Introduction

ResProt

JMIR Res Protoc

JMIR Research Protocols

1929-0748

JMIR Publications Inc.

Toronto, Canada

23611761

v1i1e3

10.2196/resprot.1930

Proposal

HIV Drug-Resistant Patient Information Management, Analysis, and Interpretation

Eysenbach

Gunther

Foster

Rosemary

Singh

Yashik

BSC, BSC(Hons), MMedSc 1

Department of TeleHealth Nelson R Mandela school of Medicine University of KwaZulu-Natal

719 Umbilo Road, Dept of Telehealth, 5th floor, Main Building, Umbilo

Durban, 4001

South Africa 27 312604117 27 312604737 singhyashik@gmail.com

Mars

Maurice

MB ChB, MD 1

¹ Department of TeleHealth Nelson R Mandela school of Medicine University of KwaZulu-Natal

Durban

South Africa

Corresponding Author: Yashik Singh singhyashik@gmail.com

Jan-Jun 2012

07 06 2012

1 1

20 09 2011 09 01 2012 27 01 2012 22 04 2012

©Yashik Singh, Maurice Mars. Originally published in JMIR Research Protocols (http://www.researchprotocols.org), 07.06.2012.

2012

This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Research Protocols, is properly cited. The complete bibliographic information, a link to the original publication on http://www.researchprotocols.org, as well as this copyright and license information must be included.

Introduction

The science of information systems, management, and interpretation plays an important part in the continuity of care of patients. This is becoming more evident in the treatment of human immunodeficiency virus (HIV) and acquired immune deficiency syndrome (AIDS), the leading cause of death in sub-Saharan Africa. The high replication rates, selective pressure, and initial infection by resistant strains of HIV infer that drug resistance will inevitably become an important health care concern. This paper describes proposed research with the aim of developing a physician-administered, artiﬁcial intelligence-based decision support system tool to facilitate the management of patients on antiretroviral therapy.

Methods

This tool will consist of (1) an artificial intelligence computer program that will determine HIV drug resistance information from genomic analysis; (2) a machine-learning algorithm that can predict future CD₄count information given a genomic sequence; and (3) the integration of these tools into an electronic medical record for storage and management.

Conclusion

The aim of the project is to create an electronic tool that assists clinicians in managing and interpreting patient information in order to determine the optimal therapy for drug-resistant HIV patients.

Medical Informatics Bioinformatics HIV drug resistance Machine Learning

Introduction

The current trend in patient health care is personalized medicine where treatment is individualized, rather than a response to set physical presentations. Thus, access to and interpretation of personal patient information is vital in order to provide a sustainable and useful medical service. The science of information systems, management, and interpretation plays an important role in the continuity of care of patients. This is becoming more evident in the treatment of human immunodeficiency virus (HIV) and acquired immune deficiency syndrome (AIDS). This paper describes proposed research where the aim is to develop a physician-administered artiﬁcial intelligence-based decision support system tool that will facilitate the management of patients on antiretroviral therapy.

The enveloped human immunodeficiency virus infects and destroys the human immune system over a long period of time [1]. The two known strains of HIV are HIV-1 and HIV-2. The rate of replication and infection of the HIV-2 is substantially slower than that of the HIV-1 and accounts for 95% of all HIV infections [2]. HIV-1 is subdivided into four groups representing four separate introductions of simian immunodeficiency virus into humans:

1. Group M is the major HIV-1 group with respect to prevalence (the number of people infected) and incidence (the number of new infections) of the virus;

2. Group O is the outlier group and is mostly restricted to west-central Africa;

3. Group N was discovered in 1998 in Cameroon and is extremely rare; and

4. Group P is a strain closely resembling the gorilla simian immunodeficiency virus.

Currently, Group M is subdivided into nine subtypes or clades—A, B, C, D, F, G, H, J, and K—based on variations in genetic sequence characteristics [3]. However, it is possible for viruses from different subtypes to form mosaic genomes called circulation recombinant forms (CRF). In sub-Saharan Africa, HIV/AIDS is the leading cause of death [4] and it is one of the fastest growing epidemics in South Africa [5-8], where currently there are 5.7million confirmed cases of HIV/AIDS [9]. Demographic information on confirmed HIV-infected patients in South Africa is presented in Table 1.

Table 1

Estimated HIV prevalence rates in South Africa [9].

	Women	Men	Total population
Age range	20–64	20–64	15–46	20–64	All ages
% with HIV	18.1	17.7	18.8	17.9	11.1

HIV infection can be effectively managed with antiretroviral (ARV) drugs, usually in the form of highly active antiretroviral therapy (HAART), which is comprised of a regimen of three drugs from at least two of the following five drug classes [10-13]: reverse transcriptase inhibitors (RTI), nucleoside reverse transcriptase inhibitors (NRTI), protease inhibitors (PI), integrase inhibitors (II), and fusion inhibitors (FI).

Factors that influence treatment of HIV/AIDS with ARVs include poor treatment regimen prescribed by the physician; the World Health Organization (WHO) stage of the disease, which is related to the progression of the disease; levels of plasma drug concentration achieved; how strictly the patient adheres to the regimen; drug resistance [14]; and toxic effects of the drug. Drug resistance is the most critical aspect of treatment. Three common reasons leading to the development of HIV antiretroviral drug resistance are high replication rates, selective pressure, and initial infection by resistant strains of HIV. Thus, it is inevitable that drug resistance will become a reality in most patients’ treatment.

