The pathophysiology of delirium is incompletely understood, including what molecular pathways are involved in brain vulnerability to delirium. This study examined whether preoperative plasma neurodegeneration markers were elevated in patients who subsequently developed postoperative delirium through a retrospective case-control study.
Inclusion criteria were patients of 65 yr of age or older, undergoing elective noncardiac surgery with a hospital stay of 2 days or more. Concentrations of preoperative plasma P-Tau181, neurofilament light chain, amyloid β1-42 (Aβ42), and glial fibrillary acidic protein were measured with a digital immunoassay platform. The primary outcome was postoperative delirium measured by the Confusion Assessment Method. The study included propensity score matching by age and sex with nearest neighbor, such that each patient in the delirium group was matched by age and sex with a patient in the no-delirium group.
The initial cohort consists of 189 patients with no delirium and 102 patients who developed postoperative delirium. Of 291 patients aged 72.5 ± 5.8 yr, 50.5% were women, and 102 (35%) developed postoperative delirium. The final cohort in the analysis consisted of a no-delirium group (n = 102) and a delirium group (n = 102) matched by age and sex using the propensity score method. Of the four biomarkers assayed, the median value for neurofilament light chain was 32.05 pg/ml for the delirium group versus 23.7 pg/ml in the no-delirium group. The distribution of biomarker values significantly differed between the delirium and no-delirium groups (P = 0.02 by the Kolmogorov–Smirnov test) with the largest cumulative probability difference appearing at the biomarker value of 32.05 pg/ml.
These results suggest that patients who subsequently developed delirium are more likely to be experiencing clinically silent neurodegenerative changes before surgery, reflected by changes in plasma neurofilament light chain biomarker concentrations, which may identify individuals with a preoperative vulnerability to subsequent cognitive decline.
Postoperative delirium is common in older patients
Cognitive impairment and frailty are known predictors for the development of postoperative delirium
Preoperative plasma neurofilament light may be useful for identifying patients at risk for developing postoperative delirium
Postoperative delirium is a common yet serious cognitive condition that affects 10 to 60% of patients after major surgery.1 Delirium is an acute confusional state defined by alterations in attention, consciousness, and disorganized thinking.1,2 Delirium has also been shown to be associated with decreased long-term physical and cognitive functioning.3,4 Patients and families frequently are concerned that exposure to major surgery and anesthesia will result in postoperative delirium, which has been hypothesized as a prodrome for subsequent long-term cognitive decline. Understanding the pathophysiology of postoperative delirium and its association with long-term cognitive changes is important for designing the appropriate research studies that address the mechanism for the association. We hypothesized that patients who subsequently developed postoperative delirium have a pre-existent brain vulnerability secondary to prodromal neurodegenerative changes that can be tracked with biofluid markers of neurodegeneration. Although a preoperative cognitive screen may identify risk for older individuals, neuropathology exists even before the onset of cognitive changes in about 30% of cases.5 Previous studies have focused on neuroinflammation and biomarkers that may be elevated after surgery. However, few studies have evaluated whether preoperative biomarkers of brain vulnerability contribute to postoperative cognitive events. Of the studies that assessed the association between preoperative biomarkers and postoperative delirium, conflicting results have been found to be caused by small sample sizes, heterogeneous biomarkers being examined, and methods of statistical analysis.6–10 Accordingly, we conducted a study to examine the prevalence of established markers of neurodegeneration in older surgical patients undergoing elective major surgery and to determine whether the presence of preoperative biomarkers of neurodegeneration is associated with postoperative delirium.
Materials and Methods
The study was approved by the institutional review board for human research and informed consent was obtained preoperatively from each study patient. The study was conducted at the University of California, San Francisco Medical Center between January 2002 and December 2010. The data for this study were collected from two separate studies. One study assessed the effects of nitrous oxide on postoperative delirium, showing that there was no effect of nitrous oxide, the intervention, on postoperative delirium.11 The other study was an observational study of risks associated with postoperative delirium.12 Inclusion criteria for both studies were identical, including consecutive men or women who were 65 yr of age or older, undergoing major noncardiac surgery requiring general anesthesia, who were expected to remain in the hospital postoperatively for more than 48 h and also had plasma banked preoperatively for the biomarkers assay. Additional inclusion criteria for this study included preoperative banking of blood and separation of plasma that was stored at the appropriate temperature. The exclusion criteria for both studies were patients who could not complete the delirium testing, such as those who were expected to remain intubated postoperatively, particularly if they would be sedated for postoperative ventilation. Patients were not excluded based on their preoperative cognitive performance, and our study cohort was largely cognitively unimpaired. The demographic data of the two studies, which included age, sex and preoperative cognitive status, was similar between the two studies; hence, the data were combined.
