Additive and Multiplicative Interactions Between Genetic Risk Score and Family History and Lifestyle in Relation to Risk of Type 2 Diabetes

Abstract

We examined interactions between lifestyle factors and genetic risk of type 2 diabetes (T2D-GR), captured by genetic risk score (GRS) and family history (FH). Our initial study cohort included 20,524 European-ancestry participants, of whom 1,897 developed incident T2D, in the Nurses’ Health Study (1984–2016), Nurses’ Health Study II (1989–2016), and Health Professionals Follow-up Study (1986–2016). The analyses were replicated in 19,183 European-ancestry controls and 2,850 incident T2D cases in the Women’s Genome Health Study (1992–2016). We defined 2 categories of T2D-GR: high GRS (upper one-third) with FH and low GRS or without FH. Compared with participants with the healthiest lifestyle and low T2D-GR, the relative risk of T2D for participants with the healthiest lifestyle and high T2D-GR was 2.24 (95% confidence interval (CI): 1.76, 2.86); for participants with the least healthy lifestyle and low T2D-GR, it was 4.05 (95% CI: 3.56, 4.62); and for participants with the least healthy lifestyle and high T2D-GR, it was 8.72 (95% CI: 7.46, 10.19). We found a significant departure from an additive risk difference model in both the initial and replication cohorts, suggesting that adherence to a healthy lifestyle could lead to greater absolute risk reduction among those with high T2D-GR. The public health implication is that a healthy lifestyle is important for diabetes prevention, especially for individuals with high GRS and FH of T2D.

Abbreviations

BMI

body mass index

CI

confidence interval

FH

family history

GR

genetic risk

GRS

genetic risk score

GWAS

genome-wide association study

HPFS

Health Professionals Follow-up Study

NHS

Nurses’ Health Study

NHSII

Nurses’ Health Study II

RERI

relative excess risk due to interaction

T2D

type 2 diabetes mellitus

WGHS

Women’s Genome Health Study

Type 2 diabetes mellitus (T2D) is a chronic metabolic disease caused by genetic and obesogenic lifestyle factors (1–3). The distribution of T2D risk across strata defined by genetic and lifestyle factors can suggest novel biological hypotheses or be used to evaluate existing hypotheses on the mechanisms contributing to T2D incidence. It can also be used to identify subsets of the population defined by genetic risk factors who might disproportionately benefit from lifestyle interventions. Previous studies examining the distribution of T2D risk across genetic and lifestyle factors have focused on “multiplicative interactions”—that is, whether the relative risk comparing individuals with both genetic and lifestyle exposures to those with baseline genetic and lifestyle values is statistically significantly different than the product of the relative risk for individuals with only genetic factors and the relative risk for individuals with only lifestyle factors (4–7). The presence or absence of such multiplicative interactions is difficult to interpret either in terms of biological mechanism or clinical or public health implications (8). In this study, we used a prospective cohort to estimate the incidence of T2D across strata defined by genetic and lifestyle factors and to assess whether the risk difference between the healthiest and least healthy lifestyle strata differs between high and low genetic risk strata.

Prior to the introduction of the genome-wide association study (GWAS), family history (FH) of T2D was the primary surrogate measure of genetic contribution to T2D risk, and individuals with a FH in first-degree relatives have approximately 2- to 6-fold higher risks of T2D (9–11). In early studies combining FH and genetic risk score (GRS) based upon the published GWAS, genetic variants showed that FH in first-degree relatives was associated with higher risk of T2D independent of GRS (12, 13). Although GWAS have identified more risk variants in recent studies, GWAS-identified variants still explain only a small proportion of T2D heritability (14), which suggests that FH still contains unidentified shared genetic or environmental information that is complementary to an updated GRS for T2D. Although FH indicates shared genetic and environmental factors, previous studies showed that FH of T2D in parents and siblings largely derives from genetic effects (15, 16). However, few studies have investigated the potential interaction between FH and lifestyle factors in relation to risk of T2D. Thus, in addition to GRS, we will further examine whether FH of T2D modifies the associations between lifestyle factors and risk of T2D.

In this study, we examined the interaction between lifestyle factors and genetic risk of T2D indicated by GRS and FH, and we assessed the magnitude of interaction on both additive and multiplicative scales. Interaction on the additive scale assesses whether the difference in absolute risk of T2D between unhealthy and healthy lifestyles differs between the high and low genetic risk groups. As such it is directly relevant to public health and clinical applications, because it suggests that adherence to a healthy lifestyle could lead to greater absolute risk reduction among those with high T2D-GR (genetic risk). Interaction on the multiplicative scale assesses whether the relative risk of unhealthy versus healthy lifestyle varies across genetic risk groups.

We conducted the analyses among 20,524 participants in a prospective study, the Women’s Genome Health Study (WGHS), and replicated the findings among 22,033 participants in 3 independent cohorts: the Nurses’ Health Study (NHS), Nurses’ Health Study II (NHSII), and Health Professionals Follow-up Study (HPFS).

