Context The goal of US health care reform is to extend access. In England, with a universal access health system, coronary heart disease (CHD) mortality rates have decreased by more than two-fifths in the last decade, but variations in rates between local populations persist.
Objective To identify which features of populations and primary health care explain variations in CHD mortality rates between the 152 primary care trust populations in England.
Design, Setting, and Participants A cross-sectional study in England of all 152 primary care trusts (total registered population, 54.3 million in 2008) using a hierarchical regression model with age-standardized CHD mortality rate as the dependent variable, and population characteristics (index of multiple deprivation, smoking, ethnicity, and registers of individuals with diabetes) and service characteristics (level of provision of primary care services, levels of detected hypertension, pay for performance data) as candidate explanatory variables.
Main Outcome Measures Age-standardized CHD mortality rates in 2006, 2007, and 2008.
Results The mean age-standardized CHD mortality rates per 100 000 European Standard Population were 97.9 (95% confidence interval [CI], 94.9-100.9) in 2006, 93.5 (95% CI, 90.4-96.5) in 2007, and 88.4 (95% CI, 85.7-91.1) in 2008. In all 3 years, 4 population characteristics were significantly positively associated with CHD mortality (index of multiple deprivation, smoking, white ethnicity, and registers of individuals with diabetes), and 1 service characteristic (levels of detected hypertension) was significantly negatively associated with CHD mortality (adjusted r2 = 0.66 in 2006, adjusted r2 = 0.68 in 2007, and adjusted r2 = 0.67 in 2008). Other service characteristics did not contribute significantly to the model.
Conclusion In England, variations in CHD mortality are predominantly explained by population characteristics; however, greater detection of hypertension is associated with lower CHD mortality.
Although mortality from coronary heart disease (CHD) has been steadily decreasing since the 1970s, it is still responsible for 15% of all deaths and nearly half of all circulatory disease deaths in England.1,2 A national policy was launched in 2000 to reduce the CHD mortality rate by two-fifths in those individuals aged younger than 75 years by 2010.2 This goal was achieved nationally, but regional variations in CHD mortality rates persist. In England, populations and primary care services are grouped geographically into primary care trusts (PCTs, 152 trusts in place between 2006 and 2010; available at http://www.erpho.org.uk/Download/Public/14465/1/PCTs%20England%202006.pdf) and, in 2008, the age-standardized CHD mortality between trusts varied between 45.3 and 147.1 per 100 000 European Standard Population.1
The goal of our study was to identify factors that explain variations in CHD mortality between PCTs. Potential causes for variability include characteristics relating to the population and to the service. Population characteristics include demographic, biological, and sociobehavioral factors (eg, deprivation, prevalence of hypertension and diabetes, and rates of smoking and obesity).3-6 The effect of some of these characteristics may be offset by health care interventions (eg, screening for and treatment of hypertension or smoking cessation programs). The consistent delivery of such interventions is influenced by the characteristics of the health care service, such as the level of provision of primary care services,7,8 the organization of primary care teams,9 or performance in terms of the management of patients and achievement of relevant performance targets. Evidence on service characteristics associated with reduced coronary mortality would, therefore, inform the development of improved delivery of health care by trusts, as well as the evolution of health care systems in other countries, including the United States, who are seeking to improve population health.10
A pay for performance scheme (the quality and outcomes framework) was implemented in 2004. This framework introduced substantial financial incentives for the achievement of performance targets by primary care practices (referred to as general practices in England) across a range of clinical and organizational domains.11 The framework includes performance indicators for processes (eg, measurement of blood pressure and serum cholesterol in patients with CHD) and intermediate outcomes (eg, control of blood pressure and cholesterol to specified targets). For asthma and diabetes, the rate of performance improvement accelerated significantly following introduction of the quality and outcomes framework, although this was not the case for CHD.12 Nevertheless, the gap in achievement between the least and most socioeconomically deprived localities13-15 has progressively narrowed across many clinical indicators, such as achieving target blood pressure levels.13,16,17 The quality and outcomes framework may therefore be associated with greater performance equality, but the effect on CHD mortality has not yet been demonstrated.