For both studies, preoperatively on the day of surgery, blood was collected by phlebotomy in ethylenediaminetetraacetic acid tubes, and centrifuged at 2,500g for 10 min at room temperature. The plasma was then aliquoted in 1.5-mL cryogenic tubes and stored at ˗80°C until analyses. For blood samples in which the plasma was isolated and stored, the plasma was analyzed according to vendor protocols, for neurofilament light chain, glial fibrillary acidic protein, amyloid β1–42 (Aβ42), and phosphorylated Tau 181 (P-Tau181), using commercially available kits for single molecule arrays in an HD-X analyzer (Quanterix, USA). The lowest level of quantification and average coefficient of variations were 0.4 pg/ml, 4.5% for neurofilament light chain; 0.38 pg/ml, 2.7% for Aβ42; 2.8 pg/ml, 7.4% for glial fibrillary acidic protein; and 0.08 pg/ml, 5.4% for P-Tau181, respectively. The analyzed samples underwent only one thaw cycle before use. The samples were run in duplicate, with kits from the same lot by an investigator blinded to group allocation.
P-Tau181, neurofilament light chain, Aβ42, and glial fibrillary acidic protein were chosen in this study because they are robust markers associated with neurodegeneration detectable in blood. Their clinical utility as markers of neurodegeneration has been validated in multiple large cohorts. High P-Tau181 discriminates individuals with underlying Alzheimer’s disease pathology. Blood P-Tau181 correlates with both amyloid plaque and neurofibrillary tangle burden, as detected by florbetapir or flortaucipir brain positron emission tomography imaging, and with Tau brain deposition as determined by Braak neuropathological staging. Blood P-Tau181 is elevated in the setting of Alzheimer’s disease pathology, including asymptomatic individuals.13 Neurofilament light chain is a sensitive but nonspecific marker of axonal injury. Blood neurofilament light chain is elevated in individuals with Alzheimer’s disease and frontotemporal dementia.14 High blood neurofilament light chain is increased in prodromal stages of Alzheimer’s disease15 and has also been detected in asymptomatic individuals at short-term risk of progression to symptomatic frontotemporal dementia.16,17 Blood neurofilament light chain concentrations are responsive to inflammatory disease activity and therapeutic interventions.18 Aβ42 is found in amyloid plaques, and it is believed that Aβ42 triggers pathologic changes in Tau and that at later stages Tau becomes unyoked from Aβ42.19 Glial fibrillary acidic protein is a marker of astroglial activation that mirrors inflammation at different stages of neurodegeneration.20 Glial fibrillary acidic protein has shown to be a marker of white matter integrity and executive function in cognitively intact older adults.21
For both studies, baseline cognitive status was measured preoperatively using the Telephone Interview of Cognitive Status instrument,22 which was adapted from the Mini Mental State Examination. For the occurrence of delirium, we used the Confusion Assessment Method rating scale,23 which was developed as a screening instrument based on operationalization of DSM-III-R criteria for use by nonpsychiatric clinicians in high-risk settings. The presence of postoperative delirium was based on Confusion Assessment Method criteria, which include the following three types of symptoms: (1) acute onset of change in mental status compared to the pre-operative Confusion Assessment Method assessment, (2) inattention, and (3) either disorganized thinking or altered level of consciousness. The Confusion Assessment Method assessments were conducted by research associates whose training was guided by the Confusion Assessment Method Training Manual and Coding Guide and whose assessments were validated by either Dr. Leung or Dr. Sands. The Confusion Assessment Method was administered before surgery and daily after surgery for up to 3 days. No patients met Confusion Assessment Method criteria for delirium before surgery. The primary outcome was incident delirium on any of the first 3 postoperative inpatient days. For the clinical trial, both patients and the research associates were blinded as to whether the patient was in the intervention or control group. The intervention was shown not to affect incident delirium.
