Background New classification criteria for inclusion of gout patients in clinical studies are being devised based on a points system. The question is what threshold to use for defining an eligible subject for inclusion in a study for gout. Misclassification will occur almost inevitably. However, in studies (interventions) with low risk and low costs, misclassification is less a problem than in studies (interventions) with high risk and high costs. Therefore, preferred cut points for classification criteria may differ by study type.
Objectives To determine the acceptable level of positive predictive value (PPV) and negative predictive value (NPV) for classification criteria for gout, given the type of study.
Methods We held an international web-based survey with 91 general practitioners and rheumatologists, comfortable with the diagnosis and treatment of gout. Conjoint analysis was used as the framework for devising and analyzing questions. Panel members were anonymously given 20 pairs of two profiles describing a study type and a positive predictive value and a negative predictive value. The panelists were asked to make a choice which of the 2 study profiles they would prefer to enroll a patient. There were 5 study types presented: a phase 3 RCT of an NSAID versus prednisone for acute gout; a phase 3 RCT of a biologic agent for acute gout; a phase 2 RCT of a novel uricosuric drug of unknown efficacy and limited toxicity data; a case-control genome-wide-association (GWAS) study of gout; a cohort study examining long term outcomes of gout. PPV and NPV both had five levels: 60%, 70%, 80%, 90%, 99%. The data were analyzed using Sawtooth software. A binary logit model was used to calculate the relative importance of study type, PPV and NPV. Importance of each level of NPV given a specific study type and PPV was plotted in a line-graph to read the point at which the function crosses the x-axis (i.e. utility =0) above which there is preference. By varying the level of NPV and study type one can see whether this cut point is dependent on study type and NPV.
Results There were 91 panellists: 35% female; 7% trainees; 93% rheumatologists and 5% general practitioners; with an average of 19 years of practice and mostly seeing 5 to 60 gout patients monthly. PPV was most highly weighted in decision making: the relative importance was 59% for PPV; 29% for NPV and 13% for study type. Although the overall (“univariate”) minimal preferred NPV was 90%, this depended on PPV and study type. The minimal preferred NPV was 60% or 70% if PPV is 90% (figure) and similar for several study types. NPV should be at least 80% if PPV is 80% for a phase 3 RCT on NSAIDS, and NPV should be at least 70% for a cohort study if PPV is 80% (not shown).
Conclusions A positive predictive value of 90% with a negative predictive value of 70% was preferred for most study types; the cut point in the classification criteria for gout should not produce PPV and NPV lower than these values. While it has been hypothesized that different cut-points may be needed based on study type, a single cut-point can be a reasonable approach for all study types if the cut point approximates a PPV of 90% and NPV of 80%.
Disclosure of Interest None declared