Urbanization and Non-communicable disease mortality in Thailand: an ecological correlation study

Authors


Corresponding Author Chaisiri Angkurawaranon, London School of Hygiene and Tropical Medicine, Keppel Street, London WC1E 7HT, UK. E-mail: chaisiri.angkurawaranon@lshtm.ac.uk; chaisiri@med.cmu.ac.th

Abstract

This study provides strong evidence from an LMIC that urbanization is associated with mortality from three lifestyle-associated diseases at an ecological level. Furthermore, our data suggest that both average household income and number of doctors per population are important factors to consider in ecological analyses of mortality.

Abstract

Cette étude fournit des preuves solides à partir d'un pays à revus faibles et intermédiaires que l'urbanisation est associée à la mortalité dans trois maladies associées au style de vie à un niveau écologique. En outre, nos données suggèrent que le revenu moyen du ménage et le nombre de médecins par habitant sont des facteurs importants à prendre en compte dans les analyses écologiques de la mortalité.

Abstract

Este estudio provee evidencia sólida de un país en vías de desarrollo de que la urbanización está relacionada con la mortalidad por tres enfermedades asociadas al estilo de vida a nivel ecológico. Más aún, nuestros datos sugieren que tanto los ingresos promedio del hogar como el número de médicos por habitante son factores importantes que considerar en los análisis ecológicos de mortalidad.

Introduction

Urbanization is considered a determinant of health and one of the key drivers of non-communicable diseases (NCDs), especially in low- and middle-income countries (LMICs) (Boutayeb & Boutayeb 2005; Vlahov et al. 2007). In 2008, 63% of global deaths were due to NCDs. Current projections suggest that between 2010 and 2020, NCD deaths are expected to increase by 20% in LMIC regions such as Africa and South-East Asia (Mathers et al. 2008; World Health Organization 2010). Growing evidence from LMICs shows that urbanization is associated with increased prevalence of risk factors for non-communicable disease (Hernandez et al. 2012). There is some evidence for an urban advantage in NCD mortality in high-income countries, possibility due to higher income and better access to health care, but limited data have been published on the association between urbanization and non-communicable disease mortality in LMICs (Harpham et al. 2004; Allender et al. 2008; Leon 2008).

Thailand's income category was recently upgraded from lower-middle-income to upper-middle-income economy by the The World Bank (2011). Like many developing countries, Thailand has undergone rapid urbanization within the last decades and has a growing burden of NCDs (Viravaidya & Sacks 1997; Cohen 2004). Between 1987 and 1993, the burden of disease from NCDs increased by 36% (Samutaruk 1997). By 2004, NCDs accounted for 65% of disability-adjusted life years lost in Thai people. The emergence of NCDs in Thailand results from socioeconomic, environmental and lifestyle changes associated with urbanization (Yiengprungsawan et al. 2011a). Recently published studies from Thailand found geographical variations in all-cause and cause-specific NCD mortality, but did not investigate the role of urbanization (Faramnuayphol et al. 2008; Odton et al. 2010).

This study aimed to assess the association between urbanization and specific causes of non-communicable disease mortality (cardiovascular disease (CVD), cerebrovascular disease and malignant neoplasms) along with all-cause mortality, in Thailand. We also investigated the influence of average monthly household income and number of doctors per population on mortality, and how they relate to the association between urbanization and mortality.

Methods

This ecological correlation study used information from 76 provinces in Thailand in 2009, including data on demographic structure, population density, the proportion living in an urban area, the number of doctors per population, the average monthly household income and the top 10 known causes of death.

Vital registration in Thailand

A detailed description of the Thai vital statistics system has been published (Tangcharoensatien et al. 2002). By law, each death must be notified within 24 h to the Bureau of Registration and Administration (BORA). For deaths that occur in hospitals, and unnatural deaths outside hospitals, a physician records one cause of death in Thai that is sent electronically to the national registration database (Pattaraarchachai et al. 2010). For natural deaths outside hospitals, the local registrars record one cause of death in Thai after inquiring the cause of death from the family. Mortality data from both systems are sent electronically to BORA to be complied into a database. Mortality is attributed to each province according to the deceased registered place of residence. The Ministry of Public Health is then responsible for coding the causes of death, which are specified in Thai, according to ICD-10 (Rao et al. 2010).

