ABSTRACT
Aim
To compare four electronic health record (EHR)‐based frailty tools—the Hospital Frailty Risk Score (HFRS), Johns Hopkins Adjusted Clinical Groups Frailty Indicator (CFI), Electronic Frailty Index (eFI) and a Laboratory‐based Frailty Index (FI‐Lab)—in predicting progression and all‐cause mortality in hospitalised patients with chronic kidney disease (CKD).
Methods
This retrospective study evaluated the indices in two cohorts: a single‐centre cohort (n = 5715) for CKD progression and the Medical Information Mart for Intensive Care (MIMIC) database (n = 2674) for mortality. We used Cox proportional hazards regression for association analyses. The incremental predictive value of adding frailty indices to established risk models was quantified using the area under the curve (AUC), net reclassification improvement (NRI) and integrated discrimination improvement (IDI).
Results
Correlations between the frailty indices were weak to moderate (Spearman's ρ = 0.205–0.451). In adjusted analyses, CFI, eFI and FI‐Lab were associated with a higher risk of CKD progression, whereas HFRS was not. In the MIMIC cohort, all four indices were significantly associated with all‐cause mortality. Notably, the CFI association was non‐significant in patients < 65 years for both CKD progression and 28‐day mortality. For predictive enhancement, adding FI‐Lab and eFI to established CKD risk models significantly improved progression prediction (ΔAUC p < 0.05), yielding substantial reclassification (NRI: 0.376–0.498) and discrimination (IDI: 0.032–0.048). For mortality prediction, all indices improved baseline severity scores, with FI‐Lab providing the greatest incremental value.
Conclusion
Among the evaluated EHR‐based frailty indices, FI‐Lab offers the most robust utility for risk stratification in hospitalised patients with CKD, followed closely by eFI.
Keywords: chronic renal insufficiency, electronic health records, frailty, mortality, risk assessment
Summary at a Glance
This study compared four electronic health record‐based frailty tools for predicting adverse outcomes in hospitalised patients with chronic kidney disease. The Laboratory‐based Frailty Index (FI‐Lab) and Electronic Frailty Index (eFI) demonstrated the most robust utility, significantly improving prediction for disease progression and all‐cause mortality.
1. Introduction
Frailty, a clinical hallmark of accelerated biological aging characterised by multisystem dysregulation and diminished physiological resilience, is strongly associated with adverse outcomes such as falls, functional decline and excess mortality [1, 2]. Chronic kidney disease (CKD), affecting approximately 10% of the global population with prevalence escalating with age, confers a 36% increased risk of all‐cause mortality [3]. Emerging evidence identifies frailty as an independent predictor of mortality and hospitalisation in CKD populations [4]. Both baseline frailty and longitudinal changes in frailty status correlate with reduced estimated glomerular filtration rate (eGFR), suggesting that interventions targeting frailty may attenuate renal function decline [5]. These findings highlight the imperative for integrating frailty assessment into CKD clinical management. Practical methods for assessing frailty are needed.
Despite the clinical importance of frailty assessment, there is currently no gold‐standard diagnostic tool. Two main conceptual models, the Fried Frailty Phenotype, which focuses on physical manifestations, and the Cumulative Deficit Model (frailty index, FI), guide frailty evaluation. Although both models have diagnostic validity, their complexity and time‐consuming nature pose operational challenges, restricting their routine clinical use [6]. This gap has driven the development of electronic frailty metrics that utilise routinely collected electronic health record (EHR) data. These metrics hold great promise for assessing hospitalised patients. Validated multidimensional assessment tools, including the Hospital Frailty Risk Score (HFRS) [7], Johns Hopkins Claims‐Based Frailty Indicator (CFI) [8], Electronic Frailty Index (eFI) [9] and Laboratory Variable‐Derived Frailty Index (FI‐Lab) [10], have been successfully operationalised using electronic health data. Previous research has illustrated associations between individual electronic frailty metrics and adverse events among patients with advanced CKD. However, no studies have systematically compared and characterised their application value in predicting adverse outcomes among hospitalised CKD patients. This knowledge is essential for improving CKD management and decision‐making by integrating these tools into clinical workflows.
In this study, we harnessed EHR data of a hospital and the Medical Information Mart for Intensive Care (MIMIC) database to systematically compare four established electronic frailty metrics: HFRS, CFI, eFI and FI‐Lab. Our study objectives were threefold: (1) to assess the intercorrelation among these frailty instruments; (2) to explore the associations between each frailty metric and the risk of CKD progression or mortality risk in hospitalised CKD patients and (3) to evaluate the incremental predictive value of integrating these frailty metrics into established CKD risk stratification tools and classical disease severity scores for predicting CKD progression and all cause mortality.
2. Methods
2.1. Study Design and Population
This retrospective study utilised two distinct cohorts to assess EHR‐based frailty indices. The Hospital CKD cohort, from our institution (The Third Affiliated Hospital of Sun Yat‐sen University, 2012–2018), included 5715 hospitalised patients with CKD to evaluate associations with adverse renal outcomes. The MIMIC CKD cohort comprised 2674 CKD patients from the public MIMI database (2008–2022) and was used to assess associations with short‐ and long‐term all‐cause mortality [11]. For both cohorts, we included adult patients (≥ 18 years) at their index hospitalisation. Common exclusion criteria were pre‐existing end‐stage kidney disease (ESKD) at baseline, incomplete data for frailty assessment and lack of follow‐up (Figure S1). Patients in the institutional cohort were additionally excluded if they experienced a renal outcome during the index admission. For the final analytic cohorts, we extracted demographics, International Classification of Diseases, 10th Revision (ICD‐10) codes and laboratory results.
This retrospective cohort study used de‐identified data from The Third Affiliated Hospital of Sun Yat‐sen University (Ethics Committee Approval No. [2019] 02‐427‐01) and the MIMIC‐IV database (v3.1, Human Subjects Research Training Certificate ID: 57964285) with IRB approval from Beth Israel Deaconess Medical Center. The study adhered to the Declaration of Helsinki, and all data were anonymised prior to analysis.
2.2. CKD
We estimated the glomerular filtration rate (eGFR) using the Chronic Kidney Disease Epidemiology Collaboration (CKD‐EPI) equation [12]. The definition of CKD was adapted for each cohort. In the Hospital CKD cohort, CKD was defined by the Kidney Disease: Improving Global Outcomes (KDIGO) criteria: eGFR < 60 mL/min/1.73 m2 or evidence of kidney damage for ≥ 3 months [13]. Kidney damage was defined as proteinuria (albumin‐to‐creatinine ratio [ACR] ≥ 30 mg/g or abnormal qualitative urinalysis) or glomerular hematuria (≥ 80% dysmorphic erythrocytes) [14]. In the MIMIC CKD cohort, where longitudinal data were limited, CKD was defined using a comprehensive set of ICD‐10 codes adapted from prior EHR‐based studies. For comparability, this same ICD‐10 code set also supplemented case identification in the Hospital CKD cohort (Table S1) [15, 16, 17]. Although we acknowledge some codes lack a chronicity qualifier, this inclusive strategy was intentionally adopted to mitigate the known low sensitivity of using only N18.x–N19.x codes, thereby building a more robust study population [18].
