IntroductionThere are few prior longitudinal studies of nasal and sinus symptoms (NSS) and their acute exacerbation (AENSS) in general population samples and no standardized approaches to measurement of AENSS in epidemiologic studies. NSS are common to multiple health conditions, can be relapsing and remitting, can become chronic as in the case of chronic rhinosinusitis (CRS), and have a significant individual and population impact (1–8). The European Position Paper on Rhinosinusitis and Nasal Polyps (EPOS) has operationalized a clinical definition of CRS, requiring both subjective symptoms which must be present for 12 continuous weeks and objective confirmation of sinonasal mucosal inflammation (e.g. via sinus computed tomography [CT]). For epidemiologic studies, EPOS only requires the presence of subjective symptoms (we designate as EPOSS) (1, 2).
Difficulties in obtaining objective evidence of inflammation have been an impediment to large-scale, population-based epidemiologic studies. Depending on individual characteristics, onset, duration, and season, the sudden onset or worsening of NSS could be an indication for allergic rhinitis (AR), acute rhinosinusitis (ARS), an acute exacerbation of chronic rhinosinusitis (AECRS), or other related diagnoses. Published studies of exacerbation among CRS patients have primarily focused on bacteriology (9–11), immunology (12, 13), and medical treatments (14–17), as opposed to population-based occurrence, severity, risk factors, and natural history. The International Consensus Statement on Allergy and Rhinology (ICAR), therefore, has declared a need for prevalence estimates of AECRS and more prospective studies, especially those which compare several definitions of AECRS (2).
As such, the objectives of this study were to evaluate and compare seasonal prevalence of AENSS by EPOSS CRS status (hereafter CRS status) across three definitions of AENSS; describe NSS severity by CRS and AENSS status; and identify self-reported individual characteristics associated with AENSS by CRS status. We addressed these objectives in a population-based longitudinal study using a sample of primary care patients from Geisinger who are representative of the general population in the area of central and northeastern Pennsylvania.
Materials and MethodsStudy overviewDetails of the study design have been published elsewhere (5, 18). Briefly, in 2014, adult (at least 18 years of age) primary care patients were selected from the EHR of Geisinger to participate in a study of the epidemiology of CRS. Individuals who responded to the baseline questionnaire were additionally mailed four seasonal follow-up questionnaires over the course of 16-months, to evaluate seasonal exacerbations (Table 1; for example questionnaire see online supplemental material S1). This study was approved by the Institutional Review Board (IRB) of Geisinger, which has an IRB Authorization Agreement with the Johns Hopkins Bloomberg School of Public Health. Health Insurance Portability and Accountability Act authorization and written informed consent waivers were approved by the IRB.
Study populationGeisinger provides primary care services to over 450,000 patients, with the majority residing in central and northeastern Pennsylvania. The source population for this study consisted of 200,769 adult primary care patients who had available EHR data, including race/ethnicity. Stratified sampling was utilized to over represent individuals more likely to have CRS, as well as racial/ethnic minorities (8% of Geisinger patients identify as non-white race/ethnicity). From the source population, 23,700 individuals were selected to participate in the baseline survey and baseline responders (n = 7,847) were mailed four follow-up questionnaires with four-month intervals in-between (Table 1).
Description of sampling methodThe sampling method has been reported previously (5, 18). Briefly, individuals with a greater likelihood of having CRS were over-sampled by using EHR data to categorize individuals into three groups, based on International Classification of Disease (ICD)-9 codes as well as Current Procedural Terminology (CPT) codes from patient medical records for: CRS, asthma, allergic rhinitis, sinus procedures, and related information (18). Oversampling of racial and ethnic minorities was also performed. Sampling proportions are reported elsewhere (5).
