Elsevier

European Journal of Cancer

Volume 153, August 2021, Pages 179-189
European Journal of Cancer

Original Research
Clinical decision support algorithm based on machine learning to assess the clinical response to anti–programmed death-1 therapy in patients with non–small-cell lung cancer

https://doi.org/10.1016/j.ejca.2021.05.019Get rights and content

Highlights

  • Programmed death-ligand 1 expression alone may not reflect the response to programmed cell death protein 1 (PD-1) inhibitors.

  • Various clinical characteristics are related to the anti–PD-1 response.

  • We established a machine learning–based algorithm to predict the anti–PD-1 response.

Abstract

Objective

Anti–programmed death (PD)-1 therapy confers sustainable clinical benefits for patients with non–small-cell lung cancer (NSCLC), but only some patients respond to the treatment. Various clinical characteristics, including the PD-ligand 1 (PD-L1) level, are related to the anti–PD-1 response; however, none of these can independently serve as predictive biomarkers. Herein, we established a machine learning (ML)–based clinical decision support algorithm to predict the anti–PD-1 response by comprehensively combining the clinical information.

Materials and methods

We collected clinical data, including patient characteristics, mutations and laboratory findings, from the electronic medical records of 142 patients with NSCLC treated with anti–PD-1 therapy; these were analysed for the clinical outcome as the discovery set. Nineteen clinically meaningful features were used in supervised ML algorithms, including LightGBM, XGBoost, multilayer neural network, ridge regression and linear discriminant analysis, to predict anti–PD-1 responses. Based on each ML algorithm's prediction performance, the optimal ML was selected and validated in an independent validation set of PD-1 inhibitor–treated patients.

Results

Several factors, including PD-L1 expression, tumour burden and neutrophil-to-lymphocyte ratio, could independently predict the anti–PD-1 response in the discovery set. ML platforms based on the LightGBM algorithm using 19 clinical features showed more significant prediction performance (area under the curve [AUC] 0.788) than on individual clinical features and traditional multivariate logistic regression (AUC 0.759).

Conclusion

Collectively, our LightGBM algorithm offers a clinical decision support model to predict the anti–PD-1 response in patients with NSCLC.

Introduction

Immune checkpoint inhibitors (ICIs), including programmed cell death protein 1/programmed death-ligand 1 (PD-1/PD-L1) inhibitors, have resulted in prolonged survival and were approved as first- and second-line therapies in patients with recurrent/metastatic non–small-cell lung cancer (NSCLC) [1,2]. PD-L1 protein expression has been approved as a biomarker for investigating the efficacy of PD-1/PD-L1 inhibitors. PD-L1 expression is enriched in anti–PD-1/PD-L1 inhibitor therapy responders [3,4]. Nevertheless, fewer than 30% of patients with NSCLC respond to anti–PD-1 inhibitors [5]. Moreover, a substantial proportion of PD-L1–positive patients show no response to therapy, whereas a subset of PD-L1–negative patients do show. Therefore, PD-L1 expression alone may not comprehensively reflect the complexity of the tumour microenvironment and response to PD-1 inhibitors.

With increasing use of anti–PD-1 therapy, various clinical characteristics related to treatment response in patients with NSCLC have been identified, such as neutrophil/lymphocyte ratio before immunotherapy, smoking history, performance status, sex, the presence of metastases and driver mutations and pathology [[6], [7], [8], [9], [10], [11], [12], [13], [14]]. These factors are routinely examined in clinical practice before anti–PD-1 therapy and are collected in electronic medical records (EMRs). However, none of the clinical factors can accurately predict the response to anti–PD-1 therapy; thus, a model integrating these factors is needed.

Recently, machine learning (ML)–based methods have been developed to predict disease progression and treatment response in various diseases [15,16]. ML tools can identify key features from complex data sets associated with a specific purpose and interest. The ML techniques, multilayer neural network (MNN), ensemble learning, support vector machine (SVM) and penalised regression, have been widely used in recent studies for developing predictive models to facilitate effective and accurate decision-making [8,17,18]. However, ML has not yet been extensively studied or used in ICI-treated patients with cancer.

This study explored and validated various predictive algorithms using an ML approach based on the clinicopathological factors of anti‒PD-1 therapy‒treated patients with NSCLC from prospectively collected data obtained from EMRs. We aimed to establish a clinical decision support system for prescribing anti‒PD-1 therapy in patients with NSCLC using clinicopathological information routinely collected during clinical practice.

Section snippets

Patients

Patients with histologically confirmed stage IV NSCLC who were treated with anti‒PD-1 therapy (nivolumab 2 mg/kg every 2 weeks or pembrolizumab 200 mg fixed dose every 3 weeks) at Yonsei Cancer Center (Seoul, Korea) between March 2014 and April 2020 were included (n = 192). Patients were divided into a discovery cohort—patients consecutively enrolled from March 2014 to January 2018—to explore the optimal algorithm and the validation set—patients enrolled after January 2018—to confirm the

Patient characteristics

In total, 192 patients with advanced NSCLC treated with anti‒PD-1 therapy were included for analysis. The discovery (n = 142) and validation cohorts (n = 50) showed no significant differences in patient characteristics, including age, performance status, driver mutation status, line of therapy and site of baseline metastasis (Table 1).

Treatment outcomes in the discovery set and role of PD-L1 as a predictive biomarker

In the discovery set, 56 responders and 86 non-responders were classified based on definitions. The clinical response of total patients is shown in Fig. 1A. The

Discussion

Immunohistochemistry(IHC)-based PD-L1 expression is the only approved predictive marker for anti‒PD-1 therapy, but its accuracy is not sufficient to discriminate the responders from non-responders. The associations of various clinical factors or blood test values in routine practice with treatment outcomes after anti‒PD-1 therapy are increasingly being reported. However, the practical clinical application of these findings is limited because these clinical factors and blood test results have

Authors’ contributions

Beung-Chul Ahn: Conceptualisation, Methodology, Resources, Investigation, Formal analysis, Roles/Writing - original draft, Writing - review & editing Jea-Woo So: Conceptualisation, Methodology, Resources, Investigation, Formal analysis, Roles/Writing - original draft, Writing - review & editing Byoung Chul Cho: Conceptualisation, Methodology, Funding acquisition, Project administration, Resources, Roles/Writing - original draft, Writing - review & editing Hye Ryun Kim: Conceptualisation,

Availability of data and material

All data generated or analysed during this study are included in this published article and its supplementary information files.

Conflict of interest statement

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

This research was supported by the Bio & Medical Technology Development Program of the National Research Foundation funded by the Ministry of Science and ICT [grant numbers NRF-2019M3A9B6065231, 2017M3A9E8029717, 2017M3A9E9072669]. This study was supported by a Dongin Sports research grant of Yonsei University College of Medicine (6-2019-0128).

References (29)

  • B. Liu

    Recent development in clinical applications of PD-1 and PD-L1 antibodies for cancer immunotherapy

    J Hematol Oncol

    (2017)
  • E.B. Garon

    Pembrolizumab for the treatment of non-small-cell lung cancer

    N Engl J Med

    (2015)
  • B. Li

    Impact of smoking on efficacy of PD-1/PD-L1 inhibitors in non-small cell lung cancer patients: a meta-analysis

    Onco Targets Ther

    (2018)
  • F. Pantano

    Prognostic clinical factors in patients affected by non-small-cell lung cancer receiving Nivolumab

    Expet Opin Biol Ther

    (2020)
  • Cited by (17)

    View all citing articles on Scopus
    1

    These authors contributed equally.

    View full text