Prospective study of dietary patterns and chronic obstructive pulmonary disease among US women

¹ From the Departments of Nutrition (RV, TTF, FBH, and WW) and Epidemiology (FBH, WW, and CAC), Harvard School of Public Health, Boston, MA; the Department of Nutrition, Simmons College, Boston, MA (TTF); Channing Laboratory, Department of Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, MA (RGB, FBH, WW, and CAC); the Division of General Medicine, Department of Medicine, and Department of Epidemiology, Columbia University Medical Center, New York, NY (RGB); and the Department of Emergency Medicine, Massachusetts General Hospital, Harvard Medical School, Boston, MA (CAC)

² Supported by grants CA-87969, HL-63841, HL-60712, HL-077612, HL-075476, and AI-52338 from the National Institutes of Health, Bethesda, MD; grants from the Société Franaise de Nutrition, Paris, France (to RV); and the Société de Pneumologie de Langue Franaise, Paris, France (to RV).

³ Reprints not available. Address correspondence to R Varraso, Department of Nutrition, Harvard School of Public Health, 655 Huntington Avenue, Boston, MA 02115. E-mail: rvarraso{at}hsph.harvard.edu.

ABSTRACT  
Background: Although many foods and nutrients are associated with lung function or symptoms of chronic obstructive pulmonary disease (COPD), the relation between overall diet and newly diagnosed COPD is not known.

Objective: We assessed the relation between dietary patterns and newly diagnosed COPD in women.

Design: Data were collected from a large prospective cohort of US women (Nurses' Health Study). Between 1984 and 2000, 754 self-reported confirmed cases of newly diagnosed COPD were identified among 72 043 women. With the use of principal component analysis, 2 dietary patterns were identified: a prudent pattern (fruit, vegetables, fish, whole-grain products) and a Western pattern (refined grains, cured and red meats, desserts, French fries). Patterns were categorized into quintiles, and the risk of COPD was compared between quintiles (lowest as reference) with the use of Cox proportional hazard models.

Results: After adjustments for 14 potential confounders, the prudent pattern was negatively associated with risk of newly diagnosed COPD [relative risk (RR) for highest compared with lowest quintile: 0.75; 95% CI: 0.58, 0.98; P for trend = 0.02] whereas the Western pattern was positively associated with risk of COPD (RR for highest compared with lowest quintile: 1.31; 95% CI: 0.94, 1.82; P for trend = 0.02). In contrast with findings for COPD, dietary patterns were not associated with the risk of adult-onset asthma.

Conclusion: In women, a negative association was found between a diet rich in fruit, vegetables, and fish and the risk of COPD, whereas a positive association was found between a diet rich in refined grains, cured and red meats, desserts, and French fries and the risk of COPD.

Key Words: Dietary pattern • principal component analysis • chronic obstructive pulmonary disease • COPD • prospective cohort • body mass index

INTRODUCTION  
Increases in chronic obstructive pulmonary disease (COPD) incidence in women are related to aging of the population and to smoking (1). Cigarette smoking is the main risk factor for COPD, but not all smokers develop COPD (2), an observation that suggests that other factors also are involved. Of these environmental factors, changes in diet (particularly decreased consumption of fresh fruit and vegetables) have been evoked to explain the large increase in obstructive lung disease, such as asthma and COPD (3). Most evidence about the relation between diet and COPD comes from cross-sectional studies, which suggest benefit from a diet rich in antioxidants and n–3 fatty acids on lung function (4-6). A few longitudinal studies have investigated the relation between diet and the decline in forced expiratory volume in 1 s (FEV₁) or COPD symptoms (7-10), and they reported an apparent benefit of fruit and vegetable intake.

Among patients with COPD, leanness is a major risk factor for poor prognosis (11, 12). Per the hypothesis that malnutrition, and therefore leanness, was a consequence of COPD associated with a poor prognosis, supplementation studies were conducted. Those trials have been negative, which suggests that the association between leanness and COPD is complex and that nutritional rehabilitation per se has no significant effect (13).

The assessment of dietary patterns instead of specific foods or nutrients has been proposed as a new approach in nutritional epidemiology of chronic diseases (14). Dietary patterns provide an overview of the diet. Although many foods or nutrients are identified in relation with lung function, the relation of overall diet to newly diagnosed COPD is not known. We examined this issue in a prospective cohort of >70 000 women.

