Predicting the Survival of Primary Biliary Cholangitis Patients

Round 1

Reviewer 1 Report

The paper is well-written with some minor English-language edits. The methods are sound. The paper is of interest. 

I have one question that should be addressed. The authors assign liver transplant patients to patients that survive. They might just as well be assigned to patients that would die without a transplant. What is the sensitivity of the results to this assumption? 

Author Response

First and foremost, we would like to thank the reviewer for the time and effort spent in analyzing our study and providing valuable feedback.

COMMENT: I have one question that should be addressed. The authors assign liver transplant patients to patients that survive. They might just as well be assigned to patients that would die without a transplant. What is the sensitivity of the results to this assumption? 

ANSWER: In a first iteration, patients in need of a transplant were classified as surviving patients. However, as the reviewer mentioned, a more concrete analysis of these patients would be useful, because a patient who requires a transplant may die. As a result, we intend to conduct a more in-depth study in the future to distinguish between these states. We have updated the future work section of the current paper to highlight the need to conduct such analysis.

Reviewer 2 Report

In this paper, the authors compare different data mining classification techniques, using clinical and demographic data, with the aim of predicting whether or not a patient with primary biliary cholangitis will survive. Authors has addressed an interesting research problem.

Reviews for Authors

- Introduction needs to explain the main contributions of the work clearer.

- The novelty of this paper is not clear. The difference between present work and previous Works should be highlighted. Specifically in the 'Related work' section, there are no other works that use Machine Learning techniques for the evaluation of PBC. Are there no papers in the scientific literature?

- Have other feature selection techniques (e.g. Lasso regression) been evaluated in addition to those mentioned on page 9?

- What oversampling and undersampling methods were used?

- Remove italic to eq. 2, 3, 4, 5, 6, 7

- It would be useful to add a few more results e.g. confusion matrices (since they are mentioned on line 302) and also ROC Curves

Author Response

First and foremost, we would like to thank the reviewer for the time and effort spent in analyzing our study and providing valuable feedback.

COMMENT: Introduction needs to explain the main contributions of the work clearer.

ANSWER: We have included the main contributions of the study in the Introduction.

 

COMMENT: The novelty of this paper is not clear. The difference between present work and previous Works should be highlighted. Specifically, in the 'Related work' section, there are no other works that use Machine Learning techniques for the evaluation of PBC. Are there no papers in the scientific literature?

ANSWER: The novelty of the paper and the difference regarding previous works are highlighted in the last paragraph of the Related Work section:

“In a global manner, as it can be seen from the previous articles, there is more emphasis
in the literature on the analysis of treatments for PBC and not specifically on its prognosis
or evolution. Thus, more studies are needed to broaden the knowledge and understanding
on the diagnosis of this condition as early as possible [13]. Machine learning algorithms
can be used to perform early diagnosis and risk stratification, as well. Hence, developing
diagnostic algorithms for PBC patients based on demographic variables, symptomatology
and laboratory results is a resourceful tool for improving the quality of clinical practice.
We present a novel DM approach for PBC where the emphasis is the use of classified data
to predict the survival of patients diagnosed with PBC. Several experiments were carried
out, in which different DM techniques and feature selection setups were taken into account.
Such knowledge is particularly useful to perform early diagnosis and risk stratification,
thus improving the quality of clinical practice and lowering the mortality rate of patients
diagnosed with PBC.”

Besides, the authors did not found works with the exact same objective as our study and therefore it was not possible to make a direct comparison.

 

COMMENT: Have other feature selection techniques (e.g. Lasso regression) been evaluated in addition to those mentioned on page 9?

ANSWER: All the feature selection techniques used in the study were mentioned in the paper. However, some additional techniques, such as Lasso regression, could be used as future work. We have updated the Conclusion accordingly.

 

COMMENT: What oversampling and undersampling methods were used?

ANSWER: We have updated the manuscript to include the oversampling and undersampling methods used.

 

COMMENT: Remove italic to eq. 2, 3, 4, 5, 6, 7

ANSWER: The journal template defines a different font for mathematical equations, so we cannot change it as the equation formatting is defined by the journal template and not by the authors.

 

COMMENT: It would be useful to add a few more results e.g. confusion matrices (since they are mentioned on line 302) and also ROC Curves

ANSWER: The confusion matrices and ROC curves were not included in the document since it would be cumbersome to present them for the fourteen data mining models mentioned in the results. Hence, the authors decided to use metrics that derive from the confusion matrices and ROC curves as they offer detailed information in a intuitive way.

Reviewer 3 Report

The proposed paper aims to predict survival or non-survival of Primary Biliary Cholangitis patients utilizing machine learning-based classification approaches as well as the Cross Industry Standard Process for Data Mining methodology. The same goals using many kinds of machine learning and deep learning techniques have been proposed by previous studies. In order to complete the research findings and determine which machine learning-based classification method is best in predicting the survival or non-survival of Primary Biliary Cholangitis patients, it would be great if additional machine learning techniques were used, such as Logistic Regression, Multi-Layer Perceptron, K-Neural Networks, Neural Networks, Extreme Gradient Boosting, etc. Thus, for this paper, the reviewer suggests that the authors incorporate and make use of more machine learning techniques.

Author Response

First and foremost, we would like to thank the reviewer for the time and effort spent in analyzing our study and providing valuable feedback.

COMMENT: The proposed paper aims to predict survival or non-survival of Primary Biliary Cholangitis patients utilizing machine learning-based classification approaches as well as the Cross Industry Standard Process for Data Mining methodology. The same goals using many kinds of machine learning and deep learning techniques have been proposed by previous studies. In order to complete the research findings and determine which machine learning-based classification method is best in predicting the survival or non-survival of Primary Biliary Cholangitis patients, it would be great if additional machine learning techniques were used, such as Logistic Regression, Multi-Layer Perceptron, K-Neural Networks, Neural Networks, Extreme Gradient Boosting, etc. Thus, for this paper, the reviewer suggests that the authors incorporate and make use of more machine learning techniques.

ANSWER:

The novelty of the paper and the difference regarding previous works are highlighted in the last paragraph of the Related Work section:

“In a global manner, as it can be seen from the previous articles, there is more emphasis in the literature on the analysis of treatments for PBC and not specifically on its prognosis or evolution. Thus, more studies are needed to broaden the knowledge and understanding on the diagnosis of this condition as early as possible [13]. Machine learning algorithms can be used to perform early diagnosis and risk stratification, as well. Hence, developing diagnostic algorithms for PBC patients based on demographic variables, symptomatology and laboratory results is a resourceful tool for improving the quality of clinical practice. We present a novel DM approach for PBC where the emphasis is the use of classified data to predict the survival of patients diagnosed with PBC. Several experiments were carried out, in which different DM techniques and feature selection setups were taken into account. Such knowledge is particularly useful to perform early diagnosis and risk stratification, thus improving the quality of clinical practice and lowering the mortality rate of patients diagnosed with PBC.”

Since the authors did not found works with the exact same objective as the present study, it was not possible to make a direct comparison. However, we have updated the future work to include the reviewer’s suggestion of using additional machine learning techniques to conduct a more in-depth study.

Round 2

Reviewer 2 Report

Now the work is ok

Reviewer 3 Report

The authors have addressed previous comments by adding them into their future work studies.