METHODS

Study population

The WGHS is a prospective cohort study of 28,345 initially healthy US women for whole genome genetic analysis, and the study participants were health professionals aged ≥45 years at baseline from 1992–1995 (17). In our study, we included the 20,524 participants with self-reported European ancestry who were free of T2D, cardiovascular disease, and cancer at baseline. The NHS began in 1976, when 121,700 female registered nurses aged 30–55 years residing in 11 states were recruited to complete a baseline questionnaire. The NHSII was established in 1989 and consisted of 116,671 younger female registered nurses, aged 25–42 years at baseline. The HPFS was initiated in 1986 and was composed of 51,529 male dentists, pharmacists, veterinarians, optometrists, osteopathic physicians, and podiatrists, aged 40–75 years at baseline. In the NHS, NHSII, and HPFS, questionnaires were collected at baseline and biennially thereafter on detailed medical history, lifestyle, usual diet, and the occurrence of chronic diseases. The loss to follow-up rate of the 3 cohorts was <10%. The present analysis included 15,738 women and 6,295 men who were sampled as participants of case-control studies of several complex diseases (including T2D, cardiovascular disease, and several cancers) nested in the 3 cohorts and had genome-wide data available (18–21). All of the included participants were of self-reported European ancestry and clustered with European-ancestry samples based on the first 4 genetic principal components (20). All were free of T2D, cardiovascular disease, or cancer at baseline (1984 for the NHS, 1991 for the NHSII, and 1986 for the HPFS). The study protocols were approved by the institutional review boards of Brigham and Women’s Hospital and Harvard School of Public Health.

Genotyping and genetic risk score of T2D

In the WGHS, genotyping was conducted using HumanHap300 Duo “+” genotyping arrays with the Infinium II assay (Illumina Inc., San Diego, California) (17). The 1000 Genomes, Phase 1, Version 3 (March 2012), ALL reference panel was used for imputation with MACH (22). In the NHS, NHSII, and HPFS, details describing genotyping and imputation have been reported elsewhere (18). Briefly, genotyping was conducted using 5 different platforms (Affymetrix 6.0 (Affymetrix Inc., Santa Clara, California) and Illumina HumanHap, Illumina OmniExpress, OncoArray, and HumanCore Exome (Illumina Inc.)). After basic quality control, 1000 Genomes (Phase 3, Version 5) was used as reference panel for imputation (23). In this study, we identified 68 single-nucleotide polymorphisms that were known to be associated with T2D in previous GWAS (14, 24, 25) (Web Table 1, available at https://academic.oup.com/aje). We excluded genetic variants rs9936385 due to unsuccessful extraction and rs6819243 due to low imputation quality (r² = 0.39). The imputation quality score of the remaining 66 single-nucleotide polymorphisms is shown according to platform in Web Table 2. A weighted GRS was calculated and each risk allele was weighted according to its relative effect size (β coefficient in log scale), which was derived from the largest GWAS available at the time analyses were conducted, involving 34,840 cases and 114,981 controls (14). We subtracted the mean GRS in the WGHS and divided by the standard deviation in the WGHS to facilitate interpretability. (The GRS means and standard deviation in the WGHS were thus 0 and 1, by construction. The GRS means and standard deviations in the NHS, NHSII, and HPFS were also approximately 0 and 1 (Table 1)).

Table 1.

Baseline Characteristics of Participants in the Women’s Genome Health Study (1992–2016), Nurses’ Health Study (1984–2016), Nurses’ Health Study II (1989–2016), and Health Professionals Follow-up Study (1986–2016), United States

	WGHS (n = 20,524)		NHS (n = 8,363)			NHSII (n = 7,375)			HPFS (n = 6,295)
Characteristic	No.	%	Mean (SD)	No.	%	Mean (SD)	No.	%	Mean (SD)	No.	%	Mean (SD)	P for Difference Across Cohorts a
Age, years			54.1 (7.1)			51.7 (6.7)			35.8 (4.3)			54.7 (8.7)	0.04
Normalized GRS			0.0 (1.0)			−0.05 (0.77)			−0.05 (0.76)			−0.00 (0.76)	1
Total energy intake (kcal/day)			1,732 (524)			1,756 (522)			1,790 (529)			2,027 (609)	0.98
Family history of T2D	4,974	24		2,593	31		3,245	44		2,014	32		<0.001
Genetic risk of T2D
Low score or no FH	18,655	91		7,375	88		6,168	84		5,496	87
High score and with FH	1,869	9		988	12		1,207	16		799	13		<0.001
BMIb
Normal weight (<25.0)	10,759	52		4,871	58		5,300	72		2,727	44
Overweight (25.0–29.9)	6,317	31		2,303	28		1,339	18		2,855	46
Obese (≥30.0)	3,448	17		1,189	14		10	10		603	10		<0.001
AHEI-2010c
Low	6,696	33		2,818	34		2,442	33		2,115	34
Medium	6,959	34		2,786	33		2,457	33		2,084	33
High	6,869	33		2,759	33		2,476	34		2,096	33		0.59
Physical activity (MET-mins/week)
Low intensity (<8.3)	9,234	47		4,286	51		2,698	37		2,635	42
Moderate (8.3–16.7)	4,246	21		1,702	20		1,607	22		1,123	18
Vigorous (>16.7)	6,374	32		2,375	28		3,070	42		2,527	40		<0.001
Smoking
Never	10,445	51		3,751	45		4,941	67		2,886	46
Past	7,701	38		2,960	35		1,687	23		2,852	45
Current	2,378	12		1,652	20		747	10		557	9		<0.001
Alcohol intake (g/day)
Low (<5)	15,402	75		5,272	63		1,151	16		2,798	44
Moderate (5–15 for women and 5–30 for men)	3,590	17		1,902	23		5,949	81		2,657	42
Heavy (>15 for women and >30 for men)	1,532	7		1,189	14		275	4		840	13		<0.001
Lifestyle score (range, 0–10)
Healthiest (≥8)	3,663	18		1,334	16		2,023	27		1,458	24
Moderate healthy (5–7)	11,018	56		4,547	54		4,290	58		3,299	53
Least healthy (≤4)	5,173	26		2,482	30		1,062	14		1,419	23		<0.001
Hypertension	4,830	24		660	8		408	6		1,391	22		<0.001
Hypercholesterolemia	5,931	29		331	4		848	12		844	13		<0.001
Postmenopausal	11,089	54		4,626	55		357	5		NA		<0.001
Postmenopausal hormone use	9,062	44		1,425	17		368	5		NA		<0.001

