Correlation of a clinical outcome (phenotype) with a biomarker or genotype requires accurate phenotyping and genotyping. We question the accuracy and reliability of phenotyping (determination of postoperative cognitive dysfunction [POCD] at 1 yr) by McDonagh et al.  1POCD lacks consensus diagnostic criteria, and disparate methods have been employed for its detection.2There are several methodological issues pertaining to the approach taken by McDonagh et al.  1that substantially weaken the measurement of POCD: (1) meticulously-matched control groups are imperative if cognitive decline is attributed to anesthetic/surgical events rather than underlying patient comorbidity3(there was no control group in this study); (2) clinically meaningful, objective outcome measures, such as incident dementia,4are interpretable and comparable across studies, as opposed to the artificial composite cognitive outcome measures used in this study; (3) the arbitrary statistical thresholds for POCD diagnosis employed in this study are likely inappropriate and of questionable clinical relevance;2and (4) preoperative mild dementia should have been assiduously identified, because mild dementia, although difficult to detect, is a potent confounder in relation to POCD.5 

McDonagh et al.  1reported a 46% 1-yr incidence of POCD and mention that other studies have found different results. However, no studies by other groups have found rates of persistent POCD even remotely approaching the alarming rate found in this study. Some investigators have found little evidence for persistent POCD attributable to a surgical event,3,5whereas others have found rates between 1 and 5%.6,7The investigators suggest that patient, surgical, and anesthetic factors may all be implicated in the causes of POCD.1However, they specifically provide evidence against anesthetic factors in showing no difference in POCD between regional and general anesthesia.1Could it be that surgery and anesthesia were coincidental and that patient factors alone determined POCD? Might an appropriate control group have shown equivalent cognitive decline?

McDonagh et al.  1employed a principal component analysis to derive four components representing four cognitive domains. No factor loadings on individual tests were reported. Because most principal components contain positive and negative loadings because of the orthogonality requirement, these loadings are important to provide the direction of cognitive score. Furthermore, assuming that a decline in a component score is associated with POCD, McDonagh et al.  1defined POCD as a decline of 1 SD or more in at least one of the four cognitive domains. This definition has two major drawbacks: (1) it depends heavily on the number of principal components and (2) it ignores the time interval after the surgery. By definition, the probability (p ) of POCD increases with the number of principal components used. To illustrate this, consider a hypothetical nonsurgical control group. Let q  be the probability of individuals in this group not meeting the definition of POCD for a specific cognitive domain (i.e. , not declining more than 1 SD). If we assume that the extents of change from baseline for the four cognitive domains follow the same normal distribution and are independent, then the probability in the control group of meeting the definition of POCD would be 1 −q  4. Even with the relatively conservative assumption that 90% of the hypothetical controls would not meet the diagnostic threshold for POCD on each of the cognitive domains (i.e. , q = 0.9), the apparent incidence of cognitive decline by the McDonagh et al.  1criteria would still be 34% (i.e. , 1 − 0.94) in this hypothetical group. It would therefore be useful if the investigators would clarify what incidence of POCD they expected purely by chance with their model and what their incidence of postoperative cognitive enhancement was, using more than 1 SD improvement in any cognitive domain as a threshold criterion. It would also be helpful to know how the investigators classified patients if they declined by more than 1 SD in one uncorrelated test and improved by more than 1 SD in another.

A further concern is that the reported incidence rate of POCD seems inconsistent with the reported change scores from baseline on the continuous cognitive index, which is the average of four component scores.1Among the entire sample, including subjects with and without the APOE4 allele, the 6-week change in cognitive score from baseline had a mean of only 0.05–0.07 (i.e. , very mild improvement compared with baseline) with an SD of approximately 0.27–0.28. Given these results, the estimated incidence of a decline of at least −0.5 (the decline in the cognitive index that the authors defined as clinically meaningful) would be roughly 2%, much smaller than the reported 6-week POCD incidence of 54.3%.

For a longitudinally observed function such as cognitive decline, discrete analyses (i.e. , incidence of POCD at 6 weeks and 1 yr) might not be optimal. It is crucial to understand the time course of POCD. A longitudinal analysis was not provided but is required to assess the entire longitudinal change from baseline to 6 weeks and then to 1 yr. A simple incidence rate of POCD at each individual time point does not offer much information on those who initially declined fast and then recovered or those who did well initially and then declined fast. The investigators should comment what the extent of overlap was between those who met their POCD diagnostic criteria at 6 weeks and those who met their criteria at 1 yr.

The findings of McDonagh et al.  1could have paradigm-changing implications. If 46% of patients undergoing major elective surgery were really to experience clinically meaningful persistent POCD, it is likely that informed patients would choose to forego many such procedures. If, on the other hand, considerably fewer than 46% were to have persistent POCD, the reported phenotyping would be inaccurate, and correlation with the genotype (APOE4) would be meaningless. Let's hope that the latter is true.

*Washington, University in St. Louis, St. Louis, Missouri.

McDonagh DL, Mathew JP, White WD, Phillips-Bute B, Laskowitz DT, Podgoreanu MV, Newman MF, Neurologic Outcome Research Group: Cognitive function after major noncardiac surgery, apolipoprotein E4 genotype, and biomarkers of brain injury. Anesthesiology 2010; 112:852–9
Neurologic Outcome Research Group
Newman S, Stygall J, Hirani S, Shaefi S, Maze M: Postoperative cognitive dysfunction after noncardiac surgery: A systematic review. Anesthesiology 2007; 106:572–90
Selnes OA, Grega MA, Bailey MM, Pham LD, Zeger SL, Baumgartner WA, McKhann GM: Cognition 6 years after surgical or medical therapy for coronary artery disease. Ann Neurol 2008; 63:581–90
Ehlenbach WJ, Hough CL, Crane PK, Haneuse SJ, Carson SS, Curtis JR, Larson EB: Association between acute care and critical illness hospitalization and cognitive function in older adults. JAMA 2010; 303:763–70
Avidan MS, Searleman AC, Storandt M, Barnett K, Vannucci A, Saager L, Xiong C, Grant EA, Kaiser D, Morris JC, Evers AS: Long-term cognitive decline in older subjects was not attributable to noncardiac surgery or major illness. Anesthesiology 2009; 111:964–70
Abildstrom H, Rasmussen LS, Rentowl P, Hanning CD, Rasmussen H, Kristensen PA, Moller JT: Cognitive dysfunction 1–2 years after non-cardiac surgery in the elderly. ISPOCD group. International Study of Post-Operative Cognitive Dysfunction. Acta Anaesthesiol Scand 2000; 44:1246–51
Williams-Russo P, Sharrock NE, Mattis S, Szatrowski TP, Charlson ME: Cognitive effects after epidural vs general anesthesia in older adults. A randomized trial. JAMA 1995; 274:44–50