Preventative measures must be taken in order to develop infrastructure that will aid in the management of drug-resistant HIV patients. It is essential to develop techniques that will extract valuable information from little patient data. There must be a means developed to manage, analyze, and interpret patient data.

The aim of this study is to develop a physician-administrated information system that facilitates the clinical management of HIV-positive patients on antiretroviral therapy. This system should be Web-based, patient centric, ascribe to the principles of personal medicine, promote complete health management, and incorporate continuity of care. Creation of this tool will involve:

Development of an artificial intelligence computer algorithm that analyzes HIV drug resistance data and provides singular interpretable information for a physician indicating which ARVs a patient will be resistant to;

Investigation of the application of a computer algorithm to predict current and future CD₄ lymphocyte cell count information given a genomic sequence and other data;

Integration of the above tools with an electronic medical record system such that it facilitates the storage, acquisition, and management of patient information; and

Development of a Web-based electronic tool that assists clinicians in determining the optimal therapy for drug-resistant HIV patients.

Background Medical Informatics

The appropriate application of computer science and associated technology has extended medical care beyond traditional diagnosis and patient management, resulting in extensive cost efficiencies and improved public health outcomes [15]. Areas of medical informatics application include patient records, practice management, clinical measurements, patient education, prescription writing, Web/database resources, clinical records, data collection, clinical decision support, and clinical measurements. Recently, there has been an intentional move towards investigating the synergy between medical informatics and bioinformatics [9]. Bioinformatics is the application of computer science techniques to study how information is represented and transmitted in biological systems starting at the molecular level [16].

The application of genomes in medicine has altered many aspects of medicine. Genome analysis that enhances clinical practice has been successfully applied to asthma [17], cancer [18-20], diabetes [21], and cardiovascular disease [22].

HIV Drug Resistance Prediction Algorithms

Testing for HIV drug resistance may consist of wet or dry chemistry laboratory tests, or by employing electronic computerized algorithms [23]. The use of computer algorithms falls under the field of medical informatics. Computer based interpretation algorithms using genomes can also be used to predict HIV drug resistance.

These interpretation algorithms can be generally divided into one of two groups:

Those based on known domain knowledge (ie, they are based on the fact that certain combinations of known genome mutations cause unequivocal resistance), and

Those not based on predefined domain knowledge. These algorithms include machine learning and statistical methods.

Interpretation Algorithms Based on Domain Knowledge

Domain knowledge interpretation algorithms are based on scientific and published interactions between certain mutations and/or combination of mutations with resistance. This means that all computational decisions concerning resistance are based on known mutation-resistance rules found in published scientific literature. REGA, Agence Nationale de Recherches sur le SIDA (ANRS), and Stanford’s HIVdb algorithm [24] are three examples of well-known domain knowledge interpretation algorithms. These algorithms are widely used in the field and are regarded as gold standards.

REGA and ANRS classify ARV resistance according to three levels: susceptible, intermediate, and resistant. “Susceptible” indicates that a particular ARV drug will be effective against HIV; “intermediate” indicates that the ARV drug is partially effective; and if the ARV is not effective at all, it is classified as “resistant.” HIVdb classifies HIV drug resistance according to five levels: susceptible, potential low-level resistance, low-level resistance, intermediate resistance, and high-level resistance. These algorithms employ Boolean-based rules, some with penalties, and predict resistance by determining which mutations are present and/or absent.

Two other domain-based algorithms are the Drug Resistance SEQuence ANalyzer (DR_SEQAN) and RetroGram. DR_SEQAN was coded by the Universidad Autónoma de Madrid for a Windows environment using Visual Basic. DR_SEQAN classifies three levels of resistance: high-level resistance, increased susceptibility, and no resistance. RetroGram was developed by InferMed Ltd (London, UK) and is built using Arezzo and PROforma. RetroGram generates a suitability ranking for ARV drugs using expert rules. Table 2 describes the accuracy of predicting drug resistance of some algorithms based on domain knowledge.

Table 2

Accuracy of predicting ARV drug resistance by the domain-based algorithms, Drug Resistance SEQuence ANalyzer (DR_SEQAN), RetroGram, REGA, and HIVdb [28].

Drug	Accuracy of algorithm (%)
	DR_SEQAN	RetroGram	REGA	HIVdb
Indinavir	70.8	70.0	84.0	78.0
Nelﬁnavir	66.1	70.5	77.7	73.7
Lopinavir	85.7	88.1	81.0	69.0
Lamivudine	83.1	79.7	78.0	78.0
Zidovudine	79.2	64.6	72.9	68.8
Stavudine	60.0	38.8	67.1	37.6
Didanosine	93.3	20.2	27.9	27.8
Nevirapine	87.9	65.2	72.7	90.9

Interpretation Algorithms Not Based on Known Domain Knowledge

Many different pattern recognition and machine-learning algorithms have been applied to find a predictable correlation between genotypic and phenotypic data (called “virtual phenotyping”) [25]. Machine learning may be used to develop a model that predicts virological response. Machine learning is an artificial intelligence computer science technique that tries to find a mathematical model that maps between inputs and outputs of a domain problem.

