Association of preoperative anaemia with cardiopulmonary exercise capacity and postoperative outcomes in noncardiac surgery: a substudy of the Measurement of Exercise Tolerance before Surgery (METS) Study

Abstract

Background

Preoperative anaemia is associated with elevated risks of postoperative complications. This association may be explained by confounding related to poor cardiopulmonary fitness. We conducted a pre-specified substudy of the Measurement of Exercise Tolerance before Surgery (METS) study to examine the associations of preoperative haemoglobin concentration with preoperative cardiopulmonary exercise testing performance (peak oxygen consumption, anaerobic threshold) and postoperative complications.

Methods

The substudy included a nested cross-sectional analysis and nested cohort analysis. In the cross-sectional study (1279 participants), multivariate linear regression modelling was used to determine the adjusted association of haemoglobin concentration with peak oxygen consumption and anaerobic threshold. In the nested cohort study (1256 participants), multivariable logistic regression modelling was used to determine the adjusted association of haemoglobin concentration, peak oxygen consumption, and anaerobic threshold with the primary endpoint (composite outcome of death, cardiovascular complications, acute kidney injury, or surgical site infection) and secondary endpoint (moderate or severe complications).

Results

Haemoglobin concentration explained 3.8% of the variation in peak oxygen consumption and anaerobic threshold (P<0.001). Although not associated with the primary endpoint, haemoglobin concentration was associated with moderate or severe complications after adjustment for peak oxygen consumption (odds ratio=0.86 per 10 g L⁻¹ increase; 95% confidence interval, 0.77–0.96) or anaerobic threshold (odds ratio=0.86; 95% confidence interval, 0.77–0.97). Lower peak oxygen consumption was associated with moderate or severe complications without effect modification by haemoglobin concentration (P=0.12).

Conclusion

Haemoglobin concentration explains a small proportion of variation in exercise capacity. Both anaemia and poor functional capacity are associated with postoperative complications and may therefore be modifiable targets for preoperative optimisation.

Editor's key points.

• •Impaired preoperative exercise capacity and preoperative anaemia are probably both associated with postoperative complications.
• •It is unknown whether the association between anaemia and postoperative complications can be completely explained by the impaired exercise capacity related to anaemia.
• •This study found that preoperative anaemia did not substantially explain impaired exercise capacity, and even after adjusting for impaired exercise capacity and other variables, worsening preoperative anaemia still showed a graded association with postoperative complications.
• •It is therefore important to clarify whether preoperative correction of anaemia can positively impact postoperative outcomes.

Preoperative anaemia is a common problem.¹ Anaemia has also consistently been associated with postoperative mortality and morbidity, even after accounting for the underlying cause of anaemia, associated transfusion requirements, and co-existing disease.1, 2, 3, 4, 5, 6 The underlying mechanisms that explain this association remain unclear.

As preoperative anaemia is a marker of underlying frailty or comorbidity, previously observed associations between preoperative anaemia and elevated perioperative risk may be explained, in part, by residual confounding.3, 4, 5 In addition, anaemic patients might be unable to increase oxygen delivery in response to the increased metabolic demands imposed during the perioperative period.² Consistent with this possibility, patients with anaemia have decreased exercise capacity as measured by cardiopulmonary exercise testing (CPET).² Importantly, CPET-derived measures of exercise capacity have been shown to predict complications and death after noncardiac surgery,7, 8, 9 with the most commonly studied measures being peak oxygen consumption (VO₂ peak) and anaerobic threshold (AT).⁷ For example, in the recently published Measurement of Exercise Tolerance before Surgery (METS) prospective cohort study, lower VO₂ peak, but not lower AT, was predictive of moderate or severe complications after major noncardiac surgery.¹⁰

During CPET, VO₂ peak is defined as the highest oxygen uptake measurement recorded during the incremental exercise test, typically when a subject has reached a plateau near to a maximal effort.⁷ The AT is a less effort-dependent measure that is calculated indirectly from expired gases. It is thought to reflect the point at which the venous lactate concentration begins to increase because tissue oxygen demand exceeds supply.¹¹ When taken together, VO₂ peak and AT broadly indicate the ability of the cardiovascular system to provide peripheral tissues with sufficient oxygen, and the corresponding ability of the tissues to utilise that oxygen. Haemoglobin concentration, as a primary determinant of oxygen-carrying capacity, would be expected to influence both VO₂ peak and AT in a significant manner.¹²

