Editor, I read with great interest the recent publication by Lammers-Lietz et al.1 on the association between dopaminergic system functional connectivity (FC) and postoperative cognitive dysfunction (POCD), which presents a thought-provoking approach to identifying predictors in a low event-rate sample. However, based on the supplementary materials and relevant statistical principles, it is noted that there are several key aspects of the study that warrant further academic discussion. The authors candidly discuss in the ‘Limitation’ section that, due to the small number of POCD cases (n = 26), principal component analysis (PCA) was selected to avoid overfitting. Although the Supplementary Figures S19 and S20 confirm the robustness of the PCA sub-networks, as an unsupervised method, PCA only models the variance within the independent variables and ignores their covariance with the outcome variable. This limitation may result in the exclusion of ‘weak but critical’ connectivity signals directly related to cognitive impairment, which could be discarded as noise. In future studies, in addition to the authors’ suggestion to use Ridge and LASSO regression, it is recommended that partial least squares regression (PLS) be explored, as it can maximise the association between brain connectivity and clinical outcomes while performing dimensionality reduction.2 For POCD, which involves high spatial collinearity in imaging data, PLS typically provides more biologically interpretable predictors compared to PCA. The study reports no significant systemic changes in dopaminergic connectivity postoperatively. However, the authors did not fully adjust for intra-operative blood pressure fluctuations (e.g. area under the curve for mean arterial pressure below baseline) and the acute modulation of dopamine receptors by specific anaesthetic agents (such as propofol or opioids). These physiological and pharmacological ‘noise’ factors may have obscured the true re-organisation of neural circuits postsurgery.3,4 It is suggested that future research integrate anaesthetic monitoring data (e.g. electronic anaesthesia records) and include intra-operative low blood pressure and cumulative drug doses as key covariates to better interpret whether postoperative FC stability results from neural resilience or is masked by peri-operative interference. The BioCog cohort had a relatively high dropout rate from screening to final analysis. Although sensitivity analyses were performed in the supplementary materials, the discussion regarding whether missing data were Missing Not at Random (MNAR) remains insufficient. If the most cognitively impaired postoperative patients were excluded due to drop-out or inability to tolerate scans, the results may suffer from ‘survivor bias’ and underestimate the true impact of dopaminergic dysfunction. It is recommended to use multiple imputation and conduct stress tests based on different missing data assumptions to significantly enhance the robustness of the conclusions.5 The current study reveals associations, but the development of POCD is usually a multifactorial, mediated process. It is suggested that future studies attempt to construct a Structural Equation Model (SEM). By modelling pathways such as ‘pre-operative dopamine reserve’ mediating through ‘intra-operative stress sensitivity’, and ultimately affecting ‘postoperative cognitive trajectories’, the SEM framework would provide a clearer understanding of the underlying mechanisms.3,4 The research by Lammers-Lietz et al. offers valuable insights for the prevention of POCD. By incorporating supervised learning algorithms, fine-tuning peri-operative control factors, and applying causal inference models, future studies could deepen our understanding of the core role of the dopaminergic system in the postoperative outcomes of elderly patients.
Chenchen Jiang (Wed,) studied this question.