Ethnicity, subjective wellbeing, and cardiovascular disease: insights from path and machine learning analyses using data from the UK biobank

Abstract

Objectives

The study is centred around two objectives: (1) to elucidate the effects of ethnicity on subjective wellbeing (SWB) and its interrelation with demographic factors and cardiovascular disease (CVD); and (2) to identify key predictors of SWB and CVD using machine learning.

Design

Employing data from 296,767 UK Biobank participants in a cross-sectional design, we conducted path analysis, and machine learning analysis to investigate the impact of ethnicity on CVD and SWB, addressing missing data through sensitivity analyses. Our models used logistic and linear regression, complemented by receiver operating characteristic analysis, to explore direct and indirect effects, and feature importance.

Results

Ethnicity was significantly associated to both outcomes directly and acting as a mediating variable when evaluating the association between key demographic variables and the outcomes. Ethnicity influenced CVD and SWB, with non-White groups showing higher CVD odds and lower SWB. Age, BMI, waist circumference, smoking status, depression, and sex were significant predictors of CVD, while factors like handgrip strength and alcohol intake showed protective effects.

Conclusion

This study underscores the critical need for ethnic-specific health interventions and highlights the complex interplay between demographic factors, CVD, and SWB. Our findings offer a foundation for developing targeted public health strategies and policies aimed at reducing health disparities and improving wellbeing across ethnic groups. Future research should continue to explore these relationships, emphasising the importance of culturally sensitive approaches to health promotion and disease prevention.

Introduction

Cardiovascular diseases (CVD) constitute a leading global health challenge, encompassing a spectrum of conditions related to the heart and blood circulation. CVD accounts for a approximately 32% of global deaths, representing one of the primary causes of morbidity and premature mortality worldwide [1]. Numerous modifiable risk factors for the development of CVD, such as hypertension, obesity, and smoking, have been studied extensively [2–5]. However, recent advancements in health research advocate for a holistic understanding of health outcomes that integrates both physical and psychological dimensions. For example, CVD has been linked to adverse mental health outcomes and cognitive decline, further underscoring its multidimensional impact on individuals’ well-being [6, 7]. Such an approach is imperative for addressing the complex interplay of risk factors in diverse populations.

Ethnicity is a known factor in CVD risk prediction [8], with ethnic minority populations in countries like the United Kingdom (UK) often experiencing worse CVD-related outcomes [9]. The variations in prevalence, incidence, and outcomes of CVD extend beyond environmental influences, emphasising the complex interplay of cultural and healthcare access factors [10]. A critical gap remains in the literature concerning the simultaneous exploration of subjective wellbeing (SWB) in diverse populations.

SWB encompasses components such as happiness and life satisfaction, representing an individual’s evaluation of their own life [11] and is increasingly recognised as a vital component of health [12]. Understanding how SWB relates to CVD and exploring whether ethnicity moderates this relationship is a key aspect that warrants investigation. Studies investigating this relationship yielded varying results. A study which compared ethnicity to SWB using a structural model comprised of endogenous variables such as physical activity found a significant association between SWB and ethnicity [13], however another study adjusting for education found no such association [14]. Despite these insights, the extent to which ethnicity moderates the relationship between CVD and SWB remains poorly understood, underscoring the need for research in this domain.

The limited body of evidence addressing the ethnicity-SWB relationship has produced inconsistent findings. For instance, studies have reported significant ethnic disparities in SWB in South Africa and Australia [15, 16] which are distinct populations. Conversely, research conducted in other populations adjusting for educational attainment found no significant association between ethnicity and SWB [17]. These conflicting results highlight the importance of studying how ethnicity influences SWB in diverse, large-scale populations and the need to consider potential mediators, such as CVD.

Exploring the intersection of ethnicity, SWB, and specific health conditions such as CVD is essential for a nuanced understanding of health disparities. Ethnicity not only directly affects health outcomes but may also shape the pathways linking socio-demographic factors to SWB through mediators such as CVD. For example, socio-demographic factors like age and socioeconomic status may influence SWB indirectly by increasing the likelihood of CVD, which in turn exacerbates SWB disparities across ethnic groups. Despite the logical framework supporting this hypothesis, evidence is sparse, particularly in large, ethnically diverse populations such as the UK Biobank cohort.

Ethnicity can influence SWB through multiply sociocultural and psychological pathways. Social identity theory suggests that belonging to a valued ethnic group can promote self-esteem and life satisfaction [18], whereas perceived discrimination can dimmish SWB [19]. Certain frameworks further propose that differing cultural norms, expectations, and coping strategies shape how individuals evaluate their own wellbeing [20]. Empirical evidence indicates that ethnic minority groups often face structural barriers that affect health, including differential access to social support [21], healthcare, and economic opportunity [22]. Consequently, ethnicity may not act only as a demographic descriptor, but also as a proxy for lived social experience that directly shapes wellbeing.