CRS classificationIndividuals were classified as having EPOSS CRS as previously reported (5, 18). In brief, CRS status was determined using subject responses concerning the frequency of the cardinal symptoms of CRS (nasal congestion/blockage, green/yellow nasal discharge [purulence], post-nasal drip, smell loss, facial pain, and facial pressure), as defined by EPOS (1). Based on responses to these questions at the baseline and first follow-up questionnaires, subjects were classified as “current long-term” (current CRS at both questionnaires), “current recent” (past or never CRS at baseline, current CRS at follow-up), “past” (past CRS at baseline, not current at follow-up) or “never” (no CRS at either questionnaire). Only these questionnaires were used for determining CRS status in this study because two of the follow-up questionnaires (winter and spring exacerbation) did not include questions about EPOS symptoms over the past three months and we did not want to induce reverse causality in the association of CRS status and exacerbation. We did not differentiate between CRS with and without nasal polyps since objective evidence of CRS was unavailable for all study participants and therefore no way to reliably phenotype these subjects.
Operationalization of NSS severity and AENSSNSS severity was assessed in two different ways. The first used self-reported rating of NSS on a 10-point visual analog scale while the second used self-report of having “worse” or “much worse” NSS on a five-point Likert scale (1).
Using consensus recommendations (1, 2) and prior evidence on CRS exacerbations (9–11, 13, 15, 19) we operationalized three definitions for the classification of AENSS (see online supplemental material Table S1). All definitions required participants to self-report worsened NSS in the past four weeks. “AENSS-Med” defined exacerbation was based on self-reported medication use for worsened NSS. We only used antibiotics and oral corticosteroids as qualifying medications as these are unlikely to be prescribed for viral infections, thereby minimizing potential misclassification of AENSS as common colds. This definition is also parallel to the medical management recommended for asthma control (1), since no evidence-based treatment recommendations exist for AECRS (2). We did not include inhaled corticosteroids because this would certainly misclassify AENSS as asthma exacerbations. “AENSS-Sx” was based on duration (≥ 1 week) of worsened aggregate NSS, again to minimize ascertainment of colds as AENSS, since these usually resolve within 1 week. Lastly, “AENSS-Sx-Pur” required the same criteria as “AENSS-Sx”, but additionally required self-reported worsened purulence in the past four weeks, yielding a definition with greater relative specificity. Although NSS could be worse for longer than four weeks, only a four week period was measured on questionnaires.
Evaluation of risk factors for AENSS and confounding variablesBased on previous studies (5, 18), potential risk factors and confounding variables from the EHR included: current age (years); sex; race/ethnicity (white non-Hispanic vs. all other groups); smoking status (current, former, and never); body mass index (BMI, kg/m2); Charlson comorbidity index (20); and history of receiving Medical Assistance, a surrogate for family socioeconomic status (SES) (21). Individual self-reported information was ascertained from baseline and follow-up questionnaires (Table 1).
Previous studies have shown asthma to be associated with CRS (5, 22, 23), and was therefore hypothesized to be a risk factor for AENSS. As such, individuals who experienced ≥ 1 asthma symptom (awakening at night due to wheezing; wheezing, chest tightness, or whistling in the chest when not having cold or flu; chest wheezing during or after exercise; dry cough at night apart from a cold or chest infection) at least some of the time were classified as having asthma symptoms at baseline. Migraine headaches have similarly been associated with CRS (5, 18), therefore a binary indicator for whether a subject had migraine headaches at baseline was determined as previously reported (18, 24). The continuous “Anxiety Sensitivity Index (ASI)” (25) measures how much a person fears the symptoms of anxiety, believing them to be harmful, and was created from the fall exacerbation questionnaire and included in the analysis as quintiles to help control for confounding due to an individual’s propensity to be aware of and/or over-report symptoms. Questionnaire return dates were used to define the season in which exacerbations occurred as follows: autumn, September 22 through December 21; winter, December 22 through March 21; spring, March 22 through June 21; and summer, June 22 through September 21.
Statistical analysesGiven the paucity of information regarding NSS and AENSS by EPOSS CRS status, the goals of the analysis were to 1) assess differences in NSS severity by CRS status and AENSS definition, 2) estimate the seasonal prevalence of different subgroups of AENSS (e.g., by CRS status and AENSS definition) in the source population, and 3) evaluate associations of individual self-reported risk factors and season with AENSS by CRS status.
