Health Expansion in Rural Outskirts (H.E.R.O)

Inspiration

From the himalayas to the deserts, the Ganges and the Yamana, the nearly 2 billion people in India represent one of the most culturally diverse populations in the world. Hundreds of different languages, cuisines, and attire make this country the ideal melting pot. Yet sustaining this rapidly growing population presents numerous challenges. Whether it is the climate change crisis in Delhi, or the lack of clean water, the Indian government is in a never ending search for solutions to the growing problems.

One of the largest problems within the nation is the lack of access to high quality healthcare, specifically in rural areas. The majority of the Indian population lives in rural areas, representing nearly 65% of the total population (India - Rural Population - 1960-2018 Data | 2020 Forecast, n.d.). Rural towns in India face the largest part of the healthcare crisis as more than 70% of Indians live in rural areas with few healthcare resources (Harvey, 2014). Although urban areas deal with overflow of patients, rural patients suffer with the mere lack of hospitals. In fact, the average distance between a rural resident and the nearest healthcare facility is between 5 - 20 miles, with some residents having to travel up to 62 miles to reach the nearest hospital (Kumar et al., 2014) (Player, 2019). Because of this, 1.6 million Indians died due to poor healthcare, and nearly one million died due to lack of any healthcare at all, with the majority of these deaths occurring in rural areas (Yadavar, 2018).

What it does

Healthcare Expansion in Rural Outskirts (H.E.R.O) aims to provide policymakers and health ministries in India insight into the most optimal locations for hospital placement in Indian cities. Our consulting firm currently focuses on three major cities where the lack of healthcare is a pertinent issue. These cities include Jhansi, Bihar, and Chhattisgarh. The lack of beds and hospitals for these high population rural areas in India place a massive strain on the healthcare systems. Jhansi has approximately 0.5 government hospital beds per 1000 people in its population. Bihar has one of the largest rural populations but has one of the lowest budgets for cities around the nation. In Bihar, one doctor serves an average 29,000 population and one hospital bed is available per 8,645 people (Ramashankar 2019). Chattisgarh faces a similar issue with only 1.6 hospitals per 10,000 people.

H.E.R.O uses data points from these areas and finds the best place to build the next hospital. H.E.R.O uses the population density for all points across the city, the location of the current hospitals, and the radius of area that the current hospital has the ability to treat. Using all of this information, our unsupervised machine learning model places new hospitals that cover larger populations and show their zones of coverage. This is crucial information for rural health ministries across the country as they can bring these data points and proposed locations up to policy makers to highlight the importance of increased construction of hospitals to support growing and struggling populations.

How we built it

In order to develop our solution for generating coordinates of optimal hospital/medical facility placement, we applied the K-means clustering unsupervised machine learning algorithm. This algorithm works on the concept of having a fixed number of centroids, which is the number of desired hospitals to construct in this case. This parameter can be changed based on the amount of desired hospitals to construct, which is dependent on city funding, and the model will output optimal longitude and latitude locations for those hospitals. The initial locations of the centroids were randomly selected, but as the model is training, they are iteratively shifted to spread out from each other in such a way that they are as close to clusters of training data as possible. In essence, the model classifies every point based on the centroid that is closest to it, and the model tries to decrease the difference between points and their corresponding centroids to adapt to patterns within the dataset. In this case, a set of points each containing samples of latitude, longitude, and the point’s corresponding population density was used to train the model. The higher the density for a region was, the more training points corresponding to that region were added. The training points were extracted from rectangular regions based on the bounds of the cities from Google Geocode.

Additionally, the model needed to account for already existing medical facilities, as building a new facility in close proximity to already existing facilities is not as effective in serving the general population. Rather, to focus on building centers near more unserved populations, points within the impact radius of each hospital, calculated using the hospital’s popularity, were not added to the training data to prevent the centroids from converging onto those regions.

Once the model was trained and the centroids converged onto optimal construction regions, the results were visualized in order to gain insights into the specific underserved populations the new hospitals will serve. First, a heatmap of population density was made to highlight more popular regions. The proposed coordinates of each hospital from the model’s output and the already existing hospitals we researched were overlaid onto the heatmap, along with the radius of service to highlight both the populations already served and the new areas that will be served from the new hospitals. For the radius of service of the new hospitals, which is dependent on the funding the city puts into their construction to make large-scale facilities, an approximation was made for the heatmap. The insights gained from this visualization show both the regions of coverage and the population density of those regions, which helps to evaluate the accuracy of the model since the model should converge to unserved populations with higher population densities.

Alongside the heatmap, a Leaflet, which is a scrollable geographic map object, was made containing markers with all the proposed and already existing locations, their radius of service, and the overall region highlighted. Each ticker is labeled with its corresponding hospital name, This scrollable functionality allows the user to zoom in and out to analyze different scales of regions and also look at geographic features around the area and connections to infrastructure. This zooming in and out feature for location analysis is the benefit of the Leaflet, while the benefit of the heatmap is the regions of service can be analyzed alongside its overlay on population distribution to get a better idea of impact.

