We express our gratitude to Sven Vrielinck from Ghent University for his timely and comprehensive delivery of data. Additionally, we extend our thanks to Jean-Paul Sanderson from the Université Catholique de Louvain for his collaboration in providing the more recent data.

1. Introduction

1Cardiovascular disease (CVD) currently ranks as the second leading cause of death in Belgium, highlighting a significant public health concern [STATBEL, 2022]. In the early 20^th century, CVD was a relatively uncommon cause of death in Belgium, but it increased substantially in the second half of that century. A nuanced view is needed when describing this increase in the early 20^th century as it appears to be partly related to improved diagnosis in many European countries [Reid et al., 2015]. Additionally, the proportional increase in CVD deaths is linked to the reduction of infectious disease and the subsequent rise of so-called man-made diseases such as CVD and cancer during the epidemiological transition [Omran, 1971]. After this surge until the end of the 20^th century, there was a notable decline in CVD mortality [Devos, 2006; STATBEL, 2022].

2Various factors, including sociodemographic (SD), socioeconomic (SE), contextual, and lifestyle factors contribute to an increased risk of CVD mortality. Importantly, these CVD risk factors do not affect everyone equally, resulting in both social and often-overlooked spatial inequalities in CVD mortality [Carter et al., 2019 ; Christensen et al., 2016 ; de Mestral and Stringhini, 2017 ; Diez-Roux, Link and Northridge, 2000 ; Gadeyne, 2005 ; Grimmeau et al., 2015 ; Kim et al., 2008 ; Kornitzer et al., 1979 ; Willems et al., 2007]. Evidence from the United States (US) and Europe reveals significant regional variations in CVD mortality [Lang, 1999 ; Morris et al., 2003 ; Müller-Nordhorn et al., 2004 ; Roth et al., 2017]. In Belgium, overall mortality rates were higher in the northern regions of the country until the end of the 19^th century [Eggerickx, Sanderson and Vandeschrick, 2020], but for CVD mortality, no exceptional north-south disparities were observed in 1910 [Van Rossem, Deboosere and Devos, 2018]. In the 1930s, in contrast, geographical differences emerged at the provincial level [Selleslags, 1938]. From the second half of the 20^th century, the geographical pattern displayed higher CVD mortality in southern Belgium [Eggerickx, Sanderson and Vandeschrick, 2020 ; Grimmeau et al., 2015 ; Kornitzer et al., 1979]. These diverging spatial trends in CVD mortality over time highlight the need for a more detailed examination of spatial inequalities at a finer, more localized scale.

4We will use two indicators. The first assesses the percentage of CVD deaths to total deaths, calculated over various time intervals (1889-1890-1891, 1905-1909-1910, 1929-1930-1931, 1947-1948-1949, 1969-1970-1971, 1991, and 2011). The second indicator, the standardised mortality ratio (SMR), is computed for specified years. Both indicators are visually depicted on maps.

5The structure of this paper unfolds as follows: the first section delves into the evolution of CVD mortality, followed by an overview of the SE history of Belgium. The third section provides a literature review on spatial inequalities in CVD mortality in Belgium. In the subsequent section, we present our data and methods and discuss our results. We conclude with a discussion of findings and a comprehensive conclusion.

2. Cycles of the Heart: Unravelling the History of CVD Mortality

6This section presents the evolution of CVD mortality. CVD is an umbrella term that describes conditions that affect the cardiovascular system (heart, blood vessels and blood circulation) [Olvera Lopez, Ballard and Jan, 2022]. These conditions include coronary heart disease, cerebrovascular disease, peripheral arterial disease, rheumatic heart disease, congenital heart disease, and deep vein thrombosis and pulmonary embolism [WHO, 2021]. Although each type of CVD has unique characteristics, this article will consider CVD as a whole due to the complexity and limited data availability for each specific condition within our timeframe.

7In Western countries, manmade diseases – essentially CVD and cancer – replaced infectious diseases (such as tuberculosis, cholera, and smallpox), as well as respiratory and other causes of deaths, as the most important causes of death [Devos, 2006], widely known as the epidemiological transition [Omran, 1971]. The battle against infectious diseases was a public health success story – commonly attributed to better hygiene, rising living standards and medical progress – resulting in a significant increase in health [Mackenbach, 2020 ; Mensah et al., 2017]. In Belgium, for instance, life expectancy at birth more than doubled, from ca. 37 years in 1800 to 82 years in 2019 [Devos, 2006; STATBEL, 2022]. Along with this evolution, mortality from CVD increased considerably [Devos, 2006]. In Belgium, the share of CVD deaths rose from 4% in 1850 to 9% in 1910, 29% in 1950 and 37% in 1995, followed by a decline from 36% in 2000 to 25% in 2019 [Devos, 2006; STATBEL, 2022]. Today, CVD still accounts for 23% of all deaths in Belgium, ranking as the second most important cause of death after cancer [STATBEL, 2022]. Worldwide, it is the largest cause of death, causing about one in three deaths [WHO, 2021].

