Summary of the studies included and the psychometric properties of scales used
Of the included studies one was a cohort study [
8], four were cross-sectional studies [
19,
34,
35,
37] and six were secondary analyses of data [
3,
4,
17,
20,
23,
36]. One was conducted in a developed country [
35] the rest in low to middle income countries. All studies reported an association of urbanicity or urbanisation and health outcomes except one [
4]. Four of eleven included studies used the terms urbanisation and urbanicity interchangeably [
8,
17,
19,
20]. Three of the included studies reported on the validity and reliability of the instruments used [
3,
19,
34] and only one study was assessed to be high quality [
34].
Allender et al. [
8] conducted a cohort study which aimed to evaluate the extent to which urbanisation was a risk factor for self-reported non-communicable diseases. The study was conducted in a representative sample from seven of the nine provinces in Sri Lanka (n = 4,485; response rate = 89.7%; >18 years of age). The authors developed a 7-item urbanicity scale from urban characteristics such as population size, population density, and access to markets, transportation, communications/media, economic factors, environment/sanitation, health, education, and housing quality. They assigned a maximum of 10 points to each item of the urbanicity scale resulting in score from 0 (no urbanicity) to 70 (high urbanicity). The village administrators in 100 study villages provided the relevant information for their village. These scores were grouped into tertiles of urbanicity (1 low urbanicity, 2 medium and 3 high urbanicity) for subsequent analysis. The authors found that urbanicity was positively associated with physical inactivity, high body mass index and diabetes mellitus in men and women. However, the validity and reliability of the urbanicity scale were not reported.
Vavken et al. [
35] conducted a cross-sectional study in 14,507 men and women in Austria (mean age 36 years; 45% male) which aimed to determine the burden of musculoskeletal disease by urbanicity, socioeconomic status, age and sex. They adopted the Nomenclature of Territorial Units for Statistics III classification of urbanicity ranging for 1 (rural areas) to 3 (urban areas) to measure urbanisation. The Nomenclature of Units for Territorial Statistics (NUTS) is a geocode standard for referencing the subdivisions of countries within the European Union for statistical purposes [
38]. The authors referenced the NUTS website but they did not present further detail about the validity and reliability characteristics of the NUTS scale in the paper. They found strong evidence for an association between urbanicity and arthritis and osteoporosis but not spinal conditions.
Jones-Smith et al. [
34] conducted a cross-sectional study in China which aimed to develop an urbanicity scale from existing data, test whether the scale was reliable and valid, and assess whether it provided information beyond what could be determined from the traditional urban/rural dichotomous variables. They used the procedures for building scales developed by DeVellis 2003 [
39] and Netemeyer, Beardon and Sharma 2003 [
40] to construct their scale. This involved first, consulting authoritative sources such as previous literature and content experts to establish a strong definition of the construct they intended to measure. Second, they identified which variables were available to represent those defining concepts as well as how each should be scored. Third, they tested the scales performance as a measurement tools including its uni-dimensionality, reliability, content, criterion and construct validity. They identified 12 main components thought to define urbanicity which were: population density, economic activity traditional markets, modern markets, transportation infrastructure, sanitation, communications, housing, education, diversity, health infrastructure, social services. They allocated a maximum of 10 points each to each of the 12 components. The 12 components appeared to represent a unidimensional underling construct (called urbanicity) as evidenced by high eigenvalue of only one factor in the exploratory factor analysis. The scale had good internal consistency (Cronbach alpha values = 0.85 to 0.89). The scale exhibited temporal stability in test- retest evaluations (correlations r = 0.90 to 0.94). There was some evidence for criterion validity from its comparison to the official classification of communities as urban or rural (Kappa statistic for agreement beyond chance of their scale with the “gold standard” = 0.21 to 0.48). Linear and logistic regression indicated that their scale demonstrated good construct validity: increasing scores on the urban scale were significantly associated with increases in the adjusted per capita household income and with significantly lower odds of having more than one child. They demonstrated that the scale predicted the incidence of overweight/obesity populations in China and added valuable additional information compared to the traditional measure namely the urban–rural dichotomy.