Virtual phenotyping is growing in popularity. Kuritzkes supports virtual phenotyping as a tool for interpreting viral genotypes [26]. The following are some of the algorithms that have been used:

Least absolute shrinkage and selection operator (LASSO)

Ridge regression

Neural networks, such as multilayer perceptron (MLP) and radial basis neural networks (RBNN)

Principle component analysis

Support vector machines (SVM)

Linear regression models

Hidden Markov models

Decision trees

Multiple correspondence analysis

Associative classifiers

k-nearest neighbor algorithm (kNN)

Results produced by these interpretation algorithms are shown in Tables 3–5. These interpretation algorithms have achieved various levels of success, but there are shortcomings in some of the current versions [27]:

Resistance is interpreted separately for each drug even though therapy consists of combination therapy;

There is a general lack of data, especially for non-B HIV-1 subtypes;

Rule-based interpretation is based on the algorithm creator’s knowledge;

Interpretation algorithms are not always updated even though HIV drug resistance is a rapidly evolving field; and

Other factors that contribute to treatment failure are not taken into account, such as treatment history, resistance history, viral load history, CD₄ count history, or plasma drug concentrations.

Table 3

Accuracy of predicting ARV drug resistance using the interpretation algorithms, support vector machines (SVM), multilayer perceptrons (MLP), and radial basis neural networks (RBNN) [29].

Drug	Accuracy of algorithm (%)
	SVM (Energy)^a	SVM (DEnergy)^a	MLP (Energy)^a	MLP (DEnergy)^a	RBNN
Indinavir	92.6	88.6	87.0	82.5	92.5
Nelﬁnavir	84.9	80.1	86.6	87.1	94.1
Lopinavir	88.6	82.4	92.3	87.9	94.4

^aSee Bonet et al [29] for detailed information about Energy and DEnergy

Table 4

Accuracy of predicting ARV drug resistance using k-nearest neighbor (kNN), decision tree [30], and associative classifier [31] algorithms.

Drug	Accuracy of algorithm (%)
	kNN	Decision tree	Associative classifier
Indinavir	73.6	93.9	93.0
Nelﬁnavir	81.1	89.5	92.1
Lopinavir	80.8	82.5	89.6

Table 5

Accuracy of predicting ARV drug resistance (%) or correlation coefficient (r) reported for various other algorithms and machine-learning techniques.

Algorithm	Accuracy (%)	Correlation coefficient (r)
HIVdb [32]	84.3	n/a
Visible Genetics/Bayer Diagnostics Guidelines 6.0 [32]	86.3	n/a
AntiRetroScan [32]	89.4	n/a
Committee of neural networks [33]	78.0	n/a
Geno2Pheno [34]	n/a	.6

These shortcomings have led to the creation of many different interpretation algorithms, which produce different resistance measures even if applied to the same resistance profile. These differences are because the studies each used different datasets, subtypes, analysis on drug-naive and drug-experienced patients, and so forth. Conclusions of some studies that reported on the discrepancy of the interpretation algorithms are shown in Table 6.

Summary of discrepancies reported using various interpretation algorithms.

Study	Comments
Ravela et al [23]	Studied 4 interpretation algorithms (ANRS-3-02, TRUGENE VGI-6, REGA 5.5, and HIVdb-8-02) and concluded that there was a discrepancy in interpretations in 33% of all resistance profiles tested. The most discordant were NRTIs.
Snoeck et al [35]	Confirmed that there are discordances between the algorithms tested. Suggested it may be due to subtypes.
Vergne et al [36]	Confirmed discrepancies and attributed it to the application of the interpretation algorithms to drug-naive or drug-experienced patients.
De Luca et al (2003) [37]	Concluded that discrepancies in the interpretation algorithms may influence the use of resistance testing over virological outcomes.
De Luca et al (2004) [38]	Studied the application of 13 interpretation algorithms of drug-naive patients and concluded that there are discordances.
Vercauteren and Vandamme [27]	Determined that there is a high level of discordance between the interpretation of NRTI resistance. Also suggests that there should be a “standardization of unique interpretative rules.”
Poonpiriya et al [39]	Indicated that there are discrepancies in the 7 interpretation algorithms they studied.

Collation and interpretation of the contradictory outputs of these algorithms is difficult for physicians treating complex drug-resistant HIV cases, as information is only valuable when it is presented in a clearly interpretable way.

Predicting CD4 Count

HIV can be successfully managed with ARV drugs, but information relating to the progression of HIV is vital. HIV infection may be monitored using laboratory [40,41] and clinical marker information [42,43]. Information about a patient’s CD₄lymphocyte cell counts are the most widely used data for HIV progression and is recognized as a standard measure of immunodeficiency in HIV-positive patients [44,45]. Thus, the proper use and analysis of information regarding CD₄cell counts is vital in CD₄-guided treatment of HIV [46].

Although the use of CD₄count is part of the standard of care in developing countries, the measurement of CD₄count requires many complex and expensive flow cytometric procedures, which burden the minimal resources available [45]. The ability to predict current CD₄cell count will aid in easing the burden on these resources. A physician may use an electronic tool to economically determine an approximate CD₄cell count. If the predicted count is low or indicates that a change in treatment is required, then the physician might order the more expensive laboratory procedure to determine the exact CD₄cell count. The ability to obtain information about future CD₄count changes will have many benefits to physicians. For example, it will facilitate definite treatment actions, such as changing the regimen in order to prevent opportunistic infection (eg, pneumocystis pneumonia) and delay the onset of AIDS.

