A Prospective Study of Serum 25-Hydroxyvitamin D Levels and Mortality Among African Americans and Non-African Americans

Lisa B. Signorello; Xijing Han; Qiuyin Cai; Sarah S. Cohen; Elizabeth L. Cope; Wei Zheng; William J. Blot


Am J Epidemiol. 2013;177(2):171-179. 

In This Article

Materials and Methods

The Southern Community Cohort Study (SCCS) was approved by institutional review boards at Vanderbilt University and Meharry Medical College, and all participants provided written, informed consent.

Study Population

The SCCS is a prospective cohort study designed to investigate racial disparities in cancer that enrolled over 85,000 men and women in 12 southeastern US states from 2002 to 2009. Comprehensive study details are provided elsewhere.[23] In brief, individuals were eligible for enrollment if they were aged 40–79 years, English speaking, and had not been under treatment for cancer within the past year. Approximately two-thirds of SCCS participants self-reported their race as African American; the remainder were mostly non-Hispanic white. Recruitment took place primarily (86%) at community health centers, with a smaller proportion (14%) enrolling through the mail in the same geographic area. This analysis was restricted to community health center-enrolled participants, as blood samples were not collected from participants who enrolled through the mail.

At the community health centers, trained study interviewers administered a computer-assisted personal interview to collect baseline data on demographic characteristics, lifestyle factors, body size, and medical history. More than half (54%) of the participants provided a 20-mL nonfasting baseline blood sample, which was kept chilled until processing and freezing at −80°C the next day.

Case Identification and Control Selection

The cohort was followed prospectively for mortality via linkage to the Social Security Administration's Death Master File and the National Death Index. Excluding deaths that occurred within 1 year of enrollment, we identified 3,953 deaths as of April 2010. Of these, 1,854 had a serum sample in frozen storage, and 25(OH)D measurement (described below) was successful for 1,852, forming the case group for this analysis with death dates ranging from May 2003 to February 2010. These deaths occurred an average of 3.6 (range: 1.0–7.8) years after study enrollment. Controls, who were alive on the death date of their corresponding case, were individually matched on sex, race (black/white/other), age at enrollment (±3 years), community health center enrollment site, and date of blood collection (±6 weeks). The latter 2 matching criteria provided tight control of the geographic location and season of blood collection. The mean absolute difference in the date of blood collection between the cases and their matched controls was 2.5 (standard deviation: 1.8) weeks. Matching on exact enrollment site could not be achieved for 10.6% of the cases, nor matching on exact race for 0.7% of the cases, but analyses restricted to the 89% with exact matching for all criteria provided the same overall results.

Of the 1,852 deaths studied, 1,798 occurred prior to January 1, 2010, the time period for which the National Death Index had cause of death data available. We obtained the underlying cause of death from the National Death Index for 1,774 (98.7%) of the deaths in this time period and created classifications based on the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) (version for 2007), as follows: cancer death (n = 477) (ICD-10 codes C00–C97); circulatory disease death (n = 537), which included hypertension (n = 86) (ICD-10 codes I10–I13), ischemic heart disease (n = 260) (ICD-10 codes I20–I25), pulmonary heart disease (n = 14) (ICD-10 codes I26–I28), cardiac arrest and arrhythmias (n = 43) (ICD-10 codes I44–I49), heart failure (n = 34) (ICD-10 code I50), cerebrovascular disease (n = 90) (ICD-10 codes I60–I69), atherosclerosis, and other diseases of the arteries, arterioles, and capillaries (n = 10) (ICD-10 codes I70–I78); and all other nonexternal causes (n = 621), which included all other deaths except those already classified as a cancer or circulatory disease death, or having ICD-10 codes beginning with S, T, V, W, X, or Y.

Serum 25(OH)D Measurement

Serum 25(OH)D levels were measured at Heartland Assays, Inc. (Ames, Iowa), by using a US Food and Drug Administration-approved direct, competitive, chemiluminescence immunoassay, the DiaSorin LIAISON 25 OH Vitamin D TOTAL assay (DiaSorin, Stillwater, Minnesota), which is co-specific for 25-hydroxyvitamin D2 and D3.[24,25] Among 10 (blinded) triplicate sets of identical serum samples included for quality control, the average intraassay coefficient of variation was 4.2%.

Statistical Analysis

Serum 25(OH)D quartiles were determined on the basis of the distribution among the controls, with 25%, 50%, and 75% cutoffs of 10.18, 15.15, and 21.64 ng/mL. Although the distributions of 25(OH)D measurements varied by race, there were sufficient overlap and adequate numbers of African Americans and non-African Americans across each quartile to avoid the need for race-specific quartiles. Mean 25(OH)D levels within each quartile were nearly identical by sex (quartile 1: 7.5, 7.2; quartile 2: 12.6, 12.4; quartile 3: 18.2, 18.3; quartile 4: 29.6, 28.9 ng/mL for males and females, respectively) and by race (quartile 1: 7.3, 7.5; quartile 2: 12.5, 12.5; quartile 3: 18.1, 18.5; quartile 4: 29.3, 29.5 ng/mL for African Americans and non-African Americans, respectively).

Mean 25(OH)D levels were compared across participant characteristics by using 1-way analysis of variance or t tests. Participant characteristics were also compared by case-control status by Pearson's χ2 test. Conditional logistic regression was used to estimate odds ratios and 95% confidence intervals for all-cause and cause-specific mortality in relation to quartiles of serum 25(OH)D. Matching factors were accounted for in the conditional analysis, and additional covariates included body mass index (using a continuous and quadratic term), smoking status (never, former smoker of below-median pack-years, former smoker of above-median pack-years, current smoker of below-median pack-years, current smoker of above-median pack-years), annual household income (<$15,000, $15,000–$24,999, $25,000–$49,999, ≥$50,000), and total work and leisure time physical activity (in quartiles of metabolic equivalent-hours/day). Other potential confounders that were examined but not retained in the final models included education, total energy intake, height, alcohol intake, health insurance status, marital status, and (for women) use of hormone replacement therapy. Time participating in sports, because of its potential for a stronger relation with outdoor exposure, was tested as a replacement for total physical activity in the models, but the results did not change. To examine the shape of the continuous relation between 25(OH)D and mortality risk, we also utilized fully adjusted restricted cubic spline models.

Interaction by race, sex, and body mass index was assessed by using the likelihood ratio test to compare models with and without the relevant interactions terms. We found no significant interaction by sex or by body mass index. Where we found significant interaction by race, race-specific results are presented. African-American versus white results were nearly identical to African-American versus non-African-American results, so only the latter are presented.

All analyses were performed by using SAS, version 9.2, statistical software (SAS Institute, Inc., Cary, North Carolina). All P values are 2 sided.