Survey-corrected methods were used for all analyses to account for the sampling design. Design weights were the inverse product of the probability of being selected into the study and probability of responding to the baseline questionnaire. Additionally, survey weights were corrected for attrition by estimating inverse probability of censoring weights (IPCW; see online supplemental material S2). Since CRS status was not ascertained at all time-points, CRS status at the first follow-up questionnaire was used for all follow-up questionnaires. Subjects who skipped a questionnaire (23.9%) were excluded from all subsequent questionnaires to avoid intermittent missingness.
Risk factor analysis consisted of inverse-probability-weighted generalized estimating equations logistic regression models assuming an independence working correlation matrix and incorporated stabilized truncated survey weights (see online supplemental material S2). Final survey weights had a median of 2.81 and range 2.45 – 43.03. Taylor linearization was used to estimate robust variances and standard errors. Lastly, item non-response for covariates was addressed by using multiple imputation by chained equations (25 imputed data sets).
Covariates were identified as being a risk factor if they retained statistical significance in adjusted models and were not a priori determined to be a confounder. Methods for assessing model fit are limited in multiply-imputed survey-based regression models. However, model-fit was assessed by visual inspection of deviance residuals versus predicted probabilities (from weighted candidate final models) and using. Archer-Lemeshow tests for goodness of fit. To assess the utility of the multiple imputations, Monte Carlo error estimates were generated for all effect estimates and associated test statistics. All analyses were conducted in STATA 14.1 (StataCorp, College Station, Texas).
ResultsDescription of participantsBaseline characteristics of the study population have been described elsewhere (5, 18). A total of 558 current long-term, 273 current recent, 1,644 past, and 2,261 never EPOSS CRS individuals contributed at least one observation to the analysis (Table 2). The general trends in Table 2 suggests individuals with AENSS appeared to be younger, white, female, on medical assistance, and have greater Charlson comorbidity index values, compared to those without AENSS (Table 2). The prevalence of AENSS increased from the lowest in the never group, to intermediate in the past and current recent CRS groups, to the highest in the current long-term CRS status group (Table 2). AENSS recurrence, as identified through the four follow-up questionnaires, was the least common in the never group and the most common in the current long-term CRS group (see online supplemental material Table S2).
Severity of nasal and sinus symptomsMean NSS severity scores varied by CRS group and exacerbation status (Figure 1; online supplemental material Table S3). There were statistically significant associations between CRS status and NSS severity (Table S3). Mean NSS scores increased ordinally from the lowest score in the never CRS group to the highest score in the current long-term CRS group, where those who were having AENSS had higher NSS severity than those who were not (p < 0.001 for all CRS status groups). Mean NSS severity scores by AENSS-Med and AENSS-Sx-Pur defined exacerbations were greater than in AENSS-Sx (Figure 1; online supplemental material Table S3).
Seasonal prevalence of AENSSPrevalence estimates of AENSS by CRS status and AENSS definition were estimated for each season (Figure 2; online supplemental material Table S4). The seasonal peak prevalence for exacerbation consistently occurred in the winter for past CRS status and in spring for never CRS status. Seasonal trends were comparable between AENSS-Sx and -Sx-Pur for the current long-term and current recent CRS groups, with peak prevalence occurring in the winter for the current recent CRS group, and a modest peak in the summer for the current long-term CRS group (Figure 2; online supplemental material Table S4).
Individual characteristic and seasonal risk factors for AENSSRisk factor analysis proceeded with two of the three AENSS definitions (AENSS-Med and -Sx-Pur). We did not include AENSS-Sx since prevalence estimates were much greater from this definition compared to AENSS-Med and -Sx-Pur, which were both comparable, indicating a low relative specificity of AENSS-Sx compared to the other definitions. Tables 3 and 4 show the adjusted odds ratios (aORs) and 95% CIs for several covariates estimated from logistic regression models.