8However, these general national figures mask important differences, often driven by individual and spatial SE inequalities. The Fundamental Cause Theory (FCT), initially introduced by Link and Phelan [1995], assumes that SE inequalities are fundamental causes and drivers of health inequalities. Higher socioeconomic status (SES) provides individuals or regions with resources such as knowledge, money, and social connections enabling them to avoid disease, seek treatment, and adopt healthy behaviour [Link and Phelan, 1995]. Clouston and colleagues expanded the FCT by integrating it with the epidemiological transition theory, positing that diseases go through four stages [Clouston et al., 2016]. In the first stage, minimal inequalities exist due to a lack of knowledge about the risk factors and/or treatments. In the second stage, knowledge and prevention measures become available, but are unequally diffused, resulting in growing inequalities. The third stage sees a reduction in inequalities as health knowledge diffuses across all layers of society. Finally, in the fourth stage, both the incidence and mortality of the disease decline [Clouston et al., 2016]. This theory offers an excellent framework to help to interpret the emergence and evolution of spatial inequalities in CVD mortality through time.

9Indeed, prior to the onset of the CVD epidemic in the early 20^th century, the understanding of the risk factors and possible therapies for CVD were limited. Enhanced diagnostic methodologies led to an apparent rise in CVD mortality at the turn of the century. Deaths previously designated due to “old age” and “paralysis” were now more accurately defined as CVD-related deaths [Reid et al., 2015]. However, the big killers among the CVD, such as sudden death and acute myocardial infarction, remained unknown [Braunwald, 1997].

10The post-World War II era marked a turning point in addressing the CVD pandemic. The establishment of the National Heart Institute and research initiatives in the US, such as the Framingham Heart Study (1949) highlighted blood pressure and cholesterol as key CVD risk factors [Nabel and Braunwald, 2012]. In Europe, the World Health Organization’s (WHO’s) Belgian Heart Disease Prevention project [1983], revealed a 17.5% lower total mortality in an intervention group that received education and counselling [Kornitzer et al., 1983]. However, the decline in cardiac deaths is multifactorial, involving prevention measures, new treatments, and medical advancements [Dalen et al., 2014], such as coronary care units, external defibrillators, and beta-blockers [Braunwald, 1997]. Throughout the 1980s and 1990s, CVD mortality continued to decline due to better control of cholesterol and hypertension, along with therapies such as thrombolysis, percutaneous coronary angioplasty, statins, and angiotensin-converting enzyme inhibitors [Araújo et al., 2014]. An additional rationale for the decline in CVD mortality may be associated with a decrease in smoking prevalence [Dalen et al., 2014]. Some demographers have considered this trend as a new fourth stage in the epidemiological transition, characterised by a significant decline in mortality from coronary heart disease and strokes [Mensah et al., 2017 ; Olshansky and Ault, 1986].

11Despite medical improvements and prevention measures, CVD continued to be the leading cause of death in many Western countries at the end of the 20^th century. A recent study even indicates an increase in age-adjusted mortality rates for heart failure during the period 2012-2015 in the US [Mensah et al., 2017].

3. Belgian Context

12This article examines spatiotemporal trends in CVD mortality spanning from 1890 to 2011 at a municipal level.

13At the beginning of the 20^th century, Belgium comprised nine provinces (West Flanders, East Flanders, Antwerp, Brabant, Limburg, Hainaut, Namur, Liège, and Luxembourg) [Van Rossem, Deboosere and Devos, 2018]. Since the fourth state government reform in 1995, the province Brabant was divided into Flemish Brabant and Walloon Brabant, leading to a total of 10 provinces [Belgische Federale Overheidsdiensten, 2024]. Flanders comprises West Flanders (West-Vlaanderen), East Flanders (Oost-Vlaanderen), Antwerp (Antwerpen), Limburg, and Flemish Brabant (Vlaams-Brabant), while Wallonia includes Hainaut, Walloon Brabant (Brabant Wallon), Namur, Liège, and Luxembourg (see Figure 1).

