|Home | About | Journals | Submit | Contact Us | Français|
Current clinical classifications of chronic rhinosinusitis (CRS) have weak prognostic utility regarding treatment outcomes. Simplified discriminant analysis based upon unsupervised clustering has identified novel phenotypic subgroups in CRS, but prognostic utility is unknown.
To determine if discriminant analysis allows prognostication in patients choosing surgery versus continued medical management.
A multi-institutional, prospective study of patients with CRS who failed initial medical therapy, then self-selected continued medical management or surgical treatment was used to separate patients into 5 clusters based upon a previously described discriminant analysis using total SinoNasal Outcome Test (SNOT-22), age and missed productivity. Patients completed the SNOT-22 at baseline and for 18 months of follow-up. Baseline demographic and objective measures included olfactory testing, computed tomography and endoscopy scoring. SNOT-22 outcomes for surgical versus continued medical treatment were compared across clusters.
Data was available on 690 patients. Baseline differences in demographics, comorbidities, objective disease measures and patient reported outcomes were similar to previous clustering reports. Three of five clusters identified by discriminant analysis had improved SNOT-22 outcomes with surgical intervention when compared to continued medical management (surgery a mean of 21.2 points better across these 3 clusters at 6 months, p<0.05). These differences were sustained at 18 months of follow up. Two of five clusters had similar outcomes when comparing surgery to continued medical management.
A simplified discriminant analysis based upon three common clinical variables is able to cluster patients and provide prognostic information regarding surgical treatment versus continued medical management in patients with CRS.
Chronic rhinosinusitis (CRS) is a common chronic disease characterized by ongoing inflammation of the sinonasal mucosa. It significantly impacts individual quality of life (QOL) and personal productivity, with substantial direct and indirect costs to society. The typical treatment paradigm for CRS involves comprehensive medical therapy to reduce inflammation, eliminate pathogenic bacteria if present, and physically wash away mucus. Although many patients adequately respond to initial medical therapy, those with persistent disease despite medical therapy may be offered sinus surgery. Previous reports have suggested that patients who fail medical therapy and undergo sinus surgery have better outcomes compared to those who continue only with medical therapy(1–3). However, these studies typically report average outcomes for an entire cohort and it remains possible that certain subsets of patients may derive more benefit from a particular treatment strategy than others. Understanding which patients are most likely to improve following surgery and which are better served with continued medical management would inform patient, clinician, and policy-level decision-making.
Diagnostic criteria for CRS require persistence of cardinal sinonasal symptoms for a minimum of 12 weeks and objective evidence of mucosal inflammation(4). Although all patients fulfilling these criteria would be considered to have CRS, it is widely believed that subsets of disease exist and may contribute to inconsistent responses to treatment and variable long-term clinical outcomes. We recently reported a study wherein unsupervised hierarchical clustering was used to separate a cohort of patients with CRS into clusters using baseline disease characteristics(5). Clusters differed across a number of features, with patient-reported outcome measures driving much of the separation among clusters. Although usage of medications was found to differ among clusters, the study was cross-sectional in nature and thus was not designed to determine if treatment outcomes differ among clusters. The goal of this follow-up study was to determine whether this clustering algorithm, when applied to a larger prospective cohort, would identify differences in outcomes between those treated surgically and those who continue only with medical therapy.
Data was derived from a multi-institutional, prospective cohort study. Study participants were recruited from 4 tertiary medical centers across North America (Medical University of South Carolina, Stanford University, University of Calgary, and Oregon Health and Sciences University). Each participant had CRS as defined by consensus criteria, with at least 12 weeks of cardinal symptoms and objective evidence of sinonasal inflammation on endoscopy or imaging. Patients were enrolled at the point in time wherein they were considered to have failed appropriate medical management and were offered endoscopic sinus surgery (ESS) as a treatment option. Failure of medical management required ongoing cardinal symptoms despite ≥2 weeks of broad spectrum or culture-directed antibiotics, ≥5 days of oral steroids, and ≥1 month of topical steroids, with most patients exceeding minimums of prior treatment before surgery was considered. As per real-world clinical practice, patients were offered surgery, but allowed to choose either to continue solely with medical management or to have ESS.