Abbreviations: AHEI-2010, Alternative Healthy Eating Index 2010; BMI, body mass index; FH, family history; GRS, genetic risk score; HPFS, Health Professionals Follow-up Study; MET, metabolic equivalent of task; NHS, Nurses’ Health Study; NHSII, Nurses’ Health Study II; SD, standard deviation; T2D, type 2 diabetes mellitus; WGHS, Women’s Genome Health Study.

^a Analysis of variance was used for continuous variables and χ² test was used for categorical variables.

^b BMI calculated as weight (kg) divided by height (m)².

^c The AHEI-2010 score ranged from 0 to 100, with higher scores indicating a healthier diet.

Assessment of family history of T2D

In the WGHS, at baseline, study participants were asked to report whether any of their first-degree relatives (father, mother, and/or siblings) had ever had diabetes. Corresponding data was collected in the NHS in 1982, 1988, and 1992; in the NHSII in 1989, 1997, 2001, 2005, and 2009; and in the HPFS in 1987, 1990, 1992, and 2008.

Assessment of lifestyle factors

Dietary data was collected at baseline in the 4 cohorts (1992, 1984, 1991, and 1986 in the WGHS, the NHS, the NHSII, and the HPFS, respectively) using a food frequency questionnaire with approximately 130 items to obtain information on usual intake of foods and beverages. Since 1986, dietary data was collected every 4 years thereafter until 2010 in the NHS, NHSII, and HPFS. The validity and reproducibility of the food frequency questionnaire have been described in detail elsewhere (26–30). Questions about the consumption of alcoholic beverages (including beer, wine, and liquor) were included in each questionnaire. The Alternative Healthy Eating Index 2010 was developed based on intake levels of 10 components, which were chosen on the basis of their association with chronic disease and mortality risk in observational and interventional studies (31). The score emphasizes higher intake of fruit, vegetables, whole grains, long-chain n-3 fats, nuts and legumes, and polyunsaturated fatty acids and lower intake of sugar-sweetened beverages, red and processed meat, trans fat, and sodium. Each component was scored from 0 (unhealthy) to 10 (healthiest), and the total score ranged from 0 (nonadherence) to 100 (perfect adherence).

Information about physical activity was obtained at baseline in the 4 cohorts and was updated every 2 years in the NHS, NHSII, and HPFS. Participants were asked by questionnaire the amount of time they spent on average per week on physical activities, and weekly energy expenditure in metabolic equivalent task (MET)-hours was calculated (32). The reproducibility and validity of the physical activity questionnaire has been described elsewhere (33).

Height and weight were assessed by questionnaire at baseline in the 4 cohorts, and weight was updated every 2 years in the NHS, NHSII, and HPFS. Self-reported weight was highly correlated with measured weight (34). Participants’ smoking status (never smoker, former smoker, current smoker) was obtained at baseline in the 4 cohorts and updated every 2 years in the NHS, NHSII, and HPFS.

Creation of lifestyle score and assessment of genetic predisposition to T2D

We categorized Alternative Healthy Eating Index 2010 scores and physical activity into tertiles and assigned 0 to the lowest two and 2 to the highest tertile. In addition, we assigned 0, 1, and 2 to different categories of body mass index (BMI, calculated as weight (kg)/height (m)²: ≥30.0, 25.0–29.9, <25.0), smoking status (current smoker, former smoker, never smoker), and alcohol intake (≥15g/day, <5g/day, 5–15g/day), based on previous literature (2, 35). A lifestyle score was created based on the 5 lifestyle factors and ranged from 0–10, with a higher score indicating better adherence to a healthier lifestyle. In our analysis, we defined the healthiest lifestyle by a lifestyle score ≥8, the moderately healthy lifestyle by a score of 5–7, and the least healthy lifestyle by a score ≤4.

Because GRS and family history both contain genetic information about T2D, we created a simple composite genetic exposure variable combining GRS and FH information. We first dichotomized GRS using the upper one-third of GRS as cutoff. We chose this cutoff to better compare the binary GRS variable and FH, because the prevalence of FH in the 4 cohorts was approximately 25%–45%. Second, we defined high T2D-GR as participants with high GRS and FH of T2D and low T2D-GR as participants with low GRS or without FH of T2D.

Assessment of covariates

In the baseline questionnaires, information was collected on age, total energy intake, and self-reported hypertension and hypercholesterolemia. Menopausal status and postmenopausal hormone use in women were also ascertained.

Assessment of T2D

From baseline to 2016, participants with self-reported incident T2D were mailed a validated supplementary questionnaire regarding symptoms, diagnostic tests, and hypoglycemic therapy to confirm the diagnosis of diabetes. Cases were ascertained using the American Diabetes Association criteria (36), and only cases confirmed by the supplemental questionnaires were identified as T2D cases. In a validation study, of the 62 cases in the NHS and 59 cases in HPFS who were confirmed by the supplemental questionnaire, 61 and 57 were reconfirmed by reviewing medical records (37, 38). Participants with self-reported T2D but not confirmed by supplemental questionnaire were treated as non-T2D cases (2, 13). Among self-reported T2D cases, 89% were confirmed by supplemental questionnaire in the NHS, 94% in the NHSII, and 89% in the HPFS.

