Reply to: “Performance of reMELD-Na as a predictor of short-term mortality in patients with cirrhosis and ascites”

To the Editor:

We thank Quickert and colleagues for their thoughtful letter regarding our multicenter study on ReMELD-Na, MELD 3.0, and established prognostic models after transjugular intrahepatic portosystemic shunt (TIPS) placement for refractory ascites.¹ Their analysis is timely and clinically relevant, particularly after the recent implementation of ReMELD-Na for liver allocation in the Eurotransplant region.^2,3 We view their findings as complementary to ours. Together, both studies raise a broader issue: the transportability of allocation-based prognostic models when the clinical context — and, in the case of TIPS, the disease trajectory itself — changes.

The question is not whether ReMELD-Na is intrinsically “good” or “poor”, but in which population, for which endpoint, and for which clinical decision it is sufficiently informative. A score developed or refitted for liver allocation may perform differently in hospitalized decompensated cirrhosis and differently again after portal decompression.^4,5 This distinction is central to interpreting the apparent differences between the two studies.

In our cohort of 1,621 patients undergoing TIPS for refractory ascites, all investigated scores showed only modest discrimination for 90-day and 1-year death or liver transplantation, with AUROC values below 0.7 in the overall cohort. ReMELD-Na performed worse than Freiburg index of post-TIPS survival (FIPS), MELD 3.0, and MELD-Na for the 90-day endpoint, whereas MELD 3.0 showed performance comparable to established scores.¹ In contrast, Quickert et al. evaluated 918 hospitalized patients with cirrhosis and ascites and reported comparable discrimination of ReMELD-Na and MELD for 90-day mortality, with AUROCs of 0.738 and 0.729, respectively, while ReMELD-Na showed superior calibration.²

These findings are not contradictory. Rather, they illustrate that prognostic performance depends strongly on cohort architecture and decision context. Importantly, Quickert et al. themselves highlight that TIPS candidates are a highly selected subgroup and that portal decompression through TIPS “fundamentally alters the natural course of the disease and its complications”.² In other words, TIPS does not simply select patients; it changes the prediction target. A score calibrated to estimate mortality under the natural course of decompensated cirrhosis or on the transplant waiting list may therefore not retain identical performance after portal decompression.

For interpretation of the data by Quickert et al., it would be helpful to know how TIPS was handled in their cohort. Patients with prior TIPS, TIPS during follow-up, or planned TIPS may differ substantially from hospitalized patients managed without portal decompression. Inclusion, exclusion, or censoring of such patients could affect both discrimination and calibration of ReMELD-Na.

A second important issue concerns transplant candidacy. Because ReMELD-Na was developed for allocation, the extent to which a validation cohort approximates a transplant-candidate population is central to interpreting model transportability.³ It would therefore be informative to know whether patients in the cohort by Quickert et al. were systematically assessed for transplant eligibility, and whether contraindications such as severe comorbidity, frailty, uncontrolled infection, malignancy, active alcohol use, age, or psychosocial barriers were recorded or excluded. A broad hospitalized ascites cohort is highly relevant for prognostic assessment, but it may not fully reflect the intended allocation setting of ReMELD-Na.

Endpoint architecture is another relevant difference. Quickert et al. excluded patients transplanted within 90 days to avoid competing-risk bias, whereas our primary endpoint considered death or liver transplantation as a clinically meaningful composite outcome after TIPS.^1,2 To improve comparability, we performed an additional sensitivity analysis excluding patients with liver transplantation within 90 days after TIPS and repeated the 90-day mortality analysis.

After exclusion of patients transplanted within 90 days, 1,598 patients remained. The 90-day mortality rate was 16.4%. AUROC values as well as high-risk stratification analysis (log-rank test for 90-day mortality) results remained mostly unchanged for all scores: ReMELD-Na (AUROC: 0.625 [0.589-0.660]; log-rank p <0.001), MELD 3.0 (AUROC: 0.646 [0.610-0.682]; log-rank p <0.001), FIPS (AUROC: 0.673 [0.639-0.708]; log-rank p <0.001), MELD (AUROC: 0.628 [0.591-0.666]; log-rank p <0.001), and MELD-Na (AUROC: 0.647 [0.612-0.683]; log-rank p <0.001). ReMELD-Na remained inferior to FIPS (p <0.001), MELD 3.0 (p = 0.013) and MELD-Na (p = 0.010) in the overall cohort as assessed by DeLong’s test, suggesting that our findings were not primarily driven by inclusion of transplantation in the composite endpoint. Regarding analyses of the female and male subcohorts, only minor differences were detected. In the female subcohort, ReMELD-Na was now inferior to FIPS (p = 0.043), while in the male subcohort, ReMELD-Na remained inferior to MELD 3.0, FIPS and MELD-Na (Table 1).

