Introduction
Post-operative complications inflict significant morbidity and mortality on patients, with considerable cost to the hospital system. Thus, it is important for both surgeons and patients to understand the risks prior to surgery, to allow for a better-informed, shared decision-making process.
The National Surgical Quality Improvement Program was developed in 1994 by the Department of Veterans Affairs to prospectively collect pre-surgical, surgical and 30-day outcome data on all major surgical procedures. On the basis of the data collected, a risk-adjusted model for 30-day morbidity and mortality was developed for all non-cardiac surgical procedures.Reference Khuri, Daley, Henderson, Hur, Demakis and Aust 1 In light of the success of the National Surgical Quality Improvement Program, the American College of Surgeons adapted this risk model, leading to the inception of the American College of Surgeons' National Surgical Quality Improvement Program (‘ACS-NSQIP’), in 2004.Reference Khuri, Daley, Henderson, Hur, Demakis and Aust 1 The American College of Surgeons' National Surgical Quality Improvement Program aimed to improve surgical outcomes by providing surgeons with highly specific, risk-adjusted data, and allow for quality improvement initiatives to improve patient outcomes.Reference Stachler, Yaremchuk and Ritz 2
Head and neck cancer is diagnosed in more than 550 000 patients annually worldwide.Reference Ferlay, Shin, Bray, Forman, Mathers and Parkin 3 In Australia, the most common subsite of head and neck cancer is the oral cavity. The incidence and five-year survival rates in the Australian population are comparable to those in the American patient population, with the latter population being utilised in the formation of the American College of Surgeons' National Surgical Quality Improvement Program risk calculator.Reference Chaturvedi, Anderson, Lortet-Tieulent, Curado, Ferlay and Franceschi 4 , Reference Carvalho, Nishimoto, Califano and Kowalski 5 Current treatment of tongue malignancies involves glossectomy and neck dissection, which are both invasive with a significant disease burden.
The American College of Surgeons' National Surgical Quality Improvement Program risk calculator has been used in the improvement of patient outcomes following surgery in the USA. However, the risk calculator has not yet been used in the Australian population.Reference Stachler, Yaremchuk and Ritz 2 , Reference Bilimoria, Liu, Paruch, Zhou, Kmiecik and Ko 6 , Reference Chen, Roman, Sosa and Judson 7 This study evaluated the risk calculator for predicting morbidity and mortality outcomes of patients undergoing glossectomy in South Australia.
Materials and methods
A retrospective chart review was performed on patients with tongue malignancy treated with partial or total glossectomy, with or without adjuvant chemoradiotherapy, at Flinders Medical Centre or Royal Adelaide Hospital, between 2005 and 2015. Inclusion criteria included patients aged over 18 years with a histological diagnosis of tongue squamous cell carcinoma.
Surgical access was gained via several approaches depending on the location and extent of the tumour; this included transoral, lingual release, or via lip-splitting incision with mandibulotomy. Surgical defects were closed primarily or reconstructed with radial forearm free flaps, anterolateral free flaps, fibular free flaps, pectoralis major flaps or buccinators flaps. Patients who received radiotherapy prior to surgery were considered to have significantly higher risk for post-operative complications.
The Human Research Ethics Committee at Flinders Medical Centre and Royal Adelaide Hospital granted ethics approval. The Operating Room Management Information System (‘ORMIS’), utilised in all South Australian public hospitals, was employed in the identification of all patients in this study. This system enables real-time management of operation theatre schedules, and storage of surgical records and operation notes. A search query was conducted, using the term ‘glossectomy’, of procedures conducted between 2005 and 2015. Once all cases were identified, operation type was manually translated into Current Procedural Terminology codes for subsequent use of the American College of Surgeons' National Surgical Quality Improvement Program risk calculator.
The Current Procedural Terminology codes included: 41120 – glossectomy, less than one-half; 41130 – glossectomy, hemiglossectomy; 41135 – glossectomy, partial with unilateral radical neck dissection; 41140 – glossectomy, complete or total, with or without tracheostomy, without radical neck dissection; 41145 – glossectomy, complete or total, with or without tracheostomy, with radical neck dissection; 41150 – glossectomy, composite procedure, with floor of mouth and mandibular resection, without radical neck dissection; 41153 – glossectomy, composite procedure, with floor of mouth resection, with suprahyoid neck dissection; and 41155 – glossectomy, composite procedure, with floor of mouth and mandibular resection, and radical neck dissection (commando type).
