Journal of Cancer Research Reviews & Reports

Application

Since the introduction of minimally invasive surgery (MIS) in 1991, the laparoscopic technique has rapidly evolved across all specialities with an increasing uptake of robotic technologies [1- 4]. The benefits of the minimally invasive approach have been well documented in terms of postoperative pain, shorter hospital stay, improved cosmesis and quicker return to normal function [5-12]. However, these must be weighed against the limitations, which include a long learning curve and increased operative time particularly with advances in training/techniques allowing surgeons to operate on more complex cases [11-15]. Prolonged operative timing is likely to negatively impact clinical outcomes following laparoscopic colorectal resection [16]. A metanalysis looking at the impact of prolonged operative time across various specialties showed statistically significant data of increased complications with time increments [17]. Prolonged surgery has an additional impact on health care economics [18].

Optimising cost-effectiveness of health care is a modern challenge for the NHS. Trusts in England estimate a £2.3 billion deficit by the end of 2016 [18]. Additionally, the COVID-19 pandemic has created unprecedented challenges including great backlog of waiting lists for routine operations, estimated at 5.7 million [19,20]. Therefore, it is essential that every effort is made to make each service as efficient and as cost-effective as possible, without compromising patient care. Thus, an accurate prediction of operative timing is essential in operative room planning. Additionally, knowledge of factors that can accurately predict prolonged surgery is important for theatre planning, aiming to enhance perioperative efficiency and accurately plan theatre resources [21,22].

The application of Artificial Neural Networks (ANN) has been widely adopted across various medical specialties. Within in surgery there are promising results for applying ANN including diagnosis and predicting cancer outcomes including survival [23-25].

This study aimed to apply artificial neural networks (ANN) as a predictive tool for operating time and identify those who are likely to have prolonged laparoscopic colorectal cancer surgery.

Methods

An observational review of a dedicated, previously reported colorectal cancer patient database was performed [26]. Inclusion criteria were patients with biopsy proven colorectal adenocarcinoma undergoing elective laparoscopic surgery with curative intent, data were collected over 12 years. Exclusion criteria included metastatic disease, those who underwent open surgery or were unfit for resection. Patient and tumour demographics were captured. Total operative time was defined as the time taken from entering to the anaesthetic room to leaving the theatre to recovery to completion of skin closure. Other operative factors such as total blood loss and conversion from laparoscopic to open surgery were recorded. Conversion from laparoscopic to open surgery was defined as the inability to complete the dissection laparoscopically, including vascular ligation, and usually, but not always, requiring an incision larger than that required to remove the specimen.

Artificial Neural Network (ANN)

ANN is a computational model composed of many highly interconnected processing elements (neurons) working in unison. Neural networks process information in a similar way with the human brain and like people, ANNS learn by example. Detailed ANN descriptions have previously been provided [26-28].

Statistical Analysis

The data was analysed using SPSS (v24; SPSS Inc, Chicago, IL, USA). For categorical data, analysis included the use of cross tabulation, odds ratios & chi-square to test the difference or association between groups.

Multivariate association between variables was assessed using binary logistic regression. Only those variables identified as being significantly associated with diagnosis to treatment in the univariate analyses were included in the modelling process. The data was screened for potentially influential observations, and the extent of multicollinearity amongst predictor variables was examined using variance inflation factors. The sample size is perceived to be sufficiently large to ensure stable logistic regression parameter estimates were obtained which are not suspected on accuracy or precision. Odds ratio for individual variables is computed from the regression equation as OR = eB adjusted for all other variables simultaneously. The effect magnitude was quantified using the odds ratio (OR) with 95% confidence interval.

Multilayer Perceptron Neural Network (MLPNN) application in SPSS was used in this analysis. This involved having neurons with an input layer (independent variables), one hidden layer, and an output layer (outcome: Dichotomised Operative Time: OT => 5 hours & 0= < hours). The layers are interconnected and learns by example through adjusting the “weight” of each link between the neurons as shown in Figure 1, which allows adaptation of the model and accurate prediction [29,30].

Figure 1: Schematic of multilayer perceptron neural network for prolonged operative timing

Artificial Neural Network Cross Validity

Receiver operator characteristics curves (ROC), area under the curve (AUC), gain and lift charts, and comparison with logistic regression modelling were used for cross-validation of the ANN. The predictive quality of our ANN was tested using the patients in the data set that had not been used in the training phase.

