Research Proposal

Name: Yile Xu

Word count: 1960

Introduction:

Despite of the considerable improvement of public health conditions over the last decades, issues of health inequality and low coverage of public health services still constantly torment people in developing countries like Kenya. The disparities and inequalities of accessibility to public health care services among different regions in Kenya can be attributed into three categories: affordability, availability, and geographical accessibility. Essentially, all these disparities on health care place burdens on Kenya’s health system and prevent the continuous improvement of its overall health state. Major health burdens on Kenya’s public health system include relative high under-5 mortality rate of 51.3 per 1000 which decreased at a stagnant rate, high maternal mortality rate of 25.7 deaths per 10000, and high mortality caused by both communicable and noncommunicable diseases. Therefore, alleviating health inequality should be considered as one of the most important development goals for Kenya. New technologies are needed for analyzing the complex nature of health system, and my central research question is how data science methods can be applied to analyze and quantify health inequality issues within Kenya and provide insights for the improvement of health disparities and Kenyan’s equitable access to basic health services.

Gaps in literature:

For the optimization of health resources allocation in Kenya’s public health system, the health inequality should be examined from three different perspectives: affordability, availability, and geographical accessibility to the health care services. Based on my method investigation research, variations in geographical accessibility to health care facilities have been actively examined and mapped by different geospatial models. Based on call detail records from subscribers across Kenya, the first geospatial method applies measures like the radius of gyration and cost-distance algorithm to quantify the impact of poor physical access to healthcare on the actual travel behavior and the degree of preventive healthcare uptake (Wesolowski, et al 2015). Another model geographically maps the variations of population’s physical access to emergent hospital services like obstetrics and emergency surgical interventions based on data of transportation networks and hospitals’ geographical locations (Paul, et al. 2018). However, most of the methods fail to geospatially map the variations of the other two factors, availability and affordability. Generally, instead of variation mappings, analysis methods for availability or affordability of health services investigation produce results represented by several typical metrics, like the average availability for a selected basket of medicines and health workforce density. The lack of variation mappings for the availability or affordability of health care is considered as an important research gap to be addressed by this research plan. Moreover, all these state-of-the-art geospatial models have not been tested and applied in the policy making process yet, so a testing and validation process is also needed to confirm the applicability of model.

Research plan:

The objective of my research plan is to address those gaps in current geospatial research by developing an aggregated geospatial model which is able to map the variations of health accessibility from its three major components at the same time, and test it to confirm the applicability. For the data collection, geographical information and transportation network data of the targeted region can be derived either from satellite images or national databases; information regarding the availability of health services like drug storage and health workforce density in Kenya will also be obtained from its national databases or international organizations like OCHA, WHO, and UNICEF; information of individuals’ health conditions and other basic demographic information like the household income will be collected through survey inquiries; and population’s mobility will be then determined using call detail record data.

Due to the fact that there are sufficient geospatial mapping techniques to assess the geographical accessibility to health care services but barely any variation mappings for availability and affordability, the first step for my research is to develop mapping methods for these two perspectives based on current geospatial mapping techniques. However, unlike geographical accessibility modeling which only takes the travel time as a single metric, there are more factors involved in the other two dimensions. The disparities of affordability for all types of health care services are largely dependent on people’s living standards and educational achievements. People in low-income settlement suffer from both higher medicine prices and the health facility user fees which have to be paid before receiving health care. From an assessment of the affordability of medicine for people in low-income settlements, retail medicine prices in the low-income settlements were found to be generally higher than the corresponding international reference prices (Ongarora, et al.). The Public Expenditure Tracking Survey of 2012 found that only 45% of facilities complied with the 10/20 policy which replaced user fees for different health facilities by a flat registration fee of 10 and 20 KES, and user fees continued to take up the majority of facilities’ operating budgets (Ilinca, et al.). Thus, both factors are needed to be taken into consideration when assessing and mapping the affordability. In addition, the availability of overall healthcare can also be attributed into different components including drug and vaccination storage, health care workforce, and life-saving commodities, which all should be incorporated into its mapping method. Take the inadequate health workforce as an example, an analysis of health workforce density revealed significant regional disparities in Kenya that only 10 of 47 counties meet the WHO minimum density threshold of 22.8 workers per 10000 (Emily, et al.).

The next step is to incorporate the mappings for all three different factors into a single aggregated geospatial model. Since public health system is a complex nonlinear system whose smooth operations are closely associated with different social, economic or environmental features, it is necessary to formulate a comprehensive model considering different factors for the mapping of unequal health care. After finalizing the geospatial model, for the testing and validation purposes, my plan is to first focus on some regions with the worst health care and use this model to formulate a hypothetical policy for the improvement and reallocation of the public health services. Then, the corresponding effects and improvement of overall health uptake for this policy will be simulated and then determined based on local health conditions and other demographic information. If possible, we will also work with the local government to actually implement this hypothetical policy on a permissible range and measure its real effects on public health condition after the policy implementation. After comparing the prediction and the real impacts of this policy, the determined disparities between them can be further utilized to modify and validate the hypothetical policy and also our geospatial model.

