A health risk model for rural households based on the distribution of multi pollutants

Since the opening up of the economy in late 1991, India has become a global marketplace. It has increased per capita income among people. These increased income levels have accelerated consumerism. This rapid consumerism, accompanied by improper waste disposal, has resulted in environmental degradation. Though the degradation was more pronounced in cities, with rapid economic expansion, the degradation is also evident in villages. Environmental degradation through air, water and water pollution exposes people to various health risks. Environmental degradation and damage to public health are essential constraints in sustainable economic growth and social development. There are three significant pollution receipts found in villages: (i) air pollution due to dispersed particles, hydrocarbons, CO, CO₂, NO, NO₂, SO₃, etc., (ii) water pollution due to organic, inorganic and biological discharges at high levels and (iii) soil pollution through the release of chemicals, heavy metals, hydrocarbons and pesticides.

Studies on the effect of pollution on human health have become a global research interest over the last decade. They have proposed various assessment methodologies to reduce the chances of significant uncertainties (Cohen et al. 2004; Pope et al. 2009; Burnett et al. 2014). Several researchers have estimated health risks due to pollution in the Indian context (Madheswaran 2007; Silva 2015; Chowdhury & Dey 2016; Kumar et al. 2016; Maji et al. 2017a, 2017b, 2018; Balakrishnan et al. 2018; Saini & Sharma 2019; Bherwani et al. 2020; Manojkumar & Srimurganandam 2021). Integrated exposure-response (IER) model (Kumar et al. 2016) estimation of premature deaths due to PM_2.5 exposure, non-linear power model (Kumar et al. 2016) to estimate premature death due to air pollutants, monetary cost-based health assessment studies using methods like the cost of illness, contingent valuation, hedonic wage (Madheswaran 2007; Silva 2015; Maji et al. 2017a, 2017b; Bherwani et al. 2020) and labour output-based health risk assessment (Pandey et al. 2021) is some of the essential works in the Indian context. Most of the studies in the Indian context are based on mortality rate and monetary burden and focus on cities. Also, the models proposed by these researchers were tied to a single pollutant factor. The health risk from exposure to pollution occurs in the rural and urban populations (Garaga et al. 2018) in India. However, most of the existing works are exclusively confined to urban centres. Real-time monitoring (Bhowmik et al. 2022) can fill this gap and provide a more comprehensive understanding of the nature and distribution of health exposure in Indian villages. This work proposes a comprehensive health risk prediction model for Indian villages that integrates real-time monitoring (Mucchielli et al. 2020) of multiple pollutant factors, with a higher coherence to World Health Organization (WHO) benchmarks. Without a comprehensive study covering Indian rural households, understanding the nature and distribution of health exposure in Indian villages is very difficult. Most existing studies are based on PM_2.5; there are few works on other strong sources in the Indian context, like biomass cooking, trash burning and landfills due to agricultural pesticides and household chemicals. This works addresses this gap and proposes a comprehensive health risk prediction model for Indian villages in terms of multiple pollutant factors.

In this section, an overview of different health risk assessment models is provided, including those used in both international and Indian settings.

Pope et al. (2009) modelled the health risk due to pollution in terms of life expectancy. The changes in life expectancy are analysed in correlation with particulate matter in the air. The regression model is built between the air pollutant levels and life expectancy. The model is adjusted for socio-economic and demographic variables. The model is built at the macro level based on limited observation of air pollutants in selected metropolitan areas of the US.

Burnett et al. (2014) proposed an IER model with a relative risk (R.R.) on respiratory problems as output and ambient indoor air pollution caused by solid cooking fuels and smoking. All these pollutant factors are converted to estimate annual PM_2.5 exposure equivalents and fitted into the IER model. The model considered only air pollutant factors, and it is a macro-level indicator.

Kumar et al. (2016) did air quality mapping and health impact assessment for Mumbai city. From air quality observations made at a particular location, spatial variation over a large area is made using ArcGIS interpolation techniques. The health impact assessment was made at ward levels based on the air pollutant level of nitrogen dioxide (NO₂), sulphur dioxide (SO₂) and suspended particulate matter (SPM). The health cost was estimated for each ward. It is difficult to isolate the health cost due to air pollutants alone. In Mumbai, there are other significant factors like sewage, water quality, etc.

Silva (2015) discussed that the design and architecture must prioritize sustainable practices and take into account the needs of diverse populations in order to create healthy and functional urban environments. Along with several case studies and examples of successful sustainable design practices, including green roofs, urban agriculture and pedestrian-friendly design, the importance of interdisciplinary collaboration and community engagement in creating sustainable cities was significantly pursued.

