Dynamic prediction based on variability of a longitudinal biomarker

Abstract Background Tacrolimus is given post-kidney transplant to suppress the immune system, and the amount of drug in the body is measured frequently. Higher variability over time may be indicative of poor drug adherence, leading to more adverse events. It is important to account for the variation...

Full description

Bibliographic Details
Main Authors: Kristen R. Campbell, Rui Martins, Scott Davis, Elizabeth Juarez-Colunga
Format: Article
Language:English
Published: BMC 2021-05-01
Series:BMC Medical Research Methodology
Subjects:
Online Access:https://doi.org/10.1186/s12874-021-01294-x
id doaj-b14a5bbdce1e40b9988a8cc3ad3afba0
record_format Article
spelling doaj-b14a5bbdce1e40b9988a8cc3ad3afba02021-05-16T11:03:02ZengBMCBMC Medical Research Methodology1471-22882021-05-0121111010.1186/s12874-021-01294-xDynamic prediction based on variability of a longitudinal biomarkerKristen R. Campbell0Rui Martins1Scott Davis2Elizabeth Juarez-Colunga3Department of Pediatrics, University of Colorado Anschutz Medical CampusCentro de Estatística e Aplicações da Universidade de Lisboa (CEAUL); Faculdade de Ciências da Universidade de Lisboa, Departamento de Estatística e Investigação OperacionalDivision of Renal Diseases and Hypertension, University of Colorado Anschutz Medical CampusDepartment of Biostatistics and Informatics, University of Colorado Anschutz Medical CampusAbstract Background Tacrolimus is given post-kidney transplant to suppress the immune system, and the amount of drug in the body is measured frequently. Higher variability over time may be indicative of poor drug adherence, leading to more adverse events. It is important to account for the variation in Tacrolimus, not just the average change over time. Methods Using data from the University of Colorado, we compare methods of assessing how the variability in Tacrolimus influences the hazard of de novo Donor Specific Antibodies (dnDSA), an early warning sign of graft failure. We compare multiple joint models in terms of fit and predictive ability. We explain that the models that account for the individual-specific variability over time have the best predictive performance. These models allowed each patient to have an individual-specific random error term in the longitudinal Tacrolimus model, and linked this to the hazard of dnDSA model. Results The hazard for the variance and coefficient of variation (CV) loading parameter were greater than 1, indicating that higher variability of Tacrolimus had a higher hazard of dnDSA. Introducing the individual-specific variability improved the fit, leading to more accurate predictions about the individual-specific time-to-dnDSA. Conclusions We showed that the individual’s variability in Tacrolimus is an important metric in predicting long-term adverse events in kidney transplantation. This is an important step in personalizing the dosage of TAC post-transplant to improve outcomes post-transplant.https://doi.org/10.1186/s12874-021-01294-xDynamic predictionKidney transplantSurvival
collection DOAJ
language English
format Article
sources DOAJ
author Kristen R. Campbell
Rui Martins
Scott Davis
Elizabeth Juarez-Colunga
spellingShingle Kristen R. Campbell
Rui Martins
Scott Davis
Elizabeth Juarez-Colunga
Dynamic prediction based on variability of a longitudinal biomarker
BMC Medical Research Methodology
Dynamic prediction
Kidney transplant
Survival
author_facet Kristen R. Campbell
Rui Martins
Scott Davis
Elizabeth Juarez-Colunga
author_sort Kristen R. Campbell
title Dynamic prediction based on variability of a longitudinal biomarker
title_short Dynamic prediction based on variability of a longitudinal biomarker
title_full Dynamic prediction based on variability of a longitudinal biomarker
title_fullStr Dynamic prediction based on variability of a longitudinal biomarker
title_full_unstemmed Dynamic prediction based on variability of a longitudinal biomarker
title_sort dynamic prediction based on variability of a longitudinal biomarker
publisher BMC
series BMC Medical Research Methodology
issn 1471-2288
publishDate 2021-05-01
description Abstract Background Tacrolimus is given post-kidney transplant to suppress the immune system, and the amount of drug in the body is measured frequently. Higher variability over time may be indicative of poor drug adherence, leading to more adverse events. It is important to account for the variation in Tacrolimus, not just the average change over time. Methods Using data from the University of Colorado, we compare methods of assessing how the variability in Tacrolimus influences the hazard of de novo Donor Specific Antibodies (dnDSA), an early warning sign of graft failure. We compare multiple joint models in terms of fit and predictive ability. We explain that the models that account for the individual-specific variability over time have the best predictive performance. These models allowed each patient to have an individual-specific random error term in the longitudinal Tacrolimus model, and linked this to the hazard of dnDSA model. Results The hazard for the variance and coefficient of variation (CV) loading parameter were greater than 1, indicating that higher variability of Tacrolimus had a higher hazard of dnDSA. Introducing the individual-specific variability improved the fit, leading to more accurate predictions about the individual-specific time-to-dnDSA. Conclusions We showed that the individual’s variability in Tacrolimus is an important metric in predicting long-term adverse events in kidney transplantation. This is an important step in personalizing the dosage of TAC post-transplant to improve outcomes post-transplant.
topic Dynamic prediction
Kidney transplant
Survival
url https://doi.org/10.1186/s12874-021-01294-x
work_keys_str_mv AT kristenrcampbell dynamicpredictionbasedonvariabilityofalongitudinalbiomarker
AT ruimartins dynamicpredictionbasedonvariabilityofalongitudinalbiomarker
AT scottdavis dynamicpredictionbasedonvariabilityofalongitudinalbiomarker
AT elizabethjuarezcolunga dynamicpredictionbasedonvariabilityofalongitudinalbiomarker
_version_ 1721439947577098240