EuResist prediction system
The EuResist project has developed an integrated computational
system for clinical management of antiretroviral drug resistance. The
system requests HIV genotype and optionally a set of clinical data and
returns a prediction of response to common antiretroviral regimens,
helping the HIV specialist to choose the most effective drug
combinations for her/his patient.
Nov 2015 update: longer term prediction model
The system has been re-trained and tested in November 2015 on a set of data derived from
clinical practice records of more than 66,000 HIV patients collected in Europe. The internal
validation showed a prediction accuracy of 73-76% (AUROC) when treatment success was defined
as achieving a decrease of at least 2 log or an undetectable HIV RNA load at week 48 after
treatment start (it was week 8 in the previous model).
read more.
The EuResist system is a linear combination of three
statistical learning engines predicting response to treatment. These
were designed to work with viral genotype and intended regimen as the
only information, as well as with additional measurements (e.g. baseline
viral load) and information derived from previous treatments and
genotypes to enhance prediction performance. The models combined in the
final prediction system include: a Generative Discriminative engine
using a Bayesian Network prediction as an additional covariate, a Mixed
Effects engine modelling interactions between covariates, and an
Evolutionary engine exploiting the genetic barrier to drug resistance as
an additional predictor.
read more.
EuResist integrated database
The EuResist Integrated DataBase (EIDB) was set up to develop
the EuResist prediction system. It is available for scientific purposes
upon request.
The EIDB includes information about patient demographics, drug
therapies, AIDS defining events, CD4 and viral load measurements, and
HIV sequences. As of June 2017 it contains the following data:
| Patients |
81.227
|
| CD4 counts |
1.126.492
|
| ENV |
196
|
| GP120 sequences |
3.952
|
| GP41 sequences |
1.389
|
| IN sequences |
8.165
|
| PR sequences |
79.809
|
| RT sequences |
80.367
|
| Therapies |
172.140
|
| V3 sequences |
1.071
|
| Viral Load data |
1.118.249
|
The EIDB integrates biomedical information from the three
founding national databases: ARCA (Italy), AREVIR (Germany), Karolinska
Institute (Sweden). The original data set is continuously updated with
new data in order to improve the accuracy of the prediction system.
Following the EIDB setup, data have also been uploaded from the
Laboratoire de Rétrovirologie of CRP-Santé (Luxembourg), the Rega
Institute (Leuven - Belgium), the IrsiCaixa Foundation (Barcelona –
Spain), the Instituto de Higiene e Medicina Tropical (Portugal) and from
the DUET study (Tibotec).
read more.
Contribution to the database
The service is and will remain free of charge. Like any
data-driven engine, availability of new data is crucial for updating and
improving the system. You can contribute genotype-response data to the
EuResist initiative under the current regulations.
(database).
The EuResist data set is available upon request for scientific studies related to HIV domain.