Estimated preventable fraction of chronic disease attributed to long-term physical activity and diet quality, independent of body weight: a prospective cohort study of three US cohorts

Summary

Background

Body mass index (BMI) is often treated as a proxy for lifestyle, but this may discourage people with higher BMI from adopting risk-reducing behaviors and lead those with lower BMI—including individuals achieving weight reduction through pharmacotherapy—to overlook the value of lifestyle. We aimed to estimate incidence of chronic disease preventable through physical activity and diet quality, independent of BMI.

Methods

We followed 142,041 adults (80% [113,727] female, mean baseline age 50 years, 96% [136,998] non-Hispanic White) from the Nurses' Health Study, Nurses’ Health Study II, and Health Professionals Follow-up Study. Physical activity was assessed as metabolic equivalent of task (MET) hours per week, and diet quality as modified Alternative Healthy Eating Index. Chronic disease included type 2 diabetes, major atherosclerotic cardiovascular disease, and cancer. Regression models adjusted for BMI at multiple life stages allowed estimation of effects for sustained adulthood habits and midlife changes, independent of prior BMI and lifestyle.

Findings

Participants in the highest deciles of physical activity and diet quality had 46% (95% confidence interval [CI], 40–52%) lower risk of chronic disease for sustained habits and 34% (95% CI, 26–41%) for midlife changes. Results were similar among women (44% lower risk [95% CI, 35–51%] sustained habits, 32% [95% CI, 20–41%] midlife changes) and men (54% lower risk [95% CI, 41–64%] sustained habits, 48% [95% CI, 33–60%] midlife changes). The proportion of cases potentially preventable was 34% (95% CI, 27–41%), or 1.1 million cases annually (95% CI, 906,000–1,376,000), for sustained habits and 26% (95% CI, 18–34%), or 873,000 cases annually (95% CI, 604,000–1,141,000), for midlife changes.

Interpretation

Higher physical activity and diet quality were associated with substantially reduced chronic disease risk regardless of BMI. These findings support public health strategies that promote healthy behaviors independent of body weight.

Funding

US National Institutes of Health, Prevent Cancer Foundation.

Research in context.

Evidence before this study

We searched PubMed for studies published through October 2025 (month of search). We searched for studies estimating the preventable fraction of diabetes, cardiovascular disease, and cancer using the search terms (“Body Mass Index"[Mesh] OR BMI) AND (“Physical Activity”[Mesh] OR “Exercise”[Mesh]) AND (“Diet”[Mesh] OR “Diet, Healthy”[Mesh] OR diet quality OR “Healthy Eating Index”) AND (“Chronic Disease”[Mesh] OR “Cardiovascular Disease”[Mesh] OR “Coronary Artery Disease” OR diabetes OR cancer) AND (“Population Attributable Fraction” OR “Population Attributable Risk” OR “Attributable Risk” OR “Population Preventable Fraction” OR “Population Impact Number” OR “Preventable Cases” OR “PAR” OR “PAF” OR “etiologic fraction” OR “excess fraction”), with no language or date restrictions. We found studies showing that a substantial fraction of type 2 diabetes and cardiovascular disease could be prevented through healthy body weight, diet, and physical activity. However, no study quantified the preventable fraction of chronic disease attributable to diet quality and physical activity in adulthood, independent of BMI or weight change. This evidence gap may reinforce a focus on weight rather than the weight-independent benefits of lifestyle.

Added value of this study

To our knowledge, this is the first study to account for BMI across the lifecourse and estimate the fraction of chronic disease preventable by optimal diet quality and physical activity independent of weight. We found that up to 34% (1.1 million cases per year) of chronic disease could be prevented by sustained healthy habits, and 26% (873,000 cases per year) by adopting them in midlife.

Implications of all the available evidence

The combined evidence indicates that improving physical activity and diet quality could prevent a large share of major chronic disease even without weight change and among adults with normal BMI. Public health messages and clinical guidance should emphasize these behaviors for all adults. Policy priorities include expanding access to healthy foods, safe spaces for activity, and interventions that support midlife lifestyle improvement. Future research should evaluate cost-effective implementation and expand on weight-neutral metabolic risk markers.

Introduction

Physical activity and diet quality are well-established determinants of chronic disease risk,^1,2 yet the extent to which they provide benefits independent of body weight remains uncertain. Most prospective studies treat baseline and follow-up BMI as a single construct, making it difficult to disentangle weight-dependent from weight-independent pathways. This distinction is increasingly important. In the United States (US), 72% of adults have overweight or obesity,³ and many individuals maintain a high BMI despite making meaningful improvements in lifestyle,^4,5 raising questions about the degree to which behavior change can reduce risk independent of weight. The rapid growth of pharmacologic and surgical weight-loss therapies further underscores the importance of quantifying disease-prevention benefits of physical activity and diet quality that are not mediated through body weight.⁶

Many prospective studies have shown associations between physical activity or diet quality and chronic disease; however, a major limitation is that few have distinguished baseline BMI as a potential confounder from weight change as a mediator, despite this being the natural treatment of body weight in a randomized trial of lifestyle intervention.7, 8, 9, 10 To our knowledge, no studies have extended this framework to estimate population preventable fractions that account for pre-exposure BMI and post-exposure weight change, a critical consideration for public health.

In the present study, we used prospective observational data from the Nurses' Health Study (NHS), Nurses’ Health Study II (NHSII), and Health Professionals Follow-up Study (HPFS) to estimate associations of physical activity and diet quality with incidence of type 2 diabetes, major atherosclerotic cardiovascular disease (ASCVD), and cancer. We applied two analytic frameworks to evaluate conceptually distinct population effects. The first estimated the cumulative effects of habits established in early adulthood, adjusting for BMI at age 18–21. The second estimated the effects of habits adopted in midlife (median age 50 years at baseline), adjusting for prebaseline BMI, physical activity, and diet quality. Finally, using nationally representative data, we estimated population preventable fractions under minimal, moderate, and optimal levels of physical activity and diet quality. We hypothesized that higher long-term physical activity and diet quality would be associated with lower chronic disease risk and a substantial preventable fraction, independent of BMI.