The relative contributions of anaemia and exercise capacity with respect to preoperative risk stratification remain unclear. An improved understanding of the association between anaemia and exercise capacity, and the interaction between anaemia and exercise capacity in the prediction of postoperative outcomes may help clinicians better target interventions to improve postoperative outcomes. We therefore conducted a substudy of the METS study to (i) determine the association between haemoglobin concentration and exercise capacity as characterised by VO₂ peak and AT, and to (ii) determine the adjusted association of haemoglobin concentration and exercise capacity with postoperative outcomes.

Methods

Design, setting, participants, and data sources

This study adhered to the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) recommendations for the design and reporting of observational studies.¹³ We conducted a pre-specified substudy of the METS prospective cohort study, which was conducted at 25 hospitals in Canada, New Zealand, Australia, and the UK.10, 14 Eligible patients had to be aged ≥40 yr; scheduled for elective in-patient noncardiac surgery under general anaesthesia, regional anaesthesia, or both, and deemed to have one or more risk factors for cardiac complications or coronary artery disease (Supplementary Table S1). All participants provided written informed consent, and each centre obtained research ethics board approval before commencing recruitment, which occurred between March 1, 2013 and March 25, 2016.

During the period from recruitment to 1 day before surgery, participants underwent symptom-limited CPET on a cycle ergometer using a standardised protocol (Supplementary material—Appendix). Baseline information collected for each participant included any preoperative haemoglobin concentration measurement ordered as part of clinical care. Participants were prospectively followed daily while in hospital to ascertain the presence of specific postoperative complications (Supplementary Table S2), the severity of which were categorised as mild, moderate, severe, or fatal using a modified Clavien–Dindo scheme (Supplementary Table S3).14, 15 Other in-hospital follow-up included the validated Postoperative Morbidity Survey (POMS) instrument,¹⁶ electrocardiograms, and blood sampling to measure troponin and creatinine concentrations. ECGs and blood sampling were performed daily for the first 3 days after surgery, whereas the POMS instrument was administered on the third and fifth days after surgery. Other complications identified during follow-up included myocardial infarction (MI), heart failure, cardiac arrest, stroke, transient ischemic attack, acute kidney injury (AKI), and surgical site infection. MI was diagnosed using the Third Universal Definition of Myocardial Infarction,¹⁷ whereas AKI was diagnosed based on the Kidney Disease Improving Global Outcomes (KDIGO) criteria.¹⁸ Participants were contacted at 30 days and 1 yr after surgery to assess remaining study outcomes.

This substudy consisted of two components: a nested cross-sectional study (Component 1) that determined the association of haemoglobin concentration with VO₂ peak and AT, and a nested cohort study (Component 2) that determined the adjusted association of haemoglobin concentration, VO₂ peak, and AT with postoperative outcomes. The inclusion and exclusion criteria for this substudy were the same as those of the overall METS study,10, 14 with two important differences. Firstly, the substudy excluded participants with missing preoperative haemoglobin concentration measurements. Secondly, Component 1 included otherwise eligible consenting participants regardless of whether they underwent surgery (i.e. individuals who underwent CPET but not surgery were included). Based on these criteria, of the 1741 individuals who consented to participate in the METS study, 1279 were included in Component 1 and 1256 were included in Component 2 (Fig. 1).

Fig 1.

Participant recruitment, screening, and follow-up. CPET, cardiopulmonary exercise testing; METS, Measurement of Exercise Tolerance before Surgery.

Variables

Trained investigators at each centre determined VO₂ peak and AT using on a protocol-based evaluation of the plotted CPET data. Haemoglobin concentration measurements were obtained from preoperative blood testing performed as part of usual clinical care. Estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-Epi) formula based on the most recent creatinine value collected before operation.¹⁹ Participants with missing creatinine values, and no scored comorbidities on preoperative questionnaires were assumed to have normal renal function based on their age and sex. These missing eGFR values (n=33) were imputed using the KDIGO estimated baselines for age and weight,¹⁸ or assumed to be 10 ml min⁻¹ 1.73 m⁻² if the patient was on dialysis.

