Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Apr 1.
Published in final edited form as: Ann Epidemiol. 2021 Jan 22;56:18–25. doi: 10.1016/j.annepidem.2021.01.003

Life-long Body Mass Index Trajectories and Mortality in Two Generations

Hui Zheng 1, Paola Echave 2, Neil Mehta 3, Mikko Myrskyla 4
PMCID: PMC8009819  NIHMSID: NIHMS1665752  PMID: 33493649

Abstract

Purpose:

To identify life-long body mass index (BMI) trajectories across two related generations and estimate their associated mortality risks and population attributable deaths.

Methods:

We use prospective cohort data from the Framingham Heart Study (1948-2011) original (4576 individuals, 3913 deaths) and offspring (3753 individuals, 967 deaths) cohorts and latent trajectory models to model BMI trajectories from age 31 to 80 years. Survival models are used to estimate trajectory-specific mortality risk.

Results:

We define seven BMI trajectories among original cohort and six among offspring cohort. Among original cohort, people who are normal weight at age 31 years and gradually move to overweight status in middle or later adulthood have the lowest mortality risk even compared to those who maintain normal weight throughout adulthood, followed by overweight stable, lower level of normal weight, overweight downward, class I obese upward, and class II/III upward trajectories. Mortality risks associated with obesity trajectories have declined across cohorts, while the prevalence of high-risk trajectories has increased.

Conclusions:

The mortality impact of weight gain depends on an individual’s BMI trajectory. Population attributable deaths associated with unhealthy weight trajectories have grown over generations because the prevalence has increased, offsetting the decline in trajectory-specific mortality risks.

Keywords: Body mass index trajectories, Mortality, Obesity, Cohort, Framingham Heart Study, United States

As one of the most urgent epidemics in many societies, obesity has emerged as a key risk factor for many illnesses (e.g., diabetes, cardiovascular disease, stroke and certain cancers), morbidities, functional limitation, and mortality (1–16). The extent of obesity’s negative impact accumulates over the life course and depends on timing of onset of obesity and duration of excess body mass (12, 16). Therefore, it is essential to utilize life history information on body mass index (BMI; weight (kg)/height (m)2) to estimate the impact of obesity on disease and mortality. A growing body of literature has done so and highlighted that: (1) a dynamic measure capturing weight status changes (e.g., weight loss, large weight gain) is more predictive of disease and mortality than a static measure of weight status (e.g., baseline BMI) (1–11); (2) obesity increases the risk of mortality more profoundly when it persists over the life course (3, 12); and (3) attributable mortality risk due to obesity is larger when using obesity trajectories rather than a static measure of obesity (3). These literatures are insightful in emphasizing the importance of dynamic BMI history on health and mortality, but life-course BMI dynamics have remained poorly characterized and it remains unclear how such dynamics relate to health and mortality.

Several strategies to model the impact of weight status histories on mortality have been utilized (10–24), e.g., strictly adding BMI statuses from different lifetime points in the hazard model, constructing duration of obesity, and specifying the amount of weight changes. These approaches usually utilize limited time points of BMI information and are based on strong assumptions, e.g., BMI statuses from different lifetime points independently affect mortality, the effect of duration of obesity does not depend on timing of onset of obesity, or cutoff points in weight change are specified arbitrarily (10). Some approaches recognize that obesity levels from different lifetime points may be interactive in their effects and include interactions among BMI statuses to the models (22–24). Including interactions among BMI status at different times in the life course improves model fit, but the interpretation of these interactions is not straightforward, especially when BMI history includes multiple time points.

Recent studies have employed an alternative modeling strategy, namely the implementation of semiparametric group-based trajectory models (mixture models) (25–27), to capture latent BMI trajectories throughout the life course. This strategy considers the complex structure of BMI histories by accounting for initial BMI status, and linear and non-linear trajectories. It can easily take advantage of BMI statuses from multiple time points, avoid arbitrary cutoff points in weight changes, capture the onset and duration of each BMI status, recognize the linkage among these BMI statuses over the life course, and segregate individuals into distinct life history trajectories. These studies, however, have focused either on older (3, 28, 29) or younger populations (30, 31) due to data limitations. To extend these studies, we use the same analytical approach but employ relatively complete cohort data beginning at age 31 from the Framingham Heart Study (FHS) original and offspring cohorts to model life-long BMI trajectories and estimate trajectory-specific mortality risks. This study is the first to examine the association of life-long BMI trajectories with mortality and the first to do this across two generations, which should uncover how BMI evolves over the whole adulthood, provide a more accurate estimate of the mortality consequence of obesity than prior studies, and yield insights on the dynamics of this relationship over time.

METHODS

Data

FHS began in 1948 with a sample of adults in Framingham, Massachusetts (32–34). Beginning in 1971, the children of the original cohort and their spouses were enrolled and constituted the offspring cohort. The original and offspring cohorts are mostly non-Hispanic Whites. Blacks, Hispanics, and other minorities are part of the Omni cohort (506 individuals), which are not included in this analysis. The original cohort consisted of 5079 men and women, 28-74 years of age at the onset of the study. Participants were examined every 2-3 years from 1948 to 2010, for a total of 30 exams. Height and weight were measured at each clinical exam. Participants in the offspring cohort were examined every 3-4 years from 1971 to 2014, for a total of 9 exams. However, detailed height and weight information was collected across exams 2 to 7 (1979-2001) and our analysis was restricted to these six exams. The offspring cohort consisted of 5013 participants, 22-67 years of age at the onset of exam 2, the first study exam used here. After removing individuals with missing data on the key variables, we arrived at a final sample size of 4576 individuals for the original cohort and 3753 individuals for the offspring cohort. Supporting Information 1 provides the descriptive characteristics of the excluded sample, which is generally similar to those included in the study (Table 1).

