OBJECTIVE: To compare non-parametric tests and procedures of selection in building clinically validated genotypic scores. DESIGN AND PATIENTS: In the Jaguar study, 111 patients on a stable antiretroviral regimen experiencing virological failure were randomized in the didanosine (ddI) arm to receive ddI for 4 weeks in addition to their current combination therapy. METHODS: The virological response was HIV-1 RNA reduction from baseline to week 4. The univariate impact of each mutation associated with resistance to ddI on virological response was quantified by comparing reduction in plasma HIV-1 RNA in patients with or without the specific mutation, using a Wilcoxon-Mann-Whitney test. The next step was to select the combination of mutations most strongly associated with the virological response. Two procedures and two tests were compared using either the set of resistance mutations or the set of resistance mutations and mutations providing a better virological response. The Kruskal-Wallis and the Jonckheere test for ordered alternatives were compared in order to build a genotypic score using the two distinct procedures. RESULTS: Eight mutations were associated with a reduced virological response to ddI: M41L, D67N, T69D, L74V, V1181, L210W, T215Y/F and K219Q/E and two mutations with a better virological response: K70R and M184V/I. The Jonckheere-Terpstra test for trend provided the combination of mutations (M41L+T69D-K70R+L74V-M184V /I+T215Y/F+ K219A/E) that were the most predictive for the week 4 virological response, that is, leading to the lowest P value. The 'removing' procedure, starting from a set of mutations retained and removing mutations one by one to find the best combination, provides lower P values than the 'adding' procedure starting with a single mutation and adding mutations one by one. Whatever the set of mutations and the procedure used, the Jonckheere-Terpstra test selects combinations of mutations leading to lower P values than the Kruskal-Wallis test. CONCLUSION: The Jonckheere-Terpstra test for trend is recommended for building a genotypic score when compared with the Kruskal-Wallis. The choice of the selection procedure is discussed here and may be dependent on the objective of the score.
RCT Entities:
OBJECTIVE: To compare non-parametric tests and procedures of selection in building clinically validated genotypic scores. DESIGN AND PATIENTS: In the Jaguar study, 111 patients on a stable antiretroviral regimen experiencing virological failure were randomized in the didanosine (ddI) arm to receive ddI for 4 weeks in addition to their current combination therapy. METHODS: The virological response was HIV-1 RNA reduction from baseline to week 4. The univariate impact of each mutation associated with resistance to ddI on virological response was quantified by comparing reduction in plasma HIV-1 RNA in patients with or without the specific mutation, using a Wilcoxon-Mann-Whitney test. The next step was to select the combination of mutations most strongly associated with the virological response. Two procedures and two tests were compared using either the set of resistance mutations or the set of resistance mutations and mutations providing a better virological response. The Kruskal-Wallis and the Jonckheere test for ordered alternatives were compared in order to build a genotypic score using the two distinct procedures. RESULTS: Eight mutations were associated with a reduced virological response to ddI: M41L, D67N, T69D, L74V, V1181, L210W, T215Y/F and K219Q/E and two mutations with a better virological response: K70R and M184V/I. The Jonckheere-Terpstra test for trend provided the combination of mutations (M41L+T69D-K70R+L74V-M184V /I+T215Y/F+ K219A/E) that were the most predictive for the week 4 virological response, that is, leading to the lowest P value. The 'removing' procedure, starting from a set of mutations retained and removing mutations one by one to find the best combination, provides lower P values than the 'adding' procedure starting with a single mutation and adding mutations one by one. Whatever the set of mutations and the procedure used, the Jonckheere-Terpstra test selects combinations of mutations leading to lower P values than the Kruskal-Wallis test. CONCLUSION: The Jonckheere-Terpstra test for trend is recommended for building a genotypic score when compared with the Kruskal-Wallis. The choice of the selection procedure is discussed here and may be dependent on the objective of the score.
Authors: Anne-Genevieve Marcelin; Bernard Masquelier; Diane Descamps; Jacques Izopet; Charlotte Charpentier; Chakib Alloui; Magali Bouvier-Alias; Anne Signori-Schmuck; Brigitte Montes; Marie-Laure Chaix; Corinne Amiel; Georges Dos Santos; Annick Ruffault; Francis Barin; Gilles Peytavin; Marc Lavignon; Philippe Flandre; Vincent Calvez Journal: Antimicrob Agents Chemother Date: 2008-07-14 Impact factor: 5.191
Authors: M M Santoro; A Bertoli; P Lorenzini; F Ceccherini-Silberstein; N Gianotti; C Mussini; C Torti; G Di Perri; G Barbarini; T Bini; S Melzi; P Caramello; R Maserati; P Narciso; V Micheli; A Antinori; C F Perno Journal: Infection Date: 2009-01-23 Impact factor: 3.553
Authors: Allal Houssaïni; Lambert Assoumou; Anne Geneviève Marcelin; Jean Michel Molina; Vincent Calvez; Philippe Flandre Journal: AIDS Res Treat Date: 2012-04-03