Identifying predictive factors for neuropathic pain after breast cancer surgery using machine learning

https://doi.org/10.1016/j.ijmedinf.2020.104170Get rights and content

Abstract

Introduction

Neuropathic pain (NP) remains a major debilitating condition affecting more than 26% of breast cancer survivors worldwide. NP is diagnosed using a validated 10-items Douleur Neuropathique - 4 screening questionnaire which is administered 3 months after surgery and requires patient-doctor interaction. To develop an effective prognosis model admissible soon after surgery, without the need for patient-doctor interaction, we sought to [1] identify specific pain characteristics that can help determine which patients may be susceptible to NP after BC surgery, and 2) assess the utility of machine learning models developed in objective [1] as a knowledge discovery tool for downstream analysis.

Methods

The dataset is from a prospective cohort study of female patients scheduled to undergo breast cancer surgery for the first time at the Jewish General Hospital, Montreal, Canada between November 2014 and March 2019. NP was assessed at 3 months after surgery using Douleur Neuropathique – 4 interview scores (in short, DN4-interview; range: 0–7). For the primary analysis, we constructed six ML algorithms (least square, ridge, elastic net, random forest, gradient boosting, and neural net) to identify the most relevant predictors for DN4-interview score; and compared model performance based on root mean square error (RMSE). For the secondary analysis, we built a logistic classification model for neuropathic pain (DN4-interview score ≥ 3 versus DN4-interview score < 3) using the relevant-consensus-predictors from the primary analysis.

Results

Anxiety, type of surgery, preoperative baseline pain and acute pain on movement were identified as the most relevant predictors for DN4 - interview score. The least square regression model (RMSE = 1.43) is comparable in performance with random forest (RMSE = 1.39) and neural network model (RMSE = 1.50). The Gradient boosting model (RMSE = 1.16) outperformed the models compared including the penalized regression models (ridge regressions, RMSE = 1.28; and elastic net, RMSE = 1.31). In the secondary analysis, the preferred logistic regression classier for NP had an area under the curve (AUC) of 0.68 (95% CI = 0.57 to 0.79). Anxiety was significantly associated with the likelihood of NP (odds ratio = 2.18; 95% CI = 1.05–4.49). In comparison to their counterparts, the odds of NP were higher in participants with acute pain on movement or with present preoperative baseline pain or participants who performed total mastectomy surgery, but the differences were not statistically significant.

Conclusions

Modern machine learning models show improvements over traditional least square regression in predicting of DN4-interview score. Penalized regression methods and the Gradient boosting model out-perform other models. As a predictor discovery tool, machine learning algorithms identify relevant predictors for DN4-interview score that remain statistically significant indicators of neuropathic pain in the classification model. Anxiety, type of surgery and acute pain on movement remain the most useful predictors for neuropathic pain.

Introduction

Neuropathic pain (NP) is a severe and debilitating disorder of the peripheral and central nervous systems. It is defined as “pain arising from a direct consequence of a lesion or disease affecting the somatosensory system [1,2]. The affected nerve lesions trigger molecular changes in nociceptive neurons that become hypersensitive, resulting in spontaneous pain, shooting pain sensations and sympathetically maintained pain [[3], [4], [5]]. In cancer patients, for example, NP could be caused by the invasiveness of the tumor, radiotherapy and/or side effect of chemotherapy or surgery [6].

NP is highly prevalent and estimated to affect approximately 7–10% of the general population worldwide [7,8]. The condition is most common among women above the age of 45 and implicates both physical and psychological co-morbidities. Despite increased interest in neuropathic pain, treatment remains highly unsatisfactory [9]. Instead of comprehensive pain relief, current treatment efforts are symptom-specific, geared towards pain relief from specific conditions and intended for pain management [10].

Recent advances in medicine have resulted in exceptional improvements in Breast Cancer (BC) survival rates. However, NP remains a major debilitating symptom in BC patients and up to 26% of cancer survivors experience the condition [11]. Patients are typically under strong medication and increased hospital visits after BC surgery. This can result in impaired quality of sleep, loss of function, anxiety, depression, impaired cognition and/or overall life [7]. Thus, there is a significant interest in understanding the epidemiology of NP in breast cancer survivors.

