Chronic rhinosinusitis (CRS) is a common inflammatory

Transcription

1 ORIGINAL ARTICLE Unsupervised cluster analysis of chronic rhinosinusitis with nasal polyp using routinely available clinical markers and its implication in treatment outcomes Jeong-Whun Kim, MD, PhD 1, Gene Huh, MD 1, Chae-Seo Rhee, MD, PhD 1, Chul Hee Lee, MD, PhD 1, Jaebong Lee, MS 2, Jin-Hang Chung, MD, PhD 3 and Sung-Woo Cho, MD 1 Background: Chronic rhinosinusitis with nasal polyps (CR- SwNP) is a multidimensional disease. In this study, we performed an unsupervised cluster analysis of CRSwNP using routinely available clinical markers. Methods: We conducted a retrospective review of patients treated with endoscopic sinus surgery due to medically intractable bilateral CRSwNP from 2009 to Unsupervised cluster analysis was performed using a patient s clinical features, including age, peripheral blood eosinophil, tissue eosinophilia, Lund-Mackay computed tomography (CT) scores, ratio of the CT scores for the ethmoid sinus and maxillary sinus (E/M ratio), and comorbid asthma. Tree analysis was performed to develop a clustering algorithm. Kaplan-Meier survival analysis was performed to determine the revision surgery free probability corresponding to each cluster. Results: Data were available on 375 patients. Patients were categorized into 6 clusters comprising 2 asthmatic clusters and 4 non-asthmatic clusters. The labels for the 2 asthmatic clusters were: asthmatic non-eosinophilic polyp (cluster A1) and asthmatic eosinophilic polyp (cluster A2). The labels for the 4 non-asthmatic clusters were: non-eosinophilic polyp with older age (cluster NA1); non-eosinophilic pol yp with younger age (cluster NA2); eosinophilic polyp with lower E/M ratio (cluster NA3); and eosinophilic polyp with higher E/M ratio (cluster NA4). The 4-year revision-free rates were 100% (cluster NA1), 80.3% (NA2), 98.0% (NA3), 66.7% (NA4), 100% (A1), and 66.7% (A2). The clusters showed statistically significant differences in terms of 4-year revisionfree rates (log-rank p < 0.05). Conclusion: Cluster analysis identified 2 asthmatic clusters and 4 non-asthmatic clusters in CRSwNP. Each cluster corresponded to a different clinical outcome. C 2018 ARS- AAOA, LLC. Key Words: cluster analysis; sinusitis; nasal polyps; treatment outcome; clinical markers How to Cite this Article: Kim JW, Huh G, Rhee CS, et al. Unsupervised cluster analysis of chronic rhinosinusitis with nasal polyp using routinely available clinical markers and its implication in treatment outcomes. Int Forum Allergy Rhinol.2019;9: Department of Otorhinolaryngology Head and Neck Surgery, Seoul National University Bundang Hospital, Seoul National University College of Medicine, Seongnam, South Korea; 2 Medical Research Collaborating Center, Seoul National University Bundang Hospital, Seongnam, South Korea; 3 Department of Pathology, Seoul National University Bundang Hospital, Seoul National University College of Medicine, Seongnam, South Korea Correspondence to: Sung-Woo Cho, MD, Department of Otorhinolaryngology Head and Neck Surgery, Seoul National University Bundang Hospital, 82 Gumi-ro 173th street, Bundang-gu, Seongnam, Gyeonggi-do 13620, South Korea; iamsungu@gmail.com Potential conflict of interest: None provided. Received: 16 April 2018; Revised: 21 August 2018; Accepted: 9 September 2018 DOI: /alr View this article online at wileyonlinelibrary.com. Chronic rhinosinusitis (CRS) is a common inflammatory upper airway disease that affects approximately 12.1% of the U.S. population 1 and approximately 8.4% of the Korean population. 2 Despite maximal medical therapy, 12.3% to 24.8% of patients undergo endoscopic sinus surgery (ESS), the current standard for surgical care. 3 In the current standard, CRS is clinically defined based on the presence of symptoms for more than 3 months with positive endoscopic findings and/or positive computed tomography (CT) findings. 4 Given that CRS diagnosis is not based on the endotypes, which explains the pathogenesis of the disease, it is considered a complex disease comprising several disease variants with different underlying pathophysiologies. 5 CRS with nasal polyps (CRSwNP) is 1 of the CRS phenotypes. CRSwNP has a greater clinical impact than other CRS phenotypes without nasal polyps, and patients have International Forum of Allergy & Rhinology, Vol. 9, No. 1, January

