Prediction of intraoperative complexity from preoperative patient data for laparoscopic cholecystectomy

https://doi.org/10.1016/j.artmed.2011.04.012Get rights and content

Abstract

Objective

Different reasons may cause difficult intraoperative surgical situations. This study aims to predict intraoperative complexity by classifying and evaluating preoperative patient data. The basic prediction problem addressed in this paper involves the classification of preoperative data into two classes: easy (Class 0) and complex (Class 1) surgeries.

Methods and material

preoperative patient data were collected from 337 patients admitted to the Klinikum rechts der Isar hospital in Munich, Germany for laparoscopic cholecystectomy (LAPCHOL) in the period of 2005–2008. The data include the patient's body mass index (BMI), sex, inflammation, wall thickening, age and history of previous surgery, as well as the name and level of experience of the operating surgeon. The operating surgeon was asked to label the intraoperative complexity after the surgery: ‘0’ if the surgery was easy and ‘1’ if it was complex. For the classification task a set of classifiers was evaluated, including linear discriminant classifier (LDC), quadratic discriminant classifier (QDC), Parzen and support vector machine (SVM). Moreover, feature-selection was applied to derive the optimal preoperative patient parameters for predicting intraoperative complexity.

Results

Classification results indicate a preference for the LDC in terms of classification error, although the SVM classifier is preferred in terms of results concerning the area under the curve. The trained LDC or SVM classifier can therefore be used in preoperative settings to predict complexity from preoperative patient data with classification error rates below 17%. Moreover, feature-selection results identify bias in the process of labelling surgical complexity, although this bias is irrelevant for patients with inflammation, wall thickening, male sex and high BMI. These patients tend to be at high risk for complex LAPCHOL surgeries, regardless of labelling bias.

Conclusions

Intraoperative complexity can be predicted before surgery according to preoperative data with accuracy up to 83% using an LDC or SVM classifier. The set of features that are relevant for predicting complexity includes inflammation, wall thickening, sex and BMI score.

Introduction

Pattern recognition is gaining increasing attention in the medical domain, as it has proven more effective than common clinical statistical tools in the prediction of clinical outcomes. Pattern recognition has been used for decades for segmentation purposes in medical imaging applications [1]. Classification methods have recently been used for decision support in computer-aided diagnosis. For example, Lin [2] designed a diagnosis model for the treatment of liver disease using classification and regression trees, and Lee et al. [3] developed a computer-aided diagnosis system for evaluating pulmonary nodules using feature selection and a linear discriminant classifier (LDC).

The vast amount of preoperative patient data generated before surgery motivates the construction of pattern recognition tools that are able to improve the accuracy of predictions regarding complexity factors that may occur during surgery. Nevertheless, the use of surgical preoperative data for prediction has been overlooked in the literature. This challenge has recently been addressed by conventional clinical statistical approaches, which lack the power of pattern recognition when using ranking and classification algorithms as prediction tools.

Surgical complications have been associated with increased inpatient hospital costs [4]. As a result, reducing complications has become a desirable objective for quality-improvement initiatives aimed at improving efficiency and safety in health care. Davenport et al. [5] have shown that preoperative risk factors and surgical procedure complexity are more effective predictors of hospital costs than complications are. This dependence between risk and procedure complexity is to be expected, as these measures were designed to predict complications. Nevertheless, the use of raw preoperative patient data in predicting complexity factors has been overlooked in the clinical literature. In most clinical studies [6], complexity factors often are based on nothing more than surgical expertise. Complexity factors are pre-designed and classified according to the surgeon's knowledge about possible complications of a specific procedure. Moreover, in the clinical literature, surgical complexity is estimated in per procedure and not with regard to the relative ease or difficulty of a procedure for a given patient [5]. Nonetheless, the literature does contain a small number of studies that assess the surgical complexity of individual procedures according to readily available patient data. Jenkins et al. [7] used patient demographics to predict the operative time for ventral hernia repair. In most cases, however, these approaches are statistical, concentrating only on evaluating the significance of individual parameters as independent variables. They lack the power of pattern recognition tools in performing classification and selecting a subset of parameters for which the classification performance improves the most.

Laparoscopic cholecystectomy (LAPCHOL) is one of the most commonly performed surgical procedures worldwide [8]. LAPCHOL is accepted as the gold standard in the treatment of symptomatic gallstones. Up to 700 000 LAPCHOL procedures are performed in the US each year [8]. The preoperative assessment of complexity factors is needed for frequent procedures such as LAPCHOL in order to avoid complications and delays and to guarantee an efficient course of surgery.

