This study was approved by the institutional review board of Asan Medical Center, and informed consent was waived due to the retrospective nature of the analysis. Using our institutional database, we identified 2042 consecutive patients who underwent first-line chemotherapy for initially metastatic PDAC from January 2009 to December 2019. Patients who had history of other malignancy or had undergone locoregional or systemic treatment for PDAC were excluded. Among the 1257 patients screened, patients who became lost to follow up before or during the first cycle of first-line chemotherapy, received blinded chemotherapy protocol as parts of the randomized clinical trials, had no available baseline CT, and did not receive standard dose of chemotherapy were excluded. A total of 636 patients were included ( Figure 1 ). Baseline clinical data including age, sex, height, weight, body mass index (BMI), body surface area, Eastern Cooperative Oncology Group (ECOG) performance status, and laboratory data including carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9) were collected from electronic medical records.

Patients received either FOLFIRINOX (n=110) or gemcitabine-based treatment (n=526) as a first-line chemotherapy. In FOLFIRINOX, oxaliplatin (85 mg/m²), leukovorin (400 mg/m²), and irinotecan (180 mg/m²) were delivered intravenously on day 1 followed by 400 mg/m² (bolus) and 2400 mg/m² (continuously over a 46-h period) of 5-FU every 2 weeks. Dose modification was made in 56 patients (50.9%) to slightly reduce the dose of irinotecan (120 mg/m² or 150 mg/m²) at the treating oncologist’s decision. Among the gemcitabine-based regimens, gemcitabine plus nab-paclitaxel was most common (53.8%; 283/526), followed by gemcitabine alone (30.4%; 160/526), gemcitabine plus erlotinib (9.9%; 52/526) and gemcitabine plus capecitabine (5.9%; 31/526). In all gemcitabine-based regimens, patients received a slow intravenous infusion of gemcitabine (1000 mg/m²) on days 1, 8, and 15 with or without combined drugs of a 28-day cycle (every 4 weeks). All chemotherapy-related toxicities were evaluated based on the National Cancer Institute Common Toxicity Criteria for Adverse Events (CTCAE) version 5.0 (26). Toxicity assessments were performed at each scheduled chemotherapy session or whenever clinically indicated by the treating oncologists.

The primary endpoints were grade 3–4 toxicity based on CTCAE version 5.0 (26) and any treatment-modifying toxicity which led to a dose reduction, delayed administration, drug skip or discontinuation. In both endpoints, toxicities developed only during the first cycle were recorded.

Baseline contrast-enhanced abdomen CT images were used to assess body composition. A single, portal venous phase axial CT image at the level of lower endplate of the 3^rd lumbar vertebra was used. Details of CT scanning techniques are in Supplementary Material and Supplementary Table 1 . As previously reported, this level is considered the best reference site to assess lean muscle (27) and have shown a better correlation with whole body muscle mass (12) than other methods, including bioelectrical impedance analysis and dual x-ray absorptiometry (28). The cross-sectional areas of total abdominal wall muscle (skeletal muscle area [SMA]; including psoas, paraspinal, transversus abdominis, rectus abdominis, quadratus lumborum, and internal and external obliques), subcutaneous adipose tissue, and visceral adipose tissue were measured with the pre-established thresholds (from -29 to +150 HU for SMA and from -190 to -30 HU for SAT and VAT) (29). The low attenuated muscle area (LAMA) was obtained with the pre-established threshold (from -29 to +29 HU) (18, 30, 31). All segmentation processes were automatically performed using the web-based toolkit program (32), and reviewed and corrected as necessary by a board-certified abdominal radiologist (H.J.P. with 8 years of clinical experience) who were blinded to the patient information.

The body morphometry parameters were normalized by being divided to the patient height squared (cm²/m²) and reported as indexes including skeletal muscle index (SMI), subcutaneous adipose tissue index (SATI), and visceral adipose tissue index (VATI). LAMA/SMA index was expressed as percentage, i.e., 100 × (LAMA [cm²]/SMA [cm²]). To identify the presence of low muscle mass, the sex-specific cutoff values were set at SMI less than 41 cm²/m² for women, and less than 43 cm²/m² (if BMI is less than 25 cm/kg²) or less than 53 cm²/m² (if BMI is equal or higher than 25 cm/kg²) for men (16). Patients with LAMA/SMA index ≥ 20% were regarded as having myosteatosis. To evaluate whether the low muscle mass accompanied by myosteatosis has any impact on toxicity, we also evaluated the presence or absence of low muscle mass with myosteatosis.

Variables not normally distributed were log-transformed. To identify factors predicting the occurrence of grade 3–4 toxicity or treatment-modifying toxicity, univariate and multivariable binary logistic regression analyses were performed including the candidate predictors selected from all clinical and body composition parameters using bootstrapping with 500 interactions (i.e., selected variables in >50% of the bootstrap models) (33), as recommended by Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) statement (34). Missing data were processed with multiple imputation (35) and with the majority method for variable selection (36). To evaluate whether the chemotherapy protocol modifies the association of body morphometry parameters with the occurrence of toxicity, interaction terms (protocol × body morphometry parameters) were included in the regression analysis. The incremental differences between the models in predicting treatment-related toxicities were evaluated using net reclassification improvement (NRI) and integrated discrimination improvement (IDI). The NRI is used to evaluate the net proportion of patients who have been reclassified correctly using the new model (i.e., with body morphometry analysis) relative to the baseline model (i.e., without body morphometry analysis) whereas the IDI measures the improvement in sensitivity of the new model relative to the baseline model, without a loss in specificity (37). All statistical analyses were performed using R version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria). Two-sided P values < 0.05 were considered significant.

