Review Reports - The Forgotten Anion: Serum Chloride and Its Association with All-Cause Mortality Across Chronic Kidney Disease and Dialysis—A Systematic Review and Meta-Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This systematic review addresses an original and clinically relevant question. The manuscript is generally well written, and the separate consideration of hypochloraemia and hyperchloraemia, use of adjusted estimates, QUIPS assessment, and GRADE framework are strengths. However, important analytical and reporting problems currently limit the reliability of the pooled estimates and require major revision.

The continuous-chloride meta-analysis is not valid as presented. Valga et al. [26] report an HR per 1-unit increase in chloride, whereas Nakaya et al. [27] report an HR per 1-standard-deviation increase. These estimates are on different exposure scales and cannot be pooled directly. Reversing their direction does not resolve this incompatibility. The pooled HR of 0.83, Figure 6A, and the related statements in the Abstract, Table 2, Results, and Discussion should be removed or recalculated after conversion to a common unit. Otherwise, these studies should be presented narratively.
The precision of the primary analysis also requires reconsideration. Using the estimates reported in Table 1 reproduces the pooled HR of approximately 2.46 and the Hartung–Knapp CI of 2.14–2.83. However, this interval is substantially narrower than the reported fixed-effect CI of 1.79–3.38 because the Hartung–Knapp variance factor is below 1. It is therefore misleading to describe this method as providing “appropriately wide” confidence intervals. With only three studies, a modified Hartung–Knapp approach or another adequately justified method should be considered. Exact τ² and Q values, model settings, and, where appropriate, a cautiously interpreted prediction interval should be reported. All pooled analyses and GRADE assessments should then be updated.
Egger’s regression is not informative with only three comparable categorical studies, and the reported p = 0.95 should be removed. Figure 7 is also methodologically problematic because it combines extreme-category and per-unit estimates, which remain on different scales despite directional harmonisation. This funnel plot should be deleted.
Important PRISMA information is missing. The complete database-specific search strategies, including all terms, limits, filters, dates, and the ClinicalTrials.gov query, should be provided in a supplement. Availability from the corresponding author does not satisfy PRISMA item 7. The authors should also state how many reviewers independently performed screening, full-text assessment, extraction, QUIPS assessment, and GRADE evaluation, and how disagreements were resolved. “Verification in a second pass” does not demonstrate independent duplicate review.
Registration occurred after the search and after study selection had commenced. This is appropriately disclosed, but analyses should not be described as pre-specified without documentation that the relevant decisions preceded examination of the results. The authors should provide a clear timeline, identify the applicable protocol version, and report all deviations or post hoc additions. The PRISMA checklist currently overstates compliance with several items.
I² = 0% should not be interpreted as evidence of clinical homogeneity when only three studies with different chloride thresholds and CKD populations were pooled. Similarly, the comparison of one non-dialysis CKD study with two PD studies is descriptive rather than a meaningful subgroup analysis. Claims of consistency across the entire CKD–dialysis spectrum should be tempered.
Table 1 should include the complete adjustment set, precise reference category, baseline versus time-varying exposure, and unit of each continuous estimate. Study-level QUIPS rationales and complete Newcastle–Ottawa assessments should also be provided. The table is currently excessively compressed and difficult to read.
Finally, the included studies demonstrate associations but do not evaluate discrimination, calibration, or incremental predictive performance. Serum chloride should not yet be presented as a validated prognostic marker suitable for risk stratification. The title and conclusions should emphasise an association with mortality. Figure 3 should also distinguish macula-densa/renin signalling from WNK-mediated chloride sensing and avoid conflating hypochloraemic metabolic alkalosis with low-bicarbonate metabolic acidosis. References 14 and 15 are not meta-analyses and do not support the corresponding statement in the Introduction.

Author Response

We thank the Editor and both reviewers for their careful, expert reading of our manuscript. The comments were detailed and constructive, and they have substantially improved the rigour and honesty of the paper. We agree with the great majority of the points raised, and we have made major revisions accordingly. All changes are highlighted in the revised manuscript. Below we respond to each comment point by point; reviewer comments are shown in italics and our responses in plain text, with the location of each change indicated.