2.3. EHR‐Derived Frailty Metrics
The development and validation of the HFRS, CFI, eFI and FI‐Lab are well‐established [7, 8, 9, 10]. The HFRS, derived from cluster analysis of hospitalized older adults, utilizes ICD‐10 diagnosis codes to predict frailty [7]. The CFI, anchored to Fried's phenotype, was originally developed using ICD‐9 codes; we applied Kundi et al.'s validated mapping to ICD‐10 equivalents [8, 19]. The eFI adopts a cumulative deficits model across six domains (diagnoses, laboratory values, vital signs, medications, social history and functional status), while the FI‐Lab focuses on laboratory test deficits [9, 10, 20]. We computed all four indices using structured EHR data, with variable definitions and scoring algorithms detailed in Tables S3–S7. To harmonize the scale of frailty metrics, CFI, eFI and FI‐Lab values were linearly transformed by multiplying raw scores by 100, converting the original range to a 0–100 scale.
2.4. Outcome
Primary outcomes were defined separately for each cohort. For the Hospital CKD cohort, the primary outcome was a composite kidney endpoint: the first occurrence of a ≥ 40% decline in eGFR from baseline, serum creatinine doubling, or the onset of kidney failure (eGFR < 15 mL/min/1.73 m2). This composite is a validated surrogate for ESKD and strongly predicts adverse clinical events. For the MIMIC CKD cohort, the primary outcomes were 28‐ and 365‐day all‐cause mortality following hospital admission.
2.5. Baseline Covariates
We extracted a comprehensive set of clinically relevant variables from EHRs. Variables common to both cohorts included demographics (age, sex), clinical measurements (systolic/diastolic blood pressure, baseline eGFR) and major comorbidities (diabetes, hypertension, cardiovascular disease, cancer). Cohort‐specific data included behavioural factors (smoking, alcohol use) and ACR for the Hospital CKD cohort, and patient race for the MIMIC CKD cohort. For Hospital CKD participants lacking direct ACR measurements, ACR was estimated from urine protein‐creatinine ratio (UPCR) or dipstick protein categories using validated conversion equations [21]. Comorbidities were identified using ICD‐10 codes (Table S2) [22, 23, 24].
2.6. Statistical Analysis
Statistical analyses were performed to compare groups. We assessed continuous variables for normality using the Shapiro–Wilk test, or the Kolmogorov–Smirnov test for the large Hospital CKD cohort (n = 5715). As distributions were non‐normal, these variables are presented as median [interquartile range (IQR)] and were compared using the Mann–Whitney U test. Categorical variables are summarized as frequency (n, %) and were compared using the chi‐squared (χ 2) test. We conducted a complete‐case analysis, as missing covariate data were minimal (< 5%) and assumed to be missing completely at random.
We evaluated pairwise correlations among eFIs using Spearman's rank correlation (ρ) and assessed inter‐cohort differences in these coefficients via bootstrapping. The associations of frailty indices with outcomes were analysed using Cox proportional hazards models to calculate hazard ratios (HRs) and 95% confidence intervals (CIs). We constructed both univariable (Model 1) and multivariable‐adjusted (Model 2) models. For Model 2, both cohorts were adjusted for a common set of covariates: age, sex, systolic blood pressure, baseline eGFR, diabetes, hypertension, cardiovascular disease and cancer. To account for cohort‐specific data availability, the Hospital CKD cohort model was additionally adjusted for smoking, alcohol use and ACR, while the MIMIC CKD cohort model was additionally adjusted for race. We further explored potential non‐linear dose–response relationships in the multivariable models using restricted cubic splines with four knots.
We evaluated the incremental predictive value of four frailty indices (HFRS, CFI, eFI and FI‐Lab) when added to established risk models. For CKD progression, the baseline models were the KDIGO heatmap and the 4‐variable Kidney Failure Risk Equation (KFRE) [25, 26]. For mortality, baseline models included the Sequential Organ Failure Assessment (SOFA), Acute Physiology Score III (APSIII) and Simplified Acute Physiology Score II (SAPSII). We primarily assessed model discrimination using the time‐dependent area under the receiver operating characteristic curve (AUC). AUC curves were plotted across the entire follow‐up period, and specific values were calculated at clinically relevant time points such as the median follow‐up. To ensure robust estimation when the median coincided with the maximum follow‐up time, the AUC was calculated 1 day prior (t − 1) [27]. The significance of AUC differences between nested models was determined by a Wald test based on influence functions. Furthermore, improvements in risk stratification were quantified using the continuous net reclassification improvement (NRI) and integrated discrimination improvement (IDI). As continuous NRI is known to be sensitive to marginal model differences and may overestimate reclassification gains [28], AUC‐based discrimination was considered the primary metric for clinical interpretation, with NRI and IDI serving as supplementary evidence. Statistical analyses were conducted using R (version 4.5.0). A two‐tailed p < 0.05 defined statistical significance throughout the study.
3. Results
3.1. Participant Characteristics
The Hospital CKD cohort included 5715 participants (Table 1), with a median age of 64.0 (IQR, 51.0–75.0) years; 61.47% were male. Over a mean follow‐up of 21.38 months, 22.45% (n = 1283) developed CKD progression. Compared to non‐progressors, patients with CKD progression were older, more frequently male and had higher systolic and diastolic blood pressure, serum creatinine, ACR, along with a greater prevalence of smoking, diabetes, hypertension and cardiovascular disease. Notably, all four frailty indices were also significantly higher in the progression group (all p < 0.05 for comparisons).
TABLE 1.
Baseline characteristics of the participants, stratified by CKD progression and all‐cause mortality.