14Given the extensive temporal scope of this study, various municipalities have undergone significant changes between 1890 and 2011. At the start of our observation period, Belgium encompassed 2,617 municipalities [Vrielinck, 2013]. This number gradually increased to 2,663 municipalities by 1961 and further diminished to 595 distinct municipalities after the municipal mergers of 1977 [Plees, 2005 ; Vrielinck, 2013]. Minor adjustments continued until 2020, leading to a classification of 581 municipalities [Plees, 2005]. To ensure comparability through time, we will adopt the current municipal boundaries.

15Belgium was the second country to industrialise, after Great Britain [Clark, 1984]. Significant regional disparities existed within the country. Wallonia witnessed successful development in heavy industry during the 19^th century, whereas most regions of Flanders faced serious challenges as the rural textile industry lagged in industrialization. In 1896 the wealthiest regions were situated in the industrial belt, stretching from Hainaut near France to Verviers in the Liège province, near Germany. These areas were major steel and manufacturing centres in Europe, rich in coal and iron resources crucial to the industrial revolution [Buyst, 2009]. While the city of Ghent in East Flanders had cotton factories that provided employment since the late 1820s [Verhaegen, 1961, as cited in Van Rossem et al., 2018], East Flanders, West Flanders and Limburg were among the less wealthy regions of the country [Buyst, 2009].

16In contrast, the 20^th century saw a reversal of fortunes, with Flanders experiencing economic growth while Wallonia faced decline [Buyst, 2009]. By 1937, the economic disparity between both regions began to decrease, with structural changes favouring Flanders. Flanders leveraged its port areas in Antwerp and imported coal and copper from the Belgian Congo, while the Walloon mines faced depletion. Yet, the provinces of Hainaut and Liège still maintained an above-average per capita GDP [Buyst, 2009].

17After WWII, the Walloon manufacturing belt greatly benefited from the high demand for steel, metal products, glass, and cement during Europe’s reconstruction [Baudhuin, 1958, as cited in Buyst, 2009]. The rise of oil in the 1950s and the closure of most coal mines in the 1960s marked a significant shift in Wallonia’s economic trajectory, leading to its decline from a previously prosperous region. By 1970, Hainaut and Liège fell below the national average in capital per GDP. Conversely, Flanders capitalized on its port zones in key industrial sectors such as oil refining and chemicals, attracting foreign investments and becoming the industrial heart of Belgium. By 2000, this led to a new North-South division, with a wealthier northern region and a less affluent southern counterpart [Buyst, 2009].

18Today, industrial activities are still concentrated in the Flemish region. Specifically, northeastern Flanders hosts sectors characterized by significant capital concentration, exemplified by industries such as the chemical, whereas the west displays light industry with modest capital involvement, notably in the textile sector. Wallonia has experienced deindustrialization, leading to a rising prominence of non-profit organisations due to insufficient industrial activities or underdeveloped tradable services. Noteworthy economic sectors include coastal tourism and Antwerp, recognized as the country’s primary industrial hub [Mérenne-Schoumaker et al., 2015]. Brussels hosts a significant portion of Belgium’s financial sector, indicating a concentration of economic authority in the region. In contrast, Walloon municipalities show comparatively less presence in this sector. Belgium’s current economy demonstrates a robust level of internationalization, firmly embedded within the global economy. Remarkably, Flanders attracts a larger share of foreign direct investment, possibly due to its higher level of urbanization [Mérenne-Schoumaker et al., 2015].

4. Regional disparities in CVD mortality and their drivers

19As previously mentioned, advancements in healthcare and proactive public health initiatives contributed to an overall decline in CVD mortality during the second half of the 20^th century. Yet, those improvements were not equally advantageous for all. Several studies have observed inequalities in CVD mortality among different European countries at the end of the 20^th century, with higher CVD mortality rates observed in Central and Eastern Europe compared to Western Europe [Mackenbach, 2020 ; Sans et al., 1997]. Furthermore, regional variations have been documented within countries, including disparities at the county level in the US [Roth et al., 2017], between the northern and southern regions of France [Lang, 1999], between East and West in Germany [Müller-Nordhorn et al., 2004], and in the UK, specifically between the northern regions of England and Scotland [Morris et al., 2003].