SUBJECTS AND METHODS  
Overview
The Nurses' Health Study (NHS) began in 1976, when 121 700 female nurses aged 30–55 y living in 11 US states responded to a mailed health questionnaire (15). Follow-up questionnaires are sent every 2 y. In 1984, participants completed a 116-item food-frequency questionnaire (FFQ). Similar FFQs were sent to the women every 2–4 y. The institutional review board approved the NHS protocols, and written consent was obtained from all subjects.

Participants without a completed FFQ at baseline or participants with unreasonably high (>3500 kcal/d) or low intakes (<500 kcal/d) and those who had left >70 items blank were excluded from the analysis. Women who reported a diagnosed asthma or COPD at baseline in 1984 were also excluded from the present analysis. The final baseline population included 72 043 women. Between 1984 and 2000, >90% of this population was followed up.

Assessment of dietary patterns
Dietary intake information was collected by an FFQ designed to assess average food intake during the previous 12 mo. Standard portion sizes were listed with each food. For each food item, participants indicated their average frequency of consumption during the past year in terms of the specified serving size by checking 1 of 9 frequency categories, ranging from "almost never" to "6 times/d." The selected frequency category for each food item was converted to a daily intake. For example, a response of "1 serving/wk" was converted to 0.14 servings/d. FFQs were administrated in 1984, 1986, 1990, 1994, and 1998. To prepare for factor analysis, food items were grouped into 38 predefined foods groups. Food items that were similar in nutrient profile and culinary use were grouped. This classification follows that of another study of dietary patterns in these women (16). To assess the sensitivity of dietary patterns to use of this specific, a priori grouping of foods, we also performed principal component analysis by using the individual food items.

With the use of principal component analysis, dietary patterns were identified from FFQs administrated in 1984, 1986, 1990, 1994, and 1998. The factors were rotated by an orthogonal transformation (Varimax rotation function in SAS; SAS Institute, Cary, NC) to achieve simpler structure with greater interpretability. The number of factors to retain was determined by using the diagram of eigenvalues, the Scree plot, and the interpretability of the factors, as well as the percentage of variance explained.

Foods that loaded 0.30 were considered to be making a contribution to the factor, although the value for meaningful factor loading is arbitrary. The factor score for each pattern was constructed by summing observed intakes of the component food items weighted by factor loading. To reduce measurement errors and to represent long-term dietary patterns, the cumulative average of pattern scores was calculated and then divided into quintiles. For the analysis according to the level of physical activity, the cumulative average scores for patterns were divided into tertiles to have enough cases in each group.

Assessment of respiratory phenotypes
Self-reported COPD was defined by the affirmative response to physician-diagnosed chronic bronchitis or emphysema and by the report of a diagnostic test at diagnosis (ie, pulmonary function testing, chest radiograph, or chest computed tomography). Women also reported age at diagnosis. Between 1984 and 2000, 754 cases of newly diagnosed COPD were reported.

We previously validated this definition in a 10% random sample in this cohort (17). We were unable to obtain standardized, reliable spirometry on this random sample because NHS participants are geographically dispersed (they lived in 11 US states in 1976 and now live throughout the United States) and are contacted by mail. Instead, we obtained participants' medical records and a physician reviewed them in a blinded fashion. The diagnosis of COPD was confirmed in 80% of 218 cases who meet this case definition and 88% of cases who met this definition and denied a physician diagnosis of asthma. Results of pulmonary function testing were available in the medical records of 71% of confirmed cases; the mean FEV₁ in this group was 50% of predicted.

Asthma was also self-reported and was defined by a doctor diagnosis of asthma and the use of medication for asthma within the past 12 mo. Between 1984 and 2000, 1100 new cases of adult-onset asthma were reported and met our epidemiologic definition.