Table 1.

Descriptive Characteristics of Analytic Sample, Framingham Heart Study.

N
 Mean (or %)
 N
 Mean (or %)
Original cohort Offspring cohort
Time-invariant variables (N=4,576 individuals) Time-invariant variables (N=3,753 individuals)
Birth cohorts Birth cohorts
 1876 to 1894 524 11.45%  1903 to 1924 500 13.32%
 1895 to 1899 676 14.77%  1925 to 1929 539 14.36%
 1900 to 1904 690 15.08%  1930 to 1934 571 15.21%
 1905 to 1909 818 17.88%  1935 to 1939 536 14.28%
 1910 to 1914 920 20.10%  1940 to 1944 676 18.01%
 1915 and above 948 20.72  1945 to 1949 555 14.79%
 1950 and above 376 10.02%
Gender Gender
 Male 2,079 45.43%  Male 1,813 48.31%
 Female 2,497 54.57%  Female 1,940 51.69%
Educational attainment Years of education
 Less than high school 1,930 42.18%  0-11 292 7.78%
 High school graduate 1,358 29.68%  12 1,293 34.45%
 Some college 713 15.58%  13-15 1,007 26.83%
 College graduate 575 12.57%  16 and above 1,161 30.94%
Time-variant variables (N=44,261 observations) Time-variant variables (N=19,067 observations)
BMIab 39,736 26.12 BMIab 18,822 27.01
Body type categoriesa Body type categoriesa
 Underweightc 601 1.51%  Underweightc 158 0.84%
 Normal weightd 16,428 41.34%  Normal weightd 6,860 36.45%
 Overweighte 16,503 41.53%  Overweighte 7,597 40.36%
 Class I obesityf 4,941 12.43%  Class I obesityf 3,030 16.10%
 Class II/III obesityg 1,263 3.18%  Class II/III obesityg 1,177 6.25%
Age 44,261 64.73 Age 19,067 53.56
Smoking behavior Smoking behavior
 Non-smoker 29,805 67.34%  Non-smoker 14,680 76.99%
 Low smoking (1-9 cigarettes) 2,842 6.42%  Low smoking (1-9 cigarettes) 636 3.34%
 Moderate smoking (10-19 cigarettes) 3,072 6.94%  Moderate smoking (10-19 cigarettes) 770 4.04%
 Heavy smoking (20 or more cigarettes) 8,542 19.30%  Heavy smoking (20 or more cigarettes) 2,981 15.63%
Disease index 44,261 70.20% Disease index 19,067 80.20%
Open in a new tab
a

Total sample size is different because this variable was constructed using reported BMIs (Original N=39,736; Offspring N=18,822)

b

Weight(kg)/height(m)2

c

Underweight was defined as having less than 18.5 kg/m2

d

Normal weight was defined as having a BMI between 18.5 and 24.9 kg/m2

e

Overweight was defined as having a BMI between 25 and 29.9 kg/m2

f

Class I obesity was defined as having a BMI between 30 and 34.9 kg/m2

g

Class II/III obesity was defined as having a BMI greater than or equal to 35 kg/m2

Due to the small number of observations before age 31 years and in order to remain consistent with the trajectory and survival analysis for both cohorts, we restricted the analysis to observations that were 31 years of age or older. FHS collected information on deaths through newspapers, personal physician communications, or coroner reports. As of 2011, 3913 individuals in the original cohort and 967 individuals in the offspring cohort died. FHS supplied the days since exam 1 as the date of death, allowing us to compute the time spent at risk. For respondents who died, exposure to mortality risk was calculated as the duration from age 31 years until their date of death (in years). For the surviving respondents, we computed exposure to mortality risk as the duration from age 31 years until date of last contact or the last exam these individuals participated in. We reshaped the data to a person-year format left truncated at age 31 years or age at first survey (for those older than age 31 years) and right-censored at the age of death or last survey/contact. The total number of observations for the original and offspring cohort was 44,261 and 19,067, respectively.

Predictors of Mortality

BMI trajectoiy.

Due to small sample sizes at very old age, we constructed BMI trajectories between ages 31 and 80 years.

Sociodemographic and behavioral factors.

We use education as a proxy for an individual’s socioeconomic status, which is negatively associated with BMI and mortality. So, it is a potential confounder in BMI-mortality relationship. Educational attainment consists of four categories: less than high school, high school graduate, some college, and college graduate. Smoking tends to be negatively related to BMI while positively associated with mortality, so it also confounds the BMI-mortality link. Smoking status at each exam is based on the average number of cigarettes respondents smoked per day. It includes four categories: non-smoker, low smoking (1-9 cigarettes), moderate smoking (10-19 cigarettes), and heavy smoking (20 or more cigarettes). Smoking is treated as a time varying variable. We could not differentiate between never and former smokers because smoking information was collected in every exam and respondents might have smoked before the first exam or in-between while did not smoke at the year of exam. Birth cohorts are controlled because prior studies have reported substantial cohortbased pattern in obesity (37), obesity-related mortality (38), and mortality of all causes (39). We constructed six categorical birth cohorts for the original cohort (i.e., 1876-1894, 1895-1899, 1900-1904, 1905-1909, 1910-1914, and 1915 and above), and seven for the offspring cohort (i.e., 1903-1924, 1925-1929, 1930-1934, 1935-1939, 1940-1944, 1945-1949, 1950 and above.)

