Disruptions to the pharmaceutical supply chain (PSC) have negative implications for patients, motivating their prediction to improve risk mitigation. Although data analytics and machine learning methods have been proposed to support the characterization of probabilities to inform decisions and risk mitigation strategies, their application in the PSC has not been previously described. Further, it is unclear how well these models perform in the presence of emergent events representing deep uncertainty such as the COVID-19 pandemic. This article examines the use of data-driven models to predict PSC disruptions before and during the COVID-19 pandemic. Using data on generic drugs from the pharmacy supply chain division of a Fortune 500 pharmacy benefit management firm, we have developed predictive models based on the naïve Bayes algorithm, where the models predict whether a specific supplier or whether a specific product will experience a supply disruption in the next time period. We find statistically significant changes in the relationships of nearly all variables associated with product supply disruptions during the pandemic, despite pre-pandemic stability. We present results showing how the sensitivity, specificity, and false positive rate of predictive models changed with the onset of the COVID-19 pandemic and show the beneficial effects of regular model updating. The results show that maintaining model sensitivity is more challenging than maintaining specificity and false positive rates. The results provide unique insight into the pandemic's effect on risk prediction within the PSC and provide insight for risk analysts to better understand how surprise events and deep uncertainty affect predictive models.