In the present retrospective study, 98 patients with cervical cancer who underwent radiotherapy at Fuyang People's Hospital (Fuyang, China) between 1 January, 2023, and 31 December, 2025, were initially screened. A total of 6 patients were excluded as they did not complete the prescribed course of radiotherapy. The final cohort comprised 92 female patients, with a median age of 59 years (range, 32–84 years). The requirement for informed consent was waived due to the retrospective nature of the study. This study was approved by the Ethics Committee of Fuyang People's Hospital (approval no. 2026-22).

The inclusion criteria were as follows: i) Pathologically confirmed cervical cancer; ii) a diagnosis of International Federation of Gynecology and Obstetrics (FIGO) stage IB-IV disease (17), with planned definitive radiotherapy or postoperative adjuvant radiotherapy; and iii) a Karnofsky performance status (18) score of ≥70.

The exclusion criteria were as follows: i) Incomplete clinical, laboratory or radiotherapy planning data; and ii) failure to complete the prescribed course of radiotherapy for any reason.

All patients were instructed to empty their bladder and bowel in the morning and to drink 800 ml water before simulation. CT simulation was performed after adequate bladder filling, indicated by a clear sensation of urinary urgency. The procedure was conducted by two experienced radiotherapists. Patients were placed in the supine position and underwent CT simulation using a large-bore CT simulator (Somatom go.Sim; Siemens AG). The scanning range extended from the superior border of the 11th thoracic vertebra to 5 cm below the ischial tuberosities, with a slice thickness of 3 mm.

All CT images were transferred to the Monaco treatment planning system (version 5.11; Elekta Instrument AB). Target volumes and organs at risk (OARs) were delineated according to the recommendations of the Radiation Therapy Oncology Group (RTOG) (19) and relevant dose constraints for OARs. Delineation was performed with reference to magnetic resonance imaging and contrast-enhanced CT findings, including the primary cervical lesion, local tumour extension and metastatic lymph nodes. Target delineation was performed by an attending radiation oncologist and jointly reviewed by a senior physician with the title of associate chief physician or above.

The planning target volume (PTV) was defined as the clinical target volume with a 3-cm extension in the superior and inferior directions. Treatment plans were generated using the Monaco treatment planning system. A seven-field intensity-modulated radiotherapy (IMRT) plan was designed using 6-MV photon beams. At least 95% of the PTV was required to receive the prescribed dose, and dose hotspots outside the PTV were not allowed to exceed 110% of the prescription dose.

The OARs included the rectum, bladder, spinal cord, bilateral femoral heads, colon and small intestine. The bowel bag was defined as the intestinal region encompassing the entire small and large bowel, excluding the rectum and anus. Dose-volume histogram (DVH) (20) parameters were used to calculate the volumes of the bowel bag and rectum receiving doses of ≥40, ≥45 and ≥50 Gy, denoted as V40, V45 and V50, respectively. These DVH parameters were recorded as relative volumes (%).

All patients received external-beam radiotherapy with 6-MV X-rays. The total prescribed dose to the target volume was 45–54.4 Gy, delivered to 95% of the PTV in 25 fractions, once daily, at five fractions per week, with a fractional dose of 1.8–2.0 Gy. Positive metastatic lymph nodes could receive a simultaneous integrated boost of up to 59.94 Gy. For patients with para-aortic or common iliac lymph node metastases, extended-field irradiation was administered when clinically indicated. The use of concurrent chemotherapy was determined according to clinical stage, risk stratification and overall patient condition.

An ARE diagnosis was based on clinical manifestations, endoscopic findings and imaging features, after exclusion of infectious and other non-radiation-related intestinal disorders. ARE was graded according to the RTOG acute radiation morbidity scoring criteria (19).

ARE was evaluated according to the RTOG acute lower gastrointestinal toxicity criteria (19) as follows: Grade 0, no obvious intestinal symptoms; grade 1, mild changes in bowel habits, increased stool frequency or rectal discomfort without medication; grade 2, diarrhoea requiring medication, mucus discharge, or abdominal or rectal pain requiring analgesics; grade 3, severe diarrhoea requiring parenteral support, marked mucus or bloody discharge, or abdominal distension with radiographic evidence of bowel dilatation; grade 4, severe complications, such as obstruction, fistula, perforation, transfusion-requiring bleeding, or symptoms requiring decompression or diversion; and grade 5, death related to intestinal toxicity.