In order to visualize the created data, we had to create a website that included all of the figures and leaflets we’ve made. We utilized HTML, CSS, and Javascript in order to create a minimalistic and aesthetic website that is able to be viewed comfrotablly, while also conveying all of the information and visuals we found and created. Although we ran through numerous iterations of our website, we were able to land on a final design that was pleasing to the eye. The use of Javascript served a role of providing a smoother user experience by giving access to transitions and animations that would have been extremely complicated to achieve through any other method, including creating them through HTML or CSS.

Challenges we ran into

Our team consisted of a diverse group of people with specialties in various areas. From aerospace engineering to speech and debate, each member held a unique set of skills that contributed equally to our team's success. Aneesh’s resilience learned from the countless failures of the rocketry led to our biggest accomplishment. Pranav’s machine learning knowledge was crucial for our models development. The team's drive and passion for the project led us to take on new challenges that we had minimal experience in. For example, one of these challenges was website development. We knew that a website would make our figures accessible and would aesthetically present our hard work. Yet we soon found none of us were nearly adept to develop a website of the caliber we wanted.

To overcome this challenge, our team put our heads down to scrounge the internet for resources that could help us in creating our dream website. From Youtube tutorials to textbooks on HTML, we searched for ways that we could build this website. After using some of the knowledge garnered from the search, our scrappy website was soon coming together. From interactive maps that connect to APIs, to buttons that lead to different pages, our collaboration led to the creation of a website perfectly crafted for our target user .

Our team collectively also spent a lot of time reading through documentation in order to attain a specific look for our visualizations and website. This process took a lot of time as we had to scrutinize the large files.

Accomplishments that we're proud of

Throughout the long journey of our team in this hackathon, we are most proud of our ability to remain open minded. We began our ideation with a divergent thinking method where we brain dumped ideas onto a whiteboard. From drug delivery for the disease track to a fitness and health tracker for the fitness category, our time shared countless ideas. Some of these ideas were impossible while others were too simple. After staring at the messy whiteboard with countless blurbs, and discussing which ones we found the most potential in, we narrowed down our ideas to the hospital location optimization. From here, we again began thinking divergently. We thought of the outcomes of our project, information about the user, possible constraints, parameters for the model, and other aspects of the consulting model. This led to another laundry list of ideas from where our narrowing process restarted.

This process continued throughout the hackathon as we tried multiple ideas. We would choose one of the ideas and pick out some parameters to test, yet we failed countless times throughout this. For example, we chose to find optimal locations for areas that have sufficient hospitals or we found optimal locations for too large of an area, resulting in a crashing program. Although we reached failure multiple times, we immediately went back to the drawing board to chef up the next best idea.

What we learned

The development of H.E.R.O proved to be a learning experience for everyone on the team, allowing us to tackle challenges that were both unfamiliar and complex. We embraced novel models and throughout the development process, engaged in thoughtful discussions and brainstorming sessions We explored innovative ways to increase the impact of our project. One of the most challenging issues we faced was how to obtain location data in a manner that was efficient and generalized. However, after deep thought we found the perfect solution: The Google Maps Application Programming Interfaces (API). By leveraging the power of Google’s API, we were able to streamline the process of obtaining location data, saving both time and resources. From this, our biggest takeaway was learning how the API could be used to repurpose our project to impact more than just one city in rural India.

We also faced adversity when working to create the visualization of our project. With the goal of creating an intuitive user interface on the web for our target user, we understood the importance of visualizations. However, with the lack of experience working with integrating machine learning into a website using HTML, CSS, and JavaScript, it was a daunting task given the time frame that took a long time. After working for several hours trying to find a viable solution, we stumbled upon Leaflet, an open-source Javascript library that provided friendly interactive maps that could be easily mutable and imported into our website.

However, we learned more than just how to use an API and visualizations. The development of our project also greatly challenged our minds when it came to nontechnical thinking. H.E.R.O aims to bring a revolution to healthcare in rural communities, so maximizing our impact was crucial for us. Nevertheless, the time constraints of this hackathon allowed us to transition this passion into MVP thinking. This resulted in us discovering and understanding user based thinking and creating a product that represented our best interests in the shortest time possible.

What's next for Health Expansion in Rural Outskirts (H.E.R.O)

During the brainstorming process for H.E.R.O, our team considered many ideas that were deemed too difficult to code in the limited time provided. With more data and more time, we believe that these ideas will be plausible and generate a massive impact on optimizing the location provided for the hospitals, and the scalability for our model. One of these ideas include adding a terrain parameter for the land where the hospital will be placed. Having a binary parameter for if it would be possible to develop a hospital in this area will be useful for rough terrains. Another addition to the project would be including more cities. One of the ideas we attempted was creating an API into the website where the user can input any city in India, and our code would provide the heatmap and leaflet for that city and where the hospitals should be placed. Our team tried multiple additions to the project which all seemed implausible given the time and data, but with additions to both of these, we believe that H.E.R.O has the potential to scale the entire world, and provide the most accurate hospital optimizations. Other plans for project improvement involved considering the type of healthcare facility (general hospital, children’s hospital, etc) to find the best suited one for the selected region.