Several patient-reported screening tools are available for diagnosing NP in health care settings including Leeds Assessment of Neuropathic Symptoms and Signs (LANSS) [12], the Neuropathic Pain Questionnaire (NPQ) and the Douleur Neuropathique - 4 (short form, DN4) Screening Questionnaire among others [13]. The 10-item standard DN4 screening questionnaire, which was introduced in French, is particularly popular and has been validated in many different languages [[14], [15], [16], [17]]; it evaluates pain following central and/or peripheral lesions, and allows practitioners to determine if a pain condition is neuropathic. However, the utility of the standard 10-item DN4 questionnaire poses two major challenges for early detection of potential NP development: 1) A lapse period of >3 months after BC surgery is required to administer the DN4 questionnaire, and 2) the questionnaire requires doctor-patient interactions to complete questions related to patient examination. Hence, the development of an effective and easy-to-use prognosis tool, that doesn’t require patient-doctor interaction, and can be administered promptly post-surgery would greatly improve both the management and treatment of NP.

Clinical prediction models play a fundamental role in aiding decision-making and are constantly used in various clinical settings and practices ranging from risk stratification in intensive care units to medical image diagnostics [18]. The etiological complexity of NP after a major surgery like BC surgery presents an interesting challenge in clinical care [[19], [20], [21]]. The lack of a universally accepted and validated clinical diagnostic tool for NP makes it difficult for clinicians and policymakers to provide precise estimates of NP prevalence. Traditionally, regression models (e.g. Least square regression) have been used to identify significant predictors of binary or continuous clinical outcomes in pain studies [22,23]; however, such models do not perform well in capturing complex relationships including interactions and are not usually robust to multicollinearity [24,25], which is a fundamental feature of pain-related variables.

The objectives of the current study include 1) To examine the performance of different machine learning algorithms as predictive models for DN4-interview score (referred to DN4 score henceforth) after breast cancer surgery; 2) To use the models as a knowledge discovery and variable exploration tool in order to identify the relevant predictors for DN4 score; 3) To include the relevant predictors as covariates in the classification model for NP while leveraging residual data variability from a select component of multiple correspondence analysis of the remaining predictors.

Section snippets

Data source

The data used in this study is obtained from a 3-months prospective cohort study of breast cancer patients seen at the Segal Cancer Centre, Jewish General Hospital (JGH), Montreal, Quebec, Canada from November 2014 to March 2019. The study was approved by the Research Ethics Committee of the JGH. All the potential participants signed a written consent form prior to their inclusion.

Eligible participants included in the study were adult female patients diagnosed with breast cancer and scheduled

Results

Distributions of the covariates are presented in Table 1. The analysis included 204 eligible female participants of mean age 56.16 (SD = 14.26) years at baseline. Approximately 33% of the participants reported pre-surgical baseline pain with a mean pain intensity score of 7.46 (SD = 15.68). At baseline, 45.6% of the participant reported experiencing moderate to severe anxiety and 38.7% reported experiencing moderate to severe depression. Other comorbidities reported were diabetes (10.82%),

Discussion

The study assessed the performance of multiple machine learning models to identify predictive factors for neuropathic pain after breast cancer surgery (i.e. six models for DN4-interview score and two models for neuropathic pain classification). The models predict neuropathic pain score using self-assessed patient questionnaires and relevant covariates collected at most seven days after surgery without the traditional need for doctor-patient interaction at 3 months after surgery.

In predicting

Conclusions

In conclusion, the current study has identified relevant predictors for DN4-interview score based on six different machine learning methods. In order of relevance, anxiety, acute pain on movement, type of surgery and preoperative baseline pain were the most relevant predictors. Our work has demonstrated a general agreement between the constructed machine learning models, with the penalized regression models and tree-based models seemingly out-performing other prediction models. In the logistic

Authorship statement

All persons who meet authorship criteria are listed as authors, and all authors certify that they have participated sufficiently in the work to take public responsibility for the content, including participation in the concept, design, analysis, writing, or revision of the manuscript.