Outcome definitions

Cardiovascular disease mortality includes ICD codes, I05-I09, I120-I128 and I130-I152; cerebrovascular disease ICD codes, I10-I15 and I60-I69; and malignant neoplasms ICD codes, C00-C48.

Primary exposures

Population density is the number of (mid-year) population divided by the area (km2) for each province. The population in Thailand is defined using local administrative criteria (Archavanitkul 1988). Every person in Thailand must be registered under a household. People are classified as living in an urban area if the household they are registered at is under local municipality administration (Flood 2000). The proportion of persons living in urban areas within each province is defined as the urban population divided by total (mid-year) population of that province.

Other variables of interest

The province average monthly household income is derived from a survey carried out by the National Statistical Office annually using a stratified two-stage sampling technique. The 76 provinces are considered as individual strata, and each stratum is categorised into municipal areas and non-municipal areas. Villages are used as the primary sampling unit; individual households are the secondary sampling unit. (National Statistial Office of Thailand).

Number of doctors per population in a province is the number of medically licensed doctors registered to work in a hospital or clinic at the provincial public health office divided by the mid-year population of that province (Bureau of Policy & Strategy, Ministry of Public Health).

Data sources

Data are openly accessible from the National Statistical Office of Thailand and the Ministry of Public Health's website. Age and gender cause-specific mortality are tabulated for the National level data (http://service.nso.go.th/nso/thailand/thailand.jsp, http://service.nso.go.th/nso/nsopublish/BaseStat/tables/00000_Whole Kingdom/N28P02-income.xls and http://bps.ops.moph.go.th/Healthinformation/statistic50/statistic50.html).

Analysis

Each measure of urbanization was analysed separately. Age and gender were considered a priori as confounding factors. Indirect age-adjusted standardisation, using the country's age structure in 2009, was used to investigate the association between population density and the proportion living in an urban area with mortality. Scatter plots of standardised mortality ratios (SMRs) against measures of urbanization were used to graphically depict the relations between variables and to identify influential points and outliers. Pearson correlation coefficients were calculated to quantify associations between measures of urbanization and SMRs. Poisson regression models assessed the relationship between measures of urbanization and mortality after adjusting for age structure (10-year age bands) and the proportion of men, using the size of the population in each province as the offset variable. The Poisson models were adjusted further for the number of doctors per population and the average monthly household income. These variables are potential confounders (when trying to separate out the association between urbanization-induced life-style changes) and/or on the causal pathway between measures of urbanization and mortality outcomes. Sensitivity analysis was performed by removing the outlier, Bangkok, from the analyses. To explore the possible misclassification in causes of deaths, and to identify the main driver in all-cause mortality, further sensitivity analysis was performed by examining the other causes of death, which contribute to 55% of all-cause mortality.

Results

In 2009, the population in Thailand was 63 525 062 with an overall population density of 123.8 people per km2. Within the 76 provinces, the population varied from 181 754 people in Ranong Province to 5.7 million people in Bangkok (median = 634 202, IQR = 462 520–1033 997). The proportion living in an urban area ranged from 6.9% in Surin Province to 100% in Bangkok (median = 23.1%, IQR = 18.5–33.8%), whilst the population density varied from 19.1 people per km2 in Mae Hong Sorn Province to 3635.2 people per sq.km.in Bangkok (median = 121.9, IQR = 78.9–163.2). The average monthly household income across the 76 provinces was 18 805 baht (approx. £ 375) (median = 17, 537; IQR = 14 545–22 174). The average number of doctors per 10 000 population across the 76 provinces was 2.55 (median = 2.29; IQR = 1.82–3.12). There was a strong positive correlation between population density and proportion living in an urban area (r = 0.72, P < 0.01). There was a positive correlation between the two primary measures of urbanization and the average household income and number of doctors per population (r > 0.6 and P < 0.001 in all analysis). Of the two, the proportion living in an urban area showed a stronger correlation with average household income and number of doctors per population than population density (Figure 1).

Figure 1.

Matrix scatter plots of population density, proportion living in urban area, average household income and number of doctors per population using aggregate data across 76 provinces in Thailand.*Note: the outlier at the upper right represents Bangkok.

Urbanization and mortality

The all-cause mortality rate was 6.2 per 1000. The leading known causes of deaths and their contribution to all mortality were as follows: malignant neoplasms, 14.1%; accidents and poisoning, 8.9%; cerebrovascular disease, 5.0%; and CVD, 4.6%. The top 10 known causes of deaths accounted for 45% of all mortality. The rest were classified as other causes.

