Data source
We conducted a prospective cohort study using the public released de-identified NHANES III (1988–1994). All-cause and cause-specific mortality were assessed in all participants linked to the National Death Index (NDI) mortality data (1988–2015). The NHANES III study was conducted by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention (CDC) and employed a complex, multistage, probability sampling design that allows results to be extrapolated to the entire US population. The program is designed to examine the health and nutritional status of the US civilian, non-institutionalized population aged 2 months and older for NHANES III participants [29]. Details on the NHANES Laboratory/Medical Technologists Procedures and Anthropometry Procedures are described elsewhere [30]. The survey protocol was approved annually by the NCHS Research Ethics Review Board and all participants provided written informed consent [29]. Detailed information about the dietary interview portion has been published previously [31].
Since NHANES data are de-identified and publicly available data, the Institutional Review Board (IRB) at the researchers’ institution does not consider this to be human subject research. Therefore, human subjects’ approval was not necessary nor sought since this was a de-identified data-only study.
Study population
The current study included individuals aged 18 years or older from a nationally representative sample of NHANES III with data on mortality status (n = 19,598). As done by a previous study [32], participants who reported implausible daily energy intake levels (< 800 kcal or > 4200 kcal for men and < 500 kcal or > 3500 kcal for women) (n = 1132) and participants with missing dietary data (n = 2920) were excluded, leaving 15,546 participants in the current analysis for NHANES III participants (Fig. 1).
Assessment of mushroom consumption
Mushroom intake was estimated based on NHANES III dietary intake data obtained via a single 24-h recall obtained by a trained interviewer with the use of an automated, microcomputer-based dietary interview and coding system known as the NHANES III Dietary Data Collection (DDC) System [31, 33]. All eligible participants provided a single 24-h dietary recall and a small subsample of approximately 8% of participants was eligible for a second 24-h dietary recall [29]. The USDA Survey Nutrient Database System (SNDB) was used to determine the nutrient content of foods. The NHANES III Individual Foods File (IFF) contains USDA food codes for recipe foods and the amount eaten in grams for every food item (identified by a unique 7-digit code) and was searched to identify each food containing mushrooms. Detailed information about NHANES III dietary data collection and IFF can be found on the NHANES website (https://wwwn.cdc.gov/nchs/data/nhanes3/2a/iff-acc.pdf). As done by a previous study [34], mushroom consumption was calculated based on the intake of foods that were mostly mushrooms or mushrooms alone, for example egg omelet or scramble egg served with mushrooms, or dishes with mushrooms as a recipe component, for example mushrooms gravy. In the mixed foods with mushrooms, the US Environmental Protection Agency-USDA Food Commodity Intake Database (FCID) commodity codes were used to determine the actual amounts of mushroom intake as follow: grams of intake by USDA food code time the commodity weight of mushroom contribution per 100 g of the USDA food code [34]. Details information regarding the Food Commodity Intake Database is described elsewhere [35]. Only individuals with reliable and complete dietary records for mushroom intake as determined by NCHS were included in the current analysis. Unique USDA food codes used to identify mushroom consumers (n = 544) are presented in the supplemental Table 1.
Mortality ascertainment
The endpoints for this study were all-cause and cause-specific mortality, ascertained by NCHS using death certificates. The de-identified and anonymized data of the NHANES III participants were linked to NDI Mortality Files (n = 5826) with a probabilistic matching algorithm to determine mortality status using the NHANES III sequence number. The NCHS public-use linked mortality file provides mortality follow-up data from the date of NHANES III survey participation up until December 31, 2015 (1988–2015) [36]. Participants with no matched death record were considered to be alive during the entire follow-up period.
All cause-mortality in the current analysis included all specified causes of death recorded in the Public-use Linked Mortality files. Cause-of-mortality coding for all US mortality occurring prior to 1999 was determined using the Ninth Revision of the International Classification of Diseases (ICD-9), while for all mortality after 1998 follows the Tenth Revision of the International Classification of Diseases, (ICD-10) for mortality occurring in or after 1999. To facilitate and assist researchers with analyses, the NCHS recoded all mortality occurring prior to 1999 coded under ICD-9 guidelines into comparable ICD-10 according to the underlying cause of mortality groups [36]. All specified causes of mortality as well as underlying causes of mortality were recorded in the Public-use Linked Mortality files using the following ICD-10 codes: Cardiovascular diseases including heart diseases (I00-I09, I11, I13, I20-I51) and cerebrovascular diseases (I60-I69), malignant neoplasms. (C00-C97). Other cause-specific mortality included: chronic lower respiratory diseases (J40-J47), accidents (unintentional injuries) (V01-X59, Y85-Y86), Alzheimer’s disease (G30), diabetes mellitus (E10-E14), influenza and pneumonia (J09-J18), nephritis, nephrotic syndrome, and nephrosis (N00-N07, N17-N19, N25-N27) and residual causes.
Assessment of dietary intakes and covariates
For the present analysis, the following covariates were extracted from the existing NHANES III 24-h recall dietary intake datasets based on previous literature [25, 34]: intake of total energy (kcal/d), alcohol (g/d), energy-adjusted fat (g)/1000 kcal/d), carbohydrates (g)/1000 kcal/d), fiber (g)/1000 kcal/d), and the Healthy Eating Index (HEI-2000), a measure of overall diet quality, which was included in the NHANES III data. To compute the HEI-2000 for NHANES III participants, Food Guide Pyramid serving sizes recommended by the USDA was applied to the food servings obtained through a 24-h [37]. The HEI-2000 includes a 10-component system of five food groups including fruits, vegetables, four nutrients, and a measure of variety in food intake. The total score ranges from 0 to 100, with a higher score suggesting a healthier diet [38]. Information on age (years), sex (men/women), ethnicity-race (non-Hispanic white, non-Hispanic Black, Mexican American, others), US regions (Northeast, Midwest, South, West), place of residence (Urban/Rural), education attainment (years), marital status (married, widowed/divorced/separated, never married), smoking status (smoked 100+ cigarettes in life yes or no), and physical activity level (moderate to vigorous) were collected through self-reported. The body measurement (including weight and height) was measured at the time of physical examination in a mobile examination center (MEC) or in the participant’s home. The body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared and was categorized into 5 groups using CDC classification: underweight (< 18.5 kg/m2), normal weight (18.5–24.9), overweight (25.0–29.9), obese (30.0–34.9), and excessively obese (≥35.0). Given the small number of participants in the first and the last categories, BMI was later categorized into 3 groups: normal weight (< 24.9), overweight (25.0–29.9), obese (≥30).
