The Nurses’ Health Study data collection
The Nurses’ Health Study started in 1976, when 121
700 female registered nurses aged 30-55 living in one of 11 US states responded to a questionnaire about history of disease and demographic and lifestyle characteristics. Since baseline, follow-up questionnaires have been administered every two years to update the information on incidence of disease and lifestyle and clinical risk factors. Starting in 1980, validated food frequency questionnaires have been administered every two to four years to collect and update dietary intakes of foods and nutrients. Up to 2000 (the year in which health status was determined for the analyses presented here), the follow-up rate of the entire cohort was over 95%.
Self reports of major chronic diseases (such as cancer, diabetes, coronary heart disease, stroke, Parkinson’s disease, and multiple sclerosis) were confirmed through various methods, including review of medical records and pathology reports, telephone interview, and supplementary questionnaires to participants. The high validity and reliability of reported incidence of chronic diseases among these nurses have been previously shown.13 14 15 16
Deaths were identified by reports from next of kin, postal authorities, or by a search of the national death index. At least 98% of deaths among the participants of the Nurses’ Health Study have been identified.17
The SF-36 health status survey was included on the 1992, 1996, and 2000 questionnaires. This 36 item questionnaire measures eight health concepts, including limitations of physical activities, usual role activities, and social activities, as well as mental health, bodily pain, vitality, and the perceptions of general health.18
Its validity and reproducibility have been extensively examined, and it is commonly used to measure quality of life in different populations.18
Finally, to assess cognitive function, beginning in 1995, we identified all nurses who had reached age 70 or older. After exclusion of nurses with a previous diagnosis of stroke, 19
415 (93%) underwent the telephone interview for cognitive status, which is modelled on the mini-mental state examination.19
A strong correlation (correlation coefficient 0.94) was documented between the scores of these two methods.19
Trained nurses who were blinded to the study hypothesis and exposure status of the participants carried out the telephone interviews. The high reliability of the interviewers and the validity of telephone assessments compared with in-person examinations have been previously shown.20
Our current analysis was conducted within this subcohort of the oldest participants of the Nurses’ Health Study who were administered a cognitive function assessment.
Anthropometric measures of adiposity
Weight and height were collected on the baseline questionnaire, and weight was further requested every two years thereafter. Self reported weight was highly correlated (correlation coefficient 0.96) with measured weight in a previous validation study in 184 participants.21
We calculated the body mass index (BMI) as weight in kilograms divided by the square of height in metres (kg/m2
) to measure overall obesity. In 1980, participants were asked about their weight at age 18 (on average 36 years from age 18 to year 1980 for the study participants). The correlation coefficient between recalled weight at age 18 and measured weight in physical examination records at age 18 was 0.87 among 188 participants.22
Data on BMI at age 18 were available for 89%. We used waist circumference (umbilicus), hip circumference (the largest circumference), and waist to hip ratio as measured in 1986 to assess central obesity.
We chose to define adiposity measures at mid-life (that is, study baseline), both because we were interested in the relation of earlier life adiposity to health in later life and also because we were concerned about the possibility of reverse causation—that is, BMI or weight change being a consequence rather than the cause of health problems. Specifically, most of the components of our outcome can have long latency periods, and women beginning to develop these health problems might lose or gain weight. We addressed this potential bias by imposing a long follow-up period between baseline and outcome ascertainment and by excluding women who had had a diagnosis of the chronic diseases in our study outcome at baseline.
Assessment of end points
Although there is no consensus on the definition of successful ageing or healthy survival, the working definitions in most previous studies8 9 11 12
were based on the concept raised by Rowe and Kahn, which incorporates not only chronic diseases but also physical, cognitive, and other functions.23
We used this same concept to derive our comprehensive working definition of healthy survival. Specifically, for our primary definition, healthy survivors were participants who survived to age 70 or older and as of age 70 were free from 11 major chronic diseases—that is, cancer (except non-melanoma skin cancer), diabetes, myocardial infarction, coronary artery bypass graft surgery, congestive heart failure, stroke, kidney failure, chronic obstructive pulmonary disease, Parkinson’s disease, multiple sclerosis, and amyotrophic lateral sclerosis (because cognitive function was assessed near 2000 for 99.1% of the study population, we used the disease status up to 2000 for this domain); had no major impairment of cognitive function; had no major limitation of physical functions; and had good mental health. We defined nurses who survived to the age of ≥70 and did not meet these four criteria as “usual survivors.” In our cohort, there were 1686 (9.9%) “healthy survivors.”
The chosen chronic conditions are major age related diseases and diseases that could severely affect the quality of life among older people. We defined impairment of cognitive function as a score less than 31 points on the telephone interview for cognitive status (about 10% of our population), according to a standard definition of impairment.20
We considered impairment of physical function, based on an existing definition,24
as presence of any of the following limitations as reported by each participant: limited at least “a little” on moderate activities as assessed by the SF-36 (such as moving a table, bowling, or pushing a vacuum cleaner; climbing one flight of stairs; walking more than 1 mile (1.6 km); walking several blocks; bathing or dressing); or limited “a lot” on the SF-36 in more difficult physical performance items (such as running; lifting heavy objects; lifting or carrying groceries; climbing several flights of stairs; bending, kneeling, or stooping). In total, 74% of these older women fulfilled this established definition of physical limitations. Finally, for mental health, we used the SF-36 mental health scale, which combines five questions: have you been a very nervous person?, have you felt so down in the dumps nothing could cheer you up?, have you felt calm and peaceful?, have you felt downhearted and blue?, and have you been a happy person? There were six possible responses to each item, ranging from “none of the time” to “all the time.” Based on the response to these questions, a score between 1 and 6 was assigned to each question, with the score 6 indicating the best mental status and score 1 indicating the worst. We then summed these scores and rescaled them to a range of 0-100.25
Good mental health was defined as a mental health score greater than 84 (the median value in our cohort).
