Statistical issues in neutralizing antibodies.

Abstract

Frequently the question of the effects of neutralizing antibodies (NAbs) is stated, but not clearly. Some ask: “Are we sure NAbs are important?” And others ask: “Are we sure NAbs are NOT important?” These two questions can lead to different statistical designs and different approaches to the investigation of the impact of NAbs. In addition, one must consider the time frame in which the question is relevant. During a 6-month period, one may see divergent curves, but if they come together at 2 years, is the short-term benefit or risk really important? If there are no differences at 6 months, does that mean NAbs are not important? Both of these questions are obviously, just that, questions. Short-term studies do not answer long-term questions and vice versa. Therefore, from a design perspective, more specificity in the question is key. How the question is stated is important. Is the hypothesis the presence of NAbs worsens the effectiveness of the treatment, or is the hypothesis the presence of NAbs does not worsen the treatment effectiveness? Each of these hypotheses, although similar in nature and inference, leads to different study designs. In the first hypothesis, this is a one-sided test that looks for an increase in worsening among NAb+ patients vs NAb− patients. This is a standard null hypothesis, and the design would compute a response rate or average outcome variable for patients who develop NAb+ vs NAb− and would test the response to treatment between these groups. Although this is not a randomized trial, because we cannot assign patients to be NAb+ or NAb−, we nevertheless can make these comparisons, with the clear caution that we are only looking for associations and not cause. This classic hypothesis and study design is not capable of demonstrating causation. Whatever characteristics are associated with developing NAb+ …