Prospective study of breakfast frequency and timing and risk of myocardial infarction and coronary artery disease in community-dwelling older adults: The Cardiovascular Health Study

Abstract

Objectives

Because it is unknown whether breakfast frequency and timing are associated with long-term risk of incident myocardial infarction (MI) and coronary artery disease (CAD) among older adults, this study aimed to assess relationships between breakfast frequency/timing and MI/CAD risk among older adults and determine whether they depend on sex or cardiometabolic risk factors.

Design and setting

Prospective cohort study of older American adults.

Participants and measurements

Weekly breakfast frequency and usual daily breakfast time were assessed by questionnaire in 4070 adults aged ≥ 65 years from the Cardiovascular Health Study who were prospectively followed for up to 26 years. Cause-specific hazard ratios (HRs) with 95% confidence intervals (CIs) were estimated from multivariable-adjusted Cox proportional hazards models.

Results

During follow-up, 1617 CAD cases were documented (795 MI cases). Although consuming breakfast 7 days/week (85.3%) and ‘breaking-fast’ between 07:00 and 09:00am (72.6%) were both associated with higher education and socioeconomic status, being married, not smoking, and consuming fruits and vegetables, neither breakfast frequency nor breakfast timing was associated with risk of CAD or MI in males, females, or altogether. In pre-specified analyses, compared with participants who ate breakfast daily, those who did not eat breakfast daily had an HR for MI of 0.66 (95% CI: 0.43, 1.02) if their body mass index (BMI) was ≥30 kg/m² and of 1.17 (0.91, 1.51) if their BMI was <30 kg/m² (interaction p = 0.02). Compared with participants whose breakfast time was 07:00–09:00, those who broke their fast before 07:00 had an HR for CAD of 1.40 (1.02, 1.93) if they had type 2 diabetes and of 1.19 (1.03, 1.38) if they had high fasting insulin at baseline.

Conclusion

Breakfast frequency and timing were not associated with either higher or lower risk of MI and CAD in these older adults. Although a priori stratification by cardiometabolic risk factors may have revealed potential trends, the findings must be confirmed in a larger study.

1. Introduction

Coronary artery disease (CAD) is a leading cause of disability and death worldwide, disproportionately affecting older adults [1,2]. Nutrition is an established modifiable risk factor for CAD, and healthy eating is a cornerstone of CAD prevention and treatment programs [[3], [4], [5]]. An individual’s diet can be assessed for both quality (‘what’ they eat) and frequency/timing (‘how often’ and ‘when’ they eat), with the majority of research to date focused on quality. Official evidence-based recommendations do not exist for how often and at what time of day an adult should have meals, snacks, and calorie-containing drinks due to limited studies on the topic [[6], [7], [8]]. Guideline committees acknowledge that frequency of eating is an important factor that could alter pathophysiological processes to influence a person’s cardiometabolic health status and thus is a research priority [6,7].

Two meta-analyses of observational studies on the association between skipping breakfast and risk of cardiovascular disease both concluded that skipping breakfast increases the risk of cardiovascular disease [9,10]. This could have widespread population impacts because the frequency of skipping breakfast among adults has steadily risen over the past decades [[11], [12], [13]]. Since none of the studies included in the meta-analyses were specifically in older adults [9,10], a vital and growing demographic, it remains unknown whether these findings are generalizable to older adults. Despite the meta-analysis findings, breakfast frequency has been a topic of controversy, as researchers have found contradicting evidence for the importance of breakfast eating versus breakfast skipping and the potential role of breakfast timing and frequency in maintaining good cardiovascular health as measured by biomarkers associated with CAD risk [14]. Omitting breakfast has been reported to impair serum lipids and postprandial insulin sensitivity in multiple randomized, crossover trials [15,16] Conversely, other studies showed that skipping breakfast was protective against obesity, insulin resistance, hypertension, and other common CAD disease risk factors [[17], [18], [19]]. Potentially the impact of breakfast timing and frequency depends on the existing health status of participants, and data on the specific timing of breakfast could also potentially clarify findings to date.

A recent study reported that adults (mean age of 41 years) who generally eat later in the day had higher cardiometabolic risk factors such as body mass index (BMI), body fat, waist circumference, insulin resistance, and triglycerides compared to early eaters [20]. In younger adults, skipping meals has been shown to increase the risk of developing type 2 diabetes specifically among people with other cardiometabolic risk factors [21]. To our knowledge, no studies have previously examined the effect of the timing of breakfast on the risk of developing CAD over time in older adults. Additionally, it remains unclear if having the cardiometabolic risk factors makes older adults more vulnerable to the influence of breakfast timing, putting older adults with cardiometabolic risk factors at different risk of CAD if they have a later eating timing.

The Cardiovascular Health Study (CHS) of older adults included measurements of the frequency and timing of breakfast [22]. The present study aimed to prospectively assess the relationship between breakfast frequency and timing and long-term risk of incident myocardial infarction (MI) and CAD among older adults using these novel data. The objectives of our present analysis of CHS data were to 1) prospectively determine whether breakfast frequency and timing are associated with risk of incident MI and CAD in older adults, and 2) test whether these associations depended on sex or common baseline cardiometabolic risk factors: lower income, type 2 diabetes (T2D), BMI (in kg/m²) ≥30, impaired fasting glucose, fasting insulin ≥10 IU/mL, hypertension, and hypercholesterolemia.

2. Materials and methods

2.1. Data source and study population

Data for this study are from the CHS (registered at clinicaltrials.gov as NCT00005133), which is a prospective, observational cohort study designed with the purpose of identifying risk factors for the onset of CAD and stroke in older adults, as previously reported [23]. Starting in 1989, the CHS recruited participants who were 65 years or older from the American communities of Forsyth County in North Carolina, Sacramento County in California, Washington County in Maryland, and Pittsburgh in Pennsylvania. There were 5888 participants enrolled in the study, 3393 (57.6%) of whom were female. During 1989−90, an initial cohort of 5201 predominately White Americans was enrolled, and three years later, a second cohort of 687 predominantly Black participants was recruited. Participants were eligible to participate in the CHS if they were 65 years and older, not institutionalized, able to give informed consent without a proxy at baseline, and were planning to remain in their respective areas for the next three years. Participants who were wheelchair-users at home at baseline, receiving hospice care, or receiving cancer therapy (radiation or chemotherapy) were excluded. Each center’s institutional review board reviewed and approved the study protocol, and all participants gave written informed consent. Eligible and consenting participants underwent baseline examinations, which consisted of a home interview and clinical exams. Participants attended in-clinic evaluations annually and were contacted by telephone at 6-mo intervals between 1989 and 1999. Semi-annual follow-up was continued by telephone after 1999 to ascertain health status and events information. Data collection tools available online at https://chs-nhlbi.org/.

In this present analysis, participants were excluded (in this order) if they had prevalent CAD at baseline (n = 1154), did not answer the eating timing and frequency questions (n = 18), or if they had cancer (except nonmelanoma skin cancer) during the 5 years preceding baseline (n = 147). We excluded participants with a history of cancer because these participants may have changed their eating timing and frequency patterns owing to health concerns, treatment, treatment side effects, or symptoms. No participants were missing CAD status at baseline or during follow-up. Participants with missing covariable data (n = 499) were excluded. Most covariables had less than 1% missing data, with 1–2% missing data for fasting glucose, fasting insulin, and baseline T2D, and 6% missing data for income. After exclusions, 4070 participants were available for the analysis (Supplementary Fig. 1).

