Assessing cardiorespiratory fitness relative to sex improves surgical risk stratification

To what extent sex‐related differences in cardiorespiratory fitness (CRF) impact postoperative patient mortality and corresponding implications for surgical risk stratification remains to be established.


| INTRODUCTION
The routine determination of cardiorespiratory fitness (CRF) employing cardiopulmonary exercise testing (CPET) is increasingly accepted in patient populations undergoing major surgery, given its capacity to improve risk prediction. [1][2][3] Indeed, CRF is considered a vital clinical sign presenting health professionals with unique opportunities to improve patient management and care. 4 Preoperative CPET is used to identify patients with low CRF which is an established predictor of survival following major intra-abdominal surgery. [5][6][7][8][9] When comparing the sexes, female patients demonstrate lower preoperative CRF than males. 10 However, to what extent sex differences in CRF influence preoperative risk stratification and corresponding 'normalization' approaches to optimize the latter has not previously been investigated.
Evidence demonstrates that CRF is lower in females compared to males across the lifespan given smaller body size, skeletal muscle mass, peak cardiac output, haemoglobin concentration, conducting airways, greater oxygen cost of breathing, and accelerated vascular ageing. [11][12][13][14] Current practice frequently adopts thresholds of CRF to stratify patients as 'high risk' for colorectal surgery (combination of anaerobic threshold [AT] <11 mL oxygen [O 2 ] kg −1 min −1 and ventilatory equivalent for carbon dioxide [V̇E/VĊO 2 -AT] >34) 9,15-17 and it is recommended that thresholds evolve, particularly for heterogenous surgical populations. 18,19 However, when considering patient sex, and if a simple dose-response relationship exists between CRF and postoperative survival, females would intuitively be expected to be at increased risk given the physiological constraints. Evidence suggests otherwise as differences in post-operative outcomes following major colorectal surgery remain equivocal, with limited evidence highlighting lower 20 or not different 21 mortality with males. Therefore, sex differences in CPET metrics should also be considered during surgical risk stratification, and the existence of different thresholds in relation to sex requires investigation.
To address this knowledge gap, we retrospectively analysed a cross-sectional population-based cohort of male and female patients who underwent CPET prior to major colorectal surgery. We tested the following hypotheses. First, that CRF would account for disproportionately more deaths compared to other, more traditional CVD risk factors reinforcing its significance as an independent risk factor.
Second, compared to males, females would be characterized by lower CRF and corresponding higher mortality, justifying the need to reassess the current (indiscriminate one size fits all) CPET metrics to surgical risk stratification to further optimize patient management and care.

| Ethics
The University of South Wales Ethics Committee (LSE1636GREO), and Cardiff and Vale University Health Board (15/AIC/6352) approved the analysis. Written informed consent was waived as this study constituted a service evaluation. Procedures were conducted in accordance with guidelines set forth by the Declaration of Helsinki of the World Medical Association 22 with the exception of registration in a database.

| Design
This study constituted a retrospective cross-sectional population-based observational analysis of anonymized longitudinal hospital databases prospectively populated at a single NHS tertiary referral centre (University Hospital of Wales, Cardiff, UK), in accordance with the STROBE statement. 23 A central CPET database was collated with merged data from colorectal surgeons' databases, critical care databases, and longitudinal mortality records from the Office for National Statistics (ONS). Males and females were considered for inclusion based upon chronological order of visit date logged in the CPET database.

| Inclusion/exclusion criteria
Consecutive patients (n = 1110, 618 males/492 females) who underwent CPET at a preoperative assessment clinic between January 2010 and December 2016 were considered for inclusion ( Figure 1). Patients were excluded from final analysis if there was no record of surgery (n = 174) or if surgery was performed more than 365 days following consideration of 'sex-specific' CRF thresholds to better inform postoperative mortality and improve surgical risk stratification.