There was a weak negative correlation between population density and the SMRs (Figure 2). For every increase of 100 people per km2 in population density, there was a 0.2% decrease in overall mortality rate after adjustment for age and gender. Additional adjustments for numbers of doctors per population and average household income strengthened the association (Table 1).

Table 1. Estimated associations between measures of urbanization and percentage increase in all-cause mortality rate and their 95% confidence intervals (CI) using aggregate data across 76 provinces in Thailand
  Model 1Model 2Model 3Model 4
  1. Model 1 exposure variables in model: population density, age and gender.

  2. Model 2 exposure variables in model: population density, age, gender, average monthly household income and number of doctors per population.

  3. Model 3 exposure variables in model: proportion living in urban area, age and gender.

  4. Model 4 exposure variables in model: proportion living in urban area, age, gender, average monthly household income and number of doctors per population.

 Measures of urbanization in each province (Units)Percentage increase in rate per unit increase (95% CI)Percentage increase in rate per unit increase (95% CI)Percentage increase inrate per unit increase (95% CI)Percentage increase in rate per unit increase (95% CI)
All-cause mortalityPopulation density (100 people per population)−0.15 (−0.20 to −0.11)−0.28 (−0.39 to −0.18)
P < 0.001P < 0.001
Proportion living in urban area (10 per cent)0.34 (0.12 to 0.55)1.87 (1.41 to 2.34)
P = 0.002P < 0.001
Average monthly householdincome (1000 baht)−0.68 (−0.80 to −0.56)−1.12 (−1.22 to −1.02)
P < 0.001P < 0.001
Number of doctors per population (Doctors per 10 000)4.11 (3.70 to 4.52)2.79 (2.34 to 3.25)
P < 0.001P < 0.01
Figure 2.

Correlation between proportion living in urban area and population density with standardised mortality ratio using aggregate data across 76 provinces in Thailand.

In contrast, for the proportion living in an urban area, there was a positive correlation with mortality (r = 0.29, P = 0.01) (Figure 2). In the age- and gender-adjusted Poisson regression model, every ten per cent increase in the proportion living in an urban area was associated with a 0.3% increase in overall mortality rate (95% CI 0.1–0.6). Adjusting for number of doctors per population and average household income strengthened the association. (Table 1)

In both models, increasing average household income was associated with a decrease in mortality rate, whilst increasing number of doctors per population was associated with an increased mortality.

Urbanization and cardiovascular mortality

The overall cardiovascular mortality rate was 28.9 per 100 000. Across provinces, there was a positive correlation between both measures for urbanization and the SMR for cardiovascular disease (Figure 3).

Figure 3.

Correlation between proportion living in urban area and population density with standardised mortality ratio for cardiovascular disease using aggregate data across 76 provinces in Thailand.

Every increase of 100 people per km2 was associated with a 2.2% increase in cardiovascular mortality rate (95% CI 2.0–2.5) in age- and gender-adjusted analyses. Additional adjustment for numbers of doctors per population and average household income attenuated the association (Table 2). On its own, average household income had a negative confounding effect and the number of doctors had a positive confounding effect (data not shown).

Table 2. Estimated associations between measures of urbanization and percentage increase in cardiovascular mortality rate and their 95% confidence intervals (CI) using aggregate data across 76 provinces in Thailand
  Model 1Model 2Model 3Model 4
  1. Model 1 exposure variables in model: population density, age and gender.

  2. Model 2 exposure variables in model: population density, age, gender, average monthly household income and number of doctors per population.

  3. Model 3 exposure variables in model: proportion living in urban area, age and gender.

  4. Model 4 exposure variables in model: proportion living in urban area, age, gender, average monthly household income and number of doctors per population.

