Exploring urban health inequities: the example of non-communicable disease prevention in Indore, India

ABSTRACT

The dynamics of urban settings can increase health disparities. This paper explores techniques for examining urban health inequities of non-communicable disease risk, using data from the city of Indore, Madhya Pradesh, India as an example. We analyzed non-communicable disease indicators by gender, wealth, education, slum status, housing type, and location, using a 2018 city-level dataset collected in Indore with a final sample size of 3,070. Four techniques for equity analysis were used including bivariate ratios, concentration indices, geographic information system heat mapping, and multivariate regressions. We found that in Indore, behavioral risk factors such as tobacco, alcohol, and salt intake were more likely to be borne by those with low education and income, slum status and temporary housing type, while diseases such as hypertension were borne more equally over the population. This analysis has shown techniques that urban health researchers and planners can use to understand differentials of non-communicable disease risk and where action can be taken to reduce inequities. Use by city planners will be limited by technical feasibility and production of understandable information. We discuss implications and next steps for Indore as an example for other cities.

KEYWORDS:

• Non-communicable disease
• India
• equity
• urban

This article is related to:

Research for city practice

Introduction

The dynamics of urban settings can increase health disparities, leading to wide variations in risk of poor health within the same city (Kumaresan et al. 2010, Grant et al. 2017). However, disease prevalence data are rarely collected at the city level, and it is even rarer to explore intra-city differences in disease risk by sub-populations. Municipal decision-makers need data on these disparities in order to effectively allocate resources.

Disparities in risk point to inequitable access to health-related services and to the conditions and resources that strongly influence health – otherwise known and the ‘social determinants of health’ (Bonnefoy et al. 2011, Braveman et al. 2017). Identification of disparities in disease prevalence can detect vulnerable populations, facilitate further identification of the inequities in social determinants of health that those populations face, and rectify the inequities to reduce these disparities.

In this paper, we will explore the presence of health inequalities and inequities of non-communicable disease risk, using data from the city of Indore, Madhya Pradesh, India as an example. Health inequalities refer to differences in health regardless of whether those observed differences are fair or can be avoided by reasonable means (Arcaya et al. 2015). Health inequities are a specific type of inequality that judges that inequality to be in some way preventable and unnecessary, and therefore unjust or wrong. The goal of this paper is to provide guidance to cities on identifying and acting on preventable differences in health. This goal is served by two objectives, which are to a) show the range of techniques to look at differences in health, and b) provide an example from Indore on documenting these differences and uncovering inequities of risk after accounting for unavoidable differences between groups.

Non-communicable diseases are an important set of conditions to look at when considering inequity. In 2016, it was estimated that 71% of all mortality worldwide was caused by non-communicable diseases (NCD Countdown 2030 collaborators 2018). Much of this burden is considered preventable. In India, the site of our example city, the three leading causes of mortality are cardiovascular diseases, respiratory diseases, and diabetes, all non-communicable diseases that carry a substantial financial burden for care (Arokiasamy 2018). Previous research suggests that the burden of many of these non-communicable diseases falls more heavily on those least able to afford them (Engelgau et al. 2011, Kundu et al. 2018). We defined non-communicable diseases based on the World Health Organization non-communicable disease targets (WHO 2013). This analysis will also enable development of more nuanced recommendations on how to reach World Health Organization non-communicable disease targets in urban settings.

Background

Urban contexts can exacerbate health disparities by widening gaps between communities, households, and individuals, either by accident or by design. The World Health Organization’s (WHO) Commission on Social Determinants of Health (CSDH), which has produced several frequently cited documents in the global literature, has theorized what drives health inequity and how to measure it. CSDH has a stated goal of supporting countries and global health partners in addressing the social factors leading to ill health and health inequities (WHO and CSDH, 2008). The Commission supports a holistic approach, stating that the best way to reduce health inequities is to address the social determinants of health. These include access to health care, schools, and education, as well as the conditions of work and leisure environments in homes, communities, towns, and cities.

