Background
The American College of Surgeons National Surgical Quality Improvement Program (NSQIP®) surgical risk calculator (American College of Physicians
n.d.), released in 2013 (Bilimoria et al.
2013), was developed using data from more than 1.4 million patients, encompassing 1557 unique Current Procedural Terminology (CPT) codes. Patient-related preoperative variables include age, sex, functional status, American Society of Anesthesiologists (ASA) classification, steroid use for chronic conditions, ascites within 30 days prior to surgery, systemic sepsis within 48 h prior to surgery, ventilator dependency, disseminated cancer, diabetes (DM), hypertension (HTN), congestive heart failure 30 days prior to surgery, dyspnea, current smoker within 1 year, history of severe chronic obstructive pulmonary disease (COPD), dialysis, acute renal failure, and body mass index (BMI). The calculator also takes into account the type of surgery, based on specific CPT codes and emergency status (American College of Physicians
n.d.). Although the NSQIP surgical risk calculator demonstrated excellent performance in predicting postoperative mortality (c-statistic = 0.944), morbidity (c-statistic = 0.816), and six surgical complications (c-statistics > 0.8), it was not developed specifically for use in the older patient population, which is rapidly growing in the USA and is known to have worse surgical outcomes than younger patients (Sukharamwala et al.
2012; Raats et al.
2015; Bentrem et al.
2009). Specifically, the NSQIP surgical risk calculator does not include factors such as frailty and mobility that are known to be important predictors of surgical outcomes in older patients (Kim et al.
2016; Makary et al.
2010).
The NSQIP and the American Geriatric Society published a best practice guideline for optimal preoperative assessment of older surgical patients (Chow et al.
2012). The guideline recommends assessment of patients’ gait and mobility impairment and fall risk, using tests such as the Timed Up and Go test to assess surgical risk in older patients. While the NSQIP surgical risk calculator does include patients’ functional status as a categorical variable with three options (independent, partially dependent, and totally dependent), this variable does not capture the various degrees of functional limitation that older patients experience. Furthermore, the NSQIP surgical risk calculator does not take into account patients’ mobility status.
In a prospective study of older surgical patients, we assessed preoperative self-reported mobility using a novel tool, the Mobility Assessment Tool-short form (MAT-sf), and found it to be a good predictor for postoperative complications, hospital length of stay (LOS), and nursing home placement (Kim et al.
2016). Here, we used data from that study to develop a simplified preoperative risk assessment model that includes measures of self-reported mobility to predict postoperative outcomes in older patients. The aim of this study was to compare the performance of these simple surgical risk calculators to the NSQIP surgical risk calculator in predicting postoperative outcomes among older patients.
Discussion
We set out to develop a surgical risk calculator that was simple and specific to the older surgical patient population. We developed three models using data from our prospective cohort of 197 older surgical patients and cross-validated the models to provide robust probabilities of risk. The model using only the MAT-sf score had similar ability for predicting postoperative complications and nursing home placement as the NSQIP surgical risk calculator. Thus, the MAT-sf score provides a simple, easy to use tool to predict surgical outcomes in older patients. The model performed as well as the NSQIP surgical risk calculator in predicting LOS.
The NSQIP risk calculator was developed using data from more than 1.4 million patients and has demonstrated excellent performance in predicting postoperative outcomes (American College of Physicians
n.d.; Bilimoria et al.
2013). However, the NSQIP calculator was not specifically developed for use in an older patient population, in which unique factors, such as mobility, are important predictors of surgical outcomes. There have been efforts to develop simpler tools to predict postoperative outcomes in the older patient population. A few studies have used mobility as a marker of postoperative outcomes. Robinson et al. tested the Timed Up and Go test as a predictive tool among 272 older patients (98 patients undergoing colorectal surgery and 174 patients undergoing cardiac surgery). The Timed Up and Go test measures the time it takes for an individual to rise from a chair, ambulate 10 ft, turn around a cone marker, return to the chair, and sit back down. In patients undergoing either colorectal or cardiac surgery, the study reported the ROC AUC for predicting postoperative complications using the Timed Up and Go test as 0.775 and 0.684, respectively, compared to 0.554 and 0.552 for standard-of-care surgical risk calculators (Robinson et al.
