|Home | About | Journals | Submit | Contact Us | Français|
Although underweight and obesity have been associated with increased risk of mortality, it remains unclear whether the associations differ by hormone therapy (HT) use and smoking. The authors examined the relationship between body mass index (BMI) and mortality within the California Teachers Study (CTS), specifically considering the impact of hormone therapy (HT) and smoking. The authors examined the associations of underweight and obesity with risks of all-cause and cause-specific mortality, among 115,433 women participating in the CTS, and specifically examined whether HT use or smoking modifies the effects of obesity. Multivariable Cox proportional hazards regression provided estimates of relative risks (RRs) and 95% confidence intervals (CIs). During follow up, 10,574 deaths occurred. All-cause mortality was increased for underweight (BMI < 18.5; adjusted relative risk [RR] = 1.33, 95% confidence interval [CI] =1.20–1.47) and obese participants (BMI ≥ 30: RR = 1.27, 95% CI = 1.19–1.37) relative to BMI of 18.5 – 24.9). Respiratory disease mortality was increased for underweight and obese participants. Death from any cancer, and breast cancer specifically, and cardiovascular disease was observed only for obese participants. The obesity and mortality association remained after stratification on HT and smoking. Obese participants remained at greater risk for mortality after stratification on menopausal hormone therapy and smoking. Obesity was associated with increased all-cause mortality, as well as death from any cancer (including breast), and cardiovascular and respiratory diseases. These findings help to identify groups at risk for BMI-related poor health outcomes.
The prevalence of obesity has been rising in the United States and other western countries, as well as in developing nations 1–7. Previous population-based research has linked high body mass index (BMI, kg/m2), a measure of obesity, to all-cause mortality 8–14, as well as death from all cancers combined 10, 12, 15-18, and cardiovascular disease (CVD) 10, 12, 17, 19–21. Low BMI has been associated with deaths due to respiratory disease, although this relationship is confounded by smoking-related morbidity 10. BMI in adulthood and weight change throughout life have gender-specific effects on mortality 2, 6, 12-14, 18, 22, 23.
A recent collaborative cohort analysis of over 900,000 adults indicated that high BMI are associated with statistically significant changes in blood pressure, lipoprotein levels, and diabetes 10. Weight loss lowers blood pressure, decreases serum lipids, cholesterol, and blood glucose 1, and the risk of vascular disease 24, 25. Women with high BMI are at increased risk of coronary heart disease 19 and colon cancer 26, and tend to have more advanced stages of breast cancer at diagnosis 27. Obesity is known to increase the risk of reproductive cancers in women 28 possibly due to greater concentrations of free estradiol and free testosterone in overweight and obese women 28. However, postmenopausal women using menopausal hormone therapy (HT) have shown a reduction in all-cause mortality 29. Therefore, it is important to explicitly consider history of HT use, as well as potential confounding factors when examining the association between BMI and mortality in women.
The aim of this paper is to investigate the relationship of BMI to all-cause and cause-specific (CVD, any cancer, breast cancer, and respiratory disease) mortality in the California Teachers Study (CTS). In addition we specifically looked at the impact of HT use and smoking on the obesity and mortality relationships.
The CTS is a prospective cohort study of 133,479 female California public school teachers and administrators who were enrolled in the California State Teachers Retirement System. A detailed account of the methods employed by the CTS has been published elsewhere 30. Participants completed a baseline questionnaire and returned it by mail during 1995–1996 30. The baseline survey collected information on height, weight, history of oral contraceptive use, HT use, reproductive history, co-morbidities (including diabetes, heart attack, hypertension, stroke, and cancer) and family history of disease, as well as behavioral factors such as physical activity, caloric intake, alcohol consumption, and smoking. Institutional Review Board approval for the CTS was obtained from all participating institutions.
