If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Nutrition, Harvard T.H. Chan School of Public Health, Boston, MA, USAChanning Division of Network Medicine, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, MA, USA
Multiple psychological wellbeing factors are associated with risk of cardiovascular diseases (CVD).
Psychological wellbeing score could be used in large populations as a combination indicator.
Lower psychologic wellbeing score is associated with increased CVD risk, regardless of genetic risk.
In CVD events, women are more vulnerable to poor psychological status than men.
Background and aims
Psychologic wellbeing can impact cardiovascular health. We aimed to evaluate the joint association of multiple psychologic wellbeing factors with cardiovascular diseases (CVD) and examine whether this association was modified by genetic susceptibility.
In the UK Biobank, 126,255 participants free of CVD (coronary heart disease [CHD], stroke, and heart failure [HF]) at baseline, who completed a questionnaire on psychological factors, were included. The psychological wellbeing score was calculated by four factors: happiness, life satisfaction, broad depression, and neuroticism. Cox proportional hazard models were used to assess the association between the psychological wellbeing score and CVD risk.
During the median follow-up of 11.5 years, 10,815 participants had newly diagnosed CVDs. Low life satisfaction, the presence of depression, and neuroticism score ≥1 were significantly associated with an increased risk of CVD in the multivariable-adjusted model. Through decreasing the psychological wellbeing score, there were significant increasing linear trends in the risk of CVD, CHD, stroke, and HF (all p for trend < 0.001). Participants with the lowest psychological wellbeing score had the highest risk for CVD (HR 1.51, 95% CI 1.42–1.61). Women were more susceptible to worse psychological wellbeing status for CVD than men (p for interaction = 0.009). The associations of the psychological wellbeing score with CVD were consistent across genetic risk (p for interaction >0.05). When considered jointly, participants exposed to high-risk psychological wellbeing and genetic status had a 2.70-fold (95% CI 2.25–3.24) risk for CHD.
Joint exposure to multiple psychological wellbeing factors was associated with increased risks of incident CVD in an additive manner, regardless of genetic susceptibility.
]. Although traditional preventive strategies have been used to delay the onset and progression of CVD, including lifestyle interventions and blood pressure control, it still has the highest burden of disease in the world. Classical risk factors such as hypertension, diabetes, smoking, and high alcohol consumption may only explain approximately 60% of the variance in CVD risk [
For a long time, more attention has been given to specific physical diseases. However, psychological health has also been found to probably contribute to the development of various diseases and social outcomes [
]. A pooled analysis with more than 560,000 participants indicated that baseline depressive symptoms were associated with poor cardiovascular conditions, even after adjustment for several cardiovascular risk factors [
]. However, previous studies were often cross-sectional, assessed only a single psychological factor, and rarely considered the complexity and correlation of multiple factors. Psychologic wellbeing factors may be correlated and impact human health and quality of life in a joint manner [
]. Thus, such studies must incorporate large samples, longitudinal designs, multiple aspects of wellbeing, adequate control for various confounders, and a broad scope of outcomes.
Notably, that genetic susceptibility plays a key role in the development of CVD. However, little is known about the potential interaction between genetic risk and psychological wellbeing in adulthood on CVD risk.
In this prospective cohort from the UK Biobank, we aimed to analyze the associations between psychological wellbeing and CVD and its three sub-outcomes (coronary heart disease [CHD], stroke, and heart failure [HF]) using a combination indicator that incorporated four facets: happiness, life satisfaction, broad depression, and neuroticism. We also examined whether these associations would be modified by genetic predisposition.
2. Patients and methods
2.1 Study design and sample
The UK Biobank (UKB) is a population-based prospective cohort study, and detailed information has been described in a previous study [
]. Briefly, UKB included more than 500,000 community-dwelling adults aged 37–73 years across the UK between 2006 and 2010 (https://www.ukbiobank.ac.uk/). Information on lifestyle, medical history, physical examinations, and biological samples were collected. We declare that all data are publicly available in the UKB repository [
]. The North West Multi-Center Research Ethics Committee Study approved the UKB study, the study protocol conforms to the ethical guidelines of the 1975 Declaration of Helsinki, and all participants provided written informed consent.