2 Kim et al. higher risk of undergoing ESS despite maximal medical treatment. 3 In addition, CRSwNP has a higher chance of disease recurrence and need for secondary revision surgery. 6 Thus, categorizing CRSwNP and identifying the correlations of CRSwNP subtypes with overall disease outcomes would be beneficial for clinicians. Cluster analysis can facilitate the categorization of heterogeneous disorders into disease subtypes and has recently been used to identify various inflammatory clusters based on the production of inflammatory cytokines within the tissue. 7 However, evaluating inflammatory biomarkers remains difficult. Thus, in the present study, we performed unsupervised cluster analysis of CRSwNP using preoperative and postoperative clinical parameters, which can be easily acquired. Patients and methods Study cohort We conducted a retrospective review of the medical records of CRSwNP patients who underwent ESS at Seoul National University Bundang Hospital from May 2009 to May Diagnosis of CRSwNP was based on the European Position Paper on Rhinosinusitis with Nasal Polyps. 4 Surgical procedures were targeted to the diseased sinuses based on objective signs of inflammation that are observed on either the radiographic or endoscopic exam. 8 Inclusion criteria were as follows: bilateral CRSwNP patients of all ages who had been refractory to medical treatment. Patients who underwent primary ESS before May 2009, or patients with unilateral disease, fungal disease, antrochoanal polyps, cystic fibrosis, primary ciliary dyskinesia, or other tumorous conditions, such as sinonasal inverted papilloma, were excluded. In addition, patients who were treated with systemic corticosteroids within the 4-week surgery period were excluded. This study was approved by the Institutional Review Board of Seoul National University Bundang Hospital (B-1801/ ). Assessment of preoperative and postoperative clinical characteristics Overall disease severity was evaluated based on the Lund- Mackay (LM) scoring system in the CT scan. 9 Mean LM score for right and left was calculated. The total ethmoid sinus score (E score, anterior ethmoid sinus score, and posterior ethmoid sinus scores for both sides), total maxillary sinus score (M scores for both sides) were additionally measured to calculate E/M ratio (ratio of E and M scores). 10 To determine the presence of asthma in our cohort, we searched for International Classification of Diseases and Related Health Problems, 10th Revision (ICD- 10) J45 codes from our institution s electronic health record system. A complete blood cell count with differential cell count under the patient s regular condition was performed within 4 weeks before surgery to check peripheral blood eosinophils. Nasal polyp tissue was obtained during ESS for histopathological analysis. Postoperative specimens were reviewed by our institution s pathologists. Pathological findings were routinely reported as eosinophilic or noneosinophilic. Tissue eosinophilia was considered when eosinophils accounted for >20% of the total inflammatory cells under magnification The polyp was defined as eosinophilic or non-eosinophilic according to the tissue eosinophilia status. Measurement of treatment outcomes Postoperatively, we administered broad-spectrum oral antibiotics for 1 to 2 weeks and intranasal corticosteroid spray for at least 1 month. Nasal saline irrigation was performed for >1 month. Endoscopic findings were evaluated at least 2 to 3 months after the operation according to a modified Lund-Kenney (mlk) scoring system, 12 which assesses edema, discharge, and polyp. The latest mlk score before any revision surgery was used for analysis. Revision ESS was performed for symptomatic patients who were refractory to systemic steroids or antibiotic therapy. Kaplan- Meier survival analysis was performed to determine the revision surgery-free probability rate. Censoring occurred at 4 years postsurgery or at the time of last visit for patients who were followed less than 4 years. Variable selection for cluster analysis Six variables were used for clustering and were selected based on their contributions to disease outcome determined after literature review The 6 variables were as follows: (1) comorbid airway disease (asthma); (2) blood eosinophil (%); (3) tissue eosinophilia (based on the pathologic report: eosinophilic or non-eosinophilic); (4) E/M ratio; (5) mean LM score; and (6) basic patient demographics (age). Age, blood eosinophil (%), E/M ratio, mean LM score are continuous variables while asthma and tissue eosinophilia are binary categorical variables. Continuous variables were normalized before cluster analysis. Statistical analysis Clustering analysis was performed using 2-step cluster analysis. The number of clusters was determined when a small value of Schwarz Bayesian criterion (BIC) was obtained and the change in Schwarz Bayesian criterion between adjacent number of clusters was small. 16 Cluster analysis cannot be performed on datasets with missing variables; therefore, subjects with missing data were removed. Differences between clusters were determined using 1-way analysis of variance (ANOVA) for continuous variables and chi-square test or Fischer s exact test for categorical variables. A classification tree based on classification and regression trees (CART) was generated using the input variables that were used in clustering. The relationship between clusters and the surgical outcome was evaluated using Kaplan-Meier survival analysis. Results are presented as mean ± standard deviation (SDs) or median with 80 International Forum of Allergy & Rhinology,Vol. 9, No. 1,January 2019