Assessment of Descriptive Characteristics
Preoperative characteristics are shown in previous research to increase risk for postoperative delirium in older adults undergoing elective surgery.11,24,25 The characteristics included age, sex, history of cerebrovascular disease, the American Society of Anesthesiologists risk score, and surgical risk as shown in table 1. Surgical risk was estimated using the guidelines from the American College of Cardiology and the American Heart Association update for the perioperative cardiovascular evaluation for noncardiac surgery.26 Each of the characteristics was collected either at the preoperative interviews or abstracted from medical records.
The distributions of the preoperative characteristics biomarkers by delirium status were examined using descriptive statistics, including means, standard deviations, and percentage with the preoperative characteristic. To assess whether characteristics differed between those with and without delirium, we computed t tests for continuous valued variables and chi-square tests for categorical variables.
Our initial cohort consisted of 189 patients who did not develop postoperative delirium (no-delirium group) and 102 patients who developed postoperative delirium (delirium group). However, previous studies27,28 have shown age and sex differences in expressions of biomarkers associated with cognitive functioning. Therefore, we conducted a propensity score matching on age and sex with nearest neighbor,29 such that each patient in the delirium group was matched with a patient in the no-delirium group. To examine the age and sex balance in the resulting matching sample, we adopted two commonly used graphical tools30 : the density or histogram plot and the Love plot. The first tool plots, for a continuous covariate, the probability densities of the covariate or for a categorical covariate the side-by-side proportion histograms of the covariate, before and after the matching adjustment. To make the Love plot, a standardized mean difference30 is first calculated for each matching variable, continuous or categorical, before and after the matching adjustment. A standardized mean difference with the absolute value greater than 0.1 is considered as an indicator of imbalance. The Love plot then presents the standardized mean differences per matching covariate laid out in a horizontal fashion with the two vertical lines of standardized mean difference = 1 and standardized mean difference = ˗1 superimposed. Covariates with standardized mean differences falling out of these two lines are considered unbalanced. Once covariate balance was achieved through the matching adjustment, we then performed all the statistical analyses on these matched no-delirium and delirium groups, each with 102 patients (fig. 1).
For each biomarker, we first constructed a group-wise relative frequency plot, visually comparing the relative frequencies of the biomarker measurements for the delirium and no-delirium groups. As shown in the relative frequency plots, none of the four biomarkers could be considered as normally distributed. Therefore, nonparametric statistical methods that do not assume that the data belonged to a specific distribution family are appropriate here. Table 2 shows the median and interquartile range of each biomarker; in addition, table 2 provides the results from the Kolmogorov–Smirnov test, a nonparametric statistical test that we used to numerically compare the biomarker measurements between the no-delirium and delirium groups. The Kolmogorov–Smirnov test examines the entire distribution of each biomarker. This test first calculates the empirical cumulative distribution function for the biomarker measurements of each group. Each empirical cumulative distribution function, considered as an estimate of the true but unknown cumulative distribution function, is a nondecreasing step function on the range of biomarker measurements such that the height of the step at a biomarker value x represents the cumulative probability of the biomarker taking a value no greater than x. After obtaining the empirical cumulative distribution functions for the two groups, the Kolmogorov–Smirnov test takes the absolute difference of the two empirical cumulative distribution functions, which is also a function on the range of biomarker measurements. The test then claims a significant difference between the distributions of the biomarker measurements for the two groups when the supreme or maximum of this absolute difference function is greater than the critical value determined by the test procedure. To gain more insight on the biomarker difference between the two groups, we also made graphical representations of the Kolmogorov–Smirnov test for each biomarker. In addition to showing the test result (rejection or failure to reject), the graph also shows at which subrange(s) of biomarker values the delirium and no-delirium groups differ the most.
Due to budgetary constraints, we processed biomarkers for only 306 of the most recently recruited 809 patients who met the inclusion criteria for the current study. However, missing data in key covariates resulted in 291 patients who were included in the present analysis (fig. 1). Of 291 patients, 102 (35%) developed postoperative delirium (table 1). The propensity score matching resulted in two groups: a delirium and a no-delirium group, that were matched on age and sex. Each group included 102 patients (fig. 1). For the age covariate, the density plot (fig. 2A) showed more similar densities after the matching. The Love plot (fig. 2B) clearly showed age was slightly unbalanced before matching but became balanced after matching. For sex, it was clearly unbalanced in the original sample with a higher proportion of females in the delirium group. This imbalance was corrected in the matched sample, in which the proportions of males and females were similar in both patient groups.
