A trained blinded interviewer conducted the CAM-S, a structured 10-15 min clinical interview, to assess the presence and severity of POD. In cases where the patient was intubated or could not complete the extended CAM-S, the briefer CAM-ICU was administered. With the exception of fluctuation, which is identified as mild (score of 0) or marked (score of 1), all other symptoms are identified as absent (0), mild (1) or marked (2). The total severity score was based on a sum score that could range from 0 through 19. In this study, we report the peak post-operative severity score for each patient throughout their inpatient stay and the average severity score up to 5 post-operative days (unless discharged prior to 5 days). A diagnosis of full POD was based on the presence of either acute onset of change or symptom fluctuation in mental status, inattention, and either disorganized thinking or altered level of consciousness, in line with prior research. In this sample, full POD was associated with peak severity scores on at least 1 post-operative day that ranged from 8 to 14. Subthreshold delirium (ST-POD) was defined as those not meeting full criteria but displayed elevated severity scores (≥5) on the POD severity long form on at least 1 post-operative day. This clinically significant cutoff has been found to be associated with higher risk of increased length of stay, increased healthcare costs, and post-operative admittance to nursing homes, functional and cognitive decline, and death within 90 days post-operatively compared to lower scores (Inouye et al., 2014a). Patients with ST-POD also met a large proportion of criteria for full POD with the exception of one or two items. For patients who did not want to complete the extended interview, the short severity form was offered and these scores were subsequently weighted on the same metric as the long form. All patients were followed up by phone by a research assistant at least 1 month post-operatively and asked if the patient or a loved one of the patient noticed if there “was a short or long period of time after you got home that you felt your thinking had changed. For example, this could be changes in your ability to focus your attention and/or being able to keep track of what is being said to you, difficulty staying on one subject while speaking, feeling confused, or problems remembering things. This could even be seeing or hearing things that weren't really there.” Responses are descriptively reported.

Standard preprocessing of MRI EPI output was accomplished with statistical parametric mapping version 8 (SPM8) software, including batch processing by an SPM toolbox and custom written in-house MatLab scripts. The preprocessing included re-alignment of images, slice time correction, co-registration with the MPRAGE images, re-slicing to the MPRAGE dimensions for both approaches, smoothing and normalization into Montreal Neurological Institute (MNI) space and inclusive masking to assess gray and white matter distribution. Motion artifact was examined. Studies were rejected if motion over the conduct of the study period was greater than 3 mm in any plane. BOLD imaging was processed as 1st and 2nd level analysis by SPM. intra-operative data were concatenated and binned on a minute-by-minute basis.

The structural neuroimaging components from each study were reviewed by a board-certified neuro-radiologist, who indicated a Fazekas score for each patient (see below).

Other data were analyzed using SPSS. Bivariate correlations examined the relationship between cognitive summary scores and psychiatric severity scores as indicated by PHQ-9 and GAD-7 with continuous severity POD measures. If significant results were indicated, we conducted a bivariate linear regression model followed by a model controlling for age, education, sex and pre-operative baseline POD severity. We also conducted analyses of variance to examine mean score differences across primary diathesis and stress factors among those classified as non-POD, ST-POD, and full POD. Finally, to examine the within subjects change from pre-operative CO₂ delta during the stress test to intra-operative CO₂ delta by the between-subjects POD groups, we conducted a repeated measures analysis of covariance including age, education and sex as covariates, and additionally: (1) identified cognitive and psychiatric diathesis factors, (2) hypercapnic and inverse hypercapnic voxel responses in gray and white brain matter, and (3) contributory intra-operative factors. Because of restricted sample size, results are shown graphically, and trends are only discussed descriptively.