A total of 502,505 participants were recruited in UKB. Participants who met the following criteria were excluded from the present study: 1) had any history of CHD, stroke, or HF at baseline (n = 34,261); and 2) withdrew from follow-up (n = 44). Among the remaining participants, 159,631 had general happiness data, 155,496 had life satisfaction data, 463,619 had broad depression data, and 374,946 had neuroticism score data. The inclusion criteria for those in the main analyses were information on four psychological wellbeing factors (n = 126,255), and a subset with UKB genetic data and no kinship to other participants was further used in genetic analysis (n = 87,968) (Supplementary Fig. S1).
In UK Biobank, available mental health questionnaires included several subsets. As suggested in previous studies [
], four dimensions of psychological wellbeing at baseline have summarized that cover and reflect the main contents of this questionnaire. They were “general happiness”, “satisfaction with health, family, friendship, and financial situation”, “broad depression”, and “neuroticism score”, reflecting the tendency to experience negative emotions that included 12 subsets. Detailed questions and answers about psychological wellbeing are presented in the supplemental file (Supplementary Data). Being collected using a touchscreen questionnaire at the baseline visit, the answers for general happiness were graded from “extremely happy”, “very happy”, “moderately happy”, “moderately unhappy”, “very unhappy” to “extremely unhappy”. The satisfaction score consisted of four questions: health, family relationships, friendships, and financial situation satisfaction. The answers included “not at all”, “a little”, “a moderate amount”, “very much”, and “an extreme amount”. The presence of broad depression was determined by a previous history of medical visits for psychiatric symptoms, which has been widely utilized in the literature on depression in the UK Biobank. This phenotype has been considered to be the most genetically trackable in the database [
]. Neuroticism was measured according to 12 questions related to neurotic symptoms. The number of answers “yes” formed the neuroticism score ranging from 0 to 12.
To generate a psychological wellbeing score, we included the above four factors (general happiness, life satisfaction, broad depression, and neuroticism). According to the association between each multinomial/binomial factor and CVD in the largest sample available (Supplementary Tables S1 and S2), those with significant HRs were combined into one group. Thus, high-risk psychological wellbeing factors were defined as follows: general happiness (moderately happy or moderately to extremely unhappy), life satisfaction (satisfaction score ≤3), presence of broad depression (yes), and neuroticism (neuroticism score ≥1). For each factor, the participant received a score of 1 if he or she was classified as low risk or 0 if he or she was classified as high risk. Then, we summed the four scores (0–4 points) to represent overall psychological wellbeing status, with a higher score indicating a healthier score. We further divided the overall psychological wellbeing status into low risk (score 3–4), intermediate risk (score 1–2) and high risk (score 0). In the sensitivity analyses, weighted scores were also formed by adding four scores weighted by the multivariable-adjusted risk estimates (β coefficients) in CHD, stroke, or HF risk. The equation was (β1*factor1+., β4*factor4) * (4/β1+ … +β4). This weighted score reflects the magnitudes of the adjusted risk for each psychological wellbeing factor, and a higher score indicates exposure to healthier psychological wellbeing.
The outcome, incident CHD, stroke, and HF (ICD 10 code I20–I25 for CHD, I60–I64, and I69 for stroke, and I50 for heart failure), was extracted from the “first occurrence of health outcomes defined by a 3-character International Statistical Classification of Diseases and Related Health Problems 10th Revision code” (category ID in UKB 1712). The diagnosis of CHD, stroke, and HF was obtained by using linkage with register of dead people, primary care, and hospital inpatient records. Detailed information regarding the linkage procedure is available online (https://biobank.ctsu.ox.ac.uk/crystal/exinfo.cgi?src=diag_xtabs_HES).
2.4 Genetic risk
Detailed genotyping information in UK Biobank was reported previously [
]. The weighted genetic risk score (GRS) was created for CHD, stroke, and HF using SNPs associated with CHD, stroke and HF at the genome-wide association significance in genome-wide association studies that do not include UK Biobank participants [
]. Information on the 74 SNPs selected for CHD, 28 SNPs for stroke, and 12 SNPs for HF is listed in Supplementary Table S3. Individual SNPs were coded as 0, 1, and 2 according to the number of risk alleles. The regression coefficient for each SNP was taken from the above genome-wide association studies. GRS was calculated as (β1*SNP1 + β2*SNP2 + … βN*SNPN) * (N/sum of the β-coefficients) [
]. We determined whether participants were at high (highest quartile), intermediate (mid-two quartiles), or low (lowest quartile) genetic risk for CHD, stroke, and HF.