Several significant and elevated odds ratios were identified in relation to AENSS-Med (Table 3) for higher BMI, being a current smoker, having asthma or migraine symptoms at baseline, doctor diagnosed hay fever, having had two or more sinus surgeries, and winter season. As CRS status was found to modify associations of season with AENSS-Med, associations are displayed within strata of CRS status (Table 3).
Elevated odds ratios of risk factors with AENSS-Sx-Pur (Table 4) were found for white race/ethnicity, BMI, having asthma symptoms at baseline, doctor diagnosed hay fever, history of having two or more sinus surgeries, and season (winter and spring). Age modified associations of CRS status with AENSS, therefore CRS status associations are displayed at the grand mean age (55.1 years). Subjects with either past or current long-term CRS had increased odds of AENSS-Sx-Pur. The interaction between age and CRS status was observed as a linear reduction in odds of AENSS-Sx-Pur with higher ages for all CRS status groups, except current long-term CRS (Table 4).
DiscussionTo our knowledge, this is the first study of the epidemiology of AENSS by EPOSS CRS status, while evaluating three definitions of AENSS. There were several potentially important findings by season and CRS status, offering possible etiologic and diagnostic insights relevant to clinical management of AENSS. NSS and AENSS were common among all subjects; NSS and AENSS severity were worst in subjects with current, long-term CRS; prevalence of AENSS as measured by AENSS-Sx was almost 2-fold higher than by -Sx-Pur and -Med; there were clear seasonal prevalence differences observed using the different definitions of AENSS; and risk factor analysis showed differing associations depending on the definition of AENSS, particularly that odds of AENSS-Sx-Pur did not decline with increasing age in current long-term EPOSS subjects but did in all other EPOSS groups.
In the absence of consensus on how to measure AENSS, we operationalized three definitions that first identified worsening of symptoms (e.g., NSS in past four weeks reported as worse or much worse than “usual”) and then applied criteria that would differentially maximize sensitivity (the proportion of people with an exacerbation who met the AENSS definition), positive predictive value (PPV; the proportion of people who met the AENSS definition who had an exacerbation), and clinical similarity to how asthma exacerbation is often defined in epidemiologic studies. AENSS-Sx was the most sensitive (and by definition, least specific) definition, and is useful for researchers wanting to estimate the prevalence of AENSS while avoiding under-estimation. Of the three definitions, AENSS-Sx-Pur should have the highest PPV, and therefore may be the best for risk factor analysis since its lower misclassification will minimize bias in effect estimates towards the null. Lastly, a medication-based AENSS definition (AENSS-Med) for CRS requires care-seeking behavior for symptoms that are generally not life-threatening, thus making a medication-based definition much more reliant on health care access and delivery.
Although overall prevalence estimates for AENSS-Med and -Sx-Pur were comparable, there was little overlap in individuals ascertained by the two definitions, with only 31% of AENSS-Sx-Pur events additionally meeting criteria for AENSS-Med (see online supplemental material Table S5). Discordance could be due to AENSS-Med being influenced by an individual’s propensity to seek and be provided with medical care.
AENSS occurred in all CRS status groups, but prevalence was higher and severity worse among subjects with past or current (long-term and recent) CRS. The absolute change in severity during an AENSS was largest among subjects who never met EPOSS CRS criteria, possibly due to a ceiling effect in NSS severity among individuals with current or past CRS.
AENSS prevalence was greatest in the winter and spring for the past and never CRS groups, respectively, across all three AENSS definitions. This suggests exacerbations might be driven by viral infections in the winter (e.g. rhinoviruses (26–30)) or seasonal allergens and allergic rhinitis, for those with or without a history of CRS, respectively. No consistent seasonal patterns of AENSS were observed for the current CRS status groups across all three definitions of AENSS; however, a peak prevalence occurred in the winter or summer (AENSS-Med and AENSS-Sx/Sx-Pur, respectively) for the current recent CRS group. Prevalence of AENSS-Med was greatest in the current long-term CRS status group and occurred in the winter season, yet no major seasonal changes in AENSS-Sx/Sx-Pur were observed for this group, apart from modest associations with summer season. This could be due to residual selection bias due to loss-to-follow-up unaccounted for by the weighting procedure, or could reflect specific seasonal triggers relevant to this subgroup (e.g. ragweed). It is possible that individuals with a long-term history of current CRS are more likely to be prescribed medications for NSS in the winter, although NSS may not necessarily be more severe (given the lack of observed associations between season and AENSS-Sx/-Sx-Pur in this group). This may also reflect underlying pathobiology relevant to triggers of exacerbation in this group, since medical management would depend on the trigger (e.g. infections vs. grass pollen).