This study seeks to bridge a critical gap by conducting a comprehensive examination of the relationship between ethnicity, CVD, and SWB. Amidst CVD’s significant global health burden, this research advances beyond traditional physical health metrics to explore the multifaceted impacts of ethnicity on health outcomes. There remains a notable void in understanding the concurrent effects of SWB across diverse populations which plays an essential role in overall health yet have been insufficiently studied. Previous investigations into the ethnicity-SWB nexus have produced mixed findings, underscoring the complexity of this relationship and highlighting the necessity for a more thorough analysis that incorporates a variety of factors using advanced analytical methodologies.

Materials and methods

Data source

Data were obtained from the UK Biobank (UKB) [23], a large prospective study of around 500,000 participants in the UK who were enrolled at ages from 40 to 69 years between 2006 and 2010. The UK Biobank was originally designed to explore the relationship between environmental, genetic, and lifestyle factors and health outcomes. For the purposes of this cross-sectional study, this study focused specifically on variables related to CVD and SWB, along with demographic, mental health, general well-being, and physical activity indicators.

In this cross-sectional study of UKB data, pertinent data were related to CVD and SWB. This cohort contained 296,767 participants with valid SWB as part of UKB application number 103,665. Beyond wellbeing, the present study utilised data on demographic characteristics, health-related outcomes, mental health, general wellbeing, and physical activity variables. As ethnicity was a key factor in the analysis, there were no restrictions on ethnicity of participants. Researchers can access the data through an established application process.

Data collection occurred over a period spanning significant societal events, including the 2008 financial crisis. This event may have influenced participants’ subjective well-being and cardiovascular health outcomes, particularly among ethnic minorities, who faced heightened economic stress and constrained access to resources during this time [24].

The UK Biobank does not provide exact dates for when each participant was surveyed; however, we acknowledge that contextual factors like the financial crisis could have introduced variability in responses, especially for measures related to economic satisfaction or health status. To account for this, the analysis includes measures such as financial satisfaction and general well-being, which may partially capture these effects. We also acknowledge the potential impact of external stressors, such as the financial crisis, as a limitation of the study.

Consent to participate declaration

All participants provided written informed consent to participate in the UK Biobank study.

Patient and public involvement

In our secondary data analysis utilising UK Biobank data, we recognise the importance of patient and public involvement (PPI) despite the absence of direct participant engagement. We engaged with relevant groups during planning, ensuring our research questions were meaningful. We plan to involve the public in interpreting and disseminating findings, ensuring relevance and accessibility. Ethical considerations include protecting participant confidentiality, while acknowledging limitations in PPI effectiveness and viewing this as an opportunity for future research refinement.

Key variables

Cardiovascular disease types

UKB participants were asked whether they had experienced various cardiovascular events, and the dates that they were experienced. A variable was created which was assigned the value ‘1’ if a UKB participant experienced any CVD and ‘0’ if they had not. Binary variables were created for each cardiovascular subtype as stated in the International Statistical Classification of Diseases and Related Health problems 10th Revision (ICD-10) [25].

Ethnicity

The ethnicities of UKB participants were self-reported and recorded in six categories: White, Mixed, South Asian, Black, East Asian, and Other. Each of the categories were divided into sub-categories such as British, Irish, and other White background for the White ethnic group. The chosen ethnic categorisation aims to capture diverse ethnic backgrounds and is integral to addressing the research objectives and are broadly similar to the ones used in the UK census [26].

Subjective wellbeing measures

SWB was assessed using diverse measures within the UKB, encompassing various components. The study employed an approach of aggregating scores from five domains, specifically assessing participants’ general happiness, and satisfaction with family, friendship, health, and financial situation. This resulted in a continuous variable ranging from 0 to 100, where higher scores indicated elevated levels of SWB. This composite variable serves as the primary outcome measure for the study. The conceptualisation of SWB as a multi-faceted construct aligning with satisfaction across life domains is consistent with definitions from previous research [27, 28], which view SWB as encompassing both emotional and cognitive evaluations of one’s life. This approach to SWB has been taken previously [29–31] due to the lack of a SWB measure in the UKB.

To validate the appropriateness of combining these five variables into a single composite score, statistical analyses were conducted, including factor analysis to assess the underlying structure of the combined variables and confirm if they indeed contribute to a shared construct of SWB, and reliability analysis such as Cronbach’s alpha, which can gauge the internal consistency of the composite score, providing insights into the reliability of the measurement.

