The study was designed as a prospective, non-randomized observational study. We invited all patients who were scheduled for robot-assisted laparoscopic prostatectomy surgery at Akershus University Hospital from June 10th to December 23rd, 2022. Patients who met the exclusion criteria listed in Supplementary Table S1 were not included.

After oral and written information, written consent was obtained before the first presurgical cognitive test. Our recruitment process and protocol were in accordance with the Declaration of Helsinki and approved by the Regional Ethics Committee (Oslo, Norway; ref. No.: 230764) and the Institutional Data Protection Officer (ref. No.: 2021_14).

The study was purely observational. Premedication, anesthesia, mechanical ventilation, and postoperative analgesia were given according to our department's standard regime for robot-assisted laparoscopic prostatectomy. Oral premedication was given approximately 60 min before surgery, consisting of paracetamol (2 g, 1.5 g if age ≥75 years or weight ≤60 kg), dexamethasone (16 mg, 12 mg if age ≥75 years or weight ≤60 kg), oxycodone, prolonged release formulation (10 mg, 5 mg if age ≥75 years or weight ≤60 kg). Dexamethasone was omitted in patients with diabetes mellitus unless they were deemed well-regulated with HbA1C ≤ 48 mmol/mol or had type II diabetes regulated by low-dose metformin (500 mg × 2–3/day) or managed by dietary restrictions alone.

Anesthesia was induced with fentanyl, 100–150 µg, remifentanil and propofol using target control anesthesia. Muscle relaxation was provided using rocuronium (0.6 mg/kg). Anesthesia was maintained with an infusion of propofol and remifentanil, targeting a Bispectral Index^TM (BIS^TM, Medtronic Inc, Dublin, Ireland) of 40–60. Bolus doses of rocuronium were used to maintain deep neuromuscular blockade throughout the procedure (TOF = 0 and PTC ≤4). Sugammadex was used as reversal agent if TOF < 90% at the end of surgery. Bolus doses of ephedrine 5–10 mg i.v. and phenylephrine 0.05–0.1 mg i.v. were given in the event of hypotension. In select cases, an infusion of norepinephrine was started. Parecoxib 40 mg i.v. was administered approximately 30 min before extubation unless contraindicated. Ondansetron 4 mg i.v. was administered to patients with a high risk of developing postoperative nausea and vomiting. Normovolemia was maintained with an infusion of isotonic Ringer's Acetate with a rate of 3 ml/kg.

Baseline measurements and recordings were obtained before induction of anesthesia (T1). Further measurements and recordings were taken immediately after induction of anesthesia and mechanical ventilation (T2), after the introduction of pneumoperitoneum and tilting the patient in the Trendelenburg position (T3), after 60 min in Trendelenburg position (T4), just before the end of surgery (T5), and after the patient was tilted back to horizontal position, but still under general anesthesia and mechanical ventilation (T6) (Table 1).

Standard anesthesia monitoring for the procedure included an arterial catheter for continuous blood pressure measurements, continuous electrocardiogram, finger pulse oximetry and measurements of end-tidal CO₂, oxygen concentrations, and mean airway pressure from the ventilator circuit (Zeus, Drägerwerk AG & Co. KGaA, Lübeck, Germany).

Results from arterial blood gas analyses collected for study purposes were concealed from the treating anesthesia team if values did not exceed predetermined safety thresholds (Supplementary Table S2).

Central venous pressure was measured through a short 1.1 mm diameter cannula placed in the left external jugular vein (20). Standard breath-to-breath spirometry from the ventilator circuit provided tidal volumes, ventilatory minute volumes (L/min), peak airway pressures, and mean airway pressures. The laparoscope (AirSeal® iFS, CONMED Corporation, Largo. USA) continuously measured and displayed intraabdominal pressure. The degree of tilt of the operating table was electronically controlled and displayed on the table panel.

Doppler ultrasound measurements were obtained using a high-frequency linear probe over the right internal carotid artery (ICA) (Venue^TM, General Electric Company, Boston, USA). We obtained time-averaged mean frequency (TAMEAN) and luminal cross-sectional area in both systole and diastole. All ultrasound measurements were repeated 3 times at each measurement time point (T1-T6). As recordings were of maximum 4 s, the ventilator frequency was set to a minimum of 15 breaths per minute during the recording to ensure that the ultrasound measurements included a full ventilatory cycle. The ventilation frequency was returned to their prior setting as soon as ultrasound measurements were completed. Arterial line signals were analyzed using LiDCO Rapid^TM (Masimo Corporation), which provided a nominal cardiac output, pulse pressure variation, nominal stroke volume, and stroke volume variation.

Optic nerve sheath diameter (ONSD) was measured using ultrasound after induction of anesthesia (T2) and near the end of surgery (T5). We measured ONSD in the transverse and sagittal plane in both eyes, the median value was reported.

Data was collected and stored using an electronic database (REDCap®, Vanderbilt University, Nashville, USA).