We identified clinical and seasonal factors associated with AENSS. CRS status, increased BMI, asthma symptoms, hay fever, migraine symptoms, history of sinus surgeries, and season were associated with AENSS by both Med and Sx-Pur defined exacerbation. Our findings with BMI are similar to those found previously with CRS (31, 32) and other otorhinolaryngological (32) diseases, possibly due to chronic low-grade inflammation associated with obesity (33, 34). Asthma (1, 2, 22, 23) and hay fever (1, 2, 22, 23) have been associated with CRS; however, symptom overlap between these conditions could indicate measurement error in EPOSS criteria. To address this issue, we evaluated whether hay fever or asthma modified associations of CRS status with AENSS. As we found no evidence for this, we included hay fever and asthma diagnoses as covariates in regression models without further stratification and statistical significance suggests indication of the unified airways disease concept. The relationship between migraines and NSS has been observed in previous studies (5, 23), but could be due to misclassification of overlapping symptoms or biologic pathways (35–37), or both. Sinus surgery was also associated with AENSS and could be due to bacterial infections in some CRS patients (38), or be a proxy for individuals with recalcitrant CRS or persistent ARS, who are more likely to be aware of the severity of sinus symptoms over time.
Females were more likely to have AENSS-Med than males, possibly due to residual confounding associated with medical-seeking and -prescribing behaviors (39), since this association was only modestly observed in the AENSS-Sx-Pur model; however, female sex has been associated with CRS symptoms in other studies (5, 40, 41). Non-white race/ethnicity was associated with reduced odds of both AENSS definitions, though only statistically significant in the AENSS-Sx-Pur model. Lastly, never smokers were less likely to have AENSS-Med, compared to current smokers, although no association with smoking status and AENSS-Sx-Pur was observed. The odds of AENSS-Sx-Pur declined with higher ages, excluding the current long-term CRS status group, possibly due to differential susceptibility to viral infections which precede bacterial infections and decrease with increasing age (42). Yet, individuals with long-term CRS may be at risk of developing viral respiratory infections even at older ages due to compromised epithelial barrier function (43, 44), which can accompany CRS (1, 2, 45), suggestive of a disease progressive model in those with persistent CRS.
Our study had several strengths, including study of the general population in the region representing the full spectrum of diseases with NSS, longitudinal design (the first to our knowledge), large sample size, and evaluation of a relevant set of individual-reported potential risk factors for AENSS, as well as season. We also used several definitions of AENSS to comparatively assess their utility in epidemiologic research, as advised by ICAR (2). Our study is not without limitations, however. We used a definition of CRS which did not include confirmation of inflammation by endoscopy or CT scan so we were unable to classify individuals with clinical CRS. Second, both CRS status and self-reported individual characteristics were selected from the same questionnaires; as such there is the potential for spurious associations between them, since they are dependent on how an individual interprets and responds to the questions. However, a strength of this study is the inclusion of the ASI as a covariate, which adjusts for an individual’s propensity to over-report symptoms and comorbidities. Therefore, the possibility of false associations from same source bias was mitigated. Furthermore, we used weighting methods and multiple imputation to adjust for non-response and potential selection bias.
In summary, our study found that NSS and AENSS were common in the general population. NSS and AENSS severity were worse across categories of EPOSS CRS, peaking among current long-term CRS. Seasonal exacerbation prevalence depended on the AENSS definition and differed by EPOSS CRS status. Results suggest that a high PPV definition (e.g., AENSS-Sx-Pur) may provide the best balance between a sensitive definition and one which is clinically meaningful.