To inform empirical work on health equity, CSDH tested 12 domains of equity for their technical feasibility, reliability, and validity, and their communicability and usefulness to policymakers (Blas et al. 2016). The main findings indicated that one key barrier to regular measurement of equity was producing understandable and useful information that can be translated for decision-makers. The domains found to have the highest programmatic relevance included income/poverty, knowledge/education, housing/infrastructure, travel and gender norms. The technical feasibility of producing indicators in these domains varied widely. In another paper put out by CSDH, Bonnefoy et al. (2011) defined specific ways to measure and monitor equity, based on the CSDH theoretical framework. The most common stratifiers included socioeconomic measures like education, occupation, and income; gender; social groupings (e.g., ethnicity or caste); and place of residence.

Refocusing on urban India to provide context for our Indore example, Ravindran and Gaitonde (2018) note that the evidence base on health inequities in India is limited and narrow. However, their 2018 book provides a synthesis of what literature does exist and a contextualization of equity theories in India (Ravindran and Gaitonde 2018, Srinivas 2018).

Narrowing the scope further to empirical evidence on inequality and inequity of non-communicable disease (NCD) burden in urban Indian contexts, the primary finding appears to be that most often those with lower incomes bear a heavier risk burden (Hosseinpoor et al. 2012, Bachani 2017). However, for individual NCDs the relationship between income and risk may not always work in the same direction (Vellakkal et al. 2013). Education is also often mentioned as a factor, but again the direction of the relationship changes by individual NCD (Gupta et al. 2010, Ravikumar et al. 2011, Joy et al. 2017, Tripathy et al. 2017). The same is true for gender: women in urban India often have higher rates of overweight and diabetes, while men often have higher rates of smoking and hypertension (Allender et al. 2010, Thankappan et al. 2010, Thakur et al. 2019). While theory suggests that the built environment, including housing and infrastructure, plays a large role in creating inequity, little empirical work has been done on the inequities of risk for NCDs in India or elsewhere between slum and non-slum status, or by housing type. When these have been included, the results were not significant (Thankappan et al. 2010, Rooban et al. 2012). Neighborhood differences in risk were found for individual NCDs in Bhopal, India, as well as the states of Rajasthan and Kerala (Gupta et al. 2018, Valamparampil et al. 2018, Banerjee et al. 2020).

Within Indore specifically, no published studies were found in the last 20 years on citywide prevalence of any of the primary NCDs, nor was there any published research on dimensions of urban health equity for any other diseases in Indore. The lack of data on these topics is in part why Indore Smart City Development Limited (ISCDL), the governing body of Smart City work in Indore, asked the United States Agency for International Development-funded Building Healthy Cities (BHC) project to collect data on these issues. Further description of this partnership’s efforts on NCD prevention can be found in WHO’s ‘Power of Cities’ report (WHO 2020). BHC’s broader social determinants of health work are described at www.jsi.com/buildinghealthycities. The analysis done for this paper not only provides an example of equity analysis for urban settings, but also fills a crucial information gap for Indore.

Materials and methods

Dataset

We used a cluster-randomized survey, representative to the city level, using the internationally validated WHO STEPwise approach to NCD risk factor surveillance (STEPS) (World Health Organization n.d.). STEPS surveys are generally household-based and interviewer-administered, with scientifically selected samples. A 2015 review of STEPS surveys found that 93 countries had carried out national surveys (including India), and 21 of those had conducted some form of subnational STEPS surveys (Riley et al. 2015).

The data collection in Indore was carried out by trained investigators following STEPS procedures in May and June 2018. Thirty wards were randomly selected using a probability proportional to size (PPS) method after stratification by presence of slums. Within each ward, three colonies were then selected by PPS; within each colony, 35 households were randomly selected, with one adult between 18 and 69 years old responding for the household. Of the total 3,150 households selected, 97.4% agreed to participate. The final sample included 30 wards, 90 colonies, and 3,070 households/respondents (n = 3,070). Data were collected on all three sections of the STEPS module, which included the main interview questionnaire, physical measurements, and biochemical measurements.

Environmental data at the community-level were also collected via a neighborhood environmental assessment tool, which was adapted from Environmental Profile of Community Health and Community Health Environment Scan Survey tools and previously validated in a study done in Ballabhgarh (Wong et al. 2011, Corsi et al. 2012, Rath et al. 2018). These environmental data were collected from the same 30 wards and most (93%) of the same colonies as the individual dataset (there were some logistical issues reaching the remaining six colonies). GPS coordinates were recorded by using the Differential Global Positioning System (DGPS) instrument. Ward-level shape files were provided to the authors by ISCDL.

