Diet patterns and risk of sepsis in community-dwelling adults: a cohort study

Study population and participants

The REGARDS study is a population-based investigation of stroke incidence in black and white US adults ≥45 years of age. Details of the study design have been reviewed elsewhere [23]. Briefly, the study was designed to provide approximately equal representation of men and women and oversampled individuals who were black as well as individuals living in eight Southeastern US states that have disproportionately high stroke mortality, termed the US stroke belt/buckle (North Carolina, South Carolina, Georgia, Tennessee, Mississippi, Alabama, Louisiana, and Arkansas). Trained personnel conducted computer-assisted telephone interviews to obtain information including participants’ sociodemographics, cardiovascular risk factors, cigarette smoking, physical activity, and use of medications. Following this call, trained health professionals conducted an in-home study visit that included an electrocardiograph (ECG) recording, blood pressure, height and weight measurements, inventory of medications and collection of blood and urine samples. Several questionnaires, including the 1998 Block food frequency questionnaire (Block 98 FFQ, NutritionQuest, Berkeley, CA), were left with participants to be completed after the in-home visit and mailed back to the study center. Overall, 30,239 black and white adults were enrolled between January 2003 and October 2007. The REGARDS study protocol was approved by the Institutional Review Boards (IRBs) governing research in human subjects at participating centers, and all participants provided informed consent. The University of Alabama at Birmingham served as the primary IRB of record for the study. The study was also approved by the University of Vermont, Wake Forest University, and University of Cincinnati.

Diet assessment

Diet data were collected using the Block 98 FFQ, a semi-quantitative, 110-item FFQ that assessed a person’s usual diet over the past year, including frequency of consumption (average times per day, week, or month) and the portion size of specific foods or beverages (e.g., ½ cup of carrots, 2 slices of bacon) [24]. FFQs received by the study center were checked for completeness and scanned. Scanned files were then sent to NutritionQuest (http://www.nutritionquest.com) for analysis of nutrient contents using proprietary algorithms.

Primary exposures

The exposures of interest were empirically derived diet pattern scores. Food and beverage questions from the FFQ were collapsed into 56 investigator-defined individual food groups. A principal components analysis (PCA) was used to derive diet patterns and factor loadings for each of the 56 individual food groups, as described in detail elsewhere [20]. The retained patterns (Convenience, Plant-based, Sweets/Fats, Southern, Alcohol/Salads) were named according to the highest food group loadings within each factor. In general, the Convenience pattern was characterized by high factor loadings for Chinese and Mexican food, pasta dishes, pizza, soup and other mixed dishes including frozen or take-out meals; the Plant-based pattern by fruits, vegetables, and fish; the Sweets/Fats pattern by desserts and other sugary foods; the Southern pattern by added fats, organ meats, fried foods, processed meats, sugar-sweetened beverages and greens typical of southern cuisines; and the Alcohol/Salads pattern by alcohol, green leafy vegetables, tomatoes, and salad dressing. A factor score for each of the patterns was calculated for each study participant by summing observed intakes of component food groups weighted by their respective factor loadings. Factor analysis differs from cluster analysis in that individuals may adhere to more than one dietary pattern identified in this analysis [18].

Ascertainment of outcome

The outcome of interest was sepsis, ascertained via medical record review as described previously [22]. Briefly, trained research staff retrieved and reviewed medical records for all hospital admissions and emergency department visits attributed to a serious infection. Two abstractors independently reviewed all relevant clinical and laboratory data to confirm the presence of a serious infection on initial hospital presentation and the fulfillment of sepsis criteria. The abstractors adjudicated discordances, with additional physician-level review as needed.

Consistent with international consensus definitions [25], sepsis was defined as a presentation to the hospital with a serious infection plus two or more systemic inflammatory response criteria, including (1) heart rate > 90 beats/minute, (2) fever or hypothermia (temperature >38.3 °C or <36 °C), (3) tachypnea (>20 breaths/minute) or pCO₂ < 32 mmHg, and (4) leukocytosis or leukopenia (white blood cells >12,000 or <4,000 cells/mm³ or >10 % band forms). A serious infection was defined according to a previously published taxonomy [1]. Vital status was determined based upon medical chart review. Because of our focus on community-acquired (vs. hospital acquired) sepsis, we utilized the worst diagnostic and laboratory values appearing during the first 28 h of hospitalization, allowing for 4 h of emergency department care plus 24 h of hospitalization. If a participant had more than one sepsis event during follow-up, then we selected the first event for analysis. Classification of sepsis was not based upon ICD-9 discharge diagnoses.