Three main groups of potential confounders were considered in the analyses: (1) demographic variables: age, sex, ethnicity (White/others), education (university or college degree/others), the Townsend index reflecting socioeconomic status (continuous); (2) lifestyle factors: smoking status (current, ever, never), drinking status (drinks, continuous), physical activity at goal or not (≥150 min/week of moderate intensity, or ≥75 min/week of vigorous intensity, or an equivalent combination), healthy diet score (score ≥4, detailed information was in supplemental file [
]); (3) cardiometabolic factors: body mass index (BMI) (continuous), total cholesterol (continuous), systolic blood pressure (continuous), diabetes (yes/no), use of antipsychotic and sleep medications (yes/no), blood pressure-lowering medications (yes/no), cholesterol-lowering medications (yes/no), sleep duration (continuous), chronotype (morning/evening) and C-reactive protein (CRP). A diet score was calculated based on the following factors: vegetable intake ≥4 tablespoons each day (median); fresh fruit intake ≥2 pieces each day (median); oily and nonoily fish intake at least twice each week (median); unprocessed red meat intake no more than twice each week (median); and processed meat intake no more than twice each week (median); each favorable diet factor received one point, with a total score ranging from 0 to 5 21. Antipsychotic medications or sleep medications include medications for schizophrenia, such as clozapine and risperidone, antidepressant medications, such as selective serotonin reuptake inhibitors (SSRIs), and tranquillizers. If the covariate information was missing, we imputed the mean (normally distributed) or median values (nonnormally distributed) for continuous variables or used a missing-indicator approach for categorical variables.
2.6 Statistical analyses
Data analyses were performed using IBM SPSS Statistics, Version 25 (IBM Corporation, Armonk, NY, USA). A p value < 0.05 indicated statistical significance (two-sided). Baseline characteristics of the study population are reported as the means or percentages by psychological wellbeing score. The follow-up time was determined from the baseline date (date of attending the assessment center) to the diagnosis of CHD, stroke and HF, death, or censoring date (July 30, 2021), whichever came first.
The Cox proportional hazards model was used to estimate the hazard ratio (HR) and 95% confidence interval (CI). The proportional hazards assumption was tested using Schoenfeld residuals. Model 1 was adjusted for demographic variables, including age, sex, ethnicity (White/others), education (university or college degree/others), and the Townsend index. Model 2 was adjusted for terms in Model 1 and lifestyle factors, including smoking status (current, ever, never), drinking, physical activity at goal (yes/no), and a healthy diet score (≥4). Model 3 was adjusted for terms in Model 2, body mass index (continuous), total cholesterol (continuous), systolic blood pressure (continuous), and diabetes (yes/no).
We then classified participants according to the joint categories of psychological wellbeing score and genetic risk for CHD, stroke, and HF (low, intermediate, and high). Using participants with psychological wellbeing scores of 3–4 (healthiest group) and low genetic risk as a reference, HRs of CHD, stroke, and HF was obtained in the left categories. The interaction analysis was performed by using the likelihood ratio test comparing models with and without a cross-product term.
In preplanned secondary analyses, we also examined the association between psychological wellbeing and CVD in subgroups of sex, age, smoking, diet, physical activity, weight status, and receiving antipsychotics and sleep medications to look for clues to interaction.
We conducted several sensitivity analyses to minimize bias. We further adjusted for several medications usage, sleep duration (continuous), chronotype (morning/evening), C-reactive protein, and genetic risk scores. We restricted subjects with follow-up time ≥1 year, ≥ 2 years, and ≥5 years to perform the regression. Moreover, participants with work satisfaction data were added, and weighted psychological wellbeing scores instead of unweighted scores were used in the analyses. Finally, the time-varying Cox regression model with age as the time scale was also used to consider the variability in patient evaluation.