Twelve patients completed the protocol. A total of 30 patients undergoing high-risk surgeries were approached in the PAC for enrollment. Thirteen were unwilling to participate, leaving 17 who consented. The majority of patients that were approached and unwilling to participate lived in rural Manitoba, and could not make an additional trip to Winnipeg prior to their surgery. Of the remaining, 5 patients were excluded; 2 had contraindications to the MRI; 1 had surgery canceled; 1 had a scheduling conflict for MRI scan time; and 1 was too nervous to be scanned.

Individual patient characteristics are shown in Table 1 (sociodemographic and diathesis factors for all participants) and Table 2 (intra-operative stress factors for all participants). The mean age of the entire sample was 62 ± 10 years and mean education was 13 years ± 2. There were 7 males and 5 females. The mean duration of surgery was 221 ± 39 min. In 6 of 12 patients the mean BP fell below 60 mmHg on at least one occasion for a minimum of 1 min. In 2 of 12 the cerebral saturation was less than 60% for a minimum of 1 min. The mean CO₂ delta was 13.2 ± 3.0 mmHg.

The end-tidal gas control for each patient and the comparison between the non-POD patients and the combined POD and ST-POD patients is shown in Supplementary Material.

A comparison between a patient deemed low risk and a patient high risk for POD following MRI BOLD CO₂ stress testing is shown in Figures 1A,B. The expected response to the hypercapnic stimulus at the p = 0.001 level is shown in the hot voxels (shades of orange). The inverse response to the hypercapnic stimulus—that is a decrease in BOLD signal with an increase in CO₂ or vice versa is shown by the cold voxels (shades of blue). See the Supplemental Figures 1A,B for a full description of the approach. The blue voxels are indicative of regions of intracranial steal.

Figure 2A demonstrates 2nd Level Analysis in SPM comparing the non-POD group and ST-POD group. The hot voxels represent where the BOLD response to CO₂ is significantly more pronounced in the non-POD group. The cold voxels indicate where the BOLD response is significantly less in ST-POD. Figure 2B demonstrates 2nd Level Analysis in SPM comparing non-POD group and the POD group. The hot voxels represent where the BOLD response to CO₂ is significantly more pronounced in the non-POD or control group. The cold voxels indicate where the BOLD response is significantly less in the POD group. The color bar indicates significant t-values at the p = 0.05 level in both circumstances.

Axial FLAIR images were examined for evidence of leucoaraiosis—a white matter lesion associated with BOLD image CVR changes and later life dementia (Sam et al., 2016a,b). Figure 3A demonstrates an axial FLAIR image of one of the non-POD patients and Figure 3B a POD patient. The Figure 3B shows areas of hyperintensity in the periventricular regions and in the deep white matter.

Table 1 displays percentiles of each significant neuropsychological factor, and the presence or absence of psychiatric illness across all participants. Not shown, bivariate correlations revealed that certain neuropsychological measures, including Animal Fluency scaled score (r = −0.90, p < 0.001), DKEFS Color Word Interference Condition 1 (r = −0.76, p = 0.006) and Condition 2 (r = −0.73, p = 0.01) scaled scores, and CLOX 1 z-score (r = −0.83, p = 0.02) were significantly correlated with peak POD score, such that worse scores on these measures pre-operatively were associated with higher post-operative peak POD scores. Bivariate correlations also indicate Animals (r = −0.81, p = 0.002) and DKEFS Color Word Interference Condition 1 (r = −0.76, p = 0.007) and 2 (r = −0.70, p = 0.002) were significantly associated with average POD severity score. Linear regressions corroborated these findings. Multiple linear regressions controlling for age, education, sex, and baseline CAM-S revealed that Animal Fluency (β = −1.2, p < 0.001), DKEFS Condition 1 (β = −1.6, p = 0.003) and Condition 2 (β = −1.0, p = 0.01) were significantly associated with peak POD score, but CLOX 1 was non-significant. Results were corroborated for multiple linear regressions with average delirium score as the dependent variable.

With respect to psychiatric factors, results revealed that PHQ-9 summary score was bivariately correlated to average delirium severity score (r = 0.58, p = 0.049). Adjusted linear regressions were non-significant.

