An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms

Michael D. Parkes, Arezu Z. Aliabadi, Martin Cadeiras, Maria G. Crespo-Leiro, Mario Deng, Eugene C. Depasquale, Johannes Goekler, Daniel H. Kim, Jon Kobashigawa, Alexandre Loupy, Peter Macdonald, Luciano Potena, Andreas Zuckermann, Philip F. Halloran

Research output: Contribution to journalArticle

2 Citations (Scopus)

Abstract

BACKGROUND: We previously reported a microarray-based diagnostic system for heart transplant endomyocardial biopsies (EMBs), using either 3-archetype (3AA)or 4-archetype (4AA)unsupervised algorithms to estimate rejection. In the present study we examined the stability of machine-learning algorithms in new biopsies, compared 3AA vs 4AA algorithms, assessed supervised binary classifiers trained on histologic or molecular diagnoses, created a report combining many scores into an ensemble of estimates, and examined possible automated sign-outs. METHODS: We studied 889 EMBs from 454 transplant recipients at 8 centers: the initial cohort (N = 331)and a new cohort (N = 558). Published 3AA algorithms derived in Cohort 331 were tested in Cohort 558, the 3AA and 4AA models were compared, and supervised binary classifiers were created. RESULTS: A‘lgorithms derived in Cohort 331 performed similarly in new biopsies despite differences in case mix. In the combined cohort, the 4AA model, including a parenchymal injury score, retained correlations with histologic rejection and DSA similar to the 3AA model. Supervised molecular classifiers predicted molecular rejection (areas under the curve [AUCs]>0.87)better than histologic rejection (AUCs <0.78), even when trained on histology diagnoses. A report incorporating many AA and binary classifier scores interpreted by 1 expert showed highly significant agreement with histology (p < 0.001), but with many discrepancies, as expected from the known noise in histology. An automated random forest score closely predicted expert diagnoses, confirming potential for automated signouts. CONCLUSIONS: Molecular algorithms are stable in new populations and can be assembled into an ensemble that combines many supervised and unsupervised estimates of the molecular disease states.

Original languageEnglish (US)
Pages (from-to)636-646
Number of pages11
JournalJournal of Heart and Lung Transplantation
Volume38
Issue number6
DOIs
StatePublished - Jun 2019
Externally publishedYes

Fingerprint

Molecular Pathology
Transplants
Biopsy
Histology
Area Under Curve
Diagnosis-Related Groups
Noise
Wounds and Injuries
Population

Keywords

  • antibody-mediated rejection
  • heart transplant
  • injury
  • microarray
  • T-cell‒mediated rejection

ASJC Scopus subject areas

  • Surgery
  • Pulmonary and Respiratory Medicine
  • Cardiology and Cardiovascular Medicine
  • Transplantation

Cite this

An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms. / Parkes, Michael D.; Aliabadi, Arezu Z.; Cadeiras, Martin; Crespo-Leiro, Maria G.; Deng, Mario; Depasquale, Eugene C.; Goekler, Johannes; Kim, Daniel H.; Kobashigawa, Jon; Loupy, Alexandre; Macdonald, Peter; Potena, Luciano; Zuckermann, Andreas; Halloran, Philip F.

In: Journal of Heart and Lung Transplantation, Vol. 38, No. 6, 06.2019, p. 636-646.

Research output: Contribution to journalArticle

Parkes, MD, Aliabadi, AZ, Cadeiras, M, Crespo-Leiro, MG, Deng, M, Depasquale, EC, Goekler, J, Kim, DH, Kobashigawa, J, Loupy, A, Macdonald, P, Potena, L, Zuckermann, A & Halloran, PF 2019, 'An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms', Journal of Heart and Lung Transplantation, vol. 38, no. 6, pp. 636-646. https://doi.org/10.1016/j.healun.2019.01.1318
Parkes, Michael D. ; Aliabadi, Arezu Z. ; Cadeiras, Martin ; Crespo-Leiro, Maria G. ; Deng, Mario ; Depasquale, Eugene C. ; Goekler, Johannes ; Kim, Daniel H. ; Kobashigawa, Jon ; Loupy, Alexandre ; Macdonald, Peter ; Potena, Luciano ; Zuckermann, Andreas ; Halloran, Philip F. / An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms. In: Journal of Heart and Lung Transplantation. 2019 ; Vol. 38, No. 6. pp. 636-646.
@article{84c4dab9a588427c96a20e7482cd01ec,
title = "An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms",
abstract = "BACKGROUND: We previously reported a microarray-based diagnostic system for heart transplant endomyocardial biopsies (EMBs), using either 3-archetype (3AA)or 4-archetype (4AA)unsupervised algorithms to estimate rejection. In the present study we examined the stability of machine-learning algorithms in new biopsies, compared 3AA vs 4AA algorithms, assessed supervised binary classifiers trained on histologic or molecular diagnoses, created a report combining many scores into an ensemble of estimates, and examined possible automated sign-outs. METHODS: We studied 889 EMBs from 454 transplant recipients at 8 centers: the initial cohort (N = 331)and a new cohort (N = 558). Published 3AA algorithms derived in Cohort 331 were tested in Cohort 558, the 3AA and 4AA models were compared, and supervised binary classifiers were created. RESULTS: A‘lgorithms derived in Cohort 331 performed similarly in new biopsies despite differences in case mix. In the combined cohort, the 4AA model, including a parenchymal injury score, retained correlations with histologic rejection and DSA similar to the 3AA model. Supervised molecular classifiers predicted molecular rejection (areas under the curve [AUCs]>0.87)better than histologic rejection (AUCs <0.78), even when trained on histology diagnoses. A report incorporating many AA and binary classifier scores interpreted by 1 expert showed highly significant agreement with histology (p < 0.001), but with many discrepancies, as expected from the known noise in histology. An automated random forest score closely predicted expert diagnoses, confirming potential for automated signouts. CONCLUSIONS: Molecular algorithms are stable in new populations and can be assembled into an ensemble that combines many supervised and unsupervised estimates of the molecular disease states.",
keywords = "antibody-mediated rejection, heart transplant, injury, microarray, T-cell‒mediated rejection",
author = "Parkes, {Michael D.} and Aliabadi, {Arezu Z.} and Martin Cadeiras and Crespo-Leiro, {Maria G.} and Mario Deng and Depasquale, {Eugene C.} and Johannes Goekler and Kim, {Daniel H.} and Jon Kobashigawa and Alexandre Loupy and Peter Macdonald and Luciano Potena and Andreas Zuckermann and Halloran, {Philip F.}",
year = "2019",
month = "6",
doi = "10.1016/j.healun.2019.01.1318",
language = "English (US)",
volume = "38",
pages = "636--646",
journal = "Journal of Heart and Lung Transplantation",
issn = "1053-2498",
publisher = "Elsevier USA",
number = "6",