In the nested cross-sectional study (Component 1), haemoglobin concentration was the principal exposure, whereas VO_{2 peak} and AT were co-outcomes in a multivariate regression model (i.e. multiple outcomes modelled simultaneously). In the nested cohort study (Component 2), haemoglobin concentration was the principal exposure whereas the primary outcome was a composite endpoint that captured events potentially attributable to anaemia-related disease processes (e.g. impaired peripheral oxygen delivery). These outcome components were chosen on the basis of known associations with anaemia in the perioperative literature and relevance to impaired peripheral oxygen delivery.20, 21, 22, 23, 24 The components of the primary outcome were 30-day mortality, MI, cardiac arrest, heart failure, stroke, transient ischemic attack, AKI (Stage 2 or 3 based on the KDIGO criteria),¹⁸ and surgical site infection (Supplementary Table S2). The secondary outcome was the presence of moderate or severe (including fatal) postoperative complications (defined in Supplementary Table S3). These events included any complication (including, but not limited to, those in Supplementary Table S2) that met the definitions of a moderate, severe, or fatal complication. All outcomes were specified a priori before the completion of the METS study.

Statistical methods

SAS University Edition (release version August 2015) was used for all analyses, which were also pre-specified before the completion of the METS study. Statistical significance was defined (where needed) by a two-sided P-value ≤0.05. As missing data were very uncommon, a complete case analysis was used. Descriptive statistics were used to characterise the sample (separately in the cross-sectional analysis and nested cohort analysis), both overall and across strata defined by WHO grades of anaemia (Supplementary Table S4). Categorical variables were characterised using counts (proportions), whereas continuous variables were characterised using means (standard deviation) and medians (inter-quartile range).

Nested cross-sectional study (Component 1)

The cross-sectional analysis sought to determine the unadjusted and adjusted associations of haemoglobin concentration (exposure) with VO₂ peak and AT (outcomes). As haemoglobin concentration appeared to have a linear association with VO₂ peak and AT, multivariable linear regression modelling was first used to separately determine the association of haemoglobin concentration with each of VO₂ peak and AT. For these analyses, clinically sensible covariates were identified from the literature, including age, sex, coronary artery disease, heart failure, peripheral arterial disease, diabetes mellitus, preoperative eGFR, obstructive lung disease, preoperative chemotherapy, and arthritis.¹⁰

Subsequently, we determined the association of haemoglobin concentration (independent variable) with both VO₂ peak and AT (co-dependent outcomes) using a multivariate linear regression model. This approach allowed us to examine the impact of haemoglobin concentration on AT and VO₂ peak simultaneously, rather than separately. As the results of linear regression models for AT and VO₂ peak are unlikely to be independent of each other, a multivariate analysis allowed greater model complexity and better analysis of variable interactions. To ensure the appropriateness of a multivariate model, the correlation between AT and VO_{2 peak} was first confirmed. Multivariate tests were then used to evaluate the variance in the co-dependent variables that was explained by the pre-specified predictor variables (same covariates included in the linear regression models). All model assumptions for linear regression were verified, including normality of residuals, homogenous variances of residuals, independence of outcomes, absence of multicollinearity, and model overfitting. Residual plots were used to help examine the underlying assumptions of a linear regression model.

Nested cohort study (Component 2)

Multivariable logistic regression models were used to determine the adjusted association of haemoglobin concentration (principal exposure) with the primary and secondary endpoints, while adjusting for CPET performance (characterised as VO₂ peak and AT) and several pre-specified clinically sensible covariates (age, sex, surgery type). As VO₂ peak and AT are correlated with each other, two separate logistic regression models were developed, one including VO₂ peak and the other including AT instead. Haemoglobin concentration, VO₂ peak, AT, and age were treated as continuous variables in these regression analyses. Adjusted associations were expressed as odds ratios (ORs) with 95% confidence intervals (CIs). Assumptions of logistic regression modelling were verified, model discrimination was evaluated using the c-statistic, and model goodness-of-fit was assessed with the Hosmer–Lemeshow statistic.