Statistical analysis

Previous studies have demonstrated that ethnic differences in cardiovascular risk profiles are well-established, with minority groups often experiencing high rates of hypertension, diabetes, and obesity [32–34]. At the same time, CVD has been associated with reduced SWB and life satisfaction, reflecting its impact on physical functioning and emotional health [35–37]. These findings provide a theoretical and empirical basis for the hypothesised pathway in which ethnicity may influence SWB indirectly though cardiovascular health.

Demographic characteristics were chosen based on variables that have been found to be associated with CVD or SWB [2, 38–40]. This includes age, body mass index (BMI), waist circumference, ethnicity, sex, smoking status, short-term depression, alcohol intake, how many days a week one performs moderately intense physical activity, and handgrip strength. Continuous variables were assessed for normality using the skewness-kurtosis test [41] and using histograms. None were normally distributed; therefore, these were summarised as median and interquartile range (IQR).

To integrate hypothesis-driven and data-driven methods, the study followed a two-step approach. First, Random Forest models were employed as a data-driven exploration to identify the strongest predictors of SWB and CVD. These results informed the subsequent hypothesis-testing phase using path analysis, which was designed to evaluate theoretically grounded relationships and mediating pathways. This sequence ensured that the path analysis focused on the most relevant predictors, thereby enhancing both its conceptual relevance and statistical robustness.

Predictive modelling through machine learning were employed to further elucidate the relationship considering the findings from the path analysis. Separate random forest models were developed for each outcome (SWB and CVD) using the ‘randomForest’ package. The model for SWB was implemented as a regression model, as SWB is a continuous variable, while the model for CVD was implemented as a classification model, as CVD is a binary outcome. The same variables used in the SEM were directly incorporated into the random forest models, as the primary aim was to assess their predictive contributions to SWB and CVD based on the relationships identified in the path analysis. The dataset was split into 80% training and 20% validation sets and evaluated 1,000 times, each using a different random seed to randomly shuffle the data before splitting. This approach reduces bias introduced by any single random split and provides a more comprehensive assessment of the model’s performance. The performance of these models was assessed using R-squared values and residual analysis to check for systematic errors in the prediction. Feature importance analysis was then conducted based on the aggregated results of the random forest models. The importance score for each predictor was derived from the average decrease in impurity, Gini importance, across all 1,000 iterations. The Gini importance is calculated as the probability of incorrectly classifying a randomly chosen observation if it were assigned a random label based on the distribution of the target variable [42]. This score reflects the relative contribution of each feature to the prediction of SWB and CVD. Predictors with higher scores were deemed more influential in the prediction process, and their rankings were consistent across the iterations, ensuring robustness in the findings.

Path analysis was used to assess the mediating role of demographic factors and CVD in the relationship between ethnicity and SWB. Direct and indirect effects were estimated to understand the pathways influencing SWB, which are presented using a directed acyclic graph. Specific paths in the structural equation model (SEM) included direct effects from ethnicity to SWB and CVD, as well as mediation paths through demographic factors. The selection of variables for this analysis was guided by their hypothesised or demonstrated associations with the constructs under investigation, ensuring relevance to the theoretical framework of the study. Categorical variables in SEM were dealt with using diagonally weighted least squares methods [43].

To explore potential variations, separate models were fitted for each ethnic group. Model fit was assessed using the chi-square test, comparative fit index (CFI), and root mean square error of approximation (RMSEA). Assumptions for linearity and multicollinearity were assessed using scatter plots and variance inflation factor (VIF), respectively. Analyses were performed in RStudio v4.1.0 [44] using the ‘lavaan’ package for SEM.

The dataset underwent a thorough examination to determine the patterns of missing data. Missing data rates were assessed for each variable. In the main analysis, cases with missing data were excluded to maintain the integrity of the primary investigation. Sensitivity analyses were conducted using multiple imputation to generate several plausible datasets with imputed values for the variables with missing values. Multiple imputation is a statistical technique used to handle missing data by creating several plausible datasets, each containing imputed values for the missing entries. These imputed values are generated based on the observed data and the relationships between variables, allowing for variability and uncertainty in the imputation process. The method enables more accurate and unbiased estimates by combining results across these datasets rather than relying on a single imputed dataset or excluding cases with missing data entirely. Analyses were conducted using the imputed values to explore the influence of missing data on the primary analysis results. Statistical significance was set at 5%.

