Effects of Postoperative Blood Pressure Management on Delirium Among Patients in the Intensive Care Unit Following Cardiac Surgery: An Observational Cohort Study

Abstract

Objectives:

This study aimed to determine if blood pressure control in the early postoperative period is associated with postoperative delirium in the cardiovascular intensive care unit (ICU).

Design:

Observational cohort study.

Setting:

Single large academic institution with a high volume of cardiac surgery.

Participants:

Cardiac surgery patients admitted to the cardiovascular ICU after surgery.

Interventions:

Observational Study

Measurements and Main Results:

517 cardiac surgery patients had mean arterial pressure (MAP) data recorded minute by minute for 12 postoperative hours. We calculated the time spent in each of seven prespecified blood pressure bands and recorded the development of delirium in the ICU. A multivariable Cox regression model was developed using the least absolute shrinkage and selection operator approach to identify associations between time spent in each MAP range-band and delirium. Compared with the reference band of 60 to 69 mmHg, longer durations spent in three bands were independently associated with a lower risk of delirium: 50 to 59 mmHg band (adjusted hazard ratio [HR], 0.907 [per 10 minutes; 95% CI, 0.861 to 0.955); 70 to 79 mmHg band (adjusted HR, 0.923 [per 10 minutes; 95% CI, 0.902 to 0.944); 90 to 99 mmHg band (adjusted HR, 0.898 [per 10 minutes; 95% CI, 0.853 to 0.945).

Conclusions:

MAP-range bands both above and below our reference band of 60–69mmHg are associated with decreased risk of ICU delirium development, however this is difficult to reconcile with a plausible biologic mechanism. Therefore, our findings did not find a correlation between early postoperative MAP control and increased risk of the development of ICU delirium after cardiac surgery.

INTRODUCTION

Approximately 500,000 cardiac surgeries are performed every year in the United States.¹ Delirium occurs in 16% to 34% of patients following cardiac surgery and is strongly associated with serious adverse outcomes.^2–5 Specifically, delirium is an independent risk factor for increased length of stay in the hospital, mortality, and worse long-term cognitive outcomes.^6–8

While the etiology of delirium is uncertain and likely multifactorial, one proposed mechanism is hypotension resulting in inadequate cerebral perfusion.⁹ Patients undergoing cardiac surgery may be particularly susceptible to cerebral hypoperfusion due to a high prevalence of pre-existing hypertension (resulting in a right shift in the cerebral blood flow autoregulatory curve) and vascular disease. Additionally, underlying cardiac dysfunction, systemic vasodilation, and bleeding all promote perioperative hypotension, often necessitating the use of potent vasoactive medications to maintain mean arterial pressure (MAP) throughout the perioperative period. Several studies have sought to establish a link between blood pressure and delirium but to date findings have been conflicting and evidence limited.^10–12 Yet, if there truly is an association between blood pressure and delirium, careful management of postoperative blood pressure to optimize cerebral perfusion pressure could reduce the risk of delirium.

Based on data from septic shock literature, clinicians aim to maintain patients’ arterial pressure above 65mmHg in the early period postoperative period after cardiac surgery,^13–14 yet it is unknown whether exposure to a MAP significantly different from this pressure increases the risk for postoperative delirium. We hypothesized that increasing time spent outside of a reference MAP band (60–69 mmHg) in the first 12 hours following cardiac surgery is associated with a higher rate of postoperative delirium in the intensive care unit (ICU).

