In this retrospective cohort of high-risk patients presenting for major abdominal cancer surgery, we found that postoperative morbidity was significantly associated with twelve variables: age, BMI, WHO status, cancer stage (TNM classified), CPET-generated data (AT, VO2 max, and AT VE/VCO2), pre-existing comorbidities (chronic renal impairment, COPD, diabetes mellitus, and a previous history of TIA or stroke), and whether a minimally invasive or an open approach was undertaken. These variables were shown to have good strength in discriminating postoperative morbidity in a prospective group of major abdominal cancer surgical patients. Using a scoring system based on the significance of each of these variables on postoperative morbidity, a simple risk scoring system called the “Marsden Morbidity Index” was devised. This index can be used in our institution to predict morbidity in patients scheduled for major abdominal surgery as a means of aiding decision-making, consent, and resource allocation. These variables were broadly reflective of the functional measures which deemed the need for CPET prior to surgery in this patient cohort.

The CPET variables we found to be associated with morbidity were in keeping with findings from previous studies where CPET was evaluated as a risk prediction tool in major abdominal surgery (Snowden et al. 2010; American Thoracic Society; American College of Chest Physicians 2003b; Hennis et al. 2011). Our study demonstrated that AT and VO2 max were significant (p < 0.001) variables at the multivariate analysis level and predictive of poor surgical outcomes. In the perioperative context, both have been shown to be strong predictors of postoperative complications and mortality in a number of cohorts analysing outcomes post major abdominal and thoracic surgery (Smith et al. 2009; Nagamatsu et al. 2015; Brunelli et al. 2014). West et al. (West et al. 2014) conducted a prospective blinded observational study to investigate for any association between CPET findings and postoperative morbidity after major colonic surgery. Patients who suffered postoperative complications had significantly lower oxygen uptake at lactate threshold, lower VO2 at peak, and higher AT VE/VCO2. These variables were found to be independently predictive of morbidity post rectal cancer surgery and major colonic surgery. Lee et al. (Lee et al. 2013) demonstrated a significant association between preoperative oxygen consumption on a 6-min walk test and postoperative medical complications (p < 0.01) post elective colorectal resection.

In our analysis, VE/VCO2 at anaerobic threshold had the strongest weighting in the model for postoperative morbidity in major abdominal surgery. This is a measure of ventilatory efficiency and is elevated in conditions such as heart failure, pulmonary embolism, and chronic lung disease (Wilson et al. 2019). It is thus unsurprising that this variable is so strongly associated with morbidity. According to Junejo et al. (Junejo et al. 2012), CPET findings for preoperative risk assessment before pancreatoduodenectomy showed VE/VCO2 at AT to be the only CPET variable independently associated with postoperative morbidity, with an AUC of 0.65 (95% CI 0.53–0.77). Similar to our study, CPET was applied in patients deemed high risk, and POMS scores were used to assess postoperative morbidity. An AT VE/VCO2 of ≥ 34.5 ml/kg/min was found to have a specificity of 84% and a sensitivity of 47%, with a PPV of 76% and an NPV of 60%, for POMS-defined morbidity.

Anaerobic threshold (VO2 at AT) was a significant CPET variable associated with postoperative morbidity in this analysis of high-risk patients undergoing major abdominal cancer surgery. This is consistent with one of our previous studies that demonstrated VO2 at AT < 10.2 ml/kg/min as a significant predictor of POMS-defined morbidity on POD 3 in patients undergoing major hepatic resection (Kasivisvanathan et al. 2015). Peak VO2 was also shown to be significantly associated with morbidity, consistent with other multiple studies (Older and Levett 2017; Andrade and Lopes 2015). It should be noted that VO2 at AT and VO2 peak usually have significant interactions so this finding is not entirely unexpected.

Objective risk identification and stratification are pivotal in linking preoperative comorbidities to risk-adapted intraoperative approaches and targeted postoperative care pathways. There are multiple grading and risk stratification tools currently in use for surgical patients. However, many of these systems are largely subjective and do not take into account any objective functional status or surgery-related factors.

The “Marsden Morbidity Index” was developed on the strong advocacy for CPET as an objective risk prediction tool based on current evidence and literature (Stringer 2010). Our aim was to combine CPET variables with premorbid variables to increase acuity in risk prediction. The use of CPET was supported by our study which demonstrated that the incorporation of CPET variables into a risk prediction tool that also takes other significant clinical variables into account creates a stronger risk prediction model. The comorbidities we identified are strongly validated in other risk scoring systems (Van Diepen et al. 2014; Barnett and Moonesinghe 2011; Stones and Yates 2019; Wong et al. 2017) currently in use, reflecting the precision of this new model. For this model, the AUC to discriminate morbidity was 0.81 and 0.79 in the fitted model binary classification and the temporal validation model respectively.

