Forskning
Udskriv Udskriv
Switch language
Region Hovedstaden - en del af Københavns Universitetshospital
Udgivet

Evaluation of algorithm development approaches: Development of biomarker panels for early detection of colorectal lesions

Publikation: Bidrag til tidsskriftTidsskriftartikelForskningpeer review

Harvard

APA

CBE

MLA

Vancouver

Author

Bibtex

@article{03f177a86f8149ef84462ba79bf8c803,
title = "Evaluation of algorithm development approaches: Development of biomarker panels for early detection of colorectal lesions",
abstract = "INTRODUCTION: Colorectal cancer (CRC) is the third most common cancer in the U.S. Early detection of CRC can substantially increase survival rates. Test compliance may be improved by offering a blood-based test option.METHODS: Endoscopy II trial specimens were tested for AFP, CA19-9, CEA, hs-CRP, CyFra 21-1, Ferritin, Galectin-3, and TIMP-1 levels. These biomarkers, as well as patient demographic information (e.g., age, gender), were included in algorithm development. Six statistical methods were utilized to develop algorithms that would discriminate cancer vs. noncancers. Statistical methods included logistic regression, adaptive index modeling, partial least-squares discriminant analysis, feature vector (weighted and unweighted), and random forest. The performance of these algorithms was compared against benchmark criteria established for stool-based tests.RESULTS: Using several statistical methods, the presence of CRC and high-risk adenomas was detected with an AUCs of at least 0.65-0.76, with a few of models approaching the stool-based tests benchmark performance. Further, common markers were utilized across the different statistical techniques, with model complexities ranging from 3 to 9 markers.CONCLUSIONS: Predictive models identified subjects with CRC and high-risk adenomas with the similar levels of statistical accuracy. Clinical performance differences were minimal across the statistical techniques, although the intuitive interpretations, model complexity, clinical adoption and implementation varied.",
keywords = "Algorithm methodologies, Biomarkers, Colon cancer, Early detection",
author = "Susan Gawel and Michael Lucht and Heather Gomer and Patrick Treado and Christensen, {Ib J} and Nielsen, {Hans J} and Davis, {Gerard J} and {Danish Research Group on Early Detection of Colorectal Cancer}",
note = "Copyright {\circledC} 2019 Elsevier B.V. All rights reserved.",
year = "2019",
month = "11",
doi = "10.1016/j.cca.2019.08.007",
language = "English",
volume = "498",
pages = "108--115",
journal = "Clinica Chimica Acta",
issn = "0009-8981",
publisher = "Elsevier BV",

}

RIS

TY - JOUR

T1 - Evaluation of algorithm development approaches

T2 - Development of biomarker panels for early detection of colorectal lesions

AU - Gawel, Susan

AU - Lucht, Michael

AU - Gomer, Heather

AU - Treado, Patrick

AU - Christensen, Ib J

AU - Nielsen, Hans J

AU - Davis, Gerard J

AU - Danish Research Group on Early Detection of Colorectal Cancer

N1 - Copyright © 2019 Elsevier B.V. All rights reserved.

PY - 2019/11

Y1 - 2019/11

N2 - INTRODUCTION: Colorectal cancer (CRC) is the third most common cancer in the U.S. Early detection of CRC can substantially increase survival rates. Test compliance may be improved by offering a blood-based test option.METHODS: Endoscopy II trial specimens were tested for AFP, CA19-9, CEA, hs-CRP, CyFra 21-1, Ferritin, Galectin-3, and TIMP-1 levels. These biomarkers, as well as patient demographic information (e.g., age, gender), were included in algorithm development. Six statistical methods were utilized to develop algorithms that would discriminate cancer vs. noncancers. Statistical methods included logistic regression, adaptive index modeling, partial least-squares discriminant analysis, feature vector (weighted and unweighted), and random forest. The performance of these algorithms was compared against benchmark criteria established for stool-based tests.RESULTS: Using several statistical methods, the presence of CRC and high-risk adenomas was detected with an AUCs of at least 0.65-0.76, with a few of models approaching the stool-based tests benchmark performance. Further, common markers were utilized across the different statistical techniques, with model complexities ranging from 3 to 9 markers.CONCLUSIONS: Predictive models identified subjects with CRC and high-risk adenomas with the similar levels of statistical accuracy. Clinical performance differences were minimal across the statistical techniques, although the intuitive interpretations, model complexity, clinical adoption and implementation varied.

AB - INTRODUCTION: Colorectal cancer (CRC) is the third most common cancer in the U.S. Early detection of CRC can substantially increase survival rates. Test compliance may be improved by offering a blood-based test option.METHODS: Endoscopy II trial specimens were tested for AFP, CA19-9, CEA, hs-CRP, CyFra 21-1, Ferritin, Galectin-3, and TIMP-1 levels. These biomarkers, as well as patient demographic information (e.g., age, gender), were included in algorithm development. Six statistical methods were utilized to develop algorithms that would discriminate cancer vs. noncancers. Statistical methods included logistic regression, adaptive index modeling, partial least-squares discriminant analysis, feature vector (weighted and unweighted), and random forest. The performance of these algorithms was compared against benchmark criteria established for stool-based tests.RESULTS: Using several statistical methods, the presence of CRC and high-risk adenomas was detected with an AUCs of at least 0.65-0.76, with a few of models approaching the stool-based tests benchmark performance. Further, common markers were utilized across the different statistical techniques, with model complexities ranging from 3 to 9 markers.CONCLUSIONS: Predictive models identified subjects with CRC and high-risk adenomas with the similar levels of statistical accuracy. Clinical performance differences were minimal across the statistical techniques, although the intuitive interpretations, model complexity, clinical adoption and implementation varied.

KW - Algorithm methodologies

KW - Biomarkers

KW - Colon cancer

KW - Early detection

UR - http://www.scopus.com/inward/record.url?scp=85070947522&partnerID=8YFLogxK

U2 - 10.1016/j.cca.2019.08.007

DO - 10.1016/j.cca.2019.08.007

M3 - Journal article

VL - 498

SP - 108

EP - 115

JO - Clinica Chimica Acta

JF - Clinica Chimica Acta

SN - 0009-8981

ER -

ID: 57797982