| Characteristics | Hospital CKD cohort | MIMIC CKD cohort (28‐day mortality) | MIMIC CKD cohort (365‐day mortality) | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Non‐CKD progression | CKD progression | p | Survivor | Non‐survivors | p | Survivors | Non‐survivors | p | |
| N (%) | 4432 (75.55%) | 1283 (22.45%) | 2031 (75.95%) | 643 (24.05%) | 1469 (54.94%) | 1205 (45.06%) | |||
| Age (years), median (IQR) | 63 (50, 75) | 65 (54, 76) | < 0.001 | 72.35 (63.17, 80.78) | 75.58 (66.35, 83.22) | < 0.001 | 71.15 (61.89, 79.46) | 75.30 (66.53, 83.56) | < 0.001 |
| Age group, n% | 0.003 | < 0.001 | < 0.001 | ||||||
| ≥ 65 | 2096 (47.29%) | 668 (52.07%) | 1433 (70.56%) | 511 (79.47%) | 996 (67.80%) | 948 (78.67%) | |||
| < 65 | 2336 (52.71%) | 615 (47.93%) | 598 (29.44%) | 132 (20.53%) | 473 (32.20%) | 257 (21.33%) | |||
| Gender, n% | 0.027 | 0.342 | 0.488 | ||||||
| Female | 1742 (39.31%) | 460 (35.85%) | 761 (37.47%) | 255 (39.66%) | 549 (37.37%) | 467 (38.76%) | |||
| Male | 2690 (60.69%) | 823 (64.15%) | 1270 (62.53%) | 388 (60.34%) | 920 (62.63%) | 738 (61.24%) | |||
| Race and ethnicity, n% | / | / | / | 0.006 | 0.329 | ||||
| White | / | / | / | 1258 (61.94%) | 379 (58.94%) | 881 (59.97%) | 756 (62.74%) | ||
| Black | / | / | / | 288 (14.18%) | 73 (11.35%) | 207 (14.09%) | 154 (12.78%) | ||
| Other | / | / | / | 485 (23.88%) | 191 (29.70%) | 381 (25.94%) | 295 (24.48%) | ||
| Smoking status, n% | 0.008 | / | / | / | / | / | / | ||
| Never smokers | 3388 (77.49%) | 945 (73.89%) | / | / | / | / | / | / | |
| Ever smokers | 984 (22.51%) | 334 (26.11%) | / | / | / | / | / | / | |
| Drinking status, n% | 0.072 | / | / | / | / | / | / | ||
| Never drinkers | 3800 (86.92%) | 1086 (84.91%) | / | / | / | / | / | / | |
| Ever drinkers | 572 (13.08%) | 193 (15.09%) | / | / | / | / | / | / | |
| SBP (mmHg), median (IQR) | 135 (121, 149) | 142 (127, 159) | < 0.001 | 121 (104, 142) | 116 (99, 134) | < 0.001 | 122 (105, 144) | 118 (101, 137) | < 0.001 |
| DBP (mmHg), median (IQR) | 79 (70, 88) | 80 (70, 90) | 0.010 | 67 (56, 81) | 65 (55, 77) | 0.006 | 67.50 (57, 81) | 66 (55, 78) | 0.002 |
| Creatinine, median (IQR) | 104 (80, 130) | 118 (88, 167) | < 0.001 | 1.50 (1.10, 2.10) | 1.70 (1.20, 2.30) | < 0.001 | 1.40 (1, 2) | 1.60 (1.20, 2.30) | < 0.001 |
| eGFR (mL/min/1.73 m2), median (IQR) | 57.93 (45.67, 82.43) | 50.86 (33.32, 69.66) | < 0.001 | 42.60 (28.54, 62.65) | 34.62 (23.74, 52.04) | < 0.001 | 44.21 (30.05, 64.10) | 36.49 (24.82, 53.64) | < 0.001 |
| ACR (mg/g), median (IQR) | 14.10 (8.33, 80.00) | 29.17 (10.74, 285.46) | < 0.001 | / | / | / | / | / | / |
| Diabetes, n % | 1665 (37.57%) | 591 (46.06%) | < 0.001 | 916 (45.10%) | 286 (44.48%) | 0.817 | 651 (44.32%) | 551 (45.73%) | 0.490 |
| Hypertension, n% | 2288 (51.62%) | 833 (64.93%) | < 0.001 | 1673 (82.37%) | 508 (79%) | 0.063 | 1221 (83.12%) | 960 (79.67%) | 0.025 |
| Cardiovascular disease, n% | 1247 (28.14%) | 417 (32.50%) | 0.003 | 1686 (83.01%) | 552 (85.85%) | 0.102 | 1206 (82.10%) | 1032 (85.64%) | 0.016 |
| Cancer, n% | 642 (14.49%) | 201 (15.67%) | 0.315 | 316 (15.56%) | 159 (24.73%) | < 0.001 | 155 (10.55%) | 320 (26.56%) | < 0.001 |
| HFRS, median (IQR) | 0.70 (0, 2) | 0.80 (0, 2.10) | 0.010 | 9.60 (6.30, 13.80) | 11 (8.50, 14.90) | < 0.001 | 9 (5.70, 13) | 11.20 (8.10, 14.90) | < 0.001 |
| CFI, median (IQR) | 5.02 (1.45, 13.65) | 6.54 (2.02, 15.71) | < 0.001 | 12.58 (5.55, 25.91) | 17.07 (7.40, 34.07) | < 0.001 | 11.51 (4.83, 22.83) | 16.79 (7.45, 33.29) | < 0.001 |
| eFI, median (IQR) | 14.29 (11.11, 17.78) | 17.04 (13.33, 21.49) | < 0.001 | 25 (20, 30) | 27.50 (22.50, 32.50) | < 0.001 | 23.08 (19.05, 28.21) | 27.50 (22.50, 32.50) | < 0.001 |
| FI‐LAB, median (IQR) | 18.18 (11.90, 26.19) | 23.26 (16.28, 31.82) | < 0.001 | 27.45 (21.57, 34) | 34 (27.45, 40) | < 0.001 | 26.53 (20.41, 32.65) | 31.37 (26, 38.30) | < 0.001 |
Abbreviations: ACR, albumin‐to‐creatinine ratio; CFI, Johns Hopkins Claims‐Based Frailty Index; CKD, chronic kidney disease; DBP, diastolic blood pressure; eFI, electronic frailty index; eGFR, estimated glomerular filtration rate; FI‐LAB, frailty index based on laboratory variables; HFRS, Hospital Frailty Risk Score; SBP, systolic blood pressure. Creatinine levels are expressed in μmol/L for the Hospital CKD cohort and in mg/dL for the MIMIC CKD cohort.
The MIMIC CKD cohort comprised 2674 patients (Table 1), with a median age of 73.03 (IQR, 64.09–81.78) years and 62.0% being male. The 28‐ and 365‐day mortality rates were 24.05% (n = 643) and 45.06% (n = 1205), respectively. Compared to survivors, non‐survivors were older, had higher serum creatinine, a greater prevalence of cancer and lower systolic and diastolic blood pressure, while gender distribution was similar (all p < 0.05 for significant differences). As hypothesized, all four frailty indices were significantly higher in non‐survivors at both 28 and 365 days (all p < 0.05).
3.2. Correlation Analysis of the EHR‐Derived Frailty Metrics
The Spearman correlations among the four frailty indices are presented in Figure 1. The eFI demonstrated weak to moderate positive correlations with the other three metrics in both the Hospital CKD and MIMIC CKD cohorts, respectively: the HFRS (ρ = 0.205 and 0.381), the CFI (ρ = 0.263 and 0.392) and FI‐lab (ρ = 0.451 and 0.358) (all p < 0.001). Among the other indices, a weak positive correlation was observed between HFRS and CFI in both cohorts (Hospital CKD: ρ = 0.316; MIMIC CKD: ρ = 0.243; both p < 0.001). In contrast, FI‐lab exhibited negligible or very weak correlations with HFRS (Hospital CKD: ρ = 0.003, p > 0.05; MIMIC CKD: ρ = 0.142, p < 0.001) and negligible or a very weak inverse correlation with CFI (Hospital CKD: ρ = −0.059; MIMIC CKD: ρ = −0.115; both p < 0.001). Importantly, there were no statistically significant differences in these correlation coefficients between the two cohorts (Table S8).
FIGURE 1.
Correlation matrix of EHR‐derived frailty metrics. CFI, Johns Hopkins Claims‐Based Frailty Index; CKD, chronic kidney disease; eFI, electronic frailty index; FI‐LAB, frailty index based on laboratory variables; HFRS, Hospital Frailty Risk Score.
3.3. Association Between Frailty Metrics and Clinical Outcomes
3.3.1. Cox Proportional Hazards Analyses
In the Hospital CKD cohort, after full adjustment, higher CFI (adjusted HR [aHR] 1.01, 95% CI 1.00–1.02), eFI (aHR 1.07, 1.06–1.08) and FI‐lab (aHR 1.04, 1.04–1.05) were associated with an increased risk of CKD progression (Table 2). Conversely, HFRS was not associated with the outcome in any model. In an age‐stratified subgroup analysis (Table S9), these associations persisted among patients aged ≥ 65 years. However, among those aged < 65 years, the association for CFI was attenuated to non‐significance (aHR 1.02, 0.94–1.10). To statistically validate this age‐dependent performance, we conducted formal interaction testing using age as a continuous variable across the entire cohort (Table S10). This analysis confirmed a significant age‐frailty interaction for CFI (p for interaction = 0.018), whereas the associations for eFI and FI‐lab remained significant and independent of age interactions.