Survey coordinators encountered several challenges during the assessment. Data collection during the hottest season meant that investigators worked in extreme heat. In addition, Ramadan fell during the data collection period, resulting in missing data since accurate blood glucose measurements were not available for 120 participants. No cases were discarded due to single indicator missingness, or due to overall data quality issues.

Measures

The following stratifying measures were used for looking at differences in NCD outcome.

Gender

This was self-reported male or female for each respondent.

Education

Education levels were consolidated into a categorical variable that allows for no education, primary education, secondary education, or college and above.

Slum status

Categories of colonies were collapsed to generate a dichotomous variable identifying slum designation (slum or non-slum), based on advice from the Indore Municipal Corporation.

Housing type

Type of house was described by survey respondents with three response options: ‘pucca’ (permanent/solid), ‘semi-pucca’ (semi-permanent), and ‘kacha’ (temporary). For the purposes of these analyses, we grouped semi-permanent and temporary together, as an ‘other than permanent’ grouping.

Wealth index

A composite measure of household wealth was constructed following guidance from the Demographic and Health Surveys (DHS) program (Rutstein 2008).¹ A previous STEPS survey also included housing type (Khalequzzaman et al. 2017), and based on this precedent, this measure was included in the wealth index as well. The wealth index score generated by the principal components analysis was then divided into quintiles.

Geography

GPS coordinates were recorded by the field team at the end of each individual adult interview. This was matched to ward-level shape files provided by the city. Bivariate mapping used two community-level indicators relating to the built environment from the environmental module:

Park Attribute Score. A composite score of park/green space attributes was determined in each ward. For each park/greenspace found, enumerators rated them yes/no for the following attributes: free space for physical activity, completed boundary wall separating space from road, available benches and/or sitting arrangements, garbage bins present, adequate street lights, no smoking signs present, open air or indoor gym present, and if walking paths were present. A score of 0 means they have none of these, whereas 1 indicates all attributes are present.

Ward Average of Tobacco Advertising per 1000-Meter Street Length. In each colony, enumerators conducted a transect walk of the main street and counted various factors, including the number of visible ads for tobacco products. This indicator averages the colony results up to ward level to match the lowest level of shape file available for Indore.

Table 1 provides summary statistics of these measures.

Table 1. Summary Statistics of Stratification Measures.

Measure*	n	Mean	Min	Max
Individual Level (maximum n = 3070)
Gender (Male)*	3,070	36%	0	1
Education (average level)	3,062	1.56	0	3
Slum Dwellers	3,070	29%	0	1
Have Permanent Housing	3,070	71%	0	1
Average Wealth Index	3,020	3	1	5
Ward Level (Maximum n = 30)
Park Attribute Score	26	0.71	0.22	1
Tobacco Advertising per 1000-Meter Street Length	30	12.35	0.52	47.70

*Table presents data with no survey weights added. For all analysis, we applied age-gender weights to match the general age and gender breakdown of Indore City population.

NCD outcome measures

We assessed self-reported NCDs and behavioral risk factors for those diseases based on the WHO STEPs methodology and aligned with the NCD targets laid out by the Global action plan for the prevention and control of NCDs 2013–2020 (WHO 2013). We used the individual-level risk factors for NCDs, which are listed in Table 2 (we have split the WHO indicator on diabetes and overweight into two separate measures).

Table 2. WHO 2025 NCD Risk Factor Indicators.