Statistical analysis
SAS statistical software version 9.4 (SAS Institute) was used to perform all statistical analyses using 2-sided P < .05 as the significance level. Survey analysis procedures were used to account for the sample weights, clustering, and stratification of the complex sampling design as specified in the instructions for using NHANES data to ensure nationally representative estimates [39]. Univariate analyses were conducted to assess the statistical significance of differences in weighted percentages for categorical variables using the Rao-Scott χ2 test and weighted means for continuous variables using t-test. For each participant, mortality follow-up time was calculated as the time from the baseline survey participation interview date until the date of death or end of follow-up (December 31, 2015), whichever came first. The consumption of mushrooms was deemed as the primary exposure during the study period. We used time-dependent multivariable Cox proportional hazards models to assess the association of mushroom consumption with all-cause and cause-specific mortality risk during the follow-up. The Cox proportional hazards regression models were performed to calculate multivariable-adjusted hazard ratios (HRs) and 95% confidence intervals (95% CIs) for all-cause and cause-specific mortality and the proportional hazards assumption was not violated. The following potential confounders were controlled for in the multivariable Cox regression models: age (years), sex (men/women), ethnicity-race (non-Hispanic white, non-Hispanic Black, Mexican American, others) region (Northeast, Midwest, South, West), place of residence (rural/urban), education attainment (years), marital status (categorical), BMI (categorical), moderate to vigorous physical activity (yes vs. no), smoking (smoked 100+ cigarettes in life, yes vs. no) intake of total energy (kcal/d), alcohol intake (g/d), energy-adjusted fat (g)/1000 kcal/d), carbohydrate (g)/1000 kcal/d), fiber (g)/1000 kcal/d), and the HEI-2000. To further examine whether there was evidence of a dose-response relationship between greater mushroom consumption and all-cause mortality risk, we further categorized mushroom intake into 4 categories: no mushroom intake (0 g/d, n = 15,002), lowest (median intake = 10 g/d, range = 23.9, n = 346), middle (median intake = 35 g/d, range = 19.0, n = 104), and high (median intake = 72 g/d, range = 141.4, n = 94). Test for linear trend was examined for significance by using the median value for each category of mushroom intake, which was analyzed as a continuous variable in the multivariable-adjusted Cox model as done by previous researchers [40]. We did a single imputation using the fully conditional specification method for missing values for demographic and lifestyle variables [41]. As a secondary analysis, we conducted a nutritional substitution analysis to compare the health effect of substituting 1-serving/d of mushroom for 1-serving/d of red or processed meat. As done by a previous study, 1-serving of red or processed meat was defined as 3.5-oz equivalents and 1-serving of mushroom as 70 g [42, 43]. Red or processed meat includes such as beef, veal, pork, lamb, cured, and organs meat. The association of substituting 1-serving/d of mushroom for 1-serving/d of red or processed meat with all-cause mortality was examined by including both as continuous variables in the same multivariable Cox regression model adjusting for age, sex, ethnicity-race, region, place of residence, education attainment (years), marital status, BMI, moderate to vigorous physical activity (yes vs. no, smoking ( smoked 100+ cigarettes in life yes vs. no), alcohol intake (g/d), total energy (kcal/d), and other dietary variables, including poultry (oz/d), fish (oz/d), eggs (oz/d), nuts/soy (oz/d), legumes (svg/d), fruit (svg/d), dark green/ yellow vegetables (svg/d), dairyy (svg/d), discretionary fat (g/d), and added sugar (tsp/d). The difference in their regression coefficients, variances, and covariance were used to estimate the HR and 95% CIs for the substitution effect. This methodology has been widely used in providing a better solution to dietary patterns [25, 44, 45].
To further test for the robustness of our results, we conducted a series of sensitivity analyses. First, to minimize potential bias, we further adjusted for a propensity score , which was calculated by including the aforementioned covariates in the final model 3. This approach allows us to balance baseline data between participants with mushroom intake and those without mushroom intake.
Second, to understand the short- vs. long-term impact of mushroom intake on mortality, we dichotomously calculated hazard by excluding when mortality cases occurred during the first 2 years of follow-up, adjusting for preceding covariates. Third, because major chronic diseases are strongly associated with the risk of mortality [46], we conducted a sensitivity analysis by excluding participants with baseline congestive heart failure or hypertension/high blood pressure, cancer, diabetes, or changed their diet because of high blood pressure. Fourth, the interaction between mushroom intake and age, ethnicity-race, sex in association with all-cause mortality were statistically tested by including the interaction terms in the Cox regression model. Lastly, a previous study of NHANES data suggested that mushroom intake was associated with better nutrients intake including micronutrients and diet quality [34], therefore we further adjusted our final model 3 for energy-adjusted Vitamin E (mg)/1000 kcal/d), β-carotene (mcg)/1000 kcal/d), vitamin C (mg)/1000 kcal/d), copper (mg)/1000 kcal/d), and selenium (mcg)/1000 kcal/d) intake.