding between the groups (Table 3).

The LASSO model included 1072 in the training dataset and 268 in the test dataset (a total of 1340), while 91 patients were dropped due to missing baseline data. From the baseline variables, the LASSO regression model selected a final set of 15 predictors for RRT. The selected variables and their corresponding coefficients are shown in Figure 1. The 3 strongest predictors were lower creatinine clearance, higher SCAI shock stage, and a greater number of diseased coronary vessels (Table S1 in Multimedia Appendix 1).

The final model demonstrated good discrimination on the held-out testing set, with an AUC of 0.714 (Figure 2). This was significantly better than the parsimonious model containing only age and creatinine clearance, which had an AUC of 0.624 (P<.001 for the difference; Figure 3). The model was well-calibrated, as shown in the calibration plot (Figure S1 in Multimedia Appendix 2).

The model’s performance was confirmed through 2 internal validation methods. A 10-fold cross-validation on the training set yielded a mean AUC of 0.723 (SD 0.050). Furthermore, bootstrapping with 1000 iterations produced a bias-corrected AUC of 0.745 (95% CI 0.730‐0.756), indicating robust and stable performance (Figure S2 in Multimedia Appendix 3).

A nomogram was developed based on the final LASSO model (Figure 4). A nomogram can be used post angiography immediately to estimate an individual patient’s probability of requiring in-hospital RRT by summing the points assigned to each predictor variable and reading the corresponding probability from the total points scale.

In the final logistic regression model accounting for center-level clustering, several predictors remained significantly associated with the need for RRT. These included history of PCI (OR 0.649; P=.007), non–ST-elevation myocardial infarction (NSTEMI) presentation (OR 1.756; P=.02), SCAI stage D (OR 10.859; P=.02) and E (OR 14.613; P=.005), lower hemoglobin (OR 0.892; P=.03), and increased number of diseased vessels (OR 1.351; P=.001; Table 4). No significant multicollinearity was detected among the predictors (all VIF <1.5).

This study used data from the Gulf-CS registry, the first large multinational study focused on patients with CS-AMI from the Gulf region. This study explored hospital and short-term outcomes of patients with CS-AMI across 6 Gulf countries, using the novel CS staging system developed by the SCAI Cardiogenic Shock Working Group (SCAI-CSWG). We used LASSO regression to predict the need for RRT in this high-risk population. Our analysis identified several predictors of RRT, including reduced creatinine clearance, advanced SCAI stage, and significant coronary artery disease affecting multiple vessels.

Renal support was required in 13.28% of patients (190/1431), and those requiring RRT had a significantly worse prognosis. This finding is consistent with previous studies showing that the RRT incidence in patients with CS-AMI is approximately 13% [7,17]. Patients who required renal support were notably older, with a mean age of 64.17 (SD 12.14) years, whereas those who did not require RRT were 59.75 (SD 11.77) years (P<.001), which is also consistent with the literature [9,18]. Comorbidities were also more prevalent in the renal support group, with higher rates of diabetes mellitus (72.11% vs 61.89%; P=.008), PAD (11.58% vs 3.71%; P<.001), previous CVA (11.05% vs 5.72%; P=.005), and congestive heart failure (15.79% vs 10.64%; P=.037). The literature also highlights a high comorbidity burden associated with increased renal support needs [19]. Patients presenting with NSTEMI and those experiencing cardiac arrest were also more likely to require RRT. Specifically, 40.53% (77/190) of patients requiring RRT had NSTEMI, whereas 23.61% (293/1241) of patients in the no-renal-support group did (P<.001). Cardiac arrest was present in 69.47% (132/190) of patients in the RRT group, which was significantly greater than the 52.94% reported in patients who did not require RRT (P<.001). These findings highlight the association between more severe initial presentations and the subsequent need for RRT. In a study by Jeppesen et al [18], 33.5% of patients who experienced out-of-hospital cardiac arrest developed AKI, and 10.1% required RRT.