TABLE 2.
Association of frailty with risk of CKD progression and all‐cause mortality.
| Frailty tool | Hospital CKD cohort | MIMIC CKD cohort | MIMIC CKD cohort | |||
|---|---|---|---|---|---|---|
| (kidney function progression) | (28‐day mortality) | (365‐day mortality) | ||||
| HR (95% CI) | p | HR (95% CI) | p | HR (95% CI) | p | |
| Continuous, per point increase | ||||||
| Univariable analysis | ||||||
| HFRS | 1.02 (0.99, 1.05) | 0.175 | 1.04 (1.02, 1.05) | < 0.001 | 1.05 (1.04, 1.05) | < 0.001 |
| CFI | 1.01 (1.01, 1.02) | < 0.001 | 1.01 (1.01, 1.02) | < 0.001 | 1.02 (1.01, 1.02) | < 0.001 |
| eFI | 1.08 (1.07, 1.09) | < 0.001 | 1.05 (1.03, 1.06) | < 0.001 | 1.05 (1.04, 1.06) | < 0.001 |
| FI‐LAB | 1.04 (1.04, 1.04) | < 0.001 | 1.06 (1.05, 1.06) | < 0.001 | 1.05 (1.04, 1.05) | < 0.001 |
| Multivariable analysis | ||||||
| HFRS | 1.00 (0.97, 1.03) | 0.970 | 1.03 (1.02, 1.04) | < 0.001 | 1.04 (1.03, 1.05) | < 0.001 |
| CFI | 1.01 (1.00, 1.02) | 0.003 | 1.01 (1.00, 1.02) | 0.059 | 1.01 (1.01, 1.02) | < 0.001 |
| eFI | 1.07 (1.06, 1.08) | < 0.001 | 1.04 (1.02, 1.05) | < 0.001 | 1.04 (1.03, 1.05) | < 0.001 |
| FI‐LAB | 1.04 (1.04, 1.05) | < 0.001 | 1.06 (1.05, 1.06) | < 0.001 | 1.05 (1.04, 1.05) | < 0.001 |
Note: The multivariable model for the Hospital CKD Cohort was adjusted for age, gender, smoking status, drinking status, systolic blood pressure, eGFR, ACR, diabetes, hypertension, cardiovascular disease, and cancer. The model for the MIMIC CKD Cohort was adjusted for age, gender, race, systolic blood pressure, eGFR, diabetes, hypertension, cardiovascular disease, and cancer. The bold values indicate p < 0.05.
Abbreviations: CFI, Johns Hopkins Claims‐Based Frailty Index; CKD, chronic kidney disease; eFI, Electronic Frailty Index; FI‐LAB, frailty index based on laboratory variables; HFRS, Hospital Frailty Risk Score.
We then assessed these associations in the MIMIC CKD cohort using 28‐ and 365‐day all‐cause mortality as endpoints (Table 2). For 365‐day mortality, all four frailty indices were significantly associated with increased mortality risk in fully adjusted models. For the 28‐day outcome, significant associations were also observed for eFI, FI‐lab and HFRS. In contrast, the association for CFI with 28‐day mortality did not reach statistical significance (p = 0.059). Age‐stratified analysis revealed that this association was non‐significant specifically in patients aged < 65 years (Table S9). Formal interaction testing further confirmed that the association of CFI with 365‐day mortality was significantly modified by age (p for interaction = 0.028). Additionally, FI‐Lab demonstrated strong age‐frailty interactions for both 28‐day (p for interaction = 0.003) and 365‐day mortality (p for interaction < 0.001) (Table S10). The other frailty indices maintained significant associations with both outcomes across all age groups.
3.3.2. Dose–Response Relationships
The dose–response relationships between frailty indices and outcomes were visualized using restricted cubic splines in fully adjusted models (Figure 2). In the Hospital CKD cohort, the risk of CKD progression showed a linear increase with higher eFI (p for overall < 0.001) and CFI (p for overall = 0.004). In contrast, FI‐lab demonstrated a significant, although non‐linear, positive association with the risk of CKD progression (p for non‐linearity < 0.001). No significant dose–response relationship was found for HFRS in this cohort.
FIGURE 2.
Association between electronic frailty metrics and CKD progression and all‐cause mortality using restricted cubic spline analysis. The model for the Hospital CKD Cohort (CKD progression) was adjusted for age, gender, smoking status, drinking status, systolic blood pressure, eGFR, ACR, diabetes, hypertension, cardiovascular disease, and cancer. The model for the MIMIC CKD cohort (all‐cause mortality) was adjusted for age, gender, race, systolic blood pressure, eGFR, diabetes, hypertension, cardiovascular disease, and cancer. (a–d) CKD progression associations with HFRS, CFI, eFI, and FI‐LAB, respectively. (e–h) 28‐day mortality associations with HFRS, CFI, eFI, and FI‐LAB, respectively. (i–l) 365‐day mortality associations with HFRS, CFI, eFI, and FI‐LAB, respectively. ACR, albumin‐to‐creatinine ratio; CFI, Johns Hopkins Claims‐Based Frailty Index; CI, confidence interval; CKD, chronic kidney disease; eFI, electronic frailty index; eGFR, estimated glomerular filtration rate; FI‐LAB, frailty index based on laboratory variables; HFRS, Hospital Frailty Risk Score; HR, hazard ratio.
In the MIMIC CKD cohort, we observed similar positive associations with mortality but with more pronounced non‐linearity. For 365‐day mortality, CFI was associated with a linear increase in risk. In contrast, eFI, FI‐lab and HFRS all demonstrated significant non‐linear associations. Specifically, while eFI and FI‐lab maintained a monotonically increasing trend (p for non‐linearity < 0.05 for both), the relationship for HFRS was characterized by a steep initial increase in risk, which plateaued at higher frailty levels where the association was no longer statistically significant (p for non‐linearity < 0.001). These patterns were largely consistent for 28‐day mortality, with the exception that the association for CFI was not statistically significant.
3.4. Incremental Value of Frailty Metrics for Risk Stratification Predictive
3.4.1. Performance for CKD Progression in the Hospital CKD Cohort
We first evaluated the individual predictive performance of each EHR‐derived frailty metric for CKD progression. In univariate analysis, FI‐lab demonstrated the highest discrimination (AUC = 0.673, 95% CI 0.651–0.695), followed by eFI (AUC = 0.648, 0.625–0.670). In contrast, both HFRS (AUC = 0.524, 0.502–0.547) and CFI (AUC = 0.526, 0.502–0.549) showed limited discriminative capacity (Table S11). Next, we assessed the incremental value of adding these frailty metrics to two established CKD risk models: the KDIGO heatmap (AUC = 0.643, 0.621–0.665) and the Kidney Failure Risk Equation (KFRE) (AUC = 0.653, 0.630–0.675). Incorporating either FI‐lab or eFI into these tools significantly improved their 2‐year predictive accuracy (p for ΔAUC < 0.05 for all comparisons). This finding was supported by positive values for both the IDI and NRI for both the KDIGO heatmap and KFRE models (Table 3). Conversely, adding HFRS failed to improve predictive performance, while CFI provided only a marginal benefit when added to the KDIGO heatmap and no benefit to the KFRE model. Furthermore, time‐dependent AUC analysis over a 5‐year follow‐up confirmed that the addition of FI‐lab consistently provided the greatest enhancement in predictive accuracy for both baseline models, followed by eFI (Figure S2).