Health and medical history.

Respondents were examined by teams of doctors and nurses at each exam. They were also asked if they took any medication to treat a health condition and if they were diagnosed with or had any disease. We created a dummy variable to indicate whether the participant had any chronic disease, e.g., degenerative arthritis, gouty arthritis, rheumatoid arthritis, asthma or wheezing, prostate trouble, prostate disease, heart disease, hypertension, rheumatic heart disease, hypertensive cardiovascular disease, arrhythmia, aortic disease, mitral valve disease, vascular brain disease, pulmonary disease, gallbladder disease, urinary tract disease, renal disease, neurological disease, or thyroid disease. We also coded the dummy variable as 1 if the participant took medication to treat cardiovascular/heart diseases, arthritis, thyroid disease, or diabetes. We did not create a summary index of diseases because the number of diseases examined changed across exams. Diseases can be the causal pathway linking obesity to mortality but may also confound the BMI-mortality relationship because they are correlated with BMI status and the risk of dying.

Analytical Models

We used a semiparametric group-based trajectory model to capture the latent BMI trajectories between ages 31 and 80 years. This model uses a multinomial mixture modeling strategy and identifies relatively homogeneous clusters of trajectories of change over time in the presence of repeated observations on analytic units (35, 36). Different from other growth curve models (e.g., latent growth curve model, hierarchical growth curve analysis), this model does not remove missing data, but instead includes individuals with missing data at any time point in the modeling procedure, which then mitigates attrition bias. Supporting Information 2 provides the technical details of this model. We used the R lcmm package to estimate the model. As the distribution of BMI was right skewed, we modeled the logarithm of BMI (log(BMI)). We conducted latent class trajectory analysis for the original and offspring cohorts separately to reveal the differences between the cohorts and assess the time/cohort trends. After obtaining the latent trajectories for these two cohorts, we fit a Cox hazard model for each cohort adjusted for sociodemographic, smoking behavior, diseases and medical history to calculate the relative mortality risk of each trajectory, using attained age as the time metric. Considering time-varying disease characteristics as both potential confounders and mediators, we proceeded by fitting models both with and without adjustment for these factors. After obtaining the hazard ratios of death associated with BMI trajectories, we calculated the population attributable mortality risk fraction (PAF) (Supporting Information 3 provides the technical details). Overall patterns were similar by gender despite small difference in the number of individuals in each trajectory and the size of coefficient estimate of the effect of each trajectory on mortality. For this reason and for the sake of space, we report the findings for whole sample without breaking down gender.

RESULTS

Table 1 shows the descriptive statistics. Compared to the original cohort, the offspring cohort has a smaller proportion of observations that are smokers, but a larger proportion of observations that have any disease. We define four BMI groups using World Health Organization classifications: normal (BMI of 18.5-24.9 kg/m2), overweight (BMI of 25-29.9 kg/m2), class I obese (BMI of 30-34.9 kg/m2), and class II/III obese (BMI greater than or equal to 35 kg/m2). The proportion of observations that are class I obesity and class II/III obesity increases from 12.45% to 16.10%, 3.18% to 6.25%, respectively, from the original to offspring cohort.

For the original cohort, seven quadratic latent trajectories best fit the data as shown in Figure 1 (Supporting Information 4–7 describe the model selection). Since we model the trajectories based on log(BMI), we back-transform the trajectories to the original scale for presentation purpose. The topmost trajectory (open triangles, 1.88% of the sample) starts with class I obesity at age 31 years (BMI = 34.31 kg/m2) and increases to a BMI of approximately 40.37 kg/m2 at age 60. From this point onward, the trajectory decreases slightly. We call this the “class II/III obese” trajectory. The trajectory marked by solid circles (8.35% of the sample) starts with an overweight status at age 31 years (BMI = 29.21 kg/m2) and increases to the class I obese status. We call this trajectory “class I obese.” The next trajectory marked by open circles (20.89% of the sample) starts with a BMI of 26.27 kg/m2 and slowly increases, but it remains within the overweight category by age 80 years. We refer to this category as the “overweight stable.” The trajectory marked by closed triangles (21.48% of the sample) starts with a BMI of 25.90 kg/m2 at age 31 years but gradually decreases by age 80. We refer to this trajectory as “overweight downward.” Although we label it as an “overweight downward” trajectory, it can be understood as a general downward trajectory as “overweight” is the expected mean value for initial BMI for this trajectory and may include a range of BMIs. The trajectory marked by plus signs (13.31% individuals) starts with a normal weight status (BMI=22.41 kg/m2) and increases to an overweight status around age 55 years. We call this trajectory “normal weight upward.” The trajectory marked by a dashed line (23.75% individuals) starts with a BMI of 21.60 kg/m2 and remains in normal weight up to age 80 years. We call this trajectory “normal weight stable.” The bottommost trajectory (solid line, 10.34% of the sample) starts with a BMI of 19.47 kg/m2. We refer to it as “lower level of normal weight.”

Figure 1.

Figure 1.

Open in a new tab

Seven latent BMI trajectories from 31 to 80 years of age in the Framingham Heart Study original cohort 1948-2010.