Study sample size

The number of patients included in this substudy was determined by the sample size calculation for the overall METS study and the proportion of participants who underwent preoperative haemoglobin concentration measurement as part of usual clinical care.¹⁰ Instead of using unhelpful post-hoc power calculations,²⁵ we focused on observed 95% confidence limits when interpreting the results of this substudy.

Results

Nested cross-sectional study (Component 1)

The cross-sectional analysis included 1279 participants (Fig. 1), of whom 234 (18.3%) were anaemic based on WHO criteria.²⁶ Compared with non-anaemic individuals, anaemic patients had lower exercise capacity (VO₂ peak and AT), lower weight, and greater burden of comorbidity (Table 1). As expected, VO₂ peak and AT were strongly correlated with each other (Pearson's rho=0.75, P<0.001). Conversely, haemoglobin concentration was weakly correlated with AT (Pearson rho=0.22, P<0.001) and VO₂ peak (Pearson's rho=0.31, P<0.001).

Table 1.

Characteristics of study cohort, stratified by WHO anaemia class. Means and standard deviations (sd) or medians and inter-quartile ranges (IQR) shown for continuous data. Counts and proportions are shown for categorical data. Continuous data were compared between strata using either analysis of variance (anova) or the Wilcoxon rank-sum test. Categorical data were compared using Fisher's exact test. AT, anaerobic threshold; eGFR, estimated glomerular filtration rate calculated using the CKD-Epi formula.16, 27

Variables	Non-anaemic (n=1045)	Mild anaemia (n=177)	Moderate or severe anaemia (n=57)	P-value
Haemoglobin concentration (g L−1), mean (sd)	143.0 (11.4)	119.9 (5.7)	100.2 (9.0)	<0.001
Age (yr), mean (range)	64.1 (40–92)	65.7 (40–88)	64.3 (40–86)	0.18
BMI (kg m−2), median (IQR)	28.0 (25.0, 31.8)	27.5 (23.5, 31.4)	25.5 (22.1 29.6)	0.0002
VO2 peak (ml kg−1 min−1), median (IQR)	19.0 (15.0, 23.0)	16.1 (13.6, 20.0)	14.7 (12.0, 18.9)	<0.001
AT (ml kg−1 min−1), median (IQR)	12.0 (10.0, 15.0)	11.1 (9.2, 14.0)	10.6 (8.8, 12.0)	<0.001
eGFR (ml min−1 1.73 m−2)
≥60	876 (83.8%)	138 (79.0%)	36 (63.2%)	<0.001
30–59	146 (14.0%)	27 (15.3%)	8 (14.0%)
<30 or dialysis	23 (2.2%)	12 (6.8%)	13 (22.8%)
Comorbidities
Coronary artery disease	72 (6.9%)	30 (17.0%)	10 (17.5%)	<0.001
Heart failure	83 (7.9%)	33 (18.6%)	11 (19.3%)	<0.001
Diabetes mellitus	193 (18.5%)	36 (20.3%)	10 (17.5%)	0.82
Obstructive lung disease	119 (11.4%)	25 (14.1%)	9 (15.8%)	0.39
Preoperative chemotherapy	62 (5.9%)	34 (19.2%)	11 (19.3%)	<0.001
Arthritis	381 (36.5%)	57 (32.2%)	14 (24.6%)	0.21

The unadjusted association of haemoglobin concentration with VO₂ peak and AT are presented separately in Fig. 2a and b. For each 10 g L⁻¹ increase in haemoglobin concentration, VO₂ peak increased by 1.27 ml kg⁻¹ min⁻¹ (95% CI, 1.06–1.49; P<0.001) and AT increased by 0.59 ml kg⁻¹ min⁻¹ (95% CI, 0.44–0.74; P<0.001). When age, sex, and comorbidities were included as covariates, the R² for the model predicting VO₂ peak increased from 0.10 to 0.25, whereas the R² for the model predicting AT increased from 0.05 to 0.12. After adjustment, each 10 g L⁻¹ increase in haemoglobin concentration was associated with a 0.71 ml kg⁻¹ min⁻¹ (95% CI, 0.48–0.93; P<0.001) increase in VO₂ peak and a 0.32 ml kg⁻¹ min⁻¹ (95% CI, 0.16–0.48; P<0.001) increase in AT. The adjusted model results are presented in Supplementary Tables S5 and S6. The statistical evidence was weak for an interaction between haemoglobin concentration and sex in the models predicting VO₂ peak (P=0.07) or AT (P=0.22).