Cognitive testing was performed using the Cambridge Neuropsychological Test Automated Battery (CANTAB®) (Cambridge Cognition Ltd) software, which tests different domains of cognitive function (attention, memory, visuo-spatial processing, language and executive functions) (21). Testing was performed at least one day before surgery and repeated approximately 10 and 180 days after surgery. Cognitive testing was performed in a controlled and quiet environment at our hospital or at a healthcare facility closer to the patient's residence. The principal investigator or dedicated research staff had oversight over the entire test routine to ensure proper execution and standardization of testing conditions. Both the principal investigator and the research staff had received training from a neuropsychiatrist.

A total of 11 cognitive variables from the testing were analyzed. We analyzed Motor Screening Task Mean Latency, Paired Associates Learning First Attempt Memory Score, Paired Associates Learning Total Errors Adjusted, Reaction Time Task Median Five-Choice Reaction Time, Rapid Visual Information Processing, Rapid Visual Information Processing Median Response Latency, Rapid Visual Information Processing Probability of False Alarm, Spatial Working Memory Between Errors and Spatial Working Memory Strategy.

Variables were chosen based on recommendations from the CANTAB Connect Research: Test Administration User Guide (22), recommendations from the CANTAB® team and review of previous studies using CANTAB® to test for postoperative cognitive dysfunction (21, 23).

Our primary endpoints were changes in ICA flow, cerebral perfusion pressure, ONSD and postoperative changes in the cognitive testing variables. Secondary endpoints were correlations between ICA flow and other hemodynamic variables.

For time-averaged mean frequency and ICA cross-sectional area in systole and diastole, the median value of all three measurements at each time point was used. For ONSD, the median value of the ONSD for all four measurements was used.

ICA flow (ml/min) was calculated by multiplying the median time-averaged mean frequency (cm/sec) by the average cross-sectional area (cm²) of the internal carotid artery (median diastolic area · 0.67 + median systolic area · 0.33), multiplied by 60. Cerebral perfusion pressure was calculated by subtracting the central venous pressure from mean arterial pressure. Cardiac index was calculated as cardiac output/body surface area. Ideal body weight (kg) was calculated as patient height in cm – 100.

The study was by design a non-randomized prospective observational study. No formal sample size calculation was performed. Temporal changes in ICA flow, ONSD, cardiorespiratory variables and outcomes from cognitive testing were assessed using a linear mixed model analysis fit by REML with t-tests using Satterthwaite's method. Participant ID was set as the random effect to account for inter-subject correlation and measurement time was set as the fixed effect.

Independent variables potentially associated with changes in ICA flow in the timeframe T2–T6 were analyzed in a linear mixed model. They were chosen a priori and included age, body mass index, cardiac index, heart rate, stroke volume index, mean arterial pressure, central venous pressure, mean airway pressure, PaCO₂, and tidal volume divided by ideal body weight. Partial pressure of arterial oxygen (PaO₂) was not used in this analysis, as we deemed it heavily influenced by pre-oxygenation before induction and extubation. A forward stepwise regression procedure set to minimize BIC was run to select important predictors. These were subsequently included in a linear mixed model with participant ID as a random effect. The underlying relationships among the contributing variables were also integrated in a Structural Equation Model (SEM) and presented as a path diagram.

We performed a linear mixed model analysis for variables potentially associated with changes in cognitive outcomes, modeling the variation in each endpoint with time as a fixed effect, the first cognitive test as a reference point, and study ID as a random effect. Secondly, we performed a mixed model regression including the following fixed effects: Total tilt time, the integral of cerebral perfusion pressure between T3 and T5, the integral of ICA flow between T3 and T5, the time between T2 and T6 (corresponding to the duration of anesthesia), the difference in ONSD between T2 and T5, and the CANTAB testing time point.

The assumptions of normal distribution and equal variance were assessed using quantile-quantile plots and a frequency distributions test. Normally distributed data are reported as means with 95% confidence interval shown in round brackets (x-y), and otherwise as medians and quartiles shown in square brackets [x,y]. The analyses were considered exploratory in nature, and therefore claims of significance are made cautiously.

Statistical analyses were performed using dedicated statistical software (R Foundation for statistical Computing, Vienna, Austria), using the “lmer” package for linear mixed models and “lavaan” for Structural Equation Modeling. Preparation of the data, stepwise regression procedure, analysis on means, medians, distributions and preparation of figures were done using JMP (JMP Pro17 ®, JMP Statistical Discovery, Buckinghamshire, UK).

Eighty-four males were screened for eligibility, of which 44 were enrolled in the study. Of the 40 patients who were screened but not enrolled, 17 were excluded due to logistical reasons, and nine were excluded as they had a non-Scandinavian language as their native language. Nine did not consent to participate or failed to respond to contact from the principal investigator. Five were excluded due to known cerebrovascular or carotid disease. No patient was excluded because of findings of carotid stenosis after enrollment.

Of the 44 included patients, we performed intraoperative recordings on 37, the remaining seven undergoing cognitive testing only. Of the patients undergoing cognitive testing only, five were excluded from intraoperative measuring due to logistical reasons. The remaining two patients were excluded due to intraoperative complications. One due to hypotension and bradycardia, the other due to subcutaneous emphysema affecting the neck. Of the 37 patients who had intraoperative measurements, 32 also completed cognitive tests 1 and 2, and 25 patients completed all three tests. Reasons for incomplete follow-up were logistical reasons (n = 4), failure to respond to contact (n = 3), or illness/surgical complications hindering cognitive testing (n = 5).