Results

590 patients were selected, 554 were included, and 36 excluded due to incomplete data. Four hundred patients’ data sets (72.2%) were used for ANN training and 154 (27.8%) to test predictive accuracy. 59.3% male, mean age 70 years, and BMI of 26. 161 (29%) were ASA III. 261 (47%) had rectal cancer and 8.5% underwent neoadjuvant treatment. Mean operative time was 218 minutes (95% CI 210-226) with 436 (78.7%) of less than 5 hours and a conversion rate of 16%

The neural links are shown in Figure 1. Four hundred (72.2%) were selected as training cases with 154 (27.8%) used to test the predictive function of the ANN. Variables included in the MLPNN network were: age, gender, ASA, BMI, stage, location of cancer, neoadjuvant therapy, time from diagnosis, and planned stoma. Our ANN model identified; Neoadjuvant therapy, age, and BMI as the best independent predictors of prolonged operative timing (Table 1).

Validation of MLPNN

1. Classification: As Table 2 shows, our MLPNN correctly classified 111/119 (93.3%) cases that were under 5 hours operating time and predicted 23/35 (65.7%) cases accurately that were equal to or over 5 hours of operating. Overall, correctly predicted operative time in 87% of testing cases. The significant predictive properties of the MLPNN were confirmed by creating an ROC curve for both predictions of prolonged operative timing with an area under the curve of 0.893 for both < and > 5 hours (Figure 2).
2. Gains Chart: A gain chart was also created demonstrating if we use half of the sample (50% population using the diagonal line) and guess as to who is likely to have operative timing over 5 hours then are likely to get 50% true positive guesses. If we use the ANN model, we only need to have 30% of the sample to give us over 80% of true positive predictions compared to 50% by random selection by 50% of the population (Figure 3).
3. Lift Chart: The lift chart (Figure 4) illustrates that if we use the ANN and select only 10% of our sample (blue), we would be greater than 3.5 times more likely (red) to get a true positive prediction compared to using random selection
4. Artificial Neural Network Cross Validity and development of Prediction Score

Figure 2: Receiver operator characteristic curve for predictive properties of the MLPNN for prolonged operative timing

Figure 3: Gain chart for predictive properties of MLPNN for prolonged operative timing

Figure 4: Lift chart for predictive properties of MLPNN for prolonged operative timing

Logistic regression model to cross-validate the ANN results. This demonstrated the same factors to predict prolonged surgical timing including neoadjuvant therapy, age, sex, BMI, tumour location, and stoma formation (Table 3). A prediction score was then calculated from all predicting variables in the multivariate analysis (logistic regression) and the linear function of the binary logistic regression model: Predictive score = B1x1 + b2x2.....+ Constant

Predictive score based on this data: positive score OT > 5 hours = -.043xAge + .605xSex (1 male /0 female) +.056xBMI + .919x Tumour Location (1 Rectum/0 Colon) + 1.825x (Stoma 1 yes 0 No) + .001x Blood loss in mls + .006x Time Lag ? 2.345. The model has demonstrated good predictive properties with an area under the ROC amount to 0.843 (Figure 5).

Figure 5: Receiver operator characteristic curve for predictive properties of the logistic regression model for prolonged operative time. Area under the curve 0.843

Discussion

Predicting operative timing is challenging, involving multiple variables but essential for preoperative planning. Non-linear interaction between the different elements affecting operative timing makes it difficult to assess which variable will have the greatest impact on operative time. In this study, using artificial neural networks, we have predicted prolonged operative time within laparoscopic colorectal cancer and generated a prediction score that can be used in future research and service development.

Accurately predicting operative time is key to streamlining theatre management including time slots, correct staffing levels and reduced cancellations, managing expectations, and improving patient care and experience. The application of MLPNN was consistent with conventional logistical regression analysis in identifying the same factors that can predict prolonged surgery. It is interesting to note that the strongest predictor of prolonged operative timing was neoadjuvant therapy, then age and then BMI

Most of the identified predictive factors for prolonged surgery are clinically justified, as neoadjuvant therapy and male sex are generally associated with difficult rectal cancer surgery. It was interesting, however, to see that age among those factors. This could be explained by the fact that elder patients require more anaesthetic preparation, which is included within the operative time.

Operating theatres are one of the most expensive areas in a hospital to run, with an average cost of approximately £1200/hour [31]. They also represent one of the most profitable areas of healthcare delivery for NHS trusts, if delivered efficiently. Streamlining could lead to significant savings. For example, the NHS Institute for Innovation and Improvement calculated that the average trust has an opportunity to save £7 million a year in efficiency savings by running a ‘productive theatre’ [32]. Although longer operations are sometimes inevitable, knowledge of this beforehand can limit the increased cost associated with a prolonged operation. For example, the estimated operative duration is used to arrange operating room schedules and allocate staff [32]. If a case extends longer than expected, this could increase costs, as improper scheduling can lead to overtime pay, perioperative inefficiency, and personnel misallocation [21,22,33]. Overtime pay is costly, erosive to morale, contributing to burn out and lack of theatre staff retention. In our cost-minded healthcare system, predicting operative time is crucial for cost-effectiveness.