2013). The MAT-sf tool had similar ROC AUCs for postoperative complications as those reported for the Timed Up and Go test, and we did not see any difference between the MAT-sf tool and the NSQIP surgical risk calculator. One important distinction between these two simple perioperative risk score measures is that the Timed Up and Go is a performance test of physical function whereas the MAT-sf is a self-report measure. Consequently, the Timed Up and Go might not be as feasible due to space and patient limitations. Since the study by Robinson et al. limited the enrollment criteria to patients who were undergoing colorectal or cardiac surgery and analyzed them separately, the risk associated with each type of surgery could be controlled. By contrast, our study enrolled any older patients who were undergoing elective surgery (Table
5). Given the relatively small sample size of our study, we could not control for the risk associated with each type of surgery. In the models that used the covariates alone or covariates plus MAT-sf, we controlled for the surgical risk in terms of low vs. intermediate-to-high risk using the definitions from American College of Cardiology/American Heart Association guidelines (Fleisher et al.
2014).
Table 5
Summary of surgical procedures by sex
Orthopedic surgery | 42 | 72 | 114 |
Hip | 13 | 26 | |
Knee | 9 | 18 | |
Spine | 18 | 21 | |
Other | 2 | 7 | |
Urology | 25 | 4 | 29 |
Kidney | 5 | 2 | |
Prostate | 10 | 0 | |
Other | 10 | 2 | |
Intraperitoneal | 13 | 7 | 20 |
Colorectal | 5 | 4 | |
Hernia | 3 | 1 | |
Other | 5 | 2 | |
Otolaryngology | 5 | 8 | 13 |
Thyroid | 0 | 3 | |
Other | 5 | 5 | |
Vascular | 6 | 3 | 9 |
Carotid endarterectomy | 4 | 1 | |
Other | 2 | 2 | |
Gynecology | 0 | 4 | 4 |
Hysterectomy | 0 | 2 | |
Other | 0 | 2 | |
Neurosurgery | 2 | 1 | 3 |
Pituitary | 1 | 1 | |
Other | 1 | 0 | |
Other | 3 | 2 | 5 |
Reddy and colleagues examined the ability of a stair climbing activity to predict surgical outcomes in 264 patients ≥ 19 years of age who were undergoing elective abdominal surgery. They reported that stair climbing predicted morbidity better than the NSQIP surgical risk calculator (AUC = 0.81 vs. 0.62,
p < 0.0001). However, the stair climbing task caused physiologic stress, especially in those patients who were slower climbers, causing a 16.6 and 21.8% increase in heart rate and mean arterial pressure, respectively, compared to 7.4 and 7.9% for faster climbers (Reddy et al.
2016). Although this study demonstrated excellent power of stair climbing time as a tool to predict postoperative outcomes, it is an even more intensive test than the Timed Up and Go test, with higher risk (including falls and cardiac events with physiologic stress). Also, the time and space constraints, as well as personnel requirements, clearly limit the feasibility of the stair climbing test in clinical practice.
What is noteworthy is that our models were not different from the NSQIP surgical risk calculator in predicting postoperative hospital LOS. Indeed, the biggest factor that determines postoperative hospital LOS is the type of surgery; for example, patients who undergo laparoscopic cholecystectomy will have a much shorter LOS than patients who undergo open hemicolectomy. The NSQIP surgical risk calculator was developed using 1557 unique CPT codes and allows input of a single procedure CPT code. Even though our models did not include the surgical risk in the model (e.g., MAT-sf only model) or reflect the risks fully (only two degrees of surgical risks [low vs. intermediate-to-high]), the predictions for LOS using the simple MAT-sf model were remarkably similar to the universal calculator. Adequately incorporating procedure complexity and risk into future studies that use the simple MAT-sf could solidify its efficacy in forecasting surgical outcomes.
The limitations of the study include its single-center design and use of a majority white patient cohort, features of the study design that compromise the external validity of the findings. We were unable to include the full list of measures included in the NSQIP calculator due to our small sample size and lack of variability in some measures such as cancer and chronic heart failure, among others. Since this issue could have accounted for the similarities observed between the universal calculator and the MAT-sf models, the results must be interpreted with caution. Nonetheless, we expect that our estimation of postoperative risks and LOS with the MAT-sf tool may have been better if we had a larger sample with increased variability. Although we used the LOO cross-validation method, our findings still need to be tested in a separate cohort. Lastly, the models were developed from relatively small number of older patients (n = 197), which may not accurately reflect surgical risk.
Conclusion
The MAT-sf can predict postoperative outcomes (complications, LOS, and nursing home placement) in older surgical patients with predictive performance comparable to the NSQIP surgical risk calculator. Given the simplicity and noninvasive nature of the MAT-sf tool, it can be easily adopted into clinical practice for preoperative risk assessment. Future studies in a larger cohort of patients undergoing a single type of surgery are warranted to validate this new method to predict postoperative outcomes.