We note at the outset that several comments converged on the same core issue — that the original manuscript in places overstated what observational, aggregate-level data can support. We have corrected this throughout: the pooled continuous analysis has been removed, the confidence interval of the primary estimate has been widened to an honest value, the funnel plot and Egger's test have been deleted, the language has been changed from causal/independent to associational, and the results and discussion have been reorganised by kidney-function stratum. We believe the revised manuscript is considerably more defensible.

We are grateful for this rigorous methodological review. We have adopted essentially all of the recommendations.

Comment 1.1 — "The continuous-chloride meta-analysis is not valid as presented. Valga et al. report an HR per 1-unit increase in chloride, whereas Nakaya et al. report an HR per 1-standard-deviation increase. These estimates are on different exposure scales and cannot be pooled directly... The pooled HR of 0.83, Figure 6A, and the related statements in the Abstract, Table 2, Results, and Discussion should be removed or recalculated after conversion to a common unit. Otherwise, these studies should be presented narratively."

Response: We fully agree; this was a genuine error. An HR per 1 mmol/L (Valga) and an HR per 1 SD (Nakaya) are on different exposure scales, and directional reversal does not make them commensurable. Because the standard deviation of serum chloride was not reported in Valga et al., a valid unit conversion is not possible from the available data. We have therefore removed the pooled continuous estimate (HR 0.83) entirely. Figure 6A has been deleted, the corresponding row has been removed from Table 2, and the pooled continuous statement has been removed from the Abstract, Results and Discussion. The two continuous-exposure haemodialysis cohorts are now presented narratively and separately, with their differing exposure metrics stated explicitly (Results, new Section 3.6; Discussion). We are grateful for this correction.

Comment 1.2 — "The precision of the primary analysis also requires reconsideration... this interval is substantially narrower than the reported fixed-effect CI of 1.79-3.38 because the Hartung-Knapp variance factor is below 1. It is therefore misleading to describe this method as providing 'appropriately wide' confidence intervals. With only three studies, a modified Hartung-Knapp approach or another adequately justified method should be considered. Exact tau-squared and Q values, model settings, and, where appropriate, a cautiously interpreted prediction interval should be reported."

Response: We agree. The reviewer is correct that with three studies and no observed heterogeneity the Hartung-Knapp variance factor was <1 (q = 0.039, sqrt = 0.20), which inappropriately narrowed the interval; describing it as "appropriately wide" was wrong and has been removed. We now report the primary pooled estimate with the standard DerSimonian-Laird random-effects confidence interval, which coincides with the fixed-effect interval because tau-squared = 0:

Primary pooled adjusted HR 2.46 (95% CI 1.79-3.38).

We now report the exact model output in full (Methods 2.5 and Results 3.4): tau-squared = 0.000, Q = 0.079 (df = 2, p = 0.96), I-squared = 0%, DerSimonian-Laird estimator. We additionally applied a modified Hartung-Knapp approach (variance factor bounded at 1) and report the resulting, deliberately conservative interval together with a cautiously interpreted 95% prediction interval; with only three studies both are very wide (approximately 1.23-4.94) and we state explicitly that they are of limited inferential value at k = 3. The headline estimate throughout the manuscript (Abstract, Table 2, Results, Discussion) has been updated to HR 2.46 (95% CI 1.79-3.38), and the GRADE assessment has been re-derived from this wider interval.

Comment 1.3 — "Egger's regression is not informative with only three comparable categorical studies, and the reported p = 0.95 should be removed. Figure 7 is also methodologically problematic because it combines extreme-category and per-unit estimates... This funnel plot should be deleted."

Response: Agreed on both counts. Egger's regression is uninformative with three categorical studies, and the p-value has been removed. The funnel plot (former Figure 7) combined estimates on different scales and has been deleted. We now state that formal small-study/reporting-bias testing was not appropriate given fewer than ten studies on a common scale (Cochrane guidance), and we address the possibility of reporting bias narratively and carry it explicitly into the GRADE publication-bias domain (Methods 2.5; Results 3.8). Figures have been renumbered accordingly.

Comment 1.4 — "Important PRISMA information is missing. The complete database-specific search strategies... should be provided in a supplement. Availability from the corresponding author does not satisfy PRISMA item 7. The authors should also state how many reviewers independently performed screening, full-text assessment, extraction, QUIPS assessment, and GRADE evaluation, and how disagreements were resolved. 'Verification in a second pass' does not demonstrate independent duplicate review."