Results

Summary characteristics

Demographic characteristics of the UKB population with valid SWB at baseline are found in an accompanying paper by Patel et al. [45]. The median age was 57 years however the non-White ethnicities were younger on average. There were more females in the sample than males. BMI was 26.5 kg/m² which is in the overweight category, the East Asian group were the only group with a BMI in the normal range (BMI = 23.7). Median waist circumference was 89 cm which is borderline unhealthy [46]. Over half the sample had never smoked and 9% currently smoke. Most participants have not had depression in the previous six months, with the proportion in the South Asian group lower compared to the other groups. More than half of the sample have at least one drink of alcohol per week, although this is weighted more by the White ethnic group. This proportion is lower for some of the other ethnic groups such as the South Asian group where 42% never drink alcohol and only 30% drink once a week. There is a broadly equal split of the number of days participants perform moderately intense physical activity. Median dominant handgrip strength was 30 kg, with the lowest being in the South and East Asian groups at 26 kg.

Table 1 presents the number of participants in this study who had CVD including the various cardiovascular subtypes. Over 54% of total participants had CVD. This was mostly due to the White group where 54% of participants had CVD. The South Asian and Black ethnic groups had a higher proportion of participants with CVD compared to Whites, 57.7% and 59.2% respectively, while the Other ethnic group had 51.8% of participants with CVD. Conversely, most of the Mixed and East Asian groups did not have CVD. The subtype with the lowest proportion was rheumatic heart fever with around 0.3% of cases, and the subtype with the highest proportion was hypertensive disease with 37.5% of cases.

Table 1.

Number (%) of UKB participants with any CVD and CVD subtypes; overall and by ethnicity

	Overall (N = 296,767)		White (N = 279,011)		Mixed (N = 1,189)		South Asian (N = 6,894)		Black (N = 5,220)		East Asian (N = 928)		Other (N = 2,825)
Variable	N	%	N	%	N	%	N	%	N	%	N	%	N	%
CVD (any)
No	136,696	45.9%	128,237	46.0%	1,008	53.4%	2,917	42.3%	2,130	40.8%	513	55.3%	1,362	48.2%
Yes	161,245	54.1%	150,774	54.0%	881	46.6%	3,977	57.7%	3,090	59.2%	415	44.7%	1,463	51.8%
Acute rheumatic fever
No	297,000	99.7%	278,098	99.7%	1,888	99.9%	6,885	99.9%	5,216	99.9%	927	99.9%	2,818	99.8%
Yes	941	0.3%	913	0.3%	1	0.1%	9	0.1%	4	0.1%	1	0.1%	7	0.2%
Chronic rheumatic heart disease
No	291,917	98.0%	273,373	98.0%	1,866	98.8%	6,717	97.4%	5,125	98.2%	917	98.8%	2,770	98.1%
Yes	6024	2.0%	5638	2.0%	23	1.2%	177	2.6%	95	1.8%	11	1.2%	55	1.9%
Hypertensive diseases
No	186,138	62.5%	175,431	62.9%	1,278	67.7%	3,705	53.7%	2,666	51.1%	629	67.8%	1,723	61.0%
Yes	111,803	37.5%	103,580	37.1%	611	32.3%	3,189	46.3%	2,554	48.9%	299	32.2%	1,102	39.0%
Ischaemic heart diseases
No	264,253	88.7%	247,824	88.8%	1,727	91.4%	5,564	80.7%	4,766	91.3%	871	93.9%	2,493	88.2%
Yes	33,688	11.3%	31,187	11.2%	162	8.6%	1,330	19.3%	454	8.7%	57	6.1%	332	11.8%
Pulmonary heart diseases
No	290,339	97.4%	271,830	97.4%	1,850	97.9%	6,783	98.4%	5,060	96.9%	918	98.9%	2,756	97.6%
Yes	7,602	2.6%	7,181	2.6%	39	2.1%	111	1.6%	160	3.1%	10	1.1%	69	2.4%
Other forms of heart disease
No	250,849	84.2%	234,412	84.0%	1,688	89.4%	5,891	85.5%	4,591	88.0%	856	92.2%	2,460	87.1%
Yes	47,092	15.8%	44,599	16.0%	201	10.6%	1,003	14.5%	629	12.0%	72	7.8%	365	12.9%
Cerebrovascular diseases
No	283,738	95.2%	265,760	95.3%	1,814	96.0%	6,522	94.6%	4,935	94.5%	903	97.3%	2,691	95.3%
Yes	14,203	4.8%	13,251	4.7%	75	4.0%	372	5.4%	285	5.5%	25	2.7%	134	4.7%
Diseases of arteries
No	283,761	95.2%	265,578	95.2%	1,811	95.9%	6,594	95.6%	5,038	96.5%	909	98.0%	2,710	95.9%
Yes	14,180	4.8%	13,433	4.8%	78	4.1%	300	4.4%	182	3.5%	19	2.0%	115	4.1%
Diseases of veins, lymphatic vessels, and lymph nodes
No	254,620	85.5%	238,194	85.4%	1,636	86.6%	5,984	86.8%	4,549	87.1%	814	87.7%	2,448	86.7%
Yes	43,321	14.5%	40,817	14.6%	253	13.4%	910	13.2%	671	12.9%	114	12.3%	377	13.3%
Other and unspecified disorders
No	285,965	96.0%	267,795	96.0%	1,831	96.9%	6,551	95.0%	5,039	96.5%	909	98.0%	2,726	96.5%
Yes	11,976	4.0%	11,216	4.0%	58	3.1%	343	5.0%	181	3.5%	19	2.0%	99	3.5%