20In Belgium, life expectancy at birth was lower in Flanders than in Wallonia during the 19^th century [Grimmeau, Romainville and Beys, 2010], primarily due to the higher infant mortality in the economically deprived Flemish region [Eggerickx, Sanderson and Vandeschrick, 2020 ; Van Rossem, Deboosere and Devos, 2018]. Regarding CVD mortality, high CVD death rates for the elderly were identified in both Belgian regions in 1910 [Van Rossem, Deboosere and Devos, 2018]. Although spatial differences in CVD mortality were not apparent during that period, Van Rossem, and colleagues [2018] demonstrated differences in CVD mortality between men and women at the district level. At the provincial level, Selleslags [1938] showed variances in CVD mortality already by the 1930s, age-standardized CVD death rates ranging from 156 deaths per 100,000 in the province of Limburg, to 357 deaths per 100,000 in East Flanders. In Wallonia, Hainaut had the highest age-standardized CVD death rate with 317 deaths per 100,000, Luxembourg had only 164 CVD deaths per 100,000 [Selleslags, 1938, p.110]. At the regional level, higher CVD mortality was observed in Wallonia than in Flanders, particularly among men, in the second half of the 20^th and early 21^st century [Andre, 1981 ; Gossiaux, 1983 ; Grimmeau et al., 2015 ; Kornitzer et al., 1979 ; Lemonnier, 1981 ; Renard, Tafforeau and Deboosere, 2015].

21Potential drivers of geographical disparities encompass among others SD and SE factors. The distribution of the population by age, sex and migrant background significantly influences CVD outcomes, with old age and specific ethnicities being associated with higher CVD mortality [Boehme, Esenwa and Elkind, 2017 ; Kathiresan and Srivastava, 2012]. Since the 1990s, numerous studies have focused on the relationship between SES and mortality, consistently showing that more deprived groups experience higher CVD mortality [Carter et al., 2019 ; Christensen et al., 2016 ; de Mestral and Stringhini, 2017 ; Diez-Roux, Link and Northridge, 2000 ; Gadeyne, 2005 ; Kim et al., 2008]. To explain these higher CVD mortality rates, high-risk behaviours have been put forward that are more prevalent among the least privileged groups such as smoking, an unhealthy diet, excessive consumption of alcohol and a sedentary lifestyle [Eggerickx, Sanderson and Vandeschrick, 2020 ; Mackenbach et al., 2008]. Additionally, poor health hygiene related to prevention and early screening [Eggerickx, Sanderson and Vandeschrick, 2020 ; Willems et al., 2007] also play a role.

22However, individual SE and risk factors alone may not explain all regional variability of CVD. Other drivers associated with such geographical inequalities include environmental factors, but also contextual neighbourhood factors such as accessibility to emergency services [Timonin et al., 2018]. For example, CVD has been linked to air pollution [Franklin, Brook and Arden Pope, 2015] and specific metals recognized by the WHO among the top 10 concerning environmental chemicals [Cosselman, Navas-Acien and Kaufman, 2015]. Similarly, long-term exposure to traffic noise has been associated with an increased risk of mortality from coronary heart disease [Cai, Ramakrishnan and Rahimi, 2021 ; Kempen et al., 2018]. There is also evidence showing a correlation between a lower SES and residing in neighbourhoods with reduced green space and poorer health infrastructure, which in their turn increase the risk of CVD [Eggerickx, Sanderson and Vandeschrick, 2020 ; Willems et al., 2007]. These studies indicate that deprived neighbourhoods may be linked to well-established behavioural and environmental risk factors [Brousmiche et al., 2023 ; Willems et al., 2007].

23Specifically in Belgium, research has identified several key factors contributing to the north-south divide in CVD mortality at the end of the 20^th century, including SE differences between the regions, differences in health behaviour, and contextual factors [Deboosere and Gadeyne, 2002 ; Van Hemelrijck, Willaert and Gadeyne, 2016]. Cultural differences between both regions translate in dietary disparities such as varying levels of vegetable, fruit, and fish consumption alongside differences in physical activity and sport engagement which contribute to different mortality and CVD mortality risks [Grimmeau et al., 2015]. Notably, differences in dietary habits have been highlighted [Grimmeau et al., 2015 ; Gupta et al., 2006 ; Joossens, Geboers and Kesteloot, 1989 ; Kornitzer et al., 1979], for instance the higher consumption of butter in Wallonia compared to greater margarine consumption in Flanders. In addition to cultural factors, exposure to air pollution [Kangas et al., 2023] have also been identified as contributors to regional inequalities in Belgium.

5. Data

24Our study aims to provide a comprehensive historical background, capturing the onset of the CVD revolution and its evolution until the present day. Notably, the year 1890 serves as the starting point for our observation period, as it represents the first year with reliable and substantial data quality available for analysis [Preneel, 2010]. Our two indicators – the percentage of CVD deaths to total deaths and the SMRs – require data on the number of all-cause deaths, the number of CVD deaths and the population structure by age. These data have been compiled from various sources, presented in table 1.

