To develop and evaluate the prediction of a disease outcome or score and index, there are requirements to which studies must adhere. The first requirement is that the study design includes a cohort, which can be transformed into two separate cohorts or two cohorts of patients with failed or successful outcomes [2,3]. The second requirement is model validation, including internal and external validation, which uses the same data for internal validation but separate data for external validation. Various methods such as split file and bootstrapping, among others, can be used for validation. [2-5]. Thirdly, discrimination is defined as the ability to distinguish events from non-events, which is appropriately determined by area under the curve (AUC). One should always consider that a large AUC only reflects a high degree of discrimination, and it may become difficult to assess whether it is appropriate or not. Consequently, even if the AUC is statistically significant, there is a chance it won't be predictive [3]. If the qualitative interaction between the important variables is not examined, it will lead to miscalculations of results [3].

The authors concluded that preoperative PaO₂/FiO₂ ratio might predict the postoperative desaturation in aging patients after spinal anesthesia for femoral fracture surgery, while preoperative arterial blood gas analysis may be helpful in predicting early postoperative desaturation in such patients [1]. In summary, for predictive studies, the above points should be duly considered, otherwise, prediction cannot be guaranteed despite a significant association [2,3]. In this letter, we explain the drawbacks of the article presented by Jo et al. and recommend that any predictive study should adhere to the above methodological requirements [2–5].