Median subjective wellbeing was 75 out of 100 across all ethnicities, indicating that participants were moderately to very happy when combining the five domains as mentioned in Sect. 2.2.3. White participants had the highest median wellbeing score of 75 (IQR = 16.7), then East Asians (score = 73.3; IQR = 12.5), then the remaining four ethnic groups all with a score of 70.8.

Machine learning insights: deciphering the predictive power of ethnic and health dynamics

The data-driven exploration using Random Forest models revealed key predictors of subjective well-being (SWB) and cardiovascular disease (CVD). Separate models were constructed for SWB and CVD outcomes, with SWB treated as a continuous variable in regression models and CVD as a binary classification model. Feature importance analysis indicated that the most significant predictor for CVD was age, contributing more than double the importance of the next variable, BMI. This was followed by waist circumference and handgrip strength, underscoring the prominence of age and physical performance in predicting cardiovascular health.

For SWB, physical factors dominated, with BMI as the most important feature, followed by waist circumference and handgrip strength. Age and depression were the next most significant factors, suggesting that physical performance, age, and mental health strongly influence subjective well-being. These results are illustrated in Fig. 1, which depicts the relative feature importance for each factor in the predictive models (left = CVD; right = SWB). Feature importance for predictive models of each ethnicity separately is presented in Appendix Figs. 3 and 4.

Fig. 1.

Relative feature importance for the models predicting CVD and subjective wellbeing (ETH Ethnicity; BMI Body mass index, WC Waist circumference, MPA Moderate physical activity, DHS Dominant handgrip strength, SS Smoking status, Al)

When assessing the predicted values of random forest models for each of the six ethnic groups separately for subjective wellbeing, each model had a p-value of < 0.001, suggesting the overall regression model is statistically significant. Furthermore, with an R-squared value of 68.8% for the White ethnicity model and between 20 and 30% for the other ethnic groups, this suggests good fit for the White ethnic group and moderately good fit for the other groups. However, R-squared values may not fully capture classification model performance; hence, we emphasize the importance of the additional metrics presented. For the CVD outcome, all predictive models were statistically significant, with chi-squared p-values of < 0.0014. The model for the White ethnicity had the highest pseudo R-squared value (72%), with the models of other ethnicities having a pseudo R-squared of around 10%. These disparities highlight differential predictive performance across ethnic groups, potentially reflecting varying data representation or unmeasured confounders.

Path analysis: unveiling the web of ethnicity, demographics, and health outcomes

Building on the insights from the machine learning analysis, the path analysis assessed the mediating role of ethnicity, demographic factors, and CVD in shaping SWB. Non-White ethnicity was negatively associated with both SWB and CVD compared to Whites. Specifically, the direct path estimate from ethnicity to SWB was − 0.586, indicating a meaningful decrease in SWB for Non-White participants. However, the direct effect of ethnicity on CVD status, while statistically significant, was notably small, with an estimate of −0.007, corresponding to an odds ratio close to one. This suggests that the practical significance of ethnicity’s direct impact on CVD status is minimal.

For other variables, age showed a direct positive association with SWB while physical performance indicators like BMI, waist circumference, and dominant handgrip also had significant but varying effects. Mental health factors, such as short-term depression, exhibited the largest negative effect on SWB, highlighting their substantial practical significance. Additionally, behaviours like moderate physical activity and alcohol intake showed positive associations with SWB.

The indirect paths further illuminate how ethnicity mediates the relationships between other predictors and SWB. For example, the indirect effect of age through ethnicity was − 0.002 (p < 0.001), while the indirect effect of depression via ethnicity was 0.103 (p < 0.001). Although statistically significant, most indirect effects were small in magnitude, reflecting limited practical influence. The total indirect effect of all predictors mediated by ethnicity was − 0.898 (p < 0.001).

