Pancreatic cancer is an aggressive malignancy that disproportionately affects older adults, with 70% of new cases diagnosed for individuals older than 65 years.1 Surgical resection currently is the only potentially curative treatment option, providing a 5-year survival rate ranging from 15 to 25%.2,3,4 However, pancreaticoduodenectomy (PD) is associated with high complication rates, including a 30-days morbidity rate of 40–50%5,6,7 and a mortality rate of 1–3%.7,8,9

Outcomes after PD estimated by risk prediction tools rely on patient- and tumor-related factors, some of which cannot be assessed preoperatively.10 Although recent studies have devised models based on parameters measured preoperatively, they require high-quality imaging and interpretation and have not been validated with large samples of patients in the general population.11,12,13 Other predictive tools such as the Physiological and Operative Scoring System for enUmeration of Morbidity and mortality (POSSUM)14 and the Estimation of Physiologic Ability and Surgical Stress Score (E-PASS)15 are complex and have demonstrated inconsistent results in their ability to predict morbidity and mortality for patients undergoing PD.16,17,18,19 Although sarcopenia has been used, it usually is combined with other measures of decreased physiologic reserve, and its use requires complex evaluation of imaging parameters.20 A multifactorial measure of overall physiologic reserve, such as frailty, may serve as a more accurate predictor of outcomes after this high-risk procedure.

Frailty has been defined as a clinical syndrome that involves the progressive loss of physical and mental function with or without the coexistence of disease.21 With decreased physiologic reserves, frail individuals have a decreased ability to maintain homeostasis and an increased vulnerability to acute stressors, including surgery.21,22,23 Findings have shown frailty to be an independent predictor of postoperative complications, hospital length of stay, discharge to a skilled- or assisted-living facility, and mortality.23,24,25

Although frailty is highly prevalent among the elderly, chronological age alone is a poor predictor of adverse outcomes after acute stress.26,27,28 The Canadian Study of Health and Aging has created a standardized frailty index (CSHA-FI) based on a cumulative deficit model.29,30 This model defines frailty as the cumulative effect of individual deficits based on clinical signs, symptoms, disease states, and disabilities, which provide a more accurate assessment of aging than chronological age.25

The CSHA-FI has been mapped to 11 variables contained in the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) database to create a modified frailty index (mFI).31,32,33 The NSQIP mFI accurately predicts postoperative morbidity and mortality after vascular surgery or colectomy, as well as outcomes in other patient populations.31,34,35 However, no studies have specifically examined the mFI in the context of PD. Therefore, we hypothesized that the NSQIP mFI can predict postoperative outcomes after PD. A measurement of frailty that can aid in preoperative risk stratification could facilitate shared decision making, improve patient selection, and help to optimize patients preoperatively so as to reduce surgical complications.


Patient Selection

All patients who underwent PD (Common Procedural Terminology code 48150—Classic Whipple procedure, and CPT code 48153—pylorus-sparing Whipple procedure) were identified in the 2005–2012 NSQIP Participant Use File (PUF).36 The PUF contains pre-, intra-, and postoperative data collected by specially trained surgical clinical reviewers from each NSQIP-participating institution. Patients undergoing emergency procedures, those classified as American Society of Anesthesiology (ASA) 5, and those with a diagnosis of preoperative sepsis were excluded from the analysis. Institutional review board approval was obtained for the study.

Definition of the mFI

The preoperative variables within the NSQIP dataset were reviewed to determine an mFI for each patient. The mFI, described previously by Velanovich et al.,33 includes the following 11 items in the NSQIP: diabetes; functional status (not independent); chronic obstructive pulmonary disease or pneumonia; congestive heart failure; history of myocardial infarction; hypertension requiring medication; peripheral vascular disease or rest pain; impaired sensorium; history of either transient ischemic attack or cerebrovascular accident; history of cerebrovascular accident with neurologic deficit; and prior percutaneous coronary intervention, previous coronary surgery, or history of angina. Each item was allocated the same weight (1 point) in the calculation of the index. The mFI then was calculated as the proportion of the total 11 items used in the study that were expressed in a given individual patient (total points as the sum of all the items divided by 11). Although the mFI is not meant to be a dichotomous variable, a cutoff of 0.27 was used based on previous data demonstrating the overlap in deficit accumulation between persons who are “robust” and those who are “frail” to be approximately 0.25.32,37,38