For each biomarker, we assessed whether there were significant differences between the delirium and no-delirium groups. Table 2 shows that the Kolmogorov–Smirnov test for P-Tau 181 is not significant (P = 0.29), indicating that the distributions of P-Tau181 values for the no-delirium and delirium groups are not significantly different. Figure 3A and B shows that the two groups differ the most when the P-Tau181 value is 1.74, but the difference (shown in red) is far below the critical value (shown in blue). The Kolmogorov–Smirnov test for the biomarker neurofilament light chain (table 2) shows significant differences between the no-delirium and delirium groups (P value = 0.02). The median value for neurofilament light chain was 32.05 pg/ml for the delirious group versus 23.7 pg/ml in the no-delirium group. The distribution of biomarker values significantly differed between the delirium and no-delirium groups (P value = 0.02 by the Kolmogorov–Smirnov test) with the greatest difference appearing at the biomarker value of 32.55 pg/ml (fig. 3C and D). The no-delirium group had more values below this threshold compared to the delirium group. The plot also showed that the two groups differed mostly in the region of lower neurofilament light chain biomarker values. The Kolmogorov–Smirnov test for the biomarker glial fibrillary acidic protein is not significant, revealing that the distributions of glial fibrillary acidic protein values for the delirium and no-delirium groups are not significantly different. Figures 3E and F shows that the two groups differ the most when the glial fibrillary acidic protein value is 94.85 pg/ml, where the difference is far below the critical value (shown in blue). The Kolmogorov–Smirnov test for the biomarker Aβ42 is not significant, meaning that the distributions of Aβ42 values for the two groups are not significantly different. Figures 3G and H shows that the two groups differ the most when the Aβ42 value is 7.9 pg/ml, where the difference is far below the critical value (shown in blue).
Because most perioperative pathways in the prevention of postoperative delirium include the performance of a preoperative cognitive screen, we performed a secondary data analysis to determine whether preoperative cognitive status is associated with different levels of preoperative biomarker levels. The results reveal very low correlations between preoperative cognitive status measured by the Telephone Interview of Cognitive Status instrument and each biomarker. Respectively, the Pearson’s correlation coefficients for each biomarker were ˗0.08 (P-Tau181), ˗0.17 (neurofilament light chain), ˗0.08 (glial fibrillary acidic protein), and 0.02 (Aβ42).
In this prospective cohort study of older patients undergoing major noncardiac surgery, we found that one of the four analytes yielded significant results. The distribution of neurofilament light chain values significantly differed between the delirious and no-delirium groups. Specifically, the no-delirium group had more values below a threshold value compared to the delirious group. These results suggest that patients who subsequently developed delirium may be more likely to be experiencing clinically silent neurodegenerative changes before surgery.