TABLE 3.
Incremental value of electronic frailty metrics in predicting 2‐year CKD progression risk (Hospital CKD Cohort).
| Models | AUC (95% CI) | △AUC | p for △AUC | IDI (p) | NRI (p) |
|---|---|---|---|---|---|
| Base model: KDIGO heat map | 0.643 (0.621–0.665) | ||||
| +HFRS vs. KDIGO heat map | 0.648 (0.626–0.670) | 0.005 | 0.075 | < 0.001 (0.696) | 0.004 (0.896) |
| +CFI vs. KDIGO heat map | 0.651 (0.629–0.673) | 0.008 | 0.009 | < 0.001 (0.666) | 0.018 (0.638) |
| +eFI vs. KDIGO heat map | 0.697 (0.676–0.718) | 0.054 | < 0.001 | 0.032 (< 0.001) | 0.376 (< 0.001) |
| +FI‐LAB vs. KDIGO heat map | 0.710 (0.689–0.730) | 0.067 | < 0.001 | 0.034 (< 0.001) | 0.424 (< 0.001) |
| Base model: KFRE | 0.653 (0.630–0.675) | ||||
| +HFRS vs. KFRE | 0.640 (0.618–0.663) | −0.012 | 0.172 | < 0.001 (0.128) | 0.046 (0.200) |
| +CFI vs. KFRE | 0.637 (0.614–0.659) | −0.016 | 0.136 | 0.004 (0.004) | 0.080 (0.038) |
| +eFI vs. KFRE | 0.694 (0.672–0.716) | 0.041 | 0.002 | 0.040 (< 0.001) | 0.440 (< 0.001) |
| +FI‐LAB vs. KFRE | 0.713 (0.692–0.734) | 0.060 | < 0.001 | 0.048 (< 0.001) | 0.498 (< 0.001) |
Note: The bold values indicate p < 0.05.
Abbreviations: AUC, area under the receiver operating characteristic curve; CFI, Johns Hopkins Claims‐Based Frailty Index; CI, confidence interval; CKD, chronic kidney disease; eFI, electronic frailty index; FI‐LAB, frailty index based on laboratory variables; HFRS, Hospital Frailty Risk Score; IDI, integrated discrimination improvement; KDIGO, Kidney Disease: Improving Global Outcomes; KFRE, 4‐variable Kidney Failure Risk Equation; NRI, net reclassification improvement.
3.5. Predictive Performance for Mortality in the MIMIC CKD Cohort
We then extended our analysis to the MIMIC CKD cohort to evaluate the predictive value of the frailty metrics for 28‐ and 365‐day all‐cause mortality. Consistent with the findings for CKD progression, FI‐lab was the strongest individual predictor for both 28‐day (AUC = 0.684, 95% CI 0.661–0.707) and 365‐day mortality (AUC = 0.661, 0.641–0.682), followed by eFI (Table S10). The predictive performance of three established disease severity scores was also evaluated. For 28‐day mortality, the AUCs were 0.702 (0.679–0.725) for APS III, 0.696 (0.673–0.719) for SAPS II and 0.634 (0.610–0.659) for SOFA. For 365‐day mortality, the corresponding AUCs were 0.681 (0.661–0.701), 0.680 (0.660–0.700) and 0.613 (0.591–0.634), respectively. As detailed in Table 4, the addition of frailty indices to these disease scores significantly improved the predictive accuracy for both short‐ and long‐term mortality. Time‐dependent AUC analyses further confirmed these findings, showing that the addition of FI‐lab provided the greatest incremental benefit to the baseline models (Figure S3).
TABLE 4.
Incremental value of electronic frailty metrics in predicting 28‐day mortality and 365‐day mortality (MIMIC CKD cohort).
| Models | 28‐day mortality | 365‐day mortality | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| AUC (95% CI) | △AUC | p for △AUC | IDI (p) | NRI (p) | AUC (95% CI) | △AUC | p for △AUC | IDI (p) | NRI (p) | |
| Base model: SOFA | 0.634 (0.610, 0.659) | 0.613 (0.591, 0.634) | ||||||||
| +HFRS vs. SOFA | 0.651 (0.627, 0.674) | 0.016 | 0.003 | 0.008 (< 0.001) | 0.186 (< 0.001) | 0.644 (0.624, 0.665) | 0.032 | < 0.001 | 0.024 (< 0.001) | 0.296 (< 0.001) |
| +CFI vs. SOFA | 0.667 (0.644, 0.691) | 0.033 | < 0.001 | 0.017 (< 0.001) | 0.252 (< 0.001) | 0.660 (0.639, 0.680) | 0.047 | < 0.001 | 0.034 (< 0.001) | 0.319 (< 0.001) |
| +eFI vs. SOFA | 0.660 (0.636,0.683) | 0.025 | < 0.001 | 0.013 (< 0.001) | 0.260 (< 0.001) | 0.665 (0.644, 0.685) | 0.052 | < 0.001 | 0.041 (< 0.001) | 0.360 (< 0.001) |
| +FI‐LAB vs. SOFA | 0.688 (0.665, 0.712) | 0.054 | < 0.001 | 0.036 (< 0.001) | 0.376 (< 0.001) | 0.664 (0.643, 0.684) | 0.051 | < 0.001 | 0.045 (< 0.001) | 0.363 (< 0.001) |
| Base model: ASP3 | 0.702 (0.679, 0.725) | 0.681 (0.661, 0.701) | ||||||||
| +HFRS vs. ASP3 | 0.711 (0.689, 0.734) | 0.009 | 0.003 | 0.002 (0.050) | 0.135 (0.004) | 0.697 (0.677, 0.717) | 0.016 | < 0.001 | 0.015 (< 0.001) | 0.260 (< 0.001) |
| +CFI vs. ASP3 | 0.722 (0.700, 0.744) | 0.020 | < 0.001 | 0.014 (< 0.001) | 0.254 (< 0.001) | 0.709 (0.689, 0.728) | 0.028 | < 0.001 | 0.028 (< 0.001) | 0.298 (< 0.001) |
| +eFI vs. ASP3 | 0.713 (0.690, 0.736) | 0.011 | 0.021 | 0.008 (< 0.001) | 0.230 (< 0.001) | 0.708 (0.689, 0.728) | 0.027 | < 0.001 | 0.031 (< 0.001) | 0.340 (< 0.001) |
| +FI‐LAB vs. ASP3 | 0.724 (0.702, 0.746) | 0.022 | < 0.001 | 0.015 (< 0.001) | 0.218 (< 0.001) | 0.699 (0.679,0.719) | 0.018 | 0.001 | 0.025 (< 0.001) | 0.289 (< 0.001) |
| Base model: SAPS2 | 0.696 (0.673, 0.719) | 0.680 (0.660, 0.700) | ||||||||
| +HFRS vs. SAPS2 | 0.703 (0.681, 0.726) | 0.008 | 0.002 | 0.126 (0.006) | 0.206 (< 0.001) | 0.696 (0.676, 0.715) | 0.016 | < 0.001 | 0.013 (< 0.001) | 0.238 (< 0.001) |
| +CFI vs. SAPS2 | 0.703 (0.680, 0.726) | 0.008 | 0.018 | 0.004 (0.016) | 0.206 (< 0.001) | 0.690 (0.670, 0.710) | 0.010 | 0.006 | 0.013 (< 0.001) | 0.256 (< 0.001) |
| +eFI vs. SAPS2 | 0.705 (0.683,0.728) | 0.010 | 0.011 | 0.004 (0.020) | 0.195 (< 0.001) | 0.704 (0.684, 0.723) | 0.024 | < 0.001 | 0.022 (< 0.001) | 0.289 (< 0.001) |
| +FI‐LAB vs. SAPS2 | 0.729 (0.708, 0.751) | 0.034 | < 0.001 | 0.027 (< 0.001) | 0.324 (< 0.001) | 0.707 (0.688, 0.726) | 0.027 | < 0.001 | 0.028 (< 0.001) | 0.305 (< 0.001) |
Note: The bold values indicate p < 0.05.