All subjects provided written informed consent to participate in this observational study (Clinicaltrials# NCT00799097; NIH: R01 DC005805; Pro 12409). Study coordinators administered baseline questionnaires that assessed demographic information (age, gender, race) and medical comorbidities. Patients were considered to have medical comorbidities if they had been given a diagnosis by a physician in the past (asthma, aspirin sensitivity, chronic obstructive pulmonary disease, obstructive sleep apnea (OSA), fibromyalgia, depression) or based on patient self-report (current smoking, alcohol intake, prior sinus surgery). Patients were considered to have allergic rhinitis based on prior positive skin prick test or allergen-specific IgE antibody test. Allergic fungal rhinosinusitis (AFRS) was defined using classic Bent-Kuhn criteria (6). At baseline, objective olfactory function was evaluated using the Brief Smell Identification Test (B-SIT). Sinonasal endoscopy was performed by the treating physician and scored using the Lund-Kennedy scale, with presence of absence of polyps recorded. All patients had computed tomography (CT) scans performed after maximal medical therapy, prior to enrollment and these were scored using the Lund-Mackay system (4). Physicians scoring sinonasal endoscopy and CT scans were blinded to all other patient-reported variables for the study duration.
Patients chose to either continue solely with medical therapy or undergo ESS at baseline enrollment. Because this was an observational study, specifics of ongoing medical therapy for either group were not dictated by study protocol, but occurred in a real-world fashion and were aimed at maximally-controlling disease in line with patient preferences. For those undergoing surgery, the specific procedure performed was left to the discretion of the treating surgeon based on CT findings and overall clinical picture, as occurs in everyday clinical practice. All surgery was done by fellowship trained rhinologists in accordance with commonly accepted principles of functional ESS(4).
Sinus-specific QOL was the primary outcome and was assessed first at baseline and then again at 6 months, 12 months, and 18 months after study enrollment. Sinus-specific QOL was assessed using two different instruments, the 22-item Sino-Nasal Outcome Test (SNOT-22) and Rhinosinusitis Disability Index (RSDI). The SNOT-22 is a validated CRS-specific QOL instrument containing 22 questions (total score range 0–110), with higher scores representing more severe impact. Total scores and individual domains were evaluated, including rhinologic, extranasal-rhinologic, psychiatric, ear/facial, and sleep domains as has been described previously(7). The RSDI is a validated 30-question survey comprised of three individual subscales to measure the impact of sinus disease on the physical, functional, and emotional domains (range: 0–120), with higher RSDI total and subscale scores representing a greater impact of disease. These instruments were chosen because they quantify the impact of each symptom specific to CRS, namely nasal congestion, nasal drainage, facial pain, and olfactory disturbance, as well as “extra-rhinologic” manifestations of disease. These instruments are used worldwide and have been shown to discriminate between subjects with CRS and healthy controls as well as identify significant differences after medical and surgical treatments(8). Sleep quality was assessed using the Pittsburgh Sleep Quality Index (PSQI).(9) Depressed mood and anhedonia were assessed using the Patient Health Questionnaire-2 (PHQ-2).(10) Patients were also asked how many days in the last 90 days they missed work.
The input variables and hierarchical clustering methods utilized in the original cohort of 382 patients to derive clusters have been previously reported in detail(5). Briefly, 103 variables encompassing demographic, comorbidity, objective CRS metrics and patient reported outcome measures were reduced to meaningful factors with a high degree of correlation. This permitted us to reduce 103 variables to 32. Ward’s minimum-variance hierarchical method was then used to perform the cluster analysis. This analysis places subjects into groups, or clusters, suggested by the data, not defined a priori and then simple discriminant analysis is performed to obtain simple algorithms for clinical use and future studies.
Cluster analysis was not performed again in this study. Rather, we used a discriminant analysis and simple clinical algorithm based upon this clustering technique to classify patients into the 5 statistical clusters with a high likelihood of success. The clinical algorithm was utilized to place each patient from the current expanded cohort into one of five clusters, based on age, SNOT-22 score, and lost productivity over 90 days (Clusters 1-5) (Figure 1).
Descriptive statistics are presented for all demographic, comorbidity/exposure, and CRS severity measures. Differences between clusters at baseline upon entry into the study were assessed using Chi-square tests and analysis of variance (ANOVA) for categorical and continuous measures, respectively. To assess differences between medical and surgical treatments, post-hoc t-tests were performed within each cluster based on linear mixed effects models which controlled for site and baseline QOL. The normality was examined in each analysis. When normality was questionable appropriate transformations were considered. Logistic regression was then used to assess the odds of achieving a minimal clinically important difference (MCID) for each QOL measure after either medical or surgical treatments, controlling for baseline value and follow-up duration. An MCID was defined as an improvement from baseline to last follow-up of 9 or greater for SNOT22 or 10.3 or greater for RSDI as previously defined(8). Statistical significance was assessed at alpha = 0.05. For the secondary objective on likelihood of achieving an MCID, statistical significance was adjusted using Bonferroni correction for multiple comparisons for simultaneously testing 10 hypotheses (5 clusters and two outcomes per cluster). All analyses were performed using SAS 9.4 (c).