Limitations

The definition of prolonged operative timing is arbitrary (5 hours) based on a previous study indicating that procedures taking longer than 5 hours were associated with a significant increase in length of stay and blood loss [16].

Although widely used by technology and analytical companies, ANN presents some considerations when applied to healthcare scenarios. Even for common clinical situations, the size of datasets is unlikely to match those analysed by the industry, therefore, may not be able to accommodate a large number of input variables or achieve the highest level of accuracy. To a degree, this could be offset by performing state-wide registry-based ANN testing, subject to data protection and confidentiality considerations. This could inform benchmarking and quality assurance processes as well as informing public health and healthcare policy. Central limit theorem could also be expected to balance out many factors that can influence patient care. Matched institutional specific ANN could help inform planning for individual patients as this network would incorporate local provider factors that are likely to influence time to surgery. Our findings now require larger scale study and validation in external centres.

Conclusion

Artificial neural network using patient demographic and tumour data successfully predicted the timing of surgery and the likelihood of prolonged laparoscopic procedures. This finding could assist the personalisation of peri-operative care to enhance the efficiency of theatre utilisation.

Conflict of Interest: The authors declare that they have no conflict of interest

Ethical consideration: For this retrospective study already held, anonymised patient data, formal ethical approval for this type of study was not required. The patient database was originally created with local research ethics and data governance committee approval.