Goodness-of-fit tests concluded that the model structure was a good fit to the data. While many pathways showed statistical significance, their practical implications vary. For example, the direct effect of ethnicity on CVD status is negligible, highlighting the importance of contextual interpretation beyond statistical thresholds. The structural pathways are depicted in Fig. 2, with solid lines indicating direct paths and dash lines representing mediating effects. Full results are presented in Appendix Table 2.

Fig. 2.

The path analysis structure. Solid lines denote direct paths that exist from ethnicity to subjective wellbeing and cardiovascular disease. The dashed line denotes the demographic factors being mediated by ethnicity; *P < 0.01, **P < 0.001

Model performance and diagnostics

A detailed evaluation of model performance, including accuracy and area under the curve, are available in the accompanying paper, Patel et al. 2024 [45].

Discussion

General

The present study revealed that ethnicity, alongside other factors, plays a pivotal in shaping the complex relationship between CVD and subjective wellbeing. To the authors’ knowledge, this research represents the first attempt to apply path and machine learning analyses to UKB data for investigating this association.

In this sample of UKB participants, non-White ethnic groups were significantly associated with higher CVD odds and lower SWB compared to the White population. This finding aligns with studies that have shown higher cardiovascular risks among South Asian and Black populations, underscoring the interplay of genetic, environmental, and socio-economic factors in influencing health outcomes [19, 47, 48]. Previous literature have found that South Asians, in particular, were associated with higher CVD odds [49, 50], with a few finding this association in the Black group [51] when adjusting for only age or sex, or without any adjustments.

Past research have found significant associations between ethnicity and SWB in countries such as South Africa [16] and Australia [15]. However, a study by Omosehin et al. [52] found no such difference in SWB between ethnicities when studying 316 working adults, opting rather for the implementation of objective wellbeing measures instead of subjective ones. Despite the acknowledged important of ethnicity in shaping SWB, there is a notable absence of studies examining this relationship in the UK context after adjusted for multiple confounding factors. The contrasting findings of Omosehin et al. highlight the need to consider the methodological differences—such as sample size, subjective versus objective wellbeing measures, and cultural settings—when interpreting ethnic disparities in SWB. Moreover, no previous study has established a comprehensive conceptual path model to elucidate how CVD and SWB are intricately linked with ethnicity.

Indirect paths, mediated by factors such as age, unravelled intricate relationships, offering insights into the nuanced dynamics in the model. Furthermore, machine learning analyses highlighted the notable variations in predictive performance across distinct ethnic groups. Specifically, predictors such as waist circumference, handgrip strength, and depression consistently demonstrated their importance in determining CVD and SWB outcomes. These results are consistent with findings from other large datasets, such as the Framingham Heart Study, which emphasise the predictive power of anthropometric and mental health factors [53, 54].

The findings of the CVD model were broadly consistent with previous research in terms of age, gender, smoking status, and depression. This study reveals a significant association between higher waist circumferences and increased odds of CVD, indicating a potential link between abdominal adiposity and adverse health outcomes. Specifically, each 1 cm increase in waist circumference is associated with elevated CVD odds by 1%. Conversely, our findings also highlight the significance of handgrip strength as a positive indicator of physical health. Higher handgrip strength is associated with a 1% decrease in the odds of CVD, suggesting a potential protective effect. These findings align with those from meta-analyses [55] emphasizing the role of muscular strength as a protective factor against CVD-related mortality.

This dual observation underscores the complex interplay between anthropometric measures and physical health outcomes, emphasising the need for a comprehensive understanding of multiple indicators when assessing cardiovascular risk. Moderate physical activity was not associated with CVD in this analysis, contrary to previous studies that found even 30 min of moderate physical activity for most days of the week is associated with decreases in CVD risk and mortality [56, 57]. Several factors may have influenced the non-association observed in our study. Firstly, variations in the intensity, duration, and type of physical activity might play a role in the discrepancies between these findings and previous research. It is possible that other studies included a broader spectrum of physical activity levels or different types of activities that may impact cardiovascular health differently. Additionally, individual differences in participants’ baseline health status, genetic predispositions, or other unmeasured confounding variables could contribute to the lack of association. The diverse nature of our study population, encompassing different ethnicities and socio-economic backgrounds, further adds complexity to the interpretation of these results. Moreover, self-reported data on physical activity may introduce recall bias and influence the accuracy of the reported frequency and intensity of activities. Participants may have overestimated or underestimated their levels of physical activity, potentially affecting the observed association with CVD.