K E Y W O R D S
cardiopulmonary exercise test, cardiorespiratory fitness, colorectal surgery, sex, survival CPET (n = 191). In the case of multiple CPET visits, the visit closest to subsequent surgery was included. Patients undergoing minor or exploratory procedures were excluded (n = 105). Inability to undergo CPET was defined as having completed static pulmonary function tests but inability to perform cycle ergometry with expiratory gas analysis which was primarily due to musculoskeletal complications (n = 21; 6 males vs. 15 females).

| Demographics
Patient demographics including risk factors were gathered from medical notes at the preoperative evaluation by the consultant anaesthetist and included age, body mass index (BMI), cancer diagnosis, smoking history, hypertension (HTN), diabetes, ischemic heart disease (IHD), chronic obstructive pulmonary disease (COPD), cerebrovascular accident (CVA), statin use, renal disease, anaemia, and American Society of Anesthesiologists (ASA) physical status classification. 24 Only variables with less than 20% missing data were included.

| CRF
Following clinical examination and flow loop spirometry, preoperative CPET was conducted using an electromagnetically-braked cycle ergometer (Lode, Gronigen, The Netherlands) and a Medgraphics Ultima metabolic cart (MedGraphics™, Gloucester, UK) as previously described. [25][26][27] Calibration was conducted in accordance with the manufacturer's guidelines using a 3-litre volume syringe (Hans Rudolph) and reference calibration gases. During data collection, the middle five of seven breaths were averaged. Following 3 min of resting data collection, patients cycled at 60 revolutions per minute for 3 min in an unloaded freewheeling state. A progressively ramped period of exercise (5-15 W min −1 based on body mass, stature, age and sex) was then undertaken to volitional exhaustion or symptom limited termination, followed by a three-minute recovery period. 28 The Medgraphics Breeze™ software automatically determined peak oxygen uptake (VȮ 2 peak) as the highest VȮ 2 during the final 30 s of exercise. Oxygen uptake efficiency slope (OUES) 29 and peak oxygen pulse (O 2 pulse) were also automatically determined by the software. The AT was manually interpreted by the supervising consultant using the V-slope method, 30 also confirmed by a second consultant, and supported by comparison of ventilatory equivalents for oxygen (V̇E/VȮ 2 ) and carbon dioxide (V̇E/VĊO 2 ) plots. Subcategories 'Unable to CPET' and 'AT not detected' were recorded if patients were unable to perform a CPET because of their clinical status, or if insufficient data was available for clear identification of the AT.
At this centre, patients were classified 'high risk' if CPET indicated an AT <11 mL O 2 kg −1 min −1 combined F I G U R E 1 Patient flow. Consort diagram. 23 CPET, cardiopulmonary exercise test.
with V̇E/VĊO 2 -AT >34 because patients with reduced AT or ventilatory deficiency are more likely to die after major intra-abdominal surgery, whether or not they have associated cardiovascular risk factors. 9,[15][16][17] However, the AT is subject to high levels of natural (mostly biological) variation and requires caution during interpretation. 25 Therefore, optimized metrics of CRF were calculated to improve patient risk stratification. Patients unable to perform CPET, typically characterized by poor postoperative outcome, 6 were also classified as high risk.

| Surgical procedures
Procedures were obtained from the surgical database and included right or left hemicolectomy, transverse, sigmoid or subtotal colectomy, anterior or abdominoperineal resection and Hartmann's procedures. Open or laparoscopic technique were recorded. Major procedures outside these categories (such as panproctocolectomy) were classified as 'Other'. The approaches were classified to allow for comparison as surgery per se carries arguably the greatest risk.

| Postoperative morbidity
Postoperative morbidity was recorded by analysis of the surgical and critical care databases detailing length of stay for hospital, high dependency and intensive therapy unit, cardiopulmonary complications, all complications, postoperative destinations, and returns to theatre. The extent of complications were retrospectively used to construct Clavien-Dindo scores. 31

| Postoperative mortality
Postoperative mortality was determined by review of ONS records calculated by comparison of surgery date with 30-day, 90-day, 1-year, and 2-year follow-up status, tracking National Bowel Cancer Audit reference points. 32 Mortality at 1 year was selected as the primary endpoint. Whilst 90-day mortality may be the preferred surgical perspective, sufficient events for meaningful survival analysis was unlikely. Furthermore, mortality relating to the surgical response may continue for 1 year. 33

| Analysis
Statistical analysis was conducted using IBM SPSS Statistics for Windows (Version 28.0, Armonk, NY).
Distribution normality was confirmed using repeated Shapiro-Wilk W tests. Continuous data are presented as mean (standard deviation), and dichotomous variables as number (percentage). Significance for all two-tailed tests was established at p < 0.050.

