Cystatin C and the difference between cystatin C and serum creatinine: Improved metrics to predict waitlist mortality among patients with decompensated cirrhosis

Abstract

Among patients with decompensated cirrhosis, serum creatinine (sCr) is biased by sex, frailty, and hepatic synthetic function, while Cystatin C (cysC) is not. We found that sCr would better associate with waitlist mortality and that the difference between cysC and sCr (cysCsCr_diff) would quantify this bias and be independently associated with outcomes. We measured cysC levels at ambulatory liver transplant visits among 525 consecutive patients seen at our center. We defined the cysCsCr_diff as the difference between cysC minus sCr. We compared demographics and clinical characteristics in patients with low, intermediate, and high cysCsCr_diff, divided by tertile. We used Cox regression to compare the association between sCr and cysC and waitlist mortality and demonstrate the independent association between cysCsCr_diff and waitlist mortality. In Cox regression, cysC was significantly more associated with waitlist mortality than sCr (p < 0.001). We found that as compared to those with a low cysCsCr_diff, those with an intermediate or high cysCsCr_diff were more likely to be female, have ascites, have higher frailty, and have higher MELD 3.0 scores (p < 0.05 for all). Compared to those with a low cysCsCr_diff, we found that those in the intermediate and high groups were more likely to die during follow-up (low: 6% vs. intermediate: 8% vs. high: 11%, p = 0.007). We found that after adjusting for the components of the MELD 3.0 score, each 1-point increase in the cysCsCr_diff was associated with 1.72× (1.27–2.32) the hazard of waitlist mortality. Our study demonstrates that not only is cysC more associated with waitlist mortality than sCr, but that cysCsCr_diff represents a novel independent metric associated with waitlist mortality.

Graphical Abstract

INTRODUCTION

Since the introduction of the Model for End-Stage Liver Disease (MELD) score, measuring kidney function has been central to stratifying risk for mortality among patients with decompensated cirrhosis.^[1–3] From its inception, this calculation has included serum creatinine (sCr). However, sCr has consistently been shown to be an inaccurate marker of kidney dysfunction in patients with cirrhosis.^[4–15] There are factors intrinsic to the patient with cirrhosis—frailty, sarcopenia, sex, and decreased hepatic synthesis of creatinine—that lower sCr levels in the population.^{[5,8–10,15–18]} Further, given that these factors may impact patients with cirrhosis differently, it is a metric that cannot easily be standardized, as done in recent MELD modifications like MELD 3.0.^[1]

Cystatin C (cysC), a cysteine protease inhibitor produced by nearly all human cells, with no intrinsic biases by hepatic synthetic function, frailty, sex, or muscle mass, may be a better metric to capture kidney function in the patient with cirrhosis.^[5,19–22] Although preliminary studies among patients with cirrhosis have demonstrated that cysC is less biased in estimating glomerular filtration rate (GFR) and is associated with outcomes, no study has shown the significance of cysC among candidates for liver transplant in the United States or the clinical utility of the difference between cysC and sCr (cysCsCr_diff) to quantitate these biases (eg, frailty, sex, and hepatic synthetic function) and capture risk for mortality beyond the components of the MELD 3.0 score.^[5,8,15,23]

Herein, we tested the hypothesis that cysC is more linked to waitlist mortality than sCr and evaluated the clinical utility of the cysCsCr_diff, a metric recently described in the general nephrology literature, among candidates for liver transplant.^[4,24–26]

METHODS

Population

This study represents an expansion of the Functional Assessment of the Study of Liver Disease (FrAILT) cohort. The FrAILT study is an ongoing prospective cohort of ambulatory patients listed for liver transplantation at the University of California, San Francisco (UCSF).^[27,28] Since 2017, the FrAILT study has included a biorepository of biospecimens (eg, serum and urine) measured at the participant’s ambulatory liver transplant evaluation. This study utilizes the biospecimens for 525 consecutive participants included in this biorepository.