Table 1.

AUROC & DeLong´s test.

Compared score/AUROC	p (90-day)	AUROC difference (90-day)
Overall
MELD 3.0/0.646	0.013	-0.021
FIPS/0.673	<0.001	-0.048
MELD/0.628	0.727	-0.003
MELD-Na/0.647	0.010	-0.022
Female
MELD 3.0/0.705	0.138	-0.021
FIPS/0.722	0.043	-0.038
MELD/0.684	0.980	0
MELD-Na/0.708	0.117	-0.023
Male
MELD 3.0/0.612	0.030	-0.022
FIPS/0.645	<0.001	-0.054
MELD/0.596	0.650	-0.006
MELD-Na/0.612	0.038	-0.022

Comparison of the AUROC of ReMELD-Na in the entire study cohort (0.625) and the female (0.684) and male (0.590) subcohort with MELD 3.0, FIPS, MELD-Na and MELD in the entire study cohort after exclusion of patients undergoing liver transplantation within 90 days after TIPS. Comparisons were performed using DeLong’s test. P values and C-index differences are reported for 90-day mortality. FIPS, Freiburg index of post-TIPS survival; MELD, model for end-stage liver disease; ReMELD, refitted MELD.

The sex-specific findings further support the context-dependent nature of score performance. In our post-TIPS cohort, AUROCs were consistently higher in women than in men, whereas Quickert et al. observed higher AUROCs in men in their hospitalized ascites cohort.^1,2 This suggests that sex effects are not fixed properties of the score alone, but may depend on cohort composition, intervention status, endpoint definition, and the intended clinical use of the model.

We agree that ReMELD-Na may provide useful prognostic information in patients with cirrhosis and ascites, particularly for estimating absolute mortality risk. However, our findings indicate that its validity should not be automatically transported to the post-TIPS setting. Risk prediction after TIPS likely requires multidimensional tools incorporating liver and renal function, portal pressure response, procedural factors, frailty, inflammation, and sex-specific effects.

In conclusion, Quickert et al. validate ReMELD-Na in a broader hospitalized ascites population, whereas our study highlights its limitations after TIPS for refractory ascites. Together, these data support a more nuanced framework: prognostic models should be judged not only by their AUROC, but by their intended population, endpoint, intervention context, and clinical use.

Authors' contributions

Conceptualization: Markus Kimmann, Michael Praktiknjo. Methodology: Markus Kimmann, Dominik Bettinger, Michael Praktiknjo. Data curation: Markus Kimmann, Nancy Farouk, Johannes Chang, Dominik Bettinger, Michael Praktiknjo. Investigation: Markus Kimmann, Nancy Farouk, Dominik Bettinger, Michael Praktiknjo. Validation: Markus Kimmann, Nancy Farouk, Michael Praktiknjo. Formal analysis: Markus Kimmann, Nancy Farouk, Michael Praktiknjo. Supervision: Jonel Trebicka, Michael Praktiknjo. Funding acquisition: Michael Praktiknjo. Visualization: Markus Kimmann, Nancy Farouk, Michael Praktiknjo. Project administration: Jonel Trebicka, Michael Praktiknjo. Resources: Jonel Trebicka, Michael Praktiknjo. Writing - original draft: Markus Kimmann, Michael Praktiknjo. Writing - review & editing: Markus Kimmann, Nancy Farouk, Dominik Bettinger, Jonel Trebicka, Michael Praktiknjo.

Data availability

Requests for data access can be directed to the corresponding author and the German Cirrhosis Study Group and will be granted upon reasonable request.

Financial support

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Conflict of interest

Dominik Bettinger has received grants from German Research Foundation and Schwiete Foundation, honoraria from W.L. Gore & Associates and Falk Foundation and support for attending meetings/travel from Abbvie. Jonel Trebicka has been supported by the German Research Foundation (DFG), German Federal Ministry of Education and Research, Hessian Ministry of Higher Education, Research and the Arts (HMWK) and European Union – Horizon 2020 (MICROB-PREDICT ID 825694, DECISION ID 847949, GALAXY ID 668031, LIVERHOPE ID 731875, IHMCSA ID964590) and has received consulting fees and honoraria from AstraZeneca, W.L. Gore & Associates, Boehringer Ingelheim, Falk, Grifols, Genfit, CSL Behring and Versantis, Michael Praktiknjo has been supported by the German Research Foundation (DFG), BONFOR and the Ernst-und Berta Grimmke Foundation and has received consulting fees, honoraria and support for attending meetings/travel from speaking fees from W.L. Gore and Associates, Orphalan, Gilead, Falk, Univar, Ipsen, Roche, AstraZeneca, Boston Scientific and MicroTec.

Please refer to the accompanying ICMJE disclosure forms for further details.

Supplementary data

The following are the Supplementary data to this article.