We conduct a longitudinal analysis of health misinformation to understand its temporal patterns and impact on public health behaviors. Informed by the Health Belief Model, we analyzed 12521 fact-checked claims related to COVID-19 posted between Jan 03, 2020, when the pandemic emerged, and May 11, 2023, when the public health emergency ended. Keeping up with the computational theory construction paradigm, we employed several methods to identify health belief topics, claim veracity (i.e., true, false, or misleading), media of origination (i.e., social or web media), and modality (i.e., lean or rich). Following the temporal bracketing approach, we analyze the patterns of misinformation (i.e., false and misleading claims) during eight stages encompassing several surge and recovery periods, as well as emergence and wind-down. Furthermore, we correlated misinformation patterns with vaccination trends to determine the impact of misinformation on public health behavior. Based on the empirical evidence, we develop propositions to theorize about the phenomenon of misinformation and health behavior. This study contributes by conducting a longitudinal analysis of health misinformation and provides insights into the patterns of health beliefs that originate on social and web media in different modalities. Further, this study contributes by providing insights into the impact of misinformation on health behaviors. The insights might interest several public health experts and policymakers in designing better communication and intervention strategies to counter the false narrative about the pandemic. The study could also inform the development of better approaches to identifying, monitoring, and fact-checking misinformation. Finally, the study lays the ground to examine further motivations, mechanisms, and impacts of sharing health misinformation on online platforms.