Fig 2.

Fit plots for unadjusted linear regression modelling of haemoglobin concentration with VO₂ peak and anaerobic threshold (AT). (a) Association between haemoglobin concentration and VO₂ peak. (b) Association between haemoglobin concentration and anaerobic threshold.

As VO₂ peak and AT were strongly correlated with each other, we proceeded to use multivariate linear regression modelling to determine the simultaneous association of haemoglobin concentration with VO₂ peak and AT. In an unadjusted multivariate model, 11% of the variance in the co-outcomes (AT and VO₂ peak) was explained by haemoglobin concentration. In an adjusted multivariate model (Table 2), haemoglobin concentration explained 3.8% of the variance in the co-outcomes, whereas sex explained 9.3% and age explained 3.4%. The other covariates each accounted for less than 3% of the variance in VO₂ peak and AT.

Table 2.

Adjusted multivariate modelling (prediction of VO₂ peak and AT simultaneously). AT, anaerobic threshold; VO₂ peak, peak oxygen consumption; CAD, coronary artery disease; eGFR, estimated glomerular filtration rate (calculated using CKD-Epi formula¹⁸); CKD-Epi, Chronic Kidney Disease Epidemiology Collaboration.

Variable	Proportion of variance explained	P-value
Age	0.034	<0.001
Female sex	0.093	<0.001
Haemoglobin concentration (g L−1)	0.038	<0.001
Coronary artery disease	0.004	0.11
Heart failure	0.003	0.14
Diabetes mellitus	0.023	<0.001
eGFR (ml min−1 1.73 m−2)	0.002	0.29
Obstructive lung disease	0.004	0.11
Preoperative chemotherapy	0.006	0.03
Arthritis	0.003	0.18

Nested cohort study (Component 2)

The nested cohort analysis included 1256 participants. Overall, 93 (7.4%) participants experienced the composite primary endpoint, 182 (14.5%) experienced moderate or severe complications, and 276 (22.0%) experienced one or more POMS morbidity events. The participants' baseline characteristics in this sample (Supplementary Table S7) were similar to those of the sample included in Component 1. There was no discernible difference in the occurrence of the primary composite endpoint between anaemic and non-anaemic patients (OR per 10 g L⁻¹ increase in haemoglobin concentration 0.94; 95% CI, 0.83–1.08; P=0.92). The event rates for individual components of the composite endpoint, stratified by anaemia, are presented in Supplementary Table S8. Compared with non-anaemic individuals, anaemic patients experienced significantly higher unadjusted risks of moderate or severe complications (OR per 10 g L⁻¹ increase in haemoglobin concentration 0.79; 95% CI, 0.69–0.90; P<0.001) and POMS morbidity events (not anaemic, 28.7% [n=206] vs anaemic, 47.6% [n=70]; P<0.001).

In multivariable logistic regression analyses (Table 3), we found no strong evidence of an association between VO₂ peak, AT, or haemoglobin concentration with the composite primary endpoint. However, haemoglobin concentration did have a strong association with moderate or severe complications (Table 4), after adjustment for either VO₂ peak (OR per 10 g L⁻¹ increase in haemoglobin concentration 0.86; 95% CI, 0.77–0.96; P=0.007) or AT (OR per 10 g L⁻¹ increase in haemoglobin concentration 0.86; 95% CI, 0.77–0.97; P=0.01).

Table 3.

Multivariable logistic regression models predicting the composite endpoint, with separate model results for VO₂ peak and anaerobic threshold. CI, confidence interval; VO₂ peak, peak oxygen consumption.