Intra-operative stress is summarized in Table 2 for each participant. Contributory trends emerged for the following stressors—mean blood pressure less than 60 mmHg, end-tidal CO₂ delta greater than 10 mmHg, cerebral saturation less than 60% and surgical duration greater than 120 min. Anesthetic stress defined by these factors descriptively appear to contribute to POD outcomes, acting as either a catalyst in the case of few diathesis risk factors and high intra-operative stress, or a protective factor in the case of diathesis risk and low intra-operative stress. There were no significant differences in post-operative narcotic dosing intra-operatively or post-operatively between the 3 groups (shown in Table 3).

Table 3 demonstrates mean scores of primary variables associated with POD across the 3 POD groups. Significant differences in peak POD scores and average POD scores were demonstrated for non-POD, ST-POD, and full POD groups. Corroborating previous regression findings, results also indicated significant differences across POD groups for depressive symptoms, and percentile scores (displayed for interpretability) on semantic fluency and processing speed/executive functioning on neuropsychological testing. A chi-square analysis indicated a significant difference for cerebral saturation (SAT), indicating only those with full POD had cerebral SAT less than 60%.

Figures 4A-D demonstrates graphical results from the repeated measure analysis of covariance. The Y-axis in all graphs represents marginal means, and the X-axis represents pre-operative CO₂ delta as indicated by the stress test, and intra-operative CO₂ delta. Figure 4A includes age, sex, and education as covariates in the model, demonstrating larger intra-operative CO₂ delta for ST-POD and POD compared to non-POD. Figure 4B controls for sociodemographics, significant neuropsychological factors, and presence of a psychiatric disorder. As demonstrated, intra-operative stress is attenuated for the POD group, while maintaining significant diathesis CO₂ elevation pre-operatively. Intra-operative stress of CO₂ delta is significantly higher for ST-POD, than non-POD and POD groups. Figure 4C controls for sociodemographics, and pre-operative voxel count for gray and white matter in hypercapnic response and inverse hypercapnic responses. Controlling for this diathesis results in pre-operative elevation in CO₂ delta for ST-POD and POD groups relative to non-POD, and a risk gradient for intra-operative CO₂ delta, where ST-POD demonstrated the largest marginal mean, followed by POD, and non-POD. Figure 4D controls for socio-demographics and intra-operative stressors, results dramatically change where POD has higher intra-operative CO₂ delta marginal means compared to non-POD and ST-POD groups. Figure 5 displays the proposed stress-diathesis model of POD and ST-POD incorporating these factors.

All 12 patients were successfully reached at follow-up (time since surgery range = 1 to 6 months). In the non-POD group, 5 participants reported no change in thinking, and one discussed occasional age-related lapses in short-term memory that was not noticed by loved ones. In the ST-POD group, 1 participant reported no change in cognition, and two indicated marked changes in cognition since surgery, providing examples such as “when picking up groceries I would forget what store I was in,” “I tell the same story over and over.” Of the full POD group, two participants described significant changes in memory and one did not. One POD patient reported that their spouse noticed significant cognitive changes and provided examples such as the more minor incident of “forgetting appointments,” to the more extreme of “forgetting I am taking care of a newborn.” The other POD patient was nonsensical, and had significant difficulty responding to any direct questions. This participant was able to indicate that they did not recall being home post-discharge for the first number of days. This person was assessed 4.5 months post-surgery.