Demographics and clinical characteristics

We determined the following sociodemographics and clinical characteristics at the time of biospecimen collection: sex, age, race, ethnicity, etiology of cirrhosis, kidney replacement therapy status, ascites status, and HE status. According to recent guidelines for this study, we graded ascites as either none, mild/moderate, or severe.^[29] We defined HE as taking or being prescribed lactulose or rifaximin for HE at the time of the liver transplant evaluation.

Laboratory values

As part of the routine standard of care, each patient received an sCr, total bilirubin, international normalized ratio (INR), serum sodium, and serum albumin measurement on the same date of biospecimen collection. These labs were used for all analyses in this manuscript. We calculated MELD 3.0 scores according to standardized methods.^[1] We calculated Child-Pugh scores as assessed at the time of enrollment in the FrAILT and collection of biospecimens. We estimated the glomerular filtration rate using the chronic kidney dieseae epidemiology collagboration cysC-based equation.

Frailty measures

As part of enrollment in this study, patients had their Liver Frailty Index (LFI) determined on the day of their liver transplant evaluation and biospecimen collection. The LFI represents a composite of a participant’s ability to complete grip strength, chair stands, and tandem walks.^[27] We categorized patients as “frail” if their LFI was ≥4.2, a measure standardized among patients with cirrhosis.^[30]

Biospecimens

All biospecimens were collected, processed, and stored at −80°C on the collection day. For this analysis, we used serum samples for the 525 participants with biospecimens available for analysis. Maine Medical Laboratories completed all cysC levels using an enzyme-linked immunosorbent assay (R&D Systems, DSCT0). We converted cysC values to units used in clinical practice, specifically mg/L.

Primary exposure

Our primary exposure was the difference in cysC (mg/L) to sCr (mg/dL). This was calculated as follows:

$${\text{cysCscr}}_{\text{diff}}=\text{cysC}-\text{sCr}.$$

Factors associated with cysCsCr_diff

We categorized our cohort into cysCsCr_diff tertiles. “Low” was defined as a cysCsCr_diff < 0.23; “Intermediate” was defined as a cysCsCr_diff between 0.23 and 0.50; and “High” was defined as a cysCsCr_diff > 0.50. We compare the factors associated with low, intermediate, and high cysCsCr_diff using either Kruskall-Wallis, chi-squared tests, or Fisher exact test as appropriate. We demonstrate the distribution of all cysCsCr_diff values using kernel density plots.

Factors correlated with cysCsCr_diff

We used linear regression to determine the correlation between clinical variables that may bias sCr, as opposed to cysC, measurements. We determined the correlation between cysCsCr_diff and LFI, sex, HE status, ascites status, Black race, and MELD 3.0. Among the 525 participants enrolled in this study, 50 participants did not have LFI measurements because of a patient’s refusal to complete the measurements or an error in the documentation of their measurements. These participants were omitted from the analyses involving the association between LFI and cysCsCr_diff.

Comparing the association between cysC and sCr and waitlist mortality

The primary outcome of this analysis was waitlist mortality, defined as either a death on the waitlist or removal from the waitlist for sickness. For these analyses, time began at the time of biospecimen collection and ended at the time of death, liver transplant, or removal from the waitlist for sickness. Those patients lost to follow-up after their liver transplant evaluation (n = 4) were omitted from these analyses. To demonstrate that cysC has a stronger association with waitlist mortality than sCr, we compare a univariable Cox regression model with sCr-alone and one with cysC-alone using the likelihood ratio test. We generated Harrell’s C-statistics for the sCr-alone and cysC-alone models and compared them using boot-strap techniques. In sensitivity analyses, we completed competing risk analyses. For these analyses, time began at the time of biospecimen collection and ended at the time of death, liver transplant, or removal from the waitlist for sickness. Our primary outcome of interest was waitlist mortality, and receipt of a liver transplant was treated as a competing risk.