CTS participants were excluded sequentially in this analysis if they did not report their weight or height at baseline (n=6,235); if their BMI (kg/m2) value was implausibly low or high (n=61); if they were older than 95 years at baseline (n=58); if they provided insufficient dietary information (n=11,623); and if they asked to be included only in breast cancer research (n=10). Participants who died within the first year of follow-up were also excluded (n=59), since weight loss and underweight may have resulted from pre-existing illness. After all exclusions, 115,433 participants remained eligible for this analysis.
Participants contributed person-years of follow-up from the date of baseline questionnaire completion until the date of death or date of a censoring event (including date of a move outside of the United States, date of a death due to accident or suicide, or the end of follow-up on December 31, 2007). Deaths were identified via annual linkage with the State of California mortality files, the Social Security Administration Death Master File and the National Death Index. In this study we used cause of death information recorded on the death certificates as follows: deaths from any cancer (International Classification of Disease (ICD) codes: ICD-9 = 140–208, ICD-10 = C00-99.9, breast cancer (ICD-9=174, ICD-10=C50), or respiratory disease (ICD-9=460–519, ICD-10=J00-98). Cancer, CVD, and respiratory illness were chosen because they are major groupings of possible causes of death. Breast cancer was also included in this analysis because it is a particular focus of the authors and the CTS in general.
For cause-specific mortality analyses, we excluded participants reporting a co-morbidity at baseline (i.e., prevalent cases) relating to the cause of death in the analysis, in order to limit survivor advantage of those cases. For CVD mortality, we excluded participants who reported a diagnosis of heart attack, stroke, and/or high blood pressure at baseline (n=1,511). For any cancer, we excluded participants reporting any cancer diagnosis at baseline (n=12,318), while for breast cancer mortality we excluded participants who reported a breast cancer diagnosis at baseline (n=5,500). For death due to respiratory illness, we did not exclude any additional participants at baseline beyond those excluded for the all-cause mortality analysis.
BMI at baseline was calculated from self-reported height and weight and was categorized based on the World Health Organization classification 2 as <18.5 kg/m2 (underweight), 18.5–24.9 kg/m2 (normal weight), 25.0–29.9 kg/m2 (overweight), and ≥30 kg/m2 (obese). Weight change from age 18 to baseline was classified into four groups: decrease of more than 5 kg, no change (defined as +/− 5 kg), an increase of 5.1–10 kg, and >10 kg gain.
History of diabetes, hypertension, heart attack, and stroke, were obtained from self-report in the baseline questionnaire and supplemented by diagnoses listed in California Office of Statewide Health Planning and Development (OSHPD) hospital discharge records from 1991 until questionnaire completion 31. Prevalent cancer diagnoses were obtained from self-report in the baseline questionnaire and supplemented by linkages with the California Cancer Registry, which receives reports of over 99% of cancer diagnoses occurring in California 32.
Detailed information on recreational physical activity was collected in the baseline questionnaire. Physical activity was reported as the average number of hours per week of strenuous or moderate exercise. Physical activity measures in the CTS have been further described elsewhere 33.
Average daily percentage of calories from fat was determined from the food-frequency portion of the baseline questionnaire 34, categorized into 4 groups: <25%, 25–29.9%, 30–34.9%, and ≥35%. Participants were classified as never smokers, former smokers, current smokers, or unknown smoking status. Alcohol use was based on the average number of alcoholic drinks consumed per day, and grouped into three categories: none, <20 grams/day, or ≥20 grams/day. Participants’ self-reported race/ethnicity was categorized as non-Hispanic white, African American, Hispanic/Latina, Native American, Asian/Pacific-Islander, or other/unknown.
Participants were classified as either premenopausal or postmenopausal, with perimenopausal participants included in the postmenopausal group. Participants were classified as perimenopausal if they reported that their periods had stopped in the previous six months. Participants were classified as postmenopausal if they: reported natural menopause more than 6 months prior to baseline (n=34,147); had a bilateral oophorectomy (n=11,597); were 56 years or older and did not report being premenopausal (n=7,523); or if their periods had stopped due to other reasons including pituitary adenoma, medication, chemotherapy, radiation treatment, or another reason (n=4,902). The criterion for classifying women as menopausal at age 56 years or older, used in recent CTS analyses 35, 36, is based on previous work indicating that among women with natural menopause 97% were postmenopausal by age 56 years 37. Postmenopausal participants were grouped according to their HT use at baseline as never HT user, former user of any HT, current user of estrogen only, current user of estrogen and progestin, or unknown hormone therapy use 35.
