CD4 results with a bias larger than hundred cells per microliter can have a significant impact on the clinical decision during treatment initiation of HIV patients

Abstract

CD4 counts are currently used to assess HIV patients for treatment eligibility and to monitor antiretroviral response to treatment. The emerging point-of-care devices could fill an important gap in resource-limited settings. However, the accuracy of CD4-counting instruments is diverse and data on how CD4 measurement errors have an impact on clinical decisions are lacking. Clinicians were queried on the use of CD4 results in their clinical setting. Subsequently, the effect of CD4 measurement errors on treatment initiation was put in a statistical model. Based on clinical CD4 databases from Belgium, Cambodia, and Senegal, the percentage of unchanged clinical decisions was calculated (treatment initiation should start within a 3-month delay [one visit]) for escalating CD4 measurement errors, taking into account the strict or preventive application of CD4 thresholds at 350 or 500 cells/µl used by clinicians. To ensure that the treatment was initiated appropriately for at least 95% of patients, an error of 5 - 10 cells/µl was allowed. This is significantly smaller than the bias of ±50 cells/µl most clinicians considered acceptable. For limits of agreement (LOA, 1.96 x error) of 100 cells/µl, corresponding to most CD4 instrument evaluations, the misclassification rate of patients was found to be 3 - 28% at the threshold of 350 cells/µl (strict or flexible), and 13 - 20% at 500 cells/µl. The maximum allowed CD4 bias on results from new CD4 technologies should not exceed 50 cells/µl (LOA 100 cells/µl) when applied for treatment initiation, to ensure at least 72% of correct clinical decisions. © 2016 International Clinical Cytometry Society.