Computational prediction of diagnosis and feature selection on mesothelioma patient health records

Our plain-English summary of Computational prediction of diagnosis and feature selection on mesothelioma patient health records, by D. Chicco, C. Rovelli (PloS One, 2019).

The study

• Authors: D. Chicco, C. Rovelli
• Journal: PloS One
• Published: 2019;14(1):e0208737
• Cited by: 30 articles, as counted by NIH iCite
• Read it: PubMed · Publisher · Free full text

Abstract

Background: Mesothelioma is a lung cancer that kills thousands of people worldwide annually, especially those with exposure to asbestos. Diagnosis of mesothelioma in patients often requires time-consuming imaging techniques and biopsies. Machine learning can provide for a more effective, cheaper, and faster patient diagnosis and feature selection from clinical data in patient records.

Methods And Findings: We analyzed a dataset of health records of 324 patients having mesothelioma symptoms from Turkey. The patients had prior asbestos exposure and displayed symptoms consistent with mesothelioma. We compared probabilistic neural network, perceptron-based neural network, random forest, one rule, and decision tree classifiers to predict diagnosis of the patient records. We measured classifiers' performance through standard confusion matrix scores such as Matthews correlation coefficient (MCC). Random forest outperformed all models tried, obtaining MCC = +0.37 on the complete imbalanced dataset and MCC = +0.64 on the under-sampled balanced dataset. We then employed random forest feature selection to identify the two most relevant dataset traits associated with mesothelioma: lung side and platelet count. These two risk factors resulted so predictive, that decision tree focusing on them achieved the second top accuracy on the complete dataset diagnosis prediction (MCC = +0.28), outperforming all other methods and even decision tree itself applied to all features.

Conclusions: Our results show that machine learning can predict diagnoses of patients having mesothelioma symptoms with high accuracy, sensitivity, and specificity, in few minutes. Additionally, random forest can efficiently select the most important features of this clinical dataset (lung side and platelet count) in few seconds. The importance of pleural plaques in lung sides and blood platelets in mesothelioma diagnosis indicates that physicians should focus on these two features when reading records of patients with mesothelioma symptoms. Moreover, doctors can exploit our machinery to predict patient diagnosis when only lung side and platelet data are available.

In plain English

This machine-learning study reanalyzed an existing public dataset of health records from 324 patients with possible mesothelioma symptoms, collected at Dicle University Faculty of Medicine Hospital in Diyarbakir, southeastern Turkey, before October 2011. Of these, 96 had mesothelioma and 228 did not. The authors describe the region's soil as containing natural asbestos fibers, mostly tremolite but also chrysotile. They first removed a data field that duplicated the diagnosis, then compared several computer methods for predicting which patients had mesothelioma. On the full dataset the best method, random forest, scored a Matthews correlation coefficient of +0.37 (where +1 is perfect), but it correctly identified only 28% of mesothelioma patients (sensitivity 0.28). On a balanced subset of 192 patients it scored +0.64. The two most informative fields were 'lung side' and platelet count. A model using only these two scored +0.28 on the full dataset. Duration of asbestos exposure ranked sixth of 33 fields. In a conventional regression analysis, only lung side remained significant. The authors concluded that physicians should focus on these two features, and cautioned that their approach might not generalize.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Computational (machine-learning) analysis of an existing, publicly available dataset of patient health records, with a supplementary conventional regression analysis
• Who was studied: 324 patients with potential mesothelioma symptoms whose records were collected at Dicle University Faculty of Medicine Hospital (Diyarbakir, southeastern Turkey) before October 2011. 96 had mesothelioma and 228 did not.
• Exposure: Environmental asbestos in the Diyarbakır region, which the authors describe as natural soil fibers, mostly tremolite but also chrysotile. The dataset recorded asbestos exposure (yes/no) and duration of asbestos exposure (0 to 70 years). No occupational history was available.
• Main results:
  • The authors excluded the 'diagnosis method' field because it duplicated the diagnosis. With that field included, random forest reached a perfect MCC of +1.00.
  • On the complete dataset of 324 patients, random forest gave the best overall score: MCC +0.37, sensitivity 0.28 (true positive rate) and specificity 0.97 (true negative rate).
  • On the under-sampled balanced subset of 192 patients (96 with and 96 without mesothelioma), random forest scored MCC +0.64, with sensitivity 0.75 and specificity 0.86.
  • On the balanced subset, the perceptron neural network had the highest sensitivity (0.95) but the lowest specificity (0.20).
  • Random forest feature selection ranked 'lung side' and 'platelet count' as the two most predictive of the 33 features.
  • A decision tree using only lung side and platelet count scored MCC +0.28 on the complete dataset.
  • 22 patients had pleural plaques on both sides. Of these, 16 (72.72%) had mesothelioma.
  • 42 patients had a platelet count below 150k per microliter. Of these, 23 (54.76%) had mesothelioma.
  • Duration of asbestos exposure ranked sixth of 33 features.
  • In the regression analysis, univariate models identified age, gender and lung side as statistically different between patients with and without mesothelioma. In multivariate analysis, only lung side remained significant.
• The authors' conclusion: The authors concluded that random forest outperformed the other methods. They also concluded that the importance of lung side and platelet count suggests physicians should focus on these two features when reading a patient's health record.
• Limitations the authors state:
  • The approach might not generalize well in the mesothelioma context because some features are specific to this dataset (for example, 'city').
  • Feature selection relied only on random forest, so its results might be biased towards that algorithm.
  • Under-sampling discarded data, so the balanced-dataset results cover only 192 of the 324 patients.
  • The dataset was too small to set aside an additional held-out final test set.
  • No occupational history was available for the patients in the dataset.

Topics

The National Library of Medicine indexes this study under subjects our site covers: Mesothelioma.

Medical terms in this study, explained in our Lexicon: Mesothelioma, Lung Cancer, Asbestos Exposure, Pleural Plaques.

Related studies

• The use of a next-generation sequencing-derived machine-learning risk-prediction model (OncoCast-MPM) for malignant pleural mesothelioma: a retrospective study — M.G. Zauderer, A. Martin, et al. (2021)
• Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study — M. Yakar, D. Etiz, et al. (2021)
• A computational study of the interactions between anthocyans and cyclodextrins — R. Pop, A. Căta, et al. (2020)
• Enduring health effects of asbestos use in Belgian industries: a record-linked cohort study of cause-specific mortality (2001-2009) — L. Van den Borre, P. Deboosere (2015)
• The "Brescia panel" (Claudin-4 and BRCA-associated protein 1) in the differential diagnosis of mesotheliomas with epithelioid features versus metastatic carcinomas — L. Bernardi, T. Bizzarro, et al. (2021)
• Small orangiophilic squamous-like cells: an underrecognized and useful morphological feature for the diagnosis of malignant mesothelioma in pleural effusion cytology — L. Chen, S.G. Caldero, et al. (2014)

Browse more research in PloS One, or return to medical research.