This study aims to estimate the intraoperative complexity of LAPCHOL according to readily available preoperative patient data. Resource planning is a crucial topic in research on surgical efficiency. Tools are prepared in essentially the same way for all surgeries. In many cases, however, surgeons need advanced tools in case of complications. These tools can be prepared preoperatively for surgeries that are predicted to be complex. Furthermore, any complex procedure always involves the risk of conversion to an open procedure. This measure can be taken preoperatively to avoid intraoperative delays during surgery. The members of the surgical team can be considered another example of resources. If a surgery is identified as complex, it can be assigned to a more experienced team (including the surgeon, the surgeon-assistant or the operative-nurse), thereby allowing for a safer and efficient surgical procedure. Moreover, although the average time required for LAPCHOL is about 60 min, in practice, it may vary from 20 min to 5 h, depending on the level of complexity [9]. The variation in operative time is due to uncertainties regarding the complexity of the LAPCHOL procedure across a diverse patient population [7]. Estimating the level of complexity beforehand may improve the flexibility and accuracy of preoperative planning. The scheduled time can be increased for complex procedures and decreased for easy ones, thus allowing flexible preoperative planning.

In complex surgical situations, the surgeon gets into dilemma weather to continue the intended operation, or to deviate from the planned procedure. Complexity estimation before surgery can aid surgeons in decisions regarding whether to proceed with a minimally invasive approach, perform an open procedure or make a referral to a more experienced surgeon. It may also be useful as an informative tool for communicating about details of the intervention with the patient and for explaining the various risks of laparoscopic and open procedures.

This study aims to delineate the relevant demographic and sonographic patient data that are predictive of the intraoperative complexity of LAPCHOL interventions.

This work contributes in several ways. First, it introduces pattern recognition tools for processing surgical data; in the medical literature, these data are usually processed using common clinical statistical tools. Second, this study provides an evaluation of a number of classifiers based on a 337-patient dataset that was collected in the period of 2005–2008. Third, it provides an analysis indicating which set of preoperative features is effective for predicting intraoperative complexity, thus filtering out insignificant features. The study also measures objectivity bias in the assessment of surgical complexity by surgeons with various levels of experience. Finally, the study evaluates the conformity of the results with clinical practice.

Section snippets

Materials and methods

The preoperative features used in the experiments were collected in the period of 2005–2008 from N = 337 patients who had been admitted for elective LAPCHOL procedures, which involve removing the patient's gallbladder in case of symptomatic gallstones.

Experimental validation

To address the problem of predicting surgical complexity, Section 3.1 focuses on the identification of the optimal classifier for the preoperative dataset described in Table 1 and the number of samples required for optimal training. Section 3.2 aims to identify which subsets of features allow the prediction of complexity with accuracy comparable to, or better than, the complete set of features.

To answer these questions, we conducted both classification and feature selection. For the experiments

Discussion and conclusions

Intraoperative complexity can be predicted before surgery according to readily available preoperative data. The problem addressed in this paper involves the classification of preoperative data in two classes: Easy (Class 0) and complex (Class 1) surgeries. Learning-curve results indicated a preference for the LDC classifier in terms of classification error, although the SVM classifier is preferable in terms of AUC error.

Feature selection was used to identify the preoperative features that are

Acknowledgments

We would like to thank David Tax for the useful discussions and advice. The data-collection process was supported in part by the German Research Foundation DFG, FE 585/1-1. This research was supported by the Dutch STW Technology Foundation, project number: 07320.

References (21)

There are more references available in the full text version of this article.

Cited by (17)

  • Predicting a ‘difficult cholecystectomy’ after mild gallstone pancreatitis

    2019, HPB
    Citation Excerpt :

    Studies in cohorts of unselected patients have identified several risk factors for technical difficulty of cholecystectomy. Among these are male sex, previous endoscopic sphincterotomy, increasing age and inflammation of the gallbladder or pancreas.11–14 One population based study found an odds ratio of 3.2 for bile duct injury in 30 211 cholecystectomies after gallstone pancreatitis.15

  • Pre-operative prediction of surgical morbidity in children: Comparison of five statistical models

    2015, Computers in Biology and Medicine
    Citation Excerpt :

    One area of biomedical research in which data mining may be particularly useful is in outcome prediction using large clinical databases, such as the American College of Surgeons’ National Surgical Quality Improvement Program (ACS NSQIP) database [2]. Of the few studies investigating the performance of data mining algorithms for predicting surgical morbidity or mortality, most have been small (several hundred or several thousand patients) [5–10], though a few larger studies have been reported [11–17]. These studies have been inconsistent in their findings, in that some have shown data mining algorithms to perform better than traditional logistic regression in terms of overall accuracy [13,14,16,18], discrimination [13,14,16], or calibration [11], whereas some have reported similar performance according to these measures [11,18–20].

  • Intervention time prediction from surgical low-level tasks

    2013, Journal of Biomedical Informatics
    Citation Excerpt :

    Second, a chosen set of diagnostic information relevant to the intervention type can be considered. The significance of preoperative information on intervention time and complexity has been shown in related work [11,12,24]. The influence of these parameters depends on the intervention type.

  • Validation of a scoring system to predict difficult laparoscopic cholecystectomy

    2013, International Journal of Surgery
    Citation Excerpt :

    In our study it has not been found as a significant factor (p = 0.265). Patient, who require hospitalization for repeated attacks of acute cholecystitis, carry more chances of difficult laparoscopic cholecystectomy and conversion, probably due to dense adhesions at calot's triangle and gall bladder fossa.13,14 In our study also, it was found to be a significant factor for prediction of difficult laparoscopic cholecystectomy (p = 0.031).

View all citing articles on Scopus
View full text