Assessment of other variables
Information on smoking status included the categories of never smokers, exsmokers, and current smokers. For smokers, further information about the amount of tobacco smoke was available by pack-years of smoking. Exposure to secondhand tobacco smoke was also reported and defined by an exposure at home, at work, or at both locations. Menopause and hormonal replacement therapy (HRT) use were assessed every 2 y by self-reported questionnaires, and menopausal status was categorized in 5 classes (premenopause, postmenopause and never HRT use, postmenopause and past user for HRT, postmenopause and estrogen replacement therapy, postmenopause and estrogen-progesterone replacement therapy). Race-ethnicity, spouse's educational attainment, physician visits, and region were also collected. Race-ethnicity was categorized in 2 classes (white, nonwhite), spouse's educational attainment was categorized in 3 classes (graduate school, college, high school), physician examination in previous 2 y was categorized in 3 classes (no visit, screening, symptoms), and US region was categorized in 6 classes (New England, Mid-Atlantic, East North Central, South Atlantic, West South Central, Pacific). Body mass index [BMI; calculated as weight divided by height squared (kg/m²)], physical activity, and multivitamin use were assessed every 2 y by self-reported questionnaires. BMI was categorized into 7 classes: 20.0, 20.0–22.4, 22.5–24.9, 25.0–27.4, 27.5–29.9, 30.0–34.9, 35.0). Women also reported physical activity, including a variety of activities such as walking, bicycle, swimming, or tennis. The validity of the questionnaire in assessing physical activity was described elsewhere (18). Physical activity was measured in metabolic equivalents per week, whereby 1 metabolic equivalent was equal to the energy expended at the basal metabolic rate or at rest and divided into quintiles. Use of vitamin and mineral supplements was investigated every 2 y. Total calorie intake was estimated through the FFQ, expressed in kilocalorie per day (kcal/d) and categorized according to quintile.

Statistical analysis
Statistical analyses included principal component analysis, analysis of variance, and Cox proportional hazard regression models. Cox models were adjusted for age, smoking status, pack-years, pack-years², exposure to secondhand tobacco smoke, menopausal status, race-ethnicity, spouse's educational attainment, physician visits, US region, BMI, physical activity, multivitamin us, and energy intake. Women were censored at the date of last contact, and the date of diagnosis was calculated by using the date of birth and the age at diagnosis. A test for trend across the quintiles of each pattern was calculated by treating the categories as an ordinal variable in a proportional hazards model. Residual confounding by smoking remains an important issue in studies of respiratory diseases and diet. Because smoking is the main risk factor for COPD, analyses were also performed among exsmokers and current smokers. Analyses also were stratified according to BMI in 3 classes (20, 20–25, 25). Physical activity level was categorized as low compared with high by using the median. We formally tested the interaction between each dietary pattern with smoking, BMI in 3 classes, and physical activity. All analyses were conducted with the use of SAS software, version 9 (SAS Institute).

RESULTS  
Dietary patterns and characteristics of the population
With the use of the principal component analysis, 2 distinct major dietary patterns were identified at baseline (Table 1). The first pattern was loaded by a high intake of fruit, vegetables, fish, poultry, whole-grain products, and low-fat dairy products. The second pattern was loaded by a high consumption of refined grains, cured and red meats, desserts and sweets, French fries, and high-fat dairy products. According to previous studies about dietary patterns in this population, the first pattern was labeled the "prudent" pattern and the second pattern was labeled the "Western" pattern. Similar dietary patterns were identified with the use of FFQs from 1986, 1990, 1994, and 1998. The principal component analysis performed at baseline on the individual foods without a priori grouping gave similar results: the first pattern was loaded by a high intake of fruit (fresh apples or pears; oranges; peaches, apricots, or plums; strawberries; cantaloupes; blueberries; grapefruits), vegetables (broccoli, eggplant, cauliflower, coleslaw, carrots, raw spinach, celery, string beans, romaine leaf lettuce, yellow squash, cooked spinach, iceberg head lettuce, tomatoes, mushrooms, Brussels sprouts, mixed vegetables, garlic, beans lentils, beets), poultry (chicken or turkey without skin), and fish. The second pattern was loaded by a high intake of French fries, hamburger, cured meats (processed meats, hot dogs, bacon), sweets and desserts (home-baked cake, doughnuts, brownies, ready-made sweet rolls, home-baked pies, pancakes or waffles), and refined cereals (white bread, pasta).

View this table:
TABLE 1. Factor loading matrix for the prudent and Western patterns at baseline from principal component analysis¹

 
The characteristics of the population according to the quintile of both prudent and Western patterns are presented in Table 2. Compared with women with the lowest intake of the prudent diet (the lowest quintile), women with the highest intake of prudent diet (the highest quintile) were more physically active, were less likely to be current smokers, and were more frequent users of multivitamin supplements. Women with the highest intake of the prudent diet consumed more polyunsaturated fat, more proteins, and more carbohydrates, but less saturated fat and trans fatty acids.

