"Editorial: Collection of Covid-19 Induced Biases in Medical Research"

Reza Rastmanesh^1*Vasiliki Epameinondas Georgakopoulou²Martin Wolkewitz³

• ¹American Physical Society, College Park, United States
• ²Laiko General Hospital of Athens, Athens, Greece
• ³Institute of Medical Biometry and Statistics, University Medical Center Freiburg, Freiburg, Germany

general terms, and many studies were significantly impacted by biases related to time, selection, and competing risks. (Lucke, et al., 2024) demonstrated that multi-state models help mitigate these biases by simultaneously analyzing various clinical outcomes while considering their time-related nature, including ongoing cases, and accounting for competing events. A group of researchers utilized a publicly accessible dataset from COVID-19 first wave to illustrate the advantages of employing multi-state methodology in the analysis of hospital data.They evaluated the results of the data analysis conducted with multi-state models against the results obtained when different types of bias were overlooked. Additionally, Cox regression was employed to analyze the transitions between states in the multi-state model, enabling a comparison of how covariates affect transition rates between the two states. Finally, they computed the anticipated lengths of stay and state probabilities derived from the multi-state model and represented this information through stacked probability plots. Utilizing multi-state models on real-time data enables quick identification of changes in disease progression when new variants emerge. This information is crucial for guiding medical and political leaders, as well as the general public.Another three common methodological biases need to be addressed: competing risks, immortal-time bias, and confounding bias in real-world observational studies that assess treatment effectiveness. A team of researchers utilized a specific observational data example involving COVID-19 patients to evaluate the effects of these biases and suggest possible solutions. Indeed, neglecting competing risks, immortal-time bias, and confounding bias can distort treatment effect estimates.According to (Martinuka, et al., 2024), using the basic Kaplan-Meier method produced the most inaccurate results, leading to inflated probabilities for the primary outcome in studies involving COVID-19 hospital data. This inflation could misguide clinical decisions. Therefore, it is essential to tackle both immortal-time bias and confounding bias when evaluating treatment effectiveness. The trial emulation framework presents a possible approach to mitigate all three of these methodological biases.This was only a part of the issue. Tackling bias in how SARS-CoV-2 reinfection is defined is another key challenge. Traditionally, reinfection is identified as a positive test result that happens at least 90 days after a prior infection has been diagnosed. However, this lengthy timeframe might result in an undercount of reinfection cases. (Chemaitelly, et al., 2024) explored the possibility of using a different, shorter timeframe to define reinfection. The 40-day time window was appropriate for defining reinfection, irrespective of whether it was the first, second, third, or fourth occurrence. The sensitivity analysis, confined to high testers exclusively, replicated similar patterns and results. These findings will significantly impact the issue of underestimation.The comparison between immunity gained from previous natural infections and that obtained through vaccination against SARS-CoV-2 is a significant topic. In this context, we required a statistical clarification to prevent any misinterpretation. To achieve this goal, we need access to real-world data from a large population. (Weber, et al., 2024) analyzed data from over 52,000 individuals. The group that was infected tended to be younger, had a higher proportion of men, and exhibited lower morbidity compared to the vaccinated group. After the initial 90 days, these differences became more pronounced.The analysis conducted during the second 90 days revealed variations in results based on different analytical methods and age groups. There were also age-related differences in mortality rates. When considering the outcome of SARS-CoV-2 infection, the impact of vaccination compared to infection differs by age, showing a disadvantage for vaccinated individuals in the younger demographic, while no significant difference was observed in older adults. It is important to analyze two observation periods: the first and second 90day spans after infection or vaccination. Furthermore, it is necessary to implement methods to correct any imbalances. This strategy facilitates equitable comparisons, enables more thorough conclusions, and helps avoid biased interpretations. It is crucial not to mix up these results with the 40-day time frame that was proposed as suitable for identifying reinfection (Chemaitelly, et al., 2024).As for the observational studies on the effectiveness of COVID-19 vaccines, these designs have provided crucial real-world insights that have influenced global public health policies. These studies, which mainly utilize existing data sources, have been crucial for evaluating vaccine effectiveness across various populations and for creating effective vaccination strategies. Cohort designs are commonly used in this research. The swift rollout of vaccination campaigns during the pandemic led to variations in vaccination rates influenced by socio-demographic factors, public policies, perceived risks, health-promoting behaviors, and overall health status. This may have resulted in biases such as healthy user bias, healthy vaccine effect, frailty bias, differential depletion of susceptibility bias, and confounding by indication. The pressure to publish findings rapidly may have exacerbated these biases or led to their oversight, thereby affecting the reliability of the results. The extent of these biases can vary greatly depending on the context, data sources, and analytical techniques used, and they are likely to be more pronounced in low-and middle-income countries due to weaker data infrastructure. It is crucial to address and reduce these biases to obtain accurate estimates of vaccine effectiveness, inform public health strategies, and maintain public confidence in vaccination efforts. (Agampodi, et al., 2024) in their brilliant paper state that clear communication about these biases and a commitment to improving the design of future observational studies are vital.Another type of neglected bias that may obscure data analysis during the COVID-19 pandemic arises from treatment-induced differences. (Prosty, et al., 2024) demonstrated that during the pandemic, many patients received concomitant corticosteroids, which are known to broadly suppress inflammatory cytokines, including those associated with type II inflammation. This may have obscured any differences induced by omalizumab and biased the results toward the null hypothesis, while others did not receive corticosteroid therapy. Results from one of the papers submitted to our research topic suggested that the potential benefits of omalizumab in COVID-19 may be mediated independently of the modulation of the measured serum biomarkers. This finding, in itself, impacts the interpretation of many clinical trials conducted during the pandemic.Given the numerous issues addressed in this brief editorial, the significance of interdisciplinary collaboration in mitigating biases exacerbated by the pandemic must be emphasized.