 Measures of urbanization in each province (Units)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)
Cardiovascular mortalityPopulation density (100 people per population)2.25 (2.04 to 2.46)1.97 (1.48 to 2.46)
P < 0.001P < 0.001
Proportion living in urban area (10 per cent)10.52 (9.48 to 11.58)6.63 (4.32 to 9.00)
P < 0.001P < 0.001
Average monthly household income (1000 baht)−0.57 (−1.12 to −0.01)0.29 (−0.19 to 0.76)
P = 0.047P = 0.236
Number of doctors per population (Doctors per 10 000)4.53 (2.61 to 6.49)3.36 (1.20 to 5.57)
P < 0.001P = 0.002

The correlation coefficient between the proportion living in an urban area and the SMR for CVD was 0.43 (P < 0.001). Adjusting for age and gender, every ten per cent increase in proportion living in urban area was associated with a 10.5% increase in cardiovascular mortality rate (95% CI 9.5–11.6), which attenuated to 6.6% (95%CI 4.3–9.0) after adjustments for average household income and number of doctors per population (Table 2). There was a positive association between number of doctors and cardiovascular mortality rates.

Urbanization and cerebrovascular mortality

The overall cerebrovascular mortality rate was 24.6 per 100 000. There was a positive correlation between both measures for urbanization and the SMR for cerebrovascular disease (Figure 4), but the correlation between population density and SMR for cerebrovascular mortality did not reach statistical significance (r = 0.14, P = 0.244). The regression models, adjusting for population age structure and gender, suggested that every increase of 100 people per km2 in population density was associated with a 1.0% increase in cerebrovascular mortality rate. (95% CI 0.8–1.2) Additional adjustments for numbers of doctors per population and average household income attenuated the association to a small degree (Table 3).

Table 3. Estimated associations between measures of urbanization and percentage increase in cerebrovascular mortality rate and their 95% confidence intervals (CI) using aggregate data across 76 provinces in Thailand
  Model 1Model 2Model 3Model 4
  1. Model 1 exposure variables in model: population density, age and gender.

  2. Model 2 exposure variables in model: population density, age, gender, average monthly household income and number of doctors per population.

  3. Model 3 exposure variables in model: proportion living in urban area, age and gender.

  4. Model 4 exposure variables in model: proportion living in urban area, age, gender, average monthly household income and number of doctors per population.

 Measures of urbanization in each province (Units)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)
Cerebrovascular mortalityPopulation density (100 people per population)0.99 (0.76 to 1.22)0.93 (0.41 to 1.46)
P < 0.001P < 0.001
Proportion living in urban area (10 per cent)5.81 (4.72 to 6.92)5.24 (2.78 to 6.92)
P < 0.001P < 0.001
Average monthly household income (1000 baht)−2.27 (−2.86 to −1.67)−2.09 (−2.59 to −1.59)
P < 0.001P < 0.001
Number of doctors per population (Doctors per 10 000)12.02 (9.86 to 14.22)10.11 (7.66 to 12.60)
P < 0.001P < 0.001
Figure 4.

Association between proportion living in urban area and population density with standardised mortality ratio for cerebrovascular disease using aggregate data across 76 provinces in Thailand.

The correlation coefficient for the proportion living in an urban area and SMR for cerebrovascular disease was 0.42 (P < 0.001) (Figure 4). In the age- and gender-adjusted model, every ten per cent increase in proportion living in an urban area was associated with a 5.8% increase in cerebrovascular mortality rate (95% CI 4.7–6.9), with little attenuation after further adjustments for average household income and number of doctors per population. Average household income was negatively associated with cerebrovascular mortality, whilst number of doctors showed a strong positive association (Table 3).

Urbanization and malignant neoplasms mortality

The overall mortality rate from malignant neoplasms was 88.3 per 100 000. There was a positive correlation between population density and SMR for malignant neoplasm (Figure 4). In the adjusted regression model, every increase of 100 people per km2 in population density was associated with a 1.0% increase in malignant neoplasm mortality rate (95% CI 0.9–1.1) (Table 4).

Table 4. Estimated associations between measures of urbanization and percentage increase in malignant neoplasm mortality rate and their 95% confidence intervals (CI) using aggregate data across 76 provinces in Thailand
  Model 1Model 2Model 3Model 4
  1. Model 1 exposure variables in model: population density, age and gender.

  2. Model 2 exposure variables in model: population density, age, gender, average monthly household income and number of doctors per population.

  3. Model 3 exposure variables in model: proportion living in urban area, age and gender.

  4. Model 4 exposure variables in model: proportion living in urban area, age, gender, average monthly household income and number of doctors per population.