METHODS

This single-institution observational cohort study was approved by the medical center’s Human Research Protections Program (Institutional Review Boards # 201406), with a waiver of the requirement for informed consent. This manuscript was developed in adherence to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (Appendix 1).¹⁵

Patient Population

We screened all patients undergoing cardiac surgery at our institution from October 1, 2020, through May 31, 2021. We started collection in October, as this was when our data collection capacity was upgraded to allow for robust data collection. Inclusion criteria for our cohort were adult patients (≥ 18 years of age) who had invasive blood pressure monitoring in place following surgery, which was determined by the presence of at least one measured invasive blood pressure recording. Patients were also required to have at least one documented Confusion Assessment Method for the ICU (CAM-ICU) while in the ICU following the index surgery.¹⁶ A CAM-ICU was not performed if the Richmond Agitation-Sedation Scale (RASS) was −4 or −5, as this level of sedation precluded CAM-ICU assessment. We analyzed only the index surgery for each patient during the 8-month time period. Index surgery was defined as the first eligible surgery performed in the main operating rooms by a cardiac surgeon and lasting longer than 4 hours from documented anesthesia start to anesthesia stop. The 4-hour length was chosen with the intent of excluding cases that may have been an anticipated precursor to major cardiac surgery (e.g., insertion of intra-aortic balloon pump while awaiting heart transplant). Only cases performed in the main operating room complex were eligible, to prevent the inclusion of patients who underwent procedures in the cardiac catheter laboratory or bedside procedures. To further refine our sample, we included only patients who were subsequently admitted to the cardiovascular ICU after surgery. We excluded patients who had emergency surgery, defined with American Society of Anesthesiologists (ASA) emergency classification; whose primary surgery was Extra-Corporeal Membrane Oxygenation (ECMO) insertion or revision; or underwent heart-liver or heart-kidney transplant.

Power Analysis

Based on an estimated accrual of 150 patients/month and a conservative delirium incidence of 15%, we anticipated that we would require 6 months to accumulate 900 patients to capture between 100–150 patients with delirium, giving us an estimated 10–15 degrees of freedom for multivariable modeling. Our actual numbers are lower due unanticipated COVID surges decreasing patient volume. Due to this we extended two more months in effort to increase our numbers, at which point we had reached our goal enrollment of delirium patients of around 100.

Data Collection

Using Structured Query Language, patient data were acquired from our in-house database, which contains data from the electronic health record (Epic, Verona, Wisconsin). For each eligible anesthetic case, we collected minute-by-minute blood pressure readings from invasive arterial pressure monitoring in the immediate 12 hours following surgery. This window of time was chosen to ensure most individuals, with a delirious episode would have it occur after the period of measured blood pressure. Other covariates and confounding variables for potential inclusion in the multivariable model were selected based on the expert opinion of the investigators. In addition, we collected demographic and disease severity (Elixhauser) variables. The exposure variable, MAP, was calculated from minute-by-minute invasive blood pressure data stored in the in-house database. Blood pressure data was stored in two possible variables, Arterial Mean (ARTm) and Arterial Blood Pressure Mean (ABPm). The differing variable names is (likely) due to different ports into which the arterial line cable may be inserted on the ICU bedside monitor and, for the purposes of analysis, were considered interchangeable. A single MAP value was assigned for every minute through the first 12 hours in the ICU after surgery, according to the following pre-specified rules. MAP values < 30 mmHg and > 150 mmHg were considered artifactual and excluded from analysis. If a single minute had more than one reading within a single variable (ARTm or ABPm), a median value was used for that minute and, if a single minute had values for both ARTm and ABPm, the greater value was selected as the representative MAP for that minute. Additionally, we performed multiple imputation for missing invasive blood pressure readings.

Primary Outcome & Primary Exposure

The primary outcome was postoperative delirium. We defined the development of delirium and its time of onset according to the timestamp associated with the first CAM-ICU positive screening during a patient’s ICU stay following index surgery. Institutional practice requires nurses to complete a full assessment of consciousness using the RASS and then to perform a CAM-ICU assessment, at least once every 12 hours.^17,18 The clinical team could decide to complete the assessment more frequently if indicated. The CAM-ICU result was then recorded in the electronic medical record and stored in the perioperative data warehouse. The CAM-ICU is a validated algorithm that uses cardinal features of delirium, fluctuations in mental status, inattention, altered level of consciousness and disorganized thinking, to screen for the presence of delirium.²⁰