Neural network machine-learning algorithms have been used to predict viral load [7,47]. Altmann et al [48] created a machine-learning algorithm that predicts success or failure of therapy, based on viral load, with 80% success. This was later changed by predicting the probability of treatment success based on a degree of predicted HIV drug resistance [49]. However, there is not a chemical test or computer model developed yet to forecast changes to the CD₄count.

Decision Support System Tool for Managing Therapy

Although models have been created to choose treatment regimens, very few are available in the public domain and/or are easily accessed through a graphical human (user) interface. Currently, there are Web portals that allow one to determine some aspects of HIV drug resistance treatment. These information portals allow one to determine the current HIV resistance profile, graph trends in viral and CD₄counts with basic alerts, or store basic patient information. BioAfrica (www.bioafrica.net) is an African-based bioinformatics resource [50]. BioAfrica contains bioinformatics resources that can perform sequence alignments, epitope analysis, tools for proteomics, subtyping and virus genotyping, an RNA virus database, and an HIV drug resistance database and tools. The HIV drug resistance database and tools section is based on the REGA collaborative mode and the Calibrated Population Resistance Tool (CPT). REGA is a drug resistance database developed by the REGA Institute, MyBioData Biomedical IT Solutions, and the Katholieke Universiteit Leuven. It contains interpretation algorithms and stores some clinical data related to HIV treatment. CPT was developed at Stanford University and determines the prevalence of HIV drug resistance in a population.

Some of the other international Web portals for managing HIV treatment information are listed in Table 7.

Table 7

Descriptions of Web portals for managing HIV treatment information.

Web portal	Description
Stanford University HIV Drug Resistance Database [51]	This portal determines the interpretation results of the REGA Institute rules, Agence Nationale de Recherches sur le SIDA (ANRS) rules and the Stanford HIVdb rules. It also allows the use of specific user-defined rules using the Algorithm Specification Interface (ASI) and also allows one the opportunity to create a graphical record of a patient’s ARV history, viral loads, CD₄counts, and sequence data.
HIVResistanceWeb [52]	This information portal allows information sharing on ARV resistance and clinical virology. It has a store and forward email-based system that allows one to interact with experts.
Los Alamos HIV database [53]	This information portal contains data on genomes, epitopes, drug resistance mutations, and vaccine trials. It is funded by the Division of AIDS of the National Institute of Allergy and Infectious Diseases (NIAID).

These individual information portals are limited by the following:

The tools they employ in determining HIV drug resistance information. Each information portal uses its own interpretation algorithm and, if collaboration does exist, it consists of simply reporting the outputs of the various algorithms. This causes confusion as some of these interpretation algorithms are disparate, even when the same mutations are analyzed.

They do not have any means of a real-time expert consultation.

They are not integrated into a full electronic medical record, which will add the advantage of continuity of care and facilitate tele-HIV-management.

No individual portal has a variety of tools that can be used to manage HIV therapy.

Methods Part 1: Developing a Single Interpretation Algorithm

The goal of Part 1 is to develop an HIV drug resistance interpretation algorithm capable of providing a single interpretation to genomic analysis.

This part of the study is divided into three main objectives: (1) determining the extent of the disparate information provided by some gold standard interpretation algorithms using the latest version of the interpretation algorithms; (2) developing a novel algorithm to collate the HIV drug resistance interpretation information of these gold standard algorithms into a single easily understandable output; and (3) analyzing the collated algorithm in terms of specificity, sensitivity, and accuracy.

1. Determining the extent of the disparate nature of some gold standard interpretation algorithms using the latest version of these algorithms.

Over time with each new version, interpretation algorithms have improved in predicting ARV drug resistance. Previous comparisons between interpretation algorithms have had some shortcomings:

Each interpretation algorithm has different measures or levels of resistance;

Non-contemporary versions of interpretation algorithms were used in the interpretation;

The interpretation algorithms were applied to different data sets; and

Few interpretations make use of complex statistical analysis to determine if the differences are in fact significant or not.

The latest versions of different interpretation algorithms will be applied to a single data set extracted from a publicly available anonymized database, the Stanford HIV drug resistance database [51]. The measures of resistance for each interpretation algorithm will be determined, grouped, and analyzed.

2. Developing a novel algorithm to collate the HIV drug resistance interpretation of these gold standard interpretation algorithms into a single output.

The gold standard algorithms may be collated by:

Weighted output. Different levels of complexity may be applied to determine a single interpretation from multiple interpretations. A simple majority-voting scheme may be applied, where a count of the interpretations of each algorithm is kept. The single interpretation is obtained by determining the resistance outcome with the highest weighting.

Machine learning on gold standard outputs. Different machine-learning techniques may be applied to the data in order to obtain a single interpretation. Machine-learning techniques work by determining a mapping between a given set of input and desired outcomes and then, using this learnt mapping function, it predicts the output, given a set of inputs. Each interpretation produced for a single resistance profile by the different interpretation algorithms can be the input to a machine-learning algorithm. The output will be the actual HIV-ARV resistance measure determined by fold resistance values. The algorithm will then learn a mapping between the interpretation results obtained using various interpretation algorithms and the actual HIV-ARV resistance measure. One such algorithm that may be employed is a support vector machine.