25Importantly, to deal with random fluctuations in data and data quality, three-year intervals centred on the year of the census were employed (e.g., 1889-1890-1891, 1905-1909-1910, 1929-1930-1931, 1947-1948-1949, 1969-1970-1971). Conversely, recent data are not susceptible to these issues, therefore, data were centred around the census years 1991 and 2011. The calculation of SMRs requires data on the age-distribution of the population. As data on age structures were unavailable for the years 1930 and 1947, SMRs could not be calculated for the periods 1929-1930-1931, and 1947-1948-1949. Analyses for these years have to rely on the other indicator, the share of CVD deaths to total deaths.

(a) CVD deaths and all-cause deaths per municipality

28We followed the classification of CVD as defined in various sources. The Mouvement referred to CVD as “Maladies organique du Coeur”. For the years 1991 and 2011, we included all types of CVD categorized and coded according to the International Classification of Diseases (ICD), see Appendix A.

(b) Population structure by municipality

30For the years, 1889-1890-1891, 1905-1909-1910, 1929-1930-1931, and 1947-1948-1949 the mortality data refer to de facto deaths, meaning that deaths are registered at the place of death and not the place of residence. From 1969 onwards, the data are de jure, registered by the place of residence of the deceased person.

6. Methods

31The analyses include three distinct sections. In the initial step, percentages representing the share of CVD deaths to total deaths (1) are calculated to assess the significance of CVD within the broader spectrum of causes of death. Calculations are conducted for each specified point in time (1890-1910-1930-1947-1970-1991-2011) offering insight into the evolution of the proportion of CVD deaths over the years.

32While the share of a cause of death in total deaths is an interesting summary measure, it should be interpreted cautiously and complemented with other indicators to ensure a comprehensive understanding of mortality patterns and trends. Focusing solely on the share of a cause of death may overlook crucial aspects of mortality, such as the impact of age, which varies over time. For example, the significant decrease in infant and child mortality throughout the observation period, varying across municipalities, may have had a substantial impact on the distribution of causes of death and its regional patterning. That is why, ideally, we should have age-specific data on CVD mortality, focussing for example on the share of CVD death in the population aged 50 and over, but these are not readily available. In the absence of such data and given the challenge posed by small numbers at the municipal level, we will apply indirect standardization techniques and compute SMRs (2) to control for age. This involves applying a standard age-specific mortality scheme, based on the Belgian population of 1961 and CVD deaths in 1961, to the population distributions (age structures) in 1890, 1910, 1970, 1991, and 2011. Up to 1970, due to limited data availability and comparability, the population distribution at the municipality level consists of three age categories 0-14; 15-54; 55+. After 1970, a more detailed distribution, consisting of the age groups 0-14; 15-24; 25-34; 35-44; 45-54; 55-64; 65-74; 75+ is used. Data on the municipality age structures are currently not available for 1930 and 1947. Municipalities with a SMR higher than one have a higher CVD mortality than expected based on the standard mortality scheme of 1961; municipalities with a SMR lower than one have a lower mortality than the standard mortality scheme.

33As literature on CVD mortality suggests shifts in sex disparities over time [Nikiforov and Mamaev, 1998], the analyses are stratified by sex. Independent two-sample t-tests (3) are performed to compare CVD mortality by sex. This method determines whether the observed difference between men and women is statistically significant.

34The data analyses and mapping are performed using R with the assistance of the ‘ggplot2’ and ‘sf’ packages. To represent the data distribution visually on the maps, a quantile-based legend is employed. Quantile classing gives each class the same number of units. Even though there is no real correct way to categorise a data set, quantile classification method is best suited for choropleth maps in series that are intended for a wide range of map-reading tasks [Brewer and Pickle, 2002]. Most epidemiologists prefer using quantiles as they provide clearer shifts in death rates and enable more effective comparison of these rates, regardless of differences in data standardization and age adjustment [Brewer, 2006].

7. Results10

7.1. Share CVD deaths to total deaths (%)

36While the overall trend for the entire country is well-documented, patterns at the municipal level remain unclear. To address this gap, we calculated the percentage share of CVD deaths to total deaths for each municipality from 1890 to 2011 (see Figure 2).