The study cohort included 690 patients, ranging in age from 18 to 86 years. A total of 376 patients overlapped with the original cross-sectional cohort and 314 were newly enrolled. Baseline differences in demographics, comorbidities, objective disease measures, and patient-reported disease severity are shown in Table 1. The patients were equally split between genders and the average age was 58.0 years (SD = 15.9). Overall, 42% (289/690) of patients had undergone prior sinus surgery and comorbidities, such as allergy (25%, 172/690), asthma (37%, 255/690), and nasal polyps (37%, 254/690) were common. As expected and previously reported (5, 11), differences existed between clusters for demographics, comorbidities/exposures, CRS severity measures, and patient-reported outcome measures (PROMs) similar to what was previously reported for the smaller cross-sectional cohort. Given the larger sample size of the current cohort, power to detect differences increased and additional baseline variables showed significant difference, most notably olfactory function (BSIT).
Cluster 1 was comprised of 87 patients of whom 56 underwent surgery and 31 continued medical management. This group has the highest prevalence of males (69%), was the oldest (65 years) and tended to have moderate CRS severity measures. At 6 months, the surgical group had significantly improved QOL on both the SNOT-22 and RSDI total scores compared to the medical group (Figure 2). However, this difference was no longer detectable at 12 months and by 18 months scores on both instruments were equivalent between treatment groups. There were no long-term differences in SNOT-22 domain scores or RSDI subscale scores between treatment groups (Figures 3 and and4).4). The odds of achieving MCID were not significantly different between groups for either the SNOT-22 (OR=2.0; 99.5% CI 0.6-7.0) or RSDI instruments (OR=3.0; 99.5% CI: 0.7-13.4) (Table 2).
Cluster 2 (N=233) was the largest, with 182 patients undergoing surgery and 51 continued medical management. This group was split between genders and tended to be older (62 years) with some of the most severe CRS severity measures. The surgical group had significantly better SNOT-22 scores as 6, 12, and 18 month time points, with similar robust differences for rhinologic and extranasal-rhinologic SNOT-22 domains. The RSDI total score was significantly better in the surgical group at 6 months, but not at other time points. However, the RSDI physical score was significantly better at both 6 months and 18 months in the surgical group as compared to the medical group. The odds of achieving MCID were higher in the surgical group for both the SNOT-22 (OR=4.7; 99.5% CI: 1.5-15.4) and RSDI instruments (OR=5.5; 99.5% CI: 1.2-19.8).
Cluster 3 was the second largest cluster with 215 total patients, of whom 161 elected surgery and 54 continued with solely medical management. This group was split between genders, was the youngest (37 years) and had the least severe CRS severity measures. The surgical group had significantly better QOL at 6, 12, and 18 month time points compared to the medical group based upon both the SNOT-22 and RSDI total scores. Significant differences were also seen at every time point for the rhinologic, extranasal-rhinologic, and ear/facial domains of the SNOT-22 instrument, as well as RSDI physical, functional and emotional subscores. Those patient in Cluster 3 had the highest odds of achieving an MCID with surgery as compared to medical treatment for both the SNOT-22 (OR=16.9; 99.5% CI: 3.7-77.5) and RSDI (OR=7.8; 99.5% CI: 2.2-28.6).
A total of 94 patients were classified into Cluster 4, with 76 choosing to undergo surgery and 18 continuing solely with medical management. This group had 2/3 females, was relatively young (38 years) and had the worse CT and endoscopy measures. Improvements in QOL were seen in both patients undergoing surgery and in those continuing with medical management. However, those undergoing surgery had significantly better total SNOT-22 scores at 6, 12, and 18 months, as well as significantly better total RSDI scores at 6 and 18 months. SNOT-22 differences were driven mainly by improvements in rhinologic and ear/facial domains in the surgical group. RSDI differences were most notable in the physical subscore which was better in the surgical group at every follow-up time point. Although the surgical group had a greater absolute improvement, the odds of at least achieving an MCID did not reach statistical significance for either the SNOT-22(OR=6.2; 99.5% CI 0.3-152.7) or RSDI instruments (OR=4.0; 99.5% CI 0.4-36.0).