Antai et al. [
37] conducted a cross-sectional study in Nigeria among 2118 children aged less than five years, which assessed whether urban area socioeconomic disadvantage has an impact on under-five mortality. Urban under-five mortality rates were directly estimated from the 1990, 1999, and 2003 Nigeria Demographic and Health Surveys. Urban area disadvantage was measured using the urban area disadvantage index (UADI) score. The UADI scores reflect the overall level of urban area disadvantage based on eight indicators of socioeconomic disadvantage at the neighbourhood level. The UDAI scores were generated through principal component analysis using 165 out of 365 available primary sampling units (PSUs). The PSUs were administratively defined, homogenous areas used as proxies for ‘neighbourhoods’ or ‘communities’ consisting of a minimum of 50 households per PSU. The scale measured ‘urban area disadvantage’ (e.g. children living in a household without piped water, flush toilets or electricity and other amenities) rather than urbanicity (i.e.urban conditions at any given point in time). The authors found that urban area disadvantage was significantly associated with under-5 mortality after adjustment for individual child and mother level demographic and social characteristics.
Monda et al. [
23] used an existing longitudinal dataset from the China Health and Nutrition Survey (CNHS) (wave from 1991–1997; n = 8769, male = 50%; 18–55 years of age) to examine the effect of rapid urbanisation on the occupational physical activity patterns of Chinese adults. The authors utilized a multidimensional measure designed specifically for the CHNS to capture urbanisation from the physical, social, cultural and economic environments. The urbanisation variable was developed using data from community surveys and household-level information and comprised 10 components which were: communication, economic, housing-related and transportation infrastructure, the availability of schools, markets and health care environmental sanitation and population size and density. The data were used to generate a continuous variable called an ‘urbanisation score’ for each community for each data collection period where each component was assigned 10 possible points and summed for a maximum value of 100 points (100 = high urbanisation). The properties of the scale were not further commented on in the paper although readers were referred to a reference (under review) in which further details on the development of the urbanicity index could be found. They found that men had 68% greater odds, and women had 51% greater odds, of light versus heavy occupational activity given the mean change in urbanisation over the 6-year period. Further, simulations showed that light occupational activity increased linearly with increasing urbanisation. After controlling for individual-level predictors, community-level urbanisation explained 54% and 40% of the variance in occupational activity for men and women, respectively. The authors concluded that because occupational activity remains the major source of energy expenditure for adults, the Chinese population is at risk of dramatic increases in the numbers of overweight and obese individuals.
Van de Poel et al. [
17] also used longitudinal data from the CNHS (6484 adults aged >16 years and older) to investigate the role of urbanisation and the spread of non-communicable diseases in China. They developed an urbanicity index by firstly applying factor analysis to a set of 25 community level characteristics (e.g.number of bus stations, dirt roads, primary schools) that reflect a community’s level of urbanicity. They subsequently computed a rank-based measure of inequality in disease risk factors by degree of urbanicity. The first factor was retained as it explained the highest proportion of the common variance among community variables (~47%). Factor loadings were then computed which were the degree to which the remaining characteristics correlated with the first factor and range for -1 to +1. The urbanicity index was then constructed as a linear combination of all these community characteristics weighted by their factor loading using an oblique promax rotation. The authors report that the index had internal consistency, temporal stability, criteria-related validity and construct validity. The authors conclude that their urbanicity index appears to be a plausible indicator of the degree of urbanicity of communities in China. In relation to non-communicable diseases, at the individual level low engagement in physical activity and farming explain more than half of the urban concentration of overweight and a rising share (28%) of the greater prevalence of hypertension in more urbanised areas.