Regarding echocardiographic findings, RV dysfunction was significantly more prevalent in the renal support group, suggesting that RV failure and venous congestion contribute to worsening renal function. This finding is corroborated by Jain and associates, who demonstrated that RV dysfunction is common in patients with CS and is associated with increased in-hospital mortality [20]. RV failure is associated with poor outcomes and often requires MCS [21]. Furthermore, baseline laboratory marker levels, such as creatinine clearance and hemoglobin, potassium, and lactate levels, were significantly lower in patients who required RRT. The median creatinine clearance was markedly lower in the RRT group (46 mL/min vs 72 mL/min; P<.001), and the baseline lactate level was greater (2.7 mmol/L vs 2.1 mmol/L; P<.001), indicating more pronounced tissue hypoperfusion. This finding is consistent with other studies that associated worse laboratory abnormalities with renal support [22].

Finally, angiographic findings such as the SYNTAX score, number of major vessels involved, and SCAI stage were also obtained. The data suggested that the SYNTAX score was greater in the renal support group, with more significant coronary disease in the RRT group. This finding is consistent with prior findings that lesion complexity increases with declining renal function [23]. Significant differences in SCAI stages, with more severe stages (D and E), were also observed in the RRT group (P<.001).

The need for RRT was associated with a significantly higher in-hospital mortality rate. Patients requiring RRT had a mortality rate of 75.79%, whereas 38.76% (144/190 and 481/1241, respectively) of those who did not require RRT (P<.001). This finding is similar to that of a study performed by Lauridsen et al [7], which revealed that RRT in patients with CS-AMI was associated with worse mortality (62% for patients with AKI-RRT vs 36% for patients with non-AKI-RRT), along with increased long-term mortality at 5 years. This significant difference highlights the role of renal failure as a major determinant of poor outcomes in patients with CS-AMI. In addition to higher mortality, patients requiring renal support had significantly longer hospital stays (median of 10 d, IQR 4‐22 vs 6 d; P<.001) and higher rates of major bleeding complications (16.84% vs 7.33%; P<.001). Similarly, external studies have shown a longer length of stay; for example, Vallabhajosula et al [9] reported that patients with CS-AMI requiring hemodialysis had a significantly longer length of stay (mean 18, SD 19 d vs mean 9, SD 10 d). These adverse outcomes likely reflect the increased severity of illness in patients requiring RRT, as well as the risks associated with invasive procedures and anticoagulation in this population.

Other clinical outcomes, such as stroke and bleeding complications, were also more frequent in patients who required RRT. The incidence of cerebrovascular accidents was 11.05% in the renal support group, whereas it was 4.92% in those not requiring RRT (P=.001). Major bleeding occurred in 16.84% (32/190) of patients receiving renal support, which was more than double the rate in those who did not receive RRT (91/1241, 7.33%; P<.001). Freund and colleagues reported from a subanalysis of the CULPRIT-SHOCK trial that bleeding events were common in patients with infarct-related CS (21.5%), particularly in those with MCS, which is often used with renal support [24]. These complications likely contribute to the higher mortality and morbidity in this group, highlighting the complexity of managing these critically ill patients.

LASSO regression is used in this study to handle the large number of potential predictors and avoid overfitting. By penalizing the absolute size of the regression coefficients, LASSO effectively selects the most relevant variables, making it an ideal tool for this analysis. LASSO identified several predictors of renal support in patients with CS-AMI. Key predictors included lower creatinine clearance, advanced SCAI stage, and multivessel coronary artery disease. Previous studies, such as that by Marenzi et al [8], have established that mechanical ventilation (OR 22.6, 95% CI 14.2‐36.0) and admission creatinine >1.5 mg/dL (OR 16.9, 95% CI 10.4‐27.3) were predictors of renal support. Moreover, Vallabhajosyula et al [9] reported that older age and greater comorbidity burden were associated with AKI and subsequent hemodialysis in patients with CS-AMI.

Contrary to expectations, NSTEMI presentation was associated with a higher risk of RRT compared to STEMI. This counterintuitive finding may reflect selection bias, where patients with NSTEMI developing CS represent a particularly high-risk subset with more comorbidities (diabetes and PAD) and chronic cardiac dysfunction. Alternatively, competing mortality risk may play a role, as patients with STEMI with severe shock may experience early death before developing AKI requiring RRT. This finding warrants further investigation in prospective studies and may indicate that patients with NSTEMI may benefit from more aggressive renal-protective strategies. This finding may also reflect differences in patient management strategies or in the underlying pathophysiology of AKI across subgroups. Patients with STEMI might benefit from more targeted revascularization strategies to mitigate renal injury. This is further supported by the CULPRIT-SHOCK trial, which demonstrated that, compared with immediate multivessel PCI, PCI of the culprit lesion was associated with a lower 30-day risk of death or severe renal failure requiring RRT [25].