Statistical analysis

We examined associations between lifestyle factors combined and individually and binary genetic indicators (GRS, FH, and GRS and FH combined) using similar methods, which we have illustrated using lifestyle score and GRS as an example.

We used pooled logistic regression to examine joint association of lifestyle and GRS to risk of T2D (39). Because our focus was developing predictive models for future disease risk, given current covariate and lifestyle factors, we modeled risk using baseline lifestyle and covariate factors. We performed sensitivity analyses using updated lifestyle factors where available in the NHS, NHSII, and HPFS. The model was fitted using SAS (SAS Institute, Inc., Cary, North Carolina) “PROC NLMIXED” software, adjusting for age (continuous), total energy intake (continuous), self-reported hypertension (yes, no), hypercholesterolemia (yes, no), menopausal status (yes, no; women only), and postmenopausal hormone use (yes, no; women only). We assessed the relative excess risk due to interaction (RERI) as an index of additive interaction. Because RERI is defined for dichotomized exposures (33, 34), we calculated RERIs by comparing the healthiest to least healthy groups using the formula . The “estimate” command of “PROC NLMIXED” was used to obtain RERI. To examine the multiplicative interaction between lifestyle score and GRS, we modeled category of lifestyle score as an ordinal variable (0, 1, 2) and included a product term of lifestyle score and GRS in the pooled logistic model. Multiplicative interaction was indicated by the coefficient and 95% confidence interval of the product term. Fixed-effects meta-analysis was used to obtain pooled effect estimates across the 4 cohorts, and heterogeneity across cohorts was assessed by Cochrane Q test, with P < 0.1 indicating significant between-cohort heterogeneity.

Unlike participants from the NHS, NHSII, and HPFS, who were selected into nested case-control studies in the NHS, NHSII, HPFS, participants in the WGHS have been followed prospectively as a cohort with no sampling. Thus, we estimated 2-year risk of T2D within joint category of lifestyle factors and GRS in the WGHS at the median value of continuous and categorical covariates and obtained a 10-year risk of T2D using the formula under the assumption that the 2-year risk of T2D was the same over 10 years (40). In the WGHS, we further calculated 10-year attributable risks of T2D comparing the least to moderate and healthiest lifestyle in GRS low and high groups using the “estimate” command of the “PROC NLMIXED” procedure. All statistical tests were 2-sided and performed using SAS, version 9.4 (SAS Institute).

RESULTS

Our study included 20,524 participants in the WGHS, 8,363 participants in the NHS, 7,375 participants in the NHSII, and 6,295 participants in the HPFS (Table 1). During follow-up, 1,897 incident T2D cases were observed in the WGHS (median follow-up = 20 years), and 2,850 T2D cases were documented in the combined cohorts of the NHS, NHSII, and HPFS. As expected, we found moderate association of GRS with FH across the 4 cohorts (Web Table 3), indicating that FH and GRS contain overlapping but largely complementary genetic information. FH and GRS are independently associated with risk of T2D, with the strongest association among participants with high GRS and a FH of diabetes (Web Table 4). This supports our classification of “low GRS or no FH” and “high GRS and with FH” groups: High GRS with FH identified a particularly high-risk population of T2D, while high GRS and FH alone were moderately associated with risk of T2. FH was moderately associated with a less healthy lifestyle, while GRS was moderately associated with healthier lifestyle (Web Table 5).

In the WGHS, compared with participants with the healthiest lifestyle and low T2D-GR, the relative risk of T2D was 2.95 (95% confidence interval (CI): 1.93, 4.51) for participants with the healthiest lifestyle and high T2D-GR, 4.63 (95% CI: 3.72, 5.77) for participants with the least healthy lifestyle and low T2D-GR, and 11.07 (95% CI: 8.52, 14.37) for participants with the least healthy lifestyle and high T2D-GR (Table 2). We found significant additive interaction between lifestyle and genetic risk of T2D (P for additive interaction < 0.001; RERI = 2.97 (95% CI: 1.72, 4.21)); however, no significant multiplicative interaction was found (P for multiplicative interaction = 0.63). Significant additive interaction with lifestyle was found for both GRS and FH when considered individually (P for additive interaction <0.001 for both GRS and FH). As to cumulative risk of T2D, we found that among participants with low T2D-GR, the 10-year risks of T2D were 1.2% (95% CI: 1.0, 1.5) for those who adhered to the healthiest lifestyle and 5.5% (95% CI: 5.0, 5.9) for least healthy; among participants with high T2D-GR, the 10-year risks of T2D were 3.7% (95% CI: 2.4, 5.0) for those who adhered to the healthiest lifestyle and 12.5% (95% CI: 10.7, 14.4) for least healthy. The attributable risk of T2D for unhealthy lifestyle was 2.5% (95% CI: 2.2, 2.9) among participants with low T2D-GR and 4.8% (95% CI: 3.2, 6.4) among participants with high T2D-GR (P for difference in attributable risk = 0.006).

Table 2.

