Association between frailty status and complications in patients undergoing surgical excision of malignant esophageal neoplasms

Introduction

Esophageal cancer is the seventh most common type of newly diagnosed cancer worldwide, with over 600,000 cases in 2020 [1]. While squamous-cell carcinoma remains the most predominant esophageal cancer worldwide, because of the high level of tobacco and alcohol abuse [2,3], rates of esophageal adenocarcinoma are rapidly increasing in the United States in parallel with the increasing rates of obesity [3-5]. Overall, esophageal cancer accounts for 3.1% of all new cancers and 5.5% of all cancer deaths, representing 1 of every 18 cancer deaths annually [1]. This high mortality can be explained by the fact that esophageal cancer usually presents clinically at later stages, with only 18% of cancers still confined to the primary site at diagnosis [6]. Recent advances in curative intent therapy include the adoption of trimodality therapy (carboplatin, paclitaxel and radiotherapy, followed by surgery) for early-stage disease, as examined in the CROSS trial, and neoadjuvant treatment with nivolumab after surgical resection, as seen in the ChckMate577 trial, which have demonstrated improved survival rates [7,8]. Additional advances in chemotherapy for esophageal cancer include the FLOT regimen, which consists of docetaxel, oxaliplatin, leucovorin, and 5-fluorouracil. The ESPOPEC trial investigated the efficacy of neoadjuvant chemoradiation with the CROSS protocol [9]. Perioperative chemotherapy has resulted in better 5-year survival compared with surgery alone. Despite these advances, the prognosis remains poor, with an overall 5-year post-esophagectomy survival rate of around 20% (range 5-47%), with the longest survival seen among those diagnosed at earlier stages [4,5,10].

Frailty is defined as a clinically recognizable state of increased vulnerability resulting from age-associated physiological decline, leading to homeostatic imbalance [11]. It has been operationally defined by Fried et al as meeting 3 of the following 5 phenotypic criteria: low grip strength, low energy, slowed walking speed, low physical activity, and/or unintentional weight loss [11-13]. Frail patients continue to be at higher risk of poor health outcomes, including falls, disability, perioperative complications, readmissions and death. Frailty has been shown to be a more accurate predictor of outcomes than age and other patient demographics, especially in patients undergoing complex surgeries [14], as well as being a robust predictor of outcomes in patients who undergo cardiac surgery or cranial neurosurgery [15-18].

As contemporary literature suggests that frailty may be a better predictor of complications than age and other patient demographics, our study aimed to investigate the role of frailty in predicting postoperative outcomes in patients undergoing esophageal excision for malignant esophageal carcinoma. We hypothesized that frail patients would have higher rates of postoperative complications, higher rates of medical complications, and a more complicated hospital course overall, compared to non-frail patients. In addition, using predictive analytics and modeling, we aimed to demonstrate an improved performance of statistical models in predicting patient discharge disposition when frailty status was taken into account. Such an understanding would aid physicians in the appropriate management of patients with newly diagnosed esophageal cancer.

Materials and methods

Patient sample

Between 2016 and 2017, we identified a total of 78,689 inpatient admissions with ICD-10 codes for malignant esophageal neoplasm (ICD-10: C15.x). Within this cohort, appropriate coding (ICD-10: 0DB5xZZ) was used to identify 1159 patients (1.5%) who underwent esophageal resection for a malignant esophageal neoplasm (Fig. 1). Frail patients were identified using the Johns Hopkins Adjusted Clinical Groups (JHACG) frailty-defining diagnosis indicator, which uses 10 categories of ICD-10 codes (malnutrition, dementia, vision impairment, decubitus ulcer, urine control, weight loss, fecal control, social support, difficulty walking, and history of a fall) to predict a patient’s frailty status [19]. Several studies have confirmed the clinical validity of the JHACG frailty-defining diagnosis indicator [19-22].

Figure 1 Patient selection flowchart for this study

The cohort was then subdivided into frail (n=289) and propensity score-matched non-frail (n=281) patients. Nearest-neighbor propensity score matching for age, sex, insurance type, median income by ZIP code, and NRD discharge weighting was performed using the R “MatchIt” algorithm [23]. In this technique, parametric models are chosen based on the minimum “distance” parameter, which is determined through logistic regression models that minimize the propensity score with no replacement. MatchIt improves parametric statistical models and reduces model dependence by preprocessing data with semi-parametric and non-parametric matching methods. Model balance, defined as the similarity of empirical covariate distributions between the 2 groups undergoing propensity matching, is analyzed and the model with the best balance is selected to ensure best model fit (Fig. 2). Complications queried for analysis in this study included postoperative infections, acute posthemorrhagic anemia, dysphagia, mortality, readmission rates, urinary tract infection (UTI), pulmonary embolism (PE), deep vein thrombosis (DVT), inpatient length of stay (LOS), costs, and discharge disposition. Nonroutine discharges were defined as discharges to places other than home (e.g., skilled nursing facility [SNF], home health care [HHC], short-term care facility, etc.).

