A commentary on “Intra- and early postoperative predictors of delirium risk in cardiac surgery results from the prospective observational FINDERI study”

Dear Editor,

We read the article in the International Journal of Surgery with considerable interest. Paul et al^[1] conducted a prospective observational study that aimed to FIND DElirium RIsk factors (FINDERI) for patients undergoing elective cardiac surgery. While the study provides valuable insights into Postoperative delirium (POD) after elective cardiac surgery, certain limitations must be addressed to contextualize the findings and guide future research.

First, the limitations of postoperative delirium assessment methods. The study used the CAM-ICU and I-CAM tools to assess delirium twice daily, but did not include immediate postoperative assessments (such as delirium during the awakening period), which may have underestimated the incidence of delirium. Additionally, while the researchers received standardized training for the assessments, the consistency among evaluators (as indicated by the Kappa value) was not reported, potentially introducing measurement bias^[2]. Future studies are recommended to incorporate continuous EEG or objective biomarkers (such as inflammatory markers) to enhance the sensitivity and objectivity of the detection.

Second, the contamination of variables is not well controlled. Although the study adjusted for known confounding factors using multivariate regression and machine learning, it did not adequately control for potential influencing factors such as postoperative sedation protocols, pain management strategies, or ICU environments (such as noise and lighting). These variables may independently affect the risk of delirium, for example, the association between benzodiazepine use and delirium has been widely confirmed^[3]. Future studies should systematically record and adjust for these variables.

Third, machine learning models have limited clinical applicability. The AUC of the LASSO regression and decision tree models was only 0.74 and 0.71, respectively, in the validation set, indicating moderate predictive performance. Additionally, no clinical decision thresholds (such as risk score thresholds) were provided. The models did not incorporate preoperative risk factors, such as cognitive function scores, which may reduce their practicality^[4]. It is recommended to develop a dynamic prediction tool that integrates preoperative, intraoperative, and postoperative data, and to validate its clinical benefits through prospective studies.

In conclusion, the FINDERI study offers a useful basis for comprehending delirium risk in patients undergoing heart surgery; however, in order to fully contextualize the results and progress the field, future research needs to address the limitations mentioned above. We can obtain a more thorough understanding of delirium risk and create more effective interventions to improve patient outcomes by refining assessment techniques, adjusting for variable contamination, and creating more clinically applicable machine learning models. The TITAN rule, which demands openness in the reporting of AI, was mentioned in connection with all of this^[5].

Ethical approval

Not applicable.

Consent

None.

Sources of funding

None.

Author contributions

J.H., X.B., and Q.Z. collaborated on the creation and structure of this letter.

Conflicts of interest disclosure

There are no conflicts of interest.

Guarantor

Qingli Zhao.

Research registration unique identifying number (UIN)

Not applicable.

Provenance and peer review

This manuscript is a comment without being invited.

Data availability statement

No data was used in this Letter to the Editor.