Dynamic Breath VOC Profiling as a Non-invasive Tool to Assess Hemodialysis Adequacy

Rationale and Background Adequacy of hemodialysis (HD) is commonly summarized by small-solute clearance metrics such as Kt/V. However, patient-centered outcomes do not always improve with higher urea removal alone. People receiving HD often experience chronic low-grade inflammation, immune dysfunction, and broad metabolic disturbances that are not captured by urea kinetics. Breath contains trace volatile organic compounds (VOCs) produced by endogenous and exogenous metabolic processes. Prior studies suggest that chronic kidney disease (CKD) and end-stage kidney disease (ESKD) alter breath chemistry, including compounds such as ammonia, alkanes, and aldehydes. Dynamic (pre- vs. post-dialysis) VOC profiling may therefore provide a rapid, non-invasive window into immediate treatment effects and the broader uremic milieu. This study expands on prior work by (1) collecting paired breath samples before and after a routine HD session, (2) pairing breath analyses with comprehensive blood biochemistry and immune measurements, and (3) integrating public transcriptomic resources to position observed immune readouts within broader immune signaling contexts. The overarching goal is to evaluate whether breath VOC signatures-alone or combined with blood indices-can serve as practical indicators of dialysis adequacy for clinical and research use. Study Objectives Primary objective: to characterize dynamic changes in exhaled VOCs across a single HD session and evaluate whether these changes track with conventional adequacy metrics and clinically relevant blood indices. Secondary objectives include: (a) describing differences in breath VOC profiles between individuals on maintenance HD and healthy adults; (b) identifying blood indices that change over the course of a single HD session and that deviate from typical reference ranges; and (c) exploring links between breath-derived features and immune/metabolic signatures using publicly available transcriptomic datasets. Study Design Overview This is a cross-sectional observational study conducted at Shanghai Second People's Hospital (Department of Nephrology, Dialysis Unit). Participants include adults on maintenance HD and age/sex-matched healthy adults. For the HD group, breath and blood samples are collected immediately before and after a routine treatment; healthy participants provide a single breath and blood sample at one visit. No investigational drug or device is administered, and clinical care is not altered. Setting and Timeline Single-center study at Shanghai Second People's Hospital. Enrollment is planned from September 1, 2025 through December 1, 2025. Sampling and laboratory analyses are expected to occur between December 2025 and December 2026, followed by data integration and dissemination through June 30, 2027. Sampling Procedures Breath collection: Alveolar breath is captured using validated, inert collection bags or devices per institutional standard operating procedures (SOPs). For HD participants, two samples are obtained: one collected prior to the start of a routine HD session and one collected promptly after completion of the same session. Ambient/blank samples are collected alongside patient samples to monitor environmental background. Samples are transported under conditions that minimize VOC loss and processed promptly according to SOPs. Blood collection: Peripheral venous blood is collected at the same time points for HD participants (pre- and post-session) and once for healthy adults. Assays include routine biochemistry and immune-related measurements as specified in the protocol. All samples are labeled with coded identifiers only. Analytical Methods VOC analysis: Gas chromatography-mass spectrometry (GC-MS) is used for relative identification and quantification of VOCs. Analytical runs include internal standards and quality controls to monitor retention time stability, mass accuracy, and signal drift. Chromatographic deconvolution, peak alignment, and compound annotation follow laboratory SOPs; putative identifications are assigned using spectral libraries where applicable. Where feasible, candidate features are confirmed using authentic standards. Blood assays: Approximately 97 parameters encompassing general biochemistry, inflammation, and immune indices are measured using clinically validated methods. Laboratory quality control procedures and instrument maintenance schedules follow hospital standards. Outcomes and Measures (Informational, non-duplicative) To avoid duplication with the Outcome Measures module, only a high-level summary is provided here. The study evaluates (a) within-person pre/post changes in candidate VOC features among HD participants and (b) between-group differences (HD vs. healthy) in breath VOC profiles. Select blood indices are analyzed for pre/post changes and contextualized relative to customary reference intervals. Please refer to the registered Outcome Measures for prespecified endpoints and definitions. Sample Size Rationale The target sample includes approximately 22 individuals on HD and approximately 23 healthy adults. A two-proportion comparison for a representative binary VOC feature (positive detection rate) informed the sample size: assuming p0 ≈ 0.40 in healthy adults and p1 ≈ 0.80 pre-dialysis, with α = 0.05 (two-sided) and 80% power, the estimated per-group sample size is about 20, allowing for modest attrition to reach the planned numbers. Blood endpoints are considered exploratory and were not used to power the study. Data Management and Confidentiality Data capture uses double data entry for laboratory results, automated range and logic checks, and routine discrepancy management. Coded identifiers are used on all samples and case report forms; the re-identification key is stored separately with restricted access. Electronic data reside on secure, access-controlled servers with regular backups and audit trails. A data lock procedure is implemented prior to final analyses. Quality Control and Quality Assurance (QC/QA) Standardized SOPs govern participant preparation, breath and blood collection, handling, storage, instrument operation, and data processing. GC-MS runs include blanks, internal standards, and pooled quality-control samples to monitor batch effects and analytical precision. Prespecified acceptance criteria (for example, retention