Figure 2 Distribution of propensity scores following matching. Frail patients are shown as Matched Treatment Units and propensity matched non-frail patients are shown as the Matched Control Units. The Unmatched Control Units represent non-frail patients who were not chosen by the propensity matching algorithm. The comparable distribution of patients in both matched treatment and control units implies that excellent propensity score matching was achieved

Results

Demographics

The average age of the frail cohort was 65.0±9.3 years and 19.8% were female. The average age of the non-frail cohort was 64.9±9.5 years and 17.5% were female. As the 2 cohorts were propensity score matched, comparisons of age, sex, insurance type and median income quartile by ZIP code between the 2 cohorts found no statistically significant differences. No significant differences were found between hospital size (P=0.46) and teaching status (P=0.42) in the 2 cohorts. However, significant differences were found between frail and non-frail patients when discharge dispositions were compared (P=0.0020) (Table 1).

Table 1 Demographics of frail and non-frail patients

Predictive models and ROC analysis

Two sets of logistic regression models were developed: the first used age alone as the primary predictor, while the second used patient frailty status and age. These models were tested to evaluate their prediction of patient discharge status, which was found to be significant. ROCs were plotted for both the logistic regression models (Fig. 3). As seen in Fig. 3, the logistic regression models using frailty and age as the primary predictors outperformed the model using age alone.

Figure 3 ROC plot for prediction of nonroutine discharge status. The black ROC represents the logistic model using age alone as the primary predictor, and the red ROC represents the logistic model using frailty status and age as the primary predictors. A noticeable increase in predictive power occurs when frailty is jointly considered for prediction of discharge status

ROC, receiver operating characteristic curve; AUC area under the curve

Postoperative complication rates

During the primary admission, the costs associated with inpatient stay were significantly higher for frail patients compared to non-frail patients (frail: $81,994.34±$72,418.47 vs. non-frail: $47,269.95±$31,512.09, P<0.001). Similarly, frail patients had a significantly longer inpatient LOS compared to non-frail patients (frail: 20.0±15.7 days vs. non-frail: 11.7±9.2 days, P<0.001).

Complication rates were found to be significantly higher in frail patients compared to non-frail patients. More specifically, frail patients had higher rates of postoperative infection (P=0.0020), acute posthemorrhagic anemia (P<0.001), pneumonia (P=0.0079), UTI (P=0.029), DVT (P<0.001), and mortality (P=0.014) (Table 2). However, rates of PE (P=0.10), dysphagia (P=0.14), and readmission (P=0.72) did not differ significantly between the 2 cohorts.

Table 2 Management and complications in frail and non-frail patients

Discussion

In this 2-year retrospective cohort analysis, we evaluated the impact of frailty on postoperative complications in patients undergoing esophagectomy for a malignant esophageal neoplasm. By comparing a cohort of frail geriatric patients to a propensity-matched cohort of non-frail geriatric patients, we found that frail patients were associated with higher rates of medical complications immediately after surgery, including higher rates of postoperative infections, acute posthemorrhagic anemia, pneumonia, UTI, DVT, and mortality. However, no significant differences were found between frail and non-frail patients with regard to PEs, dysphagia, or readmission rates. The higher complication rates probably explain the longer LOS and higher hospital costs associated with frailty. Predictive models evaluated by ROC curves also demonstrated improved model performance with the addition of frailty when predicting discharge disposition, allowing for more accurate patient risk triage and hospital resource utilization.

Over the past decade, frailty has been more popularly utilized for assessing rates of morbidity, mortality and postoperative outcomes in the field of gastroenterology, with evidence pointing towards the importance of assessing frailty in preoperative risk-stratifying assessments [24-26]. Hodari et al previously elucidated the relationship between patient frailty and morbidity and mortality in patients undergoing esophagectomy procedures [27]. Their focus was placed on evaluating a 5-point modified frailty index (mFI) in patients who undergo esophagectomy procedures, and their study demonstrated that frailty scores of 0, 1, 2, 3, 4 and 5 had associated morbidity rates of 17.9%, 25.1%, 31.4%, 34.4%, 44.4% and 61.5%, respectively [27]. Esophagectomy has historically been associated with higher levels of morbidity and mortality, and as a result surgical management has shifted towards multidisciplinary care to improve short-term outcomes. Perioperative treatment protocols, such as enhanced recovery after surgery (ERAS) programs, have been implemented in various surgical disciplines and have been shown to decrease LOS, postoperative complications, and postoperative morbidity. However, the use of these protocols in esophageal surgery is still limited. Components of the ERAS protocol for esophageal cancer included proper preoperative nutritional support, using minimally invasive surgical techniques, and avoiding routine postoperative nasogastric tube insertion, abdominal draining, and Foley catheter insertion.