Demonstrating the independent association between cysCsCr_diff and waitlist mortality

We demonstrate the association between cysCsCr_diff and waitlist mortality in univariable Cox regression analyses. To demonstrate that cysCsCr_diff was associated with waitlist mortality independent of cysC, we completed a multivariable analysis adjusting for cysC and cysCsCr_diff. To demonstrate that cysCsCr_diff was associated with waitlist mortality independent of sCr, we completed a multivariable analysis adjusting for sCr and cysCsCr_diff. To demonstrate that cysCsCr_diff was associated with waitlist mortality independent of MELD 3.0, we completed a multivariable analysis adjusting for MELD 3.0 and cysCsCr_diff. We generated Kaplan-Meier plots with participants categorized based on the median MELD 3.0 (< and ≥ 18) and cysCsCr_diff values (< 0.35 and ≥ 0.35) into 4 categories: “low MELD 3.0 and low cysCsCr_diff,” “high MELD 3.0 and low cysCsCr_diff,” “low MELD 3.0 and high cysCsCr_diff,” and “high MELD 3.0 and high cysCsCr_diff.”

Software

Analyses were completed in R version 4.3.1 (Beagle Scouts) and R Studio using the following additional packages “flextable” and “gtsummary.”

Institutional review board

This study was approved by the IRB at the University of California, San Francisco.

RESULTS

Sociodemographics and clinical characteristics

We included 525 participants in this cohort study: 175 participants in the low, intermediate, and high cysCsCr_diff, each. As compared to those with a low cysCsCr_diff, we found that those in the intermediate and high groups were more likely to be female (low: 28%, intermediate: 46%, and high: 56%), to have ascites (low: 76%, intermediate: 90%, and high: 90%), to have HE (low: 73%, intermediate: 86%, and high: 81%), to have higher LFI measurements (low: 3.66 [3.09–4.17], intermediate: 3.79 [3.29–4.19], and high: 4.13 [3.69–4.62]), and to have higher MELD 3.0 scores (low: 17 [14–21], intermediate: 18 [15–20], and high: 20 [17–24]) (p < 0.05 for all) (Table 1). The kernel density plot shows the distribution of cysCsCrdiff for each of the 3 tertiles (Figure 1).

TABLE 1.

Characteristics by low, intermediate, and high difference between cystatin C and serum creatinine

Characteristic	Low, N = 175a	Intermediate, N = 175a	High, N = 175a	p b
Difference in cysC and sCr (mg/dL)	0.08 (−0.05, 0.16)	0.36 (0.29, 0.42)	0.69 (0.59, 0.94)	< 0.001
Sex, n (%)				< 0.001
Female	49 (28)	81 (46)	98 (56)
Male	126 (72)	94 (54)	77 (44)
Race, n (%)				0.046
Asian	11 (6.3)	7 (4.0)	15 (8.6)
Black	11 (6.3)	5 (2.9)	4 (2.3)
Other	0 (0)	4 (2.3)	5 (2.9)
White	153 (87)	159 (91)	151 (86)
Etiology, n (%)				0.14
ALD	66 (38)	75 (43)	79 (45)
Other	41 (23)	31 (18)	40 (23)
SLD	39 (22)	47 (27)	43 (25)
Viral hepatitis	29 (17)	22 (13)	13 (7.4)
Age (y)	57 (48, 63)	56 (49, 62)	57 (51, 62)	0.5
Ascites, n (%)				< 0.001
None	42 (24)	17 (9.7)	17 (9.7)
Mild/moderate	79 (45)	97 (55)	82 (47)
Severe	54 (31)	61 (35)	76 (43)
HE, n (%)				0.006
No HE	48 (27)	24 (14)	34 (19)
HE	127 (73)	151 (86)	141 (81)
Liver Frailty Index	3.66 (3.09, 4.17)	3.79 (3.29, 4.19)	4.13 (3.69, 4.62)	< 0.001
Serum sodium (mEq/L)	136 (134, 138)	136 (134, 138)	135 (132, 137)	< 0.001
Total bilirubin (mg/dL)	2.40 (1.30, 4.10)	2.90 (2.00, 3.90)	3.10 (1.70, 5.80)	0.006
INR	1.40 (1.20, 1.60)	1.50 (1.40, 1.75)	1.50 (1.30, 1.70)	< 0.001
Serum albumin (g/dL)	3.10 (2.70, 3.60)	3.00 (2.70, 3.40)	3.00 (2.60, 3.50)	0.6
cysC CKDEPI eGFR < 60 mL/min/1.73m2), n (%)	42 (8)	39 (7)	102 (20)	< 0.001
Serum creatinine (mg/dL)	0.84 (0.71, 1.23)	0.79 (0.66, 0.99)	0.95 (0.67, 1.36)	< 0.001
Cystatin C (mg/dL)	0.92 (0.77, 1.21)	1.16 (0.99, 1.34)	1.68 (1.36, 2.19)	< 0.001

