We conducted a multicentre, prospective cohort across three tertiary hospitals in Shanxi Province: Shanxi Provincial Cancer Hospital, Shanxi Provincial Bethune Hospital, and Xinzhou People’s Hospital. Consecutive adults with histologically confirmed oesophageal carcinoma scheduled for curative-intent transthoracic (Ivor-Lewis or McKeown) or minimally invasive oesophagectomy were screened via electronic theatre rosters. Eligibility required: (1) age ≥ 18 years; (2) resectable squamous-cell carcinoma or adenocarcinoma; (3) capacity to complete a Mandarin food-frequency questionnaire unaided; and (4) expected in-hospital follow-up through postoperative day (POD) 7 with planned outpatient review at week 12. We excluded patients receiving parenteral nutrition at baseline, those with inflammatory bowel disease, or with systemic antibiotic or corticosteroid exposure within 2 weeks prior to enrolment. This study is based on a prospective clinical registry database established between January 2019 and June 2024 at Shanxi Cancer Hospital and Shanxi Bethune Hospital. This database was established with the approval of the hospital and is used for research analysis. Following reviewer comments, Patient data meeting the study criteria from a prospective cohort study with identical indicators, conducted at Xinzhou People’s Hospital between 1 June 2023 and 1 July 2025, has been incorporated into the analysis of this research following approval by the Ethics Committee and the signing of informed consent forms by the patients themselves, incorporating Xinzhou People’s Hospital. Ethical approval was obtained on 4 July 2025 (Approval No.: 2025-LLSC-07-04). This centre formally commenced patient enrolment in July 2025. A portion of cases have completed follow-up and been included in the present analysis, whilst the remaining cases remain under follow-up and will be reported in subsequent studies.

All centres followed harmonised case-report forms, identical screening logs, and common operating definitions. Written informed consent was obtained from all participants; the study adhered to the Declaration of Helsinki principles. Peri-operative care followed local enhanced-recovery pathways (standardised carbohydrate loading, early mobilisation and stepwise diet advancement). To minimise centre-level practice drift, the coordinating team issued procedure manuals, hosted start-up training, and performed periodic monitoring of screening/eligibility logs. This study has been approved by the Research Ethics Committee of Xinzhou People’s Hospital (2025-LLSC-07-04). All participants signed written informed consent forms in Chinese; data underwent anonymisation prior to analysis.

Pre-operative habitual diet was assessed within 2 weeks prior to surgery using a validated 147-item semi-quantitative food-frequency questionnaire (FFQ) capturing frequency and portion size over the preceding 3 months. Nutrient intakes were computed from the Chinese Food Composition Tables. Total dietary fibre (g day⁻¹) was energy-adjusted by the residual method and categorised into sex-specific quartiles; in sensitivity analyses, fibre was dichotomised at the Chinese Dietary Reference Intake threshold of 30 g day⁻¹. To gauge temporal stability of habitual intake, participants were invited to complete a 3-day food record at week 12; where both instruments were available, we calculated intra-class correlations for fibre. None of the centres prescribed explicit fibre targets in routine dietetic counselling. Probiotic use was not protocolised and, when present, was recorded (brand/strain, timing and duration) for adjustment.

At three prespecified time-points—baseline (pre-operative, T₀), POD 7 (T₁) and week 12 (T₂)—we recorded vital signs and anthropometry (height, weight, body-mass index [BMI], mid-upper-arm and calf circumferences) and obtained routine venous bloods processed by hospital laboratories accredited to national standards. The panel comprised: complete blood count with differential (to derive absolute neutrophil and lymphocyte counts), C-reactive protein (CRP), serum albumin, fasting glucose, and basic metabolic/liver profiles. Hand-grip strength (dominant hand; best of three trials) was measured where dynamometers were available and treated as an exploratory functional metric. Post-operative infectious complications (pulmonary, urinary, wound/mediastinal) and 30-day mortality were abstracted from charts by trained staff masked to FFQ data. Medication charts were reviewed to derive antibiotic exposure (calendar “antibiotic-days” from incision to POD 7), intra-operative dexamethasone, and 0–48 h opioid dose (morphine-equivalent).

We prespecified two complementary week-12 endpoints that reflect systemic inflammatory tone and immune–nutritional recovery:

Intermediate (POD 7) values were analysed as trajectory markers. Exploratory continuous outcomes included week-12 Prognostic Nutritional Index (PNI = 10 × albumin [g dL⁻¹] + 0.005 × lymphocytes [mm⁻³]) and the Systemic Immune-Inflammation Index (SII = platelets × neutrophils / lymphocytes).