2.2. Eating timing and frequency and other dietary assessment

Usual dietary habits including breakfast frequency, meal frequency, after-dinner snacking, fruit and vegetable intake, breakfast timing, and evening food and drink consumption timing were assessed at the baseline visit for both CHS cohorts in the nutrition portion of the baseline questionnaire [24]. Participants were asked to respond to the following question: “How often do you eat breakfast? Every day, some days, rarely, weekends only, never” and their response was used to determine their breakfast frequency defined as either not daily (never, some days, rarely, or weekends only) or daily (every day). Participants were asked, “On Monday through Friday, about what time do you usually first eat or drink something after waking up?” which was used to calculate their ‘break-fast’ timing after waking up. While no definition exists for breakfast content and timing in the CHS or is agreed-upon elsewhere, according to the American Heart Association, a used definition of breakfast in epidemiologic studies is the consumption of food or beverage (excluding water) between 05:00 and 09:00 [6]. Given that only a small number of participants ate or drank for the first time of the day before 05:00, we used the 07:00 cutoff in categorizing the breakfast timing variable (before 07:00; 07:00–09:00; and after 09:00). Because participants who reported never eating breakfast still broke their overnight fast, they were still included in the ‘break-fast” timing analysis in the ‘after 9:00’ strata, although we additionally conducted a sensitivity analysis that excluded them.

Participants were also asked “On Monday through Friday, how many meals do you usually eat per day?” and “On Monday through Friday, how many snacks do you eat after dinner?” These questions were used to calculate meal frequency (categorized as <3 meals/day, ≥3 meals/day) and after dinner snacks (yes, no) respectively. Additionally, participants were asked about usual fruit and vegetable consumption servings (which were categorized as <3 and ≥3 servings/day).

2.3. Measurement of covariables

Information on age, sex, race, marital status, household income, education, smoking status, alcohol intake, exercise (physical activity), depressive symptoms, body mass index (BMI), hypertension, and hypercholesterolemia was collected at the baseline visit. Only two options for sex were presented to participants at baseline: female or male. To ascertain information on race, participants were asked to respond to the following prompt: “Please look at this card and tell me which best describes your race: White, Black, American Indian/Alaskan native, Asian/Pacific Islander, Other.” We created a new dichotomous variable for race (largest racial group in the CHS: no, yes) because some categories of the original race variable contained too few participants for analysis. Hypertension was defined in the CHS as seated average systolic blood pressure ≥160 mm Hg or seated average diastolic blood pressure ≥95 mm Hg or use of antihypertensive medication with self-reported history of hypertension. Hypercholesterolemia was defined as total cholesterol ≥240 mg/dL as per the National Cholesterol Education Program Adult Treatment Panel III guidelines [25]. Information on prescription medication use during the two preceding weeks was collected directly from the medication containers at annual clinic visits in the first 10 years and by semi-annual telephone contact thereafter. Fasting glucose was categorized into a dichotomous variable (<100 mg/dL, ≥100 mg/dL) because a value of ≥100 mg/dL is used to diagnose impaired fasting glucose, which increases an individual's risk of progressing to T2D [26].

2.4. Outcome assessment

The two outcome variables for this analysis are incident MI and incident CAD. In the CHS, incident MI is defined as fatal and non-fatal MI, and incident CAD is a broader variable that includes MI (fatal and non-fatal), angina, coronary artery bypass graft (CABG), angioplasty, and/or death due to atherosclerotic CAD [27,28]. Self-report of cardiovascular disease was validated according to standardized criteria by ascertaining medications used and by relevant standardized examinations performed on all participants [29]. Incident MI and CAD events were adjudicated by the Cardiac Subgroup of the CHS Events Subcommittee based on the satisfaction of CHS appropriate algorithms for classification [28]. Confirmation of deaths was conducted through reviews of obituaries, medical records, death certificates, in addition to interviews of contacts and proxies, resulting in a 100% complete follow-up ascertainment of mortality status [30]. Time-to-event was defined as the time in days from the date of enrollment to the earliest event (date of diagnosis of CAD or MI).

2.5. Statistical analysis

Baseline demographic characteristics were presented as mean (standard deviation) for continuous variables and as frequency (percent) for categorical variables. Continuous variables were compared across groups for breakfast frequency and breakfast timing using the Student's t-test one-way ANOVA, Wilcoxon rank-sum test or Kruskal-Wallis test for continuous variables, and Chi-square or Fisher's exact test for categorical variables.

To assess associations with the exposures of breakfast frequency and breakfast timing, the cause-specific hazards of incident MI and CAD were modelled using a series of multivariable-adjusted Cox proportional hazard regression models, producing multivariable-adjusted hazard ratios (HRs) along with their 95% confidence intervals (CIs). The proportional hazards assumption was assessed using the Kolmogorov-type supremum test [31]. The first multivariable model adjusted for age and sex. All models were adjusted for baseline covariables. The second multivariable model further adjusted for marital status (married, not married), largest racial group in the CHS (no, yes), education (≤high school graduation, >high school graduation), and annual household income (<$25,000, $25,000–$49,999, ≥$50,000). The third multivariable model further adjusted for smoking status (never, former, current), alcohol (number of alcoholic beverages/week; 0, <7, 7−14, >14), exercise (none/low, moderate/high), and Centre for Epidemiological Studies-Depression Scale score (high: ≥10, low: <10) [32,33]. The fourth multivariable model (main model) further adjusted for the following dietary factors: fruit and vegetable intake (<3, ≥3 servings/day), meal frequency (<3, ≥3 meals/day), and snacks after dinner (yes, no); when using this fourth model, the analysis with breakfast frequency as the main exposure was also adjusted for break-fast timing (before 07:00, 07:00−09:00, after 09:00), and the analysis with break-fast timing as the main exposure was also adjusted for breakfast frequency (not daily, daily). Subgroup analyses and a test of interaction were conducted for each of sex, BMI, income, and baseline T2D, impaired fasting glucose, fasting insulin (<10 IU/mL, ≥10 IU/mL), hypercholesterolemia, hypertension, and follow-up time (using the median follow-up time as a cut-point).

In a sensitivity analysis, we further adjusted for BMI, hypercholesterolemia, fasting glucose, and fasting insulin. These could potentially be mediators and should not be in the final model, but we wanted to see if /how including them in the model affected the results, in order to compare to other similar studies that have included these adjustments and to generate hypotheses regarding the possibility of these variables as confounders or mediators [34,35]. In a second sensitivity analysis, events within the study’s first two years were excluded to help reduce the possibility of reverse causation from underlying illness affecting breakfast habits at baseline. In a third sensitivity analysis, breakfast frequency was assessed using an alternate definition of irregular breakfast consumer (rarely, weekends only, never) or regular breakfast consumer (everyday, some days).

We did not use a competing risk model such as a Fine-Gray subdistribution model because the research questions had an etiological nature for which Cox models are better suited [36,37]. However, death was still dealt with in our cause-specific hazards model; follow-up time was calculated from the date of the baseline visit to the date of first diagnosis of MI, CAD, loss to follow-up, or death, whichever came first. Additionally, we conducted an analysis for non-cardiovascular mortality for contextualization. As there was less than 1% of missing data for almost all variables, complete case analysis was performed. A two-sided p-value of < 0.05 was the threshold for statistical significance. All analyses were performed using SAS statistical software version 9.4 (SAS Institute Inc., Cary, N.C, USA).