Response: Agreed. (a) The complete, database-specific search strategies for all four databases (PubMed, OpenAlex, Web of Science Core Collection, Embase), including every search line, field tag, filter, limit and the date each database was last searched, together with the ClinicalTrials.gov query, are now provided in full in Appendix B of the revised manuscript (and we are glad to place them as Supplementary Material if the Editor prefers). (b) We have corrected the description of the review process to reflect what was actually done and no longer imply independent duplicate review where it did not occur. The revised Methods (2.3) now states plainly that title/abstract screening, full-text assessment and data extraction were performed by one reviewer (K.A.A.) and independently checked in full by a second author (M.A.), with any discrepancies resolved by discussion and reference to the source article; risk-of-bias (QUIPS/Newcastle-Ottawa) and GRADE assessments were made by one author and verified by a second. We have added the single-reviewer-with-independent-verification design as an explicit limitation (Section 4.1) and have removed language implying fully independent duplicate screening.

Comment 1.5 — "Registration occurred after the search and after study selection had commenced... analyses should not be described as pre-specified without documentation... provide a clear timeline, identify the applicable protocol version, and report all deviations or post hoc additions. The PRISMA checklist currently overstates compliance with several items."

Response: Agreed. We have removed all descriptions of the analyses as "pre-specified." Methods 2.1 now gives an explicit timeline: the search and study selection commenced before the protocol was registered on the Open Science Framework (https://osf.io/9p4hj), so the registration was retrospective and the review was not prospectively registered. We identify the applicable protocol version and now list, transparently, the analyses that were post hoc (the sensitivity, continuous-narrative, and cardiovascular-mortality analyses, and the GRADE domains). The PRISMA 2020 checklist (Appendix A) has been re-scored conservatively so that it reflects actual compliance, including partial or non-compliance where relevant (e.g., single-reviewer screening under item 8, retrospective registration under item 24, no formal small-study testing under item 14).

Comment 1.6 — "I-squared = 0% should not be interpreted as evidence of clinical homogeneity when only three studies with different chloride thresholds and CKD populations were pooled. Similarly, the comparison of one non-dialysis CKD study with two PD studies is descriptive rather than a meaningful subgroup analysis. Claims of consistency across the entire CKD-dialysis spectrum should be tempered."

Response: Agreed. We now state that I-squared = 0% with three studies indicates only an absence of detectable statistical heterogeneity and is not evidence of clinical or physiological homogeneity, particularly given differing chloride thresholds and case-mix (Results 3.4; Discussion). The former "subgroup" comparison of one non-dialysis CKD cohort against two peritoneal-dialysis cohorts is now presented explicitly as a descriptive observation, not a formal subgroup analysis. Claims of consistency "across the CKD-dialysis spectrum" have been substantially tempered throughout, in line with Reviewer 2's parallel comment.

Comment 1.7 — "Table 1 should include the complete adjustment set, precise reference category, baseline versus time-varying exposure, and unit of each continuous estimate. Study-level QUIPS rationales and complete Newcastle-Ottawa assessments should also be provided. The table is currently excessively compressed and difficult to read."

Response: Agreed. Table 1 has been expanded and de-compressed. It now reports, for each study, the complete multivariable adjustment set, the precise reference/comparison category, whether the chloride exposure was measured at baseline or as a time-varying covariate, and the exact unit of each continuous estimate. Full study-level Newcastle-Ottawa item scores and the QUIPS domain rationales are now provided in a dedicated table (Appendix C / new Table 3), rather than summarised only in the figure.

Comment 1.8 — "the included studies demonstrate associations but do not evaluate discrimination, calibration, or incremental predictive performance. Serum chloride should not yet be presented as a validated prognostic marker suitable for risk stratification. The title and conclusions should emphasise an association with mortality. Figure 3 should also distinguish macula-densa/renin signalling from WNK-mediated chloride sensing and avoid conflating hypochloraemic metabolic alkalosis with low-bicarbonate metabolic acidosis. References 14 and 15 are not meta-analyses and do not support the corresponding statement in the Introduction."