The quality and outcomes framework provides incentives to general practices to maintain registers of patients diagnosed as having selected chronic conditions that include hypertension and diabetes. The Health Survey for England is an annual population survey designed to measure health and health-related behaviors in England. Recording of diabetes on practice registers indicates a level of detection relatively close to expected prevalence levels, with 3.7% of individuals aged 17 years or older being recorded on general practice registers as having diabetes in 2006 compared with 4.9% of individuals aged 16 years or older reported as having diabetes in the Health Survey for England for the same year.18 In contrast with diabetes, in 2008, only 13.1% of the population were recorded on practice registers as having hypertension, although the Health Survey for England found a prevalence of hypertension of 30.1% among adults.19 The survey revealed that in 14.7% of adults, hypertension was not treated; in 8.7%, hypertension was treated and controlled; and in 6.7%, hypertension was treated but not controlled. Therefore, in England, there is evidence that hypertension is still underdetected based on the numbers of patients recorded on practice registers.
In designing our study, we hypothesized that variations in CHD mortality between PCTs would be explained not only by provision and performance of services, but also by variations in population characteristics.
We conducted a cross-sectional study involving all PCTs in England, in which population and service characteristics were used to explain CHD mortality in a hierarchical regression model. The Leicestershire, Northamptonshire, and Rutland National Health Service Research Ethics Committee was consulted; the committee advised that ethical review for the study was not required.
The study included all 152 PCTs in England. Primary health care is delivered by general practices contracted to the trusts. In 2008, the population registered with PCTs was 54.3 million, the number of individuals registered with each trust ranging between 94 879 and 1 306 879. In 2007 and 2006, respectively, these data were 54.1 million (between 95 006 and 1 294 147) and 53.9 million (between 94 741 and 1 292 151).1
The age-standardized CHD mortality rate for European Standard Population was the dependent variable, because age is an important factor in mortality and there is considerable variation in the age structure of populations in different trusts. We obtained age-standardized rates of deaths with CHD (International Statistical Classification of Diseases, 10th Revision codes I20-I2520) as the main cause for the last 3 available years (2006, 2007, and 2008).
Population and Service Characteristics
Candidate explanatory variables relating to population or service characteristics were identified on the basis of reliable data availability across every trust and of potential relevance to CHD mortality.
Population Characteristics. We obtained population data at PCT level at mid-year in 2006, 2007, and 2008 for numbers of individuals by age and sex (in quinary bands)21; the latest modeled estimate of ethnicity (2007) derived from self-reported, principally closed option, responses in the 2001 national census supplemented by data on births, deaths, and migration in the following years22; and lifestyle (obesity and smoking rates using the most recent data available [2003-2005]).23 We obtained the quality and outcomes framework registers for diabetes and hypertension maintained by general practices (for the latest 3 corresponding business years 2008/2009, 2007/2008, and 2006/2007—the year running between April 1 and March 31). Because hypertension is underdetected,19 the hypertension register is more likely to function as a measure of detection than of prevalence. On the other hand, diabetes is much better detected; therefore, the diabetes register is more likely to reflect prevalence. Because of this difference in detection efficiency between the 2 registers, we used the diabetes register as a population characteristic variable (referred to as “registers of individuals with diabetes”) and the hypertension register as a service level variable (referred to as “levels of detected hypertension”) in our analyses.
Socioeconomic deprivation was included using the 2007 index of multiple deprivation, a UK government statistic that combines a set of indicators (in 7 domains—income, employment, health, education, barriers to housing and services, living environment, and crime) chosen to cover a range of economic, social, and housing issues into a single deprivation score for each small area in England.24 The index is calculated using a formula that weights data from each domain.25 We obtained the index at PCT level from the UK Department for Communities and Local Government.25
Service Characteristics. We obtained data on service characteristics for the years 2006, 2007, and 2008, including the provision of primary care indicated by full-time equivalent general medical practitioners and full-time equivalent practice staff per 100 000 population26 (the supply of primary care physicians has been shown in US populations to be associated with mortality rates7,8). We obtained quality and outcomes framework data on clinical performance for the 3 corresponding business years 2008/2009, 2007/2008, and 2006/2007 (where available) for all 152 English PCTs (in which 8229 practices reported in 2008/2009, 8294 reported in 2007/2008, and 8372 reported in 2006/2007).27
These data included (1) the proportion of maximum points achieved in 3 disease domains: hypertension, CHD, and diabetes. (In the pay for performance scheme, incentivized components of clinical and organizational performance are specified as indicators.17,28 Points are awarded on a sliding scale up to a maximum for achievement in each indicator. The incentive payment to each practice is calculated upon the basis of the number of points achieved, adjusted for the number of patients to whom the indicator applies. The proportion of points achieved provides, therefore, a summary of clinical performance in each disease domain.) (2) In view of their likely association with coronary mortality, an indicator of the proportion of people with hypertension or CHD whose blood pressure was controlled (most recent blood pressure ≤150/90 mm Hg) or cholesterol level was 192.99 mg/dL or lower (≤5.0 mmol/L) in the previous 9 months.28
In a descriptive analysis, we calculated means or medians for quantitative variables depending on their distribution. We then undertook univariable linear regression analyses for candidate predictors of age-standardized CHD mortality rate, calculating the adjusted r2 value and β coefficients with 95% confidence intervals (CIs). Descriptions of the outcome variables used in the analysis and the univariable analyses for 2008 are shown in Table 1.