3. Results

In this sample of 4070 older American adults, 1617 CAD cases were documented (795 MI cases) during a median follow-up of 14.2 years (maximum follow-up was 26 years). Most participants reported eating breakfast daily (85.3%) and ‘breaking-fast’ between 07:00 and 09:00am (72.6%) (Table 1). At baseline, people who did not consume breakfast daily were younger than those who did consume breakfast daily, were more likely to be female, not in the study’s major ethnic group (white), unmarried, smokers, less educated, have lower annual household income, exercise less intensely, report a higher number of depressive symptoms, have greater fasting glucose, fasting insulin, BMI, and waist circumference, and hypercholesterolemia, and eat less fruits and vegetables. Except for age and hypercholesterolemia, breakfast timing (breaking fast later than 09:00) was likewise associated with the same baseline variables that are generally indicators of lower levels of privilege and increased risk of heart disease.

Table 1.

Characteristics of the Cardiovascular Health Study (CHS) participants at baseline by breakfast frequency and timing.

Characteristics	Breakfast frequency			Break-fast timing (first eating occasion of the day)f
	Daily (7 times/wk) (n = 3473)	Not-daily (0−6 times/wk) (n = 597)	P-value	Before 07:00 (n = 678)	07:00−09:00 (n = 2955)	After 09:00 (n = 437)	P-value
Age, years	72.8 (5.55)	70.9 (4.60)	<.0001a	72.3 (5.59)	72.5 (5.39)	72.8 (5.75)	0.26a
Sex			0.007b				<.0001b
Female	2048 (59.0%)	387 (64.8%)		354 (52.2%)	1821 (61.6%)	260 (59.5%)
Male	1425 (41.0%)	210 (35.2%)		324 (47.8%)	1134 (38.4%)	177 (40.5%)
Largest racial group in the CHSe			<.0001c				<.0001c
No	471 (13.6%)	172 (28.8%)		94 (13.9%)	379 (12.8%)	170 (38.9%)
Yes	3002 (86.4%)	425 (71.2%)		584 (86.1%)	2576 (87.2%)	267 (61.1%)
Education			0.0004b				<.0001b
≤ High school	1912 (55.1%)	375 (62.8%)		412 (60.8%)	1600 (54.1%)	275 (62.9%)
>High school	1561 (44.9%)	222 (37.2%)		266 (39.2%)	1355 (45.9%)	162 (37.1%)
Household income U.S.$/yr			<.0001b				<.0001b
<25,000	2038 (58.7%)	406 (68.0%)		432 (63.7%)	1703 (57.6%)	309 (70.7%)
25,000-49,999	947 (27.3%)	123 (20.6%)		164 (24.2%)	815 (27.6%)	91 (20.8%)
≥50,000	488 (14.1%)	68 (11.4%)		82 (12.1%)	437 (14.8%)	37 (8.5%)
Married	2327 (67.0%)	350 (58.6%)	<.0001b	426 (62.8%)	2009 (68.0%)	242 (55.4%)	<.0001b
Smoking status			<.0001b				0.003b
Never	1688 (48.6%)	237 (39.7%)		323 (47.6%)	1410 (47.7%)	192 (43.9%)
Former	1409 (40.6%)	223 (37.4%)		258 (38.1%)	1206 (40.8%)	168 (38.4%)
Current	376 (10.8%)	137 (22.9%)		97 (14.3%)	339 (11.5%)	77 (17.6%)
Alcohol use, drinks/wk			0.28b				0.11b
None	1737 (50.0%)	280 (46.9%)		347 (51.2%)	1435 (48.6%)	235 (53.8%)
<7	1248 (35.9%)	228 (38.2%)		221 (32.6%)	1104 (37.4%)	151 (34.6%)
7-14	268 (7.7%)	42 (7.0%)		59 (8.7%)	224 (7.6%)	27 (6.2%)
>14	220 (6.3%)	47 (7.9%)		51 (7.5%)	192 (6.5%)	24 (5.5%)
Exercise			0.04b				0.0001b
No exercise	264 (7.6%)	57 (9.5%)		57 (8.4%)	206 (7.0%)	58 (13.3%)
Low	1653 (47.6%)	292 (48.9%)		338 (49.9%)	1401 (47.4%)	206 (47.1%)
Moderate	1184 (34.1%)	204 (34.2%)		218 (32.2%)	1027 (34.8%)	143 (32.7%)
High	372 (10.7%)	44 (7.4%)		65 (9.6%)	321 (10.9%)	30 (6.9%)
Hypertension	1449 (41.7%)	257 (43.0%)	0.54b	271 (40.0%)	1248 (42.2%)	187 (42.8%)	0.52b
Hypercholesterolemia	714 (20.6%)	147 (24.6%)	0.02b	126 (18.6%)	638 (21.6%)	97 (22.2%)	0.19b
Fasting glucose, mg/dL	100.0 (94.0, 111.0)	102.0 (95.0, 113.0)	0.02d	100.0 (94.0, 109.0)	100.0 (94.0, 111.0)	102.6 (95.0, 114.0)	0.04d
Fasting insulin, IU/mL	13.0 (10.0, 17.4)	13.4 (10.0, 19.4)	0.0002d	12.0 (9.0, 17.0)	13.0 (10.0, 18.0)	14.0 (10.0, 20.0)	<.0001d
Baseline T2D	489 (14.1%)	79 (13.2%)	0.58b	92 (13.6%)	409 (13.8%)	67 (15.3%)	0.67b
BMI, kg/m2	26.5 (4.53)	27.8 (5.42)	<.0001a	26.6 (4.68)	26.5 (4.57)	27.6 (5.41)	<.0001a
Waist circumference, cm	93.6 (12.93)	96.6 (14.44)	<.0001a	94.5 (13.54)	93.6 (13.01)	96.4 (13.72)
CES-D score			0.03b				0.04b
Low (<10)	3073 (88.5%)	509 (85.3%)		611 (90.1%)	2599 (88.0%)	372 (85.1%)
High (≥10)	400 (11.5%)	88 (14.7%)		67 (9.9%)	356 (12.0%)	65 (14.9%)
Fruits & Vegetables, servings/d			<.0001b				<.0001b
<3	1840 (53.0%)	421 (70.5%)		372 (54.9%)	1579 (53.4%)	310 (70.9%)
≥3	1633 (47.0%)	176 (29.5%)		306 (45.1%)	1376 (46.6%)	127 (29.1%)
Breakfast timing			<.0001b
Before 07:00	586 (16.9%)	92 (15.4%)
07:00-09:00	2614 (75.3%)	341 (57.1%)
After 09:00	273 (7.9%)	164 (27.5%)
Meal frequency, no./d	2.8 (0.49)	2.1 (0.64)	<.0001a	2.7 (0.62)	2.7 (0.53)	2.3 (0.64)	<.0001a
Snacks after dinner	2123 (61.1%)	387 (64.8%)	0.09b	367 (54.1%)	1841 (62.3%)	302 (69.1%)	<.0001b

Values are n (%) for categorical variables and mean (standard deviation) for continuous variables (none were skewed).

Abbreviations: BMI, body mass index; CES-D, Centre for Epidemiological Studies-Depression Scale; CVD, cardiovascular disease; T2D, type 2 diabetes.

^aUnequal variance two sample t-test.

^bChi-Square p-value.

^cFisher Exact p-value.

^dWilcoxon rank sum p-value.

^eThe largest racial group in our sample of CHS participants was white.

^fParticipants were asked “On Monday through Friday about what time do you usually first eat or drink something after waking up?”.

