Background: There is no validated case definition for human immunodeficiency virus-associated immune reconstitution inflammatory syndrome (IRIS). We measured the level of agreement of 2 published case definitions (hereafter referred to as CD1 and CD2) with expert opinion in a prospective cohort of patients who were starting antiretroviral therapy in South Africa.
Methods: A total of 498 adult patients were monitored for the first 6 months of antiretroviral therapy. All new or worsening clinical events were reviewed by 2 investigators and classified on the basis of expert opinion, CD1, and CD2. Events were categorized according to whether they were paradoxical or unmasking in presentation. We measured positive, negative, and chance-corrected agreement (kappa) with expert opinion for CD1 and CD2, and reviewed areas of disagreement.
Results: A total of 620 clinical events were recorded, of which, on the basis of expert opinion, 144 (23.2%) were defined as probable IRIS and 112 (18.1%) were defined as possible IRIS. Of the 144 probable IRIS events, 93 (64.6%) were unmasking in presentation, 99 (68.8%) were associated with dermatological or orogenital disease, and 45 (31.3%) were associated with tuberculosis or major opportunistic infections. Of the 620 clinical events recorded, 41 (6.6%) were classified as IRIS on the basis of CD1, and 156 (25.2%) were classified as IRIS on the basis of CD2. Positive agreement between CD1 and expert opinion was low for both unmasking (17.2%; kappa = 0.24) and paradoxical events (37.3%; kappa = 0.43), mainly because 1 major criterion requires IRIS to be atypical and either an opportunistic infection or a tumor, although negative agreement was >98%. In contrast, CD2 had good positive agreement (>75% for most event types), with a kappa value of 0.75 for paradoxical and 0.62 for unmasking.
Conclusions: CD2 agreed well with expert opinion, with additional clinical events, such as arthropathy and inflammatory dermatoses, being classified as IRIS and added to CD2. We propose revised case definitions for both paradoxical and unmasking IRIS.