As there is no standard definition of healthy survival, and as the criteria we used for some of our outcomes (such as physical function) might be considered somewhat arbitrary, we investigated the robustness of our definition and further considered an alternative classification of healthy survival. This was similar to our primary definition in being free of the 11 chronic diseases, but we used a different scoring system for defining physical impairment, and we categorised all the domains by median performance: cognitive status score higher than median (≥34), physical function score higher than median (≥75), and mental health score higher than median (≥84). The physical function score in this alternative classification was derived from the responses to the questions in the physical function domain of the SF-36; for each question regarding physical function, a score of 1 was assigned if the response was “yes, limited a lot,” 2 if it was “yes, limited a little,” or 3 if it was “no, not limited at all.” We then summed the score for all questions and rescaled the total score to a range of 0-100. With this alternative definition, 1436 (8.4%) women were categorised as healthy survivors. Of these participants, 882 (61.4%) met the criteria of the primary definition of healthy survival, thus there was some, but far from complete, overlap of the two definitions.
In addition, as BMI might be associated with survival itself4 5 6 7
and our exclusion in the primary analyses of all women who did not survive to age 70 could possibly bias findings, we constructed another secondary outcome of healthy survival. In this secondary outcome, we added the 9352 women in the cohort who did not survive to age 70 to the group of “usual survivors.”
Population for analysis and statistical methods
Our exclusion criteria were a history of major chronic diseases at study baseline in 1976, including cancer, diabetes, myocardial infarction, coronary artery bypass graft surgery, stroke, kidney failure, chronic obstructive pulmonary disease, Parkinson’s disease, multiple sclerosis, or amyotrophic lateral sclerosis; missing BMI at baseline; or no data on cognitive function or missing data for more than two items on the mental health scale or for more than five items on the physical function scale in the SF-36. After we excluded these participants, data from 17
065 women were available for analysis.
As the waist and hip circumferences were first assessed in 1986, we used 1986 as the study baseline for the central obesity analysis and applied the same exclusion criteria. The study population for this analysis was, therefore, a subset of the primary study population (9512 for waist circumference; 9450 for hip circumference; 9438 for waist to hip ratio).
For analysis of BMI, we grouped the nurses into six categories according to their baseline BMI: <18.5, 18.5-22.9 (reference), 23.0-24.9, 25.0-26.9, 27.0-29.9, and ≥30. For analysis of weight change, we calculated weight change between age 18 and 1976 and grouped the women into five categories: lost ≥4.0 kg, stable weight (reference), gained 4.0-9.9 kg, gained 10.0-14.9 kg, gained 15.0-19.9 kg, and gained ≥20 kg. The cut-off points for the highest categories of waist circumference (≥88 cm) and waist to hip ratio (≥0.80) were based on WHO recommendations.26
The cut-off points for the lower four categories of waist circumference and waist to hip ratio were based on quartiles of these measurements among the remaining participants. The cut-off points for hip circumference were based on quintiles.
We used logistic regressions to model the associations of each risk factor variable and the odds of healthy versus usual survival. In the current analysis, an odds ratio less than 1 denotes an “undesirable” association or reduced odds of healthy survival associated with the risk factor, while an odds ratio larger than 1 denotes a “desirable” association or an increased odds of healthy survival. In the multivariable analysis, we adjusted for baseline variables, including age at baseline (year), education (registered nurse certificate, bachelor’s degree, master’s degree, or doctoral degree), husband’s education (less than high school, some high school, high school graduate, college graduate, or graduate school), marital status (married, widowed, separated/divorced/never married), postmenopausal hormone use (never used, past user, or current user), smoking status (never smoked, past smoker, current smoker of 1-14 cigarettes a day, 15-24 cigarettes a day, or ≥25 cigarettes day), family history of heart disease (yes, no), family history of diabetes (yes, no), family history of cancer (yes, no), physical activity (hours a week), ratio of intake of polyunsaturated to saturated fat (in fifths), intakes of trans fat, alcohol, and cereal fibre (all in fifths), and intakes of fruits, vegetables, and red meat (in thirds) to control for confounding. When examining the associations for weight change, we further adjusted for BMI at age 18. In analyses of waist and hip circumference, we adjusted for BMI in 1986 and mutually adjusted for waist and hip circumferences.
Cigarette smoking could reduce body weight and has strong effects on overall health.5
To account for the possibility of residual confounding by smoking we conducted a secondary analysis among women who had never smoked. In an additional secondary analysis, we adjusted for potential confounding factors defined at age 70 rather than at baseline, but this did not materially change our findings. Finally, we repeated the analyses using the alternative definitions of healthy survival and usual survival as described above.
In our examination of the joint associations of BMI at age 18 and weight change from age 18 to baseline, we included only those who had stable weight or gained weight since age 18 and whose BMI was no less than 18.5 to yield more stable estimates because only a small number of the nurses lost more than 4 kg body weight or were underweight at age 18. A secondary analysis showed that including these women did not change the results materially. We used likelihood ratio tests to evaluate the significance of interactions between BMI and weight change. These tests are based on the difference of −2 log likelihood of models with and without interaction terms and follow the χ2 distribution with the degree of freedom equal to the number of parameters in the interaction terms.
All p values were two sided. Odds ratios were calculated with 95% confidence intervals. Data were analysed with the SAS software package, version 9.1 (SAS Institute, Cary, NC).