Medical records for all identified patients were reviewed to ensure patient demographic, clinical, radiological, pathological and treatment data were correct and up to date. Patient demographics collected included age, gender, operative details, malignancy stage, previous radiotherapy treatment, height, weight, smoking and alcohol use.
The primary aim of this study was to compare the American College of Surgeons' National Surgical Quality Improvement Program predicted 30-day outcome with actual outcomes. Outcomes of interest were based on the risk calculator and included: pneumonia, heart complications, wound infections, urinary tract infections, venous thromboembolism, renal failure, return to the operating theatre, death or discharge to care. Serious complications according to the American College of Surgeons' National Surgical Quality Improvement Program risk calculator were: cardiac arrest, myocardial infarction, pneumonia, progressive renal insufficiency, acute renal failure, pulmonary embolism, deep vein thrombosis, return to the operating theatre, deep incisional surgical site infection, organ space surgical site infection, systemic sepsis, unplanned intubation, urinary tract infection and wound disruption. The American College of Surgeons' National Surgical Quality Improvement Program has a rigorous and highly standardised set of definitions for each of these post-operative occurrences to enable its use across different types of surgery. We did not have access to these specific definitions, or hospital stay and funding models, for this study.
Statistical analysis
Statistical analysis was performed utilising SAS 9.3 software (SAS Institute, Cary, North Carolina, USA). The primary end points of this study were predicted 30-day outcomes and actual outcomes. Descriptive statistics were used to analyse demographics, glossectomy type and tumour–node–metastasis (TNM) classification.
Actual length of stay was initially compared with predicted length of stay based on the American College of Surgeons' National Surgical Quality Improvement Program risk calculator using a scatter plot. A Bland–Altman plot was then used to assess the agreement between actual and predicted length of stay. Length of stay index is a nationally developed measure of hospital efficiency, calculated by dividing actual length of stay with predicted length of stay.Reference Russell, Hakendorf and Thompson 8 As average length of stay index in Australian public hospitals is 5 per cent, this was used as the clinically significant cut-off for high length of stay index. 9
For mortality and serious complications, a Hosmer–Lemeshow graphical representation was presented for predicted versus observed rates, with limit lines ± 25 per cent on either side of the diagonal which represents perfect agreement (observed = predicted). This cut-off was based on an article by Bilimoria et al.Reference Bilimoria, Liu, Paruch, Zhou, Kmiecik and Ko 6
Brier scores were calculated, defined as the average squared difference between patients' predicted probabilities and observed outcomes. These scores are effective in measuring the accuracy of probabilistic predictions.Reference Brier 10 Brier scores have a value between 0 and 1, with a score of 0 indicating perfect prediction and a score of 0.25 representing a non-informative model.Reference Bilimoria, Liu, Paruch, Zhou, Kmiecik and Ko 6 , Reference Steyerberg, Vickers, Cook, Gerds, Gonen and Obuchowski 11
Results
Patient demographics
A total of 127 patients who underwent glossectomy were identified, of which 120 were included in the study. Seven were excluded because of the lack of documented height in their medical records. Mean age at diagnosis was 61.3 years (standard deviation (SD) = 16.2). There were 79 males (66 per cent) and 41 females (34 per cent). Seven patients (6 per cent) identified as Aboriginal Australians. Ten patients (8 per cent) underwent total glossectomy, 55 patients (46 per cent) had partial glossectomy and 55 patients (46 per cent) underwent hemiglossectomy. Ninety-one patients underwent primary reconstructions, consisting of anterolateral thigh (n = 4; 4 per cent), fibular (n = 5; 5 per cent), pectoralis major (n = 9; 10 per cent), radial forearm (n = 41; 45 per cent), ulnar (n = 3; 3 per cent), rectus abdominis (n = 5; 5 per cent) and buccinator (n = 24; 26 per cent) flaps. Nine patients (8 per cent) had previously received radiotherapy. Tumour–node–metastasis classification and neck dissection details are shown in Table I.
Table I TNM classification and neck dissection details

TNM = Tumour–node–metastasis
Length of stay
The scatter plot demonstrated a substantially longer actual length of stay compared with the predicted length of stay (Figure 1). Median length of stay was 15 days (range, 1–197 days), with median predicted length of stay of 5 days (range, 1–12 days).

Fig. 1 Scatter plot of observed versus predicted rates of length of stay.
