Associations of healthy lifestyle with life expectancy and cancer free life expectancy
Principal Investigator
Name
Eleanor Watts
Degrees
D.Phil., M.P.H.
Institution
National Cancer Institute
Position Title
Postdoctoral Fellow
Email
About this CDAS Project
Study
PLCO
(Learn more about this study)
Project ID
PLCO-1100
Initial CDAS Request Approval
Nov 21, 2022
Title
Associations of healthy lifestyle with life expectancy and cancer free life expectancy
Summary
Background
Maintaining a healthy weight, avoiding smoking, being physically active, consuming no-to-low alcohol and having a healthy diet is associated with lower risks of premature death and cancer diagnosis. However, relatively few studies have quantified the associations of these modifiable risk factors with life expectancy.
There is also high variation in life expectancies within populations, for instance US national records indicate that there was an 18-year difference between the 1st and 4th quartile by age of death in 2016. Previous prospective studies that have quantified the associations of healthy lifestyle and life expectancy have been based on relatively homogenous populations and have not investigated possible sources of heterogeneity by certain population characteristics, particularly by racial/ethnic group. Quantifying these associations may prove useful for public understanding, communicating policy, and informing targeted public health interventions.
Methods
We will pool data from NIH-AARP baseline questionnaire (N=500,000), Agricultural Health Study (AHS, complete phases I and II data; N=30,000), and the PLCO (complete baseline and dietary baseline questionnaire; N=60,000). These three datasets were selected as they collected data for the 5 components of healthy lifestyle and used established exposure measurements methods with similar questionnaires. Where data vary, we will use cohort-specific percentiles and assign midpoints of ranges.
We will develop a healthy lifestyle index comprising of a weighted score of BMI, smoking status, alcohol consumption, diet, and physical activity using the PLCO cohort. Weights will be based on maximum logit scores derived from shape constrained additive regression models to calculate indexes for each participant. This index will be categorized into quartiles (very unhealthy, unhealthy, healthy, and very healthy). The score will be cross validated using AHS and NIH-AARP datasets.
Using the pooled dataset, life expectancy for participants in each health category will be predicted using adjusted survival curves based on Cox proportional hazard models with age as the underlying variable. We will stratify by study and adjust for educational attainment and self-reported racial/ethnic group. Due to sex differences in life expectancies, all associations will be examined separately for men and women. Years of life lost will be defined as the difference between the age of 50% survival for each healthy lifestyle category in comparison with the “very healthy” category.
To investigate the distribution of life expectancies within the study population we will predict two individualized life expectancies for each participant using separate Cox models: i) estimating the area under the predicted survival curve based on baseline measurements; ii) area under the survival curve using counterfactual “very healthy lifestyle”. Years of life lost for each participant will be defined as the difference between these two predictions.
In further analyses, we will examine associations by subsets of participants (racial/ethnic group, longer follow-up, socioeconomic status, and prevalent comorbidities). Socioeconomic status will be defined using area level score of deprivation from linked census tract data (AARP and ~50% of the PLCO cohort).
Maintaining a healthy weight, avoiding smoking, being physically active, consuming no-to-low alcohol and having a healthy diet is associated with lower risks of premature death and cancer diagnosis. However, relatively few studies have quantified the associations of these modifiable risk factors with life expectancy.
There is also high variation in life expectancies within populations, for instance US national records indicate that there was an 18-year difference between the 1st and 4th quartile by age of death in 2016. Previous prospective studies that have quantified the associations of healthy lifestyle and life expectancy have been based on relatively homogenous populations and have not investigated possible sources of heterogeneity by certain population characteristics, particularly by racial/ethnic group. Quantifying these associations may prove useful for public understanding, communicating policy, and informing targeted public health interventions.
Methods
We will pool data from NIH-AARP baseline questionnaire (N=500,000), Agricultural Health Study (AHS, complete phases I and II data; N=30,000), and the PLCO (complete baseline and dietary baseline questionnaire; N=60,000). These three datasets were selected as they collected data for the 5 components of healthy lifestyle and used established exposure measurements methods with similar questionnaires. Where data vary, we will use cohort-specific percentiles and assign midpoints of ranges.
We will develop a healthy lifestyle index comprising of a weighted score of BMI, smoking status, alcohol consumption, diet, and physical activity using the PLCO cohort. Weights will be based on maximum logit scores derived from shape constrained additive regression models to calculate indexes for each participant. This index will be categorized into quartiles (very unhealthy, unhealthy, healthy, and very healthy). The score will be cross validated using AHS and NIH-AARP datasets.
Using the pooled dataset, life expectancy for participants in each health category will be predicted using adjusted survival curves based on Cox proportional hazard models with age as the underlying variable. We will stratify by study and adjust for educational attainment and self-reported racial/ethnic group. Due to sex differences in life expectancies, all associations will be examined separately for men and women. Years of life lost will be defined as the difference between the age of 50% survival for each healthy lifestyle category in comparison with the “very healthy” category.
To investigate the distribution of life expectancies within the study population we will predict two individualized life expectancies for each participant using separate Cox models: i) estimating the area under the predicted survival curve based on baseline measurements; ii) area under the survival curve using counterfactual “very healthy lifestyle”. Years of life lost for each participant will be defined as the difference between these two predictions.
In further analyses, we will examine associations by subsets of participants (racial/ethnic group, longer follow-up, socioeconomic status, and prevalent comorbidities). Socioeconomic status will be defined using area level score of deprivation from linked census tract data (AARP and ~50% of the PLCO cohort).
Aims
1. Estimate the associations of a healthy lifestyle score and individual lifestyle factors with life expectancy and life expectancy free from cancer using pooled analysis of 600,000 participants.
2. Investigate heterogeneity in the associations by population characteristics.
3. Predict individualized life expectancies to estimate the distributions of loss in life years.
Collaborators
Steven Moore, National Cancer Institute, MD, USA
Li Cheung, National Cancer Institute, MD, USA
Neal Freedman, National Cancer Institute, MD, USA
Maki Inoue-Choi, National Cancer Institute, MD, USA
Erikka Loftfield, National Cancer Institute, MD, USA
Linda Liao, National Cancer Institute, MD, USA
Charles Matthews, National Cancer Institute, MD, USA
Pedro Saint Maurice, National Cancer Institute, MD, USA
Joshua Freeman, National Cancer Institute, MD, USA
Kathleen McClain, National Cancer Institute, MD, USA
Hormuzd Katki, National Cancer Institute, MD, USA