Characteristics of participants at baseline were compared across the subgroups defined by BMI at baseline (<18.5, 18.5–24.9, 25–29.9, ≥30 kg/m2) using the chi-square statistic.
Multivariable Cox proportional hazards regression models were used to estimate the hazard rate ratio, a measurement of relative risk (RR), and 95% confidence intervals (CI) for the association of BMI with all-cause and cause-specific mortality. All models used age at cohort entry and age at death or censoring event as the time scale and were stratified by age at cohort entry. Four models were fit to examine the association between BMI and all-cause mortality. Model 1 was adjusted for race/ethnicity; Model 2 was additionally adjusted for HT use); Model 3 was further adjusted for physical activity; Model 4 was additionally adjusted for co-morbidities (including diabetes, hypertension, heart attack, stroke, and any cancer), smoking status, alcohol consumption, and percent of daily calories from fat; and Model 5 was further adjusted for weight change from age 18 to baseline. Finally, to examine the impact of smoking history and use of menopausal hormone therapy on the relationship between BMI and mortality (all-cause, and cause-specific) we stratified the analysis, using Model 5, by smoking history (never, ever), and HT history (never, ever).
The cohort consisted of 115,433 eligible women who were followed for 1,322,634 person-years. Overall, 10,594 women died during the follow-up time period. Causes of death for this study included any cancer (N=2,292); breast cancer (N=302); CVD (N=3,626); and respiratory disease (N=946). The average age at baseline was 53.1 years (standard deviation = 14.1 years).
Table 1 provides the age-adjusted distribution of several participant characteristics according to BMI at baseline. Approximately 60% of participants with normal BMI (18.5–24.9 kg/m2) at age 18 still had normal BMI at cohort entry. Approximately 80% of participants who gained more than 10 kg since age 18 were overweight or obese at cohort entry. Higher BMI was observed among African-American women (and Native American women to a smaller extent), postmenopausal women (especially those who never used HT), women reporting fewer than 3 hours per week of strenuous or moderate physical activity, never drinkers of alcohol, never or former smokers, women with greater percentage of their daily calories from fat, and women with one or more co-morbidities.
Risk for all-cause mortality was statistically significantly increased for underweight participants (RR = 1.33, 95% CI = 1.20–1.47) and for obese participants (RR = 1.27, 95% CI = 1.19–1.37) compared to normal weight participants after adjustment for age, race/ethnicity, weight change from age 18 to baseline, physical activity, history of co-morbidities, smoking status, alcohol consumption, HT, and percent daily calories from fat (Model 5, Table 2). HT use (Table 2, Model 2 compared to Model 1) did not affect the association between BMI and all-cause mortality for any BMI category. Overweight participants did not significantly differ from normal weight participants with respect to all-cause mortality.
In a restricted model, we excluded participants who reported having had a defined co-morbidity at baseline (diabetes, hypertension, heart attack, stroke, and cancer). Among the 71,144 remaining participants, the mortality risk estimate for underweight and for obese participants (RR = 1.31, 95% CI = 1.09–1.58, and RR = 1.32, 95% CI = 1.15–1.53, respectively) were only slightly attenuated (data not shown).
Compared to normal weight participants, underweight participants were at increased risk of death due to due to respiratory disease (RR = 1.79, 95% CI = 1.36–2.34) (Table 3). Obese participants, compared to participants of normal weight, were at an increased risk of death due to any cancer (RR =1.32, 95% CI = 1.14–1.53), breast cancer (RR =1.57, 95% CI = 1.07–2.31), CVD (RR =1.38, 95% CI = 1.22–1.55), and respiratory disease (RR =1.46, 95% CI = 1.15–1.85).