Methods

Study population

The NHS enrolled 121,700 registered female nurses aged 30–55 years in 1976, NHSII enrolled 116,429 nurses aged 25–42 years in 1989, and HPFS enrolled 51,529 male health professionals aged 40–75 years in 1986. Comparable validated questionnaire-based measures were used across cohorts. Participant sex was defined by cohort enrollment and no data on gender identity were collected; accordingly, we report sex rather than gender and all references to sex-specific composition and analyses refer to participant sex. Race and ethnicity were self-reported; available categories included American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White, and more than one race, with ethnicity classified as Hispanic or Latino versus not Hispanic or Latino. Participants lived in various states across the US and provided updated lifestyle habits and new diagnoses every two years and updated dietary data every four years, all via mailed questionnaires. Comprehensive data on physical activity and diet were first collected in 1986 (NHS), 1993 (NHSII), and 1986 (HPFS); therefore, we used 1990 (NHS), 1997 (NHSII), and 1990 (HPFS) as the baseline year to allow adjustment for prebaseline confounders in select models. Eligible participants were followed for incident chronic disease diagnoses starting two years after collection of baseline exposure data to reduce reverse causation.¹¹

We excluded participants who died or were diagnosed with chronic disease before start of follow-up (Figure S1). At prebaseline, 18% (46,627) of participants were missing physical activity data, 18% (48,146) missing dietary data, and 12% (32,467) missing BMI data; we excluded individuals missing these primary exposure and mediator data at baseline rather than imputing. During follow-up, missingness in physical activity, diet, and body weight measures was addressed by carrying forward the prior cycle's value for a single questionnaire cycle; participants missing more than one consecutive cycle were excluded from that interval. To limit the influence of outliers and implausible data, values for physical activity, diet score, and BMI were truncated at the 1st and 99th percentile in each questionnaire cycle. Missing covariate data for confounding variables were carried forward from prior cycles, and missing data after carrying forward (<5% for all variables) were imputed using median imputation. The primary analysis used all available data across questionnaire cycles, and participants contributed person-time in cycles for which they returned a questionnaire, regardless of whether they had missed prior cycles. Participants were followed until the first occurrence of chronic disease, death, loss to follow-up, or end of follow-up (June 2016 for NHS and HPFS; June 2017 for NHSII), whichever came first.

To contextualize the extent of missing data during follow-up, per-cycle loss to follow-up (failure to return a questionnaire, after accounting for death) averaged approximately 4%. Among returned questionnaires, missingness averaged 6% for physical activity, 10% for AHEI score, and 4% for BMI; missingness was similar or lower for other covariates. To evaluate the potential impact of missing data, we conducted a sensitivity analysis using multiple imputation, described in greater detail in the Supplementary Methods.

Physical activity and diet quality assessment

Information on physical activity and diet was collected using self-reported questionnaires administered repeatedly throughout follow-up. The use of consistent questionnaire structures and fixed scoring systems allowed measures to remain comparable across time, and suitable for estimating cumulative average values.

Average weekly time spent engaging in recreational physical activities was assessed every two years using questionnaires with demonstrated reproducibility and validity.^12,13 Weight training items were added in 2000 for NHS, 2001 for NHSII, and 1990 for HPFS. Total physical activity was estimated as metabolic equivalent task (MET) hours per week, a standardized metric that facilitates comparability across timepoints and among other studies relying on self-reported activity.¹⁴

Dietary intake was assessed every four years using a 130-item food frequency questionnaire (FFQ) with established reproducibility and validity.15, 16, 17 Diet quality was estimated using a modified Alternative Healthy Eating Index (AHEI)-2010 score that excluded alcohol and trans fatty acids to maximize population generalizability and reflect the removal of trans fats from the US food supply.18, 19, 20 The AHEI assigns a score from 0 to 10 for nine dietary components (vegetables, fruit, whole grains, nuts and legumes, omega-3 fatty acids, polyunsaturated fatty acids, sugar sweetened beverages and juice, red and processed meat, and sodium), yielding a total score from 0 to 90.

Confounder assessment

Participant demographic and lifestyle factors were self-reported biennially; alcohol use was reported every four years on FFQs; and family history was assessed periodically. Race/ethnicity was modeled as White vs other racial/ethnic identity due to sparse data in individual categories. Income was approximated from census tract median income and self-reported ZIP code. BMI was derived from self-reported weight and height. At cohort enrollment, participants reported their weight at age 18 (NHS, NHSII) or 21 (HPFS). Missing covariate data were carried forward from prior cycles, and missing data after carrying forward (<5% for all variables) were imputed using median imputation. Given the low level of covariate missingness, this approach avoids additional modeling assumptions and provides results comparable to more complex approaches such as multiple imputation.

Outcome assessment

The primary outcome was total chronic disease, defined as the first occurrence of type 2 diabetes, ASCVD, or cancer; with permission, study physicians blinded to risk factors reviewed medical records to confirm diagnoses and extract relevant details. Type 2 diabetes was confirmed according to the American Diabetes Association criteria.^21,22 ASCVD was defined as the first occurrence of fatal or nonfatal myocardial infarction (coronary heart disease) or stroke (cerebrovascular disease).^23,24 Cancer was defined as the first occurrence of any cancer type; due to the high prevalence of non-melanoma skin cancer and prostate cancer, the majority of which do not cause significant morbidity, we excluded non-melanoma skin cancer and non-lethal prostate cancer. “Obesity-related cancer” was defined as the first occurrence of esophageal, gastric, colorectal, liver, gallbladder, pancreatic, breast (postmenopausal), endometrial, ovarian, kidney, or thyroid cancer or multiple myeloma.

Type 2 diabetes, ASCVD, and cancer were combined into a composite outcome of total chronic disease incidence because they represent leading causes of chronic disease morbidity in the US, share common lifestyle- and adiposity-related risk factors, and collectively capture the public health burden attributable to physical inactivity and poor diet quality. Analogous to all-cause mortality—a widely accepted composite endpoint—this approach prioritizes total preventable burden over uniformity of effect size across components. Unlike composite endpoints in clinical trials, where concerns relate to heterogeneous treatment effects and inflation of statistical significance, our objective was to estimate preventable burden at the population level. Composite endpoints are considered valid when components are clinically relevant, of similar importance, and reflect related disease processes within the same population—conditions met in the present study.²⁵ Composite chronic disease incidence and disease-free survival have been used in prior epidemiological studies to quantify preventable morbidity and mortality, and our approach is consistent with this framework.26, 27, 28 Because the magnitude of association between lifestyle behaviors and each component was expected to differ, disease specific analyses—including absolute estimates of preventable cases by condition—were conducted and presented in parallel to allow full evaluation of heterogeneity across outcomes.