Ten-Year Estimated Risk of Type 2 Diabetes and Relative Risk of Type 2 Diabetes Within Joint Categories of Lifestyle Score and Genetic Indicators of Type 2 Diabetes in the Women’s Genome Health Study, United States, 1992–2016

Genetic Risk Category	Healthiest(Lifestyle Score ≥8)		Moderately Healthy (Lifestyle Score 5–7)		Least Healthy(Lifestyle Score ≤4)		Additive Interaction(RERI)			Multiplicative Interaction
	Estimate	95% CI	Estimate	95% CI	Estimate	95% CI	Estimate	95% CI	P Value	Estimate	95% CI	P Value
Genetic Risk Score
10-year risk of T2D, %a
Low GRS	1.0	0.7, 1.3	2.9	2.6, 3.1	4.9	4.5, 5.4
High GRS	2.0	1.5, 2.5	3.8	3.4, 4.2	7.1	6.4, 7.9
RR of T2Db
Low GRS	1.00	Referent	2.59	1.97, 3.40	5.02	3.81, 6.60	1.17	0.68, 1.66	<0.001	0.93	0.80, 1.09	0.38
High GRS	1.97	1.38, 2.82	3.79	2.87, 5.00	7.37	5.57, 9.76
Family History
10-year risk of T2D, %a
Without FH	1.0	0.8, 1.3	2.7	2.5, 3.0	4.7	4.3, 5.1
With FH	2.5	1.8, 3.1	5.2	4.6, 5.7	9.6	8.6, 10.6
RR of T2Db
Without FH	1.00	Referent	2.38	1.85, 3.07	4.63	3.59, 5.98	2.34	1.66, 3.03	<0.001	0.96	0.82, 1.12	0.60
With FH	2.48	1.73, 3.56	5.10	3.93, 6.62	9.89	7.62, 12.84
Genetic Risk Score and Family History Combined
10-year risk of T2D, %a
Low GRS or without FH	1.2	1.0, 1.5	3.1	2.8, 3.3	5.5	5.0, 5.9
High GRS and with FH	3.7	2.4, 5.0	6.5	5.5, 7.5	12.5	10.7, 14.4
RR of T2Db
Low GRS or without FH	1.00	Referent	2.36	1.90, 2.93	4.63	3.72, 5.77	2.97	1.72, 4.21	<0.001	0.95	0.79, 1.16	0.63
High GRS and with FH	2.95	1.93, 4.51	5.54	4.27, 7.18	11.07	8.52, 14.37

Abbreviations: CI, confidence interval; FH, family history; GRS, genetic risk score; RERI, relative excess risk due to interaction; RR, relative risk; T2D, type 2 diabetes mellitus.

^a 10-year risk of T2D was estimated using pooled logistic regression at median values of age, total energy intake, hypertension, hypercholesterolemia, menopausal status, and postmenopausal hormone use among whole WGHS population.

^b Pooled logistic model adjusted for age (continuous), total energy intake (continuous), hypertension (yes, no), hypercholesterolemia (yes, no), menopausal status (yes, no), and postmenopausal hormone use (yes, no). FH and GRS were adjusted for each other.

In a meta-analysis of the NHS, NHSII, and HPFS, we found similar additive interaction between lifestyle score and T2D-GR (P for additive interaction < 0.001; RERI = 2.46 (95% CI: 1.35, 3.58)) (Table 3). No significant multiplicative interaction was found (P for multiplicative interaction = 0.38; interaction relative risk = 0.94 (95% CI: 0.82, 1.08)). In a meta-analysis of all 4 cohorts, we found that genetic risk indicated by GRS (Figure 1A), FH (Figure 1B), and T2D-GR (Figure 1C) and lifestyle was jointly associated with risk of T2D. Compared with participants with the healthiest lifestyle and low T2D-GR, the relative risk of T2D was 8.72 (95% CI: 7.46, 10.19) among participants with the least healthy lifestyle and high T2D-GR. No significant interaction between lifestyle and T2D-GR in relation to risk of T2D was found on the multiplicative scale in either initial or replication cohorts (P = 0.31 by meta-analyzing the 4 cohorts; P for heterogeneity = 0.88) (Web Table 6). However, we found a significant departure from an additive risk difference model in both the initial and replication cohorts (P < 0.001 by meta-analyzing the 4 cohorts; P for heterogeneity = 0.24). Similar additive and multiplicative interactions were found for GRS and FH separately.

Table 3.

Relative Risk of Type 2 Diabetes Within Joint Categories of Lifestyle Score and Genetic Indicators of Type 2 Diabetes in the Nurses’ Health Study (1984–2016), Nurses’ Health Study II (1989–2016), and Health Professionals Follow-up Study (1986–2016), United States^a