View this table:
TABLE 2. Age-standardized baseline characteristics by quintile (Q) of the 1984 pattern score among 72 043 women¹

 
Compared with women with the lowest intake of the Western diet, women with the highest intake of the Western diet had a higher BMI, were less physically active, were more likely to smoke, and took fewer multivitamin supplements. Women with the highest intake of the Western diet consumed more saturated fat and trans fatty acids, but less carbohydrates and proteins.

Dietary patterns and COPD
On average during the study period, among the 754 cases of COPD, 62% were smokers and 26% were exsmokers. Considering different time periods, cases occurring between 1984 and 1986 (n = 103) had the highest proportion of smokers: 82% of these cases were smokers in 1984 and 12% were exsmokers.

The prudent pattern was inversely associated with the risk of newly diagnosed COPD in women after adjustment for 14 potential confounders (Table 3). By contrast, the Western pattern was positively and significantly associated with the risk of newly diagnosed COPD. When the population was restricted to women without cancer or cardiovascular disease at baseline (n = 66 005), similar associations were found [for the prudent pattern, relative risk (RR) for highest compared with lowest quintile: 0.73; 95% CI: 0.55, 0.97; P for trend = 0.02; for the Western pattern, RR for highest compared with lowest quintile: 1.22; 95% CI: 0.86, 1.73; P for trend = 0.08]. Lagged analyses were also performed and similar results were found (data not shown).

View this table:
TABLE 3. Association between quintile (Q) of the cumulative average patterns and newly diagnosed chronic obstructive pulmonary disease (COPD) (1984–2000)¹

 
The relation between dietary patterns and newly diagnosed COPD also was investigated among exsmokers. After adjustments for potential confounders, the association between the prudent pattern and newly diagnosed COPD remained statistically significant (RR for highest compared with lowest quintile: 0.50; 95% CI: 0.31, 0.82; P for trend = 0.01). The positive association between the Western pattern and the risk of newly diagnosed COPD also was present in exsmokers, but the trend was borderline significant (RR for highest compared with lowest quintile 1.59; 95% CI, 0.86, 2.94; P for trend = 0.08).

We found a significant interaction between BMI and the Western pattern (Table 4). After adjustments for potential confounders, the association between tertiles of the Western pattern and the risk of newly diagnosed COPD was stronger in lean (BMI 20) than in other women (RR for highest compared with lowest tertile for lean, normal, and overweight or obese subjects: 2.76, 1.17, and 1.50, respectively; P for interaction = 0.03). We found no significant interaction between BMI and the prudent pattern (P = 0.77).

View this table:
TABLE 4. Association between tertile (T) of the cumulative average of the Western pattern and newly diagnosed chronic obstructive pulmonary disease (COPD) according to BMI (3 classes)¹

 
To further address potential confounding by physical activity, we next examined the association between the Western pattern and newly diagnosed COPD in lean women, according to the level of physical activity. In lean women, no significant interaction was observed between the level of physical activity and the Western pattern (P = 0.32). The positive and significant association between the Western pattern and newly diagnosed COPD in lean women was seen both in participants with a low level of physical activity (RR for highest compared with lowest tertile 3.22; 95% CI: 1.20, 8.68; P for trend = 0.03), and in subjects with a high level of physical activity (RR for highest compared with lowest tertile: 4.58; 95% CI, 1.33, 15.80; P for trend = 0.02).

Dietary patterns and asthma
Although the primary outcome of this study was newly diagnosed COPD, we also examined the relation of dietary patterns with adult-onset asthma in this cohort of women, because of the potential overlap between the diagnoses of COPD and asthma. In contrast with the risk of newly diagnosed COPD, no association was found between the Western pattern and the risk of adult-onset asthma (RR for highest compared with lowest quintile: 0.90; 95% CI: 0.69, 1.18; P for trend = 0.24). Before adjustment for potential confounders, a positive association was found between the prudent pattern and the risk of adult-onset asthma (RR for highest compared with lowest quintile: 1.52; 95% CI: 1.26, 1.84; P for trend < 0.001). Adjustment for the same 14 potential confounders used in earlier analyses led to a borderline significant positive association between the prudent pattern and the risk of adult-onset asthma (RR for highest compared with lowest quintile: 1.23; 95% CI: 0.99, 1.53; P for trend = 0.07).