37The 1890 map highlights a higher concentration of CVD mortality in the municipalities located in Namur, with additional hotspots in neighbouring Hainaut and Limburg. This pattern remains consistent during the period 1905-1910 and appears relatively stable in 1929-1931 as well, with decreasing percentages in Limburg. However, a shift occurs in 1947-1949, indicating a higher share of CVD death in the central regions of the country, and Luxembourg. Almost 20 years later, municipalities in Luxembourg experience a surge in CVD mortality compared to other causes of death. In 1991, Namur and parts of Hainaut regain prominence. Moreover, West Flanders and Limburg seem to experience a surge in the share of CVD death as well. By 2011, the pattern becomes more scattered across the entire country.

7.1.1 Differences between men and women

38An independent two-sample t-test was performed to compare CVD deaths by sex. The percentage of CVD mortality to total mortality for both men and women was treated as two different variables, with each municipality serving as an individual observation. Significant differences in the share of CVD death were found between men and women for 1905-1909-1910, 1929-1930-1931, 1969-1970-1971, 1991, and 2011 (p-values < 0.05). No significant differences (p-values > 0.05) were found for the period 1889-1890-1891 and 1947-1948-1949.

39Figure 3 presents maps by sex. Few differences between men and women were apparent on the maps.

7.2. Standard mortality ratios (1890-2011)

41Given the small number of deaths in some municipalities, we opted for indirect standardisation, using the national CVD mortality schedule of 1961 as the standard mortality schedule. In 1890, the SMR was equal to 0.56 in Belgium, indicating a 44% lower CVD mortality at that time compared to the standard in 1961. Remarkably, by 1910, the SMR declined to 0.44, signifying a 56% lower mortality than expected. By 1970, the SMR rose to 1.27, indicating 27% higher CVD mortality than anticipated in 1961. Towards the end of the 20^th century, the SMR declined to 0.47.

42The maps in Figure 4 do not really suggest a clear regional pattern of CVD mortality at the beginning of the 20^th century; instead, they suggest a scattered regional distribution. In 1890, municipalities such as Mortsel (1.97) (95% confidence interval (CI) of 0.73-3.21), Merksplas (4.33) (CI 2.48-6,18), and Wuustwezel (1.74) (CI 0.90-2.57) exhibit the highest SMR, surpassing the confidence interval at the national level ranging from 0.51 to 0.57. Notably, 19 municipalities or 3% of all municipalities report no CVD deaths (e.g., De Panne, Elsene, Kalmthout, Zonhoven, Genk, Amel, Eupen, Kelmis, Malmédy, …).

43Likewise, during the years 1905-1910, no distinct clusters favouring one region over the other is evident. Municipalities with significantly high SMRs include Boortmeerbeek (1.46) (CI 0.66-2.27), Merksplas (1.39) (CI 0.70-2.08), and Uccle (1.37) (CI 1.07-1.66), compared with the CI at the national level between 0.43 and 0.48. Around 32 municipalities (5%) have zero CVD deaths, including Amel, Anderlues, Bullingen, Deerlijk, Tessenderlo, Waasmunster, among others.

44From 1969-1971 onwards, a notable concentration of CVD mortality emerges in Liège and Luxembourg. Municipalities such as Juprelle (1.99) (CI 1.40-2.57), Hannut (1.91) (CI 1.53-2.30), Dison (1.87) (CI 1.55-2.19), and Saint-Hubert (1.94) (CI 1.37-2.50) exhibit the highest SMR, while Wezembeek-Oppem (0.82) (CI 0.52-1.13), Haacht (0.87) (CI 0.57-1.18), but also Etterbeek (0.91) (CI 0.80-1.02), and Brussels (0.94) (CI 0.87-1.00) record the lowest. At the national level, the CI ranged from 1.32 to 1.35.

45By 1991, there is a notable increase in CVD mortality in Wallonia. Remarkably, municipalities shaded in dark red predominantly appear in the west of the country, deviating from the previous years when such concentrations were observed in the east. Notable municipalities with elevated SMRs included Farciennes (1.41) (CI 1.09-1.72), Lobbes (1.33) (CI 0.90-1.77), Erquelinnes (1.19) (CI 0.90-1.48), and Froidchapelle (1.19) (CI 0.65-1.73). Conversely municipalities with the lowest CVD mortality compared to standard mortality scheme included Halen (0.28) (CI 0.11-0.44), Daverdisse (0.30) (CI -0.04-0.65), Onhaye (0.35) (CI 0.04-0.65), among others. The CI at the national level was between 0.71 and 0.74.