Abbreviations: APS3, Acute Physiology Score III; AUC, area under the receiver operating characteristic curve; CFI, Johns Hopkins Claims‐Based Frailty Index; CI, confidence interval; eFI, electronic frailty index; FI‐LAB, frailty index based on laboratory variables; HFRS, Hospital Frailty Risk Score; IDI, integrated discrimination improvement; NRI, net reclassification improvement; SAPS2, Simplified Acute Physiology Score II; SOFA, Sequential Organ Failure Assessment.
4. Discussion
To our knowledge, this is the first systematic comparison of four electronic frailty metrics (HFRS, CFI, eFI and FI‐Lab) in hospitalized patients with CKD. We found low to moderate correlations among these indices, suggesting they capture partially distinct aspects of the frailty construct, though shared components such as comorbidity burden and laboratory abnormalities may still contribute overlapping variance. For CKD progression, while CFI, eFI and FI‐Lab were significant predictors (unlike HFRS), only FI‐Lab and eFI incrementally improved established risk models. For mortality, the findings were more nuanced: all four indices predicted long‐term mortality, but the association between CFI and short‐term mortality was not robust. FI‐Lab again provided the greatest incremental predictive value over standard severity scores. Notably, age‐stratified analyses revealed a key limitation of CFI: its associations with both short‐term mortality and CKD progression were attenuated in patients < 65 years, suggesting its utility is concentrated in older adults. These results highlight the marked prognostic heterogeneity among these frailty metrics in this high‐risk population.
Although EHR‐based frailty tools are widely linked to adverse outcomes, systematic comparisons in hospitalized CKD patients, particularly for predicting CKD progression, are scarce. The HFRS, developed in emergency‐admitted patients aged > 75 years, has shown inconsistent performance in other populations [7, 29, 30]. Our study highlights its divergent predictive ability: HFRS was associated with short‐ and long‐term mortality but not with CKD progression. This discrepancy likely reflects a demographic mismatch. Its success in predicting mortality was observed in the older, critically ill MIMIC‐CKD cohort (median age 73.03), which mirrors its original development population. Conversely, its failure for renal outcomes occurred in our younger, more heterogeneous Hospital‐CKD cohort. This limited utility of HFRS for predicting CKD progression may therefore stem from population differences, potentially compounded by varied diagnostic coding practices across specialties. Formal interaction testing revealed age‐dependent prognostic heterogeneity across the evaluated indices. The CFI was originally developed and validated in older adults (≥ 65 years) as a proxy for the Fried Frailty Phenotype, with algorithmic age‐weighting embedded in its construction [8]. This design inherently modifies its predictive behaviour across clinical contexts. In non‐geriatric populations, CFI's discriminative ability for CKD progression was substantially attenuated. In the acute critical care setting, the age interaction for 28‐day mortality did not reach statistical significance, likely because short‐term ICU mortality is predominantly driven by acute illness severity rather than baseline frailty. As the observational window extended to 365 days, however, the age‐modifying effect re‐emerged, with the magnitude of relative risk differing significantly by age group. The cumulative deficit approaches of eFI and FI‐Lab demonstrated greater age‐independent predictive consistency for renal outcomes. Notably, while FI‐Lab remained consistent across age groups for renal progression prediction, it exhibited a pronounced age‐frailty interaction for mortality in the critically ill MIMIC cohort. This divergence highlights a clinically meaningful distinction: baseline laboratory deficits predict longitudinal renal decline independently of age, whereas the mortality risk conferred by severe laboratory derangements in critical illness is substantially amplified in older patients. Both eFI and FI‐Lab exemplify the cumulative deficits model, where individual deficits are considered interchangeable. Consistent with this model's principle that indices with > 30 items perform robustly regardless of composition, we observed a significant correlation between eFI and FI‐Lab [31]. This association was strong in the Hospital CKD cohort (Spearman's ρ = 0.451) and moderate in the MIMIC CKD cohort (ρ = 0.358). Both metrics robustly predicted the full range of adverse outcomes—CKD progression and both short‐ and long‐term mortality—underscoring the utility of the cumulative deficits approach in this population.
In assessing incremental predictive value, the four indices showed distinct patterns. For CKD progression, we evaluated them against the KDIGO heatmap and KFRE base models; variables required for other established tools (e.g., Klinrisk) were unavailable, precluding their inclusion as comparators. Focusing on incremental improvement over baseline performance, our analysis revealed that integrating FI‐Lab or eFI significantly enhanced model discrimination and reclassification, whereas HFRS offered no predictive benefit. CFI's contribution was marginal when added to the KDIGO heatmap and non‐existent when combined with the KFRE model. In contrast to the heterogeneous findings for renal outcomes, the indices demonstrated a more consistent effect for mortality prediction. When added to established severity‐of‐illness scores (SOFA, APS III, SAPS II), all four frailty indices uniformly enhanced the prediction of both short‐ and long‐term mortality. Among them, FI‐Lab provided the most substantial performance gain. Overall, FI‐Lab and eFI emerged as the most effective tools for risk stratifying hospitalized CKD patients for both CKD progression and all‐cause mortality. Furthermore, the assessment of FI‐Lab relies exclusively on objective laboratory parameters, eliminating assessor variability and facilitating its integration into routine clinical risk stratification workflows. Emerging evidence highlights the role of laboratory biomarkers in the biological mechanisms underlying frailty, and the metrics comprising FI‐Lab may reflect a patient's current biological reserve, potentially explaining its superior predictive performance [32, 33]. However, several barriers to routine implementation warrant consideration. Comprehensive laboratory panels may not be consistently available in resource‐limited or non‐tertiary care settings. Regarding missing data, although the fractional nature of the FI allows for dynamic calculation by adjusting the denominator (i.e., the total number of deficits present divided by the total number of variables measured) even when a few tests are omitted, extreme missingness could compromise the index's stability. Therefore, standardized approaches to handling missing laboratory values, as well as the potential development of an abbreviated FI‐Lab panel for primary care, are necessary prerequisites for its broad prospective validation and adoption across diverse clinical environments.
