Table 4.

Relative Importance of Input Variables Compared between Neural Network and Logistic Regression Using SHAP Values and Odds Ratios

Anastomotic leak
Variable	SHAP value	Variable	Odds ratio*
Approach	0.016	Chemotherapy	1.32
Mechanical bowel prep	0.016	Steroid use	1.25
Antibiotic bowel prep	0.014	Antibiotic bowel prep	0.81
Emergent indication	0.011	Mechanical bowel prep	0.86
Steroid use	0.010	Approach	1.14
Chemotherapy	0.009	Emergent indication	0.94
Indication	0.009	Indication	1.01
Bile leak
Variable	SHAP value	Variable	Odds ratio*
Use of drain	0.034	Biliary reconstruction	1.88
Biliary reconstruction	0.029	Pringle maneuver	1.42
Approach	0.017	Approach	1.37
Biliary stent	0.016	Neoadjuvant chemo-embolization	1.37
Pringle maneuver	0.015	Use of drain	1.37
# of concurrent resections	0.011	Neoadjuvant chemo-infusion	0.73
Concurrent ablation	0.01	Biliary stent	1.22
Viral hepatitis	0.009	Neoadjuvant ablation	1.19
Neoadjuvant therapy	0.009	Neoadjuvant chemotherapy	1.17
Neoadjuvant chemo-embolization	0.008	Viral hepatitis	1.13
Pancreatic fistula
Variable	SHAP value	Variable	Odds ratio*
Gland texture	0.039	Drains placed	1.27
Indication	0.036	Gland texture	1.25
Drain amylase (POD1)	0.027	Chemotherapy	0.89
Reconstruction	0.010	Reconstruction	1.09
Duct size	0.008	Indication	0.92
Vascular resection	0.006	Radiation therapy	0.93
Biliary stent	0.006	Vascular resection	0.94
Jaundice	0.006	Duct size	0.94
Radiation therapy	0.006	Antibiotic	0.96
Chemotherapy	0.005	Jaundice	0.97

^*Odds ratio is sorted by distance from 1 (null value)