The Bland–Altman plot demonstrated that longer actual length of stay was associated with poorer predicted values (Figure 2). The length of stay index was right-skewed, with a median value of 2.9 (actual length of stay being 2.9 times higher than predicted length of stay). As a 5 per cent difference was used as a clinically significant cut-off, eight patients (6.7 per cent) had an acceptable length of stay index of less than 1.05.

Fig. 2 Bland–Altman plot, demonstrating the agreement between actual and predicted length of stay.
Flap repairs were found to be associated with increased length of stay: median length of stay was 6 days (range, 1–29 days) in patients who did not undergo flap surgery, compared with 19 days (range, 6–197 days) in patients who received flap surgery. The median length of stay index for the free flap group was 3.43 and 1.71 in the non-free flap group. There was a statistically significant difference between actual length of stay and predicted length of stay in both free flap and non-free flap groups (Wilcoxon signed rank test, p < 0.0001 and p < 0.0001). Seven per cent of patients had an acceptable length of stay index in the free flap group compared to 14 per cent in the non-free flap group (cut-off = 1.05).
The mean age of patients discharged to a nursing home was 74 years (SD = 12.56), with a median length of stay of 23 days (range, 13–197 days). In comparison, patients discharged home had a mean age of 65 years (SD = 16.78), and a median length of stay of 14 days (range, 1–54 days).
Complications
Complications occurred in 50 per cent of patients, of which wound infection was the most common (Table II). Hosmer–Lemeshow graphical representations of predicted versus observed rates were plotted for mortality and serious complications, and those for mortality had excellent agreement (Figure 3).

Fig. 3 Scatter plots of observed versus predicted rates of: (a) mortality and (b) serious complications.
Table II Complication rates with brier scores

The Brier score is a measure of the accuracy of predictions. The lower the Brier score, the better the predictions are calibrated.
Brier scores were calculated from these binary outcomes demonstrating all complications with scores of less than 0.25 (Table II). Brier scores were then stratified by neck dissection factors (yes or no, unilateral or bilateral, and radical or selective), revealing superior scores for patients without neck dissection compared to those with neck dissection, of 0.15 and 0.23 respectively for serious complications. Twenty-six patients (59 per cent) who underwent bilateral neck dissections suffered from post-operative complications, compared with 28 patients (45 per cent) who underwent unilateral neck dissection. Brier scores were also stratified by rural or urban location, and free flap or non-free flap, for all of the complications in Table II. All of these Brier scores were less than 0.25, demonstrating varying degrees of prediction (0.25 being a non-informative model).Reference Steyerberg, Vickers, Cook, Gerds, Gonen and Obuchowski 11
Discussion
This multi-institutional, retrospective chart review reports the predicted and actual outcomes of 120 patients treated with total, hemi- or partial glossectomy using the American College of Surgeons' National Surgical Quality Improvement Program risk calculator. Complications following surgery increase treatment costs, delay adjuvant therapy, affect a patient's quality of life and prolong length of stay in hospital. Increasing the understanding and knowledge of risks associated with post-operative complications allows for improved counselling pre-operatively and management in the post-operative setting.Reference Downey, Friedlander, Groeger, Kraus, Schantz and Spiro 12 This study demonstrated low rates of mortality and moderate rates of morbidity, which correlated with both American College of Surgeons' National Surgical Quality Improvement Program predicted calculations and the published literature.Reference de Melo, Ribeiro, Kowalski and Deheinzelin 13 These results show comparable levels of surgical performance and post-operative complications assessed using the American College of Surgeons' National Surgical Quality Improvement Program risk calculator in patients undergoing surgery for tongue cancer in South Australia.
In the early stages of its development, the American College of Surgeons' National Surgical Quality Improvement Program risk calculator was based on data from general and vascular surgery, with a small percentage of data from otolaryngological head and neck surgical procedures.Reference Khuri, Daley, Henderson, Hur, Demakis and Aust 1 The current American College of Surgeons' National Surgical Quality Improvement Program risk calculator may not be specific enough for head and neck surgical procedures, as it does not take into consideration factors such as previous radiotherapy and free flap reconstructions.Reference Bilimoria, Liu, Paruch, Zhou, Kmiecik and Ko 6 Lewis et al. have attempted to improve the risk calculator by producing a specialty-specific programme, based on the American College of Surgeons' National Surgical Quality Improvement Program platform, to identify head and neck surgery specific variables and outcomes.Reference Lewis, Aloia, Shi, Martin, Lai and Selber 14
This study demonstrated significant discrepancies between American College of Surgeons' National Surgical Quality Improvement Program predicted length of stay and actual length of stay in South Australian patients. A review of the literature revealed similar incidences and five-year survival rates of patients with oral cavity malignancies in Australia and USA; thus, it appears there are some similarities between the two populations.Reference Chaturvedi, Anderson, Lortet-Tieulent, Curado, Ferlay and Franceschi 4 , Reference Carvalho, Nishimoto, Califano and Kowalski 5 Potential confounding factors, not taken into account by the risk calculator, include age, gender, rural or urban location, TNM classification, previous chemoradiotherapy, neck dissection, free flap reconstruction, and surgical risk grading. The median length of stay index ranged from 1.33 (T1 tumour classification) to 4.43 (T3 tumour classification).