Next, we undertook a 2-level hierarchical multiple linear regression (not a multilevel model) in which population and service characteristics in each year (or nearest available) were used to explain trust mortality in that year (all 3 outcome variables had a Gaussian distribution). In hierarchical regression, the researcher decides on the order of entry of variables into the multiple regression model. The variables can be split into blocks on the basis of some common attribute, and analysis can then be performed using backward, forward, or stepwise options.29,30 The first level included population characteristics known either to reflect cardiovascular disease risk factors or to be associated with CHD. The second level included the service characteristics in the domains of provision of services and clinical performance.
Because multivariable analysis for 152 PCTs could not support an excessive number of candidate explanatory variables, a selection process was undertaken to identify the second-level variables most relevant to health care and their possible effects on CHD mortality rates. We first selected those variables identified in the univariable analysis that had been significantly associated with CHD mortality. We then eliminated certain highly correlated variables on a priori theoretical grounds of being conceptually less representative of the characteristics in which we were interested. Thus, we eliminated obesity because it was highly correlated with levels of hypertension detected and registers of individuals with diabetes, and with smoking and white ethnicity. Furthermore, obesity is a risk factor for diabetes and hypertension. South Asian ethnicity was eliminated but white ethnicity retained, because this would enable comparison of white ethnicity with the combination of South Asian, black, and other ethnic groups.
The analysis was repeated separately for each of the 3 years. In step 1, the 4 first-level variables (prevalence of smoking, prevalence of diabetes, white ethnicity, and the deprivation index) were forced into the regression and estimates recorded. In step 2, up to 10 candidate second-level variables were added to the regression (data for the 3 intermediate outcome variables were only available in 2008). The 10 variables included provision of primary health care (full-time equivalent practitioners per 100 000 population and full-time equivalent staff per 100 000 population); clinical performance (levels of detected hypertension, and proportion of quality and outcomes framework clinical points achieved in 3 disease domains—hypertension, diabetes, and CHD); intermediate clinical outcome variables potentially associated with coronary mortality (proportion of patients with hypertension whose last blood pressure level was ≤150/90 mm Hg, proportion of patients with CHD whose last blood pressure level was ≤150/90 mm Hg, and proportion of patients with CHD whose last total cholesterol level was ≤192.99 mg/dL). Forward as well as backward stepwise methods were used to remove nonsignificant level 2 predictors from the final model, and the final set of predictors was the same in each case.
One thousand bootstrap samples were generated to calculate robust 95% CIs for the weakest level 1 predictor for each year from each final multivariable model, and the strongest nonsignificant level 2 predictor for each year.
Analyses were conducted using the statistical package SAS version 9.1 (SAS Institute Inc, Cary, North Carolina). P < .05 was considered statistically significant.
Results were available for all 152 PCTs in England.
Coronary heart disease mortality had a normal distribution on visual inspection with a histogram for each year, with low skewness (<0.5) and low kurtosis (<0.3) for each year. The mean age-standardized CHD mortality rates per 100 000 European Standard Population decreased from 97.9 (95% CI, 94.9-100.9) in 2006 to 93.5 (95% CI, 90.4-96.5) in 2007 to 88.4 (95% CI, 85.7-91.1) in 2008 (Table 1). This represents a decrease of approximately 5 per 100 000 population per year over the 3 years. Table 2 shows the CHD mortality rate for 2008 grouping PCTs into quarters (from lowest to highest rates), associated with selected explanatory variables.