Independent variable	Odds ratio	95% CI	P-value
Adjusted association between haemoglobin concentration and composite primary endpoint—with adjustment for VO2 peak
Age (per 10 yr)	1.00	0.98–1.03	0.71
Female sex	0.78	0.46–1.33	0.37
Haemoglobin concentration (per 10 g L−1 increase)	0.97	0.84–1.14	0.71
VO2 peak (ml kg−1 min−1)	0.99	0.95–1.03	0.59
Surgical procedure type
Intra- or retroperitoneal or intrathoracic or vascular	Reference
Urology or gynaecology	0.55	0.32–0.96	0.01
Orthopaedic	0.42	0.21–0.84
Other	1.32	0.65–2.70
Association between haemoglobin concentration and composite primary endpoint—with adjustment for anaerobic threshold
Age (per 10 yr)	1.04	0.82–1.31	0.75
Female sex	0.83	0.48–1.41	0.49
Haemoglobin concentration (per 10 g L−1)	0.98	0.84–1.15	0.79
Anaerobic threshold (ml kg−1 min−1)	1.00	0.94–1.06	0.94
Surgical procedure type
Intra- or retroperitoneal or intrathoracic or vascular	Reference
Urology or gynaecology	0.50	0.27–0.88	0.02
Orthopaedic	0.43	0.19–0.85
Other	1.08	0.47–2.25

c-Statistic=0.61; Hosmer–Lemeshow goodness-of-fit test, P=0.87. The interaction term between haemoglobin concentration and VO₂ peak was not statistically significant (P=0.66).

c-Statistic=0.61; Hosmer–Lemeshow goodness-of-fit test, P=0.48. The interaction term between haemoglobin concentration and AT was not statistically significant (P=0.51).

Table 4.

Multivariable logistic regression models predicting moderate and severe complications, with separate model results for VO₂ peak and anaerobic threshold; CI, confidence interval; VO₂ peak, peak oxygen consumption.

Independent variable	Odds ratio	95% CI	P-value
Adjusted association between haemoglobin concentration and moderate or severe complications—with adjustment for VO2 peak
Age (per 10 yr)	1.04	0.88–1.24	0.62
Female sex	0.53	0.35–0.79	0.002
Haemoglobin concentration (per 10 g L−1)	0.86	0.77–0.96	0.007
VO2 peak (ml kg−1 min−1)	0.96	0.93–0.99	0.01
Surgical procedure type
Intra- or retroperitoneal or intrathoracic or vascular	Reference
Urology or gynaecology	0.33	0.22–0.49	<0.001
Orthopaedic	0.12	0.06–0.22
Others	0.35	0.16–0.68
Adjusted association between haemoglobin concentration and moderate or severe complications—with adjustment for anaerobic threshold
Age (per 10 yr)	1.11	0.93–1.32	0.26
Female sex	0.58	0.38–0.87	0.009
Haemoglobin concentration (per 10 g L−1)	0.86	0.77–0.97	0.01
Anaerobic threshold (ml kg−1 min−1)	0.98	0.93–1.02	0.35
Surgical procedure type
Intra- or retroperitoneal or intrathoracic or vascular	Reference
Urology or gynaecology	0.31	0.20–0.47	<0.001
Orthopaedic	0.12	0.05–0.23
Others	0.32	0.14–0.64

c-Statistic=0.74; Hosmer–Lemeshow goodness-of-fit test, P=0.21; interaction term between haemoglobin concentration and VO₂ peak was not statistically significant (P=0.12).

c-Statistic=0.73; Hosmer–Lemeshow goodness-of-fit test, P=0.95; interaction term between haemoglobin concentration and AT was not statistically significant (P=0.09).

The individual clinical events that constituted moderate or severe complications are presented in Supplementary Tables S9 and S10. In multivariable logistic regression analyses, there was a strong association between VO₂ peak and moderate or severe complications (OR per 1 ml kg⁻¹ min⁻¹ increase in VO₂ peak 0.96; 95% CI, 0.94–0.99; P=0.01), but not between AT and moderate or severe complications (OR per 1 ml kg⁻¹ min⁻¹ increase in AT 0.98; 95% CI, 0.93–1.02; P=0.35). There was no strong evidence of interaction between haemoglobin concentration and VO₂ peak (P=0.12), or between haemoglobin concentration and AT (P=0.09).