Just under 9% of the cohort was classified into Cluster 5 (n=61) with 50 undergoing surgery and 11 continuing with only medical treatments. This group had the highest prevalence of females (72%), tended to be of average age (50 years) and had slightly worse than average CRS severity measures. Both medical and surgical groups appeared to have initial improvement at 6 months, with progressive worsening at 12 and 18 months. There were no differences between the surgical and medical groups for the SNOT-22 total score or any individual domain. Similarly, the RSDI total scores were not different between treatment groups, although physical subscores were better in the surgical group at 6 and 18 months. The odds of achieving at least one MCID were not different between groups for either the SNOT-22 (OR=4.9; 99.5% CI 0.2-119.8) or RSDI instruments (OR=3.9; 99.5% CI: 0.2-99.4).
Overall we had follow up on 59% of patients at a minimum of one timepoint. Follow up rate by cluster and by timepoint is shown in Table 3. There were no statistically significant differences in follow up rate between clusters. The mixed effects model we used is designed to account for lost follow, however, In order to confirm that lost follow up did not impact our findings, we also analyzed total SNOT22 comparing medical and surgical outcomes at each timepoint for each cluster only for patients with at least one follow up and this did not change our findings reported above.
Although sinus surgery results in significant QOL improvements over time, most studies report mean values for an entire cohort. Prior studies demonstrate that approximately 75–80% of patients achieve an MCID after ESS(11), but there are few traditional clinical factors that reliably predict surgical success or magnitude of QOL improvement for an individual patient. In this study, a large cohort of patients with CRS was divided into clusters and outcomes were compared between those undergoing ESS and those continuing solely with medical management. Distinct differences in outcomes were seen between clusters, with Clusters 1 and 5 failing to show added benefit with surgery, whereas robust advantages were seen with surgical treatment in Clusters 2, 3, and 4 with the greatest odds in Cluster 3. These differences in outcomes were confirmed using two different sinus-specific QOL instruments, the SNOT-22 and RSDI. These two QOL instruments have previously been shown to have a high degree of correlation(12), but given some variation in questions and differing constructs it was possible that outcomes could differ. As demonstrated in Figure 2, SNOT22 and RSDI results correlated with rare exceptions at select timepoints in clusters 2 and 4, supporting the concept of cluster prognostication for CRS QOL with 2 separate instruments. Interestingly, another benefit noted in this study was that the most symptomatic patients (clusters 4 and 5) had similar baseline SNOT-22 scores and traditional classifications may have grouped these patients together. However, discriminant algorithms are able to separate those more likely to benefit from surgery (cluster 4) from those that will probably not receive as much benefit from surgery (cluster 5).
A simple discriminant analysis with clinical algorithm (Figure 1) based upon our prior report(5) was utilized in this study to place each patient into a cluster at baseline. Utilization of this clinical algorithm, as opposed to repeating an entirely new statistical clustering, is crucial for the clinical applicability of these findings to future patients. In our previous study, this clinical algorithm placed patients into the correct cluster 89.4% of the time(5). Importantly, the clusters generated in this expanded cohort closely mirror those seen in our original study, with respect to overall proportions and qualitative clinical characteristics. The only differences were quantitative, namely some variables became significant across groups owing to larger sample size and increased power (ie BSIT). This provides further validation of our prior discriminant analysis, suggesting that the simple clinical algorithm proposed, with just 3 variables, accurately replicates the original statistical clustering based on over 100 variables. It is important to note that in addition to differences in the three variables used in the discriminant analysis (total SNOT-22, age and missed productivity) there are many other baseline differences among clusters as seen in Table 1. While statistically significant, it remains to be determined if these differences are clinically relevant. These three variables are probably best thought of as proxies for a constellation of many other clinical characteristics which collectively describe an individual cluster and theoretically are influencing outcomes.