For each patient, the primary exposure variable (MAP) was characterized by assigning the MAP value for each minute into one of a sequence of mutually exclusive blood pressure bands representing the full spectrum of recorded MAP values. Seven discrete MAP range-bands for analysis were defined as follows: less than 50 mmHg, 50 to 59 mmHg, 60 to 69 mmHg, 70 to 79 mmHg, 80 to 89 mmHg, 90 to 99 mmHg, and greater than or equal to 100 mmHg. The primary exposure of interest was the cumulative duration, in minutes, that each patient spent within each MAP range-band during the first 12 hours in the ICU. For patients who developed delirium within the first 12 hours in the ICU following surgery, the cumulative duration of time in each blood pressure band was censored at the time of the first positive CAM-ICU.

Statistical Analysis

Descriptive statistics of demographic, clinical, and procedural variables were reported to characterize the study population with means and standard deviations (SDs) for parametric variables, medians, and interquartile ranges (IQRs) for nonparametric variables, and with percentages for categorical variables, as appropriate.

To determine the association between increasing time spent within various blood pressure bands and postoperative delirium, a time-to-event analysis was performed using a Cox proportional hazards regression model to assess relative risk for the development of delirium based upon occurrence rate comparison. To account for each blood pressure band not being independent of the other blood pressure bands, the blood pressure band 60 to 69 mmHg was chosen as the reference band, as this band encompasses the typical lower limit of MAP targeted by physicians after surgery. Additional clinically plausible covariates, including patient sex, age, body mass index (BMI), procedure type, duration of surgery, postoperative stroke, norepinephrine dosing, and Elixhauser comorbidities were screened through the least absolute shrinkage and selection operator (LASSO), a machine learning approach, to determine the variables impacts on postoperative delirium.²⁰ Time spent in each of the other 6 blood pressure bands were treated as separate continuous variables. These blood pressure bands along with patient age and the Elixhauser comorbidities hypertension and diabetes were forced into the final model. The rest of the final Cox model was constructed using variables that were found to have a significant effect on postoperative delirium by LASSO. ICU discharge was considered as a competing risk and thus it was right censored in this model. The proportional hazard assumption was tested using Schoenfeld residuals and Kaplan-Meier plots, and a time-varying interaction was created and entered the model along with the original variable if the assumption was violated. Furthermore, Martingale residuals were obtained to assess potential nonlinearity in the relationship between the log hazard and the covariates.²¹ The associations were then summarized and reported using the hazard ratios (HRs) with 95% confidence intervals (Cis).

Sensitivity Analyses

Several sensitivity analyses were prespecified. First, we repeated the primary analysis after excluding patients who had a positive CAM-ICU in the first 12 hours. Second, we excluded patients who returned to the operating room prior to the development of delirium, to remove the possible confounder of the second surgery increasing the risk of delirium. Third, we excluded patients who had a stroke after their operation as stroke itself may also be a risk factor for delirium. Fourth, we excluded patients who remained intubated for longer than 12 hours postoperatively, as the additional sedative medication likely required to facilitate ongoing intubation may also confound the association between blood pressure and delirium. Finally, in a post-hoc analysis, we excluded patients who first developed delirium more than 48 hours after ICU admission. This was done for two reasons. First, new-onset delirium beyond this time-period may be more likely to reflect factors other than early postoperative blood pressure. Second, a longer ICU admission increases the number of opportunities to detect delirium and likely reflects increased severity of illness which may further bias detection of delirium that is unrelated to blood pressure.

A two-sided hypothesis testing with a p-value of less than 0.05 was used to indicate statistical significance. All statistical programming was implemented in R (R Foundation for Statistical Computing version 3.6.1, Vienna, Austria), and SAS 9.4 (SAS Institute Inc., Cary, North Carolina, USA).

