Structural disconnectivity in postoperative delirium: A perioperative two‐center cohort study in older patients

Abstract BACKGROUND Structural disconnectivity was found to precede dementia. Global white matter abnormalities might also be associated with postoperative delirium (POD). METHODS We recruited older patients (≥65 years) without dementia that were scheduled for major surgery. Diffusion kurtosis imaging metrics were obtained preoperatively, after 3 and 12 months postoperatively. We calculated fractional anisotropy (FA), mean diffusivity (MD), mean kurtosis (MK), and free water (FW). A structured and validated delirium assessment was performed twice daily. RESULTS Of 325 patients, 53 patients developed POD (16.3%). Preoperative global MD (standardized beta 0.27 [95% confidence interval [CI] 0.21–0.32] p < 0.001) was higher in patients with POD. Preoperative global MK (−0.07 [95% CI −0.11 to (−0.04)] p < 0.001) and FA (0.07 [95% CI −0.10 to (−0.04)] p < 0.001) were lower. When correcting for baseline diffusion, postoperative MD was lower after 3 months (0.05 [95% CI −0.08 to (−0.03)] p < 0.001; n = 183) and higher after 12 months (0.28 [95% CI 0.20–0.35] p < 0.001; n = 45) among patients with POD. DISCUSSION Preoperative structural disconnectivity was associated with POD. POD might lead to white matter depletion 3 and 12 months after surgery.


INTRODUCTION
Postoperative delirium (POD) is a highly prevalent complication after surgery and predominantly affects older patients. 1The exact pathomechanisms and neural substrates of POD are uncertain, and the subject of ongoing research, 2 much of which involves perioperative neuroimaging.A range of imaging techniques have identified various brain structures as potential locations of pathology underpinning POD.6][7] Such abnormal connectivity of brain networks appears to not only play a role in contributing to the risk of POD, but also provide a biomarker of POD itself. 8We hypothesized that these fMRI findings might be caused by (or predispose to) underlying structural white matter changes. 6e study examined whole-brain diffusion tensor imaging (DTI) prior to surgery. 9It revealed higher global mean diffusivity (MD) and radial diffusivity (RD) in patients who subsequently developed POD.
Notably, this exploratory analysis was conducted post-hoc and in a rather small sample of only 68 patients, of whom 17 had POD.In another study, longitudinal diffusion changes in periventricular, frontal, and temporal white matter were found to be associated with POD incidence and severity in 113 patients.POD was present in 25 patients.
Therein, higher MD was shown to be the most robust DTI parameter associated with POD. 10 The additional acquisition of postoperative diffusion weighted imaging (DWI) data is particularly beneficial as it allows insights into microstructural characteristics and development of POD and enables us to identify potential neurodegenerative progression after episodes of POD.These neurodegenerative changes might replicate or predispose to those preceding dementia.For example, baseline structural disconnectivity in terms of higher MD and lower FA was described to be associated with the development of dementia, 11 and dementia has been identified as a strong risk factor for delirium. 12Furthermore, POD might not only worsen cognitive trajectories, but also aggravate or precipitate clinical dementia. 13,14Hence, commonalities in neuroimaging findings between POD and dementia might help to understand potential overlaps between these two distinct but interrelated conditions.
To assess the cerebral microstructure, we calculated four diffusion kurtosis imaging (DKI) metrics: MD, FA, mean kurtosis (MK), and free water (FW).Taken together, these metrics are reflective of axonal neuronal and fiber integrity (FA, MK) as well as neurite density (MD). 4,15They further indicate abnormally high, undirected diffusivity in case of microstructural damage (MD). 4Higher MD and lower FA as signs of abnormal white matter integrity were frequently associated with dementia. 16Extracellular FW increases with progressing neuronal atrophy and was observed to be associated with cognition. 17Cortical FW was found to first decrease in the early preclinical stage and to later rise during the late preclinical and symptomatic stages of Alzheimer's disease (AD). 18The properties of FW might help to distinguish between mild cognitive impairment and AD. 19 sought associations between these DKI metrics and the occurrence of POD, as a means of identifying global and local microstructural abnormalities that predispose to POD.To detect postoperative and longitudinal microstructural abnormalities, we undertook an analysis at a follow-up 3 and 12 months after surgery by adjusting for preoperative DKI values.Our overarching hypothesis was that patients with POD will show signs of structural disconnectivity resembling those detected in early stages of dementia.We expected higher MD and FW, whereas FA and MK may be lower compared to surgical patients without POD preoperatively, 3 and 12 months after surgery.from 325 participants were deemed suitable for our analysis after imaging exclusions shown in Figure 1; 39 patients from Utrecht and 286 from Berlin, to form the final study cohort.

