Review Reports - Impact of Ascites on Morbidity and Length of Hospital Stay: A Large Retrospective Study from a Tertiary Referral Center

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

Dear Editor and Authors,

I reviewed with interest the study "Impact of Ascites on Morbidity and Length of Hospital Stay: A Large Retrospective Study from a Tertiary Referral Center."

There is a clear improvement in the presentation and organization of the article.

Based on a retrospective cohort study the authors demonstrate that the presence of ascites (ICD-10 code R18) translates to a higher hospital burden, greater readmission burden and with higher in-hospital mortality.

The difficulty I may have in interpreting the results stems from the lack of an etiological classification for ascites. According to the results presented, 63% of patients have associated cirrhosis, but we have a percentage of 28.5% who will have cirrhosis and cancer or only cancer, as well as 8.5% whose diagnoses are not specified. With nearly 30% of cases related to malignancy, it is difficult to attribute observed differences in length of stay, readmission, or in‑hospital mortality to ascites per se rather than underlying cancer.

There is an appropriate statistical approach for the data types. This is much more powerful than using simple linear regression for skewed hospital data. And the comparison between Model A and Model B is excellent.

I think it's important to be more objective about the real impact this study could have. When you mention "service planning," it means you believe that if ascites is present, it will be possible to predict the length of hospital stay or its cost, integrating strategies to modify this impact. If so, what strategies would they be?

In the discussion, the section on physician burnout feels slightly disconnected. You don't measure burnout, so this connection is speculative. I would remove the burnout paragraph to maintain the paper focused on patient outcomes and hospital resources.

This is an important dataset and the analytic approach is sound. Addressing the etiologic classification of ascites (or clearly acknowledging the limitation) and removing speculative links to burnout will substantially strengthen the manuscript and clarify its implications for clinical practice and service planning.

Author Response

Thank you for the feedback and the thoughtful comments.

Comment 1

Response:

We acknowledge this limitation and we thank you for pointing this out.

To address a negative effect of ascites regarding LOS and mortality in both groups (patients with or without cancer), we compared the mean LOS in admissions with or without ascites, separated for the group of patients without a diagnosis of cancer and for the group of patients with a diagnosis of cancer (independent samples T test). The effect was slightly smaller for patients with cancer, but remained significant (mean difference -2.8 days for patients without ascites [CI-3.93;-1.66], p<0.0001 in the group of neoplasic patients and mean difference -6.47 days for patients without ascites [CI-7.36;-5.57] for patients without cancer, p<0.0001). To verify, we grouped the patients based on the median LOS (staying less than 7 days or more than 8 days in the hospital) and confirmed through the χ²test that for both patients with or without neoplasms ascites was associated with longer LOS (patients with neoplasms: χ² = 19.229, p<0.0001; patients without neoplasms: χ² =202.977, p<0.0001)

Furthermore, the presence of ascites associated with a higher mortality rate for both groups (patients with neoplasms: χ² = 8.201, p=0.004; patients without neoplasms: χ² =267.099, p<0.0001).

We introduced this data in the new version of the manuscript.

Comment 2

Response

We appreciate this encouragement of approaching the subject more pragmatically. Therefore, we introduced a sequence discussing possible benefits of service planning based on a known negative effect of the R18 diagnosis. Furthermore, we adjusted the text in order to make sure that the ”service planning” claim is not too strong, highlighting the need for further studies that would verify such interventions.

Speciffically, we introduced the following paragraph in the discussion:

” Collectively, these findings support the use of ICD-10 R18-coded ascites as anadministrative marker of a subgroup of patients with higher expected resource use in Gastroenterology and Internal Medicine settings, when physiological severity measures are unavailable at scale. Similar administrative markers are used in practice for case-mix adjustment and, more cautiously, for capacity planning and for the identification of patients that might benefit from more tailored interventions[26,27]. In a metaanalysis by Mao et.al., more than half of the prediction models for readmissions were based on retrospective administrative data[27].Scores such as the Hospital Frailty Risk Score, for example, are explicitly built from ICD-10 codes to flag a high-risk group and are shown to relate to outcomesPredicting a higher risk of longer LOS or higher readmission count could be used to identify patients in which a transitional care intervention might be needed. In this context, if validated locally, R18-coded ascites could be used to flag patients who may require earlier discharge planning, closer coordination of follow-up, and anticipatory allocation of ascites-related resources. In particular, prior literature suggests that the implementation of timely diagnostic paracentesis in hospitalized patients with cirrhosis and ascites is associated with lower mortality and shorter hospital stay, and that dedicated outpatient/day-case paracentesis services may reduce admissions, emergency department utilization, inpatient bed days, and costs[28]. Furthermore, recognition of the higher-risk context in which ascites often occurs, particularly in decompensated disease, may support earlier goals-of-care and advance care planning discussions with patients and families when clinically appropriate. These applications were not tested in the present study and should therefore be interpreted as potential service-planning implications to be evaluated in future prospective work.”

Thank you for pointing this out. We eliminated this paragraph.

Thank you for the careful reading of the manuscript and for the valuable suggestions.

Author Response File: Author Response.docx

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

Overall, the clinical question is clear and interesting. Using an administrative marker to identify a high-resource subgroup is a reasonable “real-world” framing, and the outcomes the authors chose to align with that framing, but several design and analysis issues should be addressed.

Specific comments

-You state that you’re counting readmissions over the available observation window and that the window varies by index year. But you still present IRRs that will be heavily influenced by follow-up duration and calendar-time censoring. In other words, patients first admitted in 2015 simply have more opportunity to accumulate readmissions than patients first admitted in 2022/2023, regardless of clinical risk. You may consider reframing readmission as a fixed-window outcome (e.g., 30day, 90-day, or even 1-year), at least for the first admission per patient. Right now, the readmission results are too easy to misread as “ascites causes more readmissions,” when they may partially reflect “earlier calendar entry leads to more observable readmissions.”

-You work with admissions for LOS and mortality, but patients can contribute multiple admissions. If you analyze at the admission level without addressing within-patient correlation, your standard errors can be optimistic, and p-values can look stronger than they should.

-You describe deriving the CCI from ICD-10 codes over the entire 2015–2023 period and then using it for stratification. If that means a patient’s CCI includes diagnoses recorded after an earlier admission, then for early admissions you are effectively using information that would not have been known at that time. That is fine if your goal is to capture lifetime comorbidity burden over the whole dataset, but it is not appropriate if you want to present the CCI as something like baseline risk at admission. Please clarify in the Methods exactly how the CCI is built relative to the admission date.

-Finally, the Decompensation Score (DS) may need clearer justification and more careful interpretation. I understand that you are trying to account for liver-disease severity beyond the CCI when introducing the DS. However, the DS is constructed from ICD-coded complications that are plausibly on the same causal pathway as ascites. Once you adjust for those, you may be “adjusting away” the very severity signal that R18 is supposed to represent. A more explicit conceptual explanation of what the DS is doing (confounder control vs. severity decomposition vs. overlap accounting) would be helpful.

Author Response

Thank you for the thoughtful commentary and valuable feedback.

Comment 1

You state that you’re counting readmissions over the available observation window and that the window varies by index year. But you still present IRRs that will be heavily influenced by follow-up duration and calendar-time censoring. In other words, patients first admitted in 2015 simply have more opportunity to accumulate readmissions than patients first admitted in 2022/2023, regardless of clinical risk. You may consider reframing readmission as a fixed-window outcome (e.g., 30day, 90-day, or even 1-year), at least for the first admission per patient. Right now, the readmission results are too easy to misread as “ascites causes more readmissions,” when they may partially reflect “earlier calendar entry leads to more observable readmissions.”

Response:

We agree to this observation. We revised the manuscript text in order to prevent erroneous causal interpretation, and we added new statistical analysis data, evaluating readmissions at 30 days and 90 days in patients with or without ascites. The new analysis showed that patients with ascites were 2.6 (CI 2.1-3.22) times more likely than patients without ascites to be readmitted at 30 days and 2.55 (CI 2.06-3.14) times more likely to be readmitted at 90 days. We included this in the results.

Comment 2:

Response:

This observation is important and we recognise this. The analysis for LOS and mortality was based on admissions (n=16804), meaning that standard errors could be underestimated due to intra-patient clustering. For future studies, this could be adressed either by restricting the model to the first admission of each patient, or by using robust, clustered standard errors. In this study we opt for specifically presenting this limitation in the text, while also mentioning that the analysis of readmissions (which might be the most vulnerable to this issue) was re-evauated patient wise (n=13264), without being affected (the nature of the result persisted).

The next sentence was added to the ”Limitations” paragraph:

”Another limitation is that the LOS and mortality were evaluated on the admission level – therefore, if part of the patients contribute with multiple admissions, the standard errors could be slightly underestimated due to intra-patient correlations, and the results must be interpreted in this light. However, the analysis of readmissions, that might be one of the most influenced by this effect, remained valid for both patient-wise and admission-wise analyses. ”

Comment 3

Response

Thank you for pointing out the lack of clarity. The following text was added in the Materials and Methods section of the manuscript:

CCI was calculated on the patient level, gathering all ICD10 codes from all the admissions recorded in the 2015-2023 interval. This approach allows for an estimated of the cumulative comorbidity during the entire interval and is not restricted to codes present at a particular admission (especially given the fact that acute diagnoses tend to be included far more often than diagnoses that are not immediately relevant to a particular admission). Therefore, in this study, the CCI should be interpreted as a severity proxy that is temporally aggregated and not as a prospective risk score.

The following text was added to the Limitations paragraph:

Moreover, since CCI was aggregated for the whole time-interval for each patient, for patients with a longer window of observation, more diagnoses might be included (bias look-ahead), and we recognize this as a limitation of the study.

The following text was added to the Material and Methods section:

The B model must be interpreted as a stability analysis that evaluated the superposition of the administrative signals, and not as a causally-adjusted model. The components of the DS include decompensation events that could co-occur with ascites. Adjusting for DS in the model used for the presence of R18 is a super-adjustment, that is expected to attenuate the association with morbidity outcomes, and is not used to control independent confusion factors. We use model A (adjusted for age, gender, CCI) as the main analytic model, and we only include the B model to address the degree of superposition between ascites (R18) and other decompensation proxies.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

With the revision implemented by the authors, there is greater clarity and fluidity in the analysis of the results.
The conclusion proves to be valid and consistent.
Nothing more to add.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Editors and Authors,

I reviewed with interest the submitted article "Impact of Ascites on Morbidity and Length of Hospital Stay: A Large Retrospective Study from a Tertiary Referral Center" evaluating the impact of ascites on hospital length of stay (LOS) and admission frequency.

The study demonstrates that the presence of ascites is associated with increased readmissions, longer hospital stays, and higher mortality. This topic is of significant scientific interest; however, I have several concerns regarding the methodology and the presentation of the results that need to be addressed to optimize the manuscript’s validity and readability.

Major Comments

Lack of Severity Scoring (Page 3, Lines 94-96) The authors note the absence of severity scores or clinical details to objectify the severity of the analyzed cases. This is a significant limitation that may introduce bias. Please elaborate on how this bias was mitigated. If it cannot be counterbalanced, the authors must discuss how this influences the interpretation of the results.
Study Population and Etiology (Page 3, Lines 104-108) The inclusion criteria regarding the etiology of ascites are unclear. It is difficult to determine if cases of ascites not associated with liver disease (e.g., heart failure, malignancy) were excluded or kept for analysis. This distinction is critical, as the clinical course and prognosis differ significantly between cirrhosis-related ascites and other etiologies. The authors must clarify the specific etiologies included. If the study mixes different causes (e.g., cirrhosis vs. heart failure vs. cancer), this introduces significant heterogeneity. I recommend either analyzing these groups separately or excluding ambiguous/non-hepatic cases to ensure a homogeneous study population.
Presentation of Results The text should highlight the key takeaways. I recommend presenting the results more succinctly, utilizing tables or figures for objective data. The current extensive narrative across ten subsections can hinder readability.

Minor Comments

Page 2, Line 43: The sentence ending with "(...) survival, such as [3]" appears to be incomplete. Please revise.
Page 2, Lines 54-55: The citation for the "Global Burden of Disease Report" is missing.
Page 3, Lines 131-134: The manuscript describes details of specific cases. Individual case descriptions are generally not beneficial in a large retrospective study unless they represent a specific qualitative point. If these are outliers, they should be excluded or handled statistically.

The discussion section is well-structured and effectively articulates the findings in relation to the results.

In summary, I suggest enhancing clarity regarding the methods and the analyzed population. It may also be prudent to exclude ambiguous cases concerning the etiology of ascites from consideration.

Thank you for the opportunity to review this important work.

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript addresses the burden of ascites in a Romanian tertiary care hospital using ICD-10–based retrospective data. Several methodological limitations significantly restrict the study’s interpretability and generalizability.

1. Ascites Etiology Classification is Inadequate

The study identifies ascites cases based solely on ICD-10 coding (R18), without subclassifying by underlying etiology (e.g., cirrhotic vs. malignant vs. nephrogenic vs. cardiac).

Although the authors acknowledge this in their limitations (p.12), this design choice severely limits the clinical interpretability of the findings. As is well established, the natural history, management, and prognosis of ascites differ markedly depending on the underlying cause.

For example, the mortality, hospital stay, and re-admission risk for cirrhotic ascites differ substantially from those for peritoneal carcinomatosis or heart failure-induced ascites.

Recommendation: Future iterations of this study should incorporate either a chart review-based subclassification or use a combination of ICD codes (e.g., K74.6, I50.0, C22.0) to approximate etiology-specific groupings.

2. Limitations of ICD-Based Ascites Detection in Single-Center Design

The reliance on administrative data (ICD-10) without clinical validation introduces well-known biases:
Overcoding (for reimbursement),
Undercoding (due to time constraints or poor documentation),
R/O (rule-out) diagnoses being misclassified as confirmed.
This is particularly problematic in a single-center design, where coding practices may reflect local institutional patterns rather than broader clinical reality.

Recommendation: Multi-center validation or a linkage to national registries could enhance robustness in future work.

3. Study Design Lacks Novel Contribution or Insight

While the study provides descriptive statistics (LOS, mortality, readmissions), it does not offer hypothesis-driven analyses or clinically actionable conclusions.
There are no multivariable models to adjust for confounders (e.g., age, comorbidities, sex), nor stratification by disease etiology and severity (e.g., Cirrhotic ascites, Malignantic ascites, and MELD, Child-Pugh), which are standard in modern hepatology research.

4. Given the international scope of Medicina as a scientific journal, the heavily local and contextual nature of this study—centered around a single Romanian institution without broader regional validation—further reduces its generalizability and potential impact for an international readership.

5. Figures and Visual Data Presentation
Several of the figures (e.g., Figure 2 and Figure 3) are not clearly legible. The resolution appears suboptimal, and the visual layout is cluttered or lacks sufficient contrast.