| Demographics
Between-sex differences were established using independent samples t-tests for continuous data, or χ 2 tests for frequency counts.

Prognostic utility of CPET
First, we determined whether CRF was associated with postoperative mortality using Cox proportional hazards (PHs) regression models. The PH assumption was tested with Schoenfeld residuals. 34 Continuous and dichotomous variables, selected by clinical acumen and without missing values, were assessed using univariable Cox PH regression. This included age and BMI as demographic risk factors, CVA, IHD, HTN, COPD, statin therapy and diabetes as clinical risk factors, and VȮ 2 peak, VȮ 2 -AT, V̇E/VĊO 2 -AT, V̇E/VȮ 2 -AT, O 2 Pulse, OUES, Workload at AT, and peak Workload as CPET metrics. Subsequent multivariable Cox PH models were developed with variables included if p < 0.20 (from univariable analysis). Variables were removed if collinearity was identified by correlation coefficients >0.80. 35 A backward stepwise approach was employed with exclusion at p < 0.10 that yielded a final model with independent predictors of postoperative mortality. This process identified 'optimized' predictor variables for comparison with current practice.
We also determined the weighting that each risk factor contributed towards mortality using the concept of attributable fractions (AF), the estimated fraction of all deaths that would not have occurred had there been no exposure to that risk factor. 36 AF was calculated using the following equation 37 : where P c = the prevalence of exposure among cases and HR = hazard ratio for the exposure.
HRs were calculated for 1-year postoperative mortality adjusted for age, BMI, smoking history, IHD, COPD, HTN, and CVA from a separate Cox Proportional Hazards (PH) model.
Long term survival was compared between dichotomized CRF strata (high risk vs. low risk defined from the Cox PH model) and examined graphically using Kaplan-Meier plots with log-rank tests.

CRF
Mean values for peak workload, oxygen uptake at AT, VȮ 2 peak, V̇E/VĊO 2 -AT, O 2 pulse and OUES were compared between sexes using independent samples t-tests.

Postoperative outcome
A primary unadjusted Cox PH model was used to compare female and male survival with log-rank test for differences in survival curves. χ 2 tests were used to compare frequency counts for mortality, postoperative destinations, complications, and return to theatre between the sexes. Length of hospital stay was reported by median values with inter quartile range (IQR) assessed using Mann-Whitney U tests.

Defining sex-specific fitness thresholds
Area under receiver operating characteristic (AUROC) curves were constructed for markers of CRF identified as predictors of postoperative mortality. For a marker of CRF to be considered a valid predictor of mortality, an AUROC of >0.7 was used. 38 Threshold values for markers of CRF fulfilling this criterion were subsequently calculated for pooled, male, and female patients by examination of the minimal distance between AUROC plots and the upper left corner, optimizing sensitivity and specificity.
To assess the impact of sex-specific thresholds on the prediction of 1-year postoperative mortality, three risk stratification models were created. Each model used the Cox PH multivariable analysis previously described, but with scale metrics of CRF replaced with a single dichotomised value indicating patient risk (high or low). Model 1 reflected current practice (combination of AT <11 mL O 2 kg −1 min −1 and V̇E/VĊO 2 -AT >34) for baseline comparison. Model 2 included optimized metrics of CRF defined from the original multivariable Cox PH survival model to dichotomise patient risk. Model 3 included optimized metrics of CRF from Model 2, but with threshold values (re)defined relative to sex.