Creating a simulated boosted dataset both by modeling the strengths and weaknesses of the gold standards.

3. Analyzing the collated algorithm in terms of specificity, sensitivity, and accuracy.

The specificity, sensitivity, and accuracy associated with predicting ARV drug resistance will be calculated for each algorithm and then compared using statistical analysis.

Contribution

The literature does not indicate the current state of disparity between gold standard interpretation algorithms. Combining the interpretation algorithms to form one single interpretation is novel.

Part 2: Predicting CD4 Count From Genome Data

This part of the study may be divided into three parts:

Investigating the possibility of creating a machine-learning algorithm that predicts the current CD₄ count of a patient using genome sequences, viral loads, and time;

Investigating the possibility of creating a machine-learning algorithm that forecasts the medium term change in CD₄ count of a patient using current genome sequence;

It is acknowledged that genome sequencing is more expensive and resource intensive than CD₄ cell count measurement. However, the cost of genome sequencing is offset by the numerous bioinformatics applications that may be applied to the genome sequence to predict and analyze other physiological measurements and diseases. This study, however, will also investigate the possibility of creating a machine-learning algorithm that forecasts the medium term change in CD₄ count of a patient using standard of care data.

Methods

Datasets will be obtained from the Stanford HIV drug resistance database (http://hivdb.stanford.edu/), which is publically available and contains data from clinical trials. Subtype B consensus protease (PR) genome sequences, CD₄count, viral load, and the number of weeks from the baseline measure of CD₄count for each patient sample will be determined by joining individual datasets using the sample identifier (the unique number that identifies a sample) and date. Data of patient’s genome sequences and associated viral load and CD₄count data at different time points will be extracted.

The changes in CD₄count will be grouped into categories and a classification model will be built based on the changes. Different groups of inputs will be created and each will feed into the machine-learning algorithm separately, forming three models. Some of these input groups will be:

Input1: consisting only of genome sequence;

Input2: consisting of genome sequence and current viral load; and

Input3: consisting of genome sequence, current viral load, and number of weeks from the current CD₄ count to baseline CD₄ count.

Contribution

Currently, there is not a chemical test or computer model developed to forecast future changes to the CD₄count.

Part 3: Developing Web-based Tool for Determining Optimal Therapy

The goal of Part 3 is to develop a Web-based electronic tool that assists clinicians in determining the optimal therapy for patients indicative of HIV drug resistance.

There is evidence that suggests that resistance testing is beneficial:

A two-factorial (genotyping and expert advice), randomized, open label, multicenter trial [54] was undertaken to determine if there is any benefit in using genotyping rather than the expert’s direct knowledge when prescribing ARVs. The conclusion was that genotyping benefits the overall optimal care of HIV patients.

The VIRalliance SAS [55] group clearly demonstrated in their study that “resistance testing prior to initiating or switching antiretroviral therapy” is essential.

Mascolini et al [56] questioned 600 clinicians about the effect of resistance testing on their diagnosis and regimen they prescribe. They confirmed that “if the assay detected partly or multidrug-resistant virus, then the large proportions of respondents (indicated that they would) change their treatment choice.”

Hirsch et al [57] found that “resistance testing can improve virological outcome among HIV-infected individuals.”

The Can Resistance Enhance Selection of Treatment (CREST) [25] study (a 48-week follow-up randomized trial) found that genotypic drug resistance testing may be beneficial in the management of HIV infection.

Objective

The goal is to combine the tools mentioned previously, and possibly other bioinformatics tools, into one seamless application.

Methods

Java, HyperText Markup Language (HTML), PHP: Hypertext Preprocessor (PHP), and other paradigms will be used to create a Web-based portal that will integrate the different tools. An important aspect to take into account when building the model for the HIV management system is security. Dwivedi et al [58] argued that electronic medical records will only become a reality if security takes a prominent role in design considerations and during implementation. Two of the most promising techniques for incorporating security into any information system are public key infrastructure and biometrics. Public key encryption is a nondeterministic polynomial time complex technique that ensures high-level security. Biometrics use physical of behavioral traits to identify an individual. The exact means of integration and security model to be used will only be determined after the individual tools are built.

Contribution

The creation of an electronic medical record-based virtual HIV clinical support system that aids in the determination of the best HAART combination, using a combined ARV resistance interpretation, CD₄count prediction, and the other methods described is novel.

Conclusion

The outcome of this study is to facilitate the acquisition, storage, management, analysis, and interpretation of information by physicians. In personalized medicine, it is essential that information be interpreted and presented clearly and concisely. We expect that the proposed tool will aid in this aspect.

Abbreviations

AIDS

acquired immune deficiency syndrome

ANRS

Agence Nationale de Recherches sur le SIDA

ARV

antiretroviral

ASI

algorithm specification interface

CPT

Calibrated Population Resistance Tool

CREST

Can Resistance Enhance Selection of Treatment

CRF

circulation recombinant forms

DR_SEQAN

Drug Resistance SEQuence ANalyzer

fusion inhibitors

HAART

highly active antiretroviral therapy

HIV

human immunodeficiency virus

kNN

k-nearest neighbor algorithm

LASSO

least absolute shrinkage and selection operator

MLP

multilayer perceptron

NIAID

National Institute of Allergy and Infectious Diseases

NRTI

nucleoside reverse transcriptase inhibitors

PHP

PHP: Hypertext Preprocessor.

protease inhibitors

RBNN

radial basis neural networks

RTI

reverse transcriptase inhibitors

SVM

support vector machines

WHO

World Health Organization

This work was supported by the National Institutes of Health Fogarty International Centre (grant number 5D43TW007004-11).