3 Cluster analysis of nasal polyps TABLE 1. Cluster characteristics Cluster NA1 NA2 NA3 NA4 A1 A2 p Proportion, n (%) 135/375 (36.0) 74/375 (19.7) 80/375 (21.3) 40/375 (10.7) 19/375 (5.1) 27/375 (7.2) Patients with asthma, n (%) 0(0) 0(0) 0(0) 0(0) 19 (100) 27 (100) <0.001 a Patients with tissue eosinophilia, n (%) 0(0) 0(0) 80 (100) 40 (100) 1 (5.3) 27 (100) <0.001 a Age (years), mean ± SD 52.0 ± ± ± ± ± ± 8.7 <0.001 b Blood eosinophil (%), mean ± SD 3.2 ± ± ± ± ± ± 2.3 <0.001 b E/M ratio, mean ± SD 1.8 ± ± ± ± ± ± 0.9 <0.001 b LM score, mean ± SD 7.5 ± ± ± ± ± ± 2.3 <0.001 b a By Fischer s exact test. b By 1-way ANOVA. A1 = asthmatic non-eosinophilic polyp; A2 = asthmatic eosinophilic polyp; ANOVA = analysis of variance; E/M = ethmoid/maxillary; LM = Lund-Mackay; NA1 = nonasthmatic non-eosinophilic polyp with older age; NA2 = non-asthmatic non-eosinophilic polyp with younger age; NA3 = non-asthmatic eosinophilic polyp with lower E/M ratio; NA4 = non-asthmatic eosinophilic polyp with higher E/M ratio; SD = standard deviation. interquartile range (IQR). Statistical analyses were performed using SPSS 22.0 (IBM Co., Armonk, NY, USA) and R software (R Foundation for Statistical Computing; Values of p <0.05 were considered statistically significant. Results General characteristics of patients A total of 375 patients were analyzed. The median age was 43.9 years (IQR 20.8 years), and the male-to-female ratio was approximately 2:1. The median follow-up period after the operation was 6.5 months (IQR 16.9 months). Among the patients, 12.3% (46/375) had comorbid asthma, while 39.5% (148/375) exhibited tissue eosinophilia. Cluster characterization Cluster analysis identified 6 clusters, which were labeled based on the most distinguishing features. Of these, 2 were asthmatic (A1 and A2) clusters, and 4 were nonasthmatic clusters (NA1 to NA4). Clusters were also easily discriminated by the presence of tissue eosinophilia: noneosinophilic polyp (A1, NA1, and NA2) and eosinophilic polyp (A2, NA3, and NA4). All variables used for clustering were significantly different among the 6 clusters (p < 0.001). Post hoc comparisons (Tukey method) were performed for additional labeling of non-asthmatic clusters by using age (NA1 vs NA2) and E/M ratio (NA3 vs NA4). The largest cluster was cluster NA1 (non-asthmatic noneosinophilic polyp with older age), which comprises 135 patients (36%) (Table 1; Fig. 1). Cluster NA1: non-asthmatic, non-eosinophilic polyp patients with older age Patients in this cluster did not exhibit comorbid asthma or tissue eosinophilia. The mean blood eosinophil proportion, FIGURE 1. Cluster composition. Six clusters have been identified. The largest cluster was NA1 (n = 135, 36.0%), followed by NA2 (n = 74, 19.7%). A1 = asthmatic, non-eosinophilic polyp; A2 = asthmatic eosinophilic polyp; E/M = ethmoid/maxillary ratio; NA1 = non-asthmatic non-eosinophilic polyp with older age; NA2 = non-asthmatic non-eosinophilic polyp with younger age; NA3 = non-asthmatic eosinophilic polyp with lower E/M ratio; NA4 = non-asthmatic eosinophilic polyp with higher E/M ratio. mean E/M ratio, and mean LM score were 3.2 ± 2.5%, 1.8 ± 0.7, and 7.5 ± 2.2, respectively. The mean age of patients in the NA1 cluster was 52.0 ± 10.9 years. Cluster NA2: non-asthmatic, non-eosinophilic polyp patients with younger age Cluster NA2 comprised 74 patients (19.7%). No patients had asthma or tissue eosinophilia. However, in contrast to cluster NA1, patients in the NA2 cluster were younger, with a mean age of 19.4 ± 6.6 years (p < 0.001). The mean blood eosinophil proportion, E/M ratio, and mean International Forum of Allergy & Rhinology, Vol. 9, No. 1, January

