Cornelius A Thiels, DO, MBA, Denny Yu, PhD, Amro Abdelrahman, MBBS, Elizabeth Habermann, PhD, Susan Hallbeck, PhD, Juliane Bingener, MD. Mayo Clinic
INTRODUCTION: Reliable prediction of operative duration is essential for maximizing resource utilization, reducing cost, and improving patient and surgical team satisfaction. Traditional scheduling systems utilize historical operative durations by procedure type to forecast future procedures. We hypothesized that the inclusion of patient-specific factors would improve the accuracy in predicting operative duration.
METHODS: We reviewed all cases for a common operation, elective laparoscopic cholecystectomy (LC) performed between January 2007 to June 2013 at a single institution. Non-elective procedures and those with concurrent procedures were excluded. Univariate analysis evaluated the effect of the following patient factors on operative duration: age, gender, BMI, ASA, liver function test (LFT), WBC, amylase, smoking, and comorbidities. Multivariable linear regression models were constructed using the significant factors (p<0.05). The patient factors model was compared to the institutional prediction model, which uses historical surgeon specific operative duration. External validation was performed using the National Surgical Quality Improvement Program (NSQIP) database.
RESULTS: 1801 LC patients met the inclusion criteria. Univariate analysis of female sex was associated with reduced operative duration (-7.52 minutes, p<0.001 vs male sex) while increasing BMI (+5.06 minutes BMI 25-29.9, +6.89 minutes BMI 30-34.9, +10.38 minutes BMI 35-39.9, +17.02 minutes BMI 40+, all p<0.05 vs normal BMI), increasing ASA (+7.39 minutes ASA III, +38.26 minutes ASA IV, all p<0.01 vs ASA I), and elevated LFTs (+4.85 minutes, p=0.018 vs unknown LFTs) were predictive of increased operative duration. A model was then constructed using these predictive factors. Of the patients with available institutional predicted operative duration (n=633), the traditional institutional model was poorly predictive of actual operative duration (R2=0.001) compared to the patient factors model (R2=0.079). The addition of surgeon as a variable in the patient factors model further improved predictive ability of the model (R2=0.175, Figure 1). The model remained predictive on external validation using n=20,034 LC patients from the NSQIP database (R2=0.103).
CONCLUSION: The current operative scheduling standard is unreliable and may contribute to costly over- and under-estimation of operative duration. The use of routinely-available pre-operative patient factors can improve the prediction of operative duration during LC.
Figure 1. Relationship between actual and predicted operative duration for three models.