Testing Machine Learning Models to Predict Postoperative Ileus after Colorectal Surgery

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a paper that examines the effect of machine learning in trying to predict the development of ileus after surgery for colorectal cancer. It is from a well respected department of colorectal surgery with a large throughput of work. It is well written with good use of diagrams. It is a retrospective study with a reasonable number of patients. The authors suggest that machine learning might be a useful tool to predict post-operative ileus.

Could I make the following comments:

The authors acknowledge the retrospective nature of the study, however it would be sensible to detail how difficult it is to diagnose post-operative ileus in this situation. Although the authors have described what they did, this inevitably is a very limited method. These data therefore should be considered very preliminary. They could mention how the next study might be performed.

The incidence of post-operative ileus seems very low. Could the authors comment on this? Could this be a methodological issue? If this is truly the incidence, how valuable is this approach?

The authors state that the ability to predict the patients who would get an ileus would enable better prevention and treatment. However, although this is a good aspiration, is there strong evidence to suggest that ileus can be prevented? They have only examined patient related factors and have not examined anesthetic and post-operative treatments, which probably play a significant role in ileus.

Older age and obesity are mentioned as significant risk factors, however the median age and the median BMI of those who develop ileus appear to be very close to the median age  and BMI of the studied population (as in Fig 1). Is this correct and if so, are age and BMI of real relevance?

Thanks for allowing me to review the submission.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I would like to see a more extensive introduction explaining the importance of possible POI. Thou it is much more important to predict septic or anastomotic complications.
The introduction should also clarify the concept and principles of machine learning. As I, like many readers, do not work with artificial intelligence models on a daily basis, it is not clear from the introduction what the importance of using these models is and how it works.

In the description of the methodology, I would like to see a more detailed description of each of the models or its specific features. The criteria used to compare the usefulness or sensitivity of the models in predicting POIs are also unclear.

In the results I only understood the descriptive statistics tables
Table 4 is not understandable at all. I would like to see a comparison of all 8 models, but Table 4 only compares subsets of two models. What does F1 mean?
Why is it that Figure 4 only shows the results of 5 models?
The discussion has also not helped to understand the substance and nuances of this work.
The conclusions are entirely general and not supported in any way by the results presented.
The work may be understandable to mathematicians or statisticians, but it is completely beyond the reach of a doctor.

Author Response

Please see attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This retrospective study examined 316 adult patients who underwent colorectal surgery to explore patient co-morbidities contributing to postoperative ileus (POI) and compare machine learning (ML) model accuracy to existing risk instruments. Results showed that 6.33% of patients experienced POI, with age, BMI, gender, kidney disease, anemia, arrhythmia, rheumatoid arthritis, and NSQIP score identified as significant predictors. ML models, particularly AdaBoost and XGBoost tuned with grid search, demonstrated high accuracy (94.2% and 85.2%, respectively) in predicting POI risk. The study suggests that ML models offer a promising avenue for early detection and intervention in POI, potentially improving patient outcomes and reducing healthcare costs.

I think that this study is fascinating, but some points should be clarified to enhance the clarity and rigor of the study:

1.       the authors should consider moving the reporting of gender, age, and comorbidities to the Results section for better clarity and organization

2.       it's essential for the authors to clarify whether they performed a normality test for continuous variables. If the data are normally distributed, they can be expressed using mean and standard deviation (SD); otherwise, median and range should be used

3.       Figure 4 and the presentation of data in Table 1 could be improved. Figure 4 should be included in the Results section, and Table 1 should provide clearer information, specifying whether the numbers represent the frequency of patients with each type of comorbidity.

 

Addressing these points would enhance the clarity and transparency of the study and provide readers with a better understanding of the results and methods employed.

Author Response

Please see attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for responding to my comments. 

I am happy with the revised version of the paper

Author Response

Thank you for your time and constructive feedback. We appreciate your contribution to improve our manuscript. Sincerely- Authors

Reviewer 3 Report

Comments and Suggestions for Authors

Figure 1 should be in the results. The other comments are adressed

Author Response

Thank you for your time and constructive feedback. Our manuscript is much improved. Sincerely- Authors

Round 3

Reviewer 3 Report

Comments and Suggestions for Authors

all my points have been addressed. I have no other comments