Mostly due to ethical consideration, non-inferiority studies are often conducted in medical device evaluations for regulatory approval. The operating characteristics of commonly used test methods for non-inferiority on a left truncated normal endpoint are hardly reported in literature. In this paper, we note that the type I error rate using the (naïve) t-test and randomization test could be significantly different from the nominal rate with simulation study. We first propose a simple adjustment to the naïve t-test so that type I error rate could be maintained closer to the nominal level in practical use. We further explore the use of Wald test, showing type I error rate is well maintained close to the nominal level under a wide range of scenarios. Late lumen loss, an endpoint left truncated at zero in evaluating the effectiveness of drug eluting stent, will be used as an illustration example.