4 Kim et al. FIGURE 2. This decision tree was used to classify patients into 1 of the 6 clusters based on the presence of tissue eosinophilia, presence of asthma, age, and E/M ratio. The algorithm classified 95.4% of the patients into the preformed clusters. Tissue eosinophilia was considered when eosinophils accounted for >20% of the total inflammatory cells under magnification 400. E/M = ethmoid/maxillary ratio. LM score were 2.4 ± 2.0%, 1.6 ± 0.5, and 8.0 ± 2.0, respectively. TABLE 2. Confusion matrix for the performance of clinical algorithm (n = 375) Predicted Cluster NA3: non-asthmatic, eosinophilic polyp patients with lower E/M ratio Cluster NA3 comprised a total of 80 patients (21.3%). None of the patients in this cluster had any comorbid asthma but all showed tissue eosinophilia. The mean blood eosinophil proportion, mean E/M ratio, and mean LM score were 5.1 ± 2.1%, 1.9 ± 0.4, and 7.3 ± 1.3, respectively. The mean age was 45.1 ± 13.4 years. Cluster NA4: non-asthmatic, eosinophilic polyp patients with higher E/M ratio Cluster NA4 comprised 40 patients (10.7%) with a mean age of 42.5 ± 9.1 years. None of the patients had asthma, and all patients exhibited tissue eosinophilia. The mean blood eosinophil proportion, mean E/M ratio, and mean LM score were 7.9 ± 4.3%, 3.3 ± 0.8, and 9.2 ± 2.0, respectively. Mean E/M ratio was significantly higher in cluster NA4 compared to cluster NA3 (p < 0.001). Cluster A1: asthmatic and non-eosinophilic polyp patients Cluster A1 comprised 19 (5.1%) with a mean age of 53.8 ± 15.1 years. Patients in this cluster had asthma; however, only one patient showed tissue eosinophilia. Patients in this cluster had high mean blood eosinophil proportion (7.4 ± 6.5%), high E/M ratio (2.2 ± 0.8), and high LM score (8.8 ± 2.6). Actual NA1 NA2 NA3 NA4 A1 A2 NA NA NA NA A A A1 = asthmatic non-eosinophilic polyp; A2 = asthmatic eosinophilic polyp; E/M = ethmoid/maxillary; NA1 = non-asthmatic non-eosinophilic polyp with older age; NA2 = non-asthmatic non-eosinophilic polyp with younger age; NA3 = non-asthmatic eosinophilic polyp with lower E/M ratio; NA4 = non-asthmatic eosinophilic polyp with higher E/M ratio. Cluster A2: asthmatic and eosinophilic polyp patients Cluster A2 comprised 27 patients (7.2%) with asthma and tissue eosinophilia. The mean age was 49.1 ± 8.7 years. Patients in this cluster were characterized by high mean blood eosinophil proportion (7.8 ± 2.3%), high E/M ratio (2.9 ± 0.9), and high LM score (9.9 ± 2.3). Regression tree analysis Binary regression tree analysis using the selected variables was used to develop an algorithm that can categorize the patients into the appropriate clusters (Fig. 2). Patients were evaluated based on tissue eosinophilia, presence of comorbid asthma, age, and E/M ratio. The algorithm assigned 95.4% of patients to the predetermined clusters (Table 2). 82 International Forum of Allergy & Rhinology,Vol. 9, No. 1,January 2019

5 Cluster analysis of nasal polyps TABLE 3. Treatment outcomes per cluster NA1 NA2 NA3 NA4 A1 A2 p Patients having revision surgery, n (%) 0(0) 8 (10.8) 1 (1.3) 1 (2.5) 0(0) 4 (7.4) <0.001 a Polyp 0.2 ± ± ± ± ± ± 1.7 <0.001 b Edema 0.5 ± ± ± ± ± ± 1.8 <0.001 b Discharge 0.6 ± ± ± ± ± ± b Sum 1.3 ± ± ± ± ± ± 3.4 <0.001 b Follow-up (months), median (IQR) 4.7 (10.7) 8.1 (19.8) 7.1 (18.0) 5.8 (12.1) 17.9 (23.2) 15.3 (21.5) c a By Fischer s exact test. b By 1-way ANOVA. c By Kruskal-Wallis test. A1 = asthmatic non-eosinophilic polyp; A2 = asthmatic eosinophilic polyp; ANOVA = analysis of variance; E/M = ethmoid/maxillary; IQR = interquartile range; NA1 = non-asthmatic non-eosinophilic polyp with older age; NA2 = non-asthmatic non-eosinophilic polyp with younger age; NA3 = non-asthmatic eosinophilic polyp with lower E/M ratio; NA4 = non-asthmatic eosinophilic polyp with higher E/M ratio. Treatment outcomes Among our patients, 3.2% (12/375) underwent revision ESS during the follow-up period. During the follow-up period, the clusters showed significant differences in the number of revision cases and postoperative mlk scores (p < 0.05). Asthmatic clusters (A1 and A2) tended to have higher mlk scores compared to non-asthmatic clusters. Among non-asthmatic clusters, NA2 tended to have highest mlk scores. Treatment outcomes are summarized in Table 3. Kaplan-Meier survival analyses showed significant differences in 4-year revision surgery-free probabilities among the clusters (Fig. 3, Log-rank p value = 0.005). Overall actuarial 4-year revision free rate was 87.2%. Cluster NA1, which has the highest proportion in our cohort, was used as a reference for comparing the clusters because of its very low rate of revision surgery (Table 4). Cluster A2 (asthmatic eosinophilic polyp; revision-free rate = 66.7%, logrank p = 0.013), cluster NA4 (non-asthmatic eosinophilic polyp with higher E/M ratio; revision-free rate = 66.7%, log-rank p = 0.017), and cluster NA2 (non-asthmatic noneosinophilic polyp with younger age; revision-free