Predicting morbidity of liver resection
Vishal G. Shelat
Winston W. L. Woon
Yiong H. Chan
Sameer P. Junnarkar
Received: 27 July 2017 /Accepted: 25 January 2018 /Published online: 7 February 2018
Springer-Verlag GmbH Germany, part of Springer Nature 2018
Purpose Multiple models have attempted to predict morbidity of liver resection (LR). This study aims to determine the efficacy
of American College of Surgeons National Surgical Quality Improvement Program (ACS-NSQIP) surgical risk calculator and the
Physiological and Operative Severity Score in the enUmeration of Mortality and Morbidity (POSSUM) in predicting post-
operative morbidity in patients who underwent LR.
Methods A retrospective analysis was conducted on patients who underwent elective LR. Morbidity risk was calculated with the
ACS-NSQIP surgical risk calculator and POSSUM equation. Two models were then constructed for both ACS-NSQIP and
POSSUM—(1) the original risk probabilities from each scoring system and (2) a model derived from logistic regression of
variables. Discrimination, calibration, and overall performance for ACS-NSQIP and POSSUM were compared. Sub-group
analysis was performed for both primary and secondary liver malignancies.
Results Two hundred forty-five patients underwent LR. Two hundred twenty-three (91%) had malignant liver pathologies. The
post-operative morbidity, 90-day mortality, and 30-day mortality rate were 38.3%, 3.7%, and 2.4% respectively. ACS-NSQIP
showed superior discriminative ability, calibration, and performance to POSSUM (p = 0.03). Hosmer-Lemeshow plot demon-
strated better fit of the ACS-NSQIP model than POSSUM in predicting morbidity.
Conclusion In patients undergoing LR, the ACS-NSQIP surgical risk calculator was superior to POSSUM in predicting mor-
Liver resection (LR) is an established curative treatment op-
tion for both benign and malignant liver disorders [1–3].
However, morbidity risks remain high, often resulting in in-
creased length of stay and hospital costs [4–9]. Identifying
patients who are at increased risk of morbidity will aid
decisions regarding patient selection, preoperative optimiza-
tion, and employment of targeted treatment post-operatively.
This in turn will result in efficient allocation of appropriate
resources in reducing morbidity and improving overall
Several studies have evaluated the effectiveness of various
risk prediction models in predicting morbidity following LR.
Some models were not initially designed for predicting mor-
bidity in a LR population and were therefore not very accu-
rate; others were only helpful in predicting morbidity post-
operatively. The Physiological and Operative Severity Score
for the enUmeration of Mortality and Morbidity (POSSUM) is
a commonly used predictor of morbidity for gastrointestinal
surgery [10–13]. While it proved effective in certain surgical
fields, it was demonstrated to over-predict morbidity rates in
hepatobiliary surgery. Its score was subsequently revised with
logistic regression . Other scores like the Model for End-
Stage Liver Disease, American Society of Anesthesiology
(ASA) class, Child-Pugh-Turcotte class, and Charlson Index
of Comorbidity and the Estimation of Physiologic Ability and
Permissions: All tables and figures are original works and have not been
published elsewhere before.
* Sameer P. Junnarkar
Ministry of Health Holdings, 1 Maritime Square, #11-25
HarbourFront Centre, Singapore 099253, Republic of Singapore
Hepato-Pancreatico-Biliary Surgery, Department of General Surgery,
Tan Tock Seng Hospital, 11 Jalan Tan Tock Seng, Singapore 308433,
Republic of Singapore
Biostatistics Unit, National University Health System, 1E Kent Ridge
Road, Singapore 119228, Republic of Singapore
Langenbeck's Archives of Surgery (2018) 403:359–369