For the improvement of Kenya’s overall health condition, it is necessary to invest on and conduct this aggregated geospatial mapping research on health inequalities. The two major mapping techniques on which my planning methods are based was proved to be applicable for quantifying the health impact of poor physical access, and also geographically mapping the variations of physical access. The validity and applicability of these two geospatial models support the applicability of my geospatial research. After the validation and modification process, the final model can then be applied in future investigation to map and analyze the health inequalities or various accessibility to health services comprehensively considering its three major components for the targeted region within Kenya. Essentially, those subsequent researches can provide informative insights or even solutions for the optimization of public health resources and alleviation of health disparities in Kenya, which will in turn address its major health issues including relative high under-5 mortality rate, maternal mortality rate, and the prevalence of both communicable and noncommunicable diseases. As Kenya’s health system, as an important part of its development process, will improve and population with poor accessibility to health care services will benefit from my aggregated geospatial mapping research and subsequent inquiries based on it, this project should definitely be considered by the jury.

Possible concerns and solutions:

However, possible concerns and obstacles also exist for this study. Firstly, as mentioned above, unlike geographical accessibility modeling, a few different factors will be incorporated for the geospatial variation mapping of other two dimensions, the availability and affordability to health care services. Then, the mappings for geographical accessibility, availability, and affordability will be incorporated into a single aggregated geospatial model. As there are so many different features involved in this aggregated geospatial model, how to determine the weight for each of these factors will become an important problem and the final aggregated model may subject to overfitting issues due to too many features in it. The weight of each factors in the model can be randomly selected or based on findings from previous research at first, and then set as variables to be continuously modified and updated in the training process. In order to prevent overfitting, traditional techniques to reduce overfitting in machine learning like data augmentation, dropout, and regularization, can be applied and tested to improve the model.

The second major concern is related to the testing of the final aggregated geospatial model. Although simulations can provide some insights for the model’s applicability and the effects of hypothetical policy derived from the model, actual implementation is still needed for a complete assessment. However, it is clearly that the effects of this hypothetical policy can not be tested on a wide range of area as the forming and approval process for a national or state policy will take long period of time, and more time will be needed to measure the achievement of this policy, which will far exceed the one year time frame. In addition, it is reckless and impossible for policy makers to unconditionally adopt hypothetical policy formed by researchers and then implement it on a wide range before testing. Thus, to make the testing process tangible, I plan to work with local government for regions with the worst health condition that is eagerly seeking help and improvement, modify that hypothetical policy with the help of professional policy advisors, and then implement it on a small range of area for a three month period. After the implementation, the actual impact assessment will further contribute to the model validation and applicability confirmation.

Budget Plan:

Financial support is necessary for the implementation of this project over an one year exploratory phase, and considering a research funding of $100,000, I will enumerate the budget allocation on different aspects of the research plan. About half of the budget will devote to the data collection process. Roughly $25,000 will be spent on conducting geocoded surveys to get information regarding individuals’ health conditions and other demographic information like the household income. Also, obtaining other data including geographical information, transportation network data, basic statistics of Kenya’s health system, and call detail record from national databases and incumbent mobile phone provider will cost an expense at around $20,000. Another $15,000 will be spent on the purchase of necessary equipment like computers and other peripheral devices to enhance functionality for data analysis and the model development. In addition, to conduct field investigations on Kenya, at around $25,000 will be needed to cover the travel expenses and on-site accommodation and food costs of researchers. For the cooperative program with the local government to revise and implement the hypothetical policy on a relatively small region for a period of 3 months, in best cases, the local government will cover most of the spending for the policy implementation but I will still distribute $10,000 for some costs involved. The remaining $5,000 will be reserved for unexpected or supplemental spending in the research process.

Conclusion:

To conclude, my research plan focuses on the development of an aggregated geospatial mapping method for the variations of health accessibility in Kenya from its three major components. After finalization, this geospatial model and subsequent researches from it will provide informative insights for the improvement of health disparities and Kenyan’s equitable access to essential health services in order to address its major health issues.

Reference:

  1. Emily C Keats, et al. “Accelerating Kenya’s progress to 2030: understanding the determinants of under-five mortality from 1990 to 2015.” BMJ Glob Health 2018; 3:e000655.

  2. Ilinca, S., et al. Socio-economic inequality and inequity in use of health care services in Kenya: evidence from the fourth Kenya household health expenditure and utilization survey. Int J Equity Health 18, 196 (2019).

  3. Ongarora, Dennis, et al. “Medicine Prices, Availability, and Affordability in Private Health Facilities in Low-Income Settlements in Nairobi County, Kenya.” Pharmacy 2019, vol. 7, no. 2, p. 40.

  4. Paul O Ouma, et al. “Access to emergency hospital care provided by the public sector in sub-Saharan Africa in 2015: a geocoded inventory and spatial analysis.” Lancet Glob Health 2018; 6: e342–50

  5. Wesolowski, Amya, et al. “Quantifying the Impact of Accessibility on Preventive Healthcare in Sub-Saharan Africa Using Mobile Phone Data.” Epidemiology, Vol.26, 2015, pp. 223-228.