Outcome assessments
POD assessment was performed in accordance with the 5th edi-

Imaging acquisition
Whole-brain T1-weighted (T1w) structural MRI and diffusionweighted imaging were performed within 2 weeks before surgery.

RESEARCH IN CONTEXT
1. Systematic review: We reviewed the literature by using traditional databases.Human diffusion-weighted neuroimaging studies into delirium and dementia are discussed and cited in the paper.

Imaging preprocessing and feature extraction
Imaging data from all participants were preprocessed with a pipeline designed in-house 23 (Figure 2).Diffusion weighted MR images were first denoised via MPPCA, 24 and then corrected for Gibbs ringing artifacts, 25,26 eddy current distortions as well as head motion, 27 and lastly field inhomogeneities. 28Diffusion kurtosis tensors 29 were fitted with dipy to compute fractional anisotropy (FA) and MD maps. 30Subsequently, FA maps were non-linearly registered 31 to the JHU-ICBM atlas 32 and transformations found were applied to backproject 31 the JHU region atlas to patient-specific diffusion MRI space.Thereafter, mean diffusion metrics were computed for all 48 atlas regions, providing a total of 96 image derived phenotypes.

Statistical analysis
We used linear mixed-effects models to analyze association of POD with four DKI metrics (MD, FA, FW, and MK) at three time points (preoperative, 3 and 12 months) after surgery in surgical patients, resulting in twelve models.In each model, the DKI metric was treated as the dependent variable and POD as one of the independent variables of interest, as well as age, sex, baseline MMSE, comorbidity (Charlson's comorbidity index), JHU atlas region, and scanner type as covariates of no interest and a random intercept for patient.We additionally included the interaction of JHU atlas region and POD as variables to model associations of POD with regionally limited white matter alterations.The number of POD by region interactions was reduced using backward stepwise regression removing interactions using a significance criterion of p > 0.05.To account for multiple comparisons, we applied a conservative Bonferroni correction.The initial alphalevel was divided by the number of regressions models, resulting in an adjusted threshold of 0.0042 (0.05/12).DKI metrics do not have SI units.Therefore, we provide the standardized beta for global DKI metrics.This also helps to compare the metrics with each other.
To model postoperative changes in white matter rather than crosssectional associations with POD in the data collected 3 and 12 months after surgery, we added preoperative DKI metrics as an additional independent variable into the models of DKI metrics collected at follow-up visits.
Due to the high rate of cases that were loss-to-follow-up after 12 months, we were prompted to perform a sensitivity analysis which only contained patients with DKI data for all three time points.We then compared the results to our main analyses.
During the analysis stage, two outliers were identified, whose DKI metrics abnormally differed from the rest of the cohort.This phenomenon might be due to a history of major cerebral infarction.To ensure the robustness of the results, we decided on performing a sensitivity analysis without the respective outliers.
All analyses were conducted in R v4.3.0 with additional use of the packages lme4 and lmerTest (see Supplemental R code).See the Supplemental information for the sample size calculation.

RESULTS
The final cohort in this study included 325 patients with preoperative neuroimaging; 39 from Utrecht and 286 from Berlin.These participants did not differ in age, sex, MMSE score, or delirium status from those who did not reach eligibility criteria or whose DKI data were invalid for analysis (Table S1).

Preoperative DKI
A total of 324 patients with complete data were analyzed.3).Beyond global effects, several regional associations of POD with preoperative white matter alterations were observed (see Table 3): MD and FW were particularly lower in the left tapetum.