There are several reasons to think these results may prove generalizable across different patient populations. The first reason is that the original clustering methods were generated with a preliminary cohort of patients, whereas the outcome assessment reported herein utilizes a much larger cohort providing some degree of replication, although not fully independent. Additionally, patients were enrolled from four centers across North America, with multiple participating surgeons, diverse geography, and varied patient populations. Differences are thus less likely to be driven by a single center, individual surgeon, or specific patient population. The sample size is also large relative to most other prospective studies examining treatment outcomes in patients with CRS. Ultimately, widespread generalizability will need to be assessed using additional patient cohorts. As shown in Figure 1, our simple algorithm could potentially be used in clinical practice to aid in individualized decision making. However, at the present time findings should be limited to the patient population studied – namely, a heterogenous group of patients with CRS that present to tertiary centers who are surgical candidates after failing appropriate medical therapy. In our study this was defined as ≥2 weeks of broad spectrum or culture-directed antibiotics, ≥5 days of oral steroids, and ≥1 month of topical steroids. These findings should not be extrapolated to patients who present de novo with untreated CRS or to those treated with techniques other than traditional ESS. Rather this analysis is designed to serve as a foundation for future, more widespread, multi-institutional studies.
A limitation of this analysis is that patients self-selected treatment rather than being randomly allocated into groups. The lack of randomization means that unmeasured confounding can never be fully ruled out and could account for some differences seen between groups. Although one can never fully rule out this possibility, the clustering algorithm does result in near homogeneity within clusters for all input variables. Thus, within each cluster, there were no notable differences between medical and surgical treatment groups for objective disease severity measures, medical comorbidities, or baseline QOL measures. Additionally, the outcome analysis was controlled for enrollment site and length of follow-up, further reducing the risks of selection bias and follow-up bias respectively. Although a blinded randomized clinical trial would be ideal, the feasibility of such a trial would be low given patient reluctance to enroll in randomized surgical trials, ethical concerns of sham surgery, and costs involved in long-term assessments. Other study designs would also be possible, including complete standardization of medical therapy to include dosing and selection of antibiotic, oral steroid and topical steroid, standardization of surgical approach, technique and equipment and strict inclusion criteria for subtypes of CRS to enroll. While these designs would be ideal, unfortunately there is limited evidence to support specific standardization of many of these approaches and individual practices vary widely from physician to physician.
When conducting prospective outcomes research there is always concern regarding patients lost to follow up and their impact upon the findings of the study. The overall follow up of 59% at any timepoint in our study compares favorably to other cluster analyses. Prior asthma studies examining therapeutic outcomes after clustering did not explicitly state follow up rate, but from extrapolation of the data it appears to be between 49 and 72% depending upon the outcome variable and timepoint (13). Other studies containing surgical cohorts range from a low of 48% at 6 months in a single institution series (14) to a high of 63% at 12 months in a multi-institutional study previously published by our group (15). Rather than collapsing all follow up to one timepoint, we wished to examine the long term durability of our findings, realizing that at 18 months our follow up rate would continue to drop. Despite this, we still found that many of the differences identified at 6 months persisted to 18 months. The mixed effects modeling used in this study is the statistical approach ideal for longitudinal analysis with incomplete follow up (16), but in order to confirm our findings, we analyzed results using only patients with follow up at any timepoint and our conclusions were identical.
The question remains as to how these findings should impact clinical decision-making. All patients in this study were appropriate surgical candidates based on the fact that they had CRS by accepted diagnostic criteria and remained symptomatic despite appropriate medical therapy. Findings from this study suggest that those patients falling into Clusters 2, 3, and 4 have a significantly increased odds of durable improvement with surgery as opposed to continuing solely with medication, whereas those patients fitting into Clusters 1 and 5 may not. Although this data suggests certain patients may not derive additional benefit from surgery, utilization of these findings to withhold surgery for patients in Clusters 1 and 5 seems extreme based on these preliminary findings alone. Rather, this information may prove useful in patient counseling as to possible outcomes. For example, cluster 5 patients do improve with surgery, but the magnitude of improvement is similar to continued medical management. Thus other factors, such as medication side effects, may lead a cluster 5 patient to opt for surgery. Similarly, a cluster 3 patient with significant medical co-morbidities or an inability to miss work/school may opt for continued medical management despite greater odds of improvement with surgery. It is also important to point out that sample sizes varied between clusters, with Cluster 5 being the smallest and that clusters with smaller numbers may be more affected by patients lost to follow up over the 18 month study duration, although there was no difference in follow-up rates across clusters or between treatment interventions within clusters. Thus, it remains possible that a type II error exists within this group owing to lack of power. Future study with larger sample size, particularly with regard to medical treatment would help refine estimates for this particular cluster.