 Measures of urbanization in each province (Units)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)
Malignant neoplasm mortalityPopulation density (100 people per population)1.00 (0.88 to 1.12)0.96 (0.69 to 1.23)
P < 0.001P < 0.001
Proportion living in urban area (10 per cent)5.46 (4.86 to 6.07)6.27 (5.01 to 7.55)
P < 0.001P < 0.001
Average monthly household income (1000 baht)−0.93 (−1.26 to −0.60)−0.89 (−1.17 to −0.60)
P < 0.001P < 0.001
Number of doctors per population (Doctors per 10 000)4.58 (3.49 to 5.68)2.74 (1.57 to 3.92)
P < 0.001P < 0.001

There was very weak evidence for a positive correlation between the proportion living in an urban area and SMR for malignant neoplasm. (r = 0.20, P = 0.086) (Figure 5). The age- and gender-adjusted model suggests that every 10% increase in the proportion living in an urban area is associated with a 5.5% increase in malignant neoplasm mortality rate (95% CI 4.9–6.1). Additional adjustment for numbers of doctors per population and average household income strengthened the association (Table 4).

Figure 5.

Association between proportion living in urban area and population density with standardised mortality ratio for malignant neoplasms using aggregate data across 76 provinces in Thailand.

Sensitivity analyses

Removing Bangkok as the outlier did not materially change the correlations between urbanization and all-cause mortality, cardiovascular mortality and cerebrovascular mortality.

Population density was associated with a decrease in rate from other causes of death (Appendix 1). The proportion of those living in an urban area was negatively associated with the rate of other causes of death in the age- and gender-adjusted model, but this association completely attenuated after adjusting for average household income and number of doctors per populations.

Further analysis including additional adjustments for number of hospitals in each province did not change the direction of association between the two measures of urbanization and the four types of mortality. The distribution of the proportional mortality between NCD and other causes of death did not differ by number of hospitals within province (Appendix 2).

Discussion

This study found that that urbanization, measured by population density and the proportion of people living in an urban area, was associated with increased NCD mortality in Thailand. Increasing average monthly household income in each province was associated with lower NCD mortality, whilst higher density of doctors appeared to be associated with higher NCD mortality.

We found that population density was negatively associated with all-cause mortality, whilst the proportion of people living in urban area was positively associated with all-cause mortality. Discrepancy in the effects of the two measures could be due to the capture of different aspects of urbanization. Population density by definition captures crowding. The urban proportion in Thailand, by virtue of the criteria for becoming a municipality, captures some aspects of density but will also include access to high-tech health facilities and equipment, and to public health interventions such as sanitation and waste management. This notion is supported by the observation that in comparison with population density, the proportion living in an urban area shows a stronger correlation with the number of doctors per population and average household income.

All-cause mortality is made up of a variety of underlying causes of death. Apart from the top 10 causes of deaths, a variety of other causes of mortality accounted for 55% of total mortality. Each of the specific causes of death might have different associations with urbanization. For example, the association between population density and all-cause mortality was flat or slightly negative, whilst the association between population density and all NCD causes of death was positive. The sensitivity analyses showed that this different directionality of association is likely to be driven by other causes of mortality.

Our findings are consistent with other ecological studies considering urbanization and NCD mortality (Schorr et al. 1989; Smith et al. 1995; Pritchard & Evans 1997; Yang & Hseigh 1998). Petcharoen et al. carried out a similar ecological study using the same databases in 2000 to assess the relationship between socioeconomic status and cardiovascular mortality in Thailand (Petcharoen et al. 2006). In their study, the correlation between the proportion living in urban area and age-standardised cardiovascular mortality rate was 0.41, similar to the correlation found in our study (0.43). There are several plausible explanations for the associations seen between urbanization and NCD mortality. Although one must be careful not to imply causation from such a study design, it is possible that the association between urbanization and mortality is causal and mediated through other risk factors, such as individual life-style factors, social support/access to care and environmental hazards such as air pollution or exposures to possible carcinogens (Yang & Hseigh 1998; Maheswaran & Elliott 2003). Several studies have found links between urbanization and many individual risk factors for NCDs (Sleigh et al. 2008; Allender et al. 2010; Hernandez et al. 2012). Evidence from Thailand suggests that negative health behaviours such as decreasing physical activity, increasing consumption of junk food and fried food, smoking and drinking are associated with urbanization (Young 2001; Kosulwat 2002; Banwell et al. 2009; Lim et al. 2009; Yiengprungsawan et al. 2011b). Thus, it is feasible that the association between urbanization and increasing prevalence of risk factors for NCD could result in increasing NCD mortality, although these may be masked by changes in socioeconomic status. Other studies, using individual level data, found that higher socioeconomic status was associated with lower mortality in Thailand, possibly due to better health behaviour in terms of less smoking and drinking and better access to care (Sethapongkul 1992; Vapattanawong et al. 2008).