The patients’ characteristics are summarized in Table 1 . The median age of the 636 included patients was 60 years, and 58.6% (373/636) were men. Median BMI was 22.6 kg/m². Most of the patients had ECOG performance score of 0–1 (86.6%; 551/636). Low muscle mass was present in 34.0% of patients (216/636). Patients who received FOLFIRINOX were younger than those who received gemcitabine-based chemotherapy (median age, 56 vs. 61; P < 0.001). The proportion of men was higher in patients who received FOLFIRINOX (56.7% vs. 68.2%; P = 0.03). Patients received gemcitabine-based regimen had a higher CEA and lower hemoglobin level than those who received FOLFIRINOX (CEA median value, 5.0 ng/mL vs. 3.2 ng/mL; P = 0.04 and hemoglobin median value, 12.7 g/dl vs. 13.1 g/dl; P = 0.01). There was no significant difference in BMI, ECOG performance score, and the prevalence of low muscle mass and no other laboratory data were different between two regimen groups. SMI was higher in patients who received FOLFIRINOX than gemcitabine-based regimen (47.6 cm²/m² vs. 45.3 cm²/m²; P = 0.01) but no other body morphometry parameters were different between two groups. The median time from CT scan to the initiation of chemotherapy was 13 days (interquartile range, 9 to 21 days).

Among the 636 included patients, 160 patients (25.2%) experienced at least one grade 3–4 toxicity during the first cycle of the first-line chemotherapy. Neutropenia was the most common type of grade 3–4 toxicity (75.0%; 120/160), followed by nausea (14.4%, 23/160) and thrombocytopenia (5.6%, 9/160). Treatment-modifying toxicity occurred in 247 patients (38.8%). In patients who experienced treatment-modifying toxicity, treatment modification included dose reduction in 100 patients (15.7%), delayed administration in 63 patients (9.9%), both dose reduction and delayed administration in 34 patients (5.3%), drug skip in 35 patients (5.5%) and discontinuation of chemotherapy in 15 patients (2.4%). Details of the toxicity occurred in patients are provided in Table 2 . Toxicity according to each gemcitabine-based regimen is shown in Supplementary Table 2 .

Results of the univariate and multivariable analyses on the occurrence of grade 3–4 toxicity is shown in Table 3 . In univariate analysis, the type of chemotherapy regimen, white blood cell (WBC) count, platelet count, LAMI, presence of low muscle mass, and presence of both low muscle mass and myosteatosis had association with the occurrence of grade 3–4 toxicity. Multivariable analysis including the factors selected from 500 bootstrap samples (chemotherapy regimen, WBC count, and the presence of both low muscle mass and myosteatosis) revealed that all the input variables influenced the occurrence of grade 3–4 toxicity (odds ratio [OR] 2.46 [95% CI 1.55–3.91] for chemotherapy regimen [reference: gemcitabine-based]; OR 0.74 [95% CI 0.67–0.82] for WBC count; and OR 1.73 [95% CI 1.14–2.63] for low muscle mass with myosteatosis). While grade 3–4 toxicity occurred in 65 among 216 patients with low muscle mass (12.6%), 50 out of 152 patients with both low muscle mass and myosteatosis experienced grade 3–4 toxicity (32.9%) ( Figure 2 ). When testing the interaction between chemotherapy regimen and the presence of low muscle mass with myosteatosis, there was no association between those two parameters (P value for interaction = 0.85). By adding the presence of low muscle mass and myosteatosis to the WBC, the model achieved a net improvement in its classification and discrimination performance compared with the use of WBC alone (NRI, 0.196 [95% CI, 0.035–0.358, P = 0.02]; IDI, 0.012 [95% CI, 0.003–0.021, P = 0.01]). This suggested that there is a significant incremental value in using body composition analysis to predict grade 3–4 toxicity.

The factors regarding the occurrence of treatment-modifying toxicity are shown in Table 4 . In univariate analysis, WBC count, platelet count, alanine aminotransferase, and the presence of both low muscle mass and myosteatosis were associated with the occurrence of treatment-modifying toxicity. WBC count and the presence of low muscle mass with myosteatosis were the factors included in multivariable analysis selected from 500 bootstrap samples. The two factors had significant influence on the occurrence of treatment-modifying toxicity: OR of 0.996 (95% CI 0.99–0.998) for platelet and OR of 1.83 (95% CI 1.26–2.66) for the presence of both low muscle mass and myosteatosis. In terms of predicting treatment-modifying toxicities, inclusion of the presence of a low muscle mass and myosteatosis to the platelet count improved the model’s classification and discrimination ability compared with the use of the platelet count alone (NRI, 0.198 [95% CI, 0.059–0.337, P = 0.005]; IDI, 0.016 [95% CI, 0.006–0.026], P = 0.002). Treatment-modifying toxicity occurred in 95 patients among 216 patients with low muscle mass (44.0%). Of 152 patients with both low muscle mass and myosteatosis, 74 patients experienced treatment-modifying toxicity (48.7%) ( Figure 2 ). The representative cases are shown in Figure 3 .