RESULTS

From our initial cohort of 659 patients, 49 were excluded because they had fewer than 6 hours of recorded blood pressure readings, this cutoff of less than half of available recordings was decided after obtaining the initial raw data but before analysis. Also excluded were 75 ASA emergency cases, 9 non-index surgeries, and 9 cases for which the primary surgery was ECMO insertion or revision or a kidney-heart/liver-heart transplant (Figure 1). This left 517 eligible cases. The mean age at time of surgery was 59.7 years (SD = 13.5), 354 (68.5%) were male, mean BMI was 29.6 (SD = 6.4), median Elixhauser score was 7 (IQR = 1 – 13) and median ICU duration was 1.8 (IQR = 1.0–3.8) days (Table 1). For our primary exposure, the median number of minutes in which a blood pressure was recorded was 663 (IQR = 604–692) out of a total of 720 minutes possible. Median and IQR for patients with and without delirium were as follows: no delirium, 664 minutes (IQR = 604–692); with delirium, 662 minutes (IQR = 621–695). This is approximately 8% missing per patient, and of that total 19.8% were terminal stops. A terminal stop was defined a missing blood pressure reading with no resumption before the 720-minute mark. A histogram plot was created to demonstrate the distribution of blood pressure measurements for our entire cohort before multiple imputation (Figure 2), this was further broken down into patients who developed delirium (Supplemental Figure 2) and patients who did not develop delirium (Supplement Figure 3). During the first 48 hours, only one patient died in the ICU after completing the blood pressure measuring period. The multiple imputed data sets were then generated using the regression method, given the monotone missing data pattern of our cohort. Appendix 2 shows the number of missing blood pressure readings that were imputed for each eligible case. The time varying interaction was not needed and so was not included in our final regression model.

Figure 1:

The CONSORT Flow Diagram.

Table 1.

Patient Demographics and Clinical Characteristics of the Study Sample.

Variables	Total (n = 517)	No Delirium (n = 414)	Delirium (n = 103)	P-Value
Age in years, mean (SD)	59.7 (13.5)	59.6 (13.3)	60.2 (14.3)	0.657
Body Mass Index in kg/m2, mean (SD)	29.6 (6.4)	29.4 (6.3)	30.2 (6.9)	0.215
Sex (%)				0.718
Male	354 (68.5%)	285 (68.8%)	69 (67.0%)
ASA Classifications (%)				0.140
I	0 (0%)	0 (0%)	0 (0%)
II	3 (0.6%)	3 (0.7%)	0 (0%)
III	67 (13.0%)	60 (14.5%)	7 (6.8%)
IV	446 (86.3%)	351 (84.8%)	96 (93.2%)
V	1 (0.2%)	1 (0.2%)	0 (0%)
Hypertension (%)	301 (58.2%)	244 (58.9%)	57 (55.3%)	0.507
Diabetes (%)	201 (38.9%)	154 (37.2%)	47 (45.6%)	0.116
Postoperative Stroke (%)	9 (1.7%)	2 (0.5%)	7 (6.8%)	<.001
Surgery Duration in hrs., median (Interquartile Range)	5.8 (5.0–6.9)	5.6 (5.0–6.6)	6.9 (5.6–8.2)	<.001
ICU Duration in days, median (Interquartile Range)	1.8 (1.0–3.8)	1.5 (1–2.8)	5.8 (3.5–11.8)	<.001
Elixhauser Score, median  (Interquartile Range)	7.0 (1–13)	6(0–11)	14 (8.3–19)	<.001
12 hr Postoperative Total Norepinephrine Dose in mcg, median (Interquartile Range)	856 (60–2500)	756 (11.3–2160)	1650 (331–4005)	<.001
Procedure Type (%)				<.001
Coronary artery Bypass Graft (CABG)	242 (46.8%)	207 (50.0%)	35 (34.0%)
Heart Valve Procedure	138 (26.7%)	112 (27.1%)	26 (25.2%)
Aorta	29 (5.6%)	24 (5.8%)	5 (4.9%)
Heart Transplant	45 (8.7%)	24 (5.8%)	21 (20.4%)
Myectomy	9 (1.7%)	7 (1.7%)	2 (1.9%)
Combined valve and CABG	17 (3.3%)	13 (3.1%)	4 (3.9%)
Ventricular Assist Device insertion	4 (0.8%)	2 (0.5%)	2 (1.9%)
Others	33 (6.4%)	25 (6.0%)	8 (7.8%)
Selected Elixhauser Comorbidities (%)
Congestive Heart Failure	347 (67.1%)	260 (62.8%)	87 (84.5%)	<.001
Paralysis	11 (2.1%)	1 (0.2%)	10 (9.7%)	<.001
Liver Disease	38 (7.4%)	28 (6.8%)	10 (9.7%)	0.305
Cardiac Arrhythmia	387 (74.9%)	293 (70.8%)	94 (91.3%)	<.001
Valvular disease	204 (39.5%)	165 (39.9%)	39 (37.9%)	0.711
Renal Failure	164 (31.7%)	108 (26.1%)	56 (54.4%)	<.001
Peripheral Vascular disease	94 (18.2%)	73 (17.6%)	21 (20.4%)	0.516
Chronic Pulmonary Disease	103 (19.9%)	75 (18.1%)	28 (27.2%)	0.039