Maji et al. (2017a) proposed an epidemiology-based exposure-response function. The function fitted mortality and morbidity to PM_2.5 exposure over 24-year data. The fitness function is adjusted for disability-adjusted life years (DALYs). The fitness result is transformed into economic costs. The study was conducted in Mumbai city. The same author in Maji et al. (2017b) extended the work for Agra city by incorporating more air pollutant factors. The model could predict health risk in terms of health cost. Nevertheless, extending this study to the village context is impossible as no dependent metrics were available for villages.

Chowdhury & Dey (2016) developed a non-linear power law (NLP) function to estimate the relative risk in terms of mortality due to ambient PM_2.5 exposure. Satellite observations of PM_2.5 were used to predict premature death using the NLP function. Though the model was simple to apply at fine-grained district levels, it could not provide risks to other health factors like physical disabilities resulting from other pollutants.

Maji et al. (2018) correlated the PM_2.5 levels to health risk in terms of mortality using the data collected from 13 major cities. It is a macro-level study demonstrating a significant relationship between mortality and PM_2.5 levels.

Balakrishnan et al. (2018) used PM_2.5 concentration to estimate death mortality by adjusting for DALYs. The study was conducted at the macro level of states. The study can be used for budget planning but needs to be applied at the fine-grained level of villages for designing effective action plans.

Saini & Sharma (2019) predicted premature death from PM_2.5 levels using the IER model. Premature death is estimated for each specific problem of stroke, chronic obstructive pulmonary disease, lung cancer and lower respiratory infection.

Manojkumar & Srimurganandam (2021) developed a model correlating the PM concentrations to mortality and hospital admissions. The study was conducted in major Indian cities. Hospital admission count due to respiratory and cardiovascular problems is correlated using linear regression with the PM levels. With the disparity in hospitals across cities and villages, this model can only be used to assess health at the macro level.

Pandey et al. (2021) correlated premature deaths after adjusting for DALYs with indoor and outdoor particulate matter pollution. The study was conducted for each Indian state. The estimation was then used to fit the cost of illness method to provide the economic impact of air pollution.

Lu et al. (2017) used simultaneous equation modelling (SEM) to analyse the relationship between health and environmental pollution. The study was conducted across China. Air pollutants factors and wastewater emissions are collected over many years and fit the SEM model. The model was able to predict mortality in terms of pollutant factors.

Wu et al. (2020) estimated healthcare expenditure with increased pollutants. The study was conducted on the pollutant data collected for about 21 years from Taiwan. The data were transformed into time series data, and wavelet analysis was conducted. The model correlated the healthcare expenditure to influencing wavelet coefficients. The model requires a large volume of data.

Hao & Gao (2019) proposed a quantitative relationship between environmental pollution and public health using the expanded Grossman health production function. Pollutant factors in sulphur dioxide and industrial smoke dust emissions are fitted to health risks in terms of mortality rates.

Karambelas et al. (2018) designed a correlation model for the health impact due to ambient air pollution. The model was based on an analysis of levels of PM_2.5 and O₃ and their correlation to the mortality rate over the years. All the air pollutant factors were normalized to PM_2.5 levels, and linear regression was fit between mortality and PM_2.5 levels.

Ravishankara et al. (2020) estimated premature death mortality in Indian states based on satellite-derived surface PM_2.5 levels. The study was fine-grained, and death mortality was estimated for six major diseases listed in Global burden of Diseases 2017.

Koul (2021) estimated death mortality after adjustment with DALYs based on three air pollutant factors: ozone, particulate matter and indoor pollution. Like Ravishankara et al. (2020), this study was fine-grained, with death mortality estimated for all six significant diseases listed in Global Burden of Diseases 2017.

Ranzani et al. (2020) analysed the health risk of indoor household pollution in terms of bone mass. The study was conducted in five semi-urban places in India. Separate linear mixed models were fitted between the PM_2.5 levels and black carbon levels to the bone mass. The lower bone mass levels are associated with higher PM_2.5 levels.

Behera et al. (2012) estimated the health risks due to groundwater pollutants. Well-known water quality parameters like pH, R.C., turbidity, fluoride, hardness, etc., were collected from the Jagadalpur district. The impact of water quality on perceived health was analysed through a survey study. Nevertheless, the study did not provide any model correlating groundwater pollutants to health risks.

James et al. (2020) analysed the impact of cooking fuels on rural women's health. A study was a community-based cross-sectional survey across four villages in Karnataka to estimate health risk in self-reported ophthalmic, cardiovascular and dermatological symptoms with exposure to various cooking fuels. The association between cooking fuels and symptoms were modelled using regression (Rathnamala et al. 2021).

The summary of the models is presented in Table 1.