Response: Agreed on all points. (a) We now state explicitly that the included studies report associations and that none evaluated discrimination (c-statistic), calibration, or incremental predictive value over established risk factors; accordingly serum chloride is not a validated prognostic marker ready for clinical risk stratification, and we have added this as a specific limitation and a future-research priority. (b) The Conclusions now frame serum chloride as carrying prognostic information / being associated with mortality, not as a validated risk-stratification tool. (c) Figure 3 has been revised so that macula-densa / renin signalling and WNK-SPAK/OSR1 chloride sensing are shown as distinct pathways, and the figure and its legend now clearly separate hypochloraemic metabolic alkalosis from low-bicarbonate metabolic acidosis rather than conflating them; the pathways are labelled as proposed/hypothesised mechanisms. (d) We have corrected the Introduction sentence that cited references 14 and 15: those are a narrative review and a primary cohort study, not meta-analyses, and the sentence no longer describes them as such (it now describes the prior evidence accurately, with the citations retained only where they correctly support the statement).

Reviewer 2 Report

Comments and Suggestions for Authors

General comment

This systematic review addresses the role of serum chloride for predicting mortality in patients with CKD and dialysis. The manuscript is well written, and the authors appropriately acknowledge the observational nature and the scarce available evidence.

However, to my mind, several issues require substantial revision, particularly regarding the clinical interpretation of the pooled findings. My major comments are as follows.

Major comments

1) The Authors interpreted CKD, haemodialysis and peritoneal dialysis as a single physiological continuum with respect to serum chloride, and this is the main conceptual limitation. A substantial clinical and physiological heterogeneity is present among non-dialysis CKD, haemodialysis (HD), and peritoneal dialysis (PD); therefore, serum chloride may have very different determinants and biological meaning in these three settings.

In non-dialysis CKD, chloride remains influenced by residual renal handling, acid–base balance, diuretic exposure, and extracellular volume. In HD, chloride is additionally affected by an intermittent treatment characterized by large cyclic changes in water and electrolyte balance, dialysate composition, interdialytic weight gain, and residual kidney function. In PD, continuous exposure to dialysis solutions, ultrafiltration, residual kidney function, and dialysate composition, mainly about bicarbonate buffer, may produce yet another physiological context. Therefore, similar numerical associations across these populations should not necessarily be interpreted as evidence of a common chloride–mortality pathway. Conversely, the large CAPD cohort showing increased mortality with higher chloride may represent modality-specific physiology, different acid–base or volume status, dialysis prescription, or a non-linear relationship rather than merely an “exception”.

I suggest that the Results and Discussion be reorganized into three distinct sections: non-dialysis CKD, HD, and PD. The concept of a single chloride–mortality relationship across the “CKD–dialysis spectrum” should be substantially moderated.

2) All included studies are observational, and serum chloride is strongly linked to sodium, bicarbonate, volume status, diuretic use, nutritional status, and inflammation. Importantly, adjustment for these variables was inconsistent across studies; the authors themselves acknowledge that fewer studies adjusted for bicarbonate and still fewer for nutritional or inflammatory markers. Thus, the meta-analysis cannot determine whether chloride contributes causally to mortality or simply reflects underlying disease severity, acid–base disturbances, malnutrition, volume dysregulation, or dialysis-related factors.

This issue is particularly relevant in dialysis patients, in whom the renal chloride-sensing mechanisms proposed in Figure 3 cannot be straightforwardly extrapolated to patients with minimal or absent residual kidney function. The mechanistic figure should therefore distinguish non-dialysis CKD from dialysis and present the proposed pathways as hypotheses.

I would also prefer the term “multivariable-adjusted association” rather than repeatedly stating that chloride is “independently associated” with mortality.

3) The current conclusion that lower chloride is associated with a two- to three-fold higher risk of death “across most CKD and dialysis populations” appears too broad. In the Discussion, it should be clearly stated that serum chloride may be a prognostic marker rather than a causal determinant.

A more appropriate conclusion would be that serum chloride appears to carry prognostic information, but the direction and magnitude of this association may depend on kidney function, dialysis modality, acid–base status, volume status, and treatment-related factors, and causality remains unproven. Therefore, chloride biological meaning and possibly even the direction of the association appear to be context- and modality-dependent.