Compared to eating breakfast daily, not eating breakfast daily was not significantly associated with risk of MI or CAD, in the base model adjusted for only age and sex (HR = 1.04, 95% CI: 0.85–1.27 for MI and 1.04, 0.90–1.20 for CAD) or when progressively adjusted for demographic factors (1.01, 0.82–1.24 for MI and 1.00, 0.87–1.16 for CAD), lifestyle and mental health factors (1.01, 0.82–1.24 for MI and 1.00, 0.87–1.16 for CAD), and dietary factors (1.01, 0.81–1.28 for MI and 0.99, 0.85–1.17 for CAD) (Table 2). These results were not materially altered in sensitivity analyses in which the model was additionally adjusted for BMI, hypercholesterolemia, fasting glucose, and fasting insulin, in sensitivity analyses in which early cases were excluded, or in which breakfast frequency was differently dichotomized as either eating breakfast rarely, weekends only, or never (n = 302) or everyday or some days (n = 3768) did not alter the results (Supplementary Table 1). Further adjustment for baseline calories, saturated fat, protein, and fibre intake among participants who had these nutrient data did not alter the results (data not shown); however, participants without these nutrient data started the study later and were disproportionately Black compared to the participants who had these nutrient data, which means that selection bias may be present in this further adjustment, and also means that this further adjustment is less generalizable to the Black CHS participants.

Table 2.

Breakfast frequency and multivariable-adjusted hazard ratios of myocardial infarction (MI) and coronary heart disease (CAD) with 95% confidence intervals.

	Breakfast Frequency
	Not-daily (0−6 days/week) (n = 597)	Daily (7 days/week) (n = 3473)	P-value
MI cases (n)	111	684
Person-time (years, %)	8,139.14 (14.98)	461,93.43 (85.02)
Adjusted for age and sex: HR (95% CI)	1.04 (0.85, 1.27)	1 (ref)	0.72
+ Demographic factorsa	1.01 (0.82, 1.24)	1 (ref)	0.92
+ Lifestyle and mental health factorsb	1.01 (0.82, 1.24)	1 (ref)	0.92
+ Dietary factorsc(final model)	1.01 (0.82, 1.28)	1 (ref)	0.91
Sensitivity analysis: Additional adjustmentd	1.02 (0.81, 1.29)	1 (ref)	0.90
Sensitivity analysis: Excluding early eventse
Cases (n)	106	609
Person-time (years, %)	8110.3 (14.98)	46036.08 (85.02)
HRc (95% CI)	1.08 (0.85, 1.37)	1 (ref)	0.54
CAD cases (n)	228	1389
Person-time (years, %)	7,478.38 (14.99)	42,424.34 (85.01)
Adjusted for age and sex: HR (95% CI)	1.01 (0.90, 1.20)	1 (ref)	0.58
+ Demographic factorsa	1.00 (0.87, 1.16)	1 (ref)	0.95
+ Lifestyle and mental health factorsb	1.00 (0.87, 1.16)	1 (ref)	0.99
+ Dietary factorsc(final model)	0.99 (0.85, 1.17)	1 (ref)	0.94
Sensitivity analysis: Additional adjustmentd	0.98 (0.84, 1.15)	1 (ref)	0.81
Sensitivity analysis: Excluding early eventse
Cases (n)	210	1225
Person-time (years, %)	7437.25 (14.99)	42188.64 (85.01)
HRc (95% CI)	1.04 (0.88, 1.22)	1 (ref)	0.68

Abbreviations: BMI, Body Mass Index; CI, confidence interval; HR, hazard ratio.

Data are from a multivariable-adjusted analysis using Cox proportional hazard regression.

^aIn addition to age (years) and sex (female, male), this model is adjusted for demographic factors: marital status (married, not married), largest racial group in the CHS (no, yes), education (≤high school graduation, >high school graduation), and annual household income (<$25,000, $25,000-$49,999, ≥$50,000).

^bIn addition to age, sex, and demographic factors, this model is further adjusted for lifestyle and mental health factors: smoking status (never, former, current), alcohol (number of alcoholic beverages/week; 0, <7, 7−14, >14), exercise (none/low, moderate/high), and Centre for Epidemiological Studies-Depression Scale score (high: ≥10, low: <10).

^cIn addition to age, sex, demographic, lifestyle, and mental health factors, this model is further adjusted for dietary factors: fruit and vegetable intake (<3, ≥3 servings/day), breakfast timing (before 07:00, 07:00−09:00, after 09:00), meal frequency (<3, ≥3 meals/day), and snacks after dinner (yes, no). This model (bolded) is the main model reported for this analysis.

^dIn addition to age, sex, demographic, lifestyle, mental health, and dietary factors, this model is further adjusted for BMI (kg/m²) fasting insulin, fasting glucose, and hypercholesterolemia.

^eEvents within the study’s first two years were excluded to help reduce the possibility of reverse causation from underlying illness affecting breakfast habits at baseline. In this analysis, participants who had early events did not contribute any follow-up. Participants who were event-free at 2 years contributed person-time from year 2 onward.

When stratified into subgroups, the HR for risk of MI from not eating breakfast daily (compared to eating breakfast daily) in participants with a BMI ≥ 30 kg/m² was 0.66 (0.43–1.02, p = 0.06) while it was 1.17 (0.91–1.15) in participants with a BMI < 30 kg/m² in a significant interaction effect (p, interaction = 0.02) (Table 3). While not a statistically significant, the HRs for risk of MI from not eating breakfast daily were in different directions in males (0.89, 0.63–1.26) than in females (1.11, 0.84–1.47) (p, interaction = 0.30) and in people with hypertension (0.82, 0.59–1.16) and without hypertension (1.19, 0.90–1.58) (p, interaction = 0.08). However, all the corresponding findings were less pronounced for the less specific outcome of CAD.

Table 3.

Breakfast frequency and multivariable-adjusted hazard ratios of myocardial infarction (MI) and coronary heart disease (CAD) with 95% confidence intervals stratified by risk factors.