Financial disclosure of all authors

LJ received funding from Mitacs. PS-C’s research is supported by Natural Sciences and Engineering Research Council of Canada (RGPIN-2017−06100), and Fonds de la recherche en santé du Québec (Salary Award). NA, MG, and AMV have no financial disclosures to report.

Summary points

What was already known?

  • Neuropathic pain is diagnosed at 3 months after breast cancer (BC) surgery (with the aid of a primary care physician) using a validated DN4 questionnaire.

  • Previous studies have shown neuropathic pain

CRediT authorship contribution statement

Lamin Juwara: Formal analysis, Writing - original draft, Writing - review & editing. Navpreet Arora: Data curation, Writing - original draft, Writing - review & editing. Mervyn Gornitsky: Writing - original draft, Writing - review & editing. Paramita Saha-Chaudhuri: Supervising analysis, Writing - original draft, Writing - review & editing. Ana M. Velly: Writing - original draft, Writing - review & editing.

Declaration of Competing Interest

Authors declare no conflict of interest.

Acknowledgments

We thank Harsimrat Kaur, Gurveen Gill and Neha Aggarwal (Department of Dentistry, Jewish General Hospital) for their assistance with patient recruitment and data collection; Jennifer Baoteng for her thoughtful comments; and the anonymous reviewers for their insightful recommendations that has considerably improved our manuscript.

References (44)

  • M. Gevrey et al.

    Review and comparison of methods to study the contribution of variables in artificial neural network models

    Ecol. Model.

    (2003)
  • B. Emir et al.

    (414) Predictors of response to pregabalin for broad neuropathic pain: results from 11 machine learning methods from a 6-week German observational study

    J. Pain

    (2016)
  • D.E. Moulin et al.

    Pharmacological management of chronic neuropathic pain - consensus statement and guidelines from the Canadian Pain Society

    Pain Res. Manag.

    (2007)
  • A. Binder et al.

    Sodium channels in neuropathic pain--friend or foe?

    Nat. Clin. Pract. Neurol.

    (2007)
  • L. Colloca et al.

    Neuropathic pain

    Nat. Rev. Dis. Primer.

    (2017)
  • D. Bouhassira et al.

    Prevalence and incidence of chronic pain with or without neuropathic characteristics in patients with cancer

    Pain

    (2017)
  • Chronic Pain Has Arrived in the ICD-11 - IASP [Internet]

    (2019)
  • M. Bennett

    The LANSS pain scale: the Leeds assessment of neuropathic symptoms and signs

    Pain.

    (2020)
  • P. Sykioti et al.

    Validation of the greek version of the DN4 diagnostic questionnaire for neuropathic pain

    Pain Pract. Off. J. World Inst. Pain

    (2015)
  • H.-J. Kim et al.

    Validation of the korean version of the DN4 diagnostic questionnaire for neuropathic pain in patients with lumbar or lumbar-radicular pain

    Yonsei Med. J.

    (2016)
  • S.P. Madani et al.

    Validity and reliability of the persian (Farsi) version of the DN 4 (Douleur neuropathique 4 questions) questionnaire for differential diagnosis of neuropathic from non-neuropathic pains

    Pain Pract.

    (2014)
  • E.W. Steyerberg et al.

    Assessing the performance of prediction models: a framework for some traditional and novel measures

    Epidemiol. Camb. Mass

    (2010)
  • Cited by (22)

    • Artificial intelligence in anesthesiology

      2023, Artificial Intelligence in Clinical Practice: How AI Technologies Impact Medical Research and Clinics
    • The Peripheral Nerve Surgeon's Role in the Management of Neuropathic Pain

      2023, Plastic and Reconstructive Surgery - Global Open
    View all citing articles on Scopus
    View full text