None declared.

Albert

Altfeld

Aries

Behrens

Bredeek

Buhk

Dickinson

Eggers

Fätkenheuer

Korsman

HIV medicine 2007 2007

2012-05-15

Paris

Flying Publisher

Pathogenesis of HIV-1 Infectionhttp://www.hivmedicine.com/hivmedicine2007.pdf

67fr16R85

Quinn

Molecular variants of HIV-1 and their impact on vaccine development

Int J STD AIDS 1998 9 Suppl 1 2

9874106

Kanki

Hamel

Sankalé

Hsieh

Thior

Barin

Woodcock

Guèye-Ndiaye

Zhang

Montano

Siby

Marlink

NDoye

Essex

MBoup

Human immunodeficiency virus type 1 subtypes differ in disease progression

J Infect Dis 1999 01 179 1 68 73

10.1086/314557

9841824

JID980530

Campbell

Nair

Maimane

Sibiya

Supporting people with AIDS and their carers in rural South Africa: possibilities and challenges

Health Place 2008 09 14 3 507 18

10.1016/j.healthplace.2007.10.002

18023238

S1353-8292(07)00087-1

Walker

They (ARV's) are my life, without them i'm nothing- experiences of patients attending a HIV/AIDS clinic in Johannesburg, South Africa

Health and Place 2009

10.1016/j.healthplace.2009.06.006)

Simon-Meyer

Odallo

Greater involvement of people living with HIV/AIDS in South Africa

Evaluation and Program Planning 2002 25 3850 3855

Larder

Wang

Revell

Montaner

Harrigan

De Wolf

Lange

Wegner

Ruiz

Pérez-Elías

Emery

Gatell

Monforte

Torti

Zazzi

Lane

The development of artificial neural networks to predict virological response to combination HIV therapy

Antivir Ther 2007 12 1 15 24

17503743

United Nations country proile: Implementation of agenda, Commission of Sustainable development 1997

2012-05-15

http://www.un.org/esa/earthsummit/

67fmYmmJF

Rispel

Metcalf

Breaking the silence: South African HIV policies and the needs of men who have sex with men

Reprod Health Matters 2009 05 17 33 133 42

10.1016/S0968-8080(09)33442-4

19523590

S0968-8080(09)33442-4

Mitton

The sociological spread of HIV/AIDS in South Africa

J Assoc Nurses AIDS Care 2000 11 4 17 26

10911591

Bartlett

Gallant

Pham

Jhons Hopkins University, School of Medicine 2004

2012-05-15

Knowledge Source Solutions

Medical management of hiv infectionhttp://www.mmhiv.com/

67friNPaQ

Pierret

An analysis over time (1990-2000) of the experiences of living with HIV

Soc Sci Med 2007 10 65 8 1595 605

10.1016/j.socscimed.2007.06.017

17683838

S0277-9536(07)00366-8

Beerenwinkel

Schmidt

Walter

Kaiser

Lengauer

Hoffmann

Korn

Selbig

Diversity and complexity of HIV-1 drug resistance: A bioinformatics approach to predicting phenotype from genotype

PNAS 2002 99 12 8271 8276

Richman

A practical guide to HIV drug resistance and its implications for antiretroviral treatment strategies

International Medical Press 2000

Lovell

Celler

Information technology in primary health care

Int J Med Inform 1999 07 55 1 9 22

10471237

S1386505699000167

Shortliffe

Perreault

Wederhold

Fagan

Medical Informatics Computer applications in health care and biomedicine

Bioniformtics, pp. 638-662, 2000

Allen

Heinzmann

Noguchi

Abecasis

Broxholme

Ponting

Bhattacharyya

Tinsley

Zhang

Holt

Jones

Lench

Carey

Jones

Dickens

Dimon

Nicholls

Baker

Xue

Townsend

Kabesch

Weiland

Carr

von Mutius

Adcock

Barnes

Lathrop

Edwards

Moffatt

Cookson

Positional cloning of a novel gene influencing asthma from chromosome 2q14

Nat Genet 2003 11 35 3 258 63

10.1038/ng1256

14566338

ng1256

van de Vijver

van't Veer

Dai

Hart

Voskuil

Schreiber

Peterse

Roberts

Marton

Parrish

Atsma

Witteveen

Glas

Delahaye

van der Velde

Bartelink

Rodenhuis

Rutgers

Friend

Bernards

A gene-expression signature as a predictor of survival in breast cancer

N Engl J Med 2002 12 19 347 25 1999 2009

10.1056/NEJMoa021967

12490681

347/25/1999

Weigelt

Glas

Wessels

Witteveen

Peterse

van't Veer

Gene expression profiles of primary breast tumors maintained in distant metastases

2003 12 23

Proccedings National Academy Sciences

2003

USA

15901 5

10.1073/pnas.2634067100

Wong

Selvanayagam

Wei

Porter

Vittal

Lin

Liao

Shih

Cheung

Yim

Yip

Ngong

Siu

Chan

Kong

Kutlina

McKinnon

Denhardt

Chin

Chung

Expression genomics of cervical cancer: molecular classification and prediction of radiotherapy response by DNA microarray