Harmful use of alcohol Age-standardized prevalence of heavy episodic drinking among adults, as appropriate, within the national context

Physical inactivity Prevalence of insufficiently physically active persons aged 18+ years (defined as less than 150 minutes of moderate-intensity activity per week, or equivalent)

Sodium intake* *In lieu of WHO measure, this is prevalence of respondents reporting that they often/always add salt right before eating

Tobacco use Prevalence of current tobacco use among persons aged 18+ years

Raised blood pressure Prevalence of raised blood pressure among persons aged 18+ years (defined as systolic blood pressure ≥140 mmHg and/or diastolic blood pressure ≥90 mmHg)

Diabetes (raised blood glucose) Prevalence of raised blood glucose/diabetes among persons aged 18+ years (defined as fasting plasma glucose concentration ≥ 7.0 mmol/l (126 mg/dl) or on medication)

Overweight and obesity Prevalence of overweight and obesity in persons aged 18+ years (defined as body mass index ≥ 25 kg/m2 for overweight and body mass index ≥ 30 kg/m2 for obesity)

Analysis

As noted in Bonnefoy et al. (2011), measures of health inequities are never perfect; given the diversity of advantages and drawbacks, they suggest using several techniques. Table 3 describes the four techniques we have adapted from this source, keeping in mind our objective of providing an example of feasible equity measurements for urban planning. The first three techniques are more accessible in that they are easier to calculate and understand, but are based on bivariate comparisons and are thus measures of inequality; users will need to be cautious on applying value judgements related to inequity from just these techniques. The final technique listed is able to provide a better sense of inequity by controlling for other factors that could justifiably explain differences in risk. While it takes more effort to compute and is harder to interpret, it can inform a more nuanced strategy for tackling health risk.

Table 3. Analytical Techniques.

Technique	Outcome Measure	Dimension of Inequity	Reason for Inclusion
Bivariate Ratios	Relative occurrence of NCD risk factors	Gender, education, housing type, slum status	Easy to calculate and interpret, provides basic entry point for understanding of binary inequalities
Concentration Index	Actual vs. expected distribution of NCD risk factors	Wealth index	Provides a slightly more nuanced analysis of wealth inequality than the bivariate ratios; allows for controls on sociodemographic data
GIS Heat Mapping	Relative occurrence of NCD risk factors	Geographic location, environmental factors	Adds a visual description of an additional dimension of inequality
Multivariate Regression	Predicted probabilities of NCD risk	Gender, education, housing type, slum status, wealth index, sociodemographic factors, environmental factors	Allows consideration of multiple overlapping dimensions of inequity; enables controls on sociodemographic data to removed underlying, unavoidable differences by group.

Source: Adapted from Bonnefoy et al. (2011).

All statistical analyses to apply these techniques used Stata 16.1 SE (StataCorp 2020) except for the final steps of heat mapping, which was completed in Excel and ArcGIS (ESRI 2016). Age-gender survey weights were applied to the data to ensure representativeness of the Indore population, based on 2011 census figures.

Results

The results of these analyses, organized by the techniques described in Table 3, provide a better understanding of the heterogeneous risk across Indore. To start, Table 4 shows the average prevalence rates for all of Indore for each of the WHO target indicators.

Table 4. All-Indore Prevalence Rates for NCD Risk Factors.

Indicator	Indore Mean Prevalence, 2018
Harmful use of alcohol	2%
Physical inactivity	24%
Sodium intake	22%
Tobacco use	21%
Raised blood pressure	25%
Diabetes (raised blood glucose)	12%
Overweight and obesity	41%

These city-level prevalence rates are a simple first step toward understanding burden of disease. However, in order to create practical next steps for prevention and control, one needs to dive deeper to see differentials in risk. We can break these averages down by our stratifying measures to begin to investigate inequality (Figure 1(a–d)).

Figure 1. (a) NCD Risk Prevalence by Gender. (b). NCD Risk Prevalence by Education. (c) NCD Risk Prevalence by Slum Status. (d) NCD Risk Prevalence by Housing Type.

*The Education graphic compares Primary to College; No education and Secondary are left out of the graphic for visual brevity.

From these single-factor descriptive statistics, it appears that physical activity, tobacco use, and overweight were the stratifiers that have the most variation, though we cannot determine from these statistics alone whether this is due to preventable circumstances or not. Tobacco use in particular was highly gendered (Figure 1(a)) – its users are overwhelmingly men. Men were also more physically inactive, more likely to have harmful use of alcohol, or have higher rates of hypertension, whereas women had higher rates of overweight or high blood sugar. By education (Figure 1(b)), a higher percentage of those with less education used alcohol, tobacco, and excess salt, or had high blood sugar. Looking at slum status (Figure 1(c)) and housing type (Figure 1(d)), we saw those who might be considered more disadvantaged – those either in slums or with temporary housing types (both denoted by circles) had higher prevalence of usage of alcohol, salt, or tobacco. The more advantaged groups (non-slum or permanent housing type, denoted by squares) usually had higher rates of overweight. It is interesting to see that across education, slum status, and housing type, there appeared to be very little variation in hypertension.