References

1. Veldkamp R, Kuhry E, Hop WC, Jeekel J, Kazemier G, et al. (2005) COlon cancer Laparoscopic or Open Resection Study Group (COLOR). Laparoscopic surgery versus open surgery for colon cancer: short-term outcomes of a randomised trial. Lancet Oncol 6: 477-484.
2. Braga M, Vignali A, Gianotti L, Zuliani W, Radaelli G, et al. (2002) Laparoscopic versus open colorectal surgery: a randomized trial on short-term outcome. Ann Surg 236: 759- 767.
3. Tittel A, Schippers E, Anurov M, Titkova S, Ottinger A, et al. (2001) Shorter postoperative atony after laparoscopic-assisted colonic resection? An animal study. Surg Endosc 15: 508-512.
4. Kasparek MS, Muller MH, Glatzle J, Manncke K, Becker HD, et al. (2003) Postoperative colonic motility in patients following laparoscopic-assisted and open sigmoid colectomy. J Gastrointest Surg 7: 1073-1081.
5. Lacy AM, Garcia-Valdecasas JC, Delgado S, Castells A, Taura P, et al. (2002) Laparoscopy-assisted colectomy versus open colectomy for treatment of non-metastatic colon cancer: a randomised trial. Lancet 29: 2224-2229.
6. Clinical Outcomes of Surgical Therapy Study Group, Nelson H, Sargent DJ, Wieand HS, Fleshman J, et al. (2004) A comparison of laparoscopically assisted and open colectomy for colon cancer. N Engl J Med 13: 2050-2059.
7. Tekkis PP, Senagore AJ, Delaney CP, Fazio VW (2005) Evaluation of the learning curve in laparoscopic colorectal surgery: comparison of right-sided and left-sided resections. Ann Surg 242: 83-91.
8. Park IJ, Choi GS, Lim KH, Kang BM, Jun SH (2009) Multidimensional analysis of the learning curve for laparoscopic colorectal surgery: lessons from 1,000 cases of laparoscopic colorectal surgery. Surg Endosc 23: 839-846.
9. Bailey MB, Davenport DL, Vargas HD, Evers BM, McKenzie SP (2014) Longer operative time: deterioration of clinical outcomes of laparoscopic colectomy versus open colectomy. Dis Colon Rectum 57: 616-622.
10. Lujan JA, Parrilla P, Robles R, Marin P, Torralba JA, et al. (1998) Laparoscopic cholecystectomy vs open cholecystectomy in the treatment of acute cholecystitis: a prospective study. Arch Surg 133: 173-175.
11. Agha R, Muir G (2003) Does laparoscopic surgery spell the end of the open surgeon? J R Soc Med 96: 544-546.
12. Cheng H, Clymer JW, Po-Han Chen B, Sadeghirad B, FerkoNC, et al. (2018) Prolonged operative duration is associated with complications: a systematic review and meta-analysis. J Surg Res 229: 134-144.
13. NICE guideline [NG125] Surgical site infections: prevention and treatment (2020) https://www.nice.org.uk/guidance/ ng125
14. . Plowman R (2000) The socioeconomic burden of hospital acquired infection. Euro Surveill 5: 49-50.
15. The King’s Fund: NHS trusts’ deficit estimated at £2.3 billion as NHS financial crisis deepens. https://www.kingsfund.org. uk/press/press-releases/nhs-trusts-deficit
16. Sky News- NHS waiting lists: Backlog hits record high with nearly six million awaiting treatment in England. https://news.sky.com/story/nhs-waiting-lists-backlog-hitsrecord-high-with-nearly-six-million-awaiting-treatment-inengland-12433516
17. Macdonald N, Clements C, Sobti A, Rossiter D, Unnithan A, et al. (2020) Tackling the elective case backlog generated by Covid-19: the scale of the problem and solutions. J Public Health (Oxf) 42: 712-716.
18. Eijkemans MJ, van Houdenhoven M, Nguyen T, Boersma E, Steyerberg EW, et al. (2010) Predicting the unpredictable: a new prediction model for operating room times using individual characteristics and the surgeon’s estimate. Anesthesiology 112: 41-49.
19. Stepaniak PS, Heij C, Mannaerts GH, de Quelerij M, de Vries G (2009) Modeling procedure and surgical times for current procedural terminology-anesthesia-surgeon combinations and evaluation in terms of case-duration prediction and operating room efficiency: a multicenter study. Anesth Analg 109: 1232- 1245.
20. Curtis NJ, West MA, Salib E, Ockrim J, Allison AS, et al. (2018) Time from colorectal cancer diagnosis to laparoscopic curative surgery-is there a safe window for prehabilitation? Int J Colorectal Dis 33: 979-983.
21. Hashimoto DA, Rosman G, Rus D, Meireles OR (2018) Artificial Intelligence in Surgery: Promises and Perils. Ann Surg 268: 70-76.
22. Le Cun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 28: 436-444.
23. NK, Luther A, Salib E, Allanby L, Messenger D, et al. (2015) The use of artificial neural networks to predict delayed discharge and readmission in enhanced recovery following laparoscopic colorectal cancer surgery. Tech Coloproctol 19: 419-428.
24. Leong KJ, Chapman MAS (2017) Current data about the benefit of prehabilitation for colorectal cancer patients undergoing surgery are not sufficient to alter the NHS cancer waiting targets. Colorectal Dis 19: 522-524
25. The Productive Operating Theatre Improving quality and efficiency in the operating theatre. The NHS Institute for Innovation and Improvement https://alesi-surgical.com/wpcontent/uploads/2019/09/Improving-quality-and-efficiencyin-the-operating-theatre.pdf
26. Peng JH, Fang YJ, Li CX, Ou QJ, Jiang W, et al. (2016) A scoring system based on artificial neural network for predicting 10-year survival in stage II A colon cancer patients after radical surgery. Oncotarget 19: 22939-23047.
27. Shi HY, Lee KT, Wang JJ, Sun DP, Lee HH, et al. (2012) Artificial neural network model for predicting 5-year mortality after surgery for hepatocellular carcinoma: a nationwide study. J Gastrointest Surg 16: 2126-2131.
28. Delaney CL, Davis N, Tamblyn P (2010) Audit of the utilization of time in an orthopaedic trauma theatre. ANZ J Surg 80: 217-222.
29. Ansari D, Nilsson J, Andersson R, Regner S, Tingstedt B, et al. (2013) Artificial neural networks predict survival from pancreatic cancer after radical surgery. Am J Surg 205: 1-7.
30. Bottaci L, Drew PJ, Hartley JE, Hadfield MB, Farouk R, et al. (1997) Artificial neural networks applied to outcome prediction for colorectal cancer patients in separate institutions. Lancet 16: 469-472.
31. . Burke HB, Goodman PH, Rosen DB, Henson DE, Weinstein JN, et al. (1997) Artificial neural networks improve the accuracy of cancer survival prediction. Cancer 15: 857-862.
32. . Harrison OJ, Smart NJ, White P, Brigic A, Carlisle ER, et al. (2014) Operative time and outcome of enhanced recovery after surgery after laparoscopic colorectal surgery. Journal of the Soceity of Laparoscopic and Robotic Surgerons 18: 265-272.
33. Alexander Emmanuel Blackmore, Mark Te Ching Wong, Choong Leong Tang (2014) Evolution of laparoscopy in colorectal surgery: An evidence-based review. World J Gastroenterol 20: 4926-4933.
34. Yeo HL, Isaacs AJ, Abelson JS, Milsom JW, Sedrakyan A (2016) Comparison of Open, Laparoscopic, and Robotic Colectomies Using a Large National Database: Outcomes and Trends Related to Surgery Center Volume. Dis Colon Rectum 59: 535-542.
35. Chung G, Hinoul P, Coplan P, Yoo A (2015) Trends in the diffusion of robotic surgery in prostate, uterus, and colorectal procedures: a retrospective population-based study. J Robot Surg 15: 275-291.
36. Damle A, Damle RN, Flahive JM, Schlussel AT, Davids JS, et al. (2017) Diffusion of technology: Trends in robotic-assisted colorectal surgery. Am J Surg 214: 820-824.