Covariates of interest

Age, race, sex, smoking history, education and annual family income were determined by self-report. Physical activity was assessed through a single question: “How many times per week do you engage in intense physical activity, enough to work up a sweat?” Participants reported weekly television or video watching frequency on a written survey administered during the initial in-person examination with the following possible answers: none, 1–6 h/week, 1 h/day, 2 h/day, 3 h/day, and 4+ hours/day. Abdominal obesity was defined as waist circumference >88 cm for women and >102 cm for men. Hypertension was defined as a systolic blood pressure ≥140 mmHg and/or a diastolic blood pressure ≥90 mmHg, or a self-report of a prior diagnosis of hypertension or current use of anti-hypertensive medications. History of coronary heart disease (CHD) was defined as having any of the following: ECG evidence of myocardial infarction, prior history of a cardiac procedure (coronary artery bypass surgery or percutaneous coronary intervention), or self-reported history of myocardial infarction. Diabetes was as a fasting blood glucose concentration of ≥126 mg/dL, or a non-fasting blood glucose concentration of ≥200 mg/dL, or self-reported use of insulin or oral hypoglycemic agents. Chronic kidney disease (CKD) was defined as an estimated glomerular filtration rate <60 ml/min/1.73 m² or a spot urine albumin to creatinine ratio ≥30 mg/g. REGARDS did not collect information on pulmonary conditions, and therefore we defined chronic lung disease as participant use of pulmonary medications including beta agonists, leukotriene inhibitors, inhaled corticosteroids, combination inhalers, and other pulmonary medications such as ipratropium, cromolyn, aminophylline and theophylline.

Statistical analyses

Follow-up time for each participant was calculated from the date of the in-home visit to the date of death, sepsis or last telephone follow-up, updated through December 31^st, 2012. Descriptive statistics were used to compare baseline characteristics of participants across quartiles of each diet pattern. The Kaplan-Meier method was used to calculate cumulative incidence of sepsis by quartile of each diet pattern, separately. Next, after confirming the proportionality of hazards, Cox regression models were used to estimate the hazard ratio of sepsis as a function of each diet pattern, separately, in sequential models. Model 1 was adjusted for age (continuous), race (black vs. white), sex (male vs. female), geographic region of residence (within vs. outside the Southeastern US defined as residence in stroke belt/buckle states), and total energy intake (continuous). Model 2 was adjusted for variables in Model 1 plus waist circumference, lifestyle factors (self-reported frequency of exercise per week [none, 1 to 3 times per week, or 4 or more times per week], television viewing [none, <4 h per day, 4 or more hours per day], smoking [current, past, or never]), comorbidities (history of heart disease, hypertension, diabetes, chronic pulmonary disease, CKD [all yes vs. no]), annual family income (<$20,000, $20,000–$50,000 or > $50,000/year) and educational achievement (< high school diploma, high school education, >high school education). Cut-points for categorized variables were based upon thresholds used in prior studies [21, 26]. Covariates were selected based on whether they are plausibly related biologically to the diet patterns and with the outcomes of interest based on existing literature. In all Cox models, diet patterns were analyzed in quartiles (with the lowest quartile serving as the referent group) and on a continuous scale. In pre-specified analyses, we examined for effect modification by age, race, sex and diabetes by examining the statistical significance of multiplicative interaction terms in multivariable models. Due to the time lag in observations and medical record retrieval, we could not review medical records for a portion of participants with reported hospitalizations for serious infection. In a sensitivity analysis, we repeated the analyses excluding these individuals. A two-tailed P value <0.05 was considered statistically significant for all analyses, except for tests of statistical interaction (P value <0.10). All analyses were conducted using SAS software version 9.3 (SAS Institute, Cary, NC).