Certainly some of the variation by stratifier relates to inequalities in wealth. This may be the most common measure used in the literature for examining inequality, and one very useful measure of inequality by wealth is the concentration index (CI). This normalizes the scale by which we compare these indicators, which is helpful given the range of prevalence across the WHO indicators.

As shown in Table 5, A CI of −1 means the entire burden falls upon the lowest wealth group, whereas 1 would be a burden entirely on the wealthiest; zero represents true equality. The overall prevalence of alcohol use was low, but appeared to have had the heaviest burden on the poor, followed closely by tobacco use. Overweight and physical inactivity appeared to skew more toward the wealthier. Raised blood sugar seemed to be the one indicator closest to truly equitable distribution, and was not significantly different from zero. While inequality by wealth is often used to infer inequity, one cannot confirm this without controlling for other factors that might explain the difference. For instance, wealth often rises with age, and age is an unavoidable and non-preventable risk factor for NCDs.

Table 5. Concentration Indices of Wealth, by NCD Risk Factor*.

	N	Concentration Index (CI)	Standard Error	P-value
Harmful use of alcohol	3011	−0.20256	0.098005	0.0388
Physical inactivity	3020	0.119506	0.0408	0.0066
Sodium intake	3020	−0.10375	0.021363	0
Tobacco use	3020	−0.19715	0.022814	0
Hypertension	2910	0.085334	0.019021	0
Raised blood sugar	3020	0.021747	0.027479	0.4288
Overweight	2897	0.119824	0.013431	0

*Concentration index ranges from −1 to 1, with 0 representing true equality.

The final stratifier relates to the geographic distribution of risk, which can help explain how these differences might translate to burdens across the city. Figure 2 shows the geographic distribution of the more proximal, or immediate, risk factors. These appear to have had more widely varying geographic distributions than the more distal, or long-term, NCD factors of hypertension, high blood glucose, and overweight and obesity (not shown). This makes sense in that the factors affecting alcohol use, salt intake, physical activity, and tobacco use are highly influenced by the environment one currently lives in, whereas the more distal factors accumulate over a longer period of time.

Figure 2. Geographic Heat Mapping of Proximal NCD Risk Factors.

If we overlay some aspects of the physical environment for illustrative purposes, we can see some of the correlations between these proximal factors and the built environment (Figure 3(a,b)).

Figure 3. (a) Bivariate Map of Insufficient Physical Activity and Park Attribute Score. (b) Bivariate Map of Tobacco Use and Ward Average of Tobacco Advertising per 1000-Meter Street Length.

These bivariate visualizations are simple to run on most data sources and can be more engaging for lay audiences. We see some areas where there was both high NCD risk and poor built environment (here defined by park attributes and tobacco advertising), though these types of geographic analysis are more potent when data are available for every ward in the city. While these maps cannot provide explanations on why there are differences, they can identify exact locations where resources can be targeted to improve the built environment for those most at risk, which goes to the heart of health equity.

Digging into actual inequity requires multivariate analysis, which can compute the compounding nature of multiple factors on an individual’s risk. Running regressions on cross-sectional data to define the causes of NCDs is of limited utility, because these conditions develop over time and there is a risk of omitting factors or trends that occurred before the data were collected. However, cross-sectional regressions can be useful for the purpose of looking at compounded inequity across all stratifiers, subtracting out explainable inequality due to unavoidable factors. Table 6 shows the regression results for individual factors and ward level effects using multilevel logistic regression. All stratifiers are included, as well as the more immutable socio-demographic controls. An illustrative set of lifestyle factors that would theoretically have a correlation with NCD risk are also included to clarify the influence of our stratifiers.

Table 6. Regression Results, by NCD Risk Factor.