46By 2011, a resurgence of CVD mortality was evident in Namur and Hainaut, but also in West Flanders. Municipalities such as Flobecq (0.96) (CI 0.58-1.34), Houyet (0.82) (CI 0.49-1.14), Doische (0.81) (CI 0.42-1.21), Wachtebeke (0.74) (CI 0.50-0.98) had high SMRs. On the other hand, municipalities with lower SMRs include Spiere-Helkijn (0.17) (CI -0.07-0.41), Lennik (0.19) (CI 0.08-0.30), Bertogne (0.20) (CI -0.03-0.42), Hove (0.25) (CI 0.12-0.38), Zemst (0.26) (CI 0.17-0.34) and others. The CI, when considered at the national level was 0.45-0.48.

7.2.1 Differences between men and women

47An independent two-sample t-test was conducted for the SMRs to compare CVD mortality by sex. Notably, significant differences were found for each year (p-values < 0.05). A closer examination of the maps reveals only subtle disparities (see Figure 5). Particularly in 1970, there are higher SMRs in Antwerp for women and in the more recent years, 1991 and 2011, nuanced variations in geographic clusters emerge.

8. Discussion and conclusion

48The primary objective of this study is descriptive, defining spatiotemporal inequalities in CVD mortality. The spatial analyses presented in this article reveal regional hotspots and clusters in CVD mortality suggesting significant spatial variations over time. During the late 19^th and early 20^th centuries, specifically 1889-1890-1891 and 1905-1909-1910, the share of CVD mortality did not exhibit a clear regional pattern, rather a scattered distribution. By 1930, a trend began to emerge in the central parts of the country, with a noticeable shift towards the southern regions by 1947. In the late 20^th century, from 1970 onwards, CVD mortality increased in the eastern part of the country and along the coastal areas. Prominent urban centres, including Brussels, Genk, and Charleroi, initially exhibited a lower share of CVD deaths to total deaths. By 1991, this changed drastically, with Charleroi experiencing an important increase in the share of CVD death. Additionally, in 1991, there was a surge in percentage share of CVD deaths to total deaths in Flanders. In 2011, the observed pattern in share of CVD deaths becomes again more scattered.

49These trends are less obvious when looking at the SMRs. Hotspots are more prominent in Wallonia from the 1970s onwards. In the early 21^st century, specifically in 2011, there is a resurgence of CVD mortality observed in municipalities in Namur, Hainaut, and West Flanders.

50Although significant sex disparities in CVD mortality emerge, suggesting higher mortality among women, particularly in 1991 and 2011, men and women display similar regional hotspots. Notably, differences between men and women are less pronounced (though statistically significant) when analysing the SMRs.

51In summary, by the end of the 19^th and early 20^th century, no clear regional pattern had developed. Our research results at the finer-grained municipal scale align with findings from earlier studies [Van Rossem, Deboosere and Devos, 2018]. At that time, there were mainly local hotspots, resulting in a scattered pattern in 1890 and 1910. As mentioned in Selleslags’ work (1938), Limburg experienced the lowest level of CVD mortality in 1930. Later, in the 1970s and 1990s hotspots were localized in Wallonia as previously mentioned in the literature [Eggerickx and Sanderson, 2010 ; Grimmeau, Romainville and Beys, 2010].

52Several possible explanations exist for the changing regional patterns. One explanation relates to the extended FCT, which posits that diseases undergo distinct phases depending on the available knowledge about prevention and treatment of the disease. In the initial phase of the extended FCT, when knowledge and treatments are not yet available, disparities in mortality are less pronounced. This situation was observed with CVD mortality in the early 20^th century, when it began to rise as a significant cause of death. During this period, preventive measures and treatments were still in their early stages, resulting in CVD mortality occurring somewhat randomly, with no discernible spatial inequality. This is reflected in the maps, depicting a scattered pattern of CVD mortality, indicating minimal spatial inequality in CVD deaths – a characteristic of the initial stage of the extended FCT.

53However, following WWII, major risk factors for the prevention of CVD were identified [Kornitzer et al., 1983 ; Nabel and Braunwald, 2012]. The 1970s witnessed substantial advancements in treatments and medical knowledge [Araújo et al., 2014 ; Braunwald, 1997]. According to the second stage of the extended FCT, the distribution of available knowledge and treatments is uneven due to their correlation with SES of an individual or a region. SE resources such as knowledge, money, and social networks, play a pivotal role in reducing risk factors by facilitating treatment, and encouraging the adoption of healthy behaviours [Link and Phelan, 1995]. Consequently, populations in more affluent regions could increasingly adopt healthier lifestyles, thereby reducing their exposure to these risk factors, whereas those in less affluent regions were less likely to do so. This disparity results in growing spatial inequality in CVD mortality, as illustrated in maps from 1969-1971 onward.