Figure 2:

The distribution of Mean Arterial Pressure (MAP) values in the first 12 hours of ICU Stay.

A total of 103 (19.9%) patients developed delirium in the ICU. Median time to delirium onset was 27.7 hours (IQR 14.9–63.9) (Supplemental Figure 1,). In the primary analysis, an increased duration within three of the six tested MAP bands were found to be associated with a lower likelihood of developing delirium in the ICU, including the 50 to 59 mmHg band (adjusted hazard ratio [HR], 0.907 [per 10 minutes]; 95% CI, 0.861 to 0.955), the 70 to 79 mmHg band (adjusted HR, 0.923 [per 10 minutes]; 95% CI, 0.902 to 0.944), and the 90 to 99 mmHg band (adjusted HR, 0.898 [per 10 minutes]; 95% CI, 0.853 to 0.945) (Table 2, Figure 3).

Table 2.

Test Variables and Their Associated Risk of Postoperative Delirium Development using Multivariable Cox Proportional Hazard Model.

Variables	Adjusted hazard ratio (95% CI)	p-value
Patient age (per 10 years older)	1.144 (0.994 to 1.317)	0.062
MAP 50–59 (per 10 mins increase in duration)	0.907 (0.861 to 0.955)	<.001
MAP 70–79 (per 10 mins increase in duration)	0.923 (0.902 to 0.944)	<.001
MAP 90–99 (per 10 mins increase in duration)	0.898 (0.853 to 0.945)	<.001
Congestive heart failure	2.007 (1.115 to 3.613)	0.020
Hypertension	1.507 (0.958 to 2.370)	0.076
Diabetes (complicated)	1.681 (1.114 to 2.538)	0.013
Weight loss	3.058 (1.965 to 4.758)	<.001
Paralysis	3.921 (1.878 to 8.188)	0.003

Figure 3:

Visualization of the hazard ratio and 95% confidence interval of each variable included in our final proportional hazard regression model.

We also performed a Spearman rank correlation between intra-operative and postoperative blood pressure the results of which can be seen in Supplemental Table 1. Postoperative and intra-operative blood pressures were found to be essentially independent variables (correlation coefficient was < 0.2 for all but one combination).