Table 1 presents the baseline characteristics of the participants according to their psychological wellbeing scores. Among a total of 126,255 participants, 44.9% were men, and the mean age was 56.4 ± 8.1 years. Overall, individuals with higher psychological wellbeing scores were relatively older, more likely to be men, had higher education levels, and had better economic status than those with lower scores. Less antipsychotic medication use and a higher prevalence of healthy diet and physical activity at the goal were also observed among participants with higher psychological wellbeing scores. Regarding the baseline characteristics, there seemed to be little difference among participants with or without psychological wellbeing information and genetic data (Supplementary Table S4). Most of these factors were relatively stable or slightly changed during the follow-up years (first follow-up 2012–2013, n = 3695; second follow-up 2014+, n = 13,224). They were prone to have better lifestyles and lower BMI, but a higher prevalence of diabetes, dyslipidemia, and hypertension that was possibly due to increasing age (Supplementary Table S5).
Table 1Baseline characteristics of 126,255 participants according to psychological wellbeing score.
Psychological wellbeing score
55.3 ± 7.9
55.9 ± 8.2
56.8 ± 8.2
57.7 ± 8.0
University or college degree, %
Townsend deprivation index
−0.9 ± 3.0
−1.1 ± 2.9
−1.4 ± 2.8
−1.7 ± 2.7
Smoking status, %
Drinks per week
7.4 ± 10.4
7.9 ± 9.9
8.2 ± 9.7
8.5 ± 9.3
Physical activity at goal, %
Healthy diet, %
Body mass index, kg/m2
27.7 ± 5.2
27.3 ± 4.7
27.3 ± 4.6
26.8 ± 4.2
Systolic blood pressure, mmHg
137 ± 19
140 ± 19
141 ± 20
142 ± 20
Total cholesterol, mmol/L
5.8 ± 1.1
5.7 ± 1.1
5.8 ± 1.1
5.8 ± 1.1
Blood pressure-lowering medications, %
Cholesterol-lowering medications, %
Antipsychotic and sleep medication, %
Mean ± SD or median (interquartile range) for continuous variables and number (percentage) for categorical variables.
During the median follow-up of 11.5 years (approximately 1.4 million person-years), 10,815 incident cases of CVD were documented, which included 8010 CHD cases, 2168 stroke cases, and 2295 HF cases. Four factors were categorized into binary factors of low risk (reference group) vs. high risk (Supplementary Tables S1 and S2). When the four factors were mutually adjusted in the multivariable model, age, sex, ethnicity, education, Townsend index, smoking status, drinking, physical activity at goal, healthy diet score ≥4, BMI, total cholesterol, systolic blood pressure, and diabetes, life satisfaction score ≤3 (HR 1.25, 95% CI 1.18–1.33), presence of broad depression (HR 1.21, 1.16–1.26), and neuroticism score ≥1 (HR 1.08, 1.02–1.14) were significantly associated with incident CVD (Table 2).
Table 2Multivariable-adjusted hazard ratios for CVD, CHD, stroke, HF by four psychological wellbeing factors (N = 126,255).
Fig. 1 shows the associations between the psychological wellbeing score and the CVD outcomes, and Supplementary Fig. S2 presents the cumulative incidence of CVD events according to the psychological wellbeing score. We found that the psychological wellbeing score was significantly associated with a higher risk of incident CVD in the multivariable-adjusted model in a dose–response fashion. The HRs (95% CI) of CVD were 1.14 (1.07–1.12), 1.23 (1.17–1.31), and 1.51 (1.42–1.61) for psychological wellbeing scores of 2 points and 1 point to 0 points, respectively, compared with 3–4 points (p for trend<0.001). The results were similar for the risk of incident CHD, stroke, and HF (all p for trend < 0.001).
Associations of GRSs with the risk of CVD were measured. CHD-GRS, stroke-GRS and HF-GRS (per 1 standard deviation increment) were associated with 24%, 10%, and 15% higher risks of incident CHD, stroke, and HF, respectively (Supplementary Table S6). The joint associations of psychological wellbeing and GRS with CVD are shown in Fig. 2. Participants with high GRS and high-risk psychological wellbeing status had the highest risk of CHD (HR 2.70, 2.25–3.24), stroke (HR 1.23, 0.89–1.70), and HF (HR 1.67, 1.22–2.28). However, no statistically significant interaction was observed (p for interaction 0.248, 0.934 and 0.401, respectively).