For the offspring cohort, six quadratic latent trajectories best fit the data (Figure 2). Overall pattern is similar to the original cohort with five differences. First, no downward trajectories are identified, which may be because the number of individuals with weight loss is too small to be captured in the model. Second, most trajectories in the offspring cohort slowly increase from age 31 to 80 years, while the upward trajectories in the original cohort start declining around age 60 years. Third, “normal weight upward” trajectory advances to overweight status at relatively younger age (age 45 years) in the offspring cohort compared to the corresponding trajectory in the original cohort. Fourth, different from “overweight stable” in the original cohort, we identify an “overweight obesity” trajectory in the offspring cohort. This trajectory starts with overweight at age 31 years and advances to obesity around age 60. Fifth, the proportion of the sample in higher BMI trajectories systematically increase from the original to the offspring cohort.

Figure 2.

Figure 2.

Open in a new tab

Six latent BMI trajectories from 31 to 80 years of age in the Framingham Heart Study offspring cohort 1979-2001.

Next, we estimate the mortality hazards associated with these trajectories from a Cox model with “normal weight upward” as the reference group. Table 2 presents the results for the original cohort (Supporting Information 8 shows the complete table). After adjusting for birth cohort and gender, the highest mortality risk is for class II/III obese trajectory (Hazard Ratio [HR] = 2.15, 95% Confidence Interval [CI]=1.72 to 2.69), followed by class I obese (HR=1.56, 95% CI=1.35 to 1.80). The lower level of normal weight trajectory is associated with a 48% (95% CI=1.30 to 1.68) increase in the mortality risk. Overweight downward, overweight stable, and normal weight stable are associated with 37% (95% CI=1.23 to 1.53), 32% (95% CI=1.18 to 1.47), and 19% (95% CI=1.07 to 1.32) increase in mortality risk, respectively. After further adjusting for educational attainment and smoking behavior, class II/III upward trajectory continues to be associated with the highest mortality risk (HR=2.18, 95% CI=1.69 to 2.82), followed by class I obese upward (HR=1.58, 95% CI=1.36 to 1.82), overweight downward (HR=1.37, 95% CI=1.23 to 1.53), lower level of normal weight (HR=1.37, 95% CI=1.20 to 1.56), overweight stable (HR=1.34, 95% CI=1.20 to 1.49), and normal weight stable (HR=1.17, 95% CI=1.06 to 1.30). In the final model that adjusts for disease index, the associations between obesity trajectories and mortality slightly weaken but remain in the same direction, which is because comorbidity profiles are more presented in these trajectories (Supporting Information 9).

Table 2.

Adjusted Hazard Ratios of BMI Trajectories (Ages 31-80) from Cox Hazard Models in Framingham Heart Study Original Cohort 1948-2010.

Number of Persons Number of Deaths Birth Cohorts and Gender Adjusted Education Adjusteda Behavioral Factors Adjustedb Fully Adjustedc
HR (95 % CI) HR (95 % CI) HR (95 % CI) HR (95 % CI)
Body mass index trajectories
 Lower level of normald weight 473 404 1.48 (1.30, 1.68) 1.49 (1.31, 1.70) 1.37 (1.20, 1.56) 1.40 (1.23, 1.60)
 Normal weightd stable 1,087 928 1.19 (1.07, 1.32) 1.21 (1.09, 1.34) 1.17 (1.06, 1.30) 1.19 (1.07, 1.32)
 Normal weightd upward 609 499 1.00 1.00 1.00 1.00
 Overweighte downward 983 849 1.37 (1.23, 1.53) 1.38 (1.24, 1.54) 1.37 (1.23, 1.53) 1.38 (1.24, 1.53)
 Overweighte stable 956 833 1.32 (1.18, 1.47) 1.32 (1.18, 1.46) 1.34 (1.20, 1.49) 1.34 (1.20, 1.49)
 Class I obesef 382 323 1.56 (1.35, 1.80) 1.53 (1.33, 1.77) 1.58 (1.36, 1.82) 1.55 (1.34, 1.79)
 Class II/III obeseg 86 77 2.15 (1.72, 2.69) 2.11 (1.68, 2.64) 2.18 (1.69, 2.82) 2.11 (1.63, 2.73)
AIC 56,589.88 56,581.22 56,446.30 56,397.36
BIC 56,665.15 56,675.30 56,559.20 56,516.53
Observations 44,261 44,261 44,261 44,261
Open in a new tab
a

Adjusted for birth cohorts, gender, and education

b

Adjusted for birth cohorts, gender, education, and smoking behavior

c

Adjusted for birth cohorts, gender, education, smoking behavior, and disease index

d

Normal weight was defined as having a BMI between 18.5 and 24.9 kg/m2

e

Overweight was defined as having a BMI between 25 and 29.9 kg/m2

f

Class I obesity was defined as having a BMI between 30 and 34.9 kg/m2

g

Class II/III obesity was defined as having a BMI greater than or equal to 35 kg/m2

Table 3 presents the adjusted hazard ratios of the offspring cohort’s BMI trajectories with the “normal weight upward” trajectory as the reference group (Supporting Information 10 shows the complete table). After adjusting for birth cohort, gender, education and smoking, class II/III obesity trajectory is associated with an 80% (95% CI=1.21 to 2.64) increase in mortality risk, followed by class I obesity trajectory (HR=1.43, 95% CI=1.15 to 1.77), lower level of normal weight (HR=1.24, 95% CI=0.97, 1.59), normal weight stable (HR=1.04, 95% CI=0.88, 1.24), and overweight obesity (HR=1.01, 95% CI=0.84, 1.21). Compared to the original cohort, mortality risks associated with obesity trajectories have declined. In the fully adjusted model with disease index included, overall patterns remain the same.

