Determining dose-response curves is a critical first step in preclinical drug discovery. Typically, experimental dose-response data are fitted using nonlinear sigmoidal models. This process requires careful attention to fitting criteria and model comparison. Not infrequently, drug candidate compounds may lack sufficient potency at the tested concentrations or produce data that lead to unavoidable poor fitting. A key parameter to estimate in such cases is the half-maximal inhibitory concentration (IC50). This study illustrates the use of the R package drda as the state-of-the-art statistical tool to address these fitting challenges, with a reputation for having been extensively tested in fitting applications with synthetic data and under random sampling from a real dataset of 379,533 cell line and drug pairs. For model evaluation, drda provides the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Likelihood Ratio Test (LRT) to assess goodness of fit for logistic regressions with varying parameter complexities. Combined with graphical diagnostics, results from our benchmark real dataset show that AIC, BIC, and LRT are essential for the rigorous estimation of IC50. In fact, the systematic use of metrics increases reliability in dose-response studies by minimizing the risk of pursuing ineffective compounds and reducing the chance of discarding potentially active molecules, thereby supporting more efficient and innovative preclinical drug discovery.