Clin Cancer Res 2003 11 15 9 15 5486 92

14654527

Bell

Polonsky

Diabetes mellitus and genetically programmed defects in beta-cell function

Nature 2002 9 788 791

Brindle

Antti

Holmes

Tranter

Nicholson

Bethell

Clarke

Schofield

McKilligin

Mosedale

Grainger

Rapid and noninvasive diagnosis of the presence and severity of coronary heart disease using 1H-NMR-based metabonomics

Nat Med 2002 12 8 12 1439 44

10.1038/nm802

12447357

nm802

Ravela

Betts

Brun-Vézinet

Vandamme

Descamps

van Laethem

Smith

Schapiro

Winslow

Reid

Shafer

HIV-1 protease and reverse transcriptase mutation patterns responsible for discordances between genotypic drug resistance interpretation algorithms

J Acquir Immune Defic Syndr 2003 05 1 33 1 8 14

12792349

PMC2547475

de Oliveira

Deforche

Cassol

Salminen

Paraskevis

Seebregts

Snoeck

van Rensburg

Wensing

van de Vijver

Boucher

Camacho

Vandamme

An automated genotyping system for analysis of HIV-1 and other microbial sequences

Bioinformatics 2005 10 1 21 19 3797 800

10.1093/bioinformatics/bti607

16076886

bti607

Hales

Birch

Crowe

Workman

Hoy

Law

Kelleher

Lincoln

Emery

CREST investigators

A randomised trial comparing genotypic and virtual phenotypic interpretation of HIV drug resistance: the CREST study

PLoS Clin Trials 2006 1 3 e18

10.1371/journal.pctr.0010018

16878178

PMC1523224

Kuritzkes

Reliability of virtual phenotype resistance assay

Medscape HIV/AIDS 2002

Vercauteren

Vandamme

Algorithms for the interpretation of HIV-1 genotypic drug resistance information

Antiviral Research 2006 71 335 344

Garriga

Menéndez-Arias

DR_SEQAN: a PC/Windows-based software to evaluate drug resistance using human immunodeficiency virus type 1 genotypes

BMC Infect Dis 2006 6 44

10.1186/1471-2334-6-44

16524459

1471-2334-6-44

PMC1421411

Bonet

Garcia

Salazar

Sanchex

Saeys

Grau

Predicting human immunodefciency virus (hiv) drug resistance using recurrent neural networks.

2006 11 1

10th International Electronic Conference on Synthetic Organic Chemistry

2006

Switzerland

Murry

Predicting HIV type 1 drug resistance from genoptype using machine learning

Master's thesis, University of Edinburgh 2004

Srisawat

Kijsirikul

Using associative classification for predicting HIV-1 drug resistance

2004

HIS 04: Proceedings of the Fourth International Conference on Hybrid Intelligent Systems

2004

Japan

280 284

Zazzi

Romano

Venturi

Shafer

Reid

Dal Bello

Parolin

Palù

Valensin

Comparative evaluation of three computerized algorithms for prediction of antiretroviral susceptibility from HIV type 1 genotype

J Antimicrob Chemother 2004 02 53 2 356 60

10.1093/jac/dkh021

14688053

dkh021

Drăghici

Potter

Predicting HIV drug resistance with neural networks

Bioinformatics 2003 01 19 1 98 107

12499299

Beerenwinkel

Däumer

Oette

Korn

Hoffmann

Kaiser

Lengauer

Selbig

Walter

Geno2pheno: Estimating phenotypic drug resistance from HIV-1 genotypes

Nucleic Acids Res 2003 07 1 31 13 3850 5

12824435

PMC168981

Snoeck

Kantor

Shafer

Van Laethem

Deforche

Carvalho

Wynhoven

Soares

Cane

Clarke

Pillay

Sirivichayakul

Ariyoshi

Holguin

Rudich

Rodrigues

Bouzas

Brun-Vézinet

Reid

Cahn

Brigido

Grossman

Soriano

Sugiura

Phanuphak

Morris

Weber

Pillay

Tanuri

Harrigan

Camacho

Schapiro

Katzenstein

Vandamme

Discordances between interpretation algorithms for genotypic resistance to protease and reverse transcriptase inhibitors of human immunodeficiency virus are subtype dependent

Antimicrob Agents Chemother 2006 02 50 2 694 701

10.1128/AAC.50.2.694-701.2006

16436728

50/2/694

PMC1366873

Vergne

Snoeck

Aghokeng

Maes

Valea

Delaporte

Vandamme

Peeters

Van Laethem

Genotypic drug resistance interpretation algorithms display high levels of discordance when applied to non-B strains from HIV-1 naive and treated patients

FEMS Immunology and Medical Microbiology 2006 46 1 53 62

De Luca

Cingolani

Di Giambenedetto

Trotta

Baldini

Rizzo

Bertoli

Liuzzi

Narciso

Murri

Ammassari

Perno

Antinori

Variable prediction of antiretroviral treatment outcome by different systems for interpreting genotypic human immunodeficiency virus type 1 drug resistance