Results of the prespecified sensitivity are as follows. We excluded cases where delirium was identified in the first 12 hours of ICU stay (n = 23, 22.3% of delirium cases), and found that the duration within each MAP band was no longer associated with a decrease in the risk of delirium development. (Supplemental Table 2). Second, we excluded cases where a patient returned to the OR prior to delirium detection; 17 cases were excluded, and the results did not have substantial effects on the analysis (Supplemental Table 3). Third, we conducted the sensitivity analysis after excluding patients with postoperative stroke (n = 9, 0.01% of all patients); this did not change the association between MAP and delirium found in the primary analysis (Supplemental Table 4). Fourth, we performed an analysis on the cohort excluding patients remaining intubated for over 12 hours following their surgeries (n = 111, 21.5% of all patients); in this model MAP band 80 to 89 mmHg was found to have a significant protective effect against the development of delirium, but no substantial change was observed on effect estimate of other bands (Supplemental Table 5). Finally, a post-hoc exploratory analysis looking only at patients who developed delirium in the first 48 hours after surgery failed to have substantial effects on the primary analysis (Supplemental Table 6).

DISCUSSION

In this observational study of 517 cardiac surgery patients, we did not detect an increase in risk of delirium with longer time spent with a MAP outside of the range 60–69 mmHg, a typical postoperative target MAP after adult cardiac surgery. Within the observed range of MAPs our findings do not support an increase in the incidence of delirium with increased time spent either above or below a typical target MAP range.

Previous studies of blood pressure management during cardiac surgery found that a higher intraoperative MAP during cardiopulmonary bypass surgery (higher than an upper limit of normal determined by measurements of cerebral autoregulation) was associated with delirium, but a MAP below the lower limit of normal was not associated with delirium.²² Blood pressure management in the early postoperative period after cardiac surgery appears to be less well-studied. In one study, MAP outside of cerebral autoregulation limits in the postoperative period was not associated with a significantly increased risk of delirium. However, the study also reported that patients outside the cerebral autoregulation limits in the intraoperative period experienced greater rates of delirium.²³ Our study did find that time spent in specific MAP range-bands was associated with a reduced risk for the development of delirium. We did not find that increased time spent in the highest (> 100 mmHg) or lowest (< 50 mmHg) MAP ranges were associated with delirium, but this may be due to a smaller sample size at these extremes. A recent review of the topic identified intraoperative hypotension as a risk factor in 11/17 of the reviewed papers.²⁴ These studies are difficult to compare because of the different types of surgery included and because of the inconsistent definitions used for hypotension and delirium.

Our analysis provides a more comprehensive assessment than previous studies. We studied a large cohort with highly granular minute-by-minute blood pressure data, combined with a validated and pragmatic method of determining delirium, the CAM-ICU. This detail allowed a continuous assessment of blood pressure, versus previous studies which used hourly values or minimum/maximum daily values. Our method allowed us to catch even brief blood pressure events, which is especially important as blood pressures are typically tightly controlled postoperatively.

We chose to use the Cox proportional hazards regression model for our statistical analysis, as this is a widely used approach in the study of analyzing time-to-event when competing risks are present. It also allowed us to address the problem of participant heterogeneity.²⁵ We decided against doing a time-weighted analysis, as this would take away our ability to detect a threshold at either the high or low end. It would also have forced us to assume where a threshold may exist, and consequently potentially missing actual cutoffs. We chose our statistical method as it required the fewest assumptions, allowing for a non-linear relationship between MAP and delirium.²² Due to the large amount of data, we employed a combination of data-driven, machine-assisted, and clinical assessment covariate selection. All covariates were chosen because they had a potential physiologic basis or were shown in previous studies to be associated with either delirium or hypotension. The anesthetic drugs were left out due to large variability in usage during surgery. Studies have had little success showing drugs limit the development of delirium in postcardiac surgery, including dexmedetomidine.²⁶ Elixhauser comorbidities were chosen due to ease of access and reliability to screen for a variety of other common conditions. In addition, we chose to force diabetes and hypertension into the model. These two conditions are common in cardiac surgery patients, and both may have significant impact on blood vessels and autoregulation of blood pressure, which could change the threshold blood pressure at which cerebral hypoperfusion occurs. Age was also forced into the model, due to its association with delirium.