There is a robust literature discussing the relationship between biomarkers detected in the cerebrospinal fluid (CSF) and plasma and Alzheimer’s disease and related dementias31–33 ; however, this was not the focus of the present study. Rather, the primary outcome of our study is postoperative delirium. Another major distinction between previous studies and our present proposal is that we focus on preoperative baseline biomarkers and not changes in biomarker levels with surgery given the goal of our study is risk identification and not the effects of surgery. Most studies that evaluated changes in biomarkers after surgery had limited samplings of biomarkers, typically on only 1 postoperative day, heterogeneous statistical analyses, and it is unclear whether these biomarker changes were temporary and whether changes in biomarkers after surgery are associated with long-term consequences. Furthermore, previous studies of change in biomarkers did not investigate the complexity of causes for change in biomarkers, which could be due to reasons other than perioperative procedures. For example, even in the absence of surgery, older adults hospitalized for acute illness are at risk for long-term cognitive changes.34
A recently published systematic review on biomarkers of delirium concluded that there was insufficient evidence to support the use of any biomarker as a sole risk or disease marker of delirium.35 Studies that investigated preoperative biomarkers showed conflicting results (table 3). For example, for neurofilament light chain, Halaas et al.7 found an association between preoperative levels and postoperative delirium, and Fong et al.6 found an association between neurofilament light chain and the sum of scores from all days in the hospital on the Confusion Assessment Method severity scale but not for incident postoperative delirium or days of delirium. However, Casey et al.9 found no such association. These differences in results may be secondary to different patient cohorts or heterogeneous ways to analyze the results statistically. As for T-Tau, two studies in noncardiac surgical patients found no association between preoperative levels and postoperative delirium,6,40 but one small study in cardiac surgical patients found an association.10 As for glial fibrillary acidic protein, Ballweg et al.38 found no association between preoperative glial fibrillary acidic protein levels and postoperative delirium, a result similar to what we reported here. A recent study reported new Alzheimer’s disease biomarkers such as phosphorylated Tau at threonine 217 (Tau-PT217) and 181 (Tau-P-T181) to be associated with increased risk of POD.39 Finally, for Aβ42, conflicting results are found. Although a study of hip fracture patients showed that CSF Aβ42 levels were not significantly different between groups,8 other studies in patients undergoing elective arthroplasty showed that low CSF Aβ42 levels were associated with postoperative delirium40 or that those in the lowest quartile of preoperative CSF Aβ40/Tau and Aβ42/Tau ratio had a higher incidence of postoperative delirium.36 One additional study showed that delirious patients had lower ratios of Aβ42 to T-Tau relative to those without delirium.37 The previous investigations involved samples collected from CSF, and the association of plasma Aβ42 with postoperative delirium is less well investigated.
A recent review from the Alzheimer’s Association concluded that blood based markers have promise to revolutionize the diagnostic and prognostic work-up of Alzheimer’s disease and to improve the design of interventional trials.41 Several of the biomarkers investigated in the present study such as plasma P-Tau, Aβ42, neurofilament light chain, and glial fibrillary acidic protein have been proposed to be important markers that should have longitudinal measurements in prospective cohort studies.41 For many years, CSF neurofilament light chain has been used as a neuroaxonal injury marker. Its level is elevated in cognitively asymptomatic patients at risk of neurodegenerative dementia.16,42,43 Neurofilament light chain is a biomarker that has been found to be associated with myriad neurologic conditions including both peripheral nerve disorders and central nervous system disorders from traumatic brain injury to multiple sclerosis and Parkinson’s disease. The association of higher levels of neurofilament light chain preoperatively with postoperative delirium suggests that patients may show evidence of neurodegeneration. In contrast, postoperative neurofilament light chain elevation shown by Fong et al.6 may have a different etiology, such as neuronal injury of some kind. Our study differs from that of Fong et al.,6 as we aimed to determine preoperative vulnerability of patients rather than detecting the effect of surgery, which may be rather nonspecific due to the general inflammatory response after major surgery. It should be noted that neurofilament light chain has a strong age relationship.41 In our analysis, we used propensity score nearest neighbor matching for age and sex, so the significant association between preoperative neurofilament light chain and postoperative delirium persists even after considering age and sex. Glial fibrillary acidic protein, on the other hand, is a nonspecific marker of astroglial activation.44 Its blood concentration is strongly reflective of Aβ accumulation in the brain before neuronal damage or in response to neuronal damage in dementia. P-Tau181, a marker of Alzheimer’s disease that is elevated in prodromal states,13 did not have strong associations with delirium. Whether this signals a lack of association between Alzheimer’s disease and postoperative delirium or a low relevance of Tau biology in delirium deserves further investigation. A limitation of this study is that it did not assess P-Tau217 or 231, which have much stronger associations with brain amyloidosis than P-Tau181 in preclinical Alzheimer’s disease. Furthermore, our study cohort also did not consist of patients with significant cognitive impairment or dementia, which might have precluded them from undergoing major elective surgery.
Our present results suggest that incipient neurodegeneration may have been present in patients with seemingly intact cognition before surgery as measured by the Telephone Interview of Cognitive Status instrument and medical records for a diagnosis of dementia. If evidence of neurodegeneration exists before surgery as our present results suggest, then postoperative delirium may only be a surrogate marker of what is to come; that is, a predisposition of developing long-term cognitive decline including conversion to Alzheimer’s disease. Our present results support that delirium may be an intermediary outcome, not independent of pre-existent vulnerability, as evidenced by markers of neurodegeneration.