	Healthiest (Lifestyle Score ≥8)		Moderate Healthy (Lifestyle Score 5–7)		Least Healthy (Lifestyle Score ≤4)		Additive Interaction (RERI)			Multiplicative Interaction
Genetic Risk Category	Estimate	95% CI	Estimate	95% CI	Estimate	95% CI	Estimate	95% CI	P Value	Estimate	95% CI	P Value
Nurses’ Health Study
GRS
Low GRS	1.00	Referent	2.51	1.73, 3.65	5.18	3.58, 7.49	0.95	−0.32, 2.22	0.14	0.81	0.68, 0.97	0.02
High GRS	2.34	1.47, 3.73	4.06	2.79, 5.90	7.46	5.12, 10.87
FH
No	1.00	Referent	1.77	1.28, 2.43	3.67	2.67, 5.05	3.86	2.37, 5.34	<0.001	1.00	0.84, 1.20	0.97
Yes	1.85	1.16, 2.94	4.50	3.27, 6.19	8.37	6.08, 11.53
GRS and FH combined
Low GRS or without FH	1.00	Referent	2.02	1.54, 2.67	4.08	3.10, 5.38	3.97	1.83, 6.12	<0.001	0.93	0.75, 1.15	0.51
High GRS and with FH	2.25	1.30, 3.90	5.11	3.76, 6.95	9.56	6.98, 13.08
Nurses’ Health Study II
GRS
Low GRS	1.00	Referent	1.81	1.25, 2.61	3.19	2.16, 4.73	1.30	−0.10, 2.70	0.07	1.06	0.80, 1.39	0.69
High GRS	1.12	0.66, 1.91	2.99	2.06, 4.34	4.62	3.04, 7.02
FH
No	1.00	Referent	2.39	1.42, 4.01	4.55	2.59, 7.98	2.85	0.31, 5.40	0.03	0.85	0.63, 1.14	0.27
Yes	2.92	1.65, 5.16	5.89	3.58, 9.67	9.38	5.61, 15.69
GRS and FH combined
Low GRS or without FH	1.00	Referent	2.10	1.50, 2.96	3.88	2.69, 5.59	1.66	−0.81, 4.13	0.19	0.91	0.68, 1.22	0.52
High GRS and with FH	2.10	1.19, 3.72	4.94	3.40, 7.15	7.21	4.70, 11.06
Health Professionals Follow-up Study
GRS
Low GRS	1.00	Referent	1.83	1.37, 2.44	3.64	2.72, 4.86	0.75	−0.15, 1.66	0.10	0.96	0.79, 1.17	0.72
High GRS	1.42	0.97, 2.07	2.48	1.85, 3.33	4.78	3.53, 6.49
FH
No	1.00	Referent	2.02	1.48, 2.76	4.16	3.04, 5.70	1.41	0.25, 2.56	0.02	0.84	0.69, 1.02	0.08
Yes	2.30	1.57, 3.36	3.68	2.70, 5.03	6.91	5.03, 9.48
GRS and FH combined
Low GRS or without FH	1.00	Referent	1.75	1.38, 2.22	3.49	2.73, 4.45	1.99	0.46, 3.53	0.01	0.97	0.77, 1.22	0.81
High GRS and with FH	1.72	1.10, 2.68	3.34	2.51, 4.44	6.38	4.71, 8.64
Pooled
GRS
Low GRS	1.00	Referent	1.96	1.61, 2.37	3.97	3.26, 4.84	0.82	0.16, 1.47	0.01	0.91	0.81, 1.02	0.11
High GRS	1.61	1.24, 2.08	3.11	2.56, 3.79	5.41	4.40, 6.66
FH
No	1.00	Referent	1.96	1.60, 2.41	3.99	3.24, 4.92	2.39	1.53, 3.25	<0.001	0.91	0.81, 1.03	0.14
Yes	2.26	1.74, 2.93	4.32	3.53, 5.29	7.86	6.40, 9.66
GRS and FH combined
Low GRS or without FH	1.00	Referent	1.92	1.63, 2.26	3.85	3.27, 4.54	2.46	1.35, 3.58	<0.001	0.94	0.82, 1.08	0.38
High GRS and with FH	2.09	1.56, 2.79	4.44	3.71, 5.33	7.59	6.24, 9.23

Abbreviations: CI, confidence interval; FH, family history; GRS, genetic risk score; HPFS, Health Professionals Follow-up Study; NHS, Nurses’ Health Study; NHSII, Nurses’ Health Study II; RERI, relative excess risk due to interaction.

^a Pooled logistic model adjusted for age (continuous), total energy intake (continuous), hypertension (yes, no), hypercholesterolemia (yes, no), case-control data sets (9 categories in the NHS, 6 categories in the NHSII, and 7 categories in the HPFS), menopausal status (yes, no, women only), and postmenopausal hormone use (yes, no, women only). FH and GRS were adjusted for each other.

Figure 1.

Relative risk (RR) of type 2 diabetes (T2D) within joint categories of lifestyle score and genetic indicators of T2D by pooling the Women’s Genome Health Study (1992–2016), Nurses’ Health Study (1984–2016), Nurses’ Health Study II (1989–2016), and Health Professionals Follow-up Study (1986–2016), United States. A) Genetic risk score (GRS); B) family history (FH) of type 2 diabetes; C) GRS and FH combined. Pooled logistic model adjusted for age (continuous), total energy intake (continuous), hypertension (yes, no), hypercholesterolemia (yes, no), case-control data sets (9 categories in the Nurses’ Health Study, 6 categories in the Nurses’ Health Study II, and 7 categories in the Health Professionals Follow-up Study), menopausal status (yes, no, women only), and postmenopausal hormone use (yes, no, women only). FH and GRS were adjusted for each other. CI, confidence interval.

We further examined interaction between individual lifestyle factors and genetic risk of T2D by meta-analyzing the 4 cohorts. Significant additive and multiplicative interactions with T2D-GR were found for BMI and physical activity (Table 4). As to additive interaction, the difference in T2D absolute risk between being obese and normal weight (being physically active and being physically inactive) was significantly higher among participants with high T2D-GR than that among participants with low T2D-GR. Because lifestyle was measured repeatedly in the replication cohorts, we evaluated the interaction with genetic predisposition to T2D using time-varying lifestyle, and the results were similar to our finding using baseline lifestyle (Web Table 7).

Table 4.