High scoring outlier PCTs (>3 SDs from the mean) were identified as Tameside and Glossop in 2008 and Blackburn and Darwen in 2007. The lowest scoring PCT in each year was Kensington and Chelsea, but the rate for this PCT was never more than 3 SDs from the mean.
The proportion of variability (adjusted r2) accounted for in the final models (all level 1 and 2 terms) was substantial and consistent across the years 2006 to 2008, and ranged from 0.66 to 0.68 (Table 3). In the final model for all 3 years, the same 4 level 1 variables (proportion of smokers, index of multiple deprivation, white ethnicity, and registers of individuals with diabetes) were significantly positively associated with CHD mortality (Table 4). In step 2, the level of detected hypertension (a level 2 variable in our hierarchical model) was significantly negatively associated with CHD mortality, but no other service characteristics were significantly associated with mortality or significantly increased the adjusted r2 value.
The β coefficients for the significant predictors of CHD mortality in the final model for 2008 were 79 (95% CI, 51-107) for white ethnicity, 0.70 (95% CI, 0.38-1.01) for index of multiple deprivation, 1704 (95% CI, 1329-2080) for registers of individuals with diabetes, 72 (95% CI, 14-129) for proportion of smokers, and −558 (95% CI, −735 to −382) for detected hypertension levels. All 5 predictors were highly significant in each year. The final models for 2007 and 2006 contained exactly the same predictors and the 95% CIs were broadly similar across all 3 models (Table 4). Therefore, higher proportions of white individuals, higher levels of deprivation, higher levels of diabetes, higher proportions of smokers, and lower levels of detected hypertension were associated with higher levels of CHD mortality at PCT level in our models.
The median (interquartile range [IQR]) proportion with detected hypertension appears to have increased gradually between 2006 and 2008, ranging from 0.126 (IQR, 0.112-0.139) in 2006 to 0.129 (IQR, 0.115-0.143) in 2007 to 0.132 (IQR, 0.119-0.147) in 2008.
The residuals from the fitted models were all approximately normally distributed. Two trusts have been identified as high scoring outliers, but because we had no reason to doubt the integrity of these values, the trusts were included in the models (sensitivity analyses showed that omission of the outlier trust from the analysis did not change which variables were significant in the final models). We bootstrapped the 95% CIs for the weakest level 1 predictor for each year (smokers for 2008 and 2006, and index of multiple deprivation for 2007) and the strongest nonsignificant level 2 predictor for each year (2008 practice staff per 100 000 population, and 2007 and 2006 general practitioners per 100 000 population). All the bootstrapped 95% CIs were consistent with the original estimates.
Our study confirms that despite overall reductions in CHD mortality rates marked variations persist a decade after the launch of a national policy to reduce both mortality and inequalities. We found that the predictors of variation in CHD mortality between PCT populations in England were primarily population characteristics. Socioeconomic deprivation, white ethnicity, levels of smoking, and registers of individuals with diabetes were all significant positive predictors.
White ethnicity had a negative univariable coefficient with respect to mortality, which was, however, significant in a univariable model only for 2008. When hypertension detection was included in the multivariable analysis, white ethnicity produced a positive and highly significant association with mortality. This is an example of a suppressive relationship in a regression model,31 possibly arising in this case because hypertension detection was higher in PCTs with higher proportions of white individuals.
Neither provision of primary health care as indicated by the numbers of physicians or staff per 100 000 population, nor clinical performance as reflected by the quality and outcomes framework indicator scores predicted mortality in any year. However, the negative association of numbers of adults recorded in practice hypertension registers (higher CHD mortality with fewer individuals included in practice hypertension registers) can be explained by the underdetection of hypertension. This supports our hypothesis that higher numbers of patients recorded in hypertension registers reflects better performance in detecting patients with increased blood pressure rather than the prevalence of hypertension in the population.
Our study has a number of limitations. The final hierarchical model only predicted 66% to 68% of variation in CHD mortality between PCTs. This introduces the possibility that 1 or more relevant factors were omitted from the model. The service factors included in the model were limited to those for which good quality data were available. We were unable to investigate the influence of other potentially relevant factors, such as access or continuity. One candidate missing population variable was sex, because the age-standardized CHD mortality rates are higher in men than in women. Unfortunately, we lacked breakdown by sex for almost all of the explanatory variables.