DISCUSSION  
With the use of principal component analysis, 2 distinct dietary patterns were identified in this large prospective cohort of US women. The prudent pattern was associated with a significantly decreased risk of newly diagnosed COPD, whereas the Western diet was associated with an increased risk of COPD. These associations were more clearly seen in exsmokers than in current smokers, although the interaction between smoking and each dietary pattern was not statistically significant. The association between the Western diet and the risk of COPD was stronger among lean women (BMI 20) than among normal, overweight, and obese women.

Although nutritional epidemiology often focuses on the intake of specific nutrients, persons do not eat isolated nutrients but instead meals consisting of a variety of foods with complex combinations of nutrients that may interact (19). In this context, it was proposed to investigate dietary patterns to address an overview of diet (14). Dietary patterns were investigated in relation to several diseases such as breast cancer (16, 20), cardiovascular diseases (21), or diabetes (22), but prior studies on the relation between dietary patterns and respiratory diseases are sparse.

The finding of the prudent pattern (loaded by fruit and vegetables) being associated with a decrease risk of newly diagnosed COPD is consistent with prior epidemiologic literature that suggests a beneficial effect of antioxidants, particularly vitamin C, and to a lesser extent vitamin E on COPD or FEV₁ values. Most of that epidemiologic literature comes from cross-sectional studies (23-27), but the few longitudinal studies have also reported a negative association between intake of fruit, vegetables, and vitamin C with the decline of FEV₁ (7-10). In 793 men from the Netherlands, the consumption of solid fruit was inversely related to the 25-y incidence of chronic lung disease (7). The prudent pattern also was loaded by a high intake of fish, one of the main sources of n–3 polyunsaturated fatty acids. Results are still inconsistent across studies (4), but the only published prospective study observed no relation between n–3 intake and the incidence of chronic lung disease (7). A few clinical trials, not designed for respiratory diseases, have provided data on respiratory phenotypes (28, 29), and only one recent trial was designed for emphysema (30). Data from the -tocopherol and ß-carotene Cancer Prevention Study in Finland (28) and from the ß-carotene and Retinol Efficacy Trial in the United States (29) showed no reduction in COPD symptoms in men receiving -tocopherol or ß-carotene (28) and no effect of vitamin A supplementation on the rate of decline of FEV₁ (29). Preliminary results from the Feasibility of Retinoids in the Treatment of Emphysema, a multicenter clinical trial in the United States that includes 150 patients with emphysema, showed no change in respiratory symptoms, lung function testing, and lung density on computed tomographic scanning after supplementation with retinoic acid (30). The effect of any individual nutrient in reducing the risk of COPD may be too small to detect as suggested by these negative results, but, when several nutrients are consumed together, the cumulative effect may be sufficient for detection. Indeed, considering diet by an overall approach rather than by specific foods or nutrients may suggest a more comprehensive approach to disease prevention.

The positive association between the Western pattern and newly diagnosed COPD is a novel finding for a US population. Recently, Butler et al (31) reported that the "meat-dim sum" pattern, loaded by a high intake of red meat, preserved foods, rice, noodles, and deep-fried foods, was associated with an increase risk of incident cough with phlegm in 52 325 adult Chinese Singaporeans. Although the diet and lifestyle of Chinese Singaporeans are different from those in US women, the findings are consistent and suggested a deleterious effect of a diet rich in meat, starchy foods, and high-fat dairy products on COPD. It was previously reported in this cohort of women that the Western pattern was positively correlated with concentrations of C-reactive protein and interleukin-6, 2 markers of systemic inflammation (32). The association between COPD and the systemic inflammation remains unclear; whereas some studies have reported that systemic inflammation is a consequence of COPD, causation also remains possible (33, 34). More studies are needed to better understand the association between the Western pattern, COPD, and inflammation.

An interesting finding was the significant interaction between BMI and the Western diet on newly diagnosed COPD. The association between the Western diet and newly diagnosed COPD was higher among lean women and was borderline significant among overweight or obese women. Leanness is associated with poor prognosis among patients with COPD, yet it remains unclear whether leanness is simply a consequence of established disease, a risk factor, or a marker of a risk factor. One consequence of systemic inflammation is to decrease fat-free mass, and studies suggest a key role of fat-free mass in COPD (35-37). Body composition is influenced by diet choice, physical activity, and genetic factors. In our study, the association between the Western diet and the risk of newly diagnosed COPD in lean participants was found both in lean women with high and low physical activity. Results should be interpreted with caution, but lean persons with a Western diet might have a higher grade of systemic inflammation than do normal or overweight persons eating a Western diet if the leanness reflects a loss of fat-free mass. The mechanism for the observed interaction requires further study.