Ethics

This study was approved by the institutional review boards at the Harvard T. H. Chan School of Public Health and Brigham and Women's Hospital (NHS and NHSII 1999P011117, approval date 07/16/2024; HPFS IRB24-0407, approval date 4/9/2026) and participating registries as required. Written informed consent was obtained from all participants.

Statistical analysis: risk ratios

We considered two approaches to exposure assessment and modeling. “Sustained adulthood habits” were defined as cumulative average habits (physical activity, diet quality, or joint physical activity and diet quality) starting at baseline and adjusted for relevant confounders and BMI at age 18 or 21. Although the average age at start of follow-up was 50 years, physical activity and diet quality reported during the study period were likely correlated with habitual patterns established in earlier adulthood. In this framework, BMI at study enrollment largely reflects the long-term consequences of adulthood lifestyle habits; therefore, adjusting for prebaseline BMI would remove part of the habitual lifestyle-disease relationship we aimed to characterize with this model. We adjusted only for BMI at age 18–21, which temporally precedes the adulthood habits of interest but captures some of the confounding effects of childhood environment.

“Midlife changes” were defined as cumulative average habits starting at baseline but with adjustment for the same confounders, plus prebaseline BMI, physical activity, and diet (prebaseline variables were assessed 2–4 years before baseline). This framework is intended to approximate the effect of adopting more optimal physical activity or diet quality in mid-adulthood, independent of earlier behaviors and weight status.²⁹ Here, prebaseline BMI and prior habits temporally precede the behavior changes and may be associated with both the likelihood of changing habits and subsequent chronic disease risk; therefore, they were treated as confounders. This approach helps distinguish associations that reflect long-term habits and their influence on body weight from associations related specifically to midlife behavior change. The latter is less often separated analytically in epidemiologic studies and is highly relevant for the many adults seeking to improve health during midlife. See Figure S2 for model assumptions and accompanying directed acyclic diagrams.

We used multivariable-adjusted pooled logistic regression models to estimate the associations of long-term physical activity and diet quality with incident chronic disease across successive two-year follow-up intervals. Pooled logistic regression was selected to accommodate the biennial updating of exposures and to allow for direct inclusion of age in population preventable fraction models. This approach treats each two-year interval as a discrete follow-up period, with observations over all intervals combined into a pooled dataset. When outcome incidence within each interval is low, as in this study, the discrete-time odds ratios from pooled logistic regression closely approximate the hazard ratios obtained from a Cox proportional hazards model, and the resulting model coefficients approximate risk ratios.³⁰ Physical activity and diet quality were modeled as cumulative averages and lagged by two years to reduce potential reverse causation.

Risk ratios were estimated by comparing individuals in the highest versus lowest deciles of physical activity and diet quality for independent associations, and by jointly comparing individuals in the highest versus lowest deciles of both behaviors for combined effects. Deciles were chosen to enable the estimation of population preventable fractions at interpretable target intervention levels and to estimate joint effects.

To estimate the proportion mediated by cumulative average BMI (sustained adulthood habits) or change in body weight over follow-up (midlife changes in habits), we used the difference-in-coefficients approach implemented in the SAS %MEDIATE macro.^31,32 Counterfactual-based mediation methods were explored but were not feasible given the high-dimensional categorical exposure. Change in body weight was defined as the difference between cumulative average weight and prebaseline weight divided by prebaseline weight.³²

Statistical analysis: population preventable fraction (PPF)

We estimated overall chronic disease and disease-specific population preventable fractions (PPFs) using the estimated associations from the NHS, NHSII, and HPFS and the distribution of physical activity and diet quality in the US adult population, derived from the National Health and Nutrition Examination Survey (NHANES) 2017–2018. PPFs were estimated using the Miettinen formula for a multicategory protective factor, applying model-based risk ratio estimates obtained from the pooled logistic regression survival models.^33,34

We modeled three hypothetical population interventions on physical activity: ≥7.5 MET-hours/week (the lower end of the US Department of Health and Human Services [HHS] recommendation, 150 min/week of moderate-intensity aerobic activity), ≥15 MET-hours/week (the upper end, 300 min/week), and ≥47 MET-hours/week (the 90th percentile in the NHS/HPFS cohorts).³⁵ Similarly, we modeled three hypothetical diet quality interventions: modified AHEI score ≥40 points (50th percentile in the NHS/HPFS cohorts), ≥45 points (70th percentile), and ≥52 points (90th percentile). For each intervention, the PPF represents the proportion of incident cases of type 2 diabetes, ASCVD, or cancer that would be prevented if the US adult population distribution of physical activity or diet quality were shifted to the specified threshold value.

Incidence of type 2 diabetes, myocardial infarction, stroke, and cancer in the US were derived from the CDC US Diabetes Surveillance System (USDSS), the 2005 to 2014 ARIC study, and the National Cancer Institute Surveillance, Epidemiology, and End Results Program (see Supplemental Methods for more details).36, 37, 38, 39 Disease-specific relative risks were applied to US incidence for each outcome (type 2 diabetes, ASCVD, and cancer), and the composite preventable fraction reflects the weighted contribution of each condition to chronic disease.

Secondary and sensitivity analyses

We conducted several secondary and sensitivity analyses to assess results by subtype of cancer and ASCVD, effect modification, and robustness of results across methods (see Supplemental Methods for more details). We evaluated heterogeneity by sex in secondary analyses rather than the main analysis because the smaller number of male participants limited power for the multi-level joint exposure analyses. We did not present race- or ethnicity-stratified estimates due to insufficient power for subgroup analyses.

Role of funding source

This study was funded by grants from the US National Institutes of Health (NIH). The funders had no role in study design, data collection, data analysis, interpretation, writing of the report. The findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the NIH.