Postoperative DKI at 12 months follow-up after surgery
The data of 45 patients were available for the 12 months followup, of whom were all patients enrolled in Berlin.The median age TA B L E 1 Patient characteristics at baseline before surgery and information on anesthesia and surgery for the entire sample and by subgroup for those who developed postoperative delirium (POD) and for those who did not develop POD.

All N = 325
No  3).Beyond global effects, several regional associations of POD with preoperative white matter alterations were observed (see Table 4): MD was particularly higher in the left anterior limb of the internal capsule, anterior corona radiata,

Sensitivity analysis of patients with available data after 12 months
We performed the same analysis for all time points with patients that had available data for the 12-month follow-up.As with the entire cohort, preoperative MD (standardized beta 0.10 [95% CI 0.03-0.16]p = 0.007) was higher and FA (standardized beta −0.17  Analyses of the data after 12 months were not affected by outliers since the two patients with major stroke did not return for follow-up at this time point.

DISCUSSION
DKI revealed signs of structural disconnection (lower FA and MK, higher MD) were present in patients that developed POD in com- Note: Regions with more pronounced effects as estimated from the global association are highlighted in bold/italic.The table illustrates results for preoperative data.The number of POD by region interactions was reduced using backward stepwise regression removing interactions using a significance criterion of p > 0.05.Please note that the corrected alpha-level was p < 0.0042.The significant regions are in bold letters.
TA B L E 3 Regional associations of POD with postoperative white matter alterations at follow-up 3 months after surgery as indicated by interaction of JHU atlas region and POD.Note: Regions with more pronounced effects as estimated from the global association are highlighted in bold/italic.The table illustrates results for the 3 months follow-up.The number of POD by region interactions was reduced using backward stepwise regression removing interactions using a significance criterion of p > 0.05.Please note that the corrected alpha-level was p < 0.0042.The significant regions are in bold letters.

Metric
TA B L E 4 Regional associations of POD with postoperative white matter alterations at follow-up 12 months after surgery as indicated by interaction of JHU atlas region and POD.FW might not be helpful in identifying brain structural correlates of POD. 4 Although preoperative FA and MK were associated with the later occurrence of POD, they were not associated with POD after surgery.The use of FA was recently discouraged due to the impaired interpretability in regions of white matter fiber crossings. 15Instead, MD was recommended.Higher preoperative MD appears to be a common observation across most delirium-related diffusion MRI studies.

Metric
This applies to areas of gray and white matter alike. 3,4,9In a longitudinal study determining perioperative changes in diffusion 1 year after surgery, FA was lower and MD higher in patients with POD.However, MD was reported to be the most robust marker for a longitudinal study setting. 3The higher preoperative global MD of POD patients is in line with previous cross-sectional findings, whereas lower global FA was not associated with POD in an exploratory study. 9This discrepancy might be caused by voxel-based approaches in past publications, which included gray matter areas, and the post-hoc design with a rather small sample size.
The inclusion of the preoperative DKI metrics to the postoperative models enables us to draw conclusions about the longitudinal development of white matter abnormalities of POD patients in comparison to patients without POD.Furthermore, our study is the first to analyze diffusion-weighted data 3 months after surgery, whereas other studies have only focused on diffusion changes after 12 months. 10The analysis of postoperative data after 3 months suggested paradoxically lower global MD in patients with POD.Since these alterations were observed after correction for baseline values, it seems reasonable to assume that these findings are sequelae of POD.Lower postoperative MD values in POD patients could perhaps be ascribed to pseudo-normalizations which were reported in a longitudinal study on traumatic brain injury. 33is sequence of events suggests an initial increase in MD due to an acute insult which might be related to surgery.In the subacute stage, MD pseudonormalizes, but the white matter insult is now responsible for persistent white matter loss.However, to detect an acute insult that correlates with a postsurgical increase in MD, neuroimage acquisition in the early days after surgery is required.Further studies are needed to better contextualize this finding.
Although the loss to follow-up was high at 12 months after surgery, we again observed globally increased MD in patients who had experienced delirium even after adjustment for baseline MD.The pattern of particularly altered brain regions after POD resembled the preoperatively observed pattern, comprising the left superior-frontooccipital fascicle and the left fornix.Due to the small resulting sample size and the low remaining fraction of patients with POD at the 12 months follow-up, the results for this time point must be cautiously interpreted.Furthermore, we would refrain from drawing conclusions about overlaps between structural disconnectivity patterns 1 year after episodes of POD and the patterns preceding dementia. 11However, it is worth noting that global MD was higher across both study populations.
Structural disconnectivity may represent diffuse brain damage that leads to increased vulnerability to both delirium and dementia.Furthermore, regional hypometabolism was discussed earlier as a potential link between delirium and dementia. 34A brain characterized by white matter abnormalities to a degree of structural disconnection might be incapable of buffering stressors such as hypometabolism as they occur during surgery induced by anesthetics. 35A principal contribution of this study to the literature is to increase our understanding of the consequences of delirium.We have shown that episodes of POD might be detrimental for white matter integrity.
Together with the loss of gray matter volume after episodes of POD, 36 our findings could represent the structural correlate of worsening cognitive trajectories in POD patients. 37Future studies need to investigate whether this white and gray matter depletion after episodes of delirium may lead to cognitive decline and eventually the development of dementia.
We further ascertained the robustness of our results by excluding two outliers that might have obfuscated our results.This is the first neuroimaging study into delirium to assess the impact of outliers.However, we recognize the need for a systematic investigation of the impact of extreme data points, especially for smaller studies.Despite our large sample size, the removal of two subjects with major stroke slightly affected our findings for FA and FW after 3 months, potentially limiting the reliability for these two modalities.However, results for MD and MK seemed relatively robust.
Taking imaging studies on delirium together, we must acknowledge that POD might be caused by variety of indicators of brain health.However, our DKI findings might help to translate imaging findings into clinical practice.In particular, MD values are easy to obtain.Preoperatively, patients at risk of POD could potentially be identified through their elevated global MD.However, whether our results can be translated into clinical practice (e.g., to predict delirium), remains a matter to be investigated by future studies.

LIMITATIONS
Our primary limitation is that only 45 patients had available DKI data for baseline and the 12 months follow-up.Of these, only 5 patients were diagnosed with POD.The large number lost to follow-up hampers drawing conclusions about structural disconnectivity 1 year after surgery and could be a source of bias.However, the data after 1 year might help to understand the polarity inversion of MD among patients with POD.We have conducted a sensitivity analysis only with the 45 patients who have available data for the follow-up after 12 months.Preoperative results resemble those of the entire cohort, only MK was not significantly associated with POD anymore.Noteworthy, only 41 of the 45 patients received an MRI at the 3-months follow-up and of whom only three had delirium after surgery.Therefore, we are convinced that this part of the analysis does not allow for drawing conclusions.Future studies should address the issue of high loss-to-follow-up rates.
We further acknowledge limitations regarding data selection.Our data were acquired across two centers in Berlin and Utrecht, which showed a strong asymmetry in collection (286 patients from Berlin, 39 patients from Utrecht).This site difference was largely due to quality control, as 107 patients from Utrecht were excluded due to image acquisition problems.This could have introduced a selection bias toward those patients who were included in the final study cohort.
However, we found no significant differences in sociodemographic and clinical factors between those patients included versus those excluded.We, therefore, suggest any influence of selection bias is minimal.
Finally, there is no current standardization of anesthesiology management.Depth of anesthesia was continuously monitored using intraoperative electroencephalogram; however, we were unable to account for possible effects of intraoperative handling.Moreover, no interrater reliability assessments were performed for POD testing, which means that deviations in delirium ratings remain a possibility.

CONCLUSION
Global functional alterations were hypothesized to be an important risk factor in POD.Microstructural abnormalities as depicted by DKI may be the structural correlate.Our findings indicate that abnormal structural disconnectivity predisposes patients to POD.By correcting for preoperative diffusion metrics, we could also show that episodes of POD may be detrimental for the cerebral microstructure.