Results of the Multilevel Regression Models (Coefficients are Odds Ratios)	Harmful Use of Alcohol	Insufficient Physical Activity	Excess Salt Intake	Tobacco Use	Hypertension	Raised Blood Sugar	Overweight and Obesity
n	967	2858	2858	2858	2810	2848	2848
Stratifiers
Gender (Male = 1)	1.00	0.95	1.36	11.32**	1.14	1.23	0.95
Gender*Education	0.80	1.17	0.88	1.63**	1.21*	0.90	1.04
Interaction between gender and education
Education (No education = 0)
Primary	2.65**	0.94	1.25	0.41**	0.78*	1.02	0.73**
Secondary	1.03	0.90	1.58**	0.11**	0.83	1.05	0.80
College and above	1.00	0.86	1.42	0.06**	0.61	1.23	0.63**
Housing Type (Permanent = 1)	1.82	1.88**	1.19	0.89	0.96	0.65**	0.53**
Slum Status (Slum = 1)	2.61**	1.30*	1.05	1.19	0.99	0.83	0.90
Wealth (Lowest quintile = 0)
Second	0.67	0.77	0.89	0.57*	0.82	1.17	1.57**
Middle	0.44	0.60**	0.83	0.56	1.06	1.33	2.53*
Fourth	0.15**	1.07	0.53**	0.40**	1.13	1.12	3.72**
Highest	0.88	1.01	0.54**	0.49*	1.16	1.08	5.35**
Sociodemographics
Marital Status (Ever married = 1)	2.28	0.55**	1.50	1.85	0.80	2.76*	1.95**
Religion (Other religions = 0)
Hinduism	18.99**	0.68	2.03	1.34	1.16	0.80	0.93
Islam	1.00	0.72	1.91	1.81	1.41	1.32	1.15
Social Group (General caste = 0)
Other disadvantaged castes	0.95	1.15	1.07	0.95	1.01	1.02	1.18
Scheduled caste or tribe	1.75	1.22	1.10	1.45	1.09	0.83	0.96
Age	1.01	1.02**	0.98**	1.00	1.06**	1.05**	1.03**
Lifestyle Factors
Weight (BMI score)	1.04	1.03**	0.99	0.99	1.11**	1.05**	N/A
Professional Advice (Ever received lifestyle advice from health professional = 1)	2.29*	1.07	1.07	0.97	1.52**	1.42	1.29**
Ever Screened (Ever measured for risk by a health professional = 1)	0.58	0.61	1.01	0.96	1.37*	1.53*	1.53**
Tobacco (Current tobacco user = 1)	N/A	N/A	N/A	N/A	1.04	1.10	0.95
Salt (Add extra salt before eating = 1)	N/A	N/A	N/A	N/A	1.11	0.66	0.82
Fruit and Vegetables (Does not meet min. daily consumption = 1)	N/A	N/A	N/A	N/A	0.92	0.76	0.70
Passive Smoke (Exposed to passive tobacco smoke at work or during travel = 1)	N/A	N/A	N/A	N/A	1.04	0.93	0.97
Constant	0.00	0.07	0.15	0.11	0.00	0.00	0.12
Ward-level Random Effects	1.03	1.11	0.39	0.22	0.04	0.18	0.00
Intra-cluster Correlation	0.24	0.25	0.10	0.06	0.02	0.05	0.00
F	7.27**	2.82**	2.09	20.39**	32.43**	16.75**	8.58**

*P values at 95% confidence or better

**P values at 90% −94% confidence

For the majority of these regressions, the stratifiers of wealth and education were significant, with risk generally decreasing with increasing wealth and education. The notable exception is that for overweight and obesity, risk rose with every quintile increase in wealth. Slum status did have some significant associations with harmful use of alcohol and insufficient physical activity, but surprisingly, having permanent housing also was associated with increased risk of insufficient physical activity. Permanent housing was also correlated to decreases in risk of high blood sugar and overweight. Our final stratifier, geography, is measured here with ward-level random effects, and the coefficient is the size of the variance of the ward-level random intercepts from the constant (fixed intercept). The wider the variance, the greater the difference in the outcome variable between wards. An easier statistic to interpret may be the intra-cluster correlation (ICC), which is a good measure of how much random effects explain variation in the outcome. For example, the ICC for alcohol use and insufficient physical activity shows that about 25% of all variation in those outcomes can be explained by variations between wards. This falls to 10% for salt intake, and continues to fall for the remaining factors.