	Number of cases	Person-time (%)	Daily breakfast (7 days/wk, n = 3473)	Not-daily breakfast (0−6 days/wk, n = 697)	P-value	Interaction (P-value)
MI
Sex						0.30
Female	407	35,459.41 (65.26)	1 (ref)	1.11 (0.84, 1.47)	0.47
Male	388	18,873.16 (34.74)	1 (ref)	0.89 (0.63, 1.26)	0.51
Income						0.70
<$25 K/year	480	30,442.50 (56.03)	1 (ref)	0.99 (0.76, 1.29)	0.93
≥$25 K/year	315	23,890.07 (43.97)	1 (ref)	1.07 (0.74, 1.55)	0.70
Baseline BMI, kg/m2						0.02
<30	627	43,591.73 (80.23)	1 (ref)	1.17 (0.91, 1.51)	0.22
≥30	168	10,740.84 (19.77)	1 (ref)	0.66 (0.43, 1.02)	0.06
Baseline T2D						0.70
No	652	48,413.06 (89.11)	1 (ref)	1.00 (0.78, 1.29)	0.98
Yes	143	5,919.51 (10.89)	1 (ref)	1.12 (0.68, 1.82)	0.66
Impaired Fasting Glucose						0.90
No	338	26,648.65 (49.05)	1 (ref)	1.02 (0.73, 1.44)	0.91
Yes	457	27,683.92 (50.95)	1 (ref)	0.99 (0.75, 1.32)	0.97
Fasting Insulin						0.78
<10 IU/mL	176	13,474.11 (24.8)	1 (ref)	1.07 (0.68, 1.69)	0.77
≥10 IU/mL	619	40,858.46 (75.2)	1 (ref)	1.00 (0.78, 1.29)	1.00
Hypercholesterolemia						0.74
No	626	42,271.8 (77.8)	1 (ref)	0.99 (0.76, 1.29)	0.94
Yes	169	12,060.77 (22.2)	1 (ref)	1.07 (0.71, 1.60)	0.74
Hypertension						0.08
No	433	33,352.25 (61.39)	1 (ref)	1.19 (0.90, 1.58)	0.23
Yes	362	20,980.32 (38.61)	1 (ref)	0.82 (0.59, 1.16)	0.26
Follow-up time						0.21
<14.2 yearsa	574	43609.73 (80.26)	1 (ref)	0.93 (0.71, 1.22)	0.6
≥14.2 yrsa	221	10722.84 (19.74)	1 (ref)	1.23 (0.85, 1.77)	0.27
CAD
Sex
Female	896	32,791.68 (65.71)	1 (ref)	1.05 (0.86, 1.27)	0.66	0.38
Male	721	17,111.04 (34.29)	1 (ref)	0.92 (0.72, 1.17)	0.48
Income						0.80
<$25 K/year	992	28,023.77 (56.16)	1 (ref)	0.98 (0.82, 1.18)	0.85
≥$25 K/year	625	21,878.95 (43.84)	1 (ref)	1.02 (0.79, 1.33)	0.87
Baseline BMI, kg/m2						0.47
<30	1264	40,096.94 (80.35)	1 (ref)	0.99 (0.82, 1.19)	0.87
≥30	353	9,805.79 (19.65)	1 (ref)	0.88 (0.67, 1.15)	0.35
Baseline T2D						0.34
No	1338	44,629.56 (89.43)	1 (ref)	0.94 (0.79, 1.12)	0.51
Yes	279	5,273.16 (10.57)	1 (ref)	1.13 (0.80, 1.59)	0.49
Impaired Fasting Glucose						0.25
No	689	24,643.07 (49.38)	1 (ref)	1.05 (0.83, 1.32)	0.71
Yes	928	25,259.66 (50.62)	1 (ref)	0.89 (0.73, 1.08)	0.22
Fasting Insulin						0.52
<10 IU/mL	351	12,486.92 (25.02)	1 (ref)	0.87 (0.63, 1.22)	0.43
≥10 IU/mL	1266	37,415.80 (74.98)	1 (ref)	0.98 (0.83, 1.17)	0.85
Hypercholesterolemia						0.65
No	1259	39,068.87 (78.29)	1 (ref)	0.98 (0.82, 1.17)	0.80
Yes	358	10,833.85 (21.71)	1 (ref)	0.91 (0.68, 1.21)	0.50
Hypertension						0.83
No	847	30,928.01 (61.98)	1 (ref)	0.95 (0.77, 1.17)	0.64
Yes	770	18,974.72 (38.02)	1 (ref)	0.98 (0.79, 1.22)	0.87
Follow-up time						0.13
<14.2 yearsa	1256	40685.05 (81.53)	1 (ref)	0.93 (0.78, 1.12)	0.45
≥14.2 yrsa	361	9217.67 (18.47)	1 (ref)	1.2 (0.9, 1.59)	0.21

Abbreviations: BMI, Body Mass Index; CI, confidence interval; HR, hazard ratio; T2D, type 2 diabetes.

Data are from a multivariable-adjusted analysis using Cox proportional hazard regression.

P-values for interaction terms between breakfast frequency and each risk factor were not significant.

Models are adjusted for age (years), sex (female, male, except when stratified by sex), marital status (married, not married), largest racial group in the CHS (no, yes), education (≤high school graduation, >high school graduation), annual household income (<$25,000, $25,000-$49,999, ≥$50,000), smoking status (never, former, current), alcohol (number of alcoholic beverages/week; 0, <7, 7−14, >14), exercise (none/low, moderate/high), Centre for Epidemiological Studies-Depression Scale score (high: ≥10, low: <10), fruit and vegetable intake (<3, ≥3 servings/day), breakfast timing (before 07:00, 07:00−09:00, after 09:00), snacks after dinner (yes, no), and meal frequency (<3, ≥3 meals/day).

^aThe median follow-up time of 14.2 years was used as the cut-point.

The HR for risk of MI and CAD associated with breakfast timing were all null and minimally altered by multivariable adjustment or the exclusion of early cases (Table 4). A sensitivity analysis in which participants who reported never eating breakfast were removed from the breakfast timing analysis did not alter results (data not shown). In stratified analyses, compared with participants whose breakfast time was 07:00–09:00am, those who broke fast before 07:00am had an HR for CAD of 1.40 (1.02, 1.93) if they had T2D and of 1.19 (1.03, 1.38) if they had high fasting insulin at baseline (Table 5). Compared with participants whose breakfast time was 07:00–09:00am, those who broke fast after 09:00am had an HR for CAD of 0.95 (0.78, 1.15) in the first follow-up period of the study and an HR for CAD of 1.45 (1.04, 2.01) in the second follow-up period.

Table 4.

Breakfast timing and multivariable-adjusted hazard ratios of myocardial infarction (MI) and coronary heart disease (CAD) with 95% confidence intervals.

	Break-fast Timing (first eating occasion of the day)
	Before 07:00	P-value	07:00−09:00	After 09:00	P-value
	(n = 678)		(n = 2955)	(n = 437)
MI cases (n)	139		579	77
Person-time (years, %)	8650.57 (15.92)		40213.30 (74.01)	5468.7 (10.07)
Adjusted for age and sex: HR (95% CI)	1.08 (0.90, 1.30)	0.41	1 (ref)	0.99 (0.78, 1.26)	0.94
+ Demographic factorsa	1.04(0.87, 1.26)	0.67	1 (ref)	0.97 (0.76, 1.23)	0.78
+ Lifestyle and mental health factorsb	1.05(0.87, 1.27)	0.60	1 (ref)	0.96 (0.75, 1.22)	0.73
+ Dietary factorsc(final model)	1.05(0.87, 1.27)	0.60	1 (ref)	0.95 (0.74, 1.22)	0.67
Sensitivity analysis: Additional adjustmentd	1.06 (0.88, 1.28)	0.52	1 (ref)	0.96 (0.74, 1.23)	0.72
Sensitivity analysis: Excluding early eventse
Cases (n)	129		519	67
Person-time (years, %)	8625.59 (15.93)		40079.46 (74.02)	5441.34 (10.05)
HRc (95% CI)	1.10(0.90, 1.34)	0.35	1 (ref)	0.91 (0.70, 1.19)	0.50
CAD cases (n)	279		1166	172
Person-time (years, %)	7981.35 (15.99)		36906.56 (73.96)	5014.82 (10.05)
Adjusted for age and sex: HR (95% CI)	1.08 (0.94, 1.23)	0.27	1 (ref)	1.09 (0.93, 1.28)	0.29
+ Demographic factorsa	1.05 (0.92, 1.19)	0.50	1 (ref)	1.05 (0.89, 1.24)	0.54
+ Lifestyle and mental health factorsb	1.06 (0.92, 1.20)	0.43	1 (ref)	1.01 (0.88, 1.23)	0.64
+ Dietary factorsc(final model)	1.06 (0.93, 1.21)	0.42	1 (ref)	1.04 (0.88, 1.23)	0.65
Sensitivity analysis: Additional adjustmentd	1.08 (0.94, 1.23)	0.29	1 (ref)	1.05 (0.88, 1.24)	0.59
Sensitivity analysis: Excluding early eventse
Cases (n)	250		1038	147
Person-time (years, %)	7941.99 (16)		36709.81 (73.97)	4974.1 (10.02)
HRc(95% CI)	1.07 (0.93, 1.23)	0.35	1 (ref)	1.00 (0.83, 1.20)	0.99

Abbreviations: BMI, Body Mass Index; CI, confidence interval; HR, hazard ratio; T2D, type 2 diabetes.