54This trend was evident in provinces such as Antwerp and Brabant during the latter half of the 20^th century, which had some of the highest relative GDP per capita [Buyst, 2009]. Municipalities in these affluent regions, such as Wezembeek-Oppem and Haacht, were better positioned to combat CVD mortality, resulting in lower mortality rates in these areas. Conversely, Walloon provinces, such as Hainaut and Namur, which had the lowest GDP per capita at the beginning of the 21^st century [Buyst, 2009], exhibited higher CVD mortality rates in municipalities such as Erquelinnes, Farciennes, and Lobbes. Overall, the increase in spatial inequality appears to be correlated with the preventability and treatability of the disease, as well as socioeconomic conditions, which reflects the second stage of the extended FCT.

55The second explanation relates to a variation in cultural habits between the North and the South regardless of the SES [Deboosere and Gadeyne, 2002 ; Van Hemelrijck, Willaert and Gadeyne, 2016]. Studies indicate a variation in dietary habits between Flanders and Wallonia, in which the Walloon population has a higher butter and animal fats consumption according to a survey of food habits in 1959 [Kornitzer et al., 1979] and the Health Survey in 1997 [Grimmeau et al., 2015]. This type of diet increases the likelihood of CVD, possibly explaining higher CVD mortality in Wallonia. However, this does not explain why certain regions in Wallonia have hotspots and others do not. And how these hotspots evolved over time. The varying pace of SE developments in municipalities, changes in age structures, and migration leading to shifts in population composition likely contribute to this phenomenon. These factors, along with contextual elements such as air pollution and accessibility to emergency services, should be included in future research.

56Given the historical scope of this research, the main limitation pertains to data quality. In the historical context, certain municipalities had relatively small populations and few (CVD) deaths possibly jeopardizing the accuracy and usefulness of the data (Krishnan, 2011). However, the aggregation of municipalities into the current administrative boundaries, partly mitigates this issue. The older data lack detailed information, as up to 1970, only three age-categories (0-14; 15-54; 55+) were available for age-standardization. Consequently, we could insufficiently control for age, especially since CVD mortality mainly occurs at older ages (+55). Data with more detailed age-categories exist per municipality, hence, it is a matter of time before these data will be digitized, enabling a finer analysis of geographic differences in CVD mortality across time in Belgium. As mentioned, the share of CVD deaths to total deaths should be examined by age, focusing on older population groups rather than on the total population. The high rates of infant and child mortality towards the end of the 19^th century and the beginning of the 20^th century significantly impact the share of CVD death relative to overall death, leading to a low share of CVD death. Given the regional patterning of infant and child mortality in Belgium, this also influences the regional patterning of the share of CVD death. Additionally, data on the population structure are unavailable for the years 1930, and the scans for 1947 from the original records are illegible, demanding ongoing efforts for their renewal. Due to the absence of population data for 1930 and 1947, SMRs could not be calculated for these years. An additional challenge pertains to the evolving International Classification of Diseases (ICD) codes. Many steps and researchers are involved in assigning ICD codes, making this process complicated and error prone [O’Malley et al., 2005]. Coding schemes and practices often evolve in response to changes in technology, medical understanding, and societal norms. As a result, data collected using different coding schemes or practices may not be directly comparable, potentially impacting the accuracy and reliability of analyses conducted across different time periods [Reid et al., 2015]. Another consideration is the distinction between de facto and de jure registrations. Until 1969, the death was recorded at the place of occurrence rather than the place of residence, reflecting a context where people predominantly did not die in hospitals. However, these days, most people die in hospitals and care homes. The shift in registration method to the deceased person’s place of residence ensures that mortality data now depict hotspots beyond hospitals and care homes. The changing municipal boundaries throughout the observation period present a significant challenge for spatiotemporal comparisons. While using consistent boundaries mitigates this concern, it is crucial to acknowledge that applying contemporary boundaries to historical data may not fully capture the historical accuracy of spatial distributions.

58Due to the absence of explanatory variables in this study, further research should link specific spatial variables to CVD mortality numbers. It has been established that variations in food habits and SE profiles partly account for disparities in CVD mortality between the prominent regions in Belgium. However, a gap persists in comprehending why certain municipalities have higher mortality numbers than others within the same region or province, and why they changed over time. Hence, we recommend exploring spatial factors identified as impactful in previous studies, such as air pollution and access to emergency services, across Belgium [Grimmeau et al., 2015 ; Kangas et al., 2023].