rate = 70.9%, log-rank p = 0.001) exhibited higher risks of revision surgery relative to cluster NA1. Patients in cluster NA3 (non-asthmatic eosinophilic polyp with lower E/M ratio) showed a similar revision surgery rate to that of patients in cluster NA1 (4-year revision-free rate = 98.0%, log-rank p = 0.230). None of the patients in cluster A1 (asthmatic non-eosinophilic polyp) underwent revision surgery during the 4-year postoperative period. However, the 4-year revision-free rate of A1 was not significantly different compared to other clusters (log-rank p > 0.05) Discussion Cluster analysis is an unsupervised learning methodology that can integrate multiple variables to identify unique FIGURE 3. Kaplan-Meier survival probability curves for risk of revision surgery according to each cluster. Kaplan-Meier survival analyses showed significantly different 48-month revision surgery-free probability according to clusters (log-rank p value = 0.005). A1 = asthmatic, non-eosinophilic polyp; A2 = asthmatic eosinophilic polyp; E/M = ethmoid/maxillary ratio; NA1 = non-asthmatic non-eosinophilic polyp with older age; NA2 = non-asthmatic non-eosinophilic polyp with younger age; NA3 = non-asthmatic eosinophilic polyp with lower E/M ratio; NA4 = non-asthmatic eosinophilic polyp with higher E/M ratio. International Forum of Allergy & Rhinology, Vol. 9, No. 1, January

6 Kim et al. TABLE 4. Kaplan-Meier survival analysis of 4-year revision-free rates Cluster Four-year revision-free rate (%) Log-rank p a NA1 100 NA NA NA A1 100 A a Compared with NA1. A1 = asthmatic non-eosinophilic polyp; A2 = asthmatic eosinophilic polyp; E/M = ethmoid/maxillary; NA1 = non-asthmatic non-eosinophilic polyp with older age; NA2 = non-asthmatic non-eosinophilic polyp with younger age; NA3 = non-asthmatic eosinophilic polyp with lower E/M ratio; NA4 = non-asthmatic eosinophilic polyp with higher E/M ratio. patient categories, making it distinct from other approaches in analyzing multivariate data. 17 Several CRS cluster analysis studies have reported inflammatory endotypes (such as interleukin 4 [IL-4], interferon γ [IFN-γ ], IL-17A) or cellular phenotypes (eg, plasma cells, lymphocytes, eosinophils, and neutrophils) from the tissue itself. 7, 18 Cluster assignments were found to be correlated with the CRS phenotype (presence of polyps or asthma) or disease recurrence after operation. Despite the observed clinical relevance, the variables that are used for clustering in these studies are difficult to obtain in practice. Therefore, more clinically available parameters should be used for clustering. In the present study, we attempted to subcategorize patients with nasal polyps by cluster analysis. In particular, we used data that are easily accessible from the electronic health records of our institution to establish a model that utilizes routinely available clinical features. Tissue eosinophilia could also be routinely available from pathological reports because our institution has a reporting system of tissue eosinophilia from surgical specimens. Previous studies have indicated that the asthma, 19 tissue eosinophilia, 14, 20 and CT score 21 are significant factors associated with treatment outcomes. However, other studies reported contradictory results that suggested that asthma, 15 tissue eosinophilia, 11 and CT score 22 are not associated with treatment outcomes. The observed inconsistencies may be partly explained by the disease heterogeneity of CRSwNP. In the present study, unsupervised cluster analysis identified distinct subgroups using different combinations of these factors, and these designated subgroups were associated with different treatment outcomes. Therefore, unsupervised cluster analysis was determined to be suitable for analyzing heterogeneous disorders, such as asthma, chronic obstructive pulmonary disease, and obstructive sleep apnea. 23, 24 In the current study, we defined revision rate and endoscopic score as primary outcomes. Our result indicates that 6 clusters (2 asthmatic clusters and 4 non-asthmatic clusters) have different clinical outcomes after ESS. Even though cluster NA1 (non-asthmatic noneosinophilic polyp with older age) comprised the largest population of patients, none of the patients in this cluster underwent revision surgery during the follow-up period. Cluster NA3 (non-asthmatic eosinophilic polyp with lower E/M ratio) was the second largest group, with only 1 patient who underwent revision surgery. By contrast, clusters A2 (asthmatic eosinophilic polyp), NA2 (nonasthmatic non-eosinophilic polyp with younger age), and NA4 (non-asthmatic eosinophilic polyp with higher E/M ratio) showed considerably higher risks for revision surgery. Compared to cluster NA1, 4-year revision surgery free rates were significantly lower in clusters NA2 and NA4, whereas no significant difference existed in clusters NA3 and A1. Patients in both the asthmatic clusters A1 and A2 showed relatively high levels of blood eosinophilia. However, asthmatic eosinophilic polyps (A2) were more strongly associated with higher E/M ratio relative to asthmatic noneosinophilic polyps (A1) and tended to have higher rates of undergoing revision surgery. Therefore, the revision rate may depend on the tissue eosinophilia and the E/M ratio even in the presence of asthma, which also indicates eosinophilic inflammation. 10 There are 2 primary asthma endotypes characterized by high or low T-helper 2 (TH2) cell and type 2 innate lymphoid cell (ILC2) activity are termed T2 high and T2 low, respectively. 25 However, in the present study, the presence of asthma was based on the physician s diagnosis and was not further subclassified. If further asthma classification had been used for cluster analysis, the results might have been more distinct. In the present study, younger patients with non-asthmatic non-eosinophilic polyps (cluster NA2) showed higher rates of revision surgery. Thus, even in non-eosinophilic polyp cases, patients can be further categorized according to age. Age-related reduction in inflammation is known to occur in non-eosinophilic polyps, 26 which may explain the current findings. In this cluster, patients under 18 years comprised more than one-half of the total number of patients (39/74, 52.7%). However, most of the studies investigating nasal polyps are limited to individuals in the study population that are older than 18 years. CRSwNP is not common in children and sometimes accompanies cystic fibrosis. 27 However, cystic fibrosis has a very low prevalence in Korea 28 and cystic fibrosis patients were excluded in our cohort. Thus patients in cluster NA2, especially those aged below 18 years, comprised a distinct CRSwNP phenotype. In our study mlk scores of each clusters does not exactly correlate with the 4-year revision-free rate. NA1 and NA3, which had relatively high 4-year revision-free rates (100% and 98%, respectively) had lower mlk scores compared to NA2, NA4, and A2 clusters, which had relatively low 4-year revision-free rates. Although none of the patients in cluster A1 underwent revision surgery, the mlk scores were relatively higher than that of NA2 and NA4 and almost as equal as that of A2. Endoscopic evaluation had been performed at different times among clusters, and since patients 84 International Forum of Allergy & Rhinology,Vol. 9, No. 1,January 2019

7 Cluster analysis of nasal polyps in A1 tended to have been followed longer (probably due to the presence of their comorbid asthma), their endoscopic score might have been higher compared to others. In addition, despite the fact mlk score is known to be correlated with symptom outcome, still there seem to be only a minimal correlation between the endoscopic scores and symptom outcomes. Thus, a weighted combination of symptom domains and endoscopic attribute to improve this correlation may be required. 29 A1 had the smallest proportion (5.1%) in our cohort and therefore more patients should be recruited and evaluated to characterize this cluster. Considering that although more than 30% of patients showed mlk scores higher than 1, only 3.2% among all patients underwent revision surgery, a number that is lower than that of another study by Wynn and Har-El. 30 The median follow-up period in our patient cohort is somewhat shorter than other studies and this may be 1 of the reasons for such a low revision rate. In South Korea, patients are free to choose physicians and can go to any medical practitioner s office. Therefore there is no guaranteed followup for patients and this seems to be the weakness of a study from a single institution. Differences in ethnicity compared to the aforementioned study (Wynn and Har-El 30 ), which was conducted among a Western population, may be another reason for a different outcome in our study. It is known that polyps from Asian population demonstrate a Th2/Th1/Th17 mixed pattern whereas polyps from Western population are Th2-biased. 31 Our patient cohort demonstrated a mixture of cellular composition (39.5% vs 60.5% for eosinophilic polyp and non-eosinophilic polyp, respectively) reflecting a heterogeneous cytokine profile. However, the overall actuarial 4-year revision-free rate was 87.2%, suggesting that more patients may need revision surgery during a long-term follow-up. This is comparable or somewhat lower (ie, higher revision rate) than the results from a recently published study which states that 5-year actuarial revision surgery free rate is 90.1% for patients with nasal polyps. 