The MRI BOLD CO₂ stress test (i.e., pre-operative exposure to graded CO₂ while undergoing neuroimaging) was employed as a separate indicator of risk in our stress-diathesis model to be used as a proxy for intra-operative stress. This procedure specifically involves a fixed vasodilatory stimulus that was administered to all patients—a CO₂ delta targeted to 5 mmHg during normoxic isoxia. This test has been successfully used to assess risk in other clinical scenarios. Risk stratification has been identified in patients with severe cerebrovascular compromise such as moya-moya disease, atherosclerotic stenosis or occlusion and more recently for concussion, white matter changes (leucoaraiosis) and dementia (Mandell et al., 2008; Fierstra et al., 2011; Han et al., 2011; Gao et al., 2013; Mutch et al., 2014, 2015; Sam et al., 2016a,b). All of these conditions have demonstrations of altered cerebrovascular reactivity, which become unmasked by controlled administration of CO₂ as a potent vasodilatory stimulus. In the present study, differences in response to CO₂ are evident between the patients without POD vs. those that manifested either full or ST-POD, with the most robust findings for full POD (see Figures 1, 2). What is evident from the small case series here is that POD patients have a decreased response to the hypercapnic signal in gray and white matter (Figure 1B) and a trend to greater inverse response or intracranial steal (Figures 1B, 2B). These findings have also been identified with a group of patients demonstrating leucoaraiosis (Pantoni, 2008; Figure 3). Quantification of these risks and their distribution requires an atlas of age matched surgical patients without POD to determine voxel thresholds for risk of POD. In this manner voxel count “cut points” can be determined. What is further evident from examination of these images is that there are various subgroups that help to define risk of POD. There is a group with essentially normal response to hypercapnia—both in gray and white matter—who did not demonstrate POD irrespective of the anesthetic course. Based on this finding there appears to be an identifiable low risk group defined by the CO₂ stress test. Another subgroup had an attenuated response to the hypercapnic stimulus and a greater signature for intracranial steal (both markers of cerebral dysregulation). This group is identified as high risk for POD. Whether or not these individuals manifested POD appeared to depend on both other pre-operative risk factors (e.g., neuropsychological functioning and psychiatric history) and their management during the anesthetic course (see Tables 1, 2, and further discussion below). A third subgroup had smaller signatures for CO₂ responsiveness and had a mixed response as to POD presentation. Importantly, in all patients the CO₂ delta was greater than a CO₂ delta targeted to 5 mmHg (during the CO₂ stress test) during the intra-operative course as shown in Supplemental Table 1. This suggests that our CO₂ stress test was very conservative to unmask risks of intracranial steal. Despite this, this change in CO₂ delta from the pre-operative stress test to the intra-operative period helped consolidate the stress-diathesis framework. As evident in Figure 4, including diathesis-related covariates (i.e., a) implicated neuropsychological vulnerabilities and presence of a clinically significant psychiatric profile, and b) intracranial steal responses during CO₂ stress test) greatly affect the full POD group, while including intra-operative stress covariates (i.e., duration of surgery, cerebral SAT < 60, and mean MAC) greatly affect the ST-POD group. This may suggest that intra-operative stress may play a larger role in ST-POD patients, a group with less pre-operative diathesis risk. With a larger sample size, rigorous prediction models will allow for a greater understanding of the diathesis and stressor contributions in POD.

With respect to pre-operative neuropsychological risk factors, preliminary data revealed that cognitive measures of semantic categorical fluency and information processing and speed/executive functioning were strongly associated with POD severity. This may correspond with vulnerabilities in temporal and frontal regions of the brain. Interestingly, POD severity was unrelated to measures of pre-operative phonemic fluency (F-A-S) despite the strong effect on semantic fluency; this discrepancy has been well documented as an indicator of Alzheimer's disease (Henry et al., 2004), which may represent an underlying contributor to POD. Relatedly, prior research has found that delirium accelerates cognitive decline in Alzheimer's disease, which may suggest related pathophysiology (Fong et al., 2009). The findings specific to these measures, which may reflect particular regional deficiencies, emphasizes the importance of examining a wide range of measures assessing specific cognitive abilities. Establishing an atlas of POD-free subjects could aid in determining if the MRI BOLD CO₂ stress test can be declarative in modeling these neuropsychological deficits based on attenuated response to the hypercapnic stimulus or an intracranial steal signature. This is important in order to use neuropsychological measures as a proxy for at-risk patients, as the administration of a pre-operative MRI BOLD CO₂ stress test is not feasible in general practice. The MMSE, which is the most widely used assessment tool in POD research, did not differentiate POD groups, nor was it associated with POD severity. Finally, as indicated in prior research, psychiatric illness emerged as being associated with POD, namely depressive symptoms. The emerging diathesis profile is similar to deficits reported in white matter hyperintensities and subcortical vascular dementia such as leukoaraiosis, characterized by depressive symptoms, motor and gait disturbances and cognitive deficits. In further support, in healthy older adults, white matter integrity is correlated with greater processing speed (Kerchner et al., 2012), which may suggest that the deficiencies in processing speed measures (DKEFS) is associated with reduced white matter integrity, which was correlated with POD.