| Patient flow
For the final analysis, we included 640 out of 1110 patients examined (Figure 1). Females accounted for 43% of the sample (n = 274). Twenty-one patients were unable to perform CPET, including 16 females with two dying within 90 days of the CPET visit date. No serious cardiovascular complications were reported during any of the CPET visits.

| Demographics
Males were characterized by higher BMI, proportion of smokers and prevalence of diabetes, IHD and COPD, whereas ASA scores were not different (Table 1). More patients underwent laparoscopically assisted surgery compared to open surgery with equal distribution between the sexes. In contrast, more females underwent right hemicolectomy (39% vs. 27%, p = 0.002) and fewer underwent anterior resection (24% vs. 38%, p = <0.001).

| CRF and mortality
Forty-nine deaths occurred within 1 year of surgery (7.7%). The Cox PH multivariable model identified

| DISCUSSION
This study has identified two novel findings. First, CRF was identified as the principal independent risk factor for postoperative mortality in patients undergoing major colorectal surgery with high risk classification accounting for disproportionately more deaths compared to other, more established CVD risk factors. Second, optimized threshold metrics of CRF predictive of postoperative mortality were identified to be sex specific highlighting the need to reassess the current indiscriminate approach to surgical risk stratification that overestimates high risk classification in females. Collectively, these findings reinforce the clinical significance of CRF and justify the need to personalize surgical risk stratification according to sex to further optimize patient management and care.

| CRF and mortality
High risk classification was associated with poor postoperative outcome following major colorectal surgery with lower survival at all postoperative timepoints, including a disproportionate 4.8-fold increased risk of 1-year mortality, a 3.8-fold greater risk of cardiopulmonary complications, and a 2-day increased hospital length of stay. This supports existing evidence in patients undergoing major colorectal surgery. 6,8,9,17 Novel application of attributable fractions highlighted that having low CRF was responsible for 17% of deaths, greater than any of the traditional CVD risk factors. This highlights low CRF as the most important independent risk factor for postoperative mortality, similar to its relationship with 'all-cause' mortality outside of the perioperative setting. 31 Importantly, we considered the confounding potential of advancing age and corresponding reduction in CRF that has an annual mortality HR of 1.06. 39 Our attributable fractions were thus age-adjusted, and furthermore, no age differences existed between the sexes.
When considering the CPET metrics employed, Wasserman and McIlroy's development of the AT was, in part, predicated on concerns that distinct patient populations either would not, or should not, be required to undergo the exhaustive exercise protocol requisite for measurement of 'maximal' oxygen uptake (VȮ 2 max). 40 Subsequently it has become evident that good standard-of-practice for verifying that an accurate VȮ 2 max has been achieved requires at least one additional exhaustive test, ideally at a higher work rate than achieved on the initial CPET. 41 Thus, what is typically measured by CPET is referred to as the VȮ 2 peak; defined simply as the highest VȮ 2 achieved on the test. Given the opportunity for underestimating the true VȮ 2 max, 42 it is surprising that the coefficient of variation (CV) estimated for VȮ 2 peak (~6%-7%) is less than AT (~7%-10%) across patient cohorts undergoing colorectal surgery. 25 That we observed elevated Hazard Ratios for V̇E/VĊO 2 -AT, despite higher CVs, may well reflect that this threshold occurs at a lower absolute VȮ 2 (than VȮ 2 peak) and thus that the CVs reported reflects a lower absolute CV in terms of mL O 2 min −1 . It is also pertinent that the VȮ 2 max-VȮ 2 peak difference, for the patient populations considered herein, has not been determined.

