Comparing Cross-Sectional and Longitudinal Study Designs for Accurate Viral Dynamics Estimation: Insights From the NBA Cohort Data

Viral load data provide critical insights into host-pathogen interactions and guide clinical and public health decisions. Because frequent testing is often infeasible, viral dynamics models are used to reconstruct infection trajectories, but optimal sampling strategies remain unclear. We compared two approaches for collecting SARS-CoV-2 viral load data: cross-sectional sampling (one measurement at symptom onset) and longitudinal sampling (every 3 days after onset) under constraints on the total number of tests and tests per individual. A viral dynamics model was first fitted to data from the National Basketball Association cohort, and the estimated parameters were treated as ground truth. Synthetic data were then generated under each sampling design, refitted, and evaluated for accuracy in estimating viral load over 30 days, peak viral load, peak time, and viral shedding duration. Longitudinal sampling consistently yielded lower root mean squared error and narrower one standard deviation interval than cross-sectional sampling. Peak timing and viral shedding duration were unbiased under both designs, but cross-sectional designs underestimated peak viral load and produced wider one standard deviation intervals. Coverage of viral load estimates was markedly higher for longitudinal designs (> 0.90) compared with cross-sectional ones (~0.10). Accuracy and coverage exceeded 0.96 even with just two tests per individual, with little additional benefit from more tests. In conclusion, longitudinal sampling—despite limited data—substantially improves accuracy and precision of viral load estimation compared with cross-sectional designs. These findings highlight efficient strategies for study design and resource allocation in infectious disease research.