Our findings that increased time spent in non-contiguous MAP ranges of 50 to 59 mmHg, 70 to 79 mmHg, and 90 to 99 mmHg were each associated with a lower risk of postoperative delirium was unexpected and the clinical significance of this finding is not entirely clear. The apparent increase in risk for delirium with increasing time spent in the reference band is unlikely to represent a causal relationship. It may reflect clinician behavior to meticulously maintain MAP in this “normal” range for patients perceived to be at increased risk for delirium or may reflect a type 1 error. Most previous studies poorly defined hypotension, examined only the lowest MAP during a given day, or other less granular measurements, compared with our method.²⁴ If there is a causal relationship between blood pressure and delirium, further studies need to ascertain the ideal MAP range to prevent delirium in post cardiac surgery patients. This would have to be balanced by the effects these different MAP ranges would have on other important postsurgical outcomes (acute kidney injury, mortality, and hospital/ICU length of stay). And maybe needs to be individualized rather than using a one size fits all approach.

There are several limitations to our study. Attempting to collect minute-by-minute blood pressure data, we inevitably had missing data in almost every patient. On average, we were missing 8% of blood pressure values per patient. However, we would highlight that this is still a substantially larger quantity of high-fidelity data than has been analyzed in previous studies, and the missing data is unlikely to represent significant periods of hypotension. Multiple imputation was performed to fill in the missing data, which also allowed us to use a reference band in the statistical analysis. We did exclude patients whose total number of readings were less than half of the total possible readings, as they accounted for fewer than 4% of our total patient population. Possible reasons for the missing or spurious data (< 30 mmHg or > 150 mmHg) are terminal removal of invasive blood pressure monitors, interruptions due to transportation to scans or tests, positional dependence of the arterial line, and zeroing or maintenance of arterial line. As these events were random and had no effect on the blood pressure itself, we felt that using multiple imputation was most appropriate. We did not account for patients’ threshold upper/lower limits of autoregulation and so we would miss an association between these limits and delirium if one exists. However, determining autoregulatory thresholds for each patient remains a research tool that is not yet applicable in the clinical environment, meaning current blood pressure management is not readily individualized. We did not directly measure end organ perfusion, which is an important area of future research. We also did not adjust for multiple medications that are administered perioperatively to patients, some of which may contribute to postoperative delirium due to a relatively small sample size limiting our choice of covariates to include in our model to avoid overfitting. One of the medications we did chose to account for was norepinephrine. Norepinephrine was chosen among the several other widely used vasoactive medications (epinephrine, dobutamine etc.) because we predicted it would be the most widely used (almost 80% of the study population received it) and it is titrated to MAP. Lastly, we focused on MAP values in the postoperative period. Previous studies have suggested that there may be an association between intraoperative blood pressure or time on cardiopulmonary bypass and postoperative delirium.²⁷ However, our analysis suggests there is little effect of adjusting for intra-operative blood pressure on our models. Of note, our dataset does not have reliable documentation of duration of cardiopulmonary bypass, a limitation of our existing electronic record. This is an important area of future study.

We did, however use a validated method for determination of delirium, the CAM-ICU score and limited the cohort to patients who had at least one CAM-ICU score recorded. A CAM-ICU is performed at least once every nursing shift (12 hours) in routine clinical care at our institution, unless the patient does not meet CAM-ICU criteria for assessment (RASS score of −4 or −5), with a high degree of reliability.¹⁷ We assumed delirium onset was at the time of CAM-ICU assessment, but because this was not continuously sampled, this likely introduced error. While the intermittent sampling nature of the CAM-ICU test, along with other limitation such as detection biases, the CAM-ICU test remains the best method for larger cohort studies. On average patients spent 1.8 days in the ICU, meaning there were on average 3 expected CAM-ICU assessments. Direct assessment and recording of ICU delirium by trained study personnel may increase the rate of delirium noted in study cohorts, but this is costly and labor intensive. Therefore, there is a chance that we underestimated our delirium positive rate, as some patients could have been positive for delirium and not assessed prior to discharge or an assessment may not have been recorded properly. Based on our institutional practice, clinically significant delirium is considered a barrier to ICU discharge, making this unlikely. We did not collect pre-operative cognitive status, dementia, or cognitive decline, as these variables are not accurately or routinely collected and stored in our database. Extensive baseline cognitive testing is a commonly accepted limitation in many even larger delirium studies. Furthermore, our rate of delirium is the range of previously reported rates of delirium in cardiac surgery patients.^2,5 In addition, future work needs to be performed to explore the effects of administration of sympatholytic drugs, e.g., dexmedetomidine,²⁸ on delirium. Finally, the single center observational cohort study design limits the generalizability of the current study and renders it subject to residual confounding.