time windows, mass accuracy, and relative standard deviation across QC replicates) are used to flag features for review or exclusion. Monthly data integrity reviews address completeness and consistency. Ethical Considerations The study follows the Declaration of Helsinki and applicable national regulations. Participants provide written informed consent before any study-specific procedures. Because the study is non-interventional and aligns with routine care for HD participants, risks are minimal and limited primarily to venipuncture discomfort and incidental findings from laboratory testing. Risk-Benefit and Participant Burden Breath sampling is non-invasive, and blood draws entail brief, minor discomforts for most participants. There may be no direct clinical benefit to individual participants. Potential societal benefit includes progress toward simple, non-invasive markers of dialysis adequacy that could inform clinical decision-making. Preprocessing and Data Cleaning (VOC and Blood) VOC feature tables undergo background subtraction using paired ambient samples, retention-time alignment, and signal normalization (for example, internal standard-based normalization and/or total ion current normalization). Intensities are log-transformed as appropriate and scaled. Features with poor reproducibility in QC samples or excessive missingness are removed. For values below detection limits, imputation may use a small constant (e.g., half the minimum non-zero value) in sensitivity analyses. Blood indices are reviewed for physiologic plausibility and analytical flags. Outliers are examined using robust methods; corrections or exclusions follow predefined rules documented in the data management plan. Handling Missing Data If missingness is ≤5% for a variable, complete-case analyses may be used with sensitivity checks. For higher levels of missingness assumed to be missing at random, multiple imputation by chained equations (MICE) may be applied for adjusted analyses. Outcomes that are fundamentally missing-not-at-random (e.g., due to technical failure) are addressed through sensitivity analyses, including best/worst-case bounds where relevant. Statistical Analysis Plan Analyses proceed according to a prespecified plan to limit multiplicity and analytic flexibility. Continuous variables are summarized as mean±SD or median (IQR); categorical variables as counts and percentages. Within-person pre/post comparisons in the HD group use paired t-tests or Wilcoxon signed-rank tests, as suitable. Between-group comparisons (HD vs. healthy) use t-tests/ANOVA or Mann-Whitney U/Kruskal-Wallis tests. False discovery rate (FDR) control via Benjamini-Hochberg is used for multiple VOC features and blood indices. For multivariable modeling, linear or generalized linear models evaluate associations between selected VOC features and clinical covariates (for example, age, sex, dialysis vintage, interdialytic weight change, comorbidities), as documented in the statistical analysis plan (SAP). Where appropriate, robust regression or transformation is used to mitigate non-normality. Associations between VOCs and blood indices are evaluated with Spearman correlation and partial correlation adjusting for key covariates. Machine learning: Candidate classifiers (e.g., support vector machines, random forests, and penalized logistic regression) are trained to discriminate pre- vs. post-dialysis states and HD vs. healthy. To avoid information leakage, cross-validation folds are created at the participant level. Model selection uses nested or repeated stratified 10-fold cross-validation with tuning on the inner loop and performance estimation on the outer loop. Primary metrics include area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and calibration (e.g., Platt scaling). Model interpretability is assessed using SHAP values to quantify feature contributions. Performance estimates are accompanied by 95% confidence intervals via bootstrapping. Sensitivity analyses include: (a) re-running models with alternative normalization or imputation strategies; (b) excluding features with potential environmental sources; (c) stratifying analyses by dialysis adequacy categories defined in the protocol according to accepted clinical practice guidelines; and (d) repeating comparisons using non-parametric approaches. Exploratory subgroup analyses compare CKD stage 5 patients categorized by adequacy and other clinical factors. Integration with External Transcriptomic Resources (Exploratory) To contextualize immune findings, public transcriptomic datasets relevant to ESKD/HD may be queried (e.g., peripheral blood or PBMC datasets). Gene set variation analysis or single-sample enrichment methods summarize pathway and cell-type signatures, which are then related to measured blood indices and breath VOC features using correlation and regression. These exploratory analyses are intended to generate mechanistic hypotheses, not confirm causality. Bias, Confounding, and Robustness Potential biases include environmental VOC contamination, dietary effects, smoking status, and comorbidities. Controls include ambient blanks, adherence to pre-collection instructions where applicable, and adjustment for key covariates. Analyses are stratified or adjusted for factors such as age and sex, with sensitivity analyses excluding participants with major protocol deviations. Data Sharing and Dissemination Findings will be summarized in peer-reviewed publications and scientific meetings. De-identified data and analysis code may be shared upon reasonable request and institutional approvals, subject to privacy, regulatory, and ethical constraints. Limitations As a single-center observational study with a modest sample size, the work is best viewed as hypothesis-generating. VOC identifications are relative unless confirmed with authentic standards, and environmental/background contributions cannot be fully eliminated. Future studies with larger, multi-center cohorts and standardized breath collection across sites will be needed to validate candidate markers. Conclusion This study provides a structured, non-invasive assessment of dynamic breath chemistry alongside blood-based indices in people receiving hemodialysis. By characterizing pre/post-dialysis changes and integrating immune context, it aims to identify practical signals of dialysis adequacy and to advance understanding of the systemic biology of uremia.