Taken together, our results and those from other investigators show that there is a role of using proteomics in the investigation of the pathophysiology of perioperative cognitive changes. Our study results are novel in that we focus on baseline patient vulnerability rather than the effects of surgery. However, our results should be considered as preliminary, even though we have a relatively large sample size, and we only examined four biomarkers. Future investigations should consider other molecular biomarkers that may be upstream from the neurodegeneration markers that are considered to be terminal neuropathology, not easily amendable to modification.
There are several potential limitations of our study. First, we included only four biomarkers in this study, and newer Alzheimer’s disease biomarkers have been recently reported.45 Second, there is a concern that long-term storage may affect the reliability of blood biomarkers for Alzheimer’s disease and neurodegeneration. However, a recent study reported that Aβ40, Aβ42, TTau, and neurofilament light chain can be measured from serum or plasma stored up to 20 yr at 80°C with only small variability in concentration.46 Third, this is a single-center study, and the results will need to be validated by future studies including other cohorts. Last, we have not conducted long-term follow-up to evaluate whether patients with preoperative evidence of neurodegeneration will have a greater decline in cognition, an area that we will pursue in future studies.
In summary, we have preliminary evidence to suggest that biofluid markers of neurodegeneration measured before surgery may have prognostic significance in predicting postoperative delirium. Plasma biomarkers may have value in monitoring the preclinical phases in neurodegenerative disease. These results need further confirmation along with long-term follow-up.
The authors thank the Perioperative Medicine Research Group for their assistance with patient recruitment, interviews, and data collection.
Supported in part by National Institutes of Health (Bethesda, Maryland) grants R21AG048456 and 1R01NR017622-01A1 (to Dr. Leung) and K23AG059888 (to Dr. Rojas).
Dr. Rojas is the site principal investigator for clinical trials sponsored by Eli Lilly (Indianapolis, Indiana) and Eisai/Boxer (Bunkyo City, Tokyo, Japan), and has served as a consultant to AGTC (Alachu, Florida), Alector (South San Francisco, California), Alzprotect (Loos, France), Amylyx (Cambridge, Massachusetts), Arkuda (Watertown, Massachusetts), Arrowhead (Pasadena, California), Arvinas (New Haven, Connecticut), Aviado (Garden Grove, California), Boehringer Ingelheim (Fremont, California), Denali (San Francisco, California), Eli Lilly, GSK (Philadelphia, Pennsylvania), Humana (Louisville, Kentucky), Life Edit (Durham, North Carolina), Merck (Rahway, New Jersey), Modalis (Waltham, Massachusetts), Oligomerix (White Plains, New York), Oscotec (Seongnam-si, Korea), Roche (Indianapolis, Indiana), Transposon (Laguna Niguel, California), and Wave (Cambridge, California). He has received research support from Biogen (Cambridge, California) and Eisai for serving as a site investigator for clinical trials, as well as from Regeneron (Tarrytown, New York). He has received research support from the National Institutes on Aging, National Institutes of Health through grants U19AG063911, R01AG073482, R56AG075744, R01AG038791, RF1AG077557, R01AG071756, and U24AG057437 and from the Rainwater Charitable Foundation (Fort Worth, Texas), Bluefield Project to Cure Frontotemporal Dementia (San Francisco, California), GHR Foundation (Minneapolis, Minnesota), Alzheimer’s Association (Chicago, Illinois), Association for Frontotemporal Degeneration (King of Prussia, Pennsylvania), Gates Ventures (Kirkland, Washington), Alzheimer’s Drug Discovery Foundation (New York, New York), UCSF Parkinson’s Spectrum Disorders Center (San Francisco, California), and the University of California Cures Alzheimer’s disease Program (San Francisco, California). The other authors declare no competing interests.
Appendix: Perioperative Medicine Research Group
Principal investigator: Jacqueline M. Leung, M.D., M.P.H.
Research associates: Christopher Tang, B.A., Devon Pleasants, B.S., Sanam Tabatabai, B.S., Danielle Tran, B.S., Stacey Chang, B.A., Gabriela Meckler, B.A., Stacey Newman, B.A., Tiffany Tsai, M.D., Vanessa Voss, M.D., Emily Youngblom, B.A.