The methodology used in this context involves the development of a new model for health risk assessment that considers multiple pollutant factors with perceived health risk assessment, and is coherent with quantitative benchmark-based health risks. The study aimed to overcome the limitations of existing models that are based on limited pollutant factors and estimate health risks in terms of mortality, which is not adequate for accounting for various health abnormalities and loss of livelihood.

To achieve this, the researchers designed a structured questionnaire with 33 questions in four dimensions: water supply, drainage, air pollutant and solid waste, explicitly tailored to the context of Indian villages. The perceived health risk factors were designed by extending the Short Form 36 Health Survey (SF-36) (Treanor & Donnelly 2015), which has been widely adopted by numerous public and private healthcare organizations across various countries (Jenkinson & Layte 1997; Gandek et al. 1998; Kodraliu et al. 2001; Hanmer et al. 2006; Guerra & Shea 2007; Kontodimopoulos et al. 2007). However, the SF-12 was chosen for extension due to its applicability to a broad group of general and vulnerable populations (Côté et al. 2004; Jakobsson 2007; Pezzilli et al. 2007; Tang et al. 2008; Wee et al. 2008). The behavioural risk factors of SF-12 were extended with risk factors specific to pollutant contexts and used as perceived health risk factors (Rathnamala et al. 2020).

The methodology used in this study involved the development of a new model for health risk assessment that accounted for the unique context of Indian villages and included perceived health risk factors related to multiple pollutant factors. The study used a structured questionnaire and extended the SF-12 to create perceived health risk factors specific to the pollutant contexts in Indian villages. This methodology aimed to provide a more comprehensive and accurate health risk assessment that accounted for various health abnormalities and loss of livelihood, which was not possible with existing models that focused solely on mortality.

Table 2 presents pollutant factors that are grouped into four categories: water supply factors, drainage factors, air pollutant factors and solid waste factors. Each category includes several sub-factors that contribute to perceived health risks in the context of Indian villages. The perceived health risk factors listed in the table include hypertension, cancer, heart disease, gastrointestinal illness, asthma/COPD, psychiatric disease, frequent diarrhoea, skin problems and frequent illness. The table provides a comprehensive list of the factors that the study considered in developing a new model for health risk assessment that is coherent with quantitative benchmark-based health risks and considers multiple pollutant factors.

Each perceived health risk factor collects responses on two scales (Yes/No). Based on the respondents' perceived health risk factors, the health risk is categorized into three levels: Level 1 (L1), Level 2 (L2) and Level 3 (L3). The mapping is given in Table 3.

Deviating from earlier works on modelling the relationship between the pollutant factors and health risk as a linear model, this work proposes a fuzzy model to estimate the health risk in terms of scores for each level (L1, L2 and L3).

is the probability density function for the variable and is the joint probability density function.

From the data collected (F1–F33) from participants, significant factors are found using S.E. analysis (Equation (1)). The significant factors are fit for L1, L2 and L3 risk prediction into Equation (7). For any response from rural household (F1–F24), the data are fit into Equation (7) for L1, L2 and L3 to get three health risk scores as output.

From the results, there are only 20% of households above 0.6 score for L1 and 40% of households below 0.4 score, 40% of households are in the score of 0.4–0.6. For L2, there are only 20% of households below 0.4 score. 50% of households are in the score of 0.4–0.6, 30% of households are above score of 0.6. For L3, there are only 30% of households below 04 score. 30% of households are in the score of 0.4–0.6. 40% of households are above 0.6. 40% of households above L3 score of 0.6 is alarming and indicates an onset of severe health risk in these households.

The obtained results suggest that the proposed health risk prediction model is consistent with existing works that use pollutant factors to estimate health risks associated with air pollution. The study compares the perceived health score provided by the proposed model with the most used pollutant factors in existing works, such as PM_2.5.

The proposed model has higher for any one of scales of L1, L2 and L3 for most of the pollutant factors as seen in Table 4.

Table 4

Correlation of THE proposed health risk score to pollutant factors

Pollutant factor .	.	Fitness .
SO2	L2 = 0.89
NO2	L1 = 0.84
Total dissolved salt (TDS)	L2 = 0.98
Fluoride (F)	L2 = 0.81
Total hardness (T.H.)	L1 = 0.86
Iron	L1 = 0.88

Pollutant factor .	.	Fitness .
SO2	L2 = 0.89
NO2	L1 = 0.84
Total dissolved salt (TDS)	L2 = 0.98
Fluoride (F)	L2 = 0.81
Total hardness (T.H.)	L1 = 0.86
Iron	L1 = 0.88

View Large

The value for most of the pollutant factors is more significant than 0.8. This signifies a higher consistency of the proposed perceived health score with most pollutant factors. The significance is achieved against one of the L1, L2 or L3 scores, justifying the reason for modelling the health risk as a fuzzy decision on pollutant factors.