Data are from a multivariable-adjusted analysis using Cox proportional hazard regression. Participants were asked, “On Monday through Friday, about what time do you usually first eat or drink something after waking up?”.

^aIn addition to age (years) and sex (female, male), this model is adjusted for demographic factors: marital status (married, not married), largest racial group in the CHS (no, yes), education (≤high school graduation, >high school graduation), and annual household income (<$25,000, $25,000-$49,999, ≥$50,000).

^bIn addition to age, sex, and demographic factors, this model is further adjusted for lifestyle and mental health factors: smoking status (never, former, current), alcohol (number of alcoholic beverages/week; 0, <7, 7−14, >14), exercise (none/low, moderate/high), and Centre for Epidemiological Studies-Depression Scale score (high: ≥10, low: <10).

^cIn addition to age, sex, demographic, lifestyle, and mental health factors, this model is further adjusted for dietary factors: fruit and vegetable intake (<3, ≥3 servings/day), daily breakfast consumption (yes, no), meal frequency (<3, ≥3 meals/day), and snacks after dinner (yes, no). This model (bolded) is the main model reported for this analysis. In an additional sensitivity analysis, the exclusion of participants who reported that they never eat breakfast (n = 122) did not materially alter any results.

^dIn addition to age, sex, demographic, lifestyle, mental health, and dietary factors, this model is further adjusted for BMI (kg/m²) fasting insulin, fasting glucose, and hypercholesterolemia.

^eEvents within the study’s first two years were excluded to help reduce the possibility of reverse causation from underlying illness affecting breakfast habits at baseline. In this analysis, participants who had early events did not contribute any follow-up. Participants who were event-free at 2 years contributed person-time from year 2 onward.

Table 5.

Break-fast timing and multivariable-adjusted hazard ratios (HR) of myocardial infarction (MI) and coronary heart disease (CAD) with 95% confidence intervals (CI)stratified by risk factors.

	Number of cases	Person time (%)	Break-fast Timing (first eating occasion of the day)a
			Before 07:00 (n = 609)		07:00−09:00 (n = 2733)	After 09:00 (n = 402)		Interaction (P-value)
			HR (95%CI)	P-value	HR (95%CI)	HR (95%CI)	P-value
MI
Sex								0.93
Female	407	35459.41 (65.26)	1.02 (0.77, 1.35)	0.90	1 (ref)	0.92 (0.65, 1.30)	0.62
Male	388	18873.16 (34.74)	1.08 (0.84, 1.39)	0.55	1 (ref)	0.98 (0.69, 1.39)	0.90
Income								0.42
<$25 K/year	480	30442.50 (56.03)	1.05 (0.83, 1.33)	0.68	1 (ref)	0.85 (0.62, 1.16)	0.30
≥$25 K/year	315	23890.07 (43.97)	1.04 (0.77, 1.42)	0.78	1 (ref)	1.18 (0.79, 1.77)	0.42
Baseline BMI, kg/m2								0.48
<30	627	43591.73 (80.23)	0.99 (0.80, 1.23)	0.94	1 (ref)	0.96 (0.72, 1.28)	0.76
≥30	168	10740.84 (19.77)	1.29 (0.88, 1.89)	0.20	1 (ref)	0.93 (0.57, 1.51)	0.76
Baseline T2D								0.16
No	652	48413.06 (89.11)	1.15 (0.94, 1.41)	0.17	1 (ref)	0.93 (0.70, 1.24)	0.62
Yes	143	5919.51 (10.89)	0.70 (0.41, 1.20)	0.19	1 (ref)	1.12 (0.67, 1.89)	0.67
Impaired Fasting Glucose								0.23
No	338	26648.65 (49.05)	1.26 (0.96, 1.65)	0.09	1 (ref)	0.99 (0.67, 1.46)	0.97
Yes	457	27683.92 (50.95)	0.91 (0.70, 1.18)	0.49	1 (ref)	0.91 (0.66, 1.25)	0.55
Fasting Insulin								0.54
<10 IU/mL	176	13474.11 (24.80)	0.92 (0.62, 1.36)	0.66	1 (ref)	1.10 (0.65, 1.86)	0.74
≥10 IU/mL	619	40858.46 (75.20)	1.11 (0.89, 1.37)	0.35	1 (ref)	0.91 (0.69, 1.21)	0.53
Hypercholesterolemia								0.05
No	626	42271.8 (77.80)	1.02 (0.82, 1.25)	0.89	1 (ref)	0.79 (0.58, 1.07)	0.13
Yes	169	12060.77 (22.20)	1.22 (0.80, 1.87)	0.36	1 (ref)	1.50 (0.97, 2.30)	0.07
Hypertension								0.33
No	433	33352.25 (61.39)	1.20 (0.94, 1.53)	0.15	1 (ref)	1.02 (0.74, 1.42)	0.89
Yes	362	20980.32 (38.61)	0.90 (0.67, 1.21)	0.49	1 (ref)	0.91 (0.62, 1.33)	0.62
Follow-up time								0.66
<14.2 yearsb	574	43609.73 (80.26)	1.04 (0.83, 1.29)	0.75	1 (ref)	0.89 (0.66, 1.19)	0.42
≥14.2 yrsb	221	10722.84 (19.74)	1.09 (0.76, 1.57)	0.63	1 (ref)	1.13 (0.72, 1.79)	0.59
CAD
Sex								0.95
Female	896	32791.68 (65.71)	1.07 (0.89, 1.29)	0.49	1 (ref)	1.02 (0.82, 1.28)	0.85
Male	721	17111.04 (34.29)	1.05 (0.87, 1.26)	0.63	1 (ref)	1.06 (0.83, 1.36)	0.62
Income								0.26
<$25 K/year	992	28023.77 (56.16)	1.06 (0.90, 1.24)	0.51	1 (ref)	0.95 (0.73, 1.17)	0.63
≥$25 K/year	625	21878.95 (43.84)	1.05 (0.84, 1.31)	0.68	1 (ref)	1.26 (0.95, 1.66)	0.11
Baseline BMI, kg/m2								0.59
<30	1264	40096.94 (80.35)	1.09 (0.94, 1.26)	0.28	1 (ref)	1.03 (0.84, 1.25)	0.79
≥30	353	9805.79 (19.65)	1.27 (0.96, 1.67)	0.09	1 (ref)	1.13 (0.84, 1.54)	0.42
Baseline T2D								0.37
No	1338	44629.56 (89.43)	1.10 (0.95, 1.27)	0.19	1 (ref)	1.05 (0.87, 1.26)	0.62
Yes	279	5273.16 (10.57)	1.40 (1.02, 1.93)	0.04	1 (ref)	1.18 (0.81, 1.72)	0.38
Impaired fasting glucose								0.45
No	689	24643.07 (49.38)	1.19 (0.97, 1.44)	0.09	1 (ref)	0.95 (0.72, 1.25)	0.71
Yes	928	25259.66 (50.62)	1.08 (0.90, 1.29)	0.40	1 (ref)	1.12 (0.90, 1.38)	0.31
Fasting insulin								0.40
<10 IU/mL	351	12486.92 (25.02)	0.96 (0.73, 1.27)	0.79	1 (ref)	0.98 (0.67, 1.44)	0.91
≥10 IU/mL	1266	37415.8 (74.98)	1.19 (1.03, 1.38)	0.02	1 (ref)	1.07 (0.89, 1.29)	0.46
Hypercholesterolemia								0.35
No	1259	39068.87 (78.29)	1.09 (0.94, 1.26)	0.26	1 (ref)	0.99 (0.82, 1.21)	0.95
Yes	358	10833.85 (21.71)	1.28 (0.96, 1.71)	0.10	1 (ref)	1.25 (0.91, 1.71)	0.17
Hypertension								0.66
No	847	30928.01 (61.98)	1.20 (1.00, 1.43)	0.05	1 (ref)	1.12 (0.90, 1.41)	0.32
Yes	770	18974.72 (38.02)	1.07 (0.88, 1.30)	0.52	1 (ref)	1.04 (0.82, 1.33)	0.73
Follow-up time								0.09
<14.2 yearsb	1256	40685.05 (81.53)	1.03 (0.89, 1.20)	0.7	1 (ref)	0.95 (0.78, 1.15)	0.59
≥14.2 yrsb	361	9217.67 (18.47)	1.15 (0.87, 1.53)	0.31	1 (ref)	1.45 (1.04, 2.01)	0.03