| Sex differences
Lower preoperative CRF in females was reflected across all CPET metrics and agrees with previous findings, 11,12 the likely consequence of comparatively smaller skeletal muscle mass and lower convective/diffusive O 2 transport. The smaller lung size of females accompanied by smaller conducting airways relative to males, predispose greater mechanical ventilatory constraints to exercise hyperpnea, and at a given ventilation, women have a higher oxygen cost of breathing. 14 Haemoglobin concentration was lower in females, closely agreeing with a reported reduction of 12% compared with males. 43 Peak cardiac output is lower in females who, even after indexing for body size, have smaller left ventricular chambers and accordingly lower stroke volumes. 13 Vascular ageing is also accelerated in females, characterized by greater arterial elastance, higher pulse pressure, smaller and stiffer aortic arches, and earlier wave reflection than men, independent of body size and heart rate. 44  The lower CRF in females culminated in a greater proportion stratified high risk compared to males. This finding agreed with a previous study in major colorectal surgery. 9 If a presumed existence of a dose-response relationship between CRF and postoperative survival, we would have expected higher mortality in females given physiological constraints in convective/diffusive O 2 transport. However, to the contrary, our data indicate no difference in (i.e., not elevated) mortality (and morbidity) between the sexes, highlighting the need to consider other CVD risk factors that ultimately contribute to the 'collective' disease burden.
Females might experience increased sex-specific 'sensitivity' to CRF reflecting greater reliance on the systemic vascular protection afforded by CRF 45 given the lower prevalence of CVD risk factors compared to (more conditioned) male counterparts. Indeed, the collective CVD burden was higher in males and therefore may have counteracted the vascular protective benefits of elevated CRF, culminating in similar survival. However, it is important to emphasize that surgical intervention was less complicated in females given fewer anterior resections and more right hemicolectomies 8 that may have also contributed to improved postoperative outcome. Furthermore, statinuse was more prevalent in the male cohort which may also counteract 'traditional CVD risk factors' observed in males independent of, or in combination with, the effects of CRF. Albeit in a different cohort, the combination of high CRF and statin use can reduce the risk of events more than high CRF or statins alone. 46

| Preoperative risk re-stratification
These findings justify a reappraisal of current CRF thresholds employed for preoperative risk stratification, highlighting the importance of defining sex-specific thresholds to better predict postoperative outcome. Consequently, when fitness was stratified using the newly defined sex-specific thresholds, fewer females were classified high risk compared with current practice, which has important implications for surgical risk stratification and perioperative care.
Indeed, this may result in an overly risk-adverse approach with excess resource utilization (e.g., HDU beds) and associated cost implications. Our postoperative destination data did not indicate any significant between-sex differences in HDU usage; however, we were not able to distinguish between electively prepared HDU provision and perioperative decision-based allocation. Therefore, we advise caution if HDU provision is allocated based on CPET alone and reinforce the view that CPET, should be used in combination with clinical review, individualized assessment and shared decision making, 19 not in isolation.  Importantly, fewer females were stratified as high risk when reformulated CRF thresholds with corresponding lower 'cut-offs' were applied. These findings are consistent with females undergoing heart transplantation who, despite having lower CRF (VȮ 2 peak), exhibited improved 1-year transplant-free survival, 47 further reinforcing the need for sex-specific CRF thresholds that extend to other major surgeries. Given the recent adoption of multimodal prehabilitation strategies with exercise training identified as the most fundamental component, 48 our findings redefine 'target thresholds' of CRF relative to sex and highlight the increased sensitivity and importance of CRF for females in whom lower levels of CRF may respond more favourably to exercise intervention. In clinical practice, age-specific considerations are well recognized and the preoperative CRF of a 60 year old would not be expected to equal that of an 80 year old. Sex differences should thus be considered in a similar manner.

| CONCLUSIONS
This study highlights CRF as the principal independent risk factor for postoperative mortality; being high risk accounted for disproportionately more deaths than other, more traditional CVD risk factors. These findings have facilitated the assignment of 'sex-specific' CRF thresholds that may better inform surgical risk stratification to further optimize clinical decision-making and patient care.  Abbreviations:AT, anaerobic threshold; AUROC, area under receiver operating characteristic curve; V̇E/VĊO 2 -AT, ventilatory equivalent for carbon dioxide at anaerobic threshold; VȮ 2 peak, peak oxygen consumption. a p < 0.05 for between-sex or between-mortality risk differences.

AUTHOR CONTRIBUTION
companies FloTBI Inc., BrainEx Inc., and OrgEx Inc. focused on the technological development of novel biomarkers of brain injury in humans.