Allender et al. [
20] conducted an investigation into the association of urbanization and non-communicable disease risk factors in 2705 men and women aged 15–64 years in Tamil Nadu, India. They adopted a modified version of a composite continuous measures of urbanicity previously used and validated for the Philippines. It comprised seven elements: population size, population density, access to markets, communications, transport, education and health services. The authors modified it by using only three variables: population size, population density, and education. They assigned a maximum of 10 points to each item to generate a modified scale (range 0 (no urbanicity) to 70 (high urbanicity). They conducted validity testing on the modified scale and obtained a Cronbach alpha reliability coefficient of 0.72. Using this scale urbanicity appeared to be consistent with exiting government definitions or ‘urban’ and ‘rural’. Face validity was discussed by the authors but no other scale properties. The scale was used in conjunction with data collected from 3705 participants in the World Health Organization’s 2003 STEPwise risk factor surveillance survey in Tamil Nadu, India. Linear and logistic regression were used to examine the relationship between urbanicity using this scale and chronic disease risk. Using the urbanicity index the authors found that increased urbanicity was positively associated with body mass index, low physical activity and mean number of servings of fruit and vegetables consumed per day (P < 0.05) in men and women.
Dahly et al. [
3] conducted a study in which they aimed to construct a scale of urbanicity using community level data from the Cebu Longitudinal Health and Nutrition Survey (CNHLS) in the Phillipines [
41]. They used the scale development method of De Velliss 2003 [
39] to validate the new measure and tested its performance against the urban–rural dichotomy. Items included in the urbanicity scale were population size, population density, communications, transportations, education facilities, health services and markets. Each item was scored 0 to 10 scale so the scale ranged from 0 (no urbanicity) to 70 (high urbanicity). The scale had high internal consistency (item spearman correlations r > 0.5, P < 0.001; Cronbach alpha range 0.87 to 0.89), high temporal stability (spearman correlations r = 0.85 to 0.97, P < 0.001) high content validity, criterion validity and construct validity. The new scale illustrated misclassification by the urban–rural dichotomy, and was able to detect differences in urbanicity, both between communities and across time, that were not apparent before. The authors concluded that the new scale was a better measure of urbanicity than the traditionally used urban–rural dichotomy. For example, in generalised linear models applied to the CNHLS data, the scale was found to explain the variation in calorie intake above and beyond that explained by the urban–rural dichotomy alone.
McDade et al. [
4] conducted an analysis of multiple definitions of urbanicity also using data from the Cebu Longitudinal and Health Nutrition Survey in the Philippines [
41]. Factor analysis was conducted on 27 household and 26 community variables using principle components analysis. This generated factor scores that summarised a household’s position with respect to access to infrastructure and health services and level of affluence. Extensive comparisons of factor scores were then made across urban and rural areas, and across settlement types to explore household and community level markers of urbanicity differentiating households in geographically defined urban and rural areas. High population density, the availability of telephone, mail, transportation services, electricity, clean water and health care facilities were found to be the correlates of urbanicity. Apart from factor analysis, the results of further reliability and validity property evaluations were not reported in this paper. The relationship between urbanicity and health outcomes was also not reported in this paper.
Liu et al. [
36] used the CHNS to investigate the effects of urbanisation on health care and health insurance in rural China among 33,404 men and women (mean age 28.9 years) using individual and community surveys. The authors used three sets of variables to construct an urbanisation index: 1) total population of the neighbourhood divided by the area of the neighbourhood; 2) infrastructure variables 3) industrialisation variables. The authors first calculated the distribution of these variables and then defined the uppermost quartile as a high level, the lowermost quartile as the low level and the middle two quartiles as the middle level. They did not describe the psychometric characteristics of the urbanisation index. The primary finding was that urbanisation leads to a significant and equitable increase in insurance coverage, which in turn plays a critical role in access to health care.
Van de Poel et al. [
20] investigated adverse health effects of rapid urbanisation in China using the CHNS panel data for 1991–2004. The authors constructed an urbanicity index using factor analysis on a broad set of characteristics from the CHNS community level data pooled across all survey sites as described previously [
17]. The urbanicity index captured information on population size, land use, transportation facilities, economic activity and public services. The validity of the urbanicity index has previously been reported [
17] and was not described in the paper. The index correlates with a subjective classification of communities as urban, suburb, town or rural. The authors found that greater urbanisation increased the likelihood of reporting of poor health.