^aMedian (IQR); n (%).

^bKruskal-Wallis rank sum test; Pearson’s chi-squared test; Fisher exact test.

Abbreviations: ALD, alcohol-associated liver disease; CKDEPI, chronic kidney disease epidemiology collaboration; cysC, Cystatin C; eGFR, estimated glomerular filtration rate; INR, international normalized ratio; sCr, serum creatinine; SLD, steatotic liver disease

FIGURE 1.

Kernel density plot for cysCsCr_diff distribution by tertile. Abbreviations: cysC, Cystatin C; cysCsCr_diff, the difference between cysC and sCr; sCr, serum creatinine.

Sociodemographics and clinical characteristics correlated with cysCsCr_diff

In univariable linear regression, we found the following were significantly associated with the cysCsCr_diff: men as compared to women (β = −0.20, p < 0.001); each 1-point increase in LFI (β = 0.10, p = 0.008); as compared to no ascites, mild/moderate ascites (β = 0.18, p = 0.04), and severe ascites (β = 0.20, p = 0.03); and as compared to no HE and HE (β = 0.16, p = 0.03) (Table 2). In multivariable linear regression models, the factors that were significantly associated with cysCsCr_diff were LFI (β = 0.09, p = 0.01) and male sex (β = −0.19, p < 0.001) (Table 2).

TABLE 2.

Covariates correlated with cysCsCr_diff on univariable and multivariable linear regression

	Univariable			Multivariable
Characteristic	β	95% CI	p	β	95% CI	p
Age per year	0.00	0.00, 0.01	0.7
Sex
Female	—	—		—	—
Male	−0.20	−0.32, −0.09	< 0.001	−0.19	−0.31, −0.08	< 0.001
Self-identified race
Non-Black	—	—
Black	−0.08	−0.37, 0.21	0.6
Liver Frailty Index per point	0.10	0.03, 0.17	0.008	0.09	0.02, 0.16	0.014
MELD 3.0	0.01	0.00, 0.02	0.055
Ascites
None	—	—
Mild/moderate	0.18	0.01, 0.35	0.043
Severe	0.20	0.02, 0.37	0.030
HE
No HE	—	—
HE	0.16	0.02, 0.30	0.025

Abbreviations: cysC, Cystatin C; cysCsCr_diff, the difference between cysC and sCr; MELD, Model for End-Stage Liver Disease; sCr, serum creatinine.

Comparing the association between cysC and sCr and waitlist mortality

Among the 521 participants who were not lost to follow-up, 132 (25%) participants died or were removed from the waitlist during follow-up, at a median of 350 days (IQR: 170–648). Compared to those with a low cysCsCr_diff, we found that those in the intermediate and high groups were more likely to die during follow-up (low” 18% vs. intermediate: 25% vs. high: 33%, p < 0.05 for both comparisons). Among our participants, 165 (32%) received a liver transplant during follow-up, at a median of 289 (168–496) days. We found no significant differences in the proportion of patients who received a liver transplant by the cysCsCr_diff tertile (low: 36% vs. intermediate: 33% vs. high: 30%; p = 0.55).

