Neural network-longitudinal assessment of the Electronic Anti-Retroviral THerapy (EARTH) cohort to follow response to HIV-treatment.

George E. Hatzakis, Moses Mathur, Louise Gilbert, George Panos, Ajay Wanchu, Atul K. Patel, J. K. Maniar, Christos M. Tsoukas

Research output: Contribution to journalArticle

3 Citations (Scopus)

Abstract

HIV infection is for the most part a chronic and asymptomatic disease. To properly monitor the health status of infected individuals it is important to use host and viral surrogate markers as well as pharmacokinetic parameters. Disease progression, assessment of the antiviral potency of the drugs and response to therapy can only be monitored by repetitive measures of viral and host parameters. To prevent the emergence of antiviral drug-resistance, long term side effects and to decide on the appropriate treatment choices, a comprehensive assessment of all contributing factors, medical and non-medical, is necessary. However, the relationship between treatment outcomes with disease markers and other contributing factors is not simple. To date, a model that accurately predicts the likelihood of disease progression or treatment failure in HIV infected patients does not exist. Extending our previous work in this area, we developed temporal Artificial Intelligence models based on Jordan-Elman networks to longitudinally follow viral surrogate markers together with demographics, biochemical and laboratory data to describe the drug-virus-host interactions in over 4000 HIV adult patients. In an international (multi-continent) study of HIV clinical and laboratory data, the profiles of drug-naïve as well as treated patients were evaluated during a 20 year follow-up. Validation of models on a subset of this cohort (n=595) estimated the sensitivity and specificity of treatment success/failure, under different management modalities for individual patients. ROC-curves predicted: virologic success from baseline (ROC=0.871) in drug-naïve previously non-treated patients, switch from virologic success/ failure to failure/success if ever and when (ROC=0.625), switch to virologic success/failure from failure/success within 6 months (ROC=0.722) following a previous switch. This tool may be helpful in the design of longitudinal clinical trials.

Original languageEnglish (US)
Pages (from-to)301-305
Number of pages5
JournalAMIA ... Annual Symposium proceedings / AMIA Symposium. AMIA Symposium
StatePublished - 2005
Externally publishedYes

Fingerprint

HIV
Biomarkers
Treatment Failure
Disease Progression
Viral Drug Resistance
Pharmaceutical Preparations
Therapeutics
Asymptomatic Diseases
Jordan
Artificial Intelligence
ROC Curve
Health Status
HIV Infections
Antiviral Agents
Chronic Disease
Pharmacokinetics
Demography
Clinical Trials
Viruses
Sensitivity and Specificity

ASJC Scopus subject areas

  • Medicine(all)

Cite this

Neural network-longitudinal assessment of the Electronic Anti-Retroviral THerapy (EARTH) cohort to follow response to HIV-treatment. / Hatzakis, George E.; Mathur, Moses; Gilbert, Louise; Panos, George; Wanchu, Ajay; Patel, Atul K.; Maniar, J. K.; Tsoukas, Christos M.

In: AMIA ... Annual Symposium proceedings / AMIA Symposium. AMIA Symposium, 2005, p. 301-305.

Research output: Contribution to journalArticle

Hatzakis, George E. ; Mathur, Moses ; Gilbert, Louise ; Panos, George ; Wanchu, Ajay ; Patel, Atul K. ; Maniar, J. K. ; Tsoukas, Christos M. / Neural network-longitudinal assessment of the Electronic Anti-Retroviral THerapy (EARTH) cohort to follow response to HIV-treatment. In: AMIA ... Annual Symposium proceedings / AMIA Symposium. AMIA Symposium. 2005 ; pp. 301-305.
@article{0e50a49cfb5a4fa3a470683af400b944,
title = "Neural network-longitudinal assessment of the Electronic Anti-Retroviral THerapy (EARTH) cohort to follow response to HIV-treatment.",
abstract = "HIV infection is for the most part a chronic and asymptomatic disease. To properly monitor the health status of infected individuals it is important to use host and viral surrogate markers as well as pharmacokinetic parameters. Disease progression, assessment of the antiviral potency of the drugs and response to therapy can only be monitored by repetitive measures of viral and host parameters. To prevent the emergence of antiviral drug-resistance, long term side effects and to decide on the appropriate treatment choices, a comprehensive assessment of all contributing factors, medical and non-medical, is necessary. However, the relationship between treatment outcomes with disease markers and other contributing factors is not simple. To date, a model that accurately predicts the likelihood of disease progression or treatment failure in HIV infected patients does not exist. Extending our previous work in this area, we developed temporal Artificial Intelligence models based on Jordan-Elman networks to longitudinally follow viral surrogate markers together with demographics, biochemical and laboratory data to describe the drug-virus-host interactions in over 4000 HIV adult patients. In an international (multi-continent) study of HIV clinical and laboratory data, the profiles of drug-na{\"i}ve as well as treated patients were evaluated during a 20 year follow-up. Validation of models on a subset of this cohort (n=595) estimated the sensitivity and specificity of treatment success/failure, under different management modalities for individual patients. ROC-curves predicted: virologic success from baseline (ROC=0.871) in drug-na{\"i}ve previously non-treated patients, switch from virologic success/ failure to failure/success if ever and when (ROC=0.625), switch to virologic success/failure from failure/success within 6 months (ROC=0.722) following a previous switch. This tool may be helpful in the design of longitudinal clinical trials.",
author = "Hatzakis, {George E.} and Moses Mathur and Louise Gilbert and George Panos and Ajay Wanchu and Patel, {Atul K.} and Maniar, {J. K.} and Tsoukas, {Christos M.}",
year = "2005",
language = "English (US)",
pages = "301--305",
journal = "AMIA ... Annual Symposium proceedings / AMIA Symposium. AMIA Symposium",
issn = "1559-4076",
publisher = "American Medical Informatics Association",

