Elsevier

Clinica Chimica Acta

Volume 487, December 2018, Pages 6-14
Clinica Chimica Acta

Diagnostic validation and interpretation of longitudinal circulating biomarkers using a biomarker response characteristic plot

https://doi.org/10.1016/j.cca.2018.09.015Get rights and content

Highlights

  • Clear guidance on the clinical usage of serum tumor biomarkers is often lacking.

  • Software was developed that supported BReC-plots and diagnostic accuracy studies.

  • The BReC plot application was demonstrated using 216 advanced lung cancer patients.

  • Obtained BReC-plots revealed relations between biomarker and clinical responses.

  • Medical tests could be designed that met pre-specified performance criteria.

Abstract

Background

Serum-based tumor biomarkers are used to monitor cancer treatment, while clear guidance on the clinical usage is often lacking. We describe a graphical presentation to support diagnostic accuracy studies and clinical interpretation of longitudinal biomarker data.

Methods

A biomarker response characteristic (BReC) plot was designed. To allow demonstration of the BReC plot application, software was developed that supported 1) dynamic generation of BReC plots, and 2) diagnostic accuracy studies of biomarker response-based medical tests. The BReC plot application was demonstrated using serial carcinoembryonic antigen (CEA) and Cyfra 21.1 results from 216 patients with metastasized non-small cell lung cancer, treated with Nivolumab in routine clinical practice.

Results

The developed software supported the generation of BReC plots and diagnostic validation of biomarker response-based medical tests by generating the sensitivity, specificity and predictive values. Obtained BReC plots showed a clear relationship between clinical outcome and CEA and Cyfra 21.1 responses. Furthermore, using BReC plots, CEA and Cyfra 21.1 based medical tests were designed with a sensitivity for detection of treatment failure of 0.34 and 0.35 and a specificity of 0.96.

Conclusions

The BReC plot appears to support diagnostic validation studies and the interpretation of longitudinal biomarkers though further validation is warranted.

Introduction

In medical oncology, serial analysis of tumor biomarkers is used to provide an early indication of changes in tumor burden [1]. Although several serum-based tumor biomarkers are available and used in clinical practice for follow-up purposes [[2], [3], [4]], clinical interpretation of individual patient results remains challenging. The observed tumor biomarker dynamics depends on various variables, including: i) biomarker half-life, ii) therapeutic intervention, iii) analytical variation of the assay, iv) pre-analytical variations, v) biological variations [5], vi) other not-tumor-related processes such as renal or liver failure [6], and vii) tumor dynamics and heterogeneity. To estimate the relevance of two successive biomarker results the reference change value, based on analytical variation and biological variation determined in healthy controls, is often recommended [1,5]; however, it is uncertain whether this value reflects all previously established variables relevant for the interpretation of consecutive tumor biomarkers. As a result clear guidance regarding the clinical meaning of consecutive tumor biomarker results is lacking for many tumor biomarkers available and used in clinical practice.

Longitudinal circulating tumor biomarkers are most often used for the follow-up of cancer treatment in order to “diagnose” response or absence of response to (systemic) treatment or “diagnose” recurrent disease after (curative) treatment. There are several challenges related to the process of diagnostic validation of longitudinal biomarkers. The first is what kind of metric to use to describe the ‘pattern’ of consecutive biomarker results and what method to use to describe results obtained over time. Approaches used to describe and validate the longitudinal biomarker response include (or are based on): logical and criteria-based rules [7,8], doubling time [9], kinetics [10], population pharmacodynamics modeling [11], random-effect models [12], and latent class growth curve modeling [12]. A second challenge is to select relevant time points for the biomarker results and the clinical reference standard. Since time intervals at which biochemical, radiological or clinical responses occur may differ [13,14], improper selection of the clinical reference time point (e.g. at the same time as the biomarker sampling) might conceal the true diagnostic properties of a biomarker. Also, methodological characteristics (e.g. study design, patient selection and populations, quality of clinical reference standards) can affect the quality of diagnostic accuracy studies [15]. All these issues complicate the diagnostic validation of longitudinal biomarkers.

To support the diagnostic validation and clinical interpretation of longitudinal (tumor) biomarkers, we present a graphical tool that relates biomarker responses to clinical reference standards later in time, i.e. the biomarker response characteristic (BReC) plot. This descriptive graphical presentation is suggested to support biomarker response based medical test design, modeling of longitudinal data and the clinical interpretation of biomarker responses. In order to be able to demonstrate its potential use, software was developed that supported the dynamic and flexible generation of BReC plots and diagnostic validation of biomarker response based medical tests. Furthermore, metastatic non-small cell lung cancer patients treated with Nivolumab immune checkpoint therapy in routine clinical practice and regularly monitored using carcinoembryonic antigen (CEA) and Cyfra 21.1, was used as illustrative patient cohort. Studies to investigate these tumor biomarkers as early response assessment tools for these patients are subject of future research.

Section snippets

Patient and laboratory data

The BReC plot application was demonstrated using tumor biomarker data obtained from 216 patients with metastatic non-small cell lung cancer treated with Nivolumab in routine practice [16]. These patients were monitored every other week for a panel of tumor biomarkers, including CEA and Cyfra 21.1, measured on a Roche cobas 6000 system. Furthermore, clinical status was scored every 3 months after start of therapy that could result in the following responses, i.e. based on radiological

Graphical interfaces for generating BReC plots and performing diagnostic accuracy studies

The interface used to generate the BReC plots is presented in Fig. 2. Subsequently, the biomarker settings can be selected, i.e. biomarker, baseline time interval, follow-up time interval, minimum value selection (noise threshold), and a smoother that determines the bin size (biomarker response interval expressed in ∆% units). Then, the clinical response settings can be selected, i.e. the clinical responses of interest (response selection) and selection of the response time. The interface used

Discussion

A BReC plot was developed as a visual tool to support the diagnostic validation and interpretation of longitudinal (tumor) biomarkers. Together with the IT infrastructure (supporting the dynamic generation of BReC plots), several variables (e.g. baseline period, follow-up period, and effect of biomarker thresholds) were examined for multiple biomarkers. The diagnostic properties of simple multiple biomarker response-based tests could also be investigated.

A simplified way of describing the tumor

References (20)

There are more references available in the full text version of this article.

Cited by (0)

View full text