The data are not entirely contemporaneous. For example, smoking and obesity rates are modeled for 2003-2005, ethnicity rates and deprivation index calculated for 2007, mortality for the calendar year (January to December), and for the quality and outcomes framework the business year (April 1 to March 31). These were the latest or best contemporaneous fit for the variables selected for analysis.
Primary care trust level data represent averages in large populations, but may not necessarily reflect variations at individual practice and patient level. Although our study's goal was not to conduct an analysis at practice level, further investigation at practice level would be informative.
These findings have a number of implications. Coronary heart disease mortality rates are predominantly explained by population characteristics. Programs to reduce mortality should address those characteristics of populations amenable to intervention, including smoking and deprivation. The importance of paying attention to population characteristics is emphasized by the finding that better detection of hypertension in the population was associated with reduced CHD mortality at the population level. In the United States in 2007-2008, 72.5% of individuals with hypertension were receiving treatment32 compared with 51.2% in England in the same year.19 Therefore, although performance is better in the United States, improvements could be made in detecting and treating hypertension in both countries. Pay for performance incentive schemes focused on clinical management indicators alone may have only limited affect on population outcomes such as mortality. By concentrating on achieving targets for patients already on a disease register, the quality and outcomes framework provides practices with incentives primarily to improve individual patient care, with less attention being given to the whole population. Incentive schemes should be designed to promote systematic, population-wide identification of individuals at risk, as well as reward the appropriate care of individuals identified.
In contrast with US studies that have shown an association between the supply of primary care physicians and mortality,7,8 we did not find such an association in England, possibly because in the country's universal access system, there is less variation between localities in the numbers of primary health care professionals per head of population.
Although a decade-long national policy to lower CHD mortality succeeded, wide variation in rates persist. This wide variation in CHD mortality is explained predominantly by population factors. A population orientation may be important in promoting further declines in CHD mortality rates. The extent to which primary health care services can affect these population factors is not certain.
Corresponding Author: Louis S. Levene, MB, BChir, FRCGP, Department of Health Sciences, University of Leicester, 22-28 Princess Rd W, Leicester LE1 6TP, England (lsl7@le.ac.uk).
Author Contributions: Dr Levene had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Levene, Baker, Bankart, Khunti.
Acquisition of data: Levene.
Analysis and interpretation of data: Levene, Baker, Bankart, Khunti.
Drafting of the manuscript: Levene, Baker, Bankart.
Critical revision of the manuscript for important intellectual content: Levene, Baker, Bankart, Khunti.
Statistical analysis: Bankart.
Obtained funding: Baker.
Administrative, technical, or material support: Baker.
Study supervision: Baker, Khunti.
Financial Disclosures: Dr Khunti reported being an advisor to the National Screening Committee and being a Clinical Advisor for the Diabetes NICE-led Quality and Outcomes Framework Panel. No other authors reported any financial disclosures.
Funding/Support: The study formed part of the program of the National Institute for Health Research (NIHR) Collaboration for Leadership in Applied Research and Care (CLARHC) in Leicestershire, Northamptonshire, and Rutland. The CLARHC is funded by the NIHR.
Role of the Sponsor: The funding organizations had no role in the design and conduct of the study, in the collection, analysis, and interpretation of the data, or in the preparation, review, or approval of the manuscript.
Disclaimer: The views expressed in this article do not necessarily reflect those of the NIHR.
1.NHS Information Centre for Health and Social Care. National Centre for Health Outcomes Development (NCHOD).
http://www.nchod.nhs.uk/. Accessed December 22, 2009 2.Unal B, Critchley JA, Capewell S. Explaining the decline in coronary heart disease mortality in England and Wales between 1981 and 2000.
Circulation. 2004;109(9):1101-110714993137
PubMedGoogle ScholarCrossref 4.Ashworth M, Medina J, Morgan M. Effect of social deprivation on blood pressure monitoring and control in England: a survey of data from the quality and outcomes framework.
BMJ. 2008;337:a203018957697
PubMedGoogle ScholarCrossref 5.Asia Pacific Cohort Studies Collaboration. The impact of cardiovascular risk factors on the age-related excess risk of coronary heart disease.