It is possible that the overall length of stay in our study was affected by the 56 per cent of patients who required free flap reconstruction, as the length of stay was increased in those patients (median, 19 days) compared to those who did not require a free flap (median, 6 days). Studies conducted in the USA on patients with free flap reconstruction on the oral cavity demonstrated median length of stay ranging between 3 and 13 days.Reference Frederick, Sweeny, Carroll, Peters and Rosenthal 15 – Reference Schwam, Sosa, Roman and Judson 17 Clark et al. conducted a study in Canada on free flap reconstruction patients following head and neck cancer resections, and demonstrated a median length of stay of 15 days.Reference Clark, McCluskey, Hall, Lipa, Neligan and Brown 18 Studies conducted in the UK and Spain have demonstrated median lengths of stay of 16 and 33 days, respectively.Reference Rogers, Lowe, Brown and Vaughan 19 , Reference Vega, Leon, Cervelli, Pons, Lopez and Fernandez 20 Our results are comparable to other studies conducted in Australia, which have reported a mean length of stay ranging from 7 to 22 days.Reference O'Brien, Nettle and Lee 21 , Reference Gavriel, Thompson, Kleid, Chan and Sizeland 22 Although our length of stay is higher than predicted, it is comparable to many institutions. Furthermore, the acceptable length of stay index was higher in the non-free flap group (14 per cent) than the free flap group (7 per cent). Thus, on the basis of these findings, the American College of Surgeons' National Surgical Quality Improvement Program is better at predicting length of stay in non-free flap patients. These results demonstrate a need for specialty-specific variables to be identified and utilised in the risk calculator. We also hypothesise that the increased length of stay could be a result of differences in healthcare systems, but this requires further research in a prospective manner to study this possibility.
Our findings demonstrate that the American College of Surgeons' National Surgical Quality Improvement Program risk calculator may be useful in the prediction of post-operative complications; however, it may not be applicable for use in an Australian population when predicting length of stay. Predicted and actual complication rates were comparable, demonstrating similar complication rates among patients treated in South Australia and the USA. Thus, the longer length of stay in South Australia is less likely to be because of post-operative complications.
This study demonstrated an overall complication rate of 50 per cent and a serious complication rate of 38 per cent, similar to current published literature.Reference de Melo, Ribeiro, Kowalski and Deheinzelin 13 Wound infection (24 per cent) was identified as the most frequent complication following treatment. Numerous studies have associated increased post-operative wound infection rates with tumour (T) classification, node (N) classification, type of resection, complexity of procedure, utilisation of flaps, prior radiotherapy, co-morbidities and use of antibiotics.Reference Robbins, Favrot, Hanna and Cole 23 – Reference Lotfi, Cavalcanti Rde, Costa e Silva, Latorre Mdo, Ribeiro Kde and Carvalho 25 Thirty-day mortality was low in this study population, with one death. This patient suffered from a recurrence of tongue squamous cell carcinoma, which had previously been treated with pharyngolaryngectomy and radiotherapy. Unfortunately, he suffered from wound breakdown following total glossectomy, leading to carotid artery rupture. Many factors must be considered when risk-stratifying patients in the development of post-operative wound infections. Prophylactic antibiotic therapy has been demonstrated to be beneficial in decreasing rates of wound infection; however, there is no consensus regarding treatment duration and specific types of antibiotics.Reference Lotfi, Cavalcanti Rde, Costa e Silva, Latorre Mdo, Ribeiro Kde and Carvalho 25 – Reference Simo and French 27 Furthermore, analysis with the scatter plots and Brier scores demonstrated accurate predictions of mortality and serious complications with the American College of Surgeons' National Surgical Quality Improvement Program risk calculator.
Neck dissections are an important component in the treatment of tongue malignancies, either for staging of the N0 neck or treatment of the N+ neck. In this study, 106 patients underwent neck dissections, of which 42 per cent were bilateral and 58 per cent were unilateral. Rates of morbidity and mortality are closely associated with the choice of neck dissection performed.Reference Kasperts, Slotman, Leemans and Langendijk 28 Fifty per cent of patients who underwent neck dissection had associated complications. In comparison, Schwam et al., in a study utilising data extracted from the American College of Surgeons' National Surgical Quality Improvement Program database, reported a complication rate of 33 per cent following neck dissection in conjunction with oral cavity resections.Reference Schwam, Sosa, Roman and Judson 17 Our study demonstrated similar rates to that by Awad et al., who reported a complication rate following neck dissection of 34 per cent.Reference Awad, Shuman, Montero, Palmer, Shah and Patel 29 However, Chen et al. reported a complication rate of 8.5 per cent when neck dissections were performed alone, utilising the National Surgical Quality Improvement Program database.Reference Chen, Roman, Sosa and Judson 7 As neck dissections are typically employed for patients with advanced disease, the prolonged operating time and involvement of various critical structures may explain the higher complication rates.Reference Schwam, Sosa, Roman and Judson 17
Pneumonia occurred in 17 per cent of patients in this study, similar to current published literature.Reference Hanasono, Friel, Klem, Hsu, Robb and Weber 16 , Reference Amendola, Wolf, Coy and Amendola 30 Patients undergoing glossectomy can experience severe dysphagia, predisposing them to high risks of aspiration.Reference Kao, Ooi and Peters 31 , Reference Kao, Peters, Krishnan and Ooi 32 Close monitoring of aspiration pneumonia, in conjunction with early mobilisation and chest physiotherapy, is required to prevent aspiration pneumonia.
This study yielded similar results to a publication by Hatcher et al.Reference Hatcher, Bell, Browne and Waltonen 33 In that study, 12 per cent of patients who underwent microvascular reconstruction for head and neck malignancies were discharged to a nursing home. They found that older patients had higher risks of discharge to a nursing home, with prolonged length of stay associated with the higher risk. Patients awaiting placement to a nursing home would also be expected to have longer length of stay compared with patients discharged home.
As demonstrated above, a large range of factors have the potential to account for discrepancies in length of stay. A source of inconsistency may be that the American College of Surgeons' National Surgical Quality Improvement Program risk calculator does not take into account the TNM classification, use of free flaps or prior chemoradiotherapy. An international multi-institutional study would be beneficial to further identify factors affecting length of stay, with the ultimate goal of improving length of stay index and post-operative outcomes.
This study has several strengths and limitations. The retrospective nature of this study is subject to recall bias, and inaccuracies in medical records and databases. Given the low rates of morbidity and mortality, it is difficult to ascertain the statistical significance of the differences between actual and predicted complications rates. Despite these limitations, this is the first study to review the applicability of the American College of Surgeons' National Surgical Quality Improvement Program risk calculator in Australian glossectomy patients.
-
• The American College of Surgeons' National Surgical Quality Improvement Program was developed to improve surgical outcomes, providing risk-adjusted data
-
• Thirty-day mortality and morbidity rates of South Australian tongue cancer patients were retrospectively reviewed to evaluate this tool's validity
-
• Length of stay discrepancies may be because the risk calculator does not consider disease classification, free flap use or prior chemoradiotherapy
-
• The results demonstrate a need for specialty-specific variable identification and utilisation in the risk calculator
-
• Increased length of stay could be a result of differences in healthcare systems
-
• The risk calculator is useful in predicting complications of South Australian glossectomy patients
This study has demonstrated the American College of Surgeons' National Surgical Quality Improvement Program risk calculator to be useful in predicting complications for glossectomy patients treated in South Australian hospitals. These aspects of the risk calculator should be incorporated to assist in counselling patients prior to surgery.
Conclusion
The American College of Surgeons' National Surgical Quality Improvement Program risk calculator has demonstrated accuracy in predicting post-operative complications in South Australian glossectomy patients. Hence, it would be beneficial to incorporate the risk calculator into patient pre-operative investigation. However, there are significant discrepancies in predicted and actual length of stay in these patients. These differences in length of stay have been hypothesised to be secondary to head and neck surgery potential confounding factors, such as use of free flap reconstruction, and may also reflect differing healthcare systems. Further research is required to investigate these differences.