Results

Study population characteristics

The study population was 80% (113,727) female and 96% (136,998) non-Hispanic White, with mean age of 50 years at baseline (Table 1). Individuals with physical activity and diet quality above the median (≥13 MET-hours/week of activity and ≥39 points on the modified AHEI) were more likely to be older, live in ZIP-codes with higher median income, and have lower BMI. During 3 million person-years of follow-up, we documented 13,557 incident cases of type 2 diabetes, 11,989 cases of ASCVD, and 24,342 cases of cancer. Combined chronic disease consisted of 42,562 cases (28% [11,804] type 2 diabetes, 22% [9432] ASCVD, 51% [21,840] cancer). Breast (28% [6837]), colorectal (9% [2150]), and lung (9% [2155]) were the most common cancer types. The average length of follow-up was 17 years.

Table 1.

Baseline characteristics of study participants by level of physical activity and diet quality (NHS, 1990; NHSII, 1997; HPFS, 1990)^a.

	Overall (n = 142,041)	NHS		NHSII		HPFS
		<50th percentileb physical activity & diet quality (n = 13,530)	≥50th percentile physical activity & diet quality (n = 13,430)	<50th percentile physical activity & diet quality (n = 22,373)	≥50th percentile physical activity & diet quality (n = 13,978)	<50th percentile physical activity & diet quality (n = 3752)	≥50th percentile physical activity & diet quality (n = 12,719)
Age, years	50.2 (9.9)	54.9 (7.1)	57.3 (7.0)	42.0 (4.6)	42.6 (4.5)	56.3 (9.6)	58.3 (9.6)
Cohort, % (n)
NHS	35 (49,947)	–	–	–	–	–	–
NHSII	45 (63,780)	–	–	–	–	–	–
HPFS	20 (28,314)	–	–	–	–	–	–
Female, % (n)	80 (113,727)	100 (13,530)	100 (13,430)	100 (22,373)	100 (13,978)	0 (0)	0 (0)
Modified AHEI score, points out of 90c	38.9 (9.3)	31.8 (4.7)	47.6 (6.1)	30.4 (5.2)	45.6 (5.2)	31.5 (4.8)	47.5 (6.3)
Physical activity, MET-hours/week	19.1 (19.1)	4.9 (3.4)	29.8 (15.5)	5.4 (3.5)	32.4 (18.6)	6.5 (3.5)	38.9 (22.3)
BMI at age 18–21, kg/m2	21.5 (2.9)	21.1 (2.7)	21.3 (2.7)	21.1 (3)	21.2 (2.9)	22.7 (2.9)	23.1 (2.5)
BMI, kg/m2	25.4 (4.3)	25.7 (4.4)	24.6 (3.7)	26.4 (5.5)	24.4 (4.3)	26.0 (3.0)	25.1 (2.6)
BMI category
Lean BMI (<25 kg/m2), % (n)	55 (77,563)	52 (7057)	62 (8348)	50 (11,184)	66 (9249)	40 (1504)	52 (6635)
Overweight BMI (25–29.9 kg/m2), % (n)	32 (45,272)	32 (4319)	29 (3879)	28 (6155)	24 (3323)	50 (1874)	43 (5455)
Obese BMI (≥30 kg/m2), % (n)	14 (19,206)	16 (2154)	9 (1203)	23 (5034)	10 (1406)	10 (374)	5 (629)
Postmenopausal, % of women (n)	41 (46,444)	71 (9661)	82 (10,955)	12 (2758)	12 (1727)	–	–
Menopausal hormone therapy
Never MHT use, % of postmenopausal women (n)	39 (18,228)	44 (4280)	39 (4261)	28 (769)	29 (509)	–	–
Past MHT use, % of postmenopausal women (n)	18 (8317)	20 (1911)	20 (2216)	7 (188)	7 (122)	–	–
Current MHT use, % of postmenopausal women (n)	43 (19,899)	36 (3470)	41 (4478)	65 (1801)	63 (1096)	–	–
Total energy intake, calories/day	1818 (513)	1822 (435)	1721 (438)	1855 (537)	1730 (534)	1977 (515)	1918 (534)
Alcohol, g/day	5.8 (9.8)	5.7 (10.0)	6.0 (8.7)	3.0 (6.6)	4.3 (6.8)	11.7 (15.7)	10.4 (12.6)
Height, cm	167.2 (8.4)	164.0 (6.0)	163.9 (6.0)	164.9 (6.5)	165.0 (6.5)	178.4 (6.4)	178.3 (6.4)
Non-Hispanic White, % (n)	96 (136,998)	98 (13,278)	97 (13,050)	96 (21,532)	95 (13,276)	96 (3610)	96 (12,178)
Median income from census data & ZIP code, USD	45,523 (16,771)	44,902 (15,562)	48,304 (18,356)	43,409 (14,100)	48,584 (17,063)	42,133 (17,380)	44,494 (19,132)
Regular multivitamin use, % (n)	55 (77,691)	43 (5751)	52 (7006)	59 (13,097)	67 (9415)	46 (1710)	53 (6749)
Regular aspirin use (>2 tabs/week), % (n)	23 (33,028)	42 (5743)	41 (5476)	8 (1871)	9 (1251)	23 (845)	26 (3259)
Pack-years of smoking	8.5 (15.0)	13.0 (19.5)	11.1 (16.6)	4.5 (9.0)	4.2 (7.6)	15.2 (20.9)	10.7 (16.4)
Smoking status
Never smoked, % (n)	56 (78,966)	47 (6374)	44 (5965)	68 (15,143)	63 (8842)	46 (1728)	51 (6460)
Past smoker‚ quit ≥10 years ago, % (n)	24 (33,635)	22 (2940)	30 (4073)	15 (3250)	21 (2983)	29 (1079)	36 (4580)
Past smoker‚ quit <10 years ago, % (n)	10 (14,040)	11 (1488)	15 (1963)	7 (1513)	8 (1118)	12 (466)	9 (1130)
Current smoker, % (n)	11 (15,400)	20 (2728)	11 (1429)	11 (2467)	7 (1035)	13 (479)	4 (549)
Ever had endoscopy for routine screening, % (n)	15 (21,550)	14 (1838)	21 (2801)	5 (1010)	5 (743)	28 (1068)	39 (5005)
Ever had a mammogram for routine screening, % of women (n)	82 (93,825)	84 (11,414)	90 (12,082)	77 (17,316)	82 (11,439)	–	–
Family history of cancer, % (n)	31 (44,601)	45 (6133)	46 (6227)	25 (5688)	26 (3671)	18 (677)	19 (2447)
Family history of cardiovascular disease, % (n)	36 (51,156)	33 (4401)	34 (4577)	40 (8912)	39 (5460)	33 (1241)	35 (4398)
Family history of type 2 Diabetes, % (n)	24 (34,124)	26 (3559)	27 (3566)	26 (5817)	24 (3396)	18 (691)	17 (2206)