Int J Epidemiol. 2006;35(4):1025-103316581818
PubMedGoogle ScholarCrossref 6.Yusuf S, Hawken S, Ôunpuu S,
et al; INTERHEART Study Investigators. Effect of potentially modifiable risk factors associated with myocardial infarction in 52 countries (the INTERHEART study): case-control study.
Lancet. 2004;364(9438):937-95215364185
PubMedGoogle ScholarCrossref 7.Shi L, Macinko J, Starfield B, Politzer R, Wulu J, Xu J. Primary care, social inequalities and all-cause, heart disease and cancer mortality in US counties: a comparison between urban and non-urban areas.
Public Health. 2005;119(8):699-71015893346
PubMedGoogle ScholarCrossref 8.Shi L, Macinko J, Starfield B, Politzer R, Wulu J, Xu J. Primary care, social inequalities, and all-cause, heart disease, and cancer mortality in US counties, 1990.
Am J Public Health. 2005;95(4):674-68015798129
PubMedGoogle ScholarCrossref 9.Gulliford MC, Jack RH, Adams G, Ukoumunne OC. Availability and structure of primary medical care services and population health and health care indicators in England.
BMC Health Serv Res. 2004;4(1):1215193157
PubMedGoogle ScholarCrossref 10.Gostin LO, Connors EE. Health care reform in transition: incremental insurance reform without an individual mandate.
JAMA. 2010;303(12):1188-118920332405
PubMedGoogle ScholarCrossref 12.Campbell SM, Reeves D, Kontopantelis E, Sibbald B, Roland M. Effects of pay for performance on the quality of primary care in England.
N Engl J Med. 2009;361(4):368-37819625717
PubMedGoogle ScholarCrossref 13.Strong M, Maheswaran R, Radford J. Socioeconomic deprivation, coronary heart disease prevalence and quality of care: a practice-level analysis in Rotherham using data from the new UK general practitioner Quality and Outcomes Framework.
J Public Health (Oxf). 2006;28(1):39-4216436452
PubMedGoogle ScholarCrossref 14.Ashworth M, Armstrong D. The relationship between general practice characteristics and quality of care: a national survey of quality indicators used in the UK Quality and Outcomes Framework, 2004-5.
BMC Fam Pract. 2006;7:6817096861
PubMedGoogle ScholarCrossref 15.Saxena S, Car J, Eldred D, Soljak M, Majeed A. Practice size, caseload, deprivation and quality of care of patients with coronary heart disease, hypertension and stroke in primary care: national cross-sectional study.
BMC Health Serv Res. 2007;7:96-10417597518
PubMedGoogle ScholarCrossref 16.Ashworth M, Seed P, Armstrong D, Durbaba S, Jones R. The relationship between social deprivation and the quality of primary care: a national survey using indicators from the UK Quality and Outcomes Framework.
Br J Gen Pract. 2007;57(539):441-44817550668
PubMedGoogle Scholar 17.Doran T, Fullwood C, Kontopantelis E, Reeves D. Effect of financial incentives on inequalities in the delivery of primary clinical care in England: analysis of clinical activity indicators for the quality and outcomes framework.
Lancet. 2008;372(9640):728-73618701159
PubMedGoogle ScholarCrossref 24.Noble M, Wright G, Smith G, Dibbins C. Measuring multiple deprivation at the small-area level.
Environ Plan A. 2006;38(1):169-185
Google ScholarCrossref 29.Lomax RG. Statistical Concepts: A Second Course for Education and the Behavioral Sciences. 3rd ed. Mahwah, NJ: Lawrence Erlbaum Associates; 2007
30.Cohen J, Cohen P, West SG, Aiken LS. Applied Multiple Regression/Correlation Analysis for the Behavioural Sciences. 2nd ed. Mahwah, NJ: Lawrence Erlbaum Associates; 1983
31.MacKinnon DP, Krull JL, Lockwood CM. Equivalence of the mediation, confounding and suppression effect.
Prev Sci. 2000;1(4):173-18111523746
PubMedGoogle ScholarCrossref 32.Egan BM, Zhao Y, Axon RN. US trends in prevalence, awareness, treatment, and control of hypertension, 1988-2008.
JAMA. 2010;303(20):2043-205020501926
PubMedGoogle ScholarCrossref