6 The primary goal of treatment for nasal polyps is to achieve clinical control by improving symptoms and maintaining a healthy or almost healthy nasal mucosa. 4 Therefore, even if the polyps have recurred, they can be managed in most cases with medical therapies without requiring revision surgery. In severe cases, short-term treatment with systemic corticosteroids can help reduce the size of the recurred polyp, which can be managed by topical corticosteroids in order to control the patients symptoms. Tissue eosinophilia is currently 1 of the most easily available biomarkers or endotypes for CRS. However, the definition of tissue eosinophilia has not been standardized. 32 Using tissue eosinophilia as a biomarker for CRSwNP means that the cutoff value for defining tissue eosinophilia should be determined by the difference in clinical outcomes. Previous studies have defined the degree of tissue eosinophilia based on the eosinophil counts under high power field. 14, 20, 33 However, this procedure is difficult because all steps must be performed manually, and interdepartmental cooperation is required between otolaryngologists and pathologists. In addition, the degree of eosinophilia is known to have an uneven distribution throughout the tissue. 34 On the other hand, defining tissue eosinophilia based on the proportion of the total inflammatory cells, which was performed in the present study, can serve as an easier approach; however, this is a semiquantitative measurement that is prone to subjective judgment of pathologists. However, other clinical features, such as high blood eosinophilia, high LM score, or high E/M ratio, can be useful for effective categorization of patients into several clusters, which are associated with different clinical outcomes. The present study has some limitations. All data analyzed in the current study were retrospectively acquired from electronic health records, and all patients did not have the same follow-up schedule. The actuarial 4-year revision-free rates are estimations from Kaplan-Meier analysis and these estimations are most accurate at the time point when most patients are still present at the time of analysis (ie, 48 months in our study). 35 Since the median follow-up period in our study is less than 48 months, our estimation may carry some risk of error. Outcome measurements were based on endoscopic scores and rate of revision surgery, however, validated subjective symptom outcomes had not been measured. The indication of revision surgeries would have been consistent within our center; however, the actual procedure is also dependent on the patient s will. Given the above limitations, more detailed and larger study cohorts with multi-institutional prospective studies are warranted for cluster analysis and validation of outcomes. Conclusion In the present study, cluster analysis identified 2 asthmatic and 4 non-asthmatic clusters among CRSwNP patients. Each cluster exhibited distinct characteristics and outcomes. Cluster analysis using routinely available markers that are easily accessible from electronic health records can help clinicians predict patient outcomes. The current clustering algorithm warrants further validation through a prospective longitudinal study design. References 1. Blackwell DL, Lucas JW, Clarke TC. Summary health statistics for U.S. adults: national health interview survey, Vital Health Stat ;(260): Ahn JC, Kim JW, Lee CH, Rhee CS. Prevalence and risk factors of chronic rhinosinusitus, allergic rhinitis, and nasal septal deviation: results of the Korean National Health and Nutrition Survey JAMA Otolaryngol Head Neck Surg. 2016;142: Chapurin N, Pynnonen MA, Roberts R, et al. CHEER national study of chronic rhinosinusitis practice patterns: disease comorbidities and factors associated with surgery. Otolaryngol Head Neck Surg. 2017;156: Fokkens WJ, Lund VJ, Mullol J, et al. EPOS 2012: European position paper on rhinosinusitis and nasal polyps A summary for otorhinolaryngologists. Rhinology. 2012;50: Akdis CA, Bachert C, Cingi C, et al. Endotypes and phenotypes of chronic rhinosinusitis: a PRACTALL document of the European Academy of Allergy and International Forum of Allergy & Rhinology, Vol. 9, No. 1, January