Three important salient features of the proposed health risk prediction model are its simplicity, effectiveness and adaptability. Compared with the IER model (Saini & Sharma 2019), NLP function (Chowdhury & Dey 2016) and epidemiology-based exposure-response function (Maji et al. 2017a), the proposed health risk prediction model does not need pollutant measurements over a long period. Over time, pollutant observations are unavailable for rural Indian areas. It is costly to collect those parameters considering the large village distribution in India. The proposed health risk prediction model evaluates health risk at a perceived level based on a 24-item questionnaire response. The questionnaire responses are straightforward to collect. Considering the recent in-depth penetration of Smartphone revolution in India, this survey question can be easily launched as a mobile application. Feedback can be collected, and health risk scores can be provided instantly. A perceived health evaluation approach (Behera et al. 2012; James et al. 2020) lacks this effectiveness as they need pollutant measurements. Also, the approaches (Behera et al. 2012; Chowdhury & Dey 2016; Maji et al. 2017a; Saini & Sharma 2019; James et al. 2020) lacks adaptability. They are inflexible to adding new pollutant factors and perceived health risk factors. However, the proposed health risk estimation model scores best in adaptability as the model can be extended for new pollutants and health risk factors.

Air pollution is a significant public health concern in India, where pollutant observations are often unavailable in rural areas due to high costs and limited resources. To address this challenge, the proposed health risk prediction model offers a simple, effective and adaptable approach to evaluating health risks at a perceived level based on a 24-item questionnaire response.

The IER model is a commonly used method to estimate the health impacts of air pollution exposure. IER uses a linear model to relate exposure to health outcomes, but it requires data on pollutant concentrations over a long period of time to estimate the exposure-response function. This approach may not be feasible in rural areas where pollutant measurements are limited, which highlights the importance of the proposed health risk prediction model's simplicity and adaptability. However, IER has been shown to be effective in estimating health risks in urban areas where pollutant measurements are available (Fann et al. 2012).

Another method that has been used to evaluate health risks associated with air pollution is the NLP function. This method assumes a non-linear relationship between exposure and health outcomes and is particularly useful for short-term exposure assessments. However, it also requires pollutant concentration data and is therefore limited in its application in areas where such data is unavailable. A study conducted in China showed that the NLP function had a better performance in predicting daily hospital admissions for respiratory diseases than other models (Liu et al. 2019).

Epidemiology-based exposure-response functions have also been used to estimate health risks associated with air pollution. These functions are based on observed associations between air pollution and health outcomes in epidemiological studies. However, they also require pollutant concentration data and may not be feasible in areas where such data is limited. A study conducted in Canada used an epidemiological approach to estimate the burden of air pollution on premature mortality (Brook et al. 2010).

Perceived health evaluation approaches have been used to assess health risks associated with air pollution in areas where pollutant measurements are limited. These approaches use self-reported health outcomes to estimate the health impacts of air pollution exposure. However, perceived health evaluation approaches lack effectiveness as they rely on subjective measures of health outcomes and may not accurately reflect actual health impacts. A study conducted in China compared perceived health status with actual health outcomes and found that the two were not always consistent (Ye et al. 2013).

The proposed health risk prediction model offers an adaptable approach to evaluating health risks associated with air pollution. The model can be extended to include new pollutant factors and perceived health risk factors, which is a significant advantage over the other methods discussed. Additionally, the use of a questionnaire-based approach to collect data on perceived health outcomes makes the proposed model easily deployable as a mobile application. This feature can significantly reduce the cost and time associated with data collection, especially in areas with limited resources.

The proposed health risk prediction model is a novel approach to estimate health risks associated with air pollution, water pollution and landfill factors in rural households in India. The model uses a 24-item questionnaire to provide three-scale qualitative scores for rural households. Compared with expensive chemical tests based on inferences, the proposed model is simple and cost-effective, making it suitable for rural Indian villages where pollutant measurements may not be readily available. Additionally, the model can be realized using semi-skilled staff, further reducing the cost and technical expertise required.

One of the main benefits of the proposed model is its simplicity and cost-effectiveness. As mentioned, existing approaches based on chemical tests and pollutant measurements can be expensive and may not be feasible in rural Indian villages due to logistical and financial constraints. The proposed model overcomes these limitations by using a questionnaire-based approach that is easy to administer and cost-effective.

Furthermore, the proposed model was found to have higher consistency compared to benchmark air pollutant, water pollutant and landfill factor methods. This indicates that the proposed model can provide accurate and reliable estimates of health risks associated with pollution in rural households in India.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.