Table 3.

Adjusted Hazard Ratios of BMI Trajectories (Ages 31-80) from Cox Hazard Models in Framingham Heart Study Offspring Cohort 1979-2001.

Number of Persons Number of Deaths Birth Cohorts and Gender Adjusted Education Adjusteda Behavioral Factors Adjustedb Fully Adjustedc
HR (95 % CI) HR (95 % CI) HR (95 % CI) HR (95 % CI)
Body mass index trajectories
 Lower level of normal weightd 325 98 1.36 (1.07, 1.73) 1.43 (1.13, 1.83) 1.24 (0.97, 1.59) 1.26 (0.99, 1.61)
 Normal weightd stable 927 251 1.05 (0.89, 1.25) 1.09 (0.92, 1.30) 1.04 (0.88, 1.24) 1.05 (0.89, 1.25)
 Normal weightd upward 1,113 293 1.00 1.00 1.00 1.00
 Overweighte obesity 867 194 1.01 (0.84, 1.21) 1.02 (0.85, 1.22) 1.01 (0.84, 1.21) 1.00 (0.84, 1.20)
 Class I obesef 402 105 1.47 (1.19, 1.82) 1.42 (1.14, 1.76) 1.43 (1.15, 1.77) 1.41 (1.14, 1.75)
 Class II/III obeseg 119 26 1.84 (1.24, 2.74) 1.70 (1.14, 2.55) 1.79 (1.21, 2.64) 1.71 (1.16, 2.54)
AIC 13,858.68 13,828.27 13,713.75 13,703.77
BIC 13,917.17 13,901.38 13,801.49 13,796.38
Observations 19,067 19,067 19,067 19,067
Open in a new tab
a

Adjusted for birth cohorts, gender, and education

b

Adjusted for birth cohorts, gender, education, and smoking behavior

c

Adjusted for birth cohorts, gender, education, smoking behavior, and disease index

d

Normal weight was defined as having a BMI between 18.5 and 24.9 kg/m2

e

Overweight was defined as having a BMI between 25 and 29.9 kg/m2

f

Class I obesity was defined as having a BMI between 30 and 34.9 kg/m2

g

Class II/III obesity was defined as having a BMI greater than or equal to 35 kg/m2

Finally, we calculate the population attributable mortality risk fraction using hazard ratios from the model adjusted for birth cohorts, gender, educational attainment, and smoking behavior. The mortality risk attributable to obese trajectories for the original cohort is 5.4%, but it rises to 6.4% for the offspring cohort. This increase in mortality attributable to obesity is a result of more individuals being in the high-risk trajectories, and this more than offsets the declining risk for specific trajectories.

DISCUSSION

Using relatively complete cohort data from the FHS original and offspring cohorts, we uncover the heterogeneity in life-long BMI trajectories across generations and estimate their associated mortality risks. We identify seven major trajectories among the original cohort including a downward trajectory, which is not identified in the offspring cohort. The overall trajectory patterns are similar between these two cohorts with some notable exceptions. First, the trajectories in the offspring cohort tend to shift upward at earlier ages relative to the original cohort. Second, the proportion of the sample in higher BMI trajectories is higher in the offspring cohort compared to the original cohort. Third, individuals in the higher weight trajectories experienced weight loss on average in the original cohort while this decline is less visible among the offspring cohort. The decline may be a result of diseased-induced weight loss, which may be better prevented and treated among the offspring cohort (40). This diseases-induced weight loss does not cause misclassification of BMI trajectory as it may have done to static BMI status (i.e., normal weight group increasingly contains frail individuals over ages). In fact, these distinct trajectories are primarily determined by their baseline BMI status (41) and weight loss is captured within each trajectory. If weight loss has affected a good number of individuals since early adulthood, latent trajectory model can identify this group of individuals as a distinct trajectory (Figure 1).

Among original cohort, people who are normal weight at age 31 years and gradually move to overweight status in middle or later adulthood have the lowest mortality risk even compared to those who maintain normal weight status throughout adulthood, followed by overweight stable, lower level of normal weight, overweight downward, class I obese upward and class II/III upward trajectories. Overall mortality patterns of these BMI trajectories are similar for offspring cohort except that some trajectories lose statistical significance (e.g., normal weight stable, overweight upward trajectory), which is probably due to smaller number of deaths. As of 2011, 3913 out of 4576 individuals in the original cohort died while only 967 out of 3753 individuals in the offspring cohort died. The overweight and obesity trajectories are more predictive of mortality than alternative measures of BMI history, including initial BMI status, maximum BMI, and BMI duration (Supporting Information 11), which further substantiates the utility of examining BMI trajectories.

This study complements a prior study that finds in later adulthood, people who started with overweight at age 51 years and remained in that status until age 77 had the lowest mortality risk (3). Together, they implicate the complexity in the relationship between weight gain and mortality risk. Prior studies have recognized that the impact of weight gain depends on baseline BMI status (24) and magnitude of weight gain (4, 5, 9). Our study further suggests that it depends on the timing of weight gain and its interaction with these two factors. For people with normal weight in early adulthood, moderate weight gain into overweight in later adulthood is associated with lower mortality risks compared to those who remain in the range of normal weight over the course of adulthood. But, for people with overweight or obesity status in early adulthood, weight gain is associated with excessive mortality risk.