After stratification by smoking status (Table 4), obese never smokers had increased risk of all cause mortality (RR=1.32, 95% CI=1.21–1.45), any cancer (RR=1.48, 95% CI= 1.21–1.82), and CVD (RR=1.50, 95% CI= 1.29–1.75) as compared to normal weight never smokers; but for obese ever smokers only risk of all-cause mortality was statistically significantly increased (RR= 1.13, 95% CI= 1.02–1.26) as compared to normal weight ever smokers. Never smokers who were underweight were at increased risk for all-cause mortality (RR =1.34, 95% CI = 1.17–1.54), and for CVD mortality (RR =1.34, 95% CI = 1.07–1.68), as compared to normal weight never smokers. For smokers, being underweight was associated with risk only for all-cause mortality (RR=1.41, 95% CI = 1.21–1.65) and respiratory disease mortality (RR = 2.22, 95% CI = 1.58–3.11). The analyses for underweight smokers and non-smokers for any cancer and breast cancer were based on small numbers.
After stratification by history of hormone therapy use (Table 5), obese never users of HT had increased risk of all cause mortality (RR=1.27, 95% CI=1.12–1.44), any cancer (RR=1.38, 95% CI= 1.03–1.84), and CVD (RR=1.41, 95% CI= 1.15–1.74) as compared to normal weight never users. Obese HT users had statistically significantly increased risk of all-cause mortality (RR= 1.19, 95% CI= 1.08–1.32), and any cancer mortality (RR=1.23, 95% CI=1.00, 1.50), as compared to normal weight ever users. Never users of HT who were underweight were at increased risk for all-cause mortality (RR =1.22, 95% CI = 1.02–1.47), and for respiratory disease (RR =1.97, 95% CI = 1.27–3.07), as compared to normal weight never users. For HT users, being underweight was associated with risk for all-cause mortality (RR=1.47, 95% CI = 1.27–1.69) and respiratory disease mortality (RR = 1.90, 95% CI = 1.28–2.82).
Further stratification on current smoking status (former and current) or type of HT use (estrogen-only and estrogen plus progestin) revealed similar associations, however the division into more strata resulted in small numbers of events in several categories of BMI (data not shown).
Being underweight or obese was associated with an increased risk of all-cause mortality among participants in the CTS. A similar U-shaped association of BMI with all-cause mortality has been observed in other studies 8, 10, 22, 38, 39. Control of HT use in our statistical models did not significantly alter the association between BMI and all-cause mortality overall or among postmenopausal women. When we included weight change from age 18 to baseline in our all-cause mortality analysis, the risk of death in obese participants slightly increased above the level of the HT-adjusted model, but was not as high as in the unadjusted model. Obese participants also had an increased risk of death from any cancer (including breast cancer), CVD, and respiratory disease; over weight participants (those with BMI of 25–29.9) did not differ in mortality risk from normal weight participants; underweight participants had increased risk of death due to respiratory disease.
A variety of biological mechanisms may be responsible for linking obesity with mortality. A large prospective cohort study recently linked high BMI with high systolic blood pressure and high lipoprotein levels 10, and several other studies have linked high BMI to vascular disease 24, 25, 40. Obesity has also been linked to increased risk of cancer 26, 41, 42. Strategies to reduce obesity, such as through physical exercise, also have apparent anti-carcinogenic effects by decreasing fat deposits, changing sex-hormone levels, improving immune function, and/or reducing inflammation 43. Obesity in postmenopausal women is associated with an increased risk of breast cancer 44, 45 perhaps because overweight and obese postmenopausal women have higher levels of circulating estrogens and androgens 46–48. Cancers of the endometrium, large intestine, pancreas, and kidney, have also been associated with obesity, implicating a variety of biologic pathways in which body adiposity can impact apoptosis and cell proliferation in target cells 10, 48. Beyond cancer risk, high BMI is a risk factor for cancer mortality 15. On the other hand, low BMI and decreased adiposity has been implicated in increased mortality from respiratory illness (especially chronic obstructive pulmonary disease or COPD) 49.