2 . 3 .
Interpretation: Structural disconnectivity in terms of cerebral diffusion abnormalities were shown to precede dementia.These diffusion alterations might also predispose to postoperative delirium (POD) and at the same time may be a consequence of it.Future directions: This article highlights opportunities for future research.(1) Preoperative structural disconnectivity may help to identify patients at risk of POD.(2) Mean diffusivity should be assessed as potential predictor of POD.(3) The detrimental effects of delirium on white matter integrity could be linked to the worsening of cognitive trajectories.(4) Future studies need to investigate whether this white matter depletion is related to the onset of dementia after delirium.Acquisition protocols were harmonized as far as possible across the two imaging sites during study design and are described below.Berlin: Participants were scanned on a 3T Siemens Trio Tim MRI scanner (Siemens Healthcare, Erlangen, Germany).T1-weighted images were acquired with an isotropic voxel size 1.0 × 1.0 × 1.0 mm 3 , repetition time (TR) 2500 ms, and echo time (TE) 4.77 ms.DKI included one non-diffusion weighted volume (b = 0 s/mm 2 ) and 60 diffusion sensitized directions evenly distributed on two shells (b = 1000, 2500 s/mm 2 ).Images were parameterized with TR = 6500 ms, TE = 100 ms, and isotropic voxel size of 2.5 × 2.5 × 2.5 mm 3 .Utrecht: Participants were scanned on a 3T Philips Achieva scanner (Philips Healthcare, Best, The Netherlands).T1-weighted images were acquired with an isotropic voxel size 1.0 × 1.0 × 1.0 mm 3 , TR = 7.9 ms, and TE = 4.50 ms.DKI included one non-diffusion weighted volume (b = 0 s/mm 2 ) and 60 diffusion sensitized directions evenly distributed on two shells (b = 1000, 3000 s/mm 2 ).Images were parameterized with TE = 68 ms, TR = 2394 ms, and isotropic voxel size of 2.5 × 2.5 × 2.5 mm 3 .

F I G U R E 1
Consort diagram of participant inclusion.Flow chart detailing patients included within the wider BioCog study, and those included in this paper.Reasons for study drop out are given in the right-hand column.Those described as "Pausation" (n = 10) at 3-month follow-up consented to later follow-up in the study, but were not included at this time point.F I G U R E 2 Image processing.Flow chart detailing each step of the processing of diffusion-weighted images.

F I G U R E 3
Standardized estimates.This plot displays the standardized beta (dot) and the 95% confidence interval (95% CI; error bars) for the association with postoperative delirium of four diffusion kurtosis imaging (DKI) metrics: mean diffusivity (MD), fractional anisotropy (FA), mean kurtosis (MK) and free water (FW).There are three time points for each metric: preoperative (preop.), 3 months after surgery (postop.3m), and 12 months after surgery (postop.12m).Covariates were age, sex, Charlson Comorbidity index, baseline Mini-Mental State Examination, and scanner type.Baseline DKI metrics were further added to the two postsurgical analyses.Of note, after correction for multiple testing postoperative FW and MK (postop.3m) were not significantly associated with postoperative delirium.and superior frontooccipital fascicle, as well as in the right fornix and stria terminalis.
The table displays the characteristics of all patients, patients without and with delirium.For continuous variables the table states mean and standard deviation (SD) in parentheses or median and interquartile range (IQR) in parentheses.For categorical variables percentages refer to the proportion of the corresponding group (all; No Delirium; Delirium).The total N of patients with available data was added to items with cases of missing data.(ASA score ≙ American Society of Anesthesiologists' Physical Status Classification; IQR ≙Interquartile Range ≙ 25th to 75th percentile).
nificant interaction effects indicate more pronounced alterations are found especially in the fornix, left superior-frontooccipital fascicle and the left tapetum, it remains unclear whether there is a regionally distinct white matter signature for delirium.Our findings suggest that POD is rather associated with a global structural disconnectivity than TA B L E 2 Regions with more pronounced effects as estimated from the global association are highlighted in bold/italic.The table illustrates results for the 12 months follow-up.The number of POD by region interactions was reduced using backward stepwise regression removing interactions using a significance criterion of p > 0.05.Please note that the corrected alpha-level was p < 0.0042.The significant regions are in bold letters.