Combination of these two important findings leads to a key implication: the impact of weight gain among overweight people in later adulthood depends on their baseline weight status in the early adulthood. If they were normal weight in early adulthood, a modest weight gain within the overweight range in later adulthood is beneficial for their survival (3); but if they were already overweight in early adulthood, additional weight gain only brings extra harms. Moderate weight gain among those who are normal weight in early adulthood may confer some survival benefits, which is consistent with the view that modest extra body weight in old ages, including lean tissue mass and fat mass, might provide protection against nutritional and energy deficiencies, metabolic stresses, the development of wasting and frailty, and loss of muscle and bone density caused by chronic diseases such as heart failure (42–44). But future research should continue this endeavor and fully explain this phenomenon.

Obesity upward trajectories have the highest mortality risks. Class I obesity upward and class II/III obesity upward trajectories in the original cohort are associated with 58% and 118% increases in mortality risk, respectively, without controlling for health factors. Among the offspring cohort, mortality risks associated with these two trajectories have declined to 43% and 80%, respectively. These declining mortality risks are not due to the change in reference group. For the same normal weight upward trajectory, the offspring cohort experiences a lower mortality risk compared to the original cohort (Supporting Information 12 shows the Cox models of the original and offspring cohorts combined). These findings are consistent with a prior study that has found a secular decline in the association between obesity and mortality in the United States (45).

This study has several limitations. First, FHS sample is rather homogeneous (i.e., Whites and within same families in one town). The advantage of the homogeneity of the sample is that it allows us to better mitigate any unmeasured confounding and better isolate the secular effects that may be influencing the obesity-mortality association over time in an epidemiological sample. The disadvantage concerns the representativeness of our sample and application of the findings to other racial groups. But FHS is the only U.S. data that has a long prospective followup on BMI and mortality. Therefore, replicating the analyses for other racial groups is not feasible at this point. Second, FHS original cohort and offspring cohort each comprise of a wide swath of birth cohorts with some overlap. But since the overlap in birth cohorts between original and offspring is small, the overall pattern across these two FHS cohorts can still portray a relatively clear cohort trend. Third, even though the latent trajectory model is a straightforward tool to uncover the underlying structure of developmental trajectories in the population, one caveat of this model is that the assignment of individuals to a distinct developmental pattern is based on their highest estimated group-membership probability to the identified pattern. Therefore, these latent patterns should not be considered as the actual developmental patterns but, rather, as approximations of more complex processes.

Fourth, we find that in the original cohort people in the weight loss trajectory face 37% increase in mortality risk compared to those in the normal weight upward trajectory, but we cannot differentiate intentional weight loss from diseases-induced weight loss. However, prior studies have found that intentional weight loss has not always been observed to have putative beneficial effects on mortality (46–48). Moreover, we have controlled for an age-varying disease index thereby reducing some of the confounding role of disease-induced weight loss. Fifth, we are unable to control for all confounding factors. To minimize potential confounding due to smoking, we constrained the sample to those who were non-smokers at all waves and found the deleterious effects of overweight and obesity trajectories became greater (Supporting Information 13 and 14). This finding is especially salient for the original cohort because a larger proportion of that cohort was smokers. We could not conduct similar sensitivity analyses for those without any disease at all waves as the resulting sample is very small (259 individuals in original cohort and 10 individuals in offspring cohort). Moreover, as diseases are potentially the pathway from obesity to death especially in old age, such analysis controls away some of the risk of dying. Sixth, we did not conduct a detailed cause-specific hazard analysis because of the limited number of deaths in the offspring cohort. Supporting Information 15 and 16 show preliminary cause-specific results for cardiovascular disease and cancer. Obesity trajectories were linked to cardiovascular disease deaths but not deaths from cancer.

Improving upon prior studies, this study reveals the heterogeneity and mortality risk of life-long BMI trajectories across two related generations. We find that people in the normal weight upward trajectory have the lowest mortality risk, followed by normal weight stable, overweight stable / upward, lower level of normal weight, overweight downward (original cohort only), class I obese upward and class II/III upward trajectories. We further reveal the dynamics in the obesity – mortality relationship across generations. Even though the mortality risks associated with obesity trajectories have declined across cohorts, their contributions to population deaths increased from 5.4% in the original cohort to 6.4% in the offspring cohort due to the increasing proportion of individuals in these trajectories. These contributions are smaller than those found in younger birth cohorts (3). These findings should have a broad implication on the current obesity epidemic in the United States, which should observe more obesity-related deaths at the population level in the years to come even though the mortality risk due to obesity has declined at the individual level.

Supplementary Material

mmc1

Highlights.

  • This study identifies the heterogeneity in the life-long BMI trajectories over the life course and across generations.

  • The mortality impact of weight gain depends on an individual’s life-course BMI trajectory.

  • People who are normal weight in early adulthood and gradually move to overweight status during later adulthood have the lowest mortality risk.

  • Mortality risks associated with obesity trajectories have declined across cohorts.

  • Population attributable deaths associated with obesity trajectories have grown over generations due to the increasing prevalence of these trajectories.

Acknowledgements:

We thank Anatoliy Yashin for helpful comments. This work was supported by the Grant R03AG053463 and P2CHD058484, funded by the National Institutes of Health. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health or the Department of Health and Human Services.