}

TY - JOUR

T1 - Neural network-longitudinal assessment of the Electronic Anti-Retroviral THerapy (EARTH) cohort to follow response to HIV-treatment.

AU - Hatzakis, George E.

AU - Mathur, Moses

AU - Gilbert, Louise

AU - Panos, George

AU - Wanchu, Ajay

AU - Patel, Atul K.

AU - Maniar, J. K.

AU - Tsoukas, Christos M.

PY - 2005

Y1 - 2005

N2 - HIV infection is for the most part a chronic and asymptomatic disease. To properly monitor the health status of infected individuals it is important to use host and viral surrogate markers as well as pharmacokinetic parameters. Disease progression, assessment of the antiviral potency of the drugs and response to therapy can only be monitored by repetitive measures of viral and host parameters. To prevent the emergence of antiviral drug-resistance, long term side effects and to decide on the appropriate treatment choices, a comprehensive assessment of all contributing factors, medical and non-medical, is necessary. However, the relationship between treatment outcomes with disease markers and other contributing factors is not simple. To date, a model that accurately predicts the likelihood of disease progression or treatment failure in HIV infected patients does not exist. Extending our previous work in this area, we developed temporal Artificial Intelligence models based on Jordan-Elman networks to longitudinally follow viral surrogate markers together with demographics, biochemical and laboratory data to describe the drug-virus-host interactions in over 4000 HIV adult patients. In an international (multi-continent) study of HIV clinical and laboratory data, the profiles of drug-naïve as well as treated patients were evaluated during a 20 year follow-up. Validation of models on a subset of this cohort (n=595) estimated the sensitivity and specificity of treatment success/failure, under different management modalities for individual patients. ROC-curves predicted: virologic success from baseline (ROC=0.871) in drug-naïve previously non-treated patients, switch from virologic success/ failure to failure/success if ever and when (ROC=0.625), switch to virologic success/failure from failure/success within 6 months (ROC=0.722) following a previous switch. This tool may be helpful in the design of longitudinal clinical trials.

AB - HIV infection is for the most part a chronic and asymptomatic disease. To properly monitor the health status of infected individuals it is important to use host and viral surrogate markers as well as pharmacokinetic parameters. Disease progression, assessment of the antiviral potency of the drugs and response to therapy can only be monitored by repetitive measures of viral and host parameters. To prevent the emergence of antiviral drug-resistance, long term side effects and to decide on the appropriate treatment choices, a comprehensive assessment of all contributing factors, medical and non-medical, is necessary. However, the relationship between treatment outcomes with disease markers and other contributing factors is not simple. To date, a model that accurately predicts the likelihood of disease progression or treatment failure in HIV infected patients does not exist. Extending our previous work in this area, we developed temporal Artificial Intelligence models based on Jordan-Elman networks to longitudinally follow viral surrogate markers together with demographics, biochemical and laboratory data to describe the drug-virus-host interactions in over 4000 HIV adult patients. In an international (multi-continent) study of HIV clinical and laboratory data, the profiles of drug-naïve as well as treated patients were evaluated during a 20 year follow-up. Validation of models on a subset of this cohort (n=595) estimated the sensitivity and specificity of treatment success/failure, under different management modalities for individual patients. ROC-curves predicted: virologic success from baseline (ROC=0.871) in drug-naïve previously non-treated patients, switch from virologic success/ failure to failure/success if ever and when (ROC=0.625), switch to virologic success/failure from failure/success within 6 months (ROC=0.722) following a previous switch. This tool may be helpful in the design of longitudinal clinical trials.

UR - http://www.scopus.com/inward/record.url?scp=39049179056&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=39049179056&partnerID=8YFLogxK

M3 - Article

C2 - 16779050

AN - SCOPUS:39049179056

SP - 301

EP - 305

JO - AMIA ... Annual Symposium proceedings / AMIA Symposium. AMIA Symposium

JF - AMIA ... Annual Symposium proceedings / AMIA Symposium. AMIA Symposium

SN - 1559-4076

ER -