Relative Risk of Type 2 Diabetes Within Joint Categories of Individual Lifestyle Factor and Genetic Indicators of Type 2 Diabetes by Pooling the Women’s Genome Health Study (1992–2016), Nurses’ Health Study (1984–2016), Nurses’ Health Study II (1989–2016), and Health Professionals Follow-up Study (1986–2016), United States^a

Score and Family History	Individual Lifestyle Factors						P for Additive Interaction	P for Multiplicative Interaction
	Normal Weight		Overweight		Obesity
	RR	95% CI	RR	95% CI	RR	95% CI
GRS							<0.001	0.08
Low	1.00	Referent	2.82	2.52, 3.15	6.27	5.60, 7.01
High	1.63	1.43, 1.85	4.52	4.03, 5.07	8.76	7.76, 9.88
FH							<0.001	<0.001
No	1.00	Referent	3.12	2.79, 3.49	6.84	6.09, 7.68
Yes	2.51	2.19, 2.86	6.00	5.34, 6.73	12.34	10.98, 13.87
GRS and FH combined							<0.001	0.003
Low GRS or without FH	1.00	Referent	2.92	2.63, 3.24	6.41	5.75, 7.14
High GRS and with FH	2.37	2.05, 2.73	6.17	5.46, 6.97	12.16	10.68, 13.84
	Healthiest Diet		Moderately Healthy Diet		Unhealthy Diet
	RR	95% CI	RR	95% CI	RR	95% CI
GRS							0.96	0.27
Low	1.00	Referent	1.23	1.11, 1.36	1.24	1.12, 1.37
High	1.69	1.51, 1.89	1.66	1.48, 1.85	1.95	1.75, 2.17
FH							0.004	0.23
No	1.00	Referent	1.12	1.01, 1.25	1.15	1.04, 1.28
Yes	1.91	1.71, 2.14	2.14	1.92, 2.38	2.42	2.17, 2.69
GRS and FH combined							0.33	0.90
Low GRS or without FH	1.00	Referent	1.23	1.11, 1.36	1.38	1.23, 1.54
High GRS and with FH	2.22	1.96, 2.51	2.46	2.18, 2.77	2.37	2.07, 2.72
GRS							0.50	0.94
Low GRS	1.00	Referent	1.03	0.91, 1.16	1.20	1.09, 1.32
High GRS	1.58	1.41, 1.77	1.49	1.31, 1.70	1.82	1.64, 2.01
FH							0.01	0.17
No	1.00	Referent	0.96	0.85, 1.09	1.16	1.05, 1.29
Yes	1.88	1.68, 2.12	1.92	1.69, 2.19	2.28	2.06, 2.52
GRS and FH combined							0.001	0.03
Low GRS or without FH	1.00	Referent	0.97	0.86, 1.09	1.15	1.05, 1.26
High GRS and with FH	1.93	1.70, 2.19	1.99	1.72, 2.30	2.38	2.13, 2.65
	Never Smoker		Former Smoker		Current Smoker
	RR	95% CI	RR	95% CI	RR	95% CI
GRS							0.70	0.42
Low GRS	1.00	Referent	1.10	1.01, 1.20	1.46	1.29, 1.65
High GRS	1.53	1.40, 1.67	1.73	1.58, 1.90	2.07	1.80, 2.38
FH							0.96	0.04
No	1.00	Referent	1.13	1.03, 1.23	1.54	1.35, 1.74
Yes	2.05	1.88, 2.23	2.26	2.05, 2.48	2.67	2.32, 3.06
GRS and FH combined							0.94	0.10
Low GRS or without FH	1.00	Referent	1.11	1.02, 1.21	1.53	1.36, 1.72
High GRS and with FH	2.07	1.88, 2.28	2.29	2.07, 2.54	2.59	2.22, 3.01
GRS							0.62	0.89
Low GRS	1.00	Referent	1.32	1.18, 1.47	0.90	0.75, 1.07
High GRS	1.68	1.47, 1.93	1.91	1.70, 2.14	1.48	1.22, 1.79
FH							0.02	0.09
No	1.00	Referent	1.34	1.19, 1.50	0.98	0.83, 1.17
Yes	2.17	1.89, 2.49	2.51	2.24, 2.82	1.73	1.42, 2.10
GRS and FH combined							0.12	0.34
Low GRS or without FH	1.00	Referent	1.26	1.14, 1.39	0.92	0.78, 1.07
High GRS and with FH	2.32	1.99, 2.71	2.59	2.29, 2.93	1.88	1.49, 2.37

Abbreviations: AHEI-2010, Alternative Healthy Eating Index 2010; BMI, body mass index; CI, confidence interval; FH, family history; GRS, genetic risk score; HPFS, Health Professionals Follow-up Study; NHS, Nurses’ Health Study; NHSII, Nurses’ Health Study II; RR, relative risk.

^a Pooled logistic model adjusted for age (continuous), total energy intake (continuous), hypertension (yes, no), hypercholesterolemia (yes, no), case-control data sets (9 categories in the NHS, 6 categories in the NHSII, and 7 categories in the HPFS), menopausal status (yes, no, women only), and postmenopausal hormone use (yes, no, women only). FH and GRS were adjusted for each other. BMI, AHEI-2010 score, physical activity, smoking status, and alcohol intake were mutually adjusted for each other.

DISCUSSION

Using 20,524 participants in the WGHS, our study found that a healthy lifestyle—including maintaining a normal weight, adhering to a healthy diet, being physically active, no smoking, and moderate alcohol drinking—might be important for diabetes prevention, especially for individuals with high GRS and family history of T2D. Our results were replicated among 22,033 participants in the NHS, NHSII, and HPFS.