Although our primary focus was on newly diagnosed COPD, the relation between dietary pattern and adult-onset asthma also was investigated because of the potential overlap between COPD and asthma and the potential misdiagnosis of COPD. In contrast with newly diagnosed COPD, no association was found between the Western pattern and adult-onset asthma, a result in agreement with Butler et al (31) who reported no association between the meat-dim sum pattern and incident asthma in Chinese Singaporeans. We reported a borderline significant positive association between the prudent pattern and adult-onset asthma. The only published longitudinal study on the relation between diet and adult-onset asthma was performed in this cohort (38) and showed no association between vitamin C intake and the risk of adult-onset asthma. Although other groups have suggested that foods and nutrients associated with a prudent dietary pattern may be beneficial (39-42), these associations are reported mostly in children, and evidence is growing that adult-onset asthma is different from childhood asthma (43).

Our study has several potential limitations. First, newly diagnosed COPD was defined by a self-reported physician-diagnosis of COPD and lung function results were not available. Nevertheless the questionnaire-based definition of newly diagnosed COPD was validated in a subset of this unique population of registered nurses (17). The main source of misclassification probably is a misdiagnosis with asthma, and our findings for asthma diagnosis were null. Moreover, cigarette smoking is the main risk factor for COPD, and only 11% of cases were nonsmokers, which provides additional support for our epidemiologic definition of COPD. Although we acknowledge the potential for some misclassification, these data allowed us to investigate the relations between diet and COPD in a large population, with repeated assessments both of diet and newly diagnosed COPD. We also acknowledge that the association between dietary patterns and COPD may be due, in part, to a residual confounding by cigarette smoking, which is a powerful risk factor. Smokers tend to eat unhealthy diets (44), and there is a risk of saying that an association between unhealthy diet and COPD risk was due to diet when, in fact, it was due to the smoking alone. To minimize this possibility, multivariate models were adjusted with multiple measures of tobacco exposure (smoking habits, pack-years, pack-years², and exposure to secondhand tobacco smoke), as we have done in prior analyses (17). Our second approach to this important issue was to assess the relation between diet and COPD in exsmokers, who would tend to have a better diet than current smokers (44). Analyses performed in exsmokers yielded comparable results, but it should be noted that the P for trend is driven by the difference between the low and high quintiles. Finally, we acknowledge that principal component analysis to derive dietary patterns involves several arbitrary decisions (19). Nevertheless, we found good reproducibility and validity over time of dietary patterns defined by factor analysis with data from an FFQ in a parallel cohort of men (45). We also performed a principal component analysis with individual foods, without grouping them a priori, and this sensitivity analysis was highly consistent with the primary results.

In summary, we report prospective data on the association of dietary patterns with the risk of newly diagnosed COPD. We identified the prudent and the Western patterns and found that both are associated, in different directions, with risk of COPD. Confirmation of these findings in other populations, particularly among men, is warranted. These data provide additional evidence about the beneficial effect of a diet rich in fruit and vegetables and suggest a deleterious effect of a more traditional Western diet. The most important public health message remains smoking cessation, but these data suggest that diet might also affect COPD risk. Guidance for nutritional education and intervention might be easier to translate for dietary patterns than guidance in term of specific nutrients.

ACKNOWLEDGMENTS  
We thank Gary Chase and Karen Corsano for invaluable assistance with the implementation of the study. We also thank Rong Chen and Rui Jiang for their help with the data set.

The author's responsibilities were as follows—RV: study conception and planning, statistical programming and data analysis, data interpretation, primary manuscript preparation, and funding; TTF: data collection, refinement of dietary pattern exposures, and data analysis; RGB: study conception and planning, creation of supplemental questionnaire data set, and funding; FBH: statistical expertise and data interpretation; WW: data collection, statistical expertise, and funding; CAC: study conception and planning, creation of supplemental questionnaire data set, data analysis, data interpretation, and funding. All authors contributed to the drafting of the report and approved the final version. None of the authors had a personal or financial conflict of interest.

REFERENCES