The stressor was defined as the intra-operative anesthetic course. High fidelity hemodynamic and end-tidal gas output and continuous bifrontal cerebral saturation by cerebral oximetry were obtained using an intra-operative data acquisition system that downloaded the data stream at 0.5 hz (TrendFace Solo). These data were then binned to a minute-by-minute frequency. The data that were included to define the intra-operative stress and binarized were hypotension (mean blood pressure < 60 mmHg for a minimum of 1 min), elevated CO₂ delta > 10 mmHg, duration of surgery greater than 120 min, and cerebral saturation <60% for greater than 1 min).

Although preliminary, our data demonstrate that inducing stress via CO₂ exposure pre-operatively allows for an index of brain response intra-operatively. Those demonstrating diathesis risk through specific cognitive vulnerabilities described above, psychiatric illness, and abnormal response to CO₂ stress are at highest risk of POD. As previously indicated, although several diathesis risk factors are linearly related to the severity of POD, it is hypothesized that the intra-operative anesthetic course will not demonstrate a clear linear relationship with outcomes, and may be more interactive.

The development of this stress-diathesis hypothesis, exhibited in Figure 5, is based on the current findings in combination with a number of recent findings that both emphasize the multifactorial nature of POD, the null findings with respect to anesthetic toxicity, studies demonstrating pre-existing structural and functional abnormalities in those with POD (Soiza et al., 2008), and the recent findings demonstrating the importance of pre-operative cognitive dysfunction in POD. One recent study examining pre-operative arterial spin labeling MRI using whole brain and globally normalized voxel wide analysis found that greater performance on neuropsychological measures correlated with cerebral blood flow, but these neuroimaging findings were not predictive of the risk of POD (Hshieh et al., 2016). Importantly, however, their study did not incorporate an assessment of intra-operative stress-related factors that would impact brain pathophysiology, which in isolation limits our understanding, as we believe it is the interaction between these brain vulnerabilities and the stress response (the stress-diathesis) that is predictive of both POD and ST-POD. Future research should aim to examine the relationship between cognitive deficiencies, and CO₂ stress test using BOLD MRI in order to understand whether these aforementioned correlations extend to emergent intracranial steal that may be anatomically indicated in part by leucoaraiosis or associated with alterations in cerebrovascular responsiveness in dementias—specifically as noted with Alzheimer's Disease.

Our findings are revealing and provocative. We acknowledge the small sample size of our study as a primary limitation. Despite this, we have identified biomarkers in neuropsychological testing and the psychiatric realm and MRI BOLD CO₂ stress test that when combined together may be predictive of POD. In addition, our sample size (n = 12) has been deemed large enough to adequately power fMRI studies for lower percent changes in activation of BOLD signal (Desmond and Glover, 2002). We have provided preliminary support for a stress-diathesis model based on these interactions and observations. Another limitation relates to defining features of subthreshold and full POD. We chose to use a newly validated POD severity measure to establish severity scores but acknowledge this may not fully encompass other potential markers such as the full spectrum of post-operative confusional states or low-grade encephalopathic states. However, our long-term follow-up provided clinical validity to these severity groups. Specifically, two-thirds of ST-POD or POD patients reported significant changes in cognition since surgery, with one who may have continued to meet criteria for POD over 4 months post-operatively. In comparison, 5/6 non-POD participants reported no change in memory following surgery, with one indicating potential mild age-related changes, which had not been corroborated by a loved one. Nonetheless, there are a number of etiological mechanisms that may manifest differently in acute confusional states in other contexts that warrants further exploration.