Abbreviations:

BMI

body mass index

FHS

Framingham Heart Study

PAF

population attributable mortality risk fraction

HR

Hazard Ratio

CI

Confidence Interval

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Conflict of Interests: None.

Contributor Information

Hui Zheng, Ohio State University.

Paola Echave, Ohio State University.

Neil Mehta, University of Michigan.

Mikko Myrskyla, Max Planck Institute for Demographic Research.

References

  • 1.Abdullah A, Wolfe R, Stoelwinder JU, et al. The number of years lived with obesity and the risk of all-cause and cause-specific mortality. Int. J. Epidemiol 2011;40(4):985–996. [DOI] [PubMed] [Google Scholar]
  • 2.Abdullah A, Stoelwinder J, Shortreed S, et al. The duration of obesity and the risk of type 2 diabetes. Public Health Nutr. 2011;14(1):119–126. [DOI] [PubMed] [Google Scholar]
  • 3.Zheng H, Tumin D, Qian Z. Obesity and Mortality Risk: New Findings From Body Mass Index Trajectories. Am. J. Epidemiol 2013; 178(11):1591–1599. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Wannamethee S, Shaper A, Walker M. Weight change, body weight and mortality: the impact of smoking and ill health. Int. J. Epidemiol 2001;30(4):777–786. [DOI] [PubMed] [Google Scholar]
  • 5.Andres R, Muller DC, Sorkin JD. Long-term effects of change in body weight on all-cause mortality. A review. Ann. Intern. Med 1993; 119(7 Pt 2):737–43. [DOI] [PubMed] [Google Scholar]
  • 6.Everhart JE, Pettitt DJ, Bennett PH, et al. Duration of obesity increases the incidence of NIDDM. Diabetes. 1992;41(2):235–240. [DOI] [PubMed] [Google Scholar]
  • 7.Mehta NK, Stenholm S, Elo IT, et al. Weight histories and mortality among finnish adults: the role of duration and peak body mass index. Epidemiology. 2014;25(5):707–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Mikkelsen KL, Heitmann BL, Keiding N, et al. Independent effects of stable and changing body weight on total mortality. Epidemiology. 1999;10(6):671–8. [PubMed] [Google Scholar]
  • 9.Myrskylä M, Chang VW. Weight change, initial BMI, and mortality among middle- and older-aged adults. Epidemiology. 2009;20(6):840–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Preston SH, Mehta NK, Stokes A. Modeling obesity histories in cohort analyses of health and mortality. Epidemiology. 2013;24(1):158–66. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Somes GW, Kritchevsky SB, Shorr RI, et al. Body mass index, weight change, and death in older adults: The systolic hypertension in the elderly program. Am. J. Epidemiol. 2002;156(2):132–138. [DOI] [PubMed] [Google Scholar]
  • 12.Zhang WB, Pincus Z. Predicting all-cause mortality from basic physiology in the Framingham Heart Study. Aging Cell. 2016;15(1):39–48. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Engeland A, Bjørge T, Tverdal A, et al. Obesity in adolescence and adulthood and the risk of adult mortality. Epidemiology. 2004;15(1):79–85. [DOI] [PubMed] [Google Scholar]
  • 14.Ferraro KF, Thorpe RJ, Wilkinson JA, et al. The life course of severe obesity: does childhood overweight matter? J. Gerontol. B. Psychol. Sci. Soc. Sci. 2003;58(2):S110–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Gray L, Lee I-M, Sesso HD, et al. Body weight in early and mid-adulthood in relation to subsequent coronary heart disease mortality: 80-year follow-up in the Harvard Alumni Study. Arch. Intern. Med. 2011;171(19):1768–70; discussion 1770. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Adams KF, Leitzmann MF, Ballard-Barbash R, et al. Body mass and weight change in adults in relation to mortality risk. Am. J. Epidemiol. 2014;179(2):135–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Lee H, Lee D, Guo G, et al. Trends in body mass index in adolescence and young adulthood in the United States: 1959–2002. J. Adolesc. Heal. 2011;49(6):601–608. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Ford ES, Williamson DF, Liu S. Weight change and diabetes incidence: Findings from a national cohort of US adults. Am. J. Epidemiol. 1997;146(3):214–222. [DOI] [PubMed] [Google Scholar]
  • 19.Wannamethee SG, Shaper AG, Walker M. Weight change, weight fluctuation, and mortality. Arch. Intern. Med 2002;162(22):2575. [DOI] [PubMed] [Google Scholar]
  • 20.Elliott AM, Aucott LS, Hannaford PC, et al. Weight change in adult life and health outcomes. Obes. Res 2005;13(10):1784–1792. [DOI] [PubMed] [Google Scholar]
  • 21.Jacobs-van der Bruggen MAM, Spijkerman A, van Baal PHM, et al. Weight change and incident diabetes: Addressing an unresolved issue. Am. J. Epidemiol 2010;172(3):263–270. [DOI] [PubMed] [Google Scholar]
  • 22.Biggs ML, Mukamal KJ, Luchsinger JA, et al. Association between adiposity in midlife and older age and risk of diabetes in older adults. JAMA. 2010;303(24):2504–12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Brancati FL, Wang NY, Mead LA, et al. Body weight patterns from 20 to 49 years of age and subsequent risk for diabetes mellitus: The Johns Hopkins precursors study. Arch. Intern. Med 1999;159(9):957–963. [DOI] [PubMed] [Google Scholar]