J Infect Dis 2003 06 15 187 12 1934 43

10.1086/375355

12792870

JID30412

De Luca

Cozzi-Lepri

Perno

Balotta

Di Giambenedetto

Poggio

Pagano

Tositti

Piscopo

Del Forno

Chiodo

Magnani

d'Arminio Monforte

ICoNA drug resistance study group ICoNA study group

Variability in the interpretation of transmitted genotypic HIV-1 drug resistance and prediction of virological outcomes of the initial HAART by distinct systems

Antivir Ther 2004 10 9 5 743 52

15535412

Poonpiriya

Sungkanuparph

Leechanachai

Pasomsub

Watitpun

Chunhakan

Chantratita

A study of seven rule-based algorithms for the interpretation of HIV-1 genotypic resistance data in Thailand

J Virol Methods 2008 07 151 1 79 86

10.1016/j.jviromet.2008.03.017

18462814

S0166-0934(08)00117-1

Fahey

Taylor

Detels

Hofmann

Melmed

Nishanian

Giorgi

The prognostic value of cellular and serologic markers in infection with human immunodeficiency virus type 1

N Engl J Med 1990 01 18 322 3 166 72

10.1056/NEJM199001183220305

1967191

Moss

Bacchetti

Osmond

Krampf

Chaisson

Stites

Wilber

Allain

Carlson

Seropositivity for HIV and the development of AIDS or AIDS related condition: three year follow up of the San Francisco General Hospital cohort

Br Med J (Clin Res Ed) 1988 03 12 296 6624 745 50

3126959

PMC2545367

Cahn

Perez

Casiro

Crinberg

Muchinik

Progression of HIV-disease: the Buenos Aires cohort study

1991

International conference on AIDS

1991

Florence

Montaner

Craib

Schechter

Application of the World Health Organization system for HIV infection in a cohort of homosexual men in developing a prognostically meaningful staging system

AIDS 1992 07 6 7 719 24

1354450

Post

Wood

Maartens

CD4 and total lymphocyte counts as predictors of HIV disease progression

QJM 1996 07 89 7 505 8

8759490

Schechter

Zajdenverg

Machado

Pinto

Lima

Perez

Predicting CD4 counts in HIV-infected Brazilian individuals: a model based on the World Health Organization staging system

J Acquir Immune Defic Syndr 1994 02 7 2 163 8

7905525

Osmond

Charlebois

Lang

Shiboski

Moss

Factors affecting changes in time from CD4 = 200 to death in two San Francisco cohorts-1992

1983

International Conf on AIDS

1993

Germany

Wang

DeGruttola

Hammer

Harrigan

Larder

Wegner

Winslow

Zazzi

A CollaborativeHIV Resistance Response Database Initiative: Predicting Virological Response Using Neural Network Models

2002 07 2

The XI International HIV Drug Resistance Workshop

2-5 July 2002

Seville

Altmann

Rosen-Zvi

Prosperi

Aharoni

Neuvirth

Schülter

Büch

Struck

Peres

Incardona

Sönnerborg

Kaiser

Zazzi

Lengauer

Comparison of classifier fusion methods for predicting response to anti HIV-1 therapy

PLoS One 2008 3 10 e3470

10.1371/journal.pone.0003470

18941628

PMC2565127

Altmann

Däumer

Beerenwinkel

Peres

Schülter

Büch

Rhee

Sönnerborg

Fessel

Shafer

Zazzi

Kaiser

Lengauer

Predicting the response to combination antiretroviral therapy: retrospective validation of geno2pheno-THEO on a large clinical database

J Infect Dis 2009 04 1 199 7 999 1006

10.1086/597305

19239365

de Oliveira

Cassol

Scourge of a Continent: The BioAfrica website

Science 2005 310 1877

Shafter

Stevenson

Chan

Human immunodeficiency virus reverse transcriptase and protease sequence database

Nucleic Acids Research 1999 27 1 348 352

Vertibrae Inc and HIVresistanceWeb

HIVresistanceWeb 2012

2012-01-16

http://www.hivresistanceweb.com/index.shtml

64jOdnGtc

Los Alamos National Laboratory 2012

2012-01-16

HIV databaseshttp://www.hiv.lanl.gov/content/index

64jOrt35o

Tural

Ruiz

Holtzer

Schapiro

Viciana

González

Domingo

Boucher

Rey-Joly

Clotet

Havana Study Group

Clinical utility of HIV-1 genotyping and expert advice: the Havana trial

AIDS 2002 01 25 16 2 209 18

11807305

VIRalliance SAS

HIV phenotyping - A technique to evaluate drug resistance availible worldwide for patients, clinicians and pharmaceutical industries

European Pharmacotherapy 2003

Mascolini

A suite of new therapies

2003 02

10th Conference on Retroviruses and Opportunistic Infections

2003

Boston

10 14

Hirsch

Brun-Vézinet

Clotet

Conway

Kuritzkes

D'Aquila

Demeter

Hammer

Johnson

Loveday

Mellors

Jacobsen

Richman

Antiretroviral drug resistance testing in adults infected with human immunodeficiency virus type 1: 2003 recommendations of an International AIDS Society-USA Panel

Clin Infect Dis 2003 07 1 37 1 113 28

10.1086/375597

12830416

CID30598

Dwivedi

Bali

Belsis

Naguib

RNG

Every

Nassar

Towards a practical healthcare information security model for healthcare institutions

Information Technology Applications in Biomedicine 2003

4th International IEEE EMBS Special Topic Conference

2003 April

Birmingham

114 117

10.1109/ITAB.2003.1222486