Discussion

In this pre-specified substudy of an international multicentre cohort study, we found that preoperative haemoglobin concentration was associated with VO₂ peak and AT. This association was modest in magnitude in unadjusted analyses, with further attenuation after adjustment for age, sex and comorbidities. Although haemoglobin concentration explained more variation in VO₂ peak and AT than many comorbidities in multivariate regression analysis, a significant proportion of the variation in VO₂ peak and AT was not accounted for by sex, haemoglobin concentration, age, or comorbidities. When evaluating the association of haemoglobin concentration with postoperative outcomes, we found no association between lower haemoglobin concentration and the primary composite endpoint that included complications thought to reflect impaired peripheral oxygen delivery. Nonetheless, we found that both lower haemoglobin concentration and lower VO₂ peak were associated with elevated adjusted odds of experiencing moderate or severe postoperative complications. These findings extend the results of the primary METS study report by showing that association of lower VO₂ peak with moderate or severe postoperative complications is independent of concomitant anaemia.¹⁰

The overarching aim of this substudy was to determine the impact of anaemia on exercise tolerance, and to determine whether exercise tolerance might influence the postoperative outcomes of anaemic patients. Although the association of anaemia with adverse perioperative outcomes is well documented,1, 3, 6 there are only limited perioperative data that sought to examine why outcomes are worse in anaemic patients. Our study provides important new insights into these issues. We have confirmed the work of Otto and colleagues²⁸ that demonstrated a modest association of haemoglobin concentration with VO₂ peak and AT in standard multivariable linear regression modelling approaches. Haemoglobin concentration explained 5.5% of variation in VO₂ peak and 3.5% of variation in AT, after adjustment for age, sex, weight, and study centre. When we conducted multivariate regression analyses that specifically allowed for simultaneous examination of VO₂ peak and AT as correlated co-dependent outcomes, haemoglobin concentration explained 3.8% of the variation in VO₂ peak and AT. The similarity of our findings suggests that haemoglobin concentration has an impact on preoperative cardiopulmonary fitness, albeit of small magnitude. A follow-up study by Otto and colleagues²⁷ found that total haemoglobin mass was a better determinant of exercise capacity, in that it explained 23% of variance in VO₂ peak and 44% of variance in AT. Although promising as a marker of preoperative exercise capacity, total haemoglobin mass measurement requires additional testing equipment that is presently not widely available in the clinical setting.

In the primary analysis of the METS study, there was a moderately strong positive correlation between Duke Activity Status Index (DASI) scores and VO₂ peak (Spearman coefficient, 0.41), and a weak negative correlation between NT pro-BNP concentrations and VO₂ peak (Spearman coefficient, –0.21).¹⁰ These findings suggest that DASI scores and NT pro-BNP concentrations measure underlying constructs aside from exercise capacity, such as sarcopenia or frailty.¹⁰ Similarly, our finding that haemoglobin concentration has relatively weak associations with VO₂ peak and AT, yet remains associated with postoperative complications, suggests anaemia exerts deleterious effects through several mechanisms aside from reduced oxygen carrying capacity. Consistent with this possibility, recent evidence suggests that the perioperative risks associated with anaemia also vary based on the specific type of anaemia (i.e. microcytic, normocytic, macrocytic).²⁹ Thus, specific underlying disease states (e.g. malnutrition, sarcopenia, frailty) may have simultaneous effects on absolute haemoglobin concentration, anaemia type, and exercise capacity. As the primary METS study dataset did not capture several of these relevant factors, future studies should replicate our analyses while also capturing additional information relevant to the interplay between haemoglobin concentration, anaemia type, and exercise capacity (e.g. nutritional status, anaemia type, iron status).