I also suggest that the authors provide the complete adjustment set for each cohort, preferably in Table 1, and report dialysis-specific variables whenever available (dialysate composition, timing of chloride measurement, residual kidney function, diuretic use, and volume-related parameters).

Author Response

We thank the reviewer for these thoughtful comments on clinical interpretation, which align closely with Reviewer 1's methodological concerns.

Comment 2.1 — "The Authors interpreted CKD, haemodialysis and peritoneal dialysis as a single physiological continuum... substantial clinical and physiological heterogeneity is present among non-dialysis CKD, HD, and PD... I suggest that the Results and Discussion be reorganized into three distinct sections: non-dialysis CKD, HD, and PD. The concept of a single chloride-mortality relationship across the 'CKD-dialysis spectrum' should be substantially moderated."

Response: Agreed. We have reorganised both the Results (Section 3) and the Discussion into three explicit strata — non-dialysis CKD, haemodialysis, and peritoneal dialysis — and we now discuss the distinct determinants of serum chloride in each setting (residual renal handling, acid-base balance, diuretic exposure and volume in non-dialysis CKD; intermittent fluid/electrolyte shifts, dialysate composition and interdialytic weight gain in HD; continuous dialysate exposure, ultrafiltration and buffer composition in PD). We have substantially moderated the "single continuum" framing. In particular, we now discuss the large CAPD cohort showing higher mortality with higher chloride as potentially reflecting modality-specific physiology, acid-base or volume status, or a non-linear relationship, rather than a mere "exception."

Comment 2.2 — "All included studies are observational, and serum chloride is strongly linked to sodium, bicarbonate, volume status, diuretic use, nutritional status, and inflammation... the meta-analysis cannot determine whether chloride contributes causally to mortality... The mechanistic figure should therefore distinguish non-dialysis CKD from dialysis and present the proposed pathways as hypotheses. I would also prefer the term 'multivariable-adjusted association' rather than repeatedly stating that chloride is 'independently associated' with mortality."

Response: Agreed. We now state clearly that, because adjustment for sodium, bicarbonate, volume, diuretics, nutrition and inflammation was inconsistent across studies, residual and unmeasured confounding cannot be excluded and causality cannot be inferred. We have replaced "independently associated" with "multivariable-adjusted association" (and equivalent associational wording) throughout the manuscript. Figure 3 now distinguishes non-dialysis CKD from dialysis — noting that renal chloride-sensing mechanisms cannot be straightforwardly extrapolated to patients with minimal or absent residual kidney function — and presents all proposed pathways explicitly as hypotheses.

Comment 2.3 — "The current conclusion that lower chloride is associated with a two- to three-fold higher risk of death 'across most CKD and dialysis populations' appears too broad... it should be clearly stated that serum chloride may be a prognostic marker rather than a causal determinant... I also suggest that the authors provide the complete adjustment set for each cohort, preferably in Table 1..."

Response: Agreed. The Conclusions have been rewritten to state that serum chloride appears to carry prognostic information but that the direction and magnitude of the association may depend on kidney function, dialysis modality, acid-base status, volume status and treatment-related factors, and that causality remains unproven — i.e., a prognostic marker rather than a causal determinant, with context- and modality-dependent meaning. As noted under Comment 1.7, Table 1 now provides the complete adjustment set for each cohort, and we report dialysis-specific variables (dialysate composition, timing of chloride measurement, residual kidney function, diuretic use and volume-related parameters) wherever the source studies reported them; where a study did not report such variables, this is stated.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

I appreciate the authors’ careful response to the previous review. The manuscript has improved substantially, and the major concerns raised in the first round have been appropriately addressed.

I have only two remaining points:

Please recheck the confidence-interval calculations across the pooled analyses. The primary analysis has been appropriately revised, but the four-study sensitivity analysis and the cardiovascular-mortality analysis appear to use a different confidence-interval approach. Please verify the calculations and ensure consistency with the Methods.
Please recheck the GRADE rating for the primary outcome. The manuscript states that certainty was downgraded for imprecision, indirectness, and possible publication bias, while the final rating is reported as “Low.” Please ensure that the domain-level judgments and final GRADE category are internally consistent.