The primary endpoint was grade ≥1 acute lower gastrointestinal toxicity during radiotherapy. As grade 1 toxicity may be mild and of limited clinical significance, the distribution of RTOG grades was summarized. Sensitivity analyses using grade ≥2 and grade ≥3 endpoints were performed to evaluate more clinically meaningful toxicity.

The mPNI was calculated using the following formula: mPNI=4.8 × total cholesterol (mmol/l)-1.5 × albumin (g/l)-7.7 × total lymphocyte count (×109/l) + 126 (21).

The directionality of this score differs from that of conventional PNI, which is usually calculated from serum albumin and lymphocyte count with positive coefficients. In the formula used in the present study, lower albumin and lower lymphocyte count increase mPNI, but higher cholesterol also increases mPNI. Therefore, a higher mPNI should be interpreted as a less favourable cholesterol-modified nutrition-immune-metabolic profile, rather than better nutritional status.

Conventional PNI was calculated for comparison as albumin (g/l) + 5 × total lymphocyte count (×10⁹/l).

All statistical analyses were performed using SPSS software (version 26.0; IBM, Corp.). Continuous variables were first assessed for normality. Normally distributed continuous variables are presented as the mean ± standard deviation, and comparisons between two groups were performed using the independent-samples t-test. Non-normally distributed continuous variables are expressed as the median (interquartile range) and were compared using the Mann-Whitney U test. Categorical variables are presented as the frequency (percentage), and between-group comparisons were performed using a χ² test or Fisher's exact test, as appropriate.

Multivariable logistic regression analysis was used to identify independent risk factors for ARE in patients with cervical cancer receiving radiotherapy. A limited multivariable model that included FIGO stage and mPNI was first constructed to reassess the original model. A clinically adjusted model was then established to account for clinically relevant confounding factors, including age, FIGO stage, surgery, concurrent chemotherapy, brachytherapy, prescription dose >50 Gy, boost treatment, bowel bag V40 (%) and rectum V40 (%). Due to the limited sample size and potential collinearity among DVH parameters, bowel bag V40 (%) and rectum V40 (%) were selected as representative dosimetric covariates. Sensitivity analyses were performed using grade ≥2 and grade ≥3 ARE as endpoints. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated.

Receiver operating characteristic (ROC) curve analysis was used to evaluate the discriminatory performance of mPNI and associated indicators. The optimal cut-off value was determined using the Youden index. Bootstrap resampling was used to estimate AUC stability and compare paired AUCs. Decision-curve analysis was performed to assess potential clinical usefulness. Two-sided P<0.05 was considered to indicate a statistically significant difference.

Among the 92 eligible patients, 25 did not develop ARE and 67 developed grade ≥1 ARE, corresponding to an incidence rate of 72.8%. The RTOG grade distribution was as follows: grade 0, 25 patients (27.2%); grade 1, 21 patients (22.8%); grade 2, 28 patients (30.4%); grade 3, 18 patients (19.6%); and grade 4/5, 0 patients (0.0%) (Table SI). The patient selection process is shown in Fig. 1.

The FIGO stages ranged from I–IV, with 62 patients having stage I–II disease and 30 having stage III–IV disease. Stage I–II disease was present in 19 patients in the non-ARE group and 43 patients in the ARE group, whereas stage III–IV disease was present in 6 and 24 patients, respectively. This distribution was not statistically significant (Pearson χ²=1.158; P=0.282). Hypertension and diabetes mellitus were present in 36 and 24 patients, respectively. In total, 63 patients underwent radical surgery for cervical cancer, 33 received brachytherapy after external-beam radiotherapy and 10 received boost treatment, with boost doses ranging from 50 to 59.94 Gy. Concurrent chemotherapy was administered to 55 patients. Before treatment, anaemia was observed in 62 patients. In total, 20 patients received >50 Gy, whereas 72 received ≤50 Gy (Table I).

The mean pre-radiotherapy mPNI was significantly higher in the ARE group than in the non-ARE group (89.72±11.14 vs. 76.45±8.59; t=−5.381; P<0.001). Conventional PNI was significantly lower in the ARE group than in the non-ARE group [43.60 (39.03–48.33) vs. 47.85 (45.05–51.30); Z=−2.83; P=0.005]. No significant differences were observed in age, BMI, hypertension, diabetes, FIGO stage, surgery, parametrial boost, concurrent chemotherapy, brachytherapy, prescription dose, anaemia or pathological type between the two groups (all P>0.05) Table I.

Comparison of radiotherapy dosimetric parameters showed no significant differences in bowel bag V40, V45 or V50, or in rectal V40, V45 or V50 between the ARE and non-ARE groups (P>0.05). These DVH parameters are expressed as relative volumes (%). These findings suggest that, in this cohort, the analysed bowel bag and rectal DVH parameters were not significantly associated with the occurrence of ARE during radiotherapy in patients with cervical cancer. Detailed results are presented in Table II.

Representative external-beam radiotherapy plans from the ARE and non-ARE groups are shown in Fig. 2. In both representative plans, adequate dose coverage of the PTV was achieved and the dose distribution to OARs, including the rectum, small intestine and bladder, met the clinical planning requirements.

In the limited multivariable logistic model that included FIGO stage and pre-radiotherapy mPNI, pre-radiotherapy mPNI was significantly associated with grade ≥1 ARE [odds ratio (OR), 1.154; 95% confidence interval (CI), 1.075–1.239; P<0.001], whereas FIGO stage was not independently associated with ARE (OR, 1.477; 95% CI, 0.426–5.126; P=0.539). To address potential confounding by clinically relevant factors, a clinically adjusted model was constructed. Given the limited sample size and potential collinearity among DVH parameters, bowel bag V40 (%) and rectum V40 (%) were selected as representative dosimetric covariates in the clinically adjusted model. In this model, mPNI remained associated with grade ≥1 ARE (OR, 1.163; 95% CI, 1.072–1.261; P<0.001). The principal regression results are summarised in Table III, and the full clinically adjusted model is provided in Table SII.

Sensitivity analyses using more clinically meaningful endpoints showed that mPNI remained associated with grade ≥2 ARE (OR, 1.069; 95% CI, 1.023–1.117; P=0.003) and grade ≥3 ARE (OR, 1.071; 95% CI, 1.013–1.133; P=0.016) in models adjusted for age, FIGO stage, surgery, concurrent chemotherapy, brachytherapy, prescription dose >50 Gy and boost treatment. The grade ≥3 model should be interpreted as exploratory due to the limited number of severe events. Additional details are shown in Table SII.

ROC curve analysis was performed to evaluate the predictive performance of pre-radiotherapy mPNI for ARE in patients with cervical cancer undergoing radiotherapy (22). ROC analysis showed that mPNI had an AUC of 0.835 (95% CI, 0.732–0.917) for grade ≥1 ARE. The optimal cut-off value was 80.406, with a sensitivity of 79.1% and a specificity of 76.0% (P<0.001). The bootstrap mean AUC was 0.835. ROC comparisons with albumin, lymphocyte count, total cholesterol and conventional PNI are shown in Fig. 3 and Table SIII, mPNI had the highest AUC, although its discriminatory performance was close to that of total cholesterol. Apparent calibration of the mPNI logistic model was acceptable in this dataset (Brier score, 0.139; Hosmer-Lemeshow χ² value, 3.004; P=0.934). Decision-curve analysis showed that the mPNI model provided a higher net benefit than the treat-all and treat-none strategies across most threshold probabilities from 0.25 to 0.80 (Fig. 4).

Compared with its individual components and conventional PNI, mPNI had the highest AUC, although its discrimination was close to that of total cholesterol. Paired bootstrap comparisons showed that the AUC of mPNI was higher than that of albumin (difference, 0.138; P=0.048), total lymphocyte count (difference, 0.226; P<0.001) and conventional PNI (difference, 0.143; P=0.022), but was not significantly higher than that of total cholesterol (difference, 0.023; P=0.644). These findings suggest that mPNI may provide additional discriminatory information beyond albumin, lymphocyte count and conventional PNI in this cohort; however, its advantage over total cholesterol alone was not statistically significant.