In conclusion, our single center observational cohort study of 517 patients undergoing cardiac surgery using minute-by-minute MAP data failed to detect an increase in delirium with greater time spent outside of typical target MAP values. We instead found a decrease in the risk of delirium for patients as they spent more time in the 50 to 59 mmHg, 70 to 79 mmHg, and/or 90 to 99 mmHg. The clinical implications of this result are not clear and no causal inference should be drawn. Therefore, further studies to investigate the relationship between blood pressure and delirium would be needed before practice changes should be implemented.

Supplementary Material

6

Supplemental Figure 3: Histogram of Time to Delirium Onset: Non-Delirium Patients.

13

Supplemental Figure 1: Histogram of Time to Delirium Onset.

14

Supplemental Figure 2: Histogram of Time to Delirium Onset: Delirium Patients.

Table 3.

The Spearman rank correlation test of intraoperative blood pressure bands and postoperative blood pressure bands.

correlation coefficients Spearmen & Pearson-values		postoperative MAP < 50	postoperative MAP 50–60	postoperative MAP 60–70	postoperative MAP 70–80	postoperative MAP 80–90	postoperative MAP 90–100	postoperative MAP > 100
intraoperative MAP < 50	Spear man correlation	0.066	0.141	0.096	−0.080	−0.121	−0.084	−0.121
	Pears on correlation	0.131	0.001	0.030	0.068	0.006	0.056	0.006
intraoperative MAP 50–60		0.029	0.144	0.130	−0.073	−0.137	−0.122	−0.143
		0.506	0.001	0.003	0.097	0.002	0.005	0.001
intraoperative MAP 60–70		0.048	0.161	0.186	−0.056	−0.219	−0.168	−0.176
		0.272	0.000	<.001	0.204	<.001	0.000	<.001
intraope rative MAP 70–80		0.066	0.051	0.076	0.025	−0.046	−0.019	−0.042
		0.133	0.251	0.085	0.575	0.301	0.672	0.337
intraoperative MAP 80–90		0.037	−0.089	−0.130	0.146	0.122	0.114	0.065
		0.403	0.043	0.003	0.001	0.006	0.010	0.139
intraoperative MAP 90–100		−0.038	−0.117	−0.185	0.120	0.157	0.185	0.151
		0.385	0.008	<.001	0.006	0.000	<.001	0.001
intraoperative MAP >100		−0.001	−0.065	−0.143	0.078	0.108	0.176	0.167
		0.984	0.142	0.001	0.076	0.014	<.001	0.000

ABBREVATIONS

MAP

Mean Arterial Pressure

ICU

Intensive Care Unit

STROBE

Strengthening the Reporting of Observational Studies in Epidemiology

CAM-ICU

Confusion Assessment Method for the ICU

RASS

Richmond Agitation-Sedation scale

ASA

American Society of Anesthesiologists

ECMO

Extracorporeal Membrane Oxygenation

ARTm

Arterial Mean

ABPm

Arterial Blood Pressure Mean

SD

Standard Deviation

IQR

Interquartile Range

BMI

Body Mass Index

LASSO

Least Absolute Shrinkage and Selection Operator

HR

Hazard Ratio

CI

Confidence Interval

CABG

Coronary Artery Bypass Graft