We further performed a stratified analysis by sex, age, smoking, diet, physical activity, BMI, and use of blood pressure-lowering and cholesterol-lowering medications. In the stratified analysis, a significant interaction with sex was found (Supplementary Table S7). Women (psychological wellbeing score 0 points, HR 1.65, 1.48–1.83) seemed to be more susceptible to psychological wellbeing than men (psychological wellbeing score 0 points, HR 1.43, 1.32–1.55) for CVD (p for interaction = 0.009).
In sensitivity analyses, further adjusting the usage of several medications, sleep duration, chronotype, CRP, and genetic risk score did not materially change the results (Supplementary Tables S8 and S9). The results were relatively stable when restricting participants with incident CVD to ≥1 year, ≥ 2 years, and ≥5 years from baseline or adding work satisfaction data into the life satisfaction factor (Supplementary Tables S10–S12). Whether the weighted psychological wellbeing score or time-varying Cox regression was used, the results were still robust (Supplementary Tables S13 and S14).
In this large-scale cohort with an approximately 11.5-year follow-up time, we found that positive affect, life satisfaction, broad depression, and neuroticism were differentially associated with CVD. A simple-to-use psychological wellbeing score was newly created, incorporating the four facets. We found that the psychological wellbeing score was significantly associated with an increased risk of incident CVD, CHD, stroke, and HF in a dose–response fashion, independent of various traditional risk factors. Moreover, these associations held regardless of low, intermediate, or high genetic risk.
There is a term called the “mind-heart-body connection”, which means that the mind could have a negative or positive impact on human health [
]. In addition, longitudinal data for Midlife in the United States (MIDUS) revealed that positive affect and life satisfaction predicted a reduced risk for incident cardiometabolic Conditions 8–11 years later. Still, only life satisfaction was statistically significant when accounting for depressive symptoms [
]; therefore, joining viewing psychologic wellbeing factors could be a more proper. Regarding negative aspects (depressive symptoms), findings from a pooled analysis with 563,255 participants in 22 prospective cohorts indicated that baseline depressive symptoms were significantly associated with incident CVDs. The HR per 1-SD higher depression score for CHD was 1.07 (1.03–1.11); stroke, 1.05 (1.01–1.10); and CVD, 1.06 (1.04–1.08) [
]. Different psychological wellbeing factors were considered in previous studies. Still these results were not entirely consistent, which may be due to the varying sample size, different subjective questionnaires, confounders selected, outcomes, and the possible potential interactions between exposure factors. Based on the above evidence and thinking, we performed a long-term large cohort study to comprehensively assess the association between exposure to various psychological wellbeing factors, including happiness, life satisfaction, broad depression, neuroticism, and CVD events. We found that a low psychological wellbeing score, which indicated poor psychological status, had the strongest association with the risk of incident CHD, followed by CVD, stroke, and HF.
To the best of our knowledge, using a synthesized indicator, this is the first large-scale prospective cohort study to measure the association between joint exposure to multiple psychological wellbeing factors and the risk of CVD events. Due to the interccorrelation and overlap of some psychological factors in health research, it has been realized to consider the contribution of the different psychological factors together [
]. Compared with individual factors, a relatively stronger association was observed between our newly developed psychological wellbeing score and the risk of CVDs. The joint effect of multiple factors may reflect a more comprehensive measure of psychological wellbeing status, and a similar psychological score has been applied to assess the risk of type 2 diabetes [
]. Correlation analyses between the evaluations at baseline and follow-up using the same mental health questionnaire showed that psychological wellbeing was relatively stable (Supplementary Tables S15 and S16, and Supplementary Fig. S3). Furthermore, similar approaches and algorithms have been clearly used to assess joint exposure to lifestyle [
], which seems to be a comparatively mature statistical method. From the public health perspective, a simple score algorithm is also easier for epidemiological findings to be understood, translated into practice, and informed to the general population.
The interactions between psychological wellbeing and various factors on CVD events were explored. We found that the above association was statistically strengthened in women. Women were more likely to have depression or anxiety symptoms than men in UKB [
]. Moreover, no study has measured the interaction between psychological wellbeing and genetic susceptibility. The nonsignificant interaction may be partially due to the small proportion of CVD risk explained by the variants [
]. We observed that a low-risk psychological wellbeing status partly offset the CVD risk brought by genetic predisposition. Conversely, individuals with low genetic risk could lose their inherent protection if they had a high-risk psychological wellbeing status. Therefore, having a healthy psychological wellbeing status (being happy, satisfied with life, free of depression and neuroticism) may have an important position in the primary prevention of CVD among the whole population regardless of their genetic risk.