Previous studies have examined interactions between individual variants and lifestyle factors in relation to risk of T2D on a multiplicative scale and have found significant interactions for transcription factor 7-like 2 (TCF7L2), solute carrier family 30 member 8 (SLC30A8), potassium inwardly rectifying channel subfamily J member 11 (KCNJ11), and peroxisome proliferator activated receptor gamma (PPARG) (41–44). Further studies examined multiplicative interactions in relation to T2D by using a GRS approach; however, the results generated from these studies were inconsistent. The HPFS found that the association of Western dietary pattern with risk of T2D was stronger among participants with higher GRS (7), while the Diabetes Prevention Program Study found no significant interaction between lifestyle intervention and GRS based on 34 T2D-associated variants (4). In contrast, the European Prospective Investigation Into Cancer and Nutrition–InterAct study showed stronger positive association between BMI and risk of T2D among participants with low GRS (5), and the Atherosclerosis Risk in Communities study found that the association between physical activity and risk of T2D was stronger among participants with low GRS (6). In our study, we found no significant multiplicative interaction between lifestyle score and genetic risk of T2D on multiplicative scales after meta-analyzing the 4 cohorts.

In the WGHS, the risk of T2D attributable to an unhealthy lifestyle was higher among participants with high genetic risk of T2D. We consistently found that the attributable risk of T2D among participants with low genetic risk was lower than that among participants with high genetic risk across the 4 cohorts. These results suggest that interventions promoting healthy lifestyles could lead to a greater proportional reduction in T2D incidence among high-genetic-risk individuals. The implication for health policy lies in that, given limited cost, T2D prevention would be more effective among participants with a high genetic risk. In fact, our results were in line with previous studies showing that a healthy lifestyle, including a high-quality diet and being physically active, attenuated the genetic predisposition to BMI (45–47), which is a leading risk factor for T2D. Specifically, by modeling BMI as a continuous outcome and adding an interaction term to a linear model, the interactions between lifestyle factors and genetic predisposition to BMI assessed in these studies were also on additive scales.

Our study found that BMI, physical activity, and Alternative Healthy Eating Index 2010 were main lifestyle factors responsible for interaction with risk of T2D. Large prospective studies have shown that the risk of T2D can be reduced substantially by maintaining a normal BMI, being physically active, and consuming a healthy diet (2, 48). The potential mechanism might be that a healthy lifestyle can inhibit insulin resistance, inflammation, and oxidative stress and slow the accumulation of cellular and organ damage (49).

Family history was the primary surrogate measure of genetic contribution to T2D risk before GWAS, and previous studies showed that after accounting for shared familial environmental factors the genetic components of FH remained to explain 50% heritability of T2D (14, 16). However, it is still worthwhile to further investigate which genes in the FH are responsible for the polygenetic effect of T2D (10). In our study, we found that FH significantly interacted with lifestyle factors, and the genetic risk of T2D measured by combining GRS and FH had the strongest interaction with lifestyle score. Given that the risk of T2D was reduced remarkably by adopting a healthy lifestyle for individuals with high genetic risk of T2D, the genetic indicator we created by combining FH and GRS has significant implications for personalized diabetes prevention.

Our study has several strengths. First, we used FH as a genetic indicator of T2D, and we created a new measure of genetic risk of T2D by combining GRS and FH. Second, we quantified the interaction on an additive scale and consistently found that the difference in relative risk of T2D, comparing an unhealthy lifestyle with a healthy one, was higher among participants with higher genetic risk of T2D, which is relevant to the potential clinical utility of genetic information for personalized lifestyle recommendations. Third, in the discovery and replication populations, lifestyle factors were measured before T2D incidence, which minimized reverse causation by change of lifestyle factors due to cardiometabolic diseases. Fourth, lifestyle factors were measured repeatedly in the NHS, NHSII, and HPFS, which represented long-term lifestyle and minimized measurement errors.

Our study also has several limitations. First, it was an observational study, and the interactions we found might still be confounded by other factors. However, we adjusted for multiple confounders in our analysis. Second, lifestyle factors were measured by questionnaire, which might be prone to measurement error. Family history of diabetes might include type 1 diabetes mellitus. However, one recent study found that type 1 diabetes accounted for only 5.6% of diagnosed diabetes in US adults (50). Furthermore, although random measurement error in exposures attenuates the interaction effect toward the null (51), our study still detected significant interactions with BMI, Alternative Healthy Eating Index 2010, physical activity, and alcohol intake. Third, because our study included only persons of European ancestry, whether our findings can be generalized to other ethnic groups needs further investigation. Fourth, as expected, family history was associated with a less-healthy lifestyle, and the reason might be due to shared lifestyle factors between parents and offspring. GRS was also associated with lifestyle, given that the associations between genetic variants and risk of T2D can be mediated by BMI and other lifestyle factors. While these correlations provide some information on the pathways through which T2D risk is inherited—specifically, they suggest that the amount mediated through healthy lifestyle is modest—the main focus of our paper is the effect of lifestyle factors, whether they themselves are influenced by genetic factors, within genetic risk strata. Our findings that lifestyle factors are associated with T2D risk across genetic strata—indeed, that the risk difference between healthy and unhealthy lifestyle factors is largest among individuals with high genetic risk—underscore that genetics is not destiny, that interventions on modifiable risk factors can reduce risk even among those at highest genetic risk.

In conclusion, our study underscores that adhering to a healthy lifestyle—including maintaining a normal weight, adhering to a healthy diet, being physically active, not smoking, and moderate alcohol drinking—is important for diabetes prevention, especially for individuals with high GRS and FH of T2D.