A recent meta-analysis provides evidence that fewer social relationships, whether structural or functional, lead to higher mortality (Holt-Lunstad et al. 2010). In Thailand, social relationships such as trust, support and interactions are less strong in urban environments (Yiengprungsawan et al. 2011b).

Reverse causality proposes that severely ill patients with chronic non-communicable disease are more likely to relocate to a more urban area, where it is assumed there will be better access to care (Bentham 1988; Phillips 1993). Although this could be possible for diseases such as cancer, it is less likely for cardiovascular mortality and cerebrovascular mortality because in Thailand the time from event to death is short (Venketasubramanian 1998; Srimahachota et al. 2007).

The positive association we found between number of doctors per population and increasing mortality is consistent with past literature, and several explanations have been offered (Cochrane et al. 1978; Young 2001). One is the fact that urban areas are able to attract more doctors, and the urban populations have higher risk factors. The association seen is not, therefore, causal in either direction. It is interesting to note that adjustment for average household income and number of doctors alters the association between urbanization and NCD mortality differently, depending on cause and which measure of urbanization is being considered, even though the direction of association remains consistent.

There were several limitations to our study. The urban proportion in 1999 was dramatically increased due to a decentralising act, which upgraded existing rural sanitation districts to urban municipalities. This resulted in transformation of more than 700 areas to “urban” overnight, even though their lifestyle and environmental surroundings could be considered rural. This misclassification could lead to underestimation of true associations. There is also potential for uncontrolled confounding by other factors, which we have not been able to adjust for. However, it is unlikely that these issues explain the direction of the findings reported here. Regarding accuracy of death registration, a medical records review for hospital deaths suggested that around 9% of deaths were unregistered (Porapakkham et al. 2010), and for vital registration records, the positive predict values for ischaemic heart disease, cerebrovascular disease, lung cancer and liver cancer (leading causes of deaths from malignant neoplasms for men and women in Thailand) were 65%, 77%, 83% and 86%, respectively (Pattaraarchachai et al. 2010). We found no clear evidence of pronounced differential misclassification at provincial level by numbers of hospitals. Because the symptoms of cardiovascular and cerebrovascular disease are relatively clear and distinguishable from other conditions, we do not consider misclassification of outcome a major problem in this study.

A major strength of this study was the use of national data. This provided consistency in measures of urbanization as well as homogeneity in terms of national culture and lifestyle. The study was also able to adjust for important potential confounders or mediators of urbanization.

Acknowledgements

The authors would like to thank Dan Altman for his helpful suggestions in the analysis plan. C.A is funded by the Faculty of Medicine Development Scholarship from Chiang Mai University, Thailand.

Appendix 1

Estimated associations between measures of urbanization and percentage increase in other causes of mortality rate and their 95% confidence intervals (CI) using aggregate data across 76 provinces in Thailand

Model 1Model 2Model 3Model 4
  1. Model 1 exposure variables in model: population density, age and gender.

  2. Model 2 exposure variables in model: population density, age, gender, average monthly household income and number of doctors per population.

  3. Model 3 exposure variables in model: proportion living in urban area, age and gender.

  4. Model 4 exposure variables in model: proportion living in urban area, age, gender, average monthly household income and number of doctors per population

Measures of urbanization in each province (Units)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)Percentage increase in risk per unit increase (95% CI)
Other causes of mortalityPopulation density (100 people per population)−0.88 (−0.95 to −0.82)−0.79 (−0.93 to −0.66)
P < 0.001 P < 0.001
Proportion living in urban area (10 per cent)−2.83 (−3.12 to −2.54)0.26 (−0.35 to 0.87)
P < 0.001 P = 0.413
Average monthly household income (1000 baht)−0.57 (−0.73 to 0-.41)−1.19 (−1.33 to −1.05)
P < 0.001 P < 0.001
Number of doctors per population (Doctors per 10 000)2.13 (1.58 to 2.68)0.99 (0.38 to 1.60)
P < 0.001 P = 0.001

.

Appendix 2

Proportional mortality by number of hospitals across 76 provinces in Thailand

Figure 6.

Ancillary