We addressed smoking as a possible confounder by adjusting for smoking status in our multivariate models. The CTS is largely a non-smoking cohort (only about 5% of participants were current smokers at baseline and about 30% were former smokers), so the adjustment had little impact on risk estimates. However, we did look at the effect of BMI on mortality separately for smokers (current and former combined) and for never smokers. A similar U-shaped all cause mortality curve for never smokers and smokers was found, likely due to large number of former smokers in the smoking category. Moreover, we observed that history or type of HT use did not modify risk of mortality among post-menopausal women.
Limitations of the present study include possible misclassification of exposures collected by self-report, including BMI at baseline and at age 18, smoking status, HT use, recreational physical activity, nutritional intake, and co-morbid conditions. However, any misclassification should be non-differential with respect to mortality outcomes because these data were collected before the patients who died were even diagnosed with the diseases of interest. A second limitation is that changes in BMI occurring after baseline, but before death or end of follow-up were not available and BMI at baseline may not reflect BMI at the time of diagnosis. Third, the CTS represents a sample of California women who are public school professionals and may not be fully representative of women across the United States. For example, the mean BMI in the CTS is slightly lower than that of a comparable age and race-matched California population (24.8 versus 25.7 kg/m2, respectively) and fewer CTS participants were current smokers compared to this same population 30. Fourth, our co-morbidity variables are based on self-report at baseline and diagnoses listed in California OSHPD hospital discharge records from 1991 through baseline. Using the OSHPD data to supplement reported co-morbidities may miss the diagnoses that did not require hospitalization. Fifth, cause of death may not be assigned in a consistent manner across the cohort. Finally, some of the stratified analyses were based on small numbers. Based on considerations of sample size in cohort studies by Breslow and Day 50, we can detect relative risks of around 1.30 with significance level at 5% and power at 80% in strata with about 200 events, indicating that all of our all-cause and cause-specific analyses, and most of the smoking- and HT- stratified analyses are sufficiently powered. However, for less common outcomes such as breast cancer, some of the smoking- and HT-stratified analyses may be underpowered in BMI categories with few events, thus formal tests of interaction may also be underpowered. Nevertheless, the large and well-characterized CTS cohort provided a resource of women in which BMI could be assessed in relation to mortality after adjusting for potentially relevant confounding exposures including HT, diet, smoking and alcohol use.
This analysis confirms an increased risk of all-cause mortality for underweight and obese women in the CTS, that is independent of HT use and smoking history; further neither of exposures served to modify the main effects of BMI on risk of dying. Low BMI was associated with increased risk of death due to cancer and respiratory disease, while obesity was associated with increased risk of death from all cancers combined, breast cancer, CVD, respiratory disease, and diabetes. Further evaluation of modifiable risk factors contributing to women’s BMI and risk of mortality from specific causes is warranted. This information may help inform groups who are at risk for CVD, cancer, respiratory disease and diabetes that may benefit from careful monitoring for these conditions as well as preventive interventions. With increasing obesity in the United States we may begin to see diminishing life expectancy in the future, further implicating obesity as a major contemporary public health problem.
The authors would like to thank the California Teachers Study Steering Committee who are responsible for the formation and maintenance of the cohort within which this study was conducted but who did not directly contribute to the current paper: Ellen T. Chang, Christina A. Clarke, Dennis Deapen, Pamela L. Horn-Ross, James V. Lacey Jr, David O. Nelson, Susan L. Neuhausen, Rich Pinder, Peggy Reynolds, Fredrick Schumacher, Daniel O. Stram, Giske Ursin, and Sophia S. Wang.
Grant support: This work was supported by a National Cancer Institute grants [grant numbers R01 CA77398 and R01 CA 58860].The collection of cancer incidence data used in this study was supported by the California Department of Health Services as part of the statewide cancer reporting program mandated by California Health and Safety Code Section 103885; The ideas and opinions expressed herein are those of the authors. Endorsement by the National Cancer Institute and the State of California, Department of Health Services, is not intended nor should be inferred.