Our study supports the finding of Hodari et al, that frail patients have overall higher rates of morbidity compared to non-frail patients. The JHACG frailty index is a binary classification system formulated to capture multiple domains of frailty (i.e., functional, cognitive, psychological and socioeconomic), and it has been shown to identify patients who have limitations in their activities of daily living [19,28]. Whereas the mFI is useful in establishing degrees of morbidities as frailty becomes more severe, the JHACG frailty index is a superior tool when it comes to comparing frail patients to non-frail counterparts. For instance, a systematic review by Ornaghi et al of patients undergoing radical cystectomy (RC) for bladder cancer found that, though the mFI tends to be the most commonly used frail index in that field, the JHACG index was the best predictor of early postoperative RC-related adverse outcomes [29]. In addition, our study implemented propensity score matching and robust predictive models to improve current paradigms of risk stratification and prediction of postoperative complications. Our data reinforce previous evidence in the field of gastroenterology and gastrointestinal surgery, though utilizing a better predictor of adverse outcomes, and demonstrate the predictive validity of frailty in patients with gastrointestinal malignancies.

Historically, gastrointestinal surgical risk analysis has relied on patient demographics alone, primarily placing a large onus on age as a predictor for complications [30,31]. Through the creation of robust predictive models, it was possible to quantify the predictive value of patient frailty status in predicting patient discharge disposition. Specifically, the addition of frailty to our predictive models using age to predict nonroutine patient discharges increased the AUC from 0.601 to 0.652. In general, an AUC of 0.50 demonstrates a random guess and AUC values greater than 0.70 are defined as strongly predictive [32]. While neither model quite achieves 0.70, the addition of frailty increased the AUC and the predictive capabilities of our model significantly. Previous studies have shown that the addition of frailty to the American Society of Anesthesiologists, Lee’s revised cardiac risk, and Eagle indices may improve the overall performance of predictive models in a heterogeneous cohort of elderly patients undergoing surgery [14,24]. However, this association has not yet been established in patients undergoing esophageal surgery for malignant neoplasms, and our study is the first to show that frailty improves the prediction of outcomes in this specific population of patients. Future work should focus on identifying additional variables that may be used to supplement models incorporating frailty, to boost the AUC past 0.70.

The findings of this study also provide an opportunity to improve the inpatient management of patients who undergo surgery for esophageal neoplasms. Using increasingly robust and predictive models, it becomes possible to predict potential patient outcomes with high accuracy, allowing for better resource utilization. Previous studies using machine learning models and predictive modeling have successfully identified factors that are correlated with discharge turnaround time and status, allowing for the prediction of discharge status and shorter inpatient LOS [33,34]. This may also be possible for patients admitted for surgical excision of esophageal neoplasms. For example, patients identified as high-risk for nonroutine discharge using our models may be prioritized for consultation with social services, such as physical and occupational therapy, and coordination of placement in SNF, HHC, or short-term hospital facilities. This may potentially save time and resources and expedite transfer to facilities that can provide patients with appropriate care. Furthermore, our study identified specific complications that are more likely to affect frail patients, highlighting increasingly common complications in frail patients, and allowing for more tailored inpatient management.

This study had several limitations. First, it was subject to the limitations of retrospective cohort analyses. Namely, the quality of analysis is dependent on the depth and accuracy of patient encounters documented in the NRD, and Berkson’s bias is present when working with inpatient databases. Furthermore, this study is limited by being a retrospective analysis during a narrow time interval (2016 and 2017 only). However, the dates were chosen based on the implementation of mandatory ICD-10 coding in late 2015, which allowed for more detailed codes to be drawn for analysis. In addition, because the database relies on ICD-10 coding, it is not possible to query TNM staging or type of esophagectomy procedure. Lastly, the NRD allows for retrospective readmission analysis within one calendar year (January to December). Therefore, additional readmissions not occurring within the same calendar year are not captured and cannot be analyzed using the NRD.

To conclude, our study suggests that patient frailty status strongly correlates with rates of medical complications, costs, LOS, and discharge disposition in patients who undergo esophagectomy for esophageal cancer. When frailty was incorporated into logistic models, it improved the prediction of patient discharge disposition compared to patient age alone. Overall, frailty represents a robust predictor of patient outcomes, and a better understanding of frailty may aid in medical and surgical decision making when considering patients with esophageal cancer. Further research, including a multicenter analysis with a large number of participants, will be necessary to fully understand the role of frailty in the outcomes of patients with malignant esophageal neoplasms.