We can see that inclusion of sociodemographic and lifestyle factors does impact the overall effect of our stratifiers. Age and body weight had significant relationships with many risk factors, with risk generally going up by incremental amounts for each increase in year or BMI score. The single biggest predictor across all regressions was the effect of religion on alcohol use; those practicing Hinduism were 19 times more likely to drink than those of other religions (not including Islam, though those practicing Islam were equally likely to drink as compared to the base category, suggesting very low rates of drinking in the base category). We also see correlations with interactions with medical professionals for either treatment or advice, though in many cases these relationships are positive, hinting at reverse causality in cross-sectional models.

Using post-estimation commands provides us with the predicted probabilities of compounded risk and inequity across our stratifiers, including both fixed effects of individual factors and random effects for the wards. By adding those up, Table 7 provides these new compounded prevalence estimates.

Table 7. Predicted Prevalence for Compounded Risk, by NCD Risk Factor.

	Harmful Use of Alcohol	Insufficient Physical Activity	Excess Salt Intake	Tobacco Use	Hypertension	Raised Blood Sugar	Overweight and Obesity
Men
Lowest education, wealth, housing, slum	4%	25%	28%	79%	30%	13%	38%
Highest education, wealth, housing, non-slum	1%	31%	12%	21%	40%	15%	57%
Women
Lowest education, wealth, housing, slum	3%	23%	18%	27%	31%	15%	43%
Highest education, wealth, housing, non-slum	2%	24%	14%	1%	27%	17%	57%

Using these results, the gap in equity of risk becomes clear. If we define our compounded risk groups as those most disadvantaged with lowest education level, lowest wealth index, slum status and temporary housing, and those most advantaged at the opposite end of the scale for those same stratifiers, Figure 4 shows the differential in odds of NCD risk factor between these two groups.

Figure 4. Equity Gap using Predicted Probabilities of NCD Risk Factor.

Tobacco use appears to be the most inequitable risk factor, and compounding the stratfiers actually increases the gap between the advantaged and disadvantaged, most particularly for women: those with lowest education, wealth, housing type, and slum status are 5300% more likely to use tobacco than the most educated, wealthy women who are living in non-slum, permanent housing. Men also see their odds jump up almost 300% between these two groups for tobacco use. Across NCD risk factors, those with compounded disadvantage are more likely to drink alcohol, have excess salt intake and, for women, have hypertension. For men, hypertension burden falls slightly more heavily on those with compounded advantages, as does overweight for both genders. Insufficient physical activity and raised blood sugar appear to have very similar burdens between these two groups, with a slightly higher correlation in the more advantaged groups.

Limitations

Limitations of our data and analysis include that using the STEPs methodology means those under the age of 18 and pregnant women were omitted. Pregnant women and adolescents are critical groups for intervention, especially related to tobacco, alcohol, salt intake, and physical activity. Regarding the GIS maps, we added this component to a cluster-randomized sample, and thus we do not have information on every ward in Indore, reducing the utility of these maps for planning purposes. We hope that including the maps as illustrative examples of the technique here can increase interest in fielding a geographic census that can be fed into city databases. We also did not collect longitudinal data, which means we do not have important time-variant data that would allow for a better fit of our regressions for slow-to-develop NCD risk factors like hypertension, blood pressure, and overweight. Without these, the utility of the individual regression results for those outcomes is somewhat diminished. Finally, additional analysis of other disease burdens beyond NCDs could further refine our understanding of health inequities in Indore.

Conclusions

This analysis has shown four techniques (bivariate ratios, concentration indices, GIS heat mapping, and multivariate regressions) that urban health researchers and planners can use to understand inequalities of non-communicable disease risk and where action can be taken to reduce inequities. When analyzing city-level data, these techniques can inform real, practical recommendations for city planners on how to target health promotion campaigns and interventions effectively across their city. As an example of some of the campaigns and interventions cities might tailor to their needs based on these analyses, The World Health Organization developed a core set of 10 municipal-level interventions to prevent non-communicable diseases (WHO 2020).

Two major barriers to embedding equity analysis into routine city planning are the technical feasibility of producing multivariate analyses, and producing understandable and useful information from those results (Blas et al. 2016). Therefore while the first three techniques may provide less rich information, they may be more attractive to city planners. Indeed, GIS mapping may be the technique most likely to be integrated into planning processes, given the wealth of GIS data most cities already collect and its visually attractive and understandable outputs. As an example from our focus city, Indore Smart City Development Limited created an integrated command and control center that includes GIS maps of multiple city services, which can be easily overlaid with health data once health information systems in the city are fully up and running.

Turning to what these results mean for Indore, we found that risk factors such as tobacco, alcohol, and salt intake were more likely to be borne by those facing multiple disadvantages, including lower education and wealth, slum status and temporary housing. This is true after controlling for unavoidable inequality, pointing to inequitable access to health. Furthermore, about 25% of variation in insufficient physical activity and alcohol intake came from ward-level variation, suggesting the built environment and other community-level factors need to be addressed to tackle these risk factors. While those with more advantages were slightly more likely to be overweight or obese, women were also a major risk group for this risk factor.

Based on these data, we suggest that Indore should target lower-income and -education populations located in slums for public health promotion campaigns to reduce intake of tobacco, excess salt, and alcohol. As a shorthand, the ward-level analysis can be included into the integrated command and control center to target services for non-communicable diseases to geographic hotspots. Those hotspots can be further analyzed for other aspects of built environments like parks, density of unhealthy outlets, safe sidewalks, and others. The city can then enforce regulations and invest in healthy infrastructure in those neighborhoods most at risk, as well as engage local community organizations to help. Building Healthy Cities and Indore Smart City Development Limited will be exploring the environmental data module further in future papers to inform these investments.

Indore can also explore how to reach women through existing city resources like anganwadi centers and self-help groups to spread the awareness of diabetes risk and overweight. Further analysis is needed to understand what other nutritional factors might be interacting with these risks in the female population, and how these may also overlap with increased pregnancy risks. Given the relatively equitable distribution of hypertension and high blood sugar shown in this analysis, Indore health planners may need to consider how to expand public non-communicable disease clinics and practitioners in the poorest areas to ensure equal access to these services for everyone, taking into account that the private sector is also providing these services (Thakur et al. 2011).

These data can provide relevant, practical insights on how to begin tackling the non-communicable disease burden in cities in an efficient way. In Indore, Building Healthy Cities and Indore Smart City Development Limited, along with city sector offices, are working to fold these findings into an Indore healthy city action plan that addresses social determinants of health to create a healthy, more livable city for all.

Acknowledgments

This study was conducted in partnership with Indore Smart City Development, Ltd. (ISCDL) and the All India Institute of Medical Science, New Delhi. We would like to thank the Government Nursing College, Indore, and the Indore School of Social Work for their help in the data collection effort, and to ISCDL for their continued support of this activity in Indore.

Data Availability Statement

The data that support the findings will be available in The USAID development data library at https://data.usaid.gov/following a 1 month embargo from the date of first publication to allow for commercialization of research findings.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

Building Healthy Cities is a five-year cooperative agreement funded by the United States Agency for International Development (USAID) under Agreement No. AID-OAA-A-17-00028, beginning 30 September 2017. The contents of this paper are the responsibility of Building Healthy Cities and do not necessarily reflect the views of USAID or the United States Government.

Notes on contributors

Amanda Pomeroy-Stevens

Amanda Pomeroy-Stevens The USAID-funded Building Healthy Cities (BHC) project refocuses city policies, planning, and services with a multi-sectoral health equity lens while improving data-driven decision-making for three Smart Cities in Asia. BHC (2017-2022) is implemented by JSI Research & Training Institute, Inc. (JSI) with partners International Organization for Migration, Thrive Networks Global, and Urban Institute, and with support from Engaging Inquiry, LLC.

Notes

1. The following variables recommended by DHS exhibited sufficient variance (<95% or >5% prevalence) and were included in a Principal Components Analysis: water source, type of toilet facility, cooking fuel, household assets, use of domestic help, number of household members, and housing type.