Definition of Morbidity and Mortality

Major complications were defined as Clavien–Dindo class 3 or 4 complications (life threatening or requiring intensive care management such as unplanned intubation, failure to wean from ventilator in >48 h, acute renal insufficiency requiring dialysis, new-onset neurologic deficit or coma, myocardial infarction, cardiac arrest, pulmonary embolism, graft failure, organ space infection). Minor complications were defined as Clavien–Dindo class 1 or 2 complications (surgical-site infection, pneumonia, deep vein thrombosis, urinary tract infection, peripheral nerve injury, postoperative bleeding requiring transfusion). Mortality was defined as death within 30 days after surgery. Failure to rescue was defined as previously described.39,40

Statistical Analysis

The sample means and standard deviations were computed for the continuous descriptive characteristics, and the count and proportions were calculated for the discrete descriptive characteristics by frailty groups (mFI ≥0.27 vs <0.27). Demographic characteristics between patients with an mFI of 0.27 or higher and those with an mFI lower than 0.27 were compared using Chi square statistics for categorical variables and two-sample t tests for continuous variables. Univariate analysis using logistic regression identified clinically significant factors associated with the development of any complication, with major complications and mortality expressed as odds ratios (ORs). Those preoperative variables shown to be clinically relevant were then used to construct multiple variable models for the aforementioned outcomes. The covariates included age, sex, ASA, albumin (≥3 vs <3), and obesity (body mass index [BMI] ≥30 vs <30). All analyses were performed using SAS version 9.4 (SAS Inc., Cary, NC, USA).


Of the 13,851 patients who underwent PD, 3865 had at least 1 of the 11 frailty items missing and were therefore excluded from the analysis. The study sample comprised 9986 patients, and 6.4% (n = 637) of these patients had a high mFI (≥0.27). The mean age of the patients was 64.1 ± 12.4 years, and 48.8% (n = 4865) were women. The patients with a high mFI tended to be older and male, to have a higher BMI and a lower serum albumin level, to be less likely preoperatively to have fully independent functional status and more likely to have an ASA classification of 3 or 4, to be diabetic, to have major cardiovascular and neurologic comorbidities, and to have a major complication or die within 30 days after surgery (Table 1).

Table 1 Patient characteristics

Most of the patients (n = 3829) had an mFI of 0. Of these patients, 33.7% (n = 1292) experienced a complication, 25.7% (n = 983) experienced a major complication, and 1.3% (n = 51) died within 30 days after surgery. As mFI increased, the total number of patients within an mFI category declined. Increasing mFI was associated with a higher incidence of any complication (p < 0.001), a major complication (p < 0.001), or 30-days mortality (p < 0.001) (Table 2; Fig. 1).

Table 2 Trends of mFI and outcomes
Fig. 1
figure 1

Percent of patients with postoperative 30-day morbidity and morality based on modified frailty index

Of 9349 patients with an mFI of 0.27 or lower, 8.3% (n = 772) experienced a minor complication, 27.7% (n = 2592) experienced a major complication, and 2.7% (n = 249) died within 30 days after surgery. Of 637 patients with an mFI higher than 0.27, 7.7% (n = 49) experienced a minor complication, 40.8% (n = 260) experienced a major complication, and 6.3% (n = 40) died within 30 days after surgery. Failure to rescue occurred for 10% (n = 289) of the patients and was significantly associated with frailty and low mFI (9.5% vs a high mFI of 15.1%; p = 0.007).

In the univariate analysis, the development of any complication was predicted by increasing age, male sex, obesity (BMI > 30 kg/m2), low albumin, more than 10% weight loss, longer operative time, longer hospital length of stay, higher ASA, loss of functional status, transfer from a place other than home, and an mFI of 0.27 or higher. Increasing age, male sex, obesity, longer operative time, longer hospital length of stay, higher ASA (3 or 4), loss of functional status, low albumin, more than 10% weight loss, and an mFI of 0.27 or higher predicted the development of a major complication (Table 3). Increasing age, obesity, longer operative time, longer hospital length of stay, higher ASA status, loss of functional status, and high mFI predicted 30-days mortality (Table 3).

Table 3 Univariate analysis of major complications and 30-days mortality

Older patients (age >75 years) represented 19.6% (n = 1947) of the patients undergoing PD. In this subgroup, high mFI was significantly more common (9.7%, n = 208) than among younger patients (5.5%, n = 427), p < 0.001.

In the multivariate analysis, age, male sex, higher ASA, albumin level lower than 3, obesity, and an mFI of 0.27 or higher remained independent preoperative predictors of any complication, major postoperative morbidity, and 30-days mortality (Table 4). The frailty index was a strong predictor of major morbidity (OR 1.54; 95% confidence interval [CI] 1.29–1.85; p < 0.001) and 30-days mortality (OR 1.54; 95% CI 1.05–2.25; p = 0.027).