Our study has several limitations. First, our use of two heterogeneous cohorts (institutional and MIMIC‐CKD) precluded cross‐validation for the same outcome. However, we established consistent inter‐correlations among the four eFIs across both populations. Crucially, the incremental predictive utility of FI‐Lab was a robust finding across both cohorts and for different adverse outcomes, bolstering the reliability of our primary conclusion. Second, our definition of CKD progression lacked 30‐ or 90‐day confirmation, which may have led to an overestimation. This was necessitated by high patient dropout rates, which made confirming outcome persistence impractical. Third, we did not evaluate the associations between the electronic frailty metrics and other frailty instruments based on in‐person clinician assessments. Future research should explore these correlations and compare the predictive validity of electronic and clinician‐administered frailty assessments. Fourth, since the EHRs were not primarily collected for frailty evaluation, ICD‐10 codes in our study population may have been partially missing or inaccurate. In the MIMIC cohort specifically, CKD was ascertained solely through administrative coding without laboratory confirmation or chronicity documentation, raising the possibility of misclassification, including inadvertent inclusion of acute kidney injury. This limitation is shared by prior large‐scale EHR‐based studies and, to the extent misclassification occurred, would likely introduce non‐differential bias that obscures true frailty‐renal outcome associations. Future studies employing laboratory‐confirmed CKD definitions, incorporating repeated eGFR measurements alongside markers of kidney damage such as proteinuria, would improve diagnostic precision. Additionally, coding practices vary across healthcare systems, which may limit generalizability. Finally, our calculations of eFIs deviated slightly from the original published methods. These methodological modifications may influence the generalizability of our findings.
Our findings demonstrate the distinct predictive performance of four EHR‐based frailty indices in hospitalized patients with CKD. While CFI, eFI and FI‐Lab predicted both CKD progression and all‐cause mortality, HFRS was only associated with mortality. The utility of CFI was age‐dependent, confining its applicability primarily to elderly patients. For risk stratification, adding eFI or FI‐Lab to established models substantially improved the prediction of CKD progression. All four indices enhanced mortality prediction over baseline severity scores, with FI‐Lab providing the greatest incremental value. These results underscore the importance of selecting the appropriate frailty tool to optimize risk stratification in this high‐risk population.
Funding
The authors have nothing to report.
Ethics Statement
This study adhered to the principles of the Declaration of Helsinki. The protocol for the institutional cohort was approved by the Institutional Ethics Committee of The Third Affiliated Hospital of Sun Yat‐sen University (Approval No. 02‐427‐01). The analysis of the MIMIC cohort utilized a public, de‐identified database whose collection was approved by the Institutional Review Boards of MIT and BIDMC. All data were anonymized prior to analysis.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: The CKD patients were diagnosed by International Classification of Diseases‐10 version (ICD‐10).
Table S2: List of ICD‐10 codes, number of points for each to create the Hospital Frailty Risk Score.
Table S3: List of appropriate ICD‐10 codes to create Johns Hopkins Claims Based Frailty Index.
Table S4: Johns Hopkins Claims‐based Frailty Index algorithm.
Table S5: Calculation of Electronic Frailty Index (eFI).
Table S6: Laboratory variables for frailty index(FI‐LAB).
Table S7: List of ICD‐10 codes to identify covariates.
Table S8: Comparison of spearman correlation coefficients among four electronic frailty indices between the institutional and MIMIC cohorts.
Table S9: Association of frailty with risk of CKD progression and all‐cause mortality stratified by age.
Table S10: Formal age‐frailty interaction testing for adverse outcomes.
Table S11: Predictive performance of individual electronic frailty metrics for CKD progression and all‐cause mortality.
Figure S1: Flowchart of participants selection.
Figure S2: Time‐dependent discriminatory performance of standard risk models for CKD progression with and without the addition of different frailty indices (hospital CKD cohort). (A) The predictive performance of the KDIGO risk classification alone (black line) compared with models combining KDIGO with four different frailty indices. (B) The predictive performance of the Kidney Failure Risk Equation (KFRE) alone (black line) compared with models combining KFRE with the same four frailty indices.
Figure S3: Time‐dependent discriminatory performance of common disease severity scores for predicting all‐cause mortality, with and without the addition of different frailty indices.
Acknowledgements
We gratefully acknowledge all those who contributed to the completion of this study.
Contributor Information
Leile Tang, Email: tleil@mail2.sysu.edu.cn.
Xun Liu, Email: naturestyle@163.com.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.
References
- 1. Fan J., Yu C., Guo Y., et al., “Frailty Index and All‐Cause and Cause‐Specific Mortality in Chinese Adults: A Prospective Cohort Study,” Lancet Public Health 5 (2020): e650–e660. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Si H., Jin Y., Qiao X., Tian X., Liu X., and Wang C., “Predictive Performance of 7 Frailty Instruments for Short‐Term Disability, Falls and Hospitalization Among Chinese Community‐Dwelling Older Adults: A Prospective Cohort Study,” International Journal of Nursing Studies 117 (2021): 103875. [DOI] [PubMed] [Google Scholar]
- 3. Tonelli M., Wiebe N., Culleton B., et al., “Chronic Kidney Disease and Mortality Risk: A Systematic Review,” Journal of the American Society of Nephrology 17 (2006): 2034–2047. [DOI] [PubMed] [Google Scholar]
- 4. Chowdhury R., Peel N. M., Krosch M., and Hubbard R. E., “Frailty and Chronic Kidney Disease: A Systematic Review,” Archives of Gerontology and Geriatrics 68 (2017): 135–142. [DOI] [PubMed] [Google Scholar]
- 5. Wang M., Sun X., Zhang W., et al., “Frailty and the Risk of Kidney Function Decline in the Elderly Population: The Rugao Longevity and Ageing Study,” Nephrology, Dialysis, Transplantation 36 (2021): 2274–2281. [DOI] [PubMed] [Google Scholar]
- 6. Buta B. J., Walston J. D., Godino J. G., et al., “Frailty Assessment Instruments: Systematic Characterization of the Uses and Contexts of Highly‐Cited Instruments,” Ageing Research Reviews 26 (2016): 53–61. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Gilbert T., Neuburger J., Kraindler J., et al., “Development and Validation of a Hospital Frailty Risk Score Focusing on Older People in Acute Care Settings Using Electronic Hospital Records: An Observational Study,” Lancet 391 (2018): 1775–1782. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Segal J. B., Chang H.‐Y., Du Y., Walston J. D., Carlson M. C., and Varadhan R., “Development of a Claims‐Based Frailty Indicator Anchored to a Well‐Established Frailty Phenotype,” Medical Care 55 (2017): 716–722. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Callahan K. E., Clark C. J., Edwards A. F., et al., “Automated Frailty Screening At‐Scale for Pre‐Operative Risk Stratification Using the Electronic Frailty Index,” Journal of the American Geriatrics Society 69 (2021): 1357–1362. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Wang Y., Zhang R., Shen Y., Su L., Dong B., and Hao Q., “Prediction of Chemotherapy Adverse Reactions and Mortality in Older Patients With Primary Lung Cancer Through Frailty Index Based on Routine Laboratory Data,” Clinical Interventions in Aging 14 (2019): 1187–1197. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Johnson A. E. W., Bulgarelli L., Shen L., et al., “MIMIC‐IV, a Freely Accessible Electronic Health Record Dataset,” Scientific Data 10 (2023): 1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Inker L. A., Schmid C. H., Tighiouart H., et al., “Estimating Glomerular Filtration Rate From Serum Creatinine and Cystatin C,” New England Journal of Medicine 367 (2012): 20–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Stevens P. E. and Levin A., “Evaluation and Management of Chronic Kidney Disease: Synopsis of the Kidney Disease: Improving Global Outcomes 2012 Clinical Practice Guideline,” Annals of Internal Medicine 158 (2013): 825–830. [DOI] [PubMed] [Google Scholar]
- 14. Dong Z.‐y., Wang Y.‐d., Qiu Q., et al., “Dysmorphic Erythrocytes Are Superior to Hematuria for Indicating Non‐Diabetic Renal Disease in Type 2 Diabetics,” Journal of Diabetes Investigation 7 (2016): 115–120. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Liu Y., Pan Y., Tian Z., et al., “Association of Mitochondrial DNA Copy Number With Chronic Kidney Disease in Older Adults,” BMC Geriatrics 23 (2023): 514. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Leiner J., Pellissier V., König S., et al., “Patient Characteristics and Outcomes of Hospitalized Chronic Kidney Disease Patients With and Without Type 2 Diabetes Mellitus: Observations From the German Claims Data‐Based Cohort of the CaReMe‐CKD Multinational Study,” Clinical Epidemiology 16 (2024): 487–500. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Kim K., Cho W.‐H., Hwang S. D., Lee S. W., and Song J. H., “Association Between Constipation and Incident Chronic Kidney Disease in the UK Biobank Study,” Scientific Reports 14 (2024): 32106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Bothe T., Fietz A.‐K., Schaeffner E., et al., “Diagnostic Validity of Chronic Kidney Disease in Health Claims Data Over Time: Results From a Cohort of Community‐Dwelling Older Adults in Germany,” Clinical Epidemiology 16 (2024): 143–154. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19. Kundi H., Coskun N., and Yesiltepe M., “Association of Entirely Claims‐Based Frailty Indices With Long‐Term Outcomes in Patients With Acute Myocardial Infarction, Heart Failure, or Pneumonia: A Nationwide Cohort Study in Turkey,” Lancet Regional Health Europe 10 (2021): 100183. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Le S. T., Liu V. X., Kipnis P., Zhang J., Peng P. D., and Cespedes Feliciano E. M., “Comparison of Electronic Frailty Metrics for Prediction of Adverse Outcomes of Abdominal Surgery,” JAMA Surgery 157 (2022): e220172. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Sumida K., Nadkarni G. N., Grams M. E., et al., “Conversion of Urine Protein‐Creatinine Ratio or Urine Dipstick Protein to Urine Albumin‐Creatinine Ratio for Use in Chronic Kidney Disease Screening and Prognosis: An Individual Participant‐Based Meta‐Analysis,” Annals of Internal Medicine 173 (2020): 426–435. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Zhang Z., Yang H., and Luo M., “Association Between Charlson Comorbidity Index and Community‐Acquired Pressure Injury in Older Acute Inpatients in a Chinese Tertiary Hospital,” Clinical Interventions in Aging 16 (2021): 1987–1995. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Thygesen S. K., Christiansen C. F., Christensen S., Lash T. L., and Sørensen H. T., “The Predictive Value of ICD‐10 Diagnostic Coding Used to Assess Charlson Comorbidity Index Conditions in the Population‐Based Danish National Registry of Patients,” BMC Medical Research Methodology 11 (2011): 83. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. König S., Pellissier V., Hohenstein S., et al., “Characteristics and Outcome of Hospitalized Patients With Heart Failure Stratified for Chronic Kidney Disease,” ESC Heart Failure 11 (2024): 3341–3349. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Tangri N., Ferguson T., Leon S. J., et al., “Validation of the Klinrisk Chronic Kidney Disease Progression Model in the FIDELITY Population,” Clinical Kidney Journal 17 (2024): sfae052. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Tangri N., Grams M. E., Levey A. S., et al., “Multinational Assessment of Accuracy of Equations for Predicting Risk of Kidney Failure: A Meta‐Analysis,” JAMA 315 (2016): 164–174. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Blanche P., Dartigues J. F., and Jacqmin‐Gadda H., “Estimating and Comparing Time‐Dependent Areas Under Receiver Operating Characteristic Curves for Censored Event Times With Competing Risks,” Statistics in Medicine 32 (2013): 5381–5397. [DOI] [PubMed] [Google Scholar]
- 28. Kerr K. F., Wang Z., Janes H., McClelland R. L., Psaty B. M., and Pepe M. S., “Net Reclassification Indices for Evaluating Risk Prediction Instruments: A Critical Review,” Epidemiology (Cambridge, Mass.) 25 (2014): 114–121. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Hägg S., Jylhävä J., Wang Y., et al., “Age, Frailty, and Comorbidity as Prognostic Factors for Short‐Term Outcomes in Patients With Coronavirus Disease 2019 in Geriatric Care,” Journal of the American Medical Directors Association 21 (2020): 1555–1559.e1552. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Harvey L. A., Toson B., Norris C., Harris I. A., Gandy R. C., and Close J., “Does Identifying Frailty From ICD‐10 Coded Data on Hospital Admission Improve Prediction of Adverse Outcomes in Older Surgical Patients? A Population‐Based Study,” Age and Ageing 50 (2021): 802–808. [DOI] [PubMed] [Google Scholar]
- 31. Searle S. D., Mitnitski A., Gahbauer E. A., Gill T. M., and Rockwood K., “A Standard Procedure for Creating a Frailty Index,” BMC Geriatrics 8 (2008): 24. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Kameda M., Teruya T., Yanagida M., and Kondoh H., “Frailty Markers Comprise Blood Metabolites Involved in Antioxidation, Cognition, and Mobility,” Proceedings of the National Academy of Sciences of the United States of America 117 (2020): 9483–9489. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33. Fujisawa C., Umegaki H., Sugimoto T., et al., “Older Adults With a Higher Frailty Index Tend to Have Electrolyte Imbalances,” Experimental Gerontology 163 (2022): 111778. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: The CKD patients were diagnosed by International Classification of Diseases‐10 version (ICD‐10).
Table S2: List of ICD‐10 codes, number of points for each to create the Hospital Frailty Risk Score.
Table S3: List of appropriate ICD‐10 codes to create Johns Hopkins Claims Based Frailty Index.
Table S4: Johns Hopkins Claims‐based Frailty Index algorithm.
Table S5: Calculation of Electronic Frailty Index (eFI).
Table S6: Laboratory variables for frailty index(FI‐LAB).
Table S7: List of ICD‐10 codes to identify covariates.
Table S8: Comparison of spearman correlation coefficients among four electronic frailty indices between the institutional and MIMIC cohorts.
Table S9: Association of frailty with risk of CKD progression and all‐cause mortality stratified by age.
Table S10: Formal age‐frailty interaction testing for adverse outcomes.
Table S11: Predictive performance of individual electronic frailty metrics for CKD progression and all‐cause mortality.
Figure S1: Flowchart of participants selection.
Figure S2: Time‐dependent discriminatory performance of standard risk models for CKD progression with and without the addition of different frailty indices (hospital CKD cohort). (A) The predictive performance of the KDIGO risk classification alone (black line) compared with models combining KDIGO with four different frailty indices. (B) The predictive performance of the Kidney Failure Risk Equation (KFRE) alone (black line) compared with models combining KFRE with the same four frailty indices.
Figure S3: Time‐dependent discriminatory performance of common disease severity scores for predicting all‐cause mortality, with and without the addition of different frailty indices.
Data Availability Statement
The data that support the findings of this study are available from the corresponding author upon reasonable request.