Abbreviations: BMI, Body Mass Index; CI, confidence interval; HR, hazard ratio; T2D, type 2 diabetes.

Data are from a multivariable-adjusted analysis using Cox proportional hazard regression.

P-values for interaction terms between breakfast timing and each risk factor were not significant (All interaction p-values > 0.48).

Models are adjusted for age (years), sex (female, male, except when stratified by sex), marital status (married, not married), largest racial group in the CHS (no, yes), education (≤high school graduation, >high school graduation), annual household income (<$25,000, $25,000-$49,999, ≥$50,000), smoking status (never, former, current), alcohol (number of alcoholic beverages/week; 0, <7, 7−14, >14), exercise (none/low, moderate/high), Centre for Epidemiological Studies-Depression Scale score (high: ≥10, low: <10), fruit and vegetable intake (<3, ≥3 servings/day), daily breakfast consumption (yes, no), snacks after dinner (yes, no), and meal frequency (<3, ≥3 meals/day).

^aParticipants were asked, “On Monday through Friday, about what time do you usually first eat or drink something after waking up?” In a sensitivity analysis, the exclusion of participants who reported that they never eat breakfast (n = 122) did not materially alter any results.

^bThe median follow-up time of 14.2 years was used as the cut-point.

Compared to eating breakfast daily, not eating breakfast daily was not significantly associated with risk of non-cardiovascular mortality prior to MI (1.08, 0.95–1.21) or prior to CAD (1.10, 0.96–1.27). Compared with participants whose breakfast time was 07:00–09:00am, those who broke fast before 07:00am had an HR for non-cardiovascular mortality prior to MI of 1.07 (0.97, 1.18), and those who broke fast after 09:00am had an HR for non-cardiovascular mortality prior to MI of 1.10 (0.97, 1.25) (Supplementary Table 2). The risk of non-cardiovascular mortality prior to CAD associated with breakfast timing was similarly null (Supplementary Table 3).

4. Discussion

4.1. Comparison to literature and potential mechanistic explanations

To our knowledge, the present study was the first to prospectively assess the relationship between breakfast timing and/or frequency and MI or CAD risk specifically in older adults. In younger adults, a 2020 meta-analysis of seven cohort studies (total of 221,732 participants) concluded that skipping breakfast was associated with a 22% elevated risk of cardiovascular disease (1.22, 1.10–1.35) compared with eating breakfast regularly [9]. The other existing meta-analysis was also in younger adults and purported a pooled 24% increased risk (1.24, 1.09–1.40), and included two case-control studies and a cross-sectional study [10]. Interestingly, this cross-sectional study, while described as aged ≥40 years, had a mean age of 71.4 years, and in this study, skipping breakfast (defined as at least three times in a week) was not associated with risk of CAD (1.06, 0.91–1.24), a finding in line with the present study of older adults [38]. However, cross-sectional studies of meal frequency in relation to heart disease have no temporal element, and the relationship could be bi-directional or reversed, especially because having incident heart disease such as MI is likely to change a person’s eating habits, including their eating frequency. Having heart disease can also bias their recall (response bias, social desirability bias) [39]. However, the only prospective study included in the meta-analyses that stratified by age (the Health Professionals Follow-up Study) also reported no association with CAD among the strata of participants aged over 60 years (1.06, 0.84–1.33), whereas in those ≤60 years of age, skipping breakfast was associated with a 55% significantly higher risk of CAD compared to eating breakfast (1.55, 1.09–2.22) (p, interaction = 0.01) [34]. Taken together, the existing evidence and our study findings align to suggest limited evidence for an association between breakfast frequency and risk of CAD in older adults, but research to date on this topic is still too limited to know with certainty. While we did observe a statistically significant risk of CAD in the second follow-up period among participants who ate breakfast after 9:00am (compared to those who ate breakfast from 7:00−9:00am), breakfast timing was assessed at baseline, and so the potential for misclassification bias is greater in the second half of follow-up.

Potentially, associations between skipping breakfast and risk of CAD and MI are only present in younger adults. This could be due to 1) early age being a stronger time window of opportunity for the effect of eating timing and frequency on cardiometabolic health, or 2) younger adults having higher biological demands for nutrients or higher work or activity loads such that missing the morning meal causes more strain in younger adults than in older adults. However, the null findings of the present study could also be due to methodological challenges inherent to studying older adults. For example, there could be survivor bias in the older adults, with those who skipped meals being excluded from the present analysis because they already had prevalent CAD at baseline.

A CHS analysis of breakfast frequency and timing with T2D as the outcome observed no association between breakfast frequency and risk of T2D, but compared with participants whose breakfast timing was 07:00–09:00, those who broke fast after 09:00 had an adjusted HR for T2D of 0.71 (0.51−0.99) [22]. This association was stronger in participants with impaired fasting glucose at baseline (0.61, 0.39−0.95) but not present in those without (0.83, 0.50–1.38), suggesting that the Dawn and Somogyi phenomena of elevated morning blood sugar [40] may have been present among participants with prediabetes so that eating earlier in the morning could have resulted in further elevated postprandial blood glucose, increasing T2D risk. The results of the present analysis supported this hypothesis because eating breakfast before 07:00 (compared to 07:00−09:00) was significantly associated with increased risk of MI in participants with T2D (those who might have had elevated blood sugar) only.

Several previous studies not specific to older adults reported that eating breakfast (versus skipping breakfast) was associated with risk of CAD and MI risk factors such as diabetes, higher bodyweight, dyslipidemia, and insulin sensitivity [15,16,41,42]. Meal timing studies are less common and tend to be single-day studies of biomarkers in younger adults [6]; however, later eaters as defined by median time of eating occasions throughout the day have been reported to have higher BMIs, higher percentages of body fat, larger waist circumferences, more insulin resistance, higher triglycerides, and higher leptin levels compared to early eaters [20]. These may be pathways through which breakfast eating influences cardiometabolic health over time. However, people with metabolic risk factors such as older age, high BMI, and hyperglycemia may have been more likely to adopt practices usually recommended by a doctor or dietitian [43], such as eating breakfast, which could have generated a reverse causation–type effect such as indication bias.