Overall, we found that each 1 mg/dL increase in sCr was associated with 1.04 times (95% CI: 0.92–1.17, p = 0.59) the hazard of waitlist mortality, while each 1 mg/dL increase in cysC was associated with 1.17 times (95% CI: 1.03–1.33, p = 0.02) the hazard of waitlist mortality. Overall, the log-likelihood (logLik) was significantly higher for the cysC model (logLik: −715.31) than the sCr model (logLik: −713.00) (p < 0.001). A model with cysC-alone, as compared to a model with sCr-alone, had a significantly higher c-statistic (0.58 vs. 0.54, p < 0.001) (Table 3). In univariable competing risk analyses, we found similar findings: each 1 mg/dL increase in sCr was associated with 1.07 times (95% CI: 0.86–1.19, p = 0.24), and each 1 mg/L increase in cysC was associated with 1.22 times (95% CI: 1.09–1.36, p < 0.001) the risk for waitlist mortality, after accounting for receipt of a liver transplant (Supplemental Table S1, http://links.lww.com/LVT/A622).

TABLE 3.

Cox model characteristics for sCr and cysC for overall mortality

Overall model	HR (95% CI)	p	logLika	C statistica
sCr	1.03 (0.92–1.17)	0.59	−715.3	0.54
cysC	1.17 (1.03–1.33)	0.02	−713.0	0.58

^aIndicates cysC significantly outperformed sCr.

Abbreviations: cysC, Cystatin C; logLik, Log Likelihood; sCr, serum creatinine.

Demonstrating the independent association between cysCsCr_diff and waitlist mortality

We next wanted to determine the association of cysCsCr_diff and waitlist mortality. In univariable Cox regression, we found that each 1 mg/dL increase in cysCsCr_diff was associated with 1.87 times (1.33–2.62) the hazard of waitlist mortality. After adjustment for age, sex, total bilirubin, INR, serum albumin, serum sodium, and sCr, we found that each 1 mg/dL increase in the cysCsCr_diff was associated with 1.60 times (1.19–2.17) the hazard of waitlist mortality. When participants were categorized into high and low MELD 3.0 and cysCsCr_diff categories, we found that as compared to the “low MELD 3.0 and low cysCsCr_diff,” each of the other categories had significantly higher waitlist mortality: “low MELD/high cysCsCr_diff”: HR 2.03 (95% CI: 1.07–3.86); “high MELD/low cysCsCr_diff”: HR 2.25 (95% CI: 1.18–4.39); “high MELD/high cysCsCr_diff”: HR 4.73 (95% CI: 2.67–8.39) (Figure 2). After adjustment for the total bilirubin, INR, serum sodium, albumin, age, sex, and cysC, we found that each 1 mg/dL increase in the cysCsCr_diff was associated with 1.30 times (1.03–1.91) the hazard of waitlist mortality.

FIGURE 2.

Kaplan-Meier plot for waitlist mortality by tertile of cysCsCr_diff. Abbreviations: cysC, Cystatin C; cysCsCr_diff, the difference between cysC and sCr; sCr, serum creatinine.

DISCUSSION

Stratifying risk for death among patients with decompensated cirrhosis is partly dependent on accurately quantifying kidney function—a finding highlighted by its continuous incorporation into the MELD score.^[1] Yet, this calculation depends on sCr, a metric known to be inaccurate among patients with cirrhosis, biased by sex, frailty, and hepatic synthetic function.^[5,8,12] Herein, we demonstrate that as compared to sCr, cysC has a significantly stronger association with waitlist mortality; we also introduce cysCsCr_diff, the difference between cysC and sCr in the cirrhosis population. This metric, recently described in the general nephrology population, quantifies factors not captured in the MELD score,^[4] factors which, based on our correlation analyses, appear to be sex, the degree of illness, and the underlying frailty. We demonstrate in our analyses that cysCsCr_diff is associated with waitlist mortality, a finding independent of cysC, sCr, and the other components of the MELD 3.0 score.