}

TY - JOUR

T1 - An integrated molecular diagnostic report for heart transplant biopsies using an ensemble of diagnostic algorithms

AU - Parkes, Michael D.

AU - Aliabadi, Arezu Z.

AU - Cadeiras, Martin

AU - Crespo-Leiro, Maria G.

AU - Deng, Mario

AU - Depasquale, Eugene C.

AU - Goekler, Johannes

AU - Kim, Daniel H.

AU - Kobashigawa, Jon

AU - Loupy, Alexandre

AU - Macdonald, Peter

AU - Potena, Luciano

AU - Zuckermann, Andreas

AU - Halloran, Philip F.

PY - 2019/6

Y1 - 2019/6

N2 - BACKGROUND: We previously reported a microarray-based diagnostic system for heart transplant endomyocardial biopsies (EMBs), using either 3-archetype (3AA)or 4-archetype (4AA)unsupervised algorithms to estimate rejection. In the present study we examined the stability of machine-learning algorithms in new biopsies, compared 3AA vs 4AA algorithms, assessed supervised binary classifiers trained on histologic or molecular diagnoses, created a report combining many scores into an ensemble of estimates, and examined possible automated sign-outs. METHODS: We studied 889 EMBs from 454 transplant recipients at 8 centers: the initial cohort (N = 331)and a new cohort (N = 558). Published 3AA algorithms derived in Cohort 331 were tested in Cohort 558, the 3AA and 4AA models were compared, and supervised binary classifiers were created. RESULTS: A‘lgorithms derived in Cohort 331 performed similarly in new biopsies despite differences in case mix. In the combined cohort, the 4AA model, including a parenchymal injury score, retained correlations with histologic rejection and DSA similar to the 3AA model. Supervised molecular classifiers predicted molecular rejection (areas under the curve [AUCs]>0.87)better than histologic rejection (AUCs <0.78), even when trained on histology diagnoses. A report incorporating many AA and binary classifier scores interpreted by 1 expert showed highly significant agreement with histology (p < 0.001), but with many discrepancies, as expected from the known noise in histology. An automated random forest score closely predicted expert diagnoses, confirming potential for automated signouts. CONCLUSIONS: Molecular algorithms are stable in new populations and can be assembled into an ensemble that combines many supervised and unsupervised estimates of the molecular disease states.

AB - BACKGROUND: We previously reported a microarray-based diagnostic system for heart transplant endomyocardial biopsies (EMBs), using either 3-archetype (3AA)or 4-archetype (4AA)unsupervised algorithms to estimate rejection. In the present study we examined the stability of machine-learning algorithms in new biopsies, compared 3AA vs 4AA algorithms, assessed supervised binary classifiers trained on histologic or molecular diagnoses, created a report combining many scores into an ensemble of estimates, and examined possible automated sign-outs. METHODS: We studied 889 EMBs from 454 transplant recipients at 8 centers: the initial cohort (N = 331)and a new cohort (N = 558). Published 3AA algorithms derived in Cohort 331 were tested in Cohort 558, the 3AA and 4AA models were compared, and supervised binary classifiers were created. RESULTS: A‘lgorithms derived in Cohort 331 performed similarly in new biopsies despite differences in case mix. In the combined cohort, the 4AA model, including a parenchymal injury score, retained correlations with histologic rejection and DSA similar to the 3AA model. Supervised molecular classifiers predicted molecular rejection (areas under the curve [AUCs]>0.87)better than histologic rejection (AUCs <0.78), even when trained on histology diagnoses. A report incorporating many AA and binary classifier scores interpreted by 1 expert showed highly significant agreement with histology (p < 0.001), but with many discrepancies, as expected from the known noise in histology. An automated random forest score closely predicted expert diagnoses, confirming potential for automated signouts. CONCLUSIONS: Molecular algorithms are stable in new populations and can be assembled into an ensemble that combines many supervised and unsupervised estimates of the molecular disease states.

KW - antibody-mediated rejection

KW - heart transplant

KW - injury

KW - microarray

KW - T-cell‒mediated rejection

UR - http://www.scopus.com/inward/record.url?scp=85061638804&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85061638804&partnerID=8YFLogxK

U2 - 10.1016/j.healun.2019.01.1318

DO - 10.1016/j.healun.2019.01.1318

M3 - Article

C2 - 30795962

AN - SCOPUS:85061638804

VL - 38

SP - 636

EP - 646

JO - Journal of Heart and Lung Transplantation

JF - Journal of Heart and Lung Transplantation

SN - 1053-2498

IS - 6

ER -