^aAll values, except age, are age-adjusted to the distribution of the cohort. Data are given as mean (standard deviation), unless otherwise specified. AHEI denotes alternative healthy eating index; BMI, body mass index; HPFS, Health Professionals Follow-up Study; MET, metabolic equivalent of task; MHT, menopausal hormone therapy; NHS, Nurses' Health Study; USD, United States dollars; ZIP, zone improvement plan.

^b50th percentile of physical activity in the overall study population was 13 MET-hours/week of activity; 50th percentile of modified AHEI score was 39 of 90 possible points.

^cModified AHEI score is the traditional AHEI-2010 score minus the alcohol and trans-fatty acids components (range = 0–90 points).

At baseline, mean BMI was 24.4 (standard deviation [SD] 3.8), and by year 16 of follow-up, cumulative average BMI was 25.9 (SD 4.3). Participants gained an average of 5.7 lb (SD 7.9), with higher average weight gain among individuals in the lowest quartile of baseline BMI (6.4 lb) and lower average weight gain among those in the highest baseline quartile (4.6 lb). Less than 0.3% of study participants reported a history of weight-loss medication.

Individual and combined associations of physical activity and diet quality

Physical activity and diet quality were independently associated with lower risk of chronic disease (Fig. 1 and Table 2). For midlife changes, the associations were slightly weaker than sustained adulthood habits for both physical activity and diet quality. When modeled as a continuous variable, each additional 17.5 MET-hours/week of physical activity (equivalent to 30 min of moderate-intensity activity per day) was associated with 10% lower risk of chronic disease (95% CI 0.89–0.91) for sustained adulthood habits or 7% lower risk (95% CI 0.92–0.94) for midlife changes. Each 10-point improvement in AHEI score was associated with 6% lower risk of chronic disease (95% CI 0.93–0.95) for sustained adulthood habits or 5% lower risk (95% CI 0.94–0.97) for midlife changes. Combined highest deciles of physical activity and diet quality were associated with risk ratios of 0.54 (95% CI 0.48–0.60) for sustained adulthood habits and 0.66 (95% CI 0.59–0.74) for midlife changes (Fig. 1 and Tables S1–S2). The magnitude of association varied by disease type, strongest for type 2 diabetes, followed by ASCVD. For cancer, a statistically nonsignificant association was found for sustained adulthood habits but not midlife changes.

Fig. 1.

Risk ratios for incident chronic disease comparing extreme deciles of physical activity, diet quality, and combined physical activity and diet quality. Panels show risk ratios for (a) total chronic disease, defined as the first occurrence of type 2 diabetes, atherosclerotic cardiovascular disease, or cancer; (b) type 2 diabetes; (c) atherosclerotic cardiovascular disease; (d) all cancer; and (e) obesity-related cancer. Extreme deciles compare high versus low physical activity, diet quality, and combined physical activity and diet quality, as defined in the figure legend. Circles indicate risk ratios and error bars indicate 95% confidence intervals. AHEI denotes alternative healthy eating index; BMI, body mass index; MET, metabolic equivalent of task. ∗ Modified AHEI score is the traditional AHEI-2010 score minus the alcohol and trans-fatty acid components (range = 0–90 points). † To estimate sustained adulthood habits, models were adjusted for cohort; age (linear and quadratic); BMI at age 18–21; total energy intake; alcohol intake; multivitamin use (yes or no); menopausal status; menopausal hormone therapy use (current, former, never); total pack-years of smoking and smoking status (current, former, never smoker); reported family history of type 2 diabetes, cardiovascular disease, or cancer; estimated median income (from census data and ZIP code). All confounders were modeled as linear variables unless otherwise noted and updated at two-year intervals over follow-up. Type 2 diabetes models were additionally adjusted for family history of type 2 diabetes; atherosclerotic cardiovascular disease models were additionally adjusted for family history of atherosclerotic cardiovascular disease and regular aspirin use; all cancer and obesity-related cancer models were additionally adjusted for family history of cancer, regular aspirin use, and height. Combined chronic disease models were additionally adjusted for family history of type 2 diabetes, atherosclerotic cardiovascular disease, or cancer; regular aspirin use; and height.‡ To estimate midlife changes in habits, models were adjusted for the same variables included in sustained adulthood habits models and additionally adjusted for prebaseline physical activity (MET-hours/week), prebaseline modified AHEI score, prebaseline BMI, prebaseline use of anti-hypertensive or cholesterol-lowering prescription drugs or history of high cholesterol or hypertension, and prebaseline diagnosis with any of the other chronic disease outcomes for individual disease categories (i.e., when type 2 diabetes was the outcome, models were adjusted for personal history of atherosclerotic cardiovascular disease or cancer at baseline).

Table 2.

Risk ratios for incident total chronic disease (type 2 diabetes, atherosclerotic cardiovascular disease, or cancer) before and after additional adjustment for cumulative BMI (sustained adulthood habits) or change in body weight over follow-up (midlife changes in habits).

Sustained adulthood habitsb	Base model	Additional adjustment for cumulative BMI
	Risk ratio (95% confidence interval)	Risk ratio (95% confidence interval)	Percent of effect mediated by cumulative BMI
Physical activity (MET-hours/week)	0.68 (0.65–0.71)	0.86 (0.82–0.90)	60% (52–68%)
Diet quality (modified AHEI scorea)	0.82 (0.78–0.86)	0.90 (0.85–0.94)	44% (34–55%)
Combined physical activity + diet quality	0.54 (0.48–0.60)	0.74 (0.67–0.83)	52% (42–61%)

Midlife changes in habitsc	Base model	Additional adjustment for change in body weight over follow-up
	Risk ratio (95% confidence interval)	Risk ratio (95% confidence interval)	Percent of effect mediated by change in body weight
Physical activity (MET-hours/week)	0.79 (0.75–0.83)	0.84 (0.80–0.88)	27% (21–34%)
Diet quality (modified AHEI score)	0.86 (0.81–0.91)	0.90 (0.85–0.95)	30% (20–42%)
Combined physical activity + diet quality	0.66 (0.59–0.74)	0.74 (0.66–0.83)	27% (20–35%)

AHEI denotes alternative healthy eating index; BMI, body mass index; MET, metabolic equivalent of task.

^aModified AHEI score is the traditional AHEI-2010 score minus the alcohol and trans-fatty acids components (range = 0–90 points).

^bTo estimate sustained adulthood habits, models were adjusted for cohort; age (linear and quadratic); BMI at age 18–21; total energy intake; alcohol intake; multivitamin use (yes or no); menopausal status; menopausal hormone therapy use (current, former, never); total pack-years of smoking and smoking status (current, former, never smoker); reported family history of type 2 diabetes, cardiovascular disease, or cancer; estimated median income (from census data and ZIP code). All confounders were modeled as linear variables unless otherwise noted and updated at two-year intervals over follow-up. Type 2 diabetes models were additionally adjusted for family history of type 2 diabetes; atherosclerotic cardiovascular disease models were additionally adjusted for family history of atherosclerotic cardiovascular disease and regular aspirin use; all cancer and obesity-related cancer models were additionally adjusted for family history of cancer, regular aspirin use, and height. Combined chronic disease models were additionally adjusted for family history of type 2 diabetes, atherosclerotic cardiovascular disease, or cancer; regular aspirin use; and height.

^cTo estimate midlife changes in habits, models were adjusted for the same variables included in sustained adulthood habits models and additionally adjusted for prebaseline physical activity (MET-hours/week), prebaseline modified AHEI score, prebaseline BMI, prebaseline use of anti-hypertensive or cholesterol-lowering prescription drugs or history of high cholesterol or hypertension, and prebaseline diagnosis with any of the other chronic disease outcomes for individual disease categories (i.e., when type 2 diabetes was the outcome, models were adjusted for personal history of atherosclerotic cardiovascular disease or cancer at baseline).

Population preventable fraction

The population preventable fraction of total chronic disease with sustained adulthood habits of combined improved physical activity and diet quality ranged from 15% (95% CI 10–19%; 503,000 cases per year) with minimal intervention to 34% (95% CI 27–41%; 1,141,000 cases per year) with optimal intervention (Fig. 2). For midlife changes, the population preventable fraction ranged from 8% (95% CI 3–13%; 268,000 cases per year) with minimal intervention to 26% (95% CI 18–34%; 873,000 cases per year) with optimal intervention (Fig. 3). These estimates incorporate the disease-specific preventable fractions for diabetes, ASCVD, and cancer weighted by their respective US incidence.

Fig. 2.

Population preventable fraction showing the estimated percent and number of cases per year of incident total chronic disease (type 2 diabetes, cardiovascular disease, or cancer) and individual major categories of chronic disease that could theoretically be prevented if current levels of physical activity and diet quality in the United States were replaced by minimal, moderate, and optimal target levels of activity and diet quality. Physical activity and diet quality are modeled as sustained adulthood habits. Error bars indicate 95% confidence intervals. AHEI denotes alternative healthy eating index; BMI, body mass index; MET, metabolic equivalent of task. ∗ Modified AHEI score is the traditional AHEI-2010 score minus the alcohol and trans-fatty acid components (range = 0–90 points). † To estimate sustained adulthood habits, models were adjusted for cohort; age (linear and quadratic); BMI at age 18–21; total energy intake; alcohol intake; multivitamin use (yes or no); menopausal status; menopausal hormone therapy use (current, former, never); total pack-years of smoking and smoking status (current, former, never smoker); reported family history of type 2 diabetes, cardiovascular disease, or cancer; estimated median income (from census data and ZIP code). All confounders were modeled as linear variables unless otherwise noted and updated at two-year intervals over follow-up. Type 2 diabetes models were additionally adjusted for family history of type 2 diabetes; atherosclerotic cardiovascular disease models were additionally adjusted for family history of atherosclerotic cardiovascular disease and regular aspirin use; all cancer and obesity-related cancer models were additionally adjusted for family history of cancer, regular aspirin use, and height. Combined chronic disease models were additionally adjusted for family history of type 2 diabetes, atherosclerotic cardiovascular disease, or cancer; regular aspirin use; and height. ‡ Since negative values for number of preventable cases are not logically meaningful, lower bounds were set to 0. Confidence intervals with a lower bound of 0 should therefore be interpreted as including the null value.

Fig. 3.

Population preventable fraction showing the estimated percent and number of cases per year of incident total chronic disease (type 2 diabetes, cardiovascular disease, or cancer) and individual major categories of chronic disease that could theoretically be prevented if current levels of physical activity and diet quality in the United States were replaced by minimal, moderate, and optimal target levels of activity and diet quality. Physical activity and diet quality are modeled as midlife changes in habits. Error bars indicate 95% confidence intervals. AHEI denotes alternative healthy eating index; BMI, body mass index; MET, metabolic equivalent of task. ∗ Modified AHEI score is the traditional AHEI-2010 score minus the alcohol and trans-fatty acid components (range = 0–90 points). † To estimate midlife changes in habits, models were adjusted for the same variables included in sustained adulthood habits models and additionally adjusted for prebaseline physical activity (MET-hours/week), prebaseline modified AHEI score, prebaseline BMI, prebaseline use of anti-hypertensive or cholesterol-lowering prescription drugs or history of high cholesterol or hypertension, and prebaseline diagnosis with any of the other chronic disease outcomes for individual disease categories (i.e., when type 2 diabetes was the outcome, models were adjusted for personal history of atherosclerotic cardiovascular disease or cancer at baseline). ‡ Since negative values for number of preventable cases are not logically meaningful, lower bounds were set to 0. Confidence intervals with a lower bound of 0 should therefore be interpreted as including the null value.

Secondary analyses

Among individual cancers, physical activity and combined physical activity and diet quality were associated with higher risk of melanoma and lower risk of postmenopausal breast cancer; other associations were not statistically significant (Figure S3). Results for ASCVD subtypes, effect modification, and alternative dietary patterns were consistent with the main analyses and are detailed in Figures S4–S7. In sex-stratified analyses, the direction of associations was generally similar in women and men, although estimates among men were less precise due to smaller sample size (Figure S5). Among individual dietary components, high coffee intake, low red and processed meat intake, and high cereal fiber intake were associated with the most substantially lowered risk of chronic disease (Figure S8).

Sensitivity analyses

The estimated 18-year cumulative incidence of chronic disease using the midlife changes model was 26% (95% CI 24–28%; 37,210 cases) if all participants were in the highest quintiles of physical activity and diet quality, compared with 19% (95% CI 17–21%; 27,215 cases) if all were in the lowest quintiles (Figure S9). The difference was more pronounced for individuals with prebaseline BMI ≥25 and remained consistent across follow-up. Results among participants with stable BMI were consistent with those following additional adjustment for cumulative BMI or change in weight over follow-up (Figure S10). Results were unchanged after censoring individuals who did not return two subsequent questionnaires. Sensitivity analysis using multiple imputation for missing exposure, mediator, and select covariates produced similar results to the complete case method used in the main analysis, though several effect sizes for all cancer and obesity cancer were larger and reached statistical significance (Figure S11).

Discussion

In three longitudinal cohorts of US female nurses and male health professionals, higher levels of physical activity and better diet quality were associated with substantially lower risk of chronic disease independent of BMI. By adjusting for BMI at multiple life stages, we were able to distinguish the influence of habits sustained across adulthood compared to midlife changes, providing evidence that both long-term and later-life improvements in lifestyle are linked with reduced risk of chronic disease. Associations persisted after additional adjustment for change in body weight, suggesting that the benefits of physical activity and diet extend beyond their effects on BMI.

Replacing current population levels of physical activity and diet quality with optimal levels could potentially prevent 34% (1,141,000 cases per year) of chronic disease incidence with sustained adulthood habits and 26% (873,000 cases per year) with midlife changes. These estimates indicate large potential health gains from improving health behaviors both early and later in life. Importantly, this composite outcome comprises conditions with varying magnitudes of association with physical activity and diet: diabetes and ASCVD show substantial relative reductions, whereas cancer, despite representing a large share of total cases, showed a more modest risk reduction in association with these lifestyle changes. The composite reflects the total preventable burden across leading causes of chronic disease, providing a summary measure directly relevant to public health planning. While the magnitude of association differed across component outcomes, the composite captures overall preventable chronic disease incidence attributable to lifestyle change.

We used the 90th percentile AHEI score and physical activity level in the cohorts as the “optimal” target intervention for estimating the PPF. The optimal diet quality intervention (52 out of 90 possible points) was well below a perfect score, whereas the optimal physical activity intervention (47 MET-hours, or roughly 600 min of moderate-intensity activity per week) far exceeded the current US guideline of 150–300 min per week. Risk declined linearly with greater activity, indicating that any activity is better than none, but our findings suggest that guideline targets may represent a minimum rather than an optimal level for chronic disease prevention.

Prior analyses of lifestyle-related preventable fractions for chronic disease incidence have reported a wide range of estimates, reflecting differences in study design and definition of optimal physical activity and diet quality. For type 2 diabetes, estimates from the NHS ranged from 24 to 45% when stratified by baseline BMI, and a study in a more diverse cohort reported preventable fractions of 13–29% for physical activity alone.^40,41 For cerebrovascular disease, recent meta-analyses generally report more modest preventable fractions—roughly 3–12% for physical activity and diet when examined individually—whereas coronary disease showed a larger preventable fraction of 28% for combined activity and diet in a prior NHS analysis.^42,43 For cancer, a large pooled analysis similarly reported low preventable fractions of 1–3% for activity and dietary components evaluated separately.⁴⁴

Our estimates for sustained adulthood habits, which are most comparable to the methods used in these earlier studies, largely fall within the ranges reported by others, and differences are likely explained by our higher “optimal” activity and diet targets and longer follow-up period.40, 41, 42, 43, 44 Overall, our study extends this literature by estimating the preventable fractions for both sustained adulthood habits and midlife behavior changes, while accounting for the roles of BMI and weight change. This approach provides a more nuanced and internally coherent estimate of the magnitude of chronic disease prevention achievable through realistic behavioral modification across adulthood.

Though it is challenging to directly compare our risk ratio estimates to those from randomized trials because long-term lifestyle trials are uncommon, the available evidence provides important support for causal effects and context within which to place our findings. Randomized interventions consistently show 50–60% reductions in type 2 diabetes incidence with physical activity and diet intervention, even over relatively short periods.^45,46 The PREDIMED study demonstrated a 28–31% reduction in major cardiovascular events with a Mediterranean-style diet, even in the absence of weight change.⁴⁷ These results align with our findings that risks of both type 2 diabetes and cardiovascular disease were substantially reduced in both sustained adulthood and midlife changes models, and are consistent with evidence that short-term changes in physical activity and diet improve multiple markers of cardiometabolic health such as insulin resistance, inflammation, and cardiorespiratory fitness.48, 49, 50

Randomized trial findings for cancer are more variable and often site-specific. The Mediterranean diet intervention in PREDIMED resulted in fewer invasive breast cancers in a secondary analysis, and the Look AHEAD trial demonstrated a suggestive reduction in obesity-related cancers with an intensive lifestyle intervention.^51,52 In contrast, long-term follow-up of the Diabetes Prevention Program trial showed no differences in cancer incidence after 21 years, which may have been influenced by attenuation in lifestyle change over time.⁵³ Taken together, these trials align with our observation that reductions in cancer risk are more evident in the context of sustained improvements in activity and diet quality across adulthood, rather than midlife changes alone. This pattern is biologically plausible, as cancer often has long latency periods, heterogeneous etiologic pathways, and potential sensitivity to exposures early in life.⁵⁴ At the same time, the signals from PREDIMED and Look AHEAD provide encouraging evidence that meaningful cancer prevention may still be achievable through lifestyle changes. Although complete case analysis was used as the primary analytic approach—in part because pooling population preventable fraction estimates across imputed datasets is methodologically non-trivial, and in part due to the computational demands of the primary analysis across three large cohorts and numerous exposure cycles—a sensitivity analysis using multiple imputation produced materially consistent results. The slightly stronger cancer effect sizes observed under multiple imputation suggest that the modest associations in the primary analysis may be conservative, and that complete case analysis did not meaningfully bias our findings. Finally, it is worth noting that bariatric surgery studies show substantial reductions in cancer risk over shorter periods, suggesting that the profound metabolic changes achieved through surgical intervention can have a meaningful influence on cancer development.⁵⁵

Physical activity and diet quality were associated with lower chronic disease risk across BMI categories, including among individuals who remained weight-stable. After further adjustment for weight change, midlife improvements in activity and diet were still linked with a 26% reduction in chronic disease risk, underscoring that benefits extend beyond weight loss. These findings align with weight-neutral approaches to public health communication, and support messaging that encourages engagement in health-promoting behaviors for the intrinsic benefits, independent of weight loss through pharmacologic therapy, bariatric surgery, or caloric restriction.⁵⁶ At the same time, our study did not directly compare healthy habits alone to healthy habits with weight loss or maintenance of a specific BMI range.

A major strength of our study is several decades of longitudinal follow-up. Repeated assessments improved the validity of physical activity and diet quality metrics and allowed us to separate the confounding effects of prebaseline BMI from the mediating effects of weight change during follow-up. The near-negligible prevalence of weight-loss medication use supports the study objective of quantifying the health benefits of activity and diet quality independent of pharmacologically induced weight loss, which may confound similar analyses in more contemporary cohorts. With 42,562 chronic disease cases, we had power to estimate risk ratios and population preventable fractions using fine stratification of combined deciles of lifestyle behaviors, which would not be feasible in smaller cohorts.

Our study has several limitations. Exposures were self-reported and subject to non-differential measurement error, which typically biases associations towards the null. Physical activity and diet quality were assessed over several decades, during which activity patterns and dietary recommendations evolved, potentially increasing misclassification. However, the use of standardized MET-hours for activity, AHEI for a broad and stable dietary pattern, and consistent questionnaires with documented reproducibility all help mitigate this bias. Risk ratios from observational data also depend on the assumption of minimal residual confounding. Although we adjusted for a wide range of covariates, several key factors were not available, including sleep duration and quality and more granular socioeconomic indicators (income was approximated by ZIP-code median income, and education and occupation were assumed relatively homogeneous within each cohort of health professionals). Individuals achieving the highest levels of recreational activity may differ in unmeasured ways, such as baseline health or access to recreational space, which could contribute to residual confounding at the upper end of the activity range. We carried forward exposure data when missing for a single questionnaire cycle, which may introduce nondifferential exposure misclassification and attenuate associations, however, missingness was low and participants with extended exposure missingness were excluded. Because the difference-in-coefficients method assumes no unmeasured mediator–outcome confounding and no strong exposure–mediator interaction, the mediated proportion should be interpreted as an approximation.

Several unmeasured or incompletely measured confounders merit additional consideration. Sleep duration and quality represent a potentially important source of residual confounding; like physical activity and diet, poor sleep tracks with unhealthy lifestyle behaviors and elevated cardiometabolic risk. Psychosocial stress, depression, and built environment factors track similarly with behavior and health and are not fully captured by our covariates and ZIP-code income measure. For diabetes and ASCVD, these factors are well-established positive confounders and would tend to modestly inflate the apparent protective associations.57, 58, 59 For cancer, the same factors show less consistent associations in the literature, and the net direction of residual confounding is less certain.^60,61 We adjusted for two major cancer screening modalities (endoscopy and mammography), somewhat mitigating concern that differential screening uptake by lifestyle behaviors accounts for observed associations; however, residual confounding from unmeasured cancer-specific risk factors—particularly for cancers for which etiologies remain incompletely understood—results in more uncertainty regarding the net direction of confounding.

Another limitation is that the study cohorts used to derive the risk ratios consisted of primarily non-Hispanic White health professionals, who tended to have higher health literacy and healthier behaviors than the general US population. Race and ethnicity are sociocultural constructs that may reflect underlying differences in structural determinants of health, including access to resources, structural discrimination, and built environment. In applying risk ratios from these cohorts to nationally representative NHANES distributions, we assumed the relative associations between lifestyle factors and chronic disease are transportable across racial/ethnic and socioeconomic groups. Prior research supports this assumption, showing that effect estimates from these cohorts are similar to those derived from pooled estimates from more diverse cohorts.⁶² In addition, our population preventable fractions are based on recent exposure distributions, improving the relevance of these estimates for US adults today.

In conclusion, improving physical activity and diet quality could prevent a substantial fraction of incident chronic disease in the US, with the most substantial reductions for type 2 diabetes and cardiovascular disease, independent of BMI. These findings underscore the importance of supporting healthy behaviors at the population level, regardless of weight status. Future work should develop strategies to increase access and adherence to physical activity and healthy diet, which could yield major public health benefits. It will also be important to identify accessible metrics that mediate the effects of healthy behaviors on chronic disease risk and could guide behavior change beyond reliance on BMI.

Contributors

AMB and MS conceptualized the study. AMB curated the data, developed the methodology, conducted the formal analyses, and drafted the original manuscript. MD contributed to the investigation, validation, and manuscript review and editing. AHE, EG, LL, MJS, ER, WW, and FBH contributed to the investigation, provided supervision, and reviewed and edited the manuscript. MS contributed to investigation, supervision, and manuscript review and editing. All authors contributed to the interpretation of the findings, critically reviewed the manuscript, approved the final version, and agreed to submit for publication.

Data sharing statement

The data that support the findings of this study are available on request at https://www.nurseshealthstudy.org/researchers (email: nhsaccess@channing.harvard.edu) and https://sites.sph.harvard.edu/hpfs/for-collaborators/. The data are not publicly available due to privacy and ethical restrictions.

Declaration of interests

FH has a research grant from Analysis Group that is not directly related to content of this paper. EBR is on the scientific advisory board for Wilder supermarket store, Alma Food, and for GoodeHealth and receives grant funding from the USDA and the National Institutes of Health; none of these are directly related to the contents of this paper. MS reports grant funding from the National Institutes of Health. All other authors declare no competing interests.