It is unsurprising that in our study, cysC outperformed sCr in predicting overall mortality, 90-day mortality, and 180-day mortality. These findings have been reported previously in non-US and nontransplant populations.^[15,22] Given the increased availability and decreasing cost of cysC measurements, our study further supports the investigation of cysC as the metric to quantify kidney function among patients with decompensated cirrhosis. Further, regardless of any price difference, the recommendation of cysC in several guidelines should provide sufficient support for cysC to be covered by insurers in our population.^[20,21] Although not powered to determine the mechanics of incorporating cysC into the MELD 3.0 score, ongoing collaborations will hopefully provide the accurate effect size estimates needed to incorporate cysC into the MELD 3.0 score.

We build on these findings by demonstrating that not only is cysC more associated with outcomes than sCr, but that the difference between the cysC and sCr may be a clinically significant biomarker. We highlight in our Cox regression analyses that each 1 mg/dL increase cysCsCr_diff was associated with a 60% increase, even after controlling for the components of the MELD 3.0 score. Similarly, when controlling for sex, total bilirubin, INR, albumin, sodium, and cysC, each 1 mg/dL increase in cysCsCr_diff was associated with a 30% increase in waitlist mortality. We hypothesized that this association would be related to the factors previously suggested to drive the difference between cysC and sCr—sex, frailty, and hepatic synthetic function. Our correlation analyses confirmed these hypotheses, demonstrating that women had, on average, a 0.2 greater cysCsCr_diff. Further, the correlation analyses also demonstrate a significant linear relationship between cysCsCr_diff and frailty. Each 1-point increase in the LFI was associated with a 0.1 increase in the cysCsCr_diff. This is an important finding, as there are several barriers to incorporating frailty, measured through the LFI, into the MELD score (eg, needing an in-person assessment and standardization among different centers); cysCsCr_diff, with its strong correlation with LFI, has the potential to serve as a serological biomarker of physical frailty, a finding which if validated would have important implications for the implementation of this key determinant of transplant outcomes into clinical practice.

This study has several limitations. The first limitation is the power. This study was not powered to provide the effect size estimates necessary to incorporate cysC into the MELD score or to fully compare a MELD score that incorporates the cysCsCr_diff and one that does not; further, our study was comprised of predominantly White patients with either alcohol-associated liver disease or steatotic liver disease. To address these shortcomings, an ongoing collaboration through the HRS-HARMONY consortium should increase the power and provide a patient profile that is representative of the United States.^[31] The second limitation is the lack of measured GFR. We did not complete assays that measure GFR (eg, iohexol clearance and iothalamate clearance); as such, our studies do not comment on calculating GFR using cysC accurately. Although this would be a critical question when stratifying the need for a simultaneous liver-kidney transplant, for our analyses to associate cysC and the cysCsCr_diff with waitlist mortality, measuring GFR was unnecessary. The third limitation is the novelty. Previous studies have highlighted the improved accuracy of cysC in estimating GFR, the development of acute kidney injury, and death among patients with cirrhosis.^[5,7,8,15] That said, these data represent the first study to associate cysC with liver transplant waitlist outcomes among patients in the United States. In addition, these data represent the first documentation of the clinical utility of cysCsCr_diff to predict waitlist mortality among patients with cirrhosis, an association we believe is driven by the strong correlation between cysCsCr_diff with measured physical frailty and sex.^[24–26,32]

Despite these limitations, our study demonstrates a significantly stronger association between cysC and waitlist outcomes than sCr and waitlist outcomes. The data introduce cysCsCr_diff as a metric that is not only strongly correlated with physical frailty and sex but also significantly associated with waitlist outcomes, even after controlling for the components of the MELD 3.0 score. Collectively, we believe these data provide added evidence for the implementation of cysC and cysCsCr_diff into the accurate assessment of the risk of death among patients with decompensated cirrhosis.

Abbreviations:

cysC

Cystatin C

cysCsCr_diff

the difference between cysC and sCr

FrAILT

Functional Assessment of the Study of Liver Disease

GFR

glomerular filtration rate

INR

international normalized ratio

LFI

Liver Frailty Index

MELD

Model for End-Stage Liver Disease

sCr

serum creatinine