Preoperative variables deemed significant in the generation of the “Marsden Morbidity Index” risk prediction tool can be further sub-grouped into baseline parameters (age, BMI, WHO category, and TNM-classified cancer stage) and chronic conditions (COPD, diabetes mellitus, chronic renal impairment, and a previous history of TIA or stroke). The majority of these variables have been strongly validated in multiple risk prediction scores, like CHA2DS2-VASc (Van Diepen et al. 2014), p-POSSUM (Barnett and Moonesinghe 2011), Lee’s Revised Cardiac Risk Index (Stones and Yates 2019), and SORT (Wong et al. 2017) where one or more of these pre-existing variables are incorporated in a multifactorial risk-score calculation tool.

An interesting finding of our analysis showed that a low BMI scored higher than a high BMI. The effect of BMI on postoperative complications have been long studied with weight taken as a reflection of general health status from a broader perspective. From a preoperative evaluation, it reflects preoperative nutritional status, functional status, and the presence of comorbidities. While obesity is generally assumed to be a risk factor for postoperative adverse events, there is no convincing data to support this assumption (Tjeertes et al. 2015). A study published by Tjeertes et al. (Tjeertes et al. 2015) to seek more understanding of the obesity paradox revealed that while obesity alone is a significant risk factor for wound infection, more surgical blood loss, and a longer operation time, being obese is also associated with improved long-term survival. Complication and mortality rates were found to be significantly worse for underweight patients, who were most at risk of major postoperative complications, including long-term mortality. We also know from current literature that many of the CPET variables, like peak or VO2 max, are highly correlated with muscle mass (Sugie et al. 2017; Kim et al. 2016). While there is no available data on the direct comparison between CPET outcomes for high versus low BMI in cancer patients, the findings are a cause of concern that patients with low BMI are likely to perform equivocally if not worse than obese patients.

In addition, our study featured patients who underwent an open laparotomy were more likely to suffer from postoperative complications (p < 0.001) when compared to minimally invasive surgery, i.e. robotic assisted or laparoscopy. These findings are in keeping with the literature where the unique benefits and superiority of minimally invasive procedures over open procedures in selected patients have been shown (Buia et al. 2015). A systematic review and meta-analysis by Wang et al. comparing the two approaches for pancreatico-duodenectomy showed significant reductions in estimated blood loss, postoperative haemorrhage, transfusion rate, wound infection, and length of hospital stay (Wang et al. 2017). Similar findings from comparison between laparoscopy and laparotomy for rectal cancer include reductions in postoperative pain, length of stay, incisional hernia, adhesive bowel obstruction, wound complications, and mortality (Kavalukas et al. 2020). The use of robotic-assisted surgery in the management of cancer continues to increase with numerous evidence in the literature of a shorter convalescence period postoperatively (Ashrafian et al. 2017).

In conclusion, we found the CPET variables of AT, VO2 max, and AT VE/VCO2, and a number of preoperative baseline demographics and comorbidities, commonly associated with increased risk of postoperative morbidity, were shown to be associated with postoperative surgical morbidity following major abdominal oncological surgery. Our study shows that the incorporation of CPET variables into a risk prediction tool produces a model with a strong ability to discriminate postoperative complications when morbidity was assessed using a combination of the Clavien–Dindo classification scoring system and the postoperative morbidity survey.

While this model has helped us create a useful institutional tool for perioperative risks, it needs further validation in other centres performing oncological surgery. In addition, further work is required to prospectively compare the Marsden Morbidity Index’s ability to predict morbidity with other validated risk calculators, and the retrospective nature of this study and real-time evolution of current calculators prevented this for the purposes of this study. To our knowledge, the Marsden Morbidity Index is unique in that it is one of only a few validated risk-scoring tools that directly incorporate CPET variables as part of their algorithms to predict perioperative outcomes.

The strength of this study is the large number of “high-risk” cancer patients that were studied (n = 1398). This makes it one of the largest published datasets looking at the association of CPET on postoperative surgical morbidity. This was a strongly validated study, and the result reflects the high-risk cohort of patients that present to the Royal Marsden Hospital as a tertiary oncological centre.

One of the major limitations of our study is that only high-risk patients based on our institutional criteria had CPET. The ideal study design would be that all patients had CPET to limit bias in the population studied. The risk calculator is thus only valid on our high-risk patient cohort. Nonetheless, when looking at real-world use of CPET, most published data is from a high-risk cohort of patients extracted from a general surgical population. The authors chose a POMS score of > 1 on POD 7, based on similar studies using POMS scores on this day as their preferred measure to discriminate morbidity in similar major surgical cohorts (Wong et al. 2017). Another major limitation was that the study did not account for individual surgical specialities, patient pathways, and the fact that the study occurred over a 10-year period where perioperative practices changed. Despite this, the variables initially derived as being associated with morbidity were strongly validated in predicting and discriminating (AUC 0.79) in the prospectively studied population. This suggests that despite a number of important factors not being accounted for in the preoperative variables, the model is a strong tool for our population. We would be interested in implementing its use in our institution which may provide further validation of the data.