Similar to the present study, which found that the factors associated with frequently eating breakfast at baseline are factors associated with privilege (higher levels of education, being in the largest racial group, higher income, higher fruit and vegetable intake, and more physical activity), eating breakfast has been repeatedly associated with many demographic and lifestyle factors [21,22,34]. This suggests that there are multiple reasons why people may eat more or less frequently, and although we were able to adjust for many covariables in the present analysis, we could not rule out confounding by demographic and lifestyle factors that were unmeasured in the CHS, such as sleep duration, stress, food insecurity, light/dark exposure, timing of waking in relation to timing of breakfast, and circadian rhythms. It is interesting to note that progressive adjustment for covariables did not change the model much, which is unlike other breakfast studies in younger adults in which the risk has changed markedly with progressive adjustment for covariables [21,34,44]. Potentially, the effect of covariables is less in a population of older adults who have many years of health patterns that have solidified.

4.2. Strengths and limitations

The present study included a large sample of older adults with a good proportion of females (who were historically underrepresented in epidemiologic and clinical research), a long duration of follow-up, a prospective design, an exclusion of prevalent CAD, a wide range of socioeconomic and health variables, and a comprehensive repeated assessment of various demographic, lifestyle, dietary, and health-related factors to serve as covariables. However, the present study had some limitations that should be considered. Breakfast frequency and timing were assessed at only one point in time (baseline) in the CHS, so including repeated measurements of eating timing and frequency in our analysis was not possible and nondifferential misclassification error may have been present in our analysis [45]. Breakfast frequency and timing patterns were self-reported, and interpretation may have differed between participants. No universal validated questions or definitions for breakfast exist for us to use. A limited number of CHS participants reported “never” eating breakfast, so we could not examine the relationship between never eating breakfast and risk of CAD or MI within subgroups. In addition, we did not have details on the composition, quality, or variety of breakfast foods consumed, and nor did we know sleeping times or exactly what time participants woke up in the morning to study the length of time between waking and breaking fast. Finally, the generalizability of the study findings has limitations given that the participants were from American communities with a low percentage of non-White participants. Unfortunately, additional studies with the appropriate data capture and design do not yet exist, so we could not add multiple cohorts to our analysis to improve diversity and generalizability.

There are other limitations in our study that are applicable to all studies of older adults, and while they may not be ideal from a theoretical biostatistical perspective, they are not adequate rationale to avoid studying older adults. For example, studies of older adults are prone to a ‘left truncation’ situation where individuals enter a study after the onset of the event of interest, or are not observed at the natural time origin [46]. Additionally, there is always the potential for studies of older adults like ours to have survivor bias, as older adults who were eligible for this study may inherently be healthier than participants who may have had CAD or died before they could have participated. Additionally, many older adults were on medications that offered cardiometabolic protection even if they have no prevalent disease, which could also cloud the results of studies of developmental disease processes [47]. Despite these limitations, conducting analyses in older adults can still provide etiologic and clinical insight into best practice for older adults and is still an important endeavour, as older adults are a vibrant and growing segment of the population. The use of cause-specific Cox models can help to provide appropriate estimates of the rate for outcomes like MI and CAD in older adults even in the presence of the competing risk of mortality [37]. The cause specific HR approach in the present study estimates risk in the population of individuals who have not yet had an incident MI or CAD event and remain alive, which is often a group of clinical interest.

4.3. Future directions

For ethical and feasibility reasons, human intervention studies of fasting or delaying meals are rare and cannot last a long enough duration to assess the slow development of most chronic diseases over years or decades [48]. Because breakfast frequency and timing are not captured by the nutrition assessment tools commonly used in prospective cohort studies, such as food frequency questionnaires, few longitudinal cohort studies have breakfast timing and frequency data. The main value of the present study is that it is the first to examine breakfast frequency and timing in relation to risk of cardiovascular outcomes in older adults over a meaningful length of time. Thus, it could contribute to the future development of the first official evidence-based dietary guidelines for the timing and frequency of eating among adults (guideline committees state this is a research priority [6,7]), which should be age-specific and/or include older adults. Although the results of the present study suggest limited evidence for an association between breakfast frequency and risk of MI and CAD in older adults, null findings are still important to acknowledge in order to avoid publication bias where only significant findings tend to get published and translated into knowledge and recommendations. Additional prospective longitudinal cohort and clinically based studies of breakfast timing and frequency that enroll participants of diverse locations, races, sexes, health statuses, and ages are needed to address the limitations and confirm the findings of the present study. With the rise of popular diet trends such as intermittent fasting, which may or may not involve skipping breakfast, future studies should examine different types of eating timing patterns in older adults to elucidate where a cardio-protection benefit may occur.

5. Conclusions

The results of this study indicated that breakfast frequency and timing were both associated with lifestyle, demographics, and metabolic factors that are risk factors for CAD and MI but were not directly associated with CAD or MI except within subgroups defined by some of these risk factors. Addressing breakfast consumption in older adults may improve other established cardiometabolic risk factors such as malnutrition, hypertension, hyperglycemia, and dyslipidemia. The results from this study can be used to generate hypotheses for future research, and if replicated in other studies of older adults, may provide evidence in the future when dietary guideline panels develop recommendations for breakfast frequency and timing for older adults.

CRediT authorship contribution statement

Leah Cahill: Conceptualization, Writing - Original draft preparation, Methodology, Funding acquisition. Navjot Sandila: Formal analysis, Software, Writing - Review & Editing. Rania Mekary: Methodology, Writing - Original draft preparation. Mary Biggs: Methodology, Project administration, Writing - Review & Editing, Funding acquisition. Allie Carew: Data curation, Project administration, Writing - Review & Editing. Ratika Parkash: Writing - Review & Editing. Karthik Tennankore: Methodology, Writing - Review & Editing. Olga Theou: Writing - Review & Editing. Robin Urquhart: Writing - Review & Editing. Luc Djoussé: Funding acquisition, Writing - Review & Editing.

Kenneth Mukamal: Conceptualization, Funding acquisition, Writing - Review & Editing.

Declaration of Generative AI and AI-assisted technologies in the writing process

Not applicable. During the preparation of this work the author(s) used no generative AI or AI-assisted technologies in the writing process.

Funding

This research was supported by a category 1 Nova Scotia Health Research Fundgrant to LEC and by contracts HHSN268201200036C,HHSN268200800007C,HHSN268201800001C,N01HC55222,N01HC85079,N01HC85080,N01HC85081,N01HC85082,N01HC85083,N01HC85086,75N92021D00006, and grantsU01HL080295 andU01HL130114 from the National Heart, Lung, and Blood Institute(NHLBI), with additional contribution from the National Institute of Neurological Disorders and Stroke(NINDS). Additional support was provided by R01AG023629 from the National Institute on Aging(NIA). A full list of principal Cardiovascular Health Study investigators and institutions can be found at CHS-NHLBI.org. The funding sources were not involved in the study design, data collection, data analysis, interpretation of data, or manuscript drafting.

Data availability

Data described in the article, code book, and analytic code will not be made available because the data are not publicly available from the CHS.

Declaration of competing interest

None to declare

Appendix A. Supplementary data

The following is Supplementary data to this article: