This study was a retrospective observational cohort study with data from the publicly available MIMIC-IV database (version 2.2). MIMIC-IV encompasses records from 73,181 ICU admissions at Beth Israel Deaconess Medical Center (Boston, MA, USA) between 2008 and 2019 (26, 27). The first author was granted access the database (Record ID: 56270677) (28, 29). The study was reported according to the STROBE guidelines (30). Any patients diagnosed with CHF were eligible for inclusion in this analysis. We employed both ICD9 and ICD10 codes for patient identification ( Supplementary Table 1 ). In instances of multiple ICU admissions, we selected the record from the first admission. Exclusion criteria were: a) patients under the age of 18 years; b) an ICU stay shorter than 24 hours; c) the absence of essential data, including demographics, the six criticality scores, and the components needed to construct FI_Lab.

The baseline patient data were extracted from the MIMIC-IV database, including demographics, admission location, and severity assessments conducted within 24 hours of ICU admission. It included six severity assessments which covered a range of measures: sequential organ failure assessment (SOFA), logistic organ dysfunction system (LODS), systemic inflammatory response syndrome (SIRS), Oxford acute severity of illness score (OASIS), acute physiology score III (APS III), and simplified acute physiology score II (SAPS II), along with the Charlson comorbidity index (CCI). According to the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines, AKI is defined by a ≥0.3 mg/dL rise in serum creatinine from baseline within 48 hours, an increase to 1.5 times the baseline level recorded during the preceding 7 days, or a urine output of less than 0.5 mL/kg/hour for 6 hours or more (31).

To assess the frailty status, we developed FI_Lab score utilizing 33 test items sourced from the MIMIC-IV database. They encompassed data from 24 hours before and after the enrolled patients’ ICU admissions. In the present study, the actual FI_Lab range was determined to be 0.06 to 0.88, with a higher FI_Lab score indicative of a more severe degree of frailty. The ranges for the FI_Lab were non-frail (<0.36), pre-frail (0.36-0.45), and frail (>0.45), according to earlier investigations (22–24). We also analyzed the FI_Lab score as a continuous variable. The detailed construction methodology and specific items are explicated in the Supplementary Materials ( Supplementary Tables 1, 2 ).

The investigation began at the time that patients were admitted to the ICU. The primary outcomes of this study were patient mortality, encompassing both in-hospital and one-year mortality. The secondary outcomes were AKI incidence 24 hours after ICU admission and the administration of renal replacement therapy (RRT) in patients experiencing concurrent AKI.

Continuous variables were reported as mean ± standard deviation (SD) or median with interquartile range (IQR), whereas categorical variables were presented as counts and percentages (%). A t-test was employed for continuous variables with normal distributions, and a Kruskal-Wallis H-test was applied otherwise. Categorical variables were compared using the Pearson χ2 test.

Kaplan-Meier survival analysis was applied to determine the primary outcomes event incidence among distinct FI_Lab levels, with the log-rank test used to ascertain inter-group statistical significance. The assumption of proportional hazards in Cox analysis, verified through Kaplan-Meier curves, was met for 1-year mortality. A Cox regression model analyzed the relationship between FI_Lab and one-year mortality. In assessing the relationship between FI_Lab and 1-year mortality, we excluded patients who died during hospitalization. For secondary outcomes, we utilized the Fine-Gray subdistribution hazard model, which accounts for competing risks of death and expresses the associations as subdistribution hazard ratios (32). The cumulative incidence function curves was used to determine the incidence rates of secondary outcome events at different levels of FI_Lab. Logistic regression examined the association between FI_Lab and in-hospital mortality, along with the occurrence of AKI 24 hours after ICU admission. Adjustments for age, sex, ethnicity, admission location, SOFA score, and CCI were implemented in the multivariate models. The association between FI_Lab and clinical outcomes was further examined using restricted cubic spline regression models.

To ensure our results’ reliability, we performed two sensitivity analyses. APS III and SAPS II were adjusted for disease severity in multivariable regression models of FI_Lab with in-hospital mortality. We evaluated the effect of FI_Lab as a categorical variable in conjunction with the six disease severity scores to predict outcomes.

Subgroup analyses, categorized by sex, age group (≤65 and >65 years), obesity, diabetes, and hypertension were conducted to assess the uniformity in the prognostic influence of FI_Lab on the primary outcome. Additionally, we examined the interactions between FI_Lab and these stratified variables to identify any potential moderating effects.

The model’s predictive efficacy was evaluated using the area under the curve (AUC), integrated discrimination improvement (IDI), and net reclassification improvement (NRI). A DeLong test was applied to compare the AUC, both incorporating and excluding the FI_Lab score, to assess the incremental value of integrating FI_Lab as a continuous variable within the six disease severity scores in predicting adverse outcomes. We additionally reported the AUC (95% CI) for all laboratory variables in the adjusted model and used the DeLong test to compare the predictive accuracy of individual laboratory variables with that of FI_Lab.

All statistical analyses were executed using SPSS 18.0 for Windows (SPSS Inc., Chicago, IL, USA) and R (version 4.2.1), with a two-tailed p-value of <0.05 deemed statistically significant.

Among the initial cohort of 11,328 patients presenting with CHF upon their inaugural admission to the ICU, our study ultimately encompassed 3,273 individuals who had surpassed a 24-hour ICU stay, and for whom there was adequate data to establish the FI_Lab ( Figure 1 ). The demographic and clinical characteristics of the study population were stratified and described according to the frailty levels defined by FI_Lab. The non-frail patients were slightly older than the frail and pre-frail patients (median ages: 75.0, 74.9, 74.6 years, respectively, p < 0.05). There were no significant disparities in the sex distribution across the groups. Patients characterized as frail or pre-frail presented with a greater burden of comorbidities than non-frail patients (CCI points, 8, 8, 7, respectively, p < 0.001). Moreover, disease severity markedly differed, with frail patients exhibiting significantly higher SOFA scores than both pre-frail and non-frail groups (9, 6, 5, respectively, p < 0.001) ( Table 1 ; Supplementary Table 3 ).

For in-hospital mortality, the Kaplan-Meier survival curves reveal a statistically significant difference across the FI_Lab groups (non-frail: 6.1%; pre-frail: 12.7%; frail: 32.5%, log-rank test p < 0.001, Figure 2A ). Additionally, FI_Lab exhibited a significant association with in-hospital mortality within a multivariable model, both as a continuous variable (per 0.01-score increase: odds ratio (OR)=1.042, 95% confidential interval (CI):1.032-1.052; Figure 3A ), and as a categorical variable (pre-frail: OR=1.58, 95% CI: 1.148-2.174; frail: OR=3.221, 95% CI 2.341–4.432; Table 2 ). Subgroup analyses revealed that the association between frailty and in-hospital mortality was more pronounced in diabetic patients than in non-diabetic patients (p-interaction<0.01, Figure 4 ).

For 1-year mortality, elevated FI_Lab scores were associated with higher 1-year mortality (non-frail: 17.2%; pre-frail: 27.6%; frail: 36.7%; log-rank test p < 0.001, Figure 2B ). After adjusting for multiple confounding factors, patients with elevated FI_Lab exhibited an increased risk of 1-year mortality (per 0.01-score: hazard ratio (HR) = 1.03, 95% CI: 1.023-1.038; Figure 3B ). Moreover, pre-frail and frail patients demonstrated significantly elevated risks of 1-year mortality in comparison to non-frail individuals (HR=1.495, 95% CI: 1.228-1.821 and HR=2.152, 95% CI: 1.730-2.678, respectively; Table 2 ). Subgroup analyses revealed a heightened risk of 1-year mortality among frail patients over 65 years of age and those without obesity (p-interaction < 0.05, Supplementary Figure 1 ).

The cumulative incidence function curves demonstrate a significant disparity in the likelihood of AKI occurrence 24 hours after ICU admission between frail and pre-frail patients compared to non-frail patients (p = 0.002, Figure 2C ). After accounting for the competing risk of death, frail and pre-frail patients with concomitant AKI exhibited a heightened risk of subsequent RRT utilization compared to non-frail patients (p < 0.001, Figure 2D ). In the fully adjusted model, frailty was associated with the occurrence of AKI (OR=1.403, 95% CI: 1.031-1.909). There was also a significant correlation between frailty and increased utilization of RRT (subdistribution HR=2.89, 95% CI: 1.79-4.65) ( Supplementary Table 4 ). After adjusting for both sex and age, the restricted cubic spline regression models demonstrated a dose-response relationship between FI_Lab and the incidence of AKI, as well as the utilization of RRT (p-nonlinear=0.362 and p-nonlinear=0.028, Figures 3C, D ).

Sensitivity analyses indicate our findings’ robustness. Multivariable regression models, adjusted for APS III and SAPS II, affirmed the continued significance of FI_Lab’s association with in-hospital mortality. The added predictive value of FI_Lab, whether used as a categorical variable in combination with the 6 scores or as a continuous variable, remained consistent ( Supplementary Tables 5, 6 ).

For the primary outcomes, the APS III scores exhibited the highest discriminatory capacity for in-hospital mortality (AUC = 0.80, 95% CI 0.78-0.82). Incorporating the FI_Lab into the APS III score improved its ability to identify individuals who had died in the hospital (ΔAUC = 0.008, p < 0.01), yielding an IDI of 0.012 (95% CI 0.006–0.017) and an NRI of 0.272 (95% CI 0.181–0.362). Moreover, the SAPS II score demonstrated superior discriminatory ability for 1-year mortality (AUC = 0.64, 95% CI 0.61-0.66). The incorporation of the FI_Lab further improved the SAPS II score’s capacity to differentiate patients with 1-year mortality (ΔAUC = 0.015, p < 0.05). This adjustment resulted in an IDI of 0.014 (95% CI 0.009 - 0.018) and an NRI of 0.28 (95% CI 0.196 - 0.365) ( Supplementary Table 7 ).

For the secondary outcomes, the OASIS score exhibited the best predictive capability for AKI occurrence (AUC = 0.75, 95% CI 0.73-0.77). The incorporation of FI_Lab into the OASIS score did not significantly improve its predictive ability (ΔAUC=0.005, p = 0.145). We observed that the SOFA score could effectively distinguish the use of RRT in CHF patients with AKI (AUC=0.80, 95% CI 0.77-0.82). Furthermore, FI_Lab contributed to the improved discriminatory and reclassification performance of the SOFA score for RRT utilization (ΔAUC=0.015, p < 0.001; NRI=0.302, 95% CI 0.19-0.414) ( Supplementary Table 7 ). In addition, the AUC for all laboratory variables included in the FI_Lab, as presented in the adjusted model, is detailed in Supplementary Tables 8, 9 . The results indicate that FI_Lab is more accurate than each individual variable, with the difference being statistically significant (p < 0.001).

This study’s limitations should not be overlooked. Firstly, our inclusion of patients with critical heart failure admitted to the ICU raises the possibility that FI_Lab may have been influenced by acute illness. Thus, the use of FI_Lab might not have entirely prevented the interpretation of disease severity rather than sole assessment of frailty in certain patients. Secondly, the lack of a uniform consensus on the diagnostic threshold for frailty with FI_Lab led us to define non-frail, pre-frail, and frailty using FI_Lab tertiles. Thirdly, the absence of information on cardiac function class and left ventricular ejection fraction in the MIMIC database, which are crucial for patients with CHF, hindered further exploration. Lastly, while the results of the sensitivity analysis demonstrated our findings’ robustness, it is important to note that our study was a retrospective single-center study. Therefore, the findings require further validation through prospective multicenter studies.