This study was a single-center, prospective, diagnostic gain evaluation with paired histological correlation. The primary objective was to assess the independent incremental value of the plasma KTR for the diagnosis and preoperative stratification of DIE beyond a baseline pathway comprising symptoms, clinical signs, CA-125, and imaging. The paired analysis of eutopic endometrial MMP-9 expression aimed to provide evidence of biological plausibility for KTR as an auxiliary biomarker, rather than a systematic investigation of pathogenesis.

Participants were consecutively enrolled along the clinical pathway. All testing was completed before surgery and pathological determination, and the study team maintained mutual blinding across all components. Ethical approval was obtained prior to study initiation (Approval No. KT2022062), and all participants provided written informed consent. Participants were enrolled from January 1, 2023 to December 31, 2024, and the data were finalized on March 31, 2025. Sample size was pre-calculated based on two primary objectives. Objective 1 considered the difference in the area under the receiver operating characteristic curve (AUC) for KTR added to the “baseline model” (DeLong method, two-sided $\alpha$ = 0.05, effect $\mathrm{\Delta }$AUC = 0.04, event proportion 0.35, correlation coefficient within the prediction population 0.60, power 0.80), requiring a total of 216 participants. Objective 2 assessed the standardized difference in KTR between DIE and non-DIE (Cohen’s d = 0.5, two-sided $\alpha$ = 0.05, power 0.90), requiring $\ge$85 participants per group. Considering both objectives and an anticipated 10% rate of loss to follow-up or unusable specimens, the planned total sample size was 270, with consecutive enrollment stratified into three categories, DIE, non-DIE endometriosis, and non-endometriosis controls, targeting an approximately balanced distribution to preserve the power of the primary analyses; no fixed quotas were set.

The study population consisted of women of reproductive age scheduled to undergo laparoscopy or laparotomy. Inclusion criteria: age 18–50 years; planned determination of the presence or absence of endometriosis in the current surgery; completion of symptom quantification, pelvic imaging, and blood sampling before surgery; and consent to obtain eutopic endometrial samples at surgery. Exclusion criteria: fever or active infection within the past four weeks; use of systemic glucocorticoids or immunosuppressants within two weeks before surgery; chronic renal insufficiency [estimated glomerular filtration rate (eGFR) 60 mL$\cdot$min^-1$\cdot$1.73 m^-2]; pregnancy or lactation; history of malignancy; inability to complete follow-up; or inadequate specimen handling.

This study used surgical exploration combined with histopathological examination as the gold standard. Participants were classified by lesion type into three mutually exclusive study groups: ① DIE group: suspicious lesions were found during surgery in deep structures such as the uterosacral ligaments, rectovaginal septum, vaginal fornix, bladder, and bowel wall; histopathology confirmed the presence of endometrial-like glands and/or stroma, and the lesions involved subperitoneal tissue with an infiltration depth $\ge$5 mm, consistent with the definition of DIE; ② non-DIE endometriosis group: histopathology confirmed the presence of endometriosis, but the lesions were confined to ovarian endometriomas and/or superficial peritoneal lesions, without involvement of the above deep structures or not meeting the DIE criterion of infiltration depth $\ge$5 mm; ③ non-endometriosis control group: reproductive-age women scheduled to undergo laparoscopy or laparotomy for benign non-inflammatory pelvic conditions such as uterine fibroids or simple ovarian cysts. All patients underwent standardized TVUS preoperatively, and, when necessary, pelvic MRI within 90 days before surgery to assess for suspicious signs of endometriosis. Intraoperatively, the lead surgeon systematically explored typical sites of involvement, including the pelvic peritoneum, ovaries, uterosacral ligaments, rectovaginal septum, vaginal fornix, bladder, and bowel wall. Suspicious lesions, such as ovarian cyst walls and peritoneal pigmented or fibrotic areas, were routinely excised or biopsied for pathology. Only when preoperative imaging did not suggest endometriosis, no typical or suspicious endometriotic lesions were seen intraoperatively, and all suspicious lesions were confirmed by pathology to show no endometriosis at any site, the participants were included in the non-endometriosis control group.

All pathological slides were independently reviewed by two senior pathologists blinded to the KTR results, and any discrepancies were resolved by discussion to reach consensus. The primary outcome was the presence or absence of DIE (binary), which was used for diagnostic modeling and gain analyses. At the same time, based on surgical records and imaging data, trained investigators completed the ENZIAN classification, recording the involvement levels of zones A, B, C, and F (grades 0–3), as well as the total number of involved sites, and calculating the overall score [10], to quantify infiltration burden and disease severity.

On the same day as blood sampling, symptom assessment was completed, including 10-point visual analog scale scores for dysmenorrhea, chronic pelvic pain, and dyspareunia [12]. Cyclical bowel or urinary symptoms were recorded as binary variables using a structured questionnaire. A senior physician performed a bimanual pelvic examination and recorded posterior fornix tenderness and palpable nodules. Serum CA-125 was measured using a chemiluminescent immunoassay platform, traceable to national reference materials and subjected to daily QC. The imaging protocol followed a TVUS-first strategy performed by trained sonographers, with pelvic MRI supplemented within 90 days before surgery when necessary. Whether DIE was suggested and the total number of involved sites were recorded in a unified manner. The imaging readers and the surgical team were mutually blinded.

The primary exposure variable was lnKTR. The primary outcome was pathology-confirmed DIE analyzed as a binary variable. Secondary outcomes included the ENZIAN grade and the count of involved sites. Pre-specified confounders included age, body mass index (BMI), smoking status, menstrual cycle phase or exogenous hormone use, high-sensitivity C-reactive protein, creatinine, and estimated glomerular filtration rate. Time-window alignment requirements were as follows: symptom scales and physical examination were completed on the same day as blood sampling; the interval between imaging and surgery did not exceed 90 days; the interval between blood sampling and surgery did not exceed 14 days; and the eutopic endometrial sampling was performed concurrently with surgery. Records exceeding the time windows were not included in the primary analyses.

A total of 356 candidate participants were assessed, of whom 272 enrolled. Of these, 257 provided samples suitable for pathological evaluation, including 94 DIE, 87 non-DIE endometriosis, and 76 non-endometriosis controls. The study identified three distinct analytical cohorts (Fig. 1). Analysis population A comprised 252 participants with valid KTR (92/85/75). Analysis population B included 234 cases were used to determine diagnostic gain (85/79/70). Analysis population C included 213 cases for the KTR–MMP-9 correlation (79/71/63) (Fig. 1). Kruskal–Wallis and Pearson $\chi$² tests showed no statistically significant differences among the three groups in baseline demographics, cycle/hormones, renal function, and smoking status (all p $>$ 0.05). Levels of high-sensitivity C-reactive protein (hsCRP) and CA-125 differed significantly among the three groups (both p 0.001). Similarly, the imaging results for suspected DIE and the total count of suspected involved sites showed significant differences (both p 0.001) (Table 1). Method validation and descriptive statistics, and an ICC two-way random-effects consistency model were used. The LC-MS/MS calibration range covered 0.50–200.00 µmol/L, with a lower limit of quantification (LLOQ) of 0.50 and 20.00, Bias ranged from –2.27% to –1.83%, while within-run and between-run CVs remained between 3.29–6.18%. The QC pass rate was $\ge$97.22% (Table 2A). The IHC scoring demonstrated good to excellent consistency, with overall H-score ICC = 0.892 (95% CI 0.858–0.920), and ICCs of 0.887 and 0.874 for glandular epithelium and stromal areas, respectively (Table 2B).

Fig. 1.

Flowchart of participant screening, inclusion, and exclusion. DIE, deep infiltrating endometriosis; KTR, kynurenine/tryptophan ratio; IHC, immunohistochemistry; LC-MS/MS, liquid chromatography-tandem mass spectrometry; CA-125, carbohydrate antigen 125.

In Analysis population A, the raw distributions of lnKTR overlapped significantly across the three diagnostic groups, with the median and upper quartile tending to be higher in the DIE group (Fig. 2). Using multivariable linear regression (ANCOVA) and ordinal logistic regression, after adjustment for age, BMI, smoking status, menstrual cycle phase/exogenous hormones, hsCRP and eGFR, the adjusted geometric means of KTR in the DIE and non-DIE endometriosis groups were significantly higher than in the non-endometriosis control group ($\beta$_diff = 0.28/0.15; both p_adj 0.05) (Table 3A). Moreover, each 1 standard deviation (SD) increase in lnKTR was associated with an increase in ENZIAN grade (OR_perSD = 1.62, p 0.001) (Table 3B).

Fig. 2.

Raw distribution of plasma KTR across the three diagnostic groups. lnKTR, log transformed KTR.

Logistic regression modeling was employed for Analysis population B. AUC values were compared using the DeLong method, while the net reclassification improvement (NRI), integrated discrimination improvement (IDI), and their 95% CIs were obtained by 1000 bootstrap resamples. Calibration was assessed with the Brier score accompanied by calibration plots. After adding KTR, AUC increased from 0.79 to 0.83 ($\mathrm{\Delta }$AUC = 0.04, 95% CI: 0.02–0.07, p = 0.003). At the pre-specified 20% primary threshold, NRI_category = 0.16 (p = 0.004), and these results were consistent at the 10% and 30% thresholds. The Brier score decreased from 0.185 to 0.173 ($\mathrm{\Delta }$–0.012), indicating improvements in discrimination and overall error (Table 4). Decision curve analysis was performed, with 95% CIs obtained by 1000 bootstrap resamples. Within Pt = 0.10–0.30, the “Baseline + KTR” curve lay above “Baseline” across the entire range, showing a sustained and stable magnitude increase in net benefit (Fig. 3A). A logistic calibration model with locally estimated scatterplot smoothing (LOESS) smoothing was used to display predicted–observed agreement. The calibration intercept and slope of the baseline model were $\alpha$ = –0.127 and $\beta$ = 0.861, indicating slight underestimation; after adding KTR, $\alpha$ approached 0 (–0.039) and $\beta$ approached 1 (0.946), with calibration markedly improved (Fig. 3B). A risk-band reclassification matrix, a signed-rank test, and the Wilson method were applied (threshold 20%). In the imaging-suspicious but atypical subgroup, after adding KTR the net correct reclassification was +14 for events and +17 for non-events (Table 5A). Pre- versus post-test probability changes showed an increase for events and a decrease for non-events, both p 0.001 (Table 5B). Based on the 20% threshold, the likelihood ratios were positive likelihood ratio (LR+) 3.14 (95% CI 2.07–4.76) and LR– 0.30 (95% CI: 0.17–0.54) (Table 5C).

Fig. 3.

Clinical benefit and calibration assessment of the models. (A) Decision curve analysis: net benefit after adding KTR. (B) Calibration plot: predicted vs. observed (Baseline vs. Baseline + KTR).

Analysis population C (n = 213) underwent multiple linear regression. MMP-9 H-score was defined as dependent variable, standardized the lnKTR as z-score, log transformed the hsCRP levels, and applied a nested F test to evaluate incremental explained variance. lnKTR was positively associated with MMP-9 in eutopic endometrium ($\beta$_std = 0.29, 95% CI: 0.15–0.43, p 0.001). Among covariates, only ENZIAN total score ($\beta$_std = 0.18), hsCRP (ln) ($\beta$_std = 0.13), and late secretory relative to follicular phase ($\beta$_std = 0.11) were significant (all p 0.05). Model yielded R² = 0.32, after adding lnKTR, $\mathrm{\Delta }$R² = 0.04 (F = 11.882, p 0.001) (Table 6). The scatter plot, colored by menstrual cycle phase, and the partial regression plot was controlled for all covariates. The overall partial regression line was consistent with the previously described direction, and the point clouds largely overlapped across all phases (Fig. 4).

Fig. 4.

Scatter plot of KTR (log) versus MMP-9 H-score and partial regression line.

ANCOVA, the DeLong method with NRI_category (20% primary threshold), and multiple linear regression consistent with the main analyses were used. In stratifications of “early follicular only”, excluding hsCRP $>$10 mg$\cdot$L⁻¹/eGFR 60 mL$\cdot$min⁻¹$\cdot$1.73 m⁻², and by imaging modality (TVUS only versus MRI only), $\beta$_diff in Chain A, $\mathrm{\Delta }$AUC and NRI in Chain B, and $\beta$_std in Chain C were all positive and significant (all p 0.05). The directions and magnitudes of effects were consistent with the main analyses, indicating robustness (Table 7).

Little’s MCAR test and MICE multiple imputation (m = 10) was assessed. Within Analysis population B, the missing data rates for each covariate remained 5%, with the exception of hsCRP, and the missingness pattern was consistent with the MCAR assumption ($\chi$² = 18.742, p = 0.282) (Table 8). Using logistic regression and the DeLong test, the odds ratio (OR) for DIE per 1 SD increase in lnKTR was similar between complete case and imputed analyses (1.57 vs. 1.62). Moreover, the AUC increased from 0.79 to 0.83, and the directions of improvement in the Brier score and calibration were consistent, indicating that multiple imputation did not alter the main conclusions (Table 9).

To avoid overstating the conclusions, several limitations should be acknowledged. The single-center prospective cohort may be affected by referral and spectrum bias, lacks external and temporal validation, and the stability of risk thresholds has not yet been prospectively tested. NRI, IDI, and decision curves are sensitive to sample size and threshold selection, and their CIs carry inherent uncertainty. KTR is influenced by concomitant inflammation, diet, and metabolic status. Although adjustments were made for hsCRP and renal function, residual confounding may remain. MMP-9 was semi-quantified by IHC, and despite good ICC, batch effects and reader drift may occur. The maximum 90-day interval from imaging to surgery may lead to disease progression. Future work should conduct multicenter, pre-registered external and temporal validation; establish cross-platform standardization of LC-MS/MS and IHC with reference materials; prospectively confirm thresholds according to clinical use scenarios; evaluate decision impact and cost-effectiveness in imaging-atypical and preoperative stratification pathways; combine plasma markers, symptoms, and structural imaging to develop deployable multimodal models; refine preanalytical standard operating procedures (SOPs) and inter-laboratory consistency; and, within the cohort, track longitudinal changes and modifiability of the kynurenine–AhR–MMP-9 pathway to clarify causal associations with clinical outcomes.