8 Kim et al. Clinical Immunology and the American Academy of Allergy, Asthma & Immunology. J Allergy Clin Immunol. 2013;131: Stein NR, Jafari A, DeConde AS. Revision rates and time to revision following endoscopic sinus surgery: a large database analysis. Laryngoscope. 2018;128: Tomassen P, Vandeplas G, Van Zele T, et al. Inflammatory endotypes of chronic rhinosinusitis based on cluster analysis of biomarkers. J Allergy Clin Immunol. 2016;137: Stammberger H, Posawetz W. Functional endoscopic sinus surgery. Concept, indications and results of the Messerklinger technique. Eur Arch Otorhinolaryngol. 1990;247: Lund VJ, Kennedy DW. Staging for rhinosinusitis. Otolaryngol Head Neck Surg. 1997;117(3 Pt 2):S35> S Meng Y, Lou H, Wang C, Zhang L. Predictive significance of computed tomography in eosinophilic chronic rhinosinusitis with nasal polyps. Int Forum Allergy Rhinol. 2016;6: Kim SY, Park JH, Rhee CS, Chung JH, Kim JW. Does eosinophilic inflammation affect the outcome of endoscopic sinus surgery in chronic rhinosinusitis in Koreans? Am J Rhinol Allergy. 2013;27:e166 e Psaltis AJ, Li G, Vaezeafshar R, Cho KS, Hwang PH. Modification of the Lund-Kennedy endoscopic scoring system improves its reliability and correlation with patient-reported outcome measures. Laryngoscope. 2014;124: Stewart MG, Donovan DT, Parke RB RB Jr, Bautista MH. Does the severity of sinus computed tomography findings predict outcome in chronic sinusitis? Otolaryngol Head Neck Surg. 2000;123(1 Pt 1): Tokunaga T, Sakashita M, Haruna T, et al. Novel scoring system and algorithm for classifying chronic rhinosinusitis: the JESREC Study. Allergy. 2015;70: Veloso-Teles R, Cerejeira R. Endoscopic sinus surgery for chronic rhinosinusitis with nasal polyps: clinical outcome and predictive factors of recurrence. Am J Rhinol Allergy. 2017;31: Norušis MJ. Cluster analysis. In: PASW Statistics 18 Statistical Procedures Companion. Upper Saddle River, NJ: Prentice Hall; 2010: Weatherall M, Shirtcliffe P, Travers J, Beasley R. Use of cluster analysis to define COPD phenotypes. Eur Respir J. 2010;36: Lou H, Meng Y, Piao Y, et al. Cellular phenotyping of chronic rhinosinusitis with nasal polyps. Rhinology. 2016;54: Mendelsohn D, Jeremic G, Wright ED, Rotenberg BW. Revision rates after endoscopic sinus surgery: a recurrence analysis. Ann Otol Rhinol Laryngol. 2011;120: Lou H, Meng Y, Piao Y, Wang C, Zhang L, Bachert C. Predictive significance of tissue eosinophilia for nasal polyp recurrence in the Chinese population. Am J Rhinol Allergy. 2015;29: Grgić MV,Ćupić H, Kalogjera L, Baudoin T. Surgical treatment for nasal polyposis: predictors of outcome. Eur Arch Otorhinolaryngol. 2015;272: Smith TL, Litvack JR, Hwang PH, et al. Determinants of outcomes of sinus surgery: a multi-institutional prospective cohort study. Otolaryngol Head Neck Surg. 2010;142: Moore WC, Meyers DA, Wenzel SE, et al. Identification of asthma phenotypes using cluster analysis in the Severe Asthma Research Program. Am J Respir Crit Care Med. 2010;181: Zinchuk AV, Jeon S, Koo BB, et al. Polysomnographic phenotypes and their cardiovascular implications in obstructive sleep apnoea. Thorax. 2018;73: Wenzel SE. Asthma phenotypes: the evolution from clinical to molecular approaches. Nat Med. 2012;18: Kim DW, Kim DK, Jo A, et al. Age-related decline of neutrophilic inflammation is associated with better postoperative prognosis in non-eosinophilic nasal polyps. PLoS One. 2016;11:e Triglia JM, Nicollas R. Nasal and sinus polyposis in children. Laryngoscope. 1997;107: Jung H, Ki CS, Koh WJ, et al. Heterogeneous spectrum of CFTR gene mutations in Korean patients with cystic fibrosis. Korean J Lab Med. 2011;31: DeConde AS, Bodner TE, Mace JC, Alt JA, Rudmik L, Smith TL. Development of a clinically relevant endoscopic grading system for chronic rhinosinusitis using canonical correlation analysis. Int Forum Allergy Rhinol. 2016;6: Wynn R, Har-El G. Recurrence rates after endoscopic sinus surgery for massive sinus polyposis. Laryngoscope. 2004;114: Wang X, Zhang N, Bo M, et al. Diversity of T H cytokine profiles in patients with chronic rhinosinusitis: a multicenter study in Europe, Asia, and Oceania. J Allergy Clin Immunol. 2016;138: Cho SW, Kim DW, Kim JW, Lee CH, Rhee CS. Classification of chronic rhinosinusitis according to a nasal polyp and tissue eosinophilia: limitation of current classification system for Asian population. Asia Pac Allergy. 2017;7: Soler ZM, Sauer D, Mace J, Smith TL. Impact of mucosal eosinophilia and nasal polyposis on quality-oflife outcomes after sinus surgery. Otolaryngol Head Neck Surg. 2010;142: Bachert C, Akdis CA. Phenotypes and emerging endotypes of chronic rhinosinusitis. J Allergy Clin Immunol Pract. 2016;4: Rich JT, Neely JG, Paniello RC, Voelker CC, Nussenbaum B, Wang EW. A practical guide to understanding Kaplan-Meier curves. Otolaryngol Head Neck Surg. 2010;143: International Forum of Allergy & Rhinology,Vol. 9, No. 1,January 2019