Alcohol intake was also associated with CVD; however, the results of this study suggest that more frequent alcohol intake is protective of CVD. This result adds a layer of complexity to the ongoing debate on the relationship between alcohol consumption and cardiovascular health. While some studies have argued against any protective effects of alcohol on cardiovascular health [58], others, in line with the present study provide support for an association between light to moderate alcohol consumption and a lower risk of CVD [59]. Reasons for this result may be related to the accuracy and reliability of the measures used for alcohol intake. For example, the number of units or type of alcohol consumed were not recorded. It could also be due to reverse causation, with individuals with better subjective wellbeing engaging in behaviours such as increased alcohol intake. The observed protective effect of more frequent alcohol intake against CVD in this study raises questions about the potential mechanisms underlying this association.

Strengths and limitations

The sample size provided by the UKB emerges as a notable strength, offering a robust foundation for analyses. The extensive collection of important factors enhances the richness of information and allows for a better understanding of complex relationships among variables. However, it is important to acknowledge the potential limitations associated with the sizes of specific ethnic groups within the dataset. While the overall sample size is extensive with over half a million participants, certain ethnic groups, such as East Asians, exhibit smaller representation within the UKB, thus caution is advised when extrapolating results to these smaller ethnic cohorts. The results reflect the characteristics of the ‘well’ UK population, which may pose limitations when extending findings to populations outside of the UK or those who are unwell, or with different genetic and environmental profiles.

While the findings indicate that CVD and ethnicity were among the least influential predictors of SWB across all ethnicities (Fig. 1; right panel), this does not necessarily imply that disease burden lacks relevance for subjective wellbeing. Rather, the limited role of CVD in our model may reflect the specific set of health conditions included as covariates in the model. Future research incorporating a more comprehensive set of disease indicators, other CVD types or other conditions entirely, may provide a fuller understanding on the dynamic between SWB, CVD, and ethnicity.

The categorisation of ethnic groups, which align with the UK census, may oversimplify the diverse ethnic groups within each category. For example, the Black ethnic group which includes Africans and Caribbeans. Future studies could benefit from more granular ethnic classifications to capture the heterogeneity within each broad category, but this runs the risk of reducing the sample size in each sub-category.

The cross-sectional design of the study is also a limitation as it restricts the ability of the analyses to establish causation, instead emphasising the correlation between variables. On a positive note, the incorporation of stratified analyses emerges as a strength to the approach taken in this paper. This method enables more detailed exploration of how a range of factors may differently impact distinct subgroups within the population, providing a more comprehensive understanding of the complex interplay between variables. Moreover, the study’s strength also lies in the real-world applicability, particularly concerning public health interventions and precision medicine.

Additionally, personality traits are known to influence SWB [60]. However, they were not controlled for in this analysis due to unavailability in the UKB data used. This omission may limit the ability of the model to isolate the effects of psychological and social variables.

Implications and future research

The findings of this study underscore the important of targeted public health interventions to address the ethnic disparities in CVD and SWB. The UK government can leverage these insights to design culturally tailored healthcare programs that acknowledge the unique challenges faced by specific ethnic groups. For instance, health promotion campaigns focusing on dietary modifications and physical activity could be designed to resonate with the cultural practices of South Asian and Black communities. Tailored strategies that consider the unique cultural, social, and lifestyle factors within specific ethnic groups can be instrumental in reducing health inequalities. These initiatives should prioritise culturally sensitive interventions that resonate with different communities. An example would be addressing the hesitation of certain ethnic groups in participating in health research [61, 62]. This would involve engaging communities directly in the development and implementation of interventions.

Healthcare practitioners should integrate demographic and lifestyle factors into routine CVD and wellbeing evaluations. For example, integrating routine waist circumference and handgrip strength assessments into clinical practice could improve early identification of individuals at risk for adverse cardiovascular outcomes. Integrating mental health assessments and interventions into cardiovascular care can contribute to a more comprehensive and patient-centred approach. Clinical settings play a pivotal role in promotion health equity. Thus, ensuring equitable access to healthcare services with culturally competent care, and addressing social determinants of health within clinical practices can contribute to reducing physical and mental health disparities.

The scientific implications of this study also extend to policy. By understanding the complex interplay between ethnicity, CVD, and SWB, the UK government could prioritize funding for community-level interventions that address the socio-economic drivers of health disparities. This includes promoting better access to nutritious foods, safe spaces for physical activity, and mental health resources in underserved communities.

Future research should delve into ethnic-specific analysis, recognising the diversity within non-White groups. A more detailed examination of cultural, socio-economic, and lifestyle, nonclinical factors, within each group could uncover unique patters influencing CVD and SWB. Moreover, exploring the effectiveness of targeted intervention/policies aimed at mitigating disparities identified in this study. Accompanying quantitative analyses with qualitative methodologies, such as in-depth interviews or focus groups can offer rich insights into the lived experiences of individuals within different ethnic groups, illuminating cultural nuances and contextual factors that quantitative models alone would not capture. Moreover, due to the abundance of cardiovascular risk prediction models in existing literature [63] the next step is to compare their accuracy and validate more extensive larger datasets to identify gaps in the models and their overall performance in different populations.