Especially as haemoglobin concentration was associated with cardiopulmonary fitness, we also sought to better understand the relative prognostic importance of anaemia and poor fitness when evaluating perioperative risk. Notably, poor cardiopulmonary fitness is itself associated with elevated risks of adverse events after surgery,¹⁰ possibly because patients with poor fitness are not able to meet the increased metabolic demands of the postoperative state.10, 30 In our study, lower preoperative haemoglobin concentration was associated with elevated adjusted odds of moderate to severe postoperative complications, but not the primary composite endpoint.⁶ The elevated risk of moderate to severe postoperative complications in anaemic patients was also independent of concomitant poor exercise tolerance, without evidence of effect modification. The differing association of haemoglobin concentration with these two outcomes might be explained, in part, by anaemia having a greater pathophysiological relationship to frequent individual events in moderate or severe complications (i.e. respiratory failure, pneumonia, unexpected critical care unit admission, re-operation, surgical site infection) vs frequent individual events in the primary composite endpoint (i.e. AKI, surgical site infection).¹⁰ Thus, the types of outcomes studied in relation to exercise capacity and preoperative anaemia may matter. In addition, although there were sufficient outcomes (182 moderate-or-worse complication events and 93 primary composite endpoint events) to permit reliable estimation of the coefficients for the seven predictor variables in the multivariable logistic regression models (Table 3, Table 4), there was likely greater statistical power in analyses pertaining to moderate or worse complications than the primary composite endpoint.³¹ Consistent with this possibility, the CIs for the adjusted association of haemoglobin concentration with the primary composite endpoint (Table 3) did not exclude a plausible effect (e.g. adjusted OR 0.84 per 10 g L⁻¹ increase in haemoglobin concentration).

Our study has important strengths, especially with respect the data source being a large multicentre prospective cohort study with minimal loss to follow-up and minimal incomplete data.¹⁰ Conversely, several important limitations should also be noted. For example, the VO₂ peak and AT data were determined by trained investigators at each centre. It is possible that central adjudication of CPET results might have altered these measurements. Additionally, as might be expected in a study requiring strenuous exercise testing, there was likely a selection bias for healthier study participants. Consistent with this possibility, 14% of the study sample had mild anaemia and only 4% had moderate or severe anaemia. Thus, further research is necessary to confirm our findings in higher-risk patients, especially as other factors (e.g. frailty) may better explain impaired functional capacity in high-risk surgical populations.³²

Conclusions

Anaemia is a potentially modifiable risk factor that should be corrected before operation for a variety of reasons.¹⁰ Although our findings support the association of anaemia with poor preoperative fitness, the relative contribution of preoperative anaemia is small. Future research could explore the possibility that correction of both anaemia and poor cardiopulmonary fitness might improve postoperative outcomes in surgical patients.

Authors' contributions

Writing of the first draft of the manuscript: JB

All authors contributed to the contributed to the conception and design of the study. All authors contributed to the acquisition, analysis and interpretation of the data. All authors revised the manuscript critically for important intellectual content. All authors read and approved the final version of the manuscript.

Handling editor: M. Avidan

Editorial decision date: 09 April 2019

Declaration of interest

The authors declare that they have no conflicts of interest.

Funding

The METS study was supported by grants from the Canadian Institutes of Health Research; Heart and Stroke Foundation of Canada; Ontario Ministry of Health and Long-Term Care; Ontario Ministry of Research, Innovation and Science; United Kingdom (UK) National Institute of Academic Anaesthesia; UK Clinical Research Collaboration; Australian and New Zealand College of Anaesthetists; and Monash University (Melbourne, Victoria, Australia). These sponsors had no role in the design and conduct of the METS study; collection, management, analysis and interpretation of the data; preparation, review or approval of this paper; and decision to submit this manuscript for publication. Professor Wijeysundera is supported in part by a New Investigator Award from the Canadian Institutes of Health Research, an Excellence in Research Award from the Department of Anesthesia at the University of Toronto, and the Endowed Chair in Translational Anesthesiology Research at St. Michael's Hospital and University of Toronto. Professor Laupacis is supported in part by a Canada Research Chair in Health Policy and Citizen Engagement from the Government of Canada.

Appendix A. Supplementary data

The following is the Supplementary data to this article: