R H I N O L O G Y
Using preoperative unsupervised cluster analysis of chronic rhinosinusitis to inform patient decision and endoscopic sinus surgery outcome
Choaib Adnane1• Taoufik Adouly1• Amine Khallouk1• Sami Rouadi1• Redallah Abada1•Mohamed Roubal1•Mohamed Mahtar1
Received: 31 May 2016 / Accepted: 20 September 2016 ÓSpringer-Verlag Berlin Heidelberg 2016
Abstract The purpose of this study is to use unsupervised cluster methodology to identify phenotype and mucosal eosinophilia endotype subgroups of patients with medical refractory chronic rhinosinusitis (CRS), and evaluate the difference in quality of life (QOL) outcomes after endo- scopic sinus surgery (ESS) between these clusters for better surgical case selection. A prospective cohort study inclu- ded 131 patients with medical refractory CRS who elected ESS. The Sino-Nasal Outcome Test (SNOT-22) was used to evaluate QOL before and 12 months after surgery.
Unsupervised two-step clustering method was performed.
One hundred and thirteen subjects were retained in this study: 46 patients with CRS without nasal polyps and 67 patients with nasal polyps. Nasal polyps, gender, mucosal eosinophilia profile, and prior sinus surgery were the most discriminating factors in the generated clusters. Three clusters were identified. A significant clinical improvement was observed in all clusters 12 months after surgery with a reduction of SNOT-22 scores. There was a significant difference in QOL outcomes between clusters; cluster 1 had the worst QOL improvement after FESS in comparison with the other clusters 2 and 3. All patients in cluster 1 presented CRSwNP with the highest mucosal eosinophilia endotype. Clustering method is able to classify CRS phe- notypes and endotypes with different associated surgical outcomes.
Keywords Cluster analysisPhenotypeEndotype Quality of lifeChronic rhinosinusitisSNOT-22
Introduction
Chronic rhinosinusitis (CRS) is a common otorhinolaryn- gologic disease of a multi-factorial origin that has a great impact on quality of life (QOL) [1]. In the United States, it affects approximately 14–16 % of the population [2].
Based on clinical presentation, CRS includes at least three distinct categories: CRS without nasal polyps (CRSsNP), CRS with nasal polyps (CRSwNP), and allergic fungal rhinosinusitis (AFRS) [3]. In spite of the recent advances, the etiologic factors, pathogenesis, and treatment of CRS remain a matter of debate [4]. Recently, functional endo- scopic sinus surgery (FESS) is became the treatment of choice for patients with CRS resistant to medical therapy [5]. Many specific instruments, such as the Sino-Nasal Outcome Test (SNOT-22) questionnaire, have been used for quantifying QOL improvement after CRS treatment [6].
Disease stratification using phenotypic characterization, which can be performed by data-driven methods, has been developed to improve clinical and therapeutic decisions [7]. However, it does not provide complete understanding about all underlying cellular and molecular pathophysio- logic mechanisms of CRS. The concept of multiple groups of biological subtypes, or ‘‘endotypes’’ which are identi- fied by corresponding biomarkers has been recently studied [8]. Classification of CRS according to phenotypes and endotypes may allow the identification of subgroups in relation to treatment response [9].
The objective of this study is to use unsupervised cluster methodology to identify phenotype and mucosal eosino- philia endotype subgroups from a prospective cohort of patients with CRS, describe clinical differences and dis- criminant analysis in these identified clusters, and identify which groups (clusters) of surgical patients experience the largest magnitude of QOL improvement following ESS.
& Choaib Adnane
adnanechoaib@gmail.com
1 Department of ENT, 20 Aouˆt Hospital, Ibn Rochd University Hospital, Casablanca, Morocco
DOI 10.1007/s00405-016-4315-8
This study attempts to use clustering analysis to predict outcomes of sinus surgery, which has enormous potential for clinical utility in identifying which patients may best benefit from surgical intervention.
Materials and methods Study design and participants
The present work is a continuation of the previously pub- lished study from our group [10]. It is a prospective obser- vational cohort study of 131 patients undergoing FESS for CRS from January 2012 to December 2013 in Ibn Rochd university hospital, in Casablanca. The research protocol was approved by the Ethics in Research with Human Beings Committee. A written consent was obtained from the patients, and no one refused to take part in the study.
Inclusion criteria were as follows: adult patients (18 years or older) who had undergoing FESS for CRS (with or without polyps) refractory to medical treatment, and had preoperative endoscopic examination and sinus CT scan. Patients were excluded if the CRS medical therapy was recently changed before surgery. Patients were also excluded if they had age less than 18 years, previous trauma, only anatomical disorders etiology, pregnancy, tumors, or other disorders, such as cystic fibrosis, primary ciliary dyskinesia, and immune deficiencies.
Demographics and CRS diagnosis
Demographics, medical, and surgical history were recorded for each patient. The diagnosis criteria of CRS were used as reported by the European Position Paper on Rhinosi- nusitis and Nasal Polyps (EPOS 2012) [11].
Disease-severity measures
Sinus CT scan was classified using the Lund–Mackay scoring system (score range, 0–24), and the endoscopic examination was scored using the Lund–Kennedy scoring system (score range, 0–20) [12]. Sinus-specific QOL was measured using the Sino-Nasal Outcome Test (SNOT-22) (score range, 0–110) to quantify the patient’s symptoms before and 12 months after surgery [6].
Definition of mucosal eosinophilia profile of CRS was based on tissue eosinophil amount[5 % of all leukocytes in five visual fields or[5 cells/HPF [13].
Treatment and follow-up
Prior to surgery, all subjects had previously failed medical treatment. It was based on topical steroid applications
‘‘budesonide spray’’ on each nostril every 12 h for a min- imum of 4 weeks, short courses of oral steroid (pred- nisolone) for 7 days, and broad spectrum antibiotics at least for 3 weeks.
FESS was performed under general anesthesia after failure of medical therapy protocol. The extent of surgery was based on CRS CT-scan extension and type: for CRS with nasal polyps (CRSwNP), bilateral maxillary antros- tomy with functional total ethmoı¨dectomy was performed, but for CRS without nasal polyps (CRSsNP), surgical extension was based on CT-scan extension. Septoplasty and/or inferior turbinoplasty were performed when indicated.
Postoperatively, the nasal packing was removed after 2 days, and all patients were given short course of antibi- otic (amoxicillin 1 g?clavulanate 125 mg) for 1 week.
Nasal saline douching was given for 1 month, and topical nasal corticosteroid was started 15 days after surgery and continued if necessary.
Cluster analysis methodology Clustering variables selection
Variables chosen for cluster modeling were selected on the basis of their considered contribution to characterizing the CRS phenotype and endotype. Demographic, comorbidity, subjective, and objective clinical findings were reduced to nine factors using factor analysis and absolute correlation at 0.80.
Clustering procedure
Cluster analysis methodology was applied using a two-step statistical approach. All measurements were standardized using z scores for continuous variables and 0 or 1 for categorical variables. According to the indication, contin- uous variables were log-transformed to approximate a normal distribution. Discriminant function analysis was performed using backward stepwise algorithm on cluster model to evaluate the input variables that were significant determinants of model clustering.
Statistical analysis
All analyses were performed using SPSS 20.0. The normal distribution was assessed using Shapiro–Wilk test and skewness kurtosis z values. Descriptive statistics were summarized by means, percentages, and standard devia- tions (SDs).
The between-cluster comparison of baseline parameters was performed to identify which of the input variables were significantly different between clusters. Comparisons were
provided using analysis of variance (ANOVA) or a Kruskal–
Wallis (KW) test for quantitative variables and Pearson’sv2 or Fischer’s exact test for qualitative variables.
The relative improvement for each preoperative SNOT-22 score was calculated using the absolute value of the formula:
[(postoperative score)-(preoperative score)/mean preoper- ative score]. Larger values indicate larger postoperative improvements. The Wilcoxon signed-rank test was used to assess improvement of SNOT-22 scores after surgery.
Correlation and multiple logistic regression analysis were used to evaluate the relationship between preopera- tive characteristics and significant relative improvement after FESS [(mean preoperative score)-(mean postoper- ative score)/mean preoperative score]. Multiple regression models were chosen using stepwise method selection. The variables significant at the 0.2 level in bivariate analyses were entered into this multivariate logistic regression.
Apvalue under 0.05 (5 %) was considered statistically significant for all analyses.
Results
Participant characteristics
A total of 131 patients with refractory CRS who elected FESS were recruited. After 12 months follow-up, only 113
(86 %) subjects were retained in the final study: 46 patients with CRSsNP and 67 patients with CRSwNP. Mean fol- low-up for the overall cohort study was 19 months (range 12–36 months).
The mean age was 38.7 years, the median was 40 years (IQR 17 and SD 12.7), and the sex ratio was 0.82 (51 male and 62 female). Baseline characteristics of patients in the study are described in Table1.
Overall QOL outcomes after FESS
A strongly statistically significant reduction was seen in SNOT-22 scores after surgery [57.6 (IQR 39) versus 31.8 (IQR 32), Wilcoxon signed-rank test, T=0, z= -9.23, r= -0.87,p\0.0001].
Cluster description
Using the clustering approach outlined above, a dendro- gram was generated, and three-cluster model was the best fit of medical refractory CRS dataset (Fig.1). Differences across clusters are presented for demographic factors, comorbidities, and disease-severity metrics (Table 1).
Three phenotypes were identified. Cluster 1 was the largest cluster (n=52; 46 %) and had more women (67.3 %) than men with slightly more mucosal eosinophilia endotype (53.8 %). All patients in this group presented
Table 1 Comparison of baseline characteristics in the three clusters
Variable Cluster 1 Cluster 2 Cluster 3 Total Significance (pvalue)
Number of patients 52 (46 %) 26 (23 %) 35 (31 %) 113 –
Gendera
Female 35 (67.3 %) 0 27 (77.1 %) 62 (54.9 %) \0.001
Male 17 (32.7 %) 26 (100 %) 8 (22.9 %) 51 (45.1 %) \0.001
Agea 38.4±12.2 40.4±15.7 38.0±11.0 38.7±12.7 0.735
Prior sinus surgerya 10 (19.2 %) 0 7 (20 %) 17 (15 %) 0.026
Asthmaa 29 (55.8 %) 7 (26.9 %) 2 (5.7 %) 38 (33.6 %) \0.001
ASA intolerancea 10 (19.2 %) 0 0 10 (8.8 %) 0.001
Polypsa 52 (100 %) 15 (57.7 %) 0 67 (59.3 %) \0.001
Endoscopy score 9.5±2.3 6.8±3.6 2.9±1.1 6.8±3.7 \0.001
CT-scan score 20.8±3.3 14.5±8.4 5.0±2.0 14.5±8.3 \0.001
Time to ESSa 33.6±32.3 33.1±23.0 17.7±8.9 28.6±25.9 0.067
Mucosal eosinophilia endotypea 28 (53.8 %) 0 13 (37.1 %) 41 (36.3 %) \0.001
SNOT-22 totala 73.0±16.7 54.5±20.1 36.9±8.7 57.6±22.1 \0.001
SNOT-22 domain 1 (rhinologic symptoms) 20.5±4.8 16.5±5.8 13.1±4.2 17.3±5.8 \0.001 SNOT-22 domain 2 (extranasal rhinologic symptoms) 9.7±2.6 8.0±3.6 6.0±1.9 8.2±3.1 \0.001 SNOT-22 domain 3 (ear/facial symptoms) 10.7±4.1 7.7±4.4 5.3±2.8 8.3±4.5 \0.001 SNOT-22 domain 4 (psychological dysfunction) 23.5±8.6 15.9±8.1 10.8±2.0 17.8±9.0 \0.001 SNOT-22 domain 5 (sleep dysfunction) 18.7±5.7 13.2±5.1 10.9±2.9 15.0±6.0 \0.001 ESSendoscopic sinus surgery,SNOTSino-Nasal Outcome Test
a Variables used to define clusters
CRSwNP, and it was the group with the worst endoscopy scores (9.5±2.3), CT scores (20.8±3.3), and SNOT-22 scores (73.0±16.7). Cluster 2 was the only cluster where all patients were male and with non-mucosal eosinophilic profile. Cluster 3 had slightly more women (54.9 %) than men. All patients in this group presented CRSsNP. This group had the better endoscopy scores (2.9±1.1), CT scores (5.0±2.0), and SNOT-22 scores (36.9±8.7).
QOL outcomes in data-driven clusters after FESS
A significant clinical improvement was observed in all clusters 12 months after surgery with a reduction of SNOT- 22 scores. According to KW test, there was a significant difference in QOL outcomes between clusters; cluster 1
had the worst QOL improvement after FESS in comparison with the other clusters 2 and 3 (Table2).
Discriminant analysis
Discriminant analysis was performed on all nine variables used in the cluster model to identify those measures which best separate patients into clusters. Nasal polyps, gender, mucosal eosinophilia profile, and prior sinus surgery were the most discriminating factors.
Predictive factors influencing QOL improvement after FESS
After bivariate analysis, multivariate logistic regression model was examined nine predictive factors that can affect QOL improvement: nasal polyps profile, prior sinus sur- gery, asthma, ASA intolerance, Lund–Kennedy endoscopy score, Lund–Mackay CT score, time to ESS, mucosal eosinophilia profile, and preoperative SNOT-22 scores.
The predictive factors were: nasal polyps profile, prior sinus surgery, mucosal eosinophilia profile, asthma, and time to surgery (Table3). This model was able to explain 55.6 % of relative change in QOL (R2=0.556 and adjusted R2=0.535). There was no collinearity within the data (VIF\10, tolerance statistics\0.1). Therefore, the presence of asthma, prior sinus surgery, nasal polyps, or mucosal eosinophilia endotype had a negative impact on QOL improvement after surgery. Time to surgery had a little significant effect to QOL improvement; long time between diagnosis and surgery had a little good effect on QOL outcome. Patients in cluster 1 had the worst QOL improvement after surgery, and patients in cluster 3 had the best QOL improvement.
Discussion
The need for classifying CRS heterogeneity has gained a great importance with the parallel development of better tools for measuring disease characteristics in clinical and pathologic biomarkers, together with novel medical and surgical therapies that are only likely to be efficacious in Fig. 1 Algorithm tree of cluster construction
Table 2 Cluster specific quality of life outcomes after FESS
Variable Cluster 1 Cluster 2 Cluster 3 Total Significance (pvalue) Number of patients 52 (46 %) 26 (23 %) 35 (31 %) 113 –
SNOT-22 preop 73.0±16.7 54.5±20.1 36.9±8.7 57.6±22.1 \0.001 SNOT-22 postop 47.2±16.5 24.4±15.6 14.2±8.9 31.8±20.6 \0.001
RelativeD(%) 35±17 57±17 62±19 48±21 \0.001
FESSfunctional endoscopic sinus surgery,SNOTSino-Nasal Outcome Test
DRelative change value of SNOT-22 score after surgery represented with percentage
particular subgroups of CRS. The data-driven analysis had shown that the three phenotype and mucosal eosinophilia endotype groups constructed in this way exhibit clinically symptoms and QOL relevant differences in outcome, with management strategy that uses FESS.
Traditional phenotype classifications and measures of disease severity have not universally been found to impact treatment outcome. Studies about clinical predictors of surgical success were contradictory. This study used unsupervised statistical methods to generate clusters based on prospectively collected data from a cohort of patients with CRS refractory to medical treatment. Interestingly, traditional measures such as polyp profile were signifi- cantly related to highly QOL, subjective, and objective measures of CRS severity. After clusters identification, a comparison concerning evolution and outcomes after FESS was studied.
Overall, patients with CRS demonstrate improvement in QOL after FESS, despite of comorbidity or other preop- erative factors analyzed. Most of studies found the same evolution after surgery [13–15]. However, according to the
three generated clusters, surgery outcome was significantly different especially between cluster 1 and 3; cluster 1 had the worst prognosis.
This result indicates that clustering according to polyp status, prior sinus surgery, and mucosal eosinophilia profile will help physicians for better surgical case selection. This is critical because maybe these generated clusters represent distinct pathophysiologies of CRS.
Several authors have described nasal polyps to have a significant negative impact on QOL patients and less improvement after surgery [16]. In contrast, Smith et al. in their study of 119 patients with CRS found nasal polyps to be a positive effect on QOL scores after FESS [17].
Of the various predictive factors affected the surgical outcome, prior sinus surgery was the most one described in the literature to be a negative effect on QOL outcome after revision sinus surgery [16,18]. This finding is consistent with the results of this paper and suggests that the first surgical intervention is extremely important.
Comorbid asthma was a secondary factor of poor prognosis after surgery. Therefore, most of authors Table 3 Predictors of disease-
specific QOL improvement after FESS (stepwise method)
Predictor B SEB ß p R2
Step 1
Constant 0.629 0.026
Nasal polyps -0.244 0.034 -0.559 \0.0001 0.312
Step 2
Constant 0.663 0.025
Nasal polyps -0.245 0.031 -0.560 \0.0001
Prior sinus surgery -0.223 0.042 -0.371 \0.0001 0.450
Step 3
Constant 0.692 0.024
Nasal polyps -0.228 0.029 -0.523 \0.0001
Prior sinus surgery -0.191 0.041 -0.318 \0.0001 0.517
Mucosal eosinophilia profile -0.119 0.031 -0.268 \0.0001 Step 4
Constant 0.693 0.024
Nasal polyps -0.199 0.032 -0.457 \0.0001
Prior sinus surgery -0.177 0.041 -0.296 \0.0001
Mucosal eosinophilia profile -0.108 0.031 -0.243 0.001
Asthma -0.072 0.034 -0.160 0.037 0.537
Step 5
Constant 0.666 0.027
Nasal polyps -0.219 0.033 -0.502 \0.0001
Prior sinus surgery -0.220 0.045 -0.366 \0.0001
Mucosal eosinophilia profile -0.086 0.032 -0.194 0.008
Asthma -0.082 0.034 -0.181 0.017
Time to ESS 0.001 0.001 -0.167 0.034 0.556
Nasal polyps, prior sinus surgery, mucosal eosinophilia profile, asthma, and time to ESS had a negative impact on relative SNOT-22 score improvement
ESSendoscopic sinus surgery
revealed no major differences in outcome after FESS in groups of asthmatic and non-asthmatic patients [19].
Recently, Hopkins et al. reported that patients with delayed surgery reported less improvement in SNOT-22 scores than patients treated at earlier time points; regardless of comorbid status [20]. This study found that patients with delayed surgery reported little more QOL improvement.
This result may be explained by the long duration of maximal medical treatment as all of the patients were recruited from a tertiary center.
Several authors have suggested that tissue eosinophilia predicted significantly less improvement of symptoms, QOL, and relapse after FESS [10,21]. Other studies have not demonstrated the same thing [22].
This study had number of limitations that should be acknowledged. First, the patient population with CRS was obtained from a tertiary care center, making external gen- eralizations to other surgical or nonsurgical patient popu- lations challenging. Second, the sample size was relatively small, and the conclusions cannot be taken for granted.
Third, patients responding to the first-line medical therapy were excluded. Last, other biomarkers of CRS disease are absent in this data set.
To date, only a few studies have reported cluster anal- ysis of CRS [23–25]. However, all of them did not deter- mine if such clustering predicts treatment outcomes. The study by Nakayama et al. found that polyp score and mucosal eosinophil count were the strongest predictors of clustering by discriminant analysis [24]. In this study, the same thing was found, because all patients in cluster 1 presented CRSwNP with the highest mucosal eosinophilia endotype and had the worst QOL improvement after FESS.
Conclusion
This study supports a role for the use of multivariate techniques in an unsupervised classification of CRS populations. Traditional clinical phenotyping of CRS patients based upon nasal polyp status, combined with mucosal eosinophilia endotyping may provide a reliable framework for treatment response, exploratory molecular and genetic studies, presently undermined by population heterogeneity. The use of clustering analysis based on nasal polyps, gender, mucosal eosinophilia profile, and prior sinus surgery has enormous clinical utility in iden- tifying which patients may best benefit from surgical intervention.
Compliance with ethical standards
Conflict of interest All authors have no conflict of interest or financial support with this article.
Informed consent Informed written consent was obtained in advance from all patients included in this study, which was approved by the hospital’s Ethics Committee.
Research involving human participants and/or animals This manuscript is not a research involving human participants and/or animals.
References
1. Piccirillo JF, Merritt MG, Richards ML (2002) Psychometric and clinimetric validity of the 20-Item Sino-Nasal Outcome Test (SNOT-20). Otolaryngol Head Neck Surg 126:41–47
2. Anand VK (2004) Epidemiology and economic impact of rhi- nosinusitis. Ann Oto Rhinol Laryngol Suppl 193:3–5
3. Meltzer EO, Hamilos DL, Hadley JA, American Academy of Allergy, Asthma and Immunology, American Academy of Oto- laryngic Allergy, American Academy of Otolaryngology–Head and Neck Surgery, American College of Allergy, Asthma and Immunology, American Rhinologic Society et al (2004) Rhi- nosinusitis: establishing definitions for clinical research and patient care. Otolaryngol Head Neck Surg 131(6 Suppl):S1–S62 4. Lane AP, Turner JH (2012) Etiologic factors in chronic rhinosi- nusitis. In: Kennedy DW, Hwang PH (eds) Rhinology: diseases of the nose, sinuses, and skull base. Thieme Medical Publishers, New York, pp 171–181
5. Kennedy DW, Ramakrishnan VR (2012) Functional endoscopic sinus surgery: concepts, surgical indications, and techniques. In:
Kennedy DW, Hwang PH (eds) Rhinology: diseases of the nose, sinuses, and skull base. Thieme Medical Publishers, New York, pp 306–335
6. Hopkins C, Gillett S, Slack R, Lund VJ, Browne JP (2009) Psychometric validity of the 22-item Sinonasal Outcome Test.
Clin Otolaryngol 34:447–454
7. Bousquet J, Anto JM, Sterk PJ et al (2011) Systems medicine and integrated care to combat chronic noncommunicable diseases.
Genome Med 3:43
8. Agache I, Akdis C, Jutel M, Virchow JC (2012) Untangling asthma phenotypes and endotypes. Allergy 67:835–846 9. Akdis CA (2012) Therapies for allergic inflammation: refining
strategies to induce tolerance. Nat Med 18:736–749
10. Adnane C, Adouly T, Zouak A, Mahtar M (2015) Quality of life outcomes after functional endoscopic sinus surgery for nasal polyposis. Am J Otolaryngol 36(1):47–51
11. Fokkens WJ, Lund V, Mullol J (2012) European position paper on rhinosinusitis and nasal polyps 2012. Rhinol Suppl 23:1–298 12. Lund V, Kennedy D (1995) Quantification for staging sinusitis.
Ann Otol Rhinol Laryngol 104(Suppl 10):1–31
13. Soler ZM, Sauer DA, Mace J, Smith TL (2009) Relationship between clinical measures and histopathologic findings in chronic rhinosinusitis. Otolaryngol Head Neck Surg 141:454–461 14. Smith KA, Smith TL, Mace JC, Rudmik L (2014) Endoscopic
sinus surgery compared to continued medical therapy for patients with refractory chronic rhinosinusitis. Int Forum Allergy Rhinol 4:823–827
15. Soler ZM, Rudmik L, Hwang PH, Mace JC, Schlosser RJ, Smith TL (2013) Patient-centered decision making in the treatment of chronic rhinosinusitis. Laryngoscope 123(10):2341–2346 16. Dursun E, Korkmaz H, Eryilmaz A, Bayiz U, Sertkaya D, Samim
E (2003) Clinical predictors of long-term success after endo- scopic sinus surgery. Otolaryngol Head Neck Surg 129:526–531 17. Smith TL, Mendolia-Loffredo S, Loehrl TA, Sparapani R, Laud PW, Nattinger AB (2005) Predictive factors and outcomes in
endoscopic sinus surgery for chronic rhinosinusitis. Laryngo- scope 115:2199–2205
18. Kennedy JL, Hubbard MA, Huyett P, Patrie JT, Borish L, Payne SC (2013) Sino-nasal outcome test (SNOT-22): a predictor of postsurgical improvement in patients with chronic sinusitis. Ann Allergy Asthma Immunol 111(4):246–251
19. Bhattacharyya N (2007) Influence of polyps on outcomes after endoscopic sinus surgery. Laryngoscope 117(10):1834–1838 20. Hopkins C, Rimmer J, Lund VJ (2015) Does time to endoscopic
sinus surgery impact outcomes in chronic rhinosinusitis?
Prospective findings from the National Comparative Audit of Surgery for Nasal Polyposis and Chronic Rhinosinusitis. Rhi- nology 53(1):10–17
21. Soler ZM, Sauer D, Mace J, Smith TL (2010) Impact of mucosal eosinophilia and nasal polyposis on quality-of-life outcomes after sinus surgery. Otolaryngol Head Neck Surg 142:64–71
22. Eweiss A, Dogheim Y, Hassab M, Tayel H, Hammad Z (2009) VCAM-1 and eosinophilia in diffuse sino-nasal polyps. Eur Arch Otorhinolaryngol 266:377–383
23. Soler ZM, Hyer JM, Ramakrishnan V et al (2015) Identification of chronic rhinosinusitis phenotypes using cluster analysis. Int Forum Allergy Rhinol 5(5):399–407
24. Nakayama T, Asaka D, Yoshikawa M et al (2012) Identification of chronic rhinosinusitis phenotypes using cluster analysis. Am J Rhinol Allergy 26(3):172–176
25. Soler ZM, Hyer JM, Rudmik L, Ramakrishnan V, Smith TL, Schlosser RJ (2016) Cluster analysis and prediction of treatment outcomes for chronic rhinosinusitis. J Allergy Clin Immunol 137(4):1054–1062