Sinus-specific QOL is an important metric for CRS outcome studies, but it should be remembered that these measures are not the only outcome of interest and may not be the sole driving force behind pursuing surgery in all patients. Some patients may elect surgery to alleviate a specific symptom, such as olfactory loss, which is not a major contributor to QOL as measured by the SNOT-22 or RSDI instruments. Other examples include orbital/skull base erosion related to disease burden, or a desire to transition from systemic to topical medication administration. With these considerations in mind, strict adherence to this clustering algorithm to guide surgical decision-making should not supersede the specific needs of each individual patient. Future studies might explore additional outcome measures such as medication usage, personal productivity, and medical costs.
It remains quite likely that inclusion of additional cluster-defining measures would further refine our cluster analysis. Clinicians inherently feel that classifications such as revision surgery status, the presence of polyps or a number of other variables impact the presentation and outcomes of CRS patients. Unfortunately, traditional classifications and CRS severity metrics such as Lund Kennedy endoscopy score and Lund McKay CT scores have not been overly useful in predicting surgical success(11). We did analyze our data controlling for revision surgery status. Total SNOT22 results were identical across all clusters and at all timepoints to data presented in this manuscript. This demonstrates that the presented clustering algorithm adequately accounts for this variability without the need to directly classify patients based upon this multitude of traditional clinical variables. Development of more precise clinical metrics, such as volumetric CT analysis or modified endoscopy scales, and refined phenotypic subclassifications may be useful in defining clusters or in developing discriminant algorithms. In addition, while we included over 100 clinical variables in our original cluster analysis, there may be unknown key clinical variables that are missing or as yet unknown. Also notably absent from our original study was inclusion of histologic or biomolecular markers. A wide array of possible measures could be included in the future, such as eosinophil levels, cytokine profiles, genotypes, or measures of the local microbiome among others. Inclusion of these measures is likely necessary for clusters to begin to mirror underlying endotypes, ie subgroups differentiated based on underlying pathophysiologic mechanisms. Certainly, identifying discrete endotypes will be crucial in order to develop and assess future targeted therapeutics. The current study can therefore be considered proof of concept that clustering methods can be utilized to predict outcomes, but should serve as a starting rather than an ending point for future research.
As mentioned above, there are numerous implications from our study. Our discriminant analysis relied upon three variables – total SNOT22, days of missed productivity and age. While age is readily available, SNOT22 and missed productivity may not be collected during routine clinical practice. It is not expected that these 3 variables will immediately be incorporated into all practices, but rather serve as a starting point for further development of clustering algorithms and prognostic studies using these and additional clinical variables. Furthermore, while practitioners can begin to use this information to better counsel patients, they must do so within the limitations of our study acknowledged above. Namely, that these findings may not apply to all CRS patients. It is hoped that this analysis is replicated by other centers around the world, applied to other patient populations and further refinement in clinical and bioimarker variables is pursued. Another important area for further study is understanding why certain clusters respond better or worse to a given therapeutic modality, for example, why cluster 3 appears to have better surgical outcomes.
Hierarchical clustering of CRS patients was reproducible and use of a novel discriminant analysis provides some prognostic utility in determining patients most likely to benefit from surgical therapy. It can serve as an aid in counseling CRS patients regarding treatment choices and likely outcomes, but is not a substitute for physician recommendations that take into account individual patient differences or preferences. Future refinements in clinical metrics and inclusion of biomarkers will likely further improve our prognostic ability and aid us in individualizing treatment algorithms.
Cluster analysis and the resulting simple discriminant algorithms may improve therapeutic recommendations for patients with CRS.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Conflict(s) of Interest / Financial Disclosures: Zachary M. Soler and Timothy L. Smith are supported for this investigation by a grant from the National Institute on Deafness and Other Communication Disorders (NIDCD), one of the National Institutes of Health, Bethesda, MD (R01 DC005805; PI/PD: TL Smith). Public clinical trial registration (http://www.clinicaltrials.gov) #NCT01332136. Zachary M. Soler is also supported for this investigation by another grant from the NIDCD (R03 DC013651-01). Timothy L. Smith is a consultant for IntersectENT, which is not associated with this manuscript. Zachary M. Soler is a consultant for Olympus, which is not affiliated with this manuscript. Rodney J. Schlosser is supported by grants from OptiNose and IntersectENT, neither are associated with this manuscript. Dr. Schlosser is also a consultant for Olympus and Arrinex which are not affiliated with this study.