As previously indicated, neuroprotection of POD through anesthetic agents has not been supported. However, other means are available to provide neuroprotection to patients undergoing anesthesia for surgery in the operating room. Neuroanesthesia principles are time-honored and designed to optimize the surgical approach to the brain. It would seem to make sense to apply such approaches to the management of patients at risk for POD. While mild hypocapnia is usually maintained in neurosurgical procedures to reduce brain bulk, our findings here suggest that maintenance of normocapnia minimizing CO₂ delta intra-operatively may be the best course to choose for patients at risk of POD. There are older animal studies to support this contention. In two studies done using a paraplegia model, spinal cord was examined as a focus of injury. In these studies clear end-points of damage and comprehensive hemodynamics and end-tidal gases were reported. The first paper shows no difference between intravenous anesthesia (methohexital) or volatile agent (isoflurane) on the incidence of paraplegia suggesting the anesthetic agent is not at issue (Mutch et al., 1993b). The second paper shows a clear advantage of a neuroanesthesia approach (Mutch et al., 1993a).

An anesthetic approach tailored to rigorous control of intra-operative CO₂ may be appropriate as we now have greater understanding of the impact of alterations on CBF with changing levels of CO_2. When the brain manifests with cerebrovascular dysregulation, conditions exist for intracranial steal with elevation of CO₂ and the possibility of augmented flow to these areas with mild hypocapnia—the so-called “Robin Hood” effect. However, this effect is lost with subsequent elevation of the CO₂ tension leading to steal with return to normocapnia. This study has revealed the magnitude of change in CO₂ that can occur with major surgery. The mean delta was 13.2 ± 2.9 mmHg; range 7.5 – 20.0 mmHg. Thus, none of the patients had a CO₂ delta as low as the hypercapnic stimulus in the brain MRI CO₂ stress test. For this reason the voxel counts reported are conservative estimates of the response to CO₂ in any given patient. The effect of CO₂ intraoperatvely could be further confounding as all patients received volatile agents—known cerebrovascular vasodilators. Patients are also at ischemic risk with hypotension and in this context the combination of hypotension with larger CO₂ deltas would be anticipated to be even more deleterious. An operative procedure where such a situation can routinely arise is with open aortic aneurysmectomy. With cross-clamp release hypotension in combination with large swings in CO₂ ensues. It has been noted that this patient group suffers with a very high incidence of POD (Salata et al., 2012). Especially important in the context of the MRI BOLD CO₂ stress test is delineating the magnitude of intracranial steal in brain at risk. Follow up studies where the brain stress test utilizes a greater CO₂ delta are being entertained; most likely using a ramp protocol of incremental CO₂ change that is better tolerated by patients, but able to examine a greater range (Fierstra et al., 2013; Sobczyk et al., 2014; Supplemental Video 1).

Importantly, maintenance of normocapnia as a prevention effort may also extend to intensive care settings, where POD rates are also high. Many of these patients will have the same diatheses, even the same stress if managed in a surgical ICU, following their operative intervention. Irrespective of management in a medical or surgical ICU, mechanical ventilation is extremely common, with swings in end-tidal CO₂ evident, often with hemodynamic instability. In fact, in intensive care units POD has been found to be associated with longer durations of mechanical ventilation (van den Boogaard et al., 2012), supporting this contention.

JF, JD, and DM have patent rights on the RespirAct™ device described in this manuscript. They and the University of Toronto stand to gain financially if the device described is sold commercially. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.