There are plausible mechanisms underlying the probable effects of psychological wellbeing on CVD across the lifespan. First, studies have associated various components of psychological wellbeing with biological processes underlying the observed link with cardiometabolic metrics such as blood pressure [
] have received attention in these associations. Second, numerous studies have found that better psychological wellbeing is associated with healthier behavior-related factors, including nonsmoking, ideal physical activity, healthy diet, and ideal weight status, which has also been presented in our studies. Third, psychosocial mechanisms matter. Having high levels of psychological wellbeing is prone to larger and better social networks when facing difficulties and helps buffer against the harmful effects of stress [
]. Greater psychological wellbeing is also associated with micro-level factors, including higher income, education, and occupational status, and macro-level factors, such as higher gross domestic product per capita, freedom to make life choices, and lower income inequality [
Compared to most previous studies in this field, this study has some strengths, including large sample size (over 120,000 participants), a relatively long follow-up duration (approximately 11.5 years), multiple aspects of psychological wellbeing measured by a simple-to-use questionnaire that are likely to be more practicable and feasible in daily work, various confounders adjusted, and a broad scope of three cardiovascular conditions. Another novelty of our study is that we constructed a psychological wellbeing score to determine the joint effect of multiple psychological wellbeing factors, and its interaction with genetic risk was further performed.
Some limitations also remain in our study. First, we cannot demonstrate the causal relationship between incident CVD and psychological wellbeing factors due to the observational nature. Future trials must reveal whether psychological well-being-promoting programs could improve CVD outcomes. Second, because the exposures tested in this study are subjective terms and recorded by self-report, individual perceptions of different levels of exposure may lead to classification errors. However, as a previous study mentioned [
Comparison of risk factor associations in UK Biobank against representative, general population based studies with conventional response rates: prospective cohort study and individual participant meta-analysis.
], the design of the prospective study suggests that this bias may be random in terms of the outcome, leading to a lower effect estimate and thus an underestimation of the true association. Third, although we carefully adjusted for various major confounders, bias from unknown and unmeasured confounding may still exist. Changes in covariates (treatments, new medical conditions, and different socioeconomic and psychological statuses) may occur over the follow-up period and could not be controlled due to the study design. Fourth, psychological wellbeing information was assessed at baseline only, so this information might change over time during follow-up. Fifth, cases without cardiovascular disease may be missed because of dependency on medical records. Cases with depression or limited access to medical services may be underdiagnosed. Finally, this cohort mainly included individuals of European ancestry, mostly White British, and the generalizability of our findings is not secured to other ethnicities. Therefore, caution should be taken when generalizing summary statistics to the general population.
In this prospective cohort study, for the first time, we found that exposure to unhealthy psychologic was associated with an increased risk of incident CVD events in a dose–response fashion, independent of traditional confounding factors. These associations were not significantly modified by genetic susceptibility. Our findings highlight the importance of considering the potential additive effects of different psychological wellbeing factors on CVD risk in future rigorous intervention trials.
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
This research has been conducted using the UK Biobank Resource under Application Number 77740. This study was supported by Shanghai Ninth People's Hospital ( YBKA201909 ), Shanghai Municipal Human Resources and Social Security Bureau ( 2020074 ), Clinical Research Plan of SHDC ( SHDC2020CR4006 ). All the funders played no role in the design or conduction of the study; in the collection, management, analysis, or interpretation of the data; or in the preparation, review, or approval of the article.
Appendix A. Supplementary data
The following is the Supplementary data to this article:
Comparison of risk factor associations in UK Biobank against representative, general population based studies with conventional response rates: prospective cohort study and individual participant meta-analysis.
The global burden of cardiovascular diseases (CVDs) still represents a challenging issue. Global Burden of Disease (GBD) Study 2019 estimated a 2-fold increase in CVDs prevalence and about 50% increase in the overall CVD deaths worldwide from 1990 to 2019 .