Larynx and hypopharynx cancer represents almost 45% of all treated diseases in this tertiary care hospital's department of otolaryngology, specialized in oncology.

Prevalence of pharyngocutaneous fistula after larynx and pharyngolarynx cancer surgery is reported to be between 9% and 25%.1,2 Its occurrence vastly increases these patients’ length of stay and consequently the treatment cost.3 Moreover, its occurrence can cause, among other physical and psychological damages, the delay in onset of complementary therapies (such as radiotherapy/chemotherapy), consequently delaying recovery.

Until now, no study has proposed a clinical decision rule (CDR) for predicting pharyngocutaneous fistula after total laryngectomy and primary or rescue pharyngolaryngectomy, with or without neck dissection, and major predictive factors were scarcely assessed.4 Thus, the goal was to derive such a CDR in a sample of patients from this hospital. Its creation is intended to encourage clinical behavior modification in order to avoid development of this complication and thus facilitate the reduction of unnecessary costs as well as improve patients’ quality of life.

A retrospective cohort study was conducted (longitudinal historical cohort study), consecutively including all the patients performing total laryngectomy/pharyngolaryngectomy, due to cancer of the larynx and hypopharynx, in the Otolaryngology Department of this hospital, from February of 2006 to June of 2011. All patients diagnosed with laryngeal and pharyngolaryngeal cancer were evaluated by the team before final treatment. Only procedures consistent with total laryngectomy with or without partial pharyngectomy and primary closure of the pharyngeal defect were considered in this study. The operations were performed by four different surgeons, two with ten or more years of experience, and two with less.

This study conduction was approved by the Ethics Committee of this hospital.

This study included 171 subjects, characterized in Table 2.

All the participants were M0 and pharyngeal closure was performed with two layers (muscles and mucosa) of continuous suture or stapler (used whenever possible).

During a mean follow-up of 30 days, 48 subjects (28.1%) developed pharyngocutaneous fistula postoperatively. They were all classified as ASA 2 or 3; 79.2% as high alcohol consumers, almost 48.9% had preoperative hemoglobin level below 12.5g/dL, 16.7% were diabetic patients, 6.2% had some degree of COPD, 22.9% showed impaired liver function or some degree of heart disease, 8.3% had received radiotherapy previously to the operation, 18.8% had undergone chemotherapy/radiotherapy prior to the operation, and in 58.3% an emergency tracheotomy before elective surgery was required. In fact, 45% were performed emergently and 13.3% were elective tracheotomies.

It was observed that patients with advanced stage (IVa and IVb) had the largest absolute number of pharyngocutaneous fistula (n=40) observed in the sample, although it was not statistically significant (p=0.25). However, the 87 remaining patients in these stages (representing almost 50% of the sample) did not develop a fistula. Univariate and multivariate analysis was performed with the inclusion of the relevant factors obtained by a previous systematic review,4 which had potential for inclusion in the predictive model being developed, as already explained (Table 3).

It was found that ASA 2 and 3, prior CRT, tracheotomy performed before surgery, and low levels of hemoglobin were associated with fistula development in the univariate analysis. In the multivariate analysis, only the ASA classification maintained statistical significance (p<0.05). Albumin, prior CRT, and tracheotomy performed before surgery showed predictive potential (p<0.2) for inclusion in the model.

Although albumin did not achieve statistical significance in the univariate analysis and there were many missing albumin values in this sample (85 missing), it was observed that its exclusion from the model substantially reduced the validity (particularly regarding sensitivity).

Alcohol consumption, preoperative hemoglobin<12.5g/dL, DM presence, N-classification, primary tumor location and extension, and no prior Rt were not associated with fistula development, but increased the CDR's prognostic accuracy (regarding the balance of sensitivity-specificity) and thus were also included.

The model therefore included all pre-operative variables analyzed that presented a p-value<0.2 and all those considered important by a systematic review previously conducted by this team.4 The best results were obtained using the following model:

This model presented an AUC of 0.74 (95% CI 0.61–0.86), as shown in Fig. 1.

Figure 1.

Model's area under the receiver operating characteristic curve. Ivory line, line of reference of AUC 0.5; blue line, score of the model; green line, four categories of the model.

(0,09MB).

The model's score stratification into four different risk categories is proposed: low risk (for a score ≤9.88), medium-low (9.881–12.25), medium-high (12.26–17.1), and high risk (score >17.1). This stratification resulted in the prognostic accuracy values in Table 4. The values for the cut-offs were selected by an assessment of AUC coordinates, using those that provided best fit.

Approximately half of the participants (48%) who developed a pharyngocutaneous fistula were classified as high risk. By associating the three highest risk groups (high+medium-high+medium-low), 92% of fistula development was predicted. It was also observed that with an increase in the risk degree, more subjects develop fistula (p<0.001 for chi-squared in an association and trend analysis).

A high specificity for the high-risk group (93%) was noted, with moderately high positive likelihood ratio (LR+) of (7.10) and positive (PPV) and negative predictive values (NPV) with high results (76% and 79% respectively) (Table 5). With the three groups of medium-low/medium-high/high risk, a sensitivity of 92%, negative likelihood ratio (LR-) of 0.25, and NPV of 89% were obtained.

The measures of prognostic validity for ASA classification were also calculated, in order to compare with the results of Farwell et al.8; these values are shown in Table 6. Thus, this derived model is statistically superior to ASA classification when compared the measures of prognostic validity.

Clinical decision rules (CDR) are instruments used to increase health professionals’ accuracy in diagnostic and prognostic assessment by summarizing the probability of a particular outcome.

Pharyngocutaneous fistula represents a major postoperative complication of laryngectomized patients. Until now, no CDR has been derived for determining the risk of fistula development; consequently, this study was conducted. As 13 variables were included, the inclusion of 130 to 195 participants was required. The present study included 171 patients, which should be an adequate number.5 It was decided to use four categories in order to maintain equilibrium between the best fit of the model and clinical application; i.e., on one hand, with a higher number of categories, the classification can be adjusted more effectively to the original score's AUC; on the other hand, clinical applicability improves with a lower number of categories.

In this analysis, the incidence of the pharyngocutaneous fistula was 28.1%, which is between the values obtained by other authors.9–14

For most authors, the male gender was more represented (in the present sample, 97.1%), and fistula developed around the 10th to the 15th postoperative day, as demonstrated in 38% of the present cases.1,15–23

This analysis has shown, in agreement with other studies,17,24,25 that for the advanced stage (IVa and IVb) there seems to be a higher fistula incidence. This relationship was not statistically significant both in the present as well as in other studies.1,10–13,15,16,18–23,26–29

The univariate analysis confirmed that CRT and prior tracheotomy are factors associated with formation of pharyngocutaneous fistula, as did other authors.9,12,13,24,26,30,31 Tracheotomy previous to the operation was not considered a risk factor for the fistulization by several authors15,18,20–22,25–28,31 contrary to the present findings. In performing the multivariate analysis, these variables remained as potential predictive factors for the fistula development, as other authors have concluded.24,30

In the univariate analysis, radiotherapy prior to surgery was not significant for pharyngocutaneous fistula development, as other authors concluded.2,11,14,18,20,21,25–29,32,33 Their inclusion in the final model did not add significant changes in accuracy, thus they were excluded.

ASA classification, unlike other studies,8,21,34 was significant as a risk factor for fistula development in the univariate and multivariate analysis, which demonstrates that the presence of pre-operative co-morbidities and functional status of the patient have a major importance in the development of intra- and post-surgical complications.

Few authors have assessed the ASA classification's predictive value8 for this outcome, although its relevance has been demonstrated in the present sample, reflecting the key role of the pre-operative control of comorbidities. Paradoxically, evaluating each comorbidity separately (DM, COPD, liver disease, and heart disease), they did not demonstrate significant association with fistula development (both in the univariate and multivariate analysis). In this aspect, the available evidence is contradictory, as some authors reported the same results11,13 while others found a significant relationship between fistula and these comorbidities.1,12,15,16,20,21,34

After the multivariate analysis, pre-operative hypoalbumin (<3.7g/L) was an independent factor for the development of the pharyngocutaneous fistula, as other authors concluded.12,15,20,34

In the available literature, it was not possible to find a CDR developed specifically for stratifying patients by their risk of pharyngocutaneous fistula formation after pharyngolaryngectomy and total laryngectomy.

For the high-risk group, a high specificity was observed, and therefore the ability to determine among patients at highest risk those who will not develop pharyngocutaneous fistula. However, the model identified with high sensitivity the groups of medium-low+medium-high+highrisk subjects who will develop a fistula. In a hospital setting, this allows the distribution of economic and personnel resources primarily to patients with higher probability for the occurrence of postoperative complications. It also allows healthcare professionals to act more intensively and earlier in the pre-operative and post-operative period in order to assist the identified the risk group.

The terrible consequences for the patient's quality of life and prognosis after development of pharyngocutaneous fistula are well-known. Moreover, the economic resources targeted to the healing this complication are a tremendous financial burden on any healthcare system.

It should be emphasized that the derived model has some limitations. The collection of the variables was performed retrospectively and therefore was not as complete and accurate as it would have been with a prospective basis. Examples include the absence of albumin in the preoperative records of 85 patients and the lack of observed records to characterize the nutritional status of the patient prior to surgery. Regardless of the limiting factors, a CDR has been developed for easy application with robust values for prognostic accuracy, namely the AUC value of 0.76 (95% CI 0.64–0.87).

We have statistically calculated a mathematical model with a good accuracy to detect the groups with medium/high risk of pharyngocutaneous fistula development. Ulteriorly, it will allow a best distribution of economic and personal resources targeting to patients with higher probability of this complication occurrence. These study’ major strengths are the fact that it represents first adequate and specifically CDR derived for the pharyngocutaneous fistula development risk stratification that additionally presented robust values of prognostic validity. The major limitations of the study are its retrospective and unicenter design and a modest sample size. Despite these limitations our CDR outperformed the ASA classification – which was validated for this outcome prediction.

This CDR is presented as a proposal for a later prospective multicenter validation, reliability assessment and optimization. We also wish to assess their applicability practice among health professionals and therefore make an economic impact and ease of use analysis of implementation among health services in our country.

The authors declare no conflicts of interest.