The mean age of the included patients was 67 years (65–69), mean BMI was 27 kg/m² (26–28) and median ASA physical status was 2 [2,2]. The most prevalent co-morbidities were hypertension (43%), coronary artery disease (16%), and diabetes mellitus (7%). Sixty-one percent had other co-morbidities. Supplementary Table S3 provides additional patient characteristics.

The total time with head-down tilt during surgery was 165 min [134,199], the duration of operation was 182 min [161,225], while general anesthesia lasted 226 min [201,266]. The total propofol and remifentanil doses were 4.9 mg/kg/h [4.5, 5.3] and 0.16 µg/kg/min [0.14, 0.18] respectively. Vasopressors were used in 78% of patients at T3, 16% at T5, and 73% at T6. Body temperatures were 36.2 C° [35.7, 36.6] and 36.5 C° [36.1, 36.9] at the start and end of the surgery respectively.

Descriptive statistics and results of a linear mixed model of variation of observed peroperative variables with time are presented in Tables 2, 3. There was a significant approximately 30% decline in ICA flow after induction of anesthesia and onset of mechanical ventilation. In the period from established head down tilt to end of surgery we observed no further changes in ICA flow, but there was a moderate increase in ICA flow when the patients were tilted back to horizontal position.

Cerebral perfusion pressure declined significantly following induction of anesthesia and did not change further immediately upon tilting. However, at all following timepoints cerebral perfusion pressure was lowered compared to immediately after tilting (Tables 2, 3). ONSD increased significantly over the course of the surgery.

We observed a significant decrease in ICA cross-sectional area from the awake state to after induction of anesthesia, with no significant changes in any of the subsequent time points. Similarly, blood velocity (time-averaged mean frequency) decreased 23% following induction of anesthesia, with no further changes apart from a trend towards recovery. Figure 1 provides an overview of changes in hemodynamic variables during surgery.

Cardiac index and mean arterial pressure were the predictors of ICA flow selected in the forward stepwise regression procedure set to minimize BIC. In a final multivariate linear mixed model, cardiac index was the strongest predictor of ICA flow. Its positive association had a log worth of 4.82 (p < 0.0001), more than twice the 2.30 log worth (p < 0.0005) of the negative association between mean arterial pressure and ICA flow.

We found that cardiac index had a moderate positive effect on ICA flow (β 0.367, p < 0.0001) with 1 L/min/m² increased cardiac index corresponding to 92 ml/min increased ICA flow. PaCO₂ had a small positive effect (β 0.145, p = 0.045), a 1 kPa increase in PaCO₂ corresponding to 49 ml/min increased ICA flow, while mean arterial pressure had a small negative effect (β −0.203, p = 0.007), a 10 mmHg increase corresponding to a decline in ICA flow of 29 ml/min. In Figure 2, the predicted ICA flow response to changes in cardiac index and mean arterial pressure is illustrated for each individual and on a group level based on mixed-model multiple regression.

Structural Equation Modeling also found ICA flow to have a significant positive relationship with cardiac index and PaCO₂, while having a significant negative relationship with mean arterial pressure (Figure 3). The general structural equation model had a moderate fit (Chi-Squared/df = 8.995, CFI = 0.882, RMSEA = 0.213), the analysis potentially being limited by heterogeneity due to unaccounted-for variation in time present in the data. The proportion of variance of ICA flow not explained by the effect of other variables was 0.83.

Focus and outcomes of the employed cognitive tests are shown in Table 4. We observed improvements 10 days after surgery in Rapid Visual Information Processing (1.1%, p = 0.015) and Motor Screening Task Mean Latency 10.7%, p < 0.001), and at 180-days for Rapid Visual Information Processing (2.2%, p < 0.001), Motor Screening Task Mean Latency (10.8%, p = 0.001), Reaction Time Task Median Five-Choice Reaction Time (4.3%, p = 0.003) and Spatial Working Memory Between Errors (28%, p = 0.043).

Total time spent under anesthesia and total Trendelenburg time were respectively negative and positive predictors of Spatial Working Memory Strategy, but these results should be interpreted with a high degree of caution considering the large number of tests and exploratory nature of the analysis. Data are presented in Supplementary Table S4.

Although ICA has no extracranial branches, they only supply 80% of the cerebral blood flow, the remainder coming from the vertebral arteries. Our study utilized unilateral measurements of the ICA since the contralateral side was used for central venous pressure measurements. All analyses of ultrasound images, which require some manual evaluation, were done in a non-blinded fashion. Vasopressors were not administered by protocol but at the discretion of the attending doctor or nurse anesthetist.

The choice to exclude non-Scandinavian native language speakers due to the cognitive testing may limit external validity. Further, the first cognitive test was performed only days before surgery, when the cancer diagnosis and anticipation of surgery may have contributed negatively to the results. Selection bias may have played a role at the 180-day testing as we had a substantial loss to follow up. This observational study may have been underpowered for statistical analyses of cognitive test results.