Overall, the revised manuscript is considerably improved, and these issues should be readily addressed with a minor revision.

Author Response

We thank the reviewer for the careful second-round assessment and for noting that the manuscript is considerably improved. We have addressed both remaining points, and the changes are highlighted in the revised manuscript.

Comment 1 — "Please recheck the confidence-interval calculations across the pooled analyses. The primary analysis has been appropriately revised, but the four-study sensitivity analysis and the cardiovascular-mortality analysis appear to use a different confidence-interval approach. Please verify the calculations and ensure consistency with the Methods."

We agree, and we are grateful for this observation. The four-study sensitivity analysis and the cardiovascular-mortality analysis had inadvertently retained the earlier Hartung–Knapp confidence-interval approach, which was inconsistent with the standard DerSimonian–Laird random-effects (inverse-variance) model used for the corrected primary analysis. We have re-estimated both with the same model applied to the primary pool. The results are now:

Sensitivity analysis (four studies, adding the composite-outcome cohort): HR 2.38 (95% CI 1.80–3.14; τ² = 0, I² = 0%), previously reported as 2.08–2.72.
Cardiovascular mortality (two studies): HR 3.01 (95% CI 1.74–5.19; τ² = 0, I² = 0%), previously reported as 1.21–7.44.

Table 2, the Results text (Sections 3.4 and 3.6) and Figures 3 and 6 have all been updated to these values, and the Methods (Section 2.5) now state explicitly that all pooled confidence intervals were derived from the same DerSimonian–Laird random-effects model. We verified each calculation independently and confirmed internal consistency across the text, tables and figures.

Comment 2 — "Please recheck the GRADE rating for the primary outcome. The manuscript states that certainty was downgraded for imprecision, indirectness, and possible publication bias, while the final rating is reported as 'Low.' Please ensure that the domain-level judgments and final GRADE category are internally consistent."

We agree, and we have corrected this. As written, three separate downgrades from an initial rating of High would yield Very low, not Low. We have therefore made the domain-level judgments and the final category explicit and consistent (Section 3.9 and Table 2). For the primary outcome, the body of evidence began at high certainty (cohort studies) and was rated down by two levels to Low: one level for imprecision (only three studies, with a confidence and prediction interval that remain wide) and one level for indirectness (differing chloride thresholds and case-mix). Risk of bias, inconsistency (I² = 0%) and publication bias were each judged not serious enough to warrant a further full downgrade; suspected publication bias is now recorded as a concern within the Low rating rather than as an additional downgrade level. Cardiovascular mortality is rated down by three levels to Very low (imprecision, indirectness and suspected publication bias). The domain judgments and final categories are now internally consistent throughout.

We also note one editorial addition: at a co-author's request, a second affiliation has been added for Dr. Mona Alshahrani (College of Medicine, King Khalid University, Abha).

We thank the reviewer again for comments that have improved the internal consistency and rigour of the manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

I have no further comments

Author Response

We thank the reviewer for the positive second-round assessment and for indicating no further comments. We are grateful for the constructive first-round review, which substantially improved the clinical interpretation of the manuscript — in particular the reorganisation of the Results and Discussion by kidney-function stratum (non-dialysis CKD, haemodialysis and peritoneal dialysis), the shift from causal to multivariable-adjusted associational language, and the more cautious framing of serum chloride as a context- and modality-dependent prognostic marker rather than a validated risk-stratification tool.

No further changes were requested by the reviewer, and we have made none in response to this report. For completeness, we note two changes made in response to Reviewer 1 in this round: (i) the four-study sensitivity analysis and the cardiovascular-mortality analysis were re-estimated with the same DerSimonian–Laird random-effects model used for the primary pool, giving HR 2.38 (95% CI 1.80–3.14) and HR 3.01 (95% CI 1.74–5.19) respectively, with Table 2 and Figures 3 and 6 updated accordingly; and (ii) the GRADE domain judgments and final certainty categories were made internally consistent (Low for the primary outcome, Very low for cardiovascular mortality). A second affiliation for Dr. Mona Alshahrani (College of Medicine, King Khalid University, Abha) was also added at a co-author's request.

We thank the reviewer for the time and expertise devoted to our manuscript.