A notable divergence exists between the variable importance rankings from the LASSO model and the significance findings from the cluster-robust logistic regression. This apparent contradiction reflects fundamental differences in the purpose and methodology of the 2 approaches rather than a true inconsistency. The LASSO model is optimized for prediction: it assigns coefficient weights based on the variable’s contribution to discriminating between patients who will and will not require RRT across the entire dataset, without regard to statistical significance. The cluster-robust logistic regression, by contrast, is designed for inference: it estimates the independent association of each variable with the outcome while explicitly accounting for the nonindependence of observations within the same center. The application of cluster-robust (sandwich) standard errors inflates the variance of coefficient estimates to correct for within-center correlation, which reduces statistical power. The 2 analyses are complementary rather than contradictory, and the clinical utility of the nomogram is not diminished by the non-significance of individual predictors in the inferential model.

Our prediction model, presented as a user-friendly nomogram (Figure 4), is intended for use immediately post angiography, once the coronary anatomy has been defined. This represents a critical decision point in the management of CS-AMI, where clinicians must rapidly decide on the aggressiveness of hemodynamic support, the initiation of renal-protective strategies, and the need for early nephrology consultation. This early identification, aided by this tool, can trigger a cascade of preventative measures, including the avoidance of nephrotoxic agents, careful fluid management, and early involvement of a multidisciplinary team, including nephrologists, which has been shown to improve outcomes and mortality [10,11]. Finally, these data align with the growing evidence that early RRT initiation improves outcomes in selected patients.

Several limitations warrant consideration. First, our complete-case analysis excluded patients with missing data, which may introduce selection bias. Second, the Cockcroft-Gault equation was used to estimate creatinine clearance; this formula assumes a steady-state serum creatinine level, which is rarely the case in the acute, dynamic setting of CS and may therefore underestimate the true degree of renal impairment. Third, the inclusion of the number of diseased vessels, an angiographic variable, means the model is intended for postcatheterization risk assessment and is not applicable for precatheterization risk stratification in the emergency department. Fourth, we did not perform a competing risks analysis to account for in-hospital mortality as a competing event for RRT because our registry lacks precise timing data for both events. Fifth, although the model was validated on an internal testing set and via bootstrapping, external validation in an independent cohort is needed before widespread clinical adoption. Sixth, a further limitation concerns the treatment of SCAI shock stage in the predictive model. The LASSO model retains a single continuous coefficient for SCAI stage, and the nomogram plots the stages on a single ordinal axis. This approach assumes equidistant and linear increments in risk between consecutive stages, which may not accurately reflect the disproportionate increase in RRT risk. However, the SCAI stage was included in the model as a single variable to simplify the nomogram for clinical use and to account for the small number of events in some SCAI stages. Finally, the wide confidence intervals observed for the SCAI stage comparisons are attributable to sparse outcome data in the reference category (stage B).

The study highlights significant differences in baseline characteristics and clinical outcomes between patients with CS who required renal support and those who did not. Key findings indicate that patients who needed renal support were older, had a greater prevalence of comorbidities, and presented with more severe clinical conditions, including higher rates of anterior STEMI and RV dysfunction. The analysis demonstrated that patients requiring renal replacement therapy had markedly higher hospital mortality rates and longer hospital stays, underscoring the gravity of their condition. Furthermore, the study identified critical predictors for RRT, including baseline creatinine clearance and advanced SCAI stage.

The development and internal validation of a 15-variable model allows accurate prediction of the need for RRT in patients with CS-AMI using only baseline data. Overall, these findings emphasize the importance of early identification and management of high-risk patients with CS to improve outcomes and tailor interventions more effectively. Further research may enhance the predictive capabilities and clinical applicability of these findings. Moreover, multidisciplinary management of CS, including nephrologists, is warranted.