Table 4 Multivariable analysis of major complications and 30-days mortality


In the current study, high mFI was associated with worse outcomes after PD. Frailty is considered to be a state of decreased physiologic reserves arising from cumulative deficits in multiple homeostatic systems that results in greater susceptibility and less resilience to physiologic stressors.41 Importantly, although frailty traditionally has been described in the form of physical weakness as a function of aging, it is well known that besides chronological age, several other factors contribute to physiologic aging and determine functional reserve and response to stress.20,21,23,42 Surgeons often rely too much a patient’s age and not enough on an objective measure of physiologic reserve, whereas patients may tend to overestimate their ability to tolerate major surgical stresses, potentially leading to unrealistic expectations of their outcome and recovery.43

Both pancreatic cancer and PD are major physiologic stressors, so the concept of frailty is particularly important in this population of cancer patients. Accordingly, several authors have advocated the routine incorporation of frailty assessment for older cancer patients, including comprehensive geriatric assessment (CGA).44,45,46

The CGA, although very specific, is complex and very time consuming, making it less than ideal as a screening tool in the everyday clinical setting.47,48,49,50 Studies have shown that the CGA is infrequently used by surgeons in preoperative frailty assessment of cancer patients.51 Therefore, abbreviated indices that efficiently assess frailty are essential for risk stratification of patients and for helping to determine eligibility for surgery. The mFI is a simple frailty assessment tool that uses easily available historical variables that can be reliably and consistently collected in the preoperative setting.

In the current study, 25.7% of the patients not considered frail (mFI = 0) had major complications, and 1.3% of these patients had postoperative mortality. This is consistent with the published data on outcomes after PD in several large series.2,3,5 We found that the mFI correlates well with the development of postoperative complications and 30-days mortality after PD. As the mFI increases, so does the percentage of patients experiencing an adverse outcome. Importantly, only 6.4% of the patients in our study cohort had a high mFI (mFI > 0.27), which likely represents a selection bias toward healthier or less frail patients for PD. The high-mFI group had significant morbidity (41%) and mortality (5.6%).

Whereas age, BMI, ASA status, and nutritional status (low albumin level) were independent determinants of adverse outcomes, mFI remained a predictor of postoperative outcomes in the multivariable analysis. The results from the current study mirror reports from prior studies using the mFI.31,33,52,53,54,55 However, to our knowledge, this is the first study to use a large national database to validate the ability of the mFI to predict development of postoperative adverse events after PD.

Our study had several important implications. First, the variables used to calculate the mFI are physical, readily available from the history, and easily reproducible. The identification of frailty allows for an objective assessment of the potential for the development of complications and recovery from them. This can in turn help in better informing the shared decision-making process, which is essential for preoperative counseling and patient selection.

Second, frailty, instead of being an irreversible state, occurs as a continuum within a spectrum, with the possibility of transitions between higher and lower states of frailty.56 Therefore, identification of frail patients and implementation of prehabilitation strategies to treat modifiable factors that determine frailty may prevent further deterioration in physical and functional impairment and potentially improve postoperative outcomes.57,58,59 These strategies include measures such as referral to specialty geriatrics clinics for an in-depth assessment and incorporation of physical and cognitive exercise, social support, and nutrition for patients in earlier stages of frailty.60 Although several individual variables within the mFI are likely not modifiable during the preoperative period, a multidisciplinary team approach can help to mitigate the risks associated with specific preoperative risk factors.

Our study had a several limitations. First, we chose 11 of the 70 variables described in the original CHSA study, which mapped to the preoperative variables used in the NSQIP dataset. Arguably, these variables may not fully represent the entire spectrum of frailty parameters. A previous analysis indicates that one need not use the same items or even the same number of items to estimate the proportion of patients that represent a certain value within the index.61 Although the inclusion of more variables does increase the precision of the estimate, the differences in the mean mFI values between similar distributions of phenotypic categories of frailty are not statistically significant.37

Second, variables used in the mFI are based on subjective assessment. The NSQIP database also was not specifically designed to assess frailty. Therefore, its use for preoperative patient assessment needs to be validated in prospective studies.

Finally, given its retrospective nature, the study was limited by an inherent selection bias, with inclusion of lower-risk patients who were de facto operative candidates. Nonetheless, within the study sample, the mFI was a strong, objective, and reproducible indicator of postoperative outcomes.


In conclusion, the mFI is a strong predictor of postoperative complications and mortality after PD. Its use in the preoperative setting can help to risk-stratify patients, identify subgroups at increased risk for the development of adverse events, and subsequently institute prehabilitative measures to potentially optimize modifiable factors and minimize complications.