Large healthcare databases and clinical research: Challenges and opportunities

P. Garbayo-Salmons J. Shourick The increasing popularity of research based on large-scale, real-world databases such as TriNetX, Explorys and other similar platforms has opened new avenues for clinical investigation, also in the field of dermatology.1-3 These tools offer researchers fast and cost-effective access to large-scale patient datasets across institutions, regions and even continents. This is particularly valuable in contexts where multicentre collaboration is logistically or economically challenging. One of the main advantages of these platforms is their inclusiveness. By integrating data from a wide range of healthcare systems, they capture substantial sociodemographic and clinical variability. In addition, they enable rapid hypothesis generation and provide access to populations often underrepresented in traditional studies. The use of advanced analytical methods, such as propensity score matching and multiple imputation, not only improves the quality of observational analyses but also lowers technical barriers. As a result, these tools help democratize research, allowing broader participation from investigators regardless of institutional setting or available resources. Despite these advantages, several methodological concerns must be acknowledged. First, data accuracy and consistency are frequently limited. Diagnostic coding practices may vary across institutions and countries, and criteria for case definitions are often unstandardized. Transparency regarding data entry, missing data handling and variable construction is generally lacking, which can impair reproducibility and comparability.4 In prevalence studies, underdiagnosis and underreporting are significant limitations. Thus, chronic or stigmatized conditions, such as hidradenitis suppurativa or sexually transmitted infections, could be underrepresented. Consequently, observed prevalence may reflect diagnostic practices, coding behaviour or disparities in access to care rather than the actual epidemiological distribution. Similar methodological challenges arise in association studies, which are also prone to various forms of bias. Even with advanced statistical techniques, residual confounding and selection bias may still occur.5, 6 It is essential to remember that these studies are often exploratory and should not be considered a substitute for well-designed prospective research. A notable subset of these studies focuses on evaluating the risk of developing new diseases following a specific condition or exposure, in which the temporality between exposure and outcome is often ambiguous, limiting causal inference. Another frequent pitfall is surveillance bias: patients with pre-existing conditions may undergo more intensive follow-up, increasing the likelihood of detecting unrelated diagnoses. For instance, a study might report a higher incidence of thyroid disease following melanoma, not due to a causal link, but because of increased medical surveillance. Considering these challenges, we propose practical interventions to improve the reliability and interpretability of studies using large real-world datasets. These recommendations, aimed at reducing bias and improving reproducibility, are summarized in Table 1. In conclusion, large real-world datasets represent a powerful tool for clinical research, especially when traditional study designs are not feasible. Nevertheless, their use must be accompanied by methodological rigour and critical interpretation. As the scientific community continues to embrace these platforms, we must ensure that enthusiasm does not outpace caution. Encouraging transparency in data collection practices and reinforcing the limitations of observational data in publications are key steps to maximize their value without compromising scientific integrity. None. We do not have conflicts of interest to declare related to the content of this article. This research did not require IRB approval because it involved analysis of publicly available, de-identified data without any direct interaction with human subjects. This study complies with the ICMJE Recommendations for patient privacy. No identifiable patient information is included in the manuscript. The data that support the findings of this study are available from the corresponding author upon reasonable request.