Table 1 reported the distribution of CVD subtypes across ethnic groups, acknowledging the diversity within the CVD category. While the primary analysis focused on CVD as a composite measure, distinct CVD subtypes may be influenced by different risk factors. Further research disentangling these pathways could offer more tailored insights for prevention and intervention strategies.

Additionally, longitudinal studies could offer insights into the causal relationships between variables, while machine learning algorithms could be used to refine predictive models for ethnic-specific risk factors. This could provide policymakers and healthcare providers with actionable tools to target interventions more effectively.

Conclusion

Utilising path and machine learning analyses on UK Biobank data, this study unveils ethnic-specific nuances in the complex relationship between CVD and subjective wellbeing, stressing the need for tailored interventions addressing ethnic-specific health disparities. The findings highlight significantly higher CVD odds and lower subjective wellbeing in non-White ethnic groups, and the most important features that are protective of negative outcomes, filling a notable gap in the literature adjusting for multiple confounders. Anthropometric measures, physical activity, handgrip strength and alcohol intake emerged as influential factors. The study underscores the importance of nuanced, culturally sensitive public health interventions and healthcare practices, acknowledging the diversity within non-White ethnic groups. Future research should delve into detailed ethnic-specific analyses, explore targeted interventions, and implement longitudinal studies to elucidate causal pathways. Integrating qualitative methodologies can offer richer insights into the lived experiences shaping health outcomes and well-being among ethnic groups. This comprehensive approach contributes to a more nuanced understanding of the intricate relationships among demographic factors, health outcomes, and well-being.

Abbreviations

BMI

Body mass index

CFI

Comparative fit index

CVD

Cardiovascular disease

IQR

Interquartile range

PPI

Patient and public involvement

RMSEA

Root means square error of approximation

SEM

Structural equation modelling

SWB

Subjective wellbeing

UKB

UK Biobank

VIF

Variance inflation factor

Appendix

Table 2.

Results of the path analysis which confirm the significant associations in the model structure

	Estimate	Standard error	P-value
Direct paths
Ethnicity -> SWB	−0.586	0.051	< 0.001
Ethnicity -> CVD status	−0.007	0.003	0.006
Age -> SWB	0.120	0.003	< 0.001
BMI -> SWB	0.035	0.006	< 0.001
Waist circumference -> SWB	−0.068	0.002	< 0.001
Moderate PA -> SWB	0.211	0.012	< 0.001
Dominant handgrip -> SWB	0.056	0.002	< 0.001
Smoking status -> SWB	−0.992	0.042	< 0.001
Alcohol intake -> SWB	0.307	0.018	< 0.001
Depression 6 m -> SWB	−7.902	0.077	< 0.001
Sex -> SWB	−0.285	0.053	< 0.001
CVD status -> SWB	−1.008	0.053	< 0.001
Indirect paths (to SWB via ethnicity)
Age	−0.002	0.000	< 0.001
BMI	0.000	0.000	< 0.001
Waist circumference	0.001	0.000	< 0.001
Moderate PA	−0.003	0.000	< 0.001
Dominant handgrip	−0.001	0.000	< 0.001
Smoking status	0.013	0.001	< 0.001
Alcohol intake	−0.004	0.000	< 0.001
Depression in past 6 months	0.103	0.010	< 0.001
Sex	0.004	0.001	< 0.001
CVD status	−1.008	0.053	< 0.001
Total indirect effect	−0.898	0.054	< 0.001

Abbreviations: BMI Body mass index, CVD Cardiovascular disease, PA Physical activity, SWB Subjective wellbeing

Fig. 3.

Feature importance for cardiovascular disease predictive models separated by ethnicity

Fig. 4.

Feature importance for subjective wellbeing predictive models separated by ethnicity

Authors’ contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Mubarak Patel and Mohammed A Buchya. The first draft of the manuscript was written by Mubarak Patel and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

Open access fees were covered by financial support from the University of Warwick Research Development Fund.

Data availability

The datasets generated and analysed during the current study are available in the UK Biobank repository which can be accessed at: https://www.ukbiobank.ac.uk/.

Declarations

Ethics approval and consent to participant

This research has been conducted using the UK Biobank Resource under the application number 103665. All methods were carried out in accordance with relevant guidelines and regulations, including the Declaration of Helsinki. Ethical approval was obtained by the UK Biobank from the North West Multi-centre Research Ethics Committee (MREC).

All participants provided written informed consent to participate in the UK Biobank study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.