Page 93 - OR-1-1

Table 2. Patient demographics and tumor characteristics lower than those of the resistant group (Figure 4). Therefore,
Characteristics Matched group Unmatched group p‑value we applied fluorescence staining data relevant to sensitivity
(n=26) (n=35) and resistance to subsequent analyses.
Age, years 57.5 (49.7 – 62.6) 57.6 (48.3 – 62.5) 0.891
Gender 3.4. Prognostic value of the PDOs-based drug
sensitivity testing
Female 11 (42.3%) 12 (34.3%) 0.710
Male 15 (57.7%) 23 (65.7%) The Kaplan–Meier curve analysis showed that RFS events
AFP, μg/L occurred in 34 of 54 patients (63.0%) in the matched
<20 25 (96.2%) 29 (82.9%) 0.228 group and in 9 of 53 patients (17.0%) in the unmatched
group. Median RFS was reached only in unmatched group
≥20 1 (3.8%) 6 (17.1%) (16.5 months [95% Confidence interval (CI), 11.1 – 46.7]),
CA19-9, U/L the log-rank test confirmed that the matched group had
<19 8 (30.8%) 13 (37.1%) 0.806 significantly better RFS (p=0.007) (Figure 4).
≥19 18 (69.2%) 22 (62.9%)
We employed XGBoost machine learning and Cox
CEA, U/L regression to analyze the impact of drug testing results
<5 17 (65.4%) 31 (88.6%) 0.061 on the prognosis of ICC. In the XGBoost model, 2-year
≥5 9 (34.6%) 4 (11.4%) recurrence events were used as the outcome, and SHAP
TBIL, μmol/L 14.3 (9.5 – 18.5) 10.9 (9.6 – 16.6) 0.406 values were utilized to illustrate the influence of different
Albumin, g/L 42.6 (39.9 – 44.3) 41.3 (39.4 – 44.5) 0.599 features on the outcome. The results indicated that the
ALT, U/L 27.5 (18.0 – 54.0) 25.0 (14.0 – 59.5) 0.594 PDOs inhibition ratio was an important feature affecting
ALP, U/L 89.5 (79.0 – 127.0) 92.0 (67.0 – 122.5) 0.610 recurrence within 2 years post-surgery (Figure 5A).
Diameter, cm Scatter plots of SHAP values and feature assignments
<5 6 (23.1%) 5 (14.3%) 0.585 further revealed distinctions in PDOs inhibition ratio,
≥5 20 (76.9%) 30 (85.7%) tumor number, and tumor size between the recurrence
Number and non-recurrence groups (Figure 5B-D). Furthermore,
Single 16 (61.5%) 17 (48.6%) 0.456 we randomly selected ten patients with recurrence
and ten patients without recurrence and constructed
Multiple 10 (38.5%) 18 (51.4%) a heatmap based on their feature values. The heatmap
Cirrhosis demonstrated that the PDOs inhibition ratio was lower
No 25 (96.2%) 30 (85.7%) 0.358 in the recurrence group compared to the non-recurrence
Yes 1 (3.8%) 5 (14.3%) group (Figure 5E).
Surgical margin, cm
<1 14 (53.8%) 27 (77.1%) 0.101 3.5. Construction a prediction model for RFS based
≥1 12 (46.2%) 8 (22.9%) on PDOs-based drug sensitivity testing
Hepatectomy Given that XGBoost cannot analyze the impact of features
Minor 6 (23.1%) 13 (37.1%) 0.372 on survival, we conducted univariate and multivariate
Major 20 (76.9%) 22 (62.9%) Cox regression analyses to assess the correlations between
Differentiation drug testing, clinical, and pathological characteristics and
Well 6 (23.1%) 10 (28.6%) 0.851 RFS in the entire cohort (Table 3). Univariate analysis
indicated that male, tumor diameter >5 cm, multiple
Poor 20 (76.9%) 25 (71.4%) tumor nodules, the presence of vascular invasion, and
Vascular invasion drug resistance determined through PDOs testing were
No 16 (61.5%) 20 (57.1%) 0.935 associated with poor RFS outcomes. In multivariate
Yes 10 (38.5%) 15 (42.9%) analysis, the inconsistency of clinical chemotherapy
LN metastasis regimens with drug testing results (unmatched group)
No 15 (57.7%) 20 (57.1%) 1.000 (hazard ratio (HR), 2.27; 95% CI, 1.12 – 4.59; p=0.023),
Yes 11 (42.3%) 15 (42.9%) multiple tumors (HR, 2.06; % CI, 1.02 – 4.17; p=0.044),
Local invasion and the presence of vascular invasion (HR, 2.15; % CI,
No 18 (69.2%) 29 (82.9%) 0.345 1.08 – 4.26; p=0.028) were identified as independent risk
Yes 8 (30.8%) 6 (17.1%) factors for RFS. Based on the results of the multivariate
Abbreviations: AFP: Alpha-fetoprotein; ALP: Alkaline phosphatase; analysis, a nomogram including the results of PDOs-
ALT: Alanine aminotransferase; CA19 – 9: Carbohydrate antigen 19 – 9; based drug sensitivity testing was constructed to predict
CEA: Carcinoembryonic antigen; LN: Lymph node; TBIL: Total bilirubin. RFS (Figure 6).

Volume 1 Issue 1 (2025) 6 doi: 10.36922/or.8571