A priori covariates were selected on biological grounds and operationalised uniformly across sites: age, sex, BMI, smoking status (never/former/current), alcohol intake (standard drinks per week), Charlson comorbidity index, tumour histology and pathologic stage (AJCC 8th), neoadjuvant chemoradiotherapy, operative approach, anaesthesia duration, intra-operative dexamethasone, 0–48 h opioid dose, baseline CRP, and antibiotic-days (POD0–7). To address potential reverse causation (pre-existing inflammatory state shaping diet), baseline white-cell count and baseline NLR were recorded for sensitivity models. Missing data were handled with chained-equations multiple imputation (m = 15) under a missing-at-random assumption.

Continuous variables are summarised as mean ± SD or median (inter-quartile range), categorical variables as counts (percentages). Baseline characteristics across fibre quartiles were compared using one-way ANOVA, Kruskal–Wallis, or χ² tests as appropriate. Primary analyses estimated relative risks (RR) for each endpoint across fibre quartiles using multivariable Poisson regression with robust variance. We fitted sequential models: Model 1 adjusted for age and sex; Model 2 added tumour histology, stage and Charlson index; Model 3 further included operative approach, anaesthesia duration, intra-operative dexamethasone, 0–48 h opioid dose, antibiotic-days, and probiotic use. Hospital site was accounted for using cluster-robust standard errors and, in sensitivity analyses, as fixed effects. For dose–response assessment, energy-adjusted fibre (g day⁻¹) was modelled continuously with restricted cubic splines (knots at the 5ᵗʰ, 50ᵗʰ and 95ᵗʰ percentiles), with p-values for trend derived from the linear term.

Repeated-measure trajectories (baseline → POD 7 → week 12) for CRP, NLR, lymphocyte count, PNI and SII were evaluated with linear mixed-effects models specifying random intercepts for participants and an autoregressive correlation structure; site was included as a random intercept in sensitivity analyses. Effect-modification was explored a priori for neoadjuvant therapy (yes/no), BMI (< 24 vs. ≥ 24 kg m⁻²), and operative approach, using cross-product terms with interaction p < 0.10 taken as evidence of heterogeneity. For secondary/exploratory outcomes and interaction tests we controlled the false-discovery rate at 5% (Benjamini–Hochberg); primary endpoints were prespecified and not multiplicity-adjusted. Two-sided α = 0.05. Analyses were performed in R 4.3.3.

We enrolled 312 adults with histologically confirmed oesophageal carcinoma across three tertiary hospitals. Mean age was 63.0 ± 8.0 y; 76% were men; mean BMI was 23.9 ± 3.2 kg m⁻². Histology was predominantly squamous-cell carcinoma (86%), and pathologic stage distribution was I/II/III: 29%/46%/25%; 41% received neoadjuvant chemoradiotherapy. Baseline characteristics across sex-specific dietary-fibre quartiles were broadly balanced with small differences in smoking and Charlson index (Table 1). Analytic choices and covariates followed the prespecified plan.

Energy-adjusted fibre (g day⁻¹, residual-method) showed the expected separation by sex-specific quartiles: median (IQR) Q1 15.22 (12.48–16.83), Q2 20.95 (19.87–22.17), Q3 26.12 (24.86–27.30), Q4 31.76 (30.33–34.79). At week 12, 82% completed a 3-day food record; the FFQ–record intra-class correlation for fibre was 0.87, indicating good temporal stability. The proportion meeting ≥ 30 g day⁻¹ was 0, 0, 0, and 79.5% across Q1–Q4 (Table 2).

Event rates improved monotonically across fibre quartiles. The favorable inflammatory profile occurred in 21/78 (26.9%), 36/78 (46.2%), 52/78 (66.7%), and 60/78 (76.9%) from Q1 to Q4; lymphocyte recovery occurred in 33/78 (42.3%), 40/78 (51.3%), 56/78 (71.8%), and 62/78 (79.5%) (Table 3). In multivariable Poisson models with robust, cluster-by-site variance (Model 3: age, sex, histology, stage, Charlson, smoking, alcohol, BMI, operative approach, anesthesia duration, intra-operative dexamethasone, 0–48 h opioid dose, antibiotic-days, probiotics, neoadjuvant therapy), adjusted RRs (vs Q1) for the favorable profile were:

For lymphocyte recovery (Model 3), RRs (vs Q1) were:

Results were directionally consistent in more parsimonious Model 1 (age, sex) and Model 2 (adds histology, stage, Charlson) (Table 3).

Modelled continuously, each 10 g day⁻¹ higher pre-operative fibre was associated with a 56% higher probability of the favorable inflammatory profile (RR per10 = 1.56; 95% CI 1.34–1.82; p trend = 6.9 × 10⁻⁹) and a 30% higher probability of lymphocyte recovery (RR per10 = 1.30; 1.17–1.45; p trend = 2.1 × 10⁻⁶) (Table 4). A quadratic term to probe non-linearity was not significant for lymphocyte recovery (p non-lin = 0.656) and was borderline for the inflammatory profile (p non-lin = 0.091). The spline-shaped adjusted risk curve rose steadily across the observed exposure window (Figure 1).

Group means (Table 5) demonstrated parallel early post-operative perturbations with subsequent recovery, accentuated in higher-fibre quartiles. From baseline to week 12, mean CRP (mg L⁻¹) decreased from 16.98 → 9.79 in Q1 (Δ − 7.19) and from 18.01 → 7.62 in Q4 (Δ − 10.39). Mean NLR fell from 3.45 → 2.20 (Δ − 1.25) in Q1 versus 3.63 → 1.65 (Δ − 1.98) in Q4. Mean lymphocyte count (×10⁹ L⁻¹) rose from 1.11 → 1.31 (+0.20) in Q1 and 1.10 → 1.41 (+0.31) in Q4. In mixed-effects models (random intercept for participant), time × quartile interactions at week 12 were significant for CRP (omnibus p = 1.68 × 10⁻⁴) and NLR (p = 1.21 × 10⁻⁴) and strongly significant for lymphocyte count (p = 2.26 × 10⁻¹⁰), indicating steeper recovery with higher fibre (Table 5; Figure 2).

Adjusted linear models showed higher PNI and lower SII with greater fibre. For Q4 vs. Q1, the adjusted mean difference in PNI was +2.16 (95% CI − 0.14 to +4.46; p = 0.066; q FDR = 0.105) and in SII was −149.6 (−205.4 to −93.9; p = 1.44 × 10⁻⁷; qFDR = 5.74 × 10⁻⁷) (Table 6).

Pre-specified modifiers did not materially alter the fibre–outcome relationship (interaction p < 0.10 considered heterogeneous). For the favorable profile (RR per10, Model 2): neoadjuvant therapy—No: 1.55 (1.34–1.79) vs. Yes: 1.55 (1.41–1.70); p interaction = 0.445. BMI— < 24 kg m⁻²: 1.45 (1.14–1.85) vs. ≥ 24 kg m⁻²: 1.69 (1.33–2.14); p interaction = 0.325. Operative approach—Open Ivor-Lewis: 1.76 (1.26–2.47), minimally invasive: 1.55 (1.10–2.18), McKeown: 1.42 (1.29–1.58); p interaction = 0.185 (Table 7). No modifier crossed the heterogeneity threshold, and Benjamini–Hochberg–adjusted q-values remained > 0.20.

Results were robust across pre-specified checks (Table 7). Dichotomizing fibre at 30 g day⁻¹ (≥ 30 vs. < 30) yielded adjusted RRs of 1.46 (1.15–1.85; p = 0.0019) for the favorable profile and 1.45 (1.33–1.57; p < 10⁻¹⁸) for lymphocyte recovery. Adding baseline CRP and baseline NLR to Model 3 minimally changed quartile effects (vs Q1: Q2 = 1.72 [1.01–2.93], Q3 = 2.76 [1.97–3.86], Q4 = 2.79 [2.44–3.20]). Treating hospital site as a fixed effect (instead of clustering) produced near-identical estimates (Q4 vs. Q1 RR = 2.57; 2.13–3.10).

Pulmonary/urinary/wound-mediastinal infections through POD 7 occurred in 21/78 (26.9%), 19/78 (24.4%), 11/78 (14.1%), and 14/78 (17.9%) across Q1–Q4 (Table 8). In fully adjusted models including operative and antimicrobial covariates, the per-10 g effect was RR = 0.84 (95% CI 0.59–1.20; p = 0.333). Thirty-day mortality was infrequent overall (3/312; 0.96%) without evidence of quartile-wise differences (Q1: 0/78; Q2: 2/78; Q3: 0/78; Q4: 1/78; Table 8).