  • 24.Corrada MM, Kawas CH, Mozaffar F, et al. Association of body mass index and weight change with all-cause mortality in the elderly. Am. J. Epidemiol 2006;163(10):938–49. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Land KC, McCall PL, Nagin DS. A comparison of Poisson, negative binomial, and semiparametric mixed Poisson regression models. Sociol. Methods Res. 1996;24(4):387–442. [Google Scholar]
  • 26.Nagin DS. Analyzing Developmental Trajectories: A Semiparametric, Group-Based Approach. Psychol. Methods 1999;4:139–157. [DOI] [PubMed] [Google Scholar]
  • 27.Nagin DS, Land KC. Age, criminal, careers, and population heterogeneity: Specification and estimation of a nonparametric, mixed Poisson model. Criminology. 1993;31(3):327–362. [Google Scholar]
  • 28.Kuchibhatla MN, Fillenbaum GG, Kraus WE, et al. Trajectory classes of body mass index in a representative elderly community sample. J. Gerontol. A. Biol. Sci. Med. Sci. 2013;68(6):699–704. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Zajacova A, Ailshire J. Body mass trajectories and mortality among older adults: A joint growth mixture–discrete-time survival analysis. Gerontologist. 2014;54(2):221. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Ostbye T, Malhotra R, Landerman LR. Body mass trajectories through adulthood: results from the National Longitudinal Survey of Youth 1979 Cohort (1981-2006). Int. J. Epidemiol. 2011;40(1):240–250. [DOI] [PubMed] [Google Scholar]
  • 31.Nonnemaker JM, Morgan-Lopez AA, Pais JM, et al. Youth BMI trajectories: Evidence from the NLSY97. Obesity. 2009;17(6):1274–1280. [DOI] [PubMed] [Google Scholar]
  • 32.Kannel WB, Feinleib M, McNamara PM, et al. An investigation of coronary heart disease in families. Am. J. Epidemiol 1979;110(3):281–290. [DOI] [PubMed] [Google Scholar]
  • 33.Dawber TR, Meadors GF, Moore FE. Epidemiological approaches to heart disease: the Framingham Study. Am. J. Public Health 1951;41(3):279–286. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Framingham Heart Study. FHS original cohort and FHS offspring cohort. Location: Biologic Specimen and data Repository Information Coordinating Center. (https://biolincc.nhlbi.nih.gov/). (Accessed August 1, 2017)
  • 35.Jones BL, Nagin DS, Roeder K. A SAS procedure based on mixture models for estimating developmental trajectories. Sociol. Methods Res. 2001;29(3):374–393. [Google Scholar]
  • 36.Nagin DS. Group-Based Modeling of Development. 2005. Cambridge, MA: Harvard University Press. [Google Scholar]
  • 37.Reither EN, Hauser RM, Yang Y. Do birth cohorts matter? Age-period-cohort analyses of the obesity epidemic in the United States. Soc Sci Med. 2009;69(10):1439–1448. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Yu Y Reexamining the declining effect of age on mortality differentials associated with excess body mass: evidence of cohort distortions in the United States. Am J Public Health. 2012;102(5):915–922. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Zheng H, Yang YC, Land KC. Age-Specific Variation in Adult Mortality Rates in Developed Countries. Popul Res Policy Rev. 2016;35(1):49–71. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Uretsky S, Messerli FH, Bangalore S, et al. Obesity Paradox in Patients with Hypertension and Coronary Artery Disease. Am J Med. 2007;120(10):863–870. [DOI] [PubMed] [Google Scholar]
  • 41.Block JP, Subramanian SV., Christakis NA, O’Malley AJ. Population Trends and Variation in Body Mass Index from 1971 to 2008 in the Framingham Heart Study Offspring Cohort. PLoS One. 2013;8(5):1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Zheng H, Yang Y. Population heterogeneity in the impact of body weight on mortality. Soc Sci Med. 2012;75(6):990–996. [DOI] [PubMed] [Google Scholar]
  • 43.Oreopoulos A, Padwal R, Kalantar-Zadeh K, Fonarow GC, Norris CM, McAlister FA. Body mass index and mortality in heart failure: A meta-analysis. Am Heart J. 2008;156(1):13–22. [DOI] [PubMed] [Google Scholar]
  • 44.Janssen I. Morbidity and mortality risk associated with an overweight BMI in older men and women. Obesity. 2007;15(7):1827–1840. [DOI] [PubMed] [Google Scholar]
  • 45.Mehta NK, Chang VW. Secular declines in the association between obesity and mortality in the United States. Popul Dev Rev. 2011;37(3):435–451. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Yaari S, Goldbourt U. Voluntary and involuntary weight loss: associations with long term mortality in 9,228 middle-aged and elderly men. Am. J. Epidemiol 1998;148(6):546–555. [DOI] [PubMed] [Google Scholar]
  • 47.Berentzen T, SØrensen TIA. Effects of intended weight loss on morbidity and mortality: Possible explanations of controversial results. Nutr. Rev. 2006;64(11):502–507. [DOI] [PubMed] [Google Scholar]
  • 48.Simonsen MK, Hundrup YA, Obel EB, et al. Intentional weight loss and mortality among initially healthy men and women. Nutr. Rev 2008;66(7):375–386. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

mmc1
NIHMS1665752-supplement-mmc1.docx (784KB, docx)

RESOURCES