Predictive Model for Antimicrobial Resistance Using Tabular Data

Arkangel AI's solution utilizes machine learning to create a predictive model for antimicrobial resistance using tabular data available in Latin American hospitals, aiming to improve early detection and address the challenge of limited genomic data in middle- and low-income countries.

Jose David Gómez Zea, CEO and co-founder at Arkangel ai

The problem we are addressing is antimicrobial resistance (AMR) in Latin America, particularly in middle- and low-income countries where genomic data is scarce. In these regions, medical information is often limited to tabular patient data, impeding the effective determination of medication resistance. AMR is a significant challenge in Latin America, with 569,000 deaths potentially related to AMR, and 141,000 directly attributable to it. In Colombia alone, there were 4,700 deaths attributable to antibiotic resistance in 2019.

The causes of AMR include the overuse and misuse of antibiotics, lack of access to quality healthcare, and inadequate diagnostic tools. Limited access to quality healthcare exacerbates the problem by hindering effective treatment and monitoring of infections. Additionally, the lack of advanced diagnostic tools in these regions makes it challenging to identify and track resistant strains, further contributing to the spread of AMR. Addressing these causes is crucial for controlling the AMR crisis and reducing its impact on public health.

Our solution targets healthcare providers and patients in middle- and low-income Latin American countries, particularly focusing on vulnerable populations like adults over 65, who are most affected by antimicrobial resistance (AMR). We aim to address the critical need for early detection of AMR, essential for effective treatment and prevention of resistant bacteria spread.

To understand our audience's needs, we engage in ongoing dialogue with healthcare professionals and conduct research on their challenges in diagnosing and treating AMR. This includes analyzing available hospital data and identifying gaps in current diagnostic tools. We also consider the economic and infrastructural constraints of healthcare settings in these regions to ensure our solution is accessible and feasible.

Our engagement with the target audience continues throughout development. We collaborate with hospitals and providers to pilot our predictive model, gather feedback, and make adjustments, ensuring our solution fits the specific needs and context of these healthcare settings.

By emphasizing early AMR detection, our solution aims to assist healthcare providers in making informed decisions about antibiotic use and treatment strategies. This approach seeks to improve patient outcomes and reduce the burden of AMR in vulnerable populations.

Our primary goal is to generate valuable knowledge through the development of our predictive model for antimicrobial resistance. Once the model is implemented and validated in the field, we plan to publish our findings in a reputable peer-reviewed scientific journal. This will ensure that the insights gained from our work are shared with the global scientific community and can be used to inform further research and policy-making in the field of antimicrobial resistance. 

While we cannot guarantee free access to our model due to operational costs, we are dedicated to making it as accessible as possible, especially for healthcare providers in Colombia and other low- and middle-income countries. Our pricing strategy will be designed to reflect the needs of the sector, ensuring that our solution can be utilized by those who need it most.

Our solution addresses antimicrobial resistance (AMR) in Latin America, focusing on adults over 65, a group with high AMR-related mortality rates. In 2019, Colombia alone saw 4,700 deaths due to antibiotic resistance. Our primary goal is early AMR detection to reduce mortality rates, starting in Colombia and expanding across Latin America. Our predictive model aims to classify at-risk patients early, allowing healthcare professionals to adapt treatments and improve personalized medicine. Our process involves data collection, cleaning, transformation, modeling, deployment, and ongoing improvement. Early detection is key to enhancing treatment outcomes and lowering AMR mortality rates. 

Previous models have demonstrated the feasibility of using tabular data for predictive modeling in microbial resistance. A study by Feretzakis et al.showed a model with ICU data achieving AUC scores between 82 and 85% using clinical data. By focusing on early AMR detection in vulnerable populations, especially older adults, our solution aims to improve access to effective treatments and reduce AMR-related deaths in the region.

Our solution aims to scale its impact through a strategic approach, starting with a pilot deployment in Colombia and expanding to other LATAM countries like Mexico and Brazil. Our objective is to deploy our model in major hospitals with access to antimicrobial resistance data. Upon validation, we plan to extend implementation to other major hospitals in the region, focusing on the most affected adult population over 65 years old.

Key challenges include data privacy, compliance, and costs. We will offer data anonymization consultancy and ensure model compliance with ISO 27001/2 standards for secure medical data handling. Financially, we aim to make our models affordable for each institution and seek partnerships to enhance project viability.

Our primary partnerships are with major LATAM hospitals. Once the pilot is validated, we aim to scale our models and infrastructure for broader implementation, aiming to transform the fight against antimicrobial resistance.

Our plan to monitor and evaluate the impact of our solution begins with establishing a baseline to determine the current status of the disease. We then set specific objectives, such as the number of patients we aim to process through the algorithm and the number of early detections we seek to achieve. Once the data is processed, we collect and update it in real time to validate the impact of our solution, in this case, in terms of patients with antimicrobial resistance detected early.

To measure our progress and impact, we use specific, measurable indicators derived directly from the number of patients processed by the algorithm. This allows us to quantify the effectiveness of our solution in detecting antimicrobial resistance early and to make data-driven decisions to improve our approach.

By continuously monitoring these indicators, we can assess the performance of our solution and make necessary adjustments to ensure that we are achieving our objectives and making a meaningful impact in the fight against antimicrobial resistance.

In the next year and the next three years, we anticipate several barriers to accomplishing our goals, including financial constraints, data privacy and compliance challenges, technical hurdles, education and awareness gaps, and infrastructure limitations. 

To overcome these barriers, we plan to pursue strategic partnerships with major hospitals and pharmaceutical companies to secure funding and resources. We will offer data anonymization consultancy to partner hospitals and ensure compliance with ISO 27001/2 standards to address data privacy and compliance issues. To tackle technical challenges, our AI models are capable of working with limited data availability in target regions. 

To address the slow adoption of AI in healthcare in low-income regions, we intend to develop educational programs and materials to inform healthcare providers and stakeholders about the benefits of AI in healthcare. We are currently developing our AI Heroes community, in which healthcare professionals can access to AI in healthcare content: https://www.aiheroes.org/ We also conduct workshops to build trust and understanding. To navigate infrastructure limitations, we will work closely with local healthcare providers to understand specific challenges and tailor our solution to fit within these constraints, exploring mobile and cloud-based solutions to increase accessibility. 

We are applying to The Trinity Challenge because it provides a unique platform to showcase our solution's potential impact on reducing the impact of antimicrobial resistance (AMR) and bacterial infections in low- and middle-income communities. The challenge aligns closely with our mission to leverage AI and machine learning for early detection and management of AMR, a critical issue in healthcare.

One of the barriers we face is the limited availability of genomic data in Latin American hospitals, which hinders the development of predictive models for AMR. The Trinity Challenge can help us overcome this barrier by providing access to a network of experts and resources focused on addressing global health challenges. Additionally, the challenge's focus on data and analytics aligns well with our expertise and allows us to contribute meaningfully to the broader conversation on combating AMR.

By participating in The Trinity Challenge, we aim to gain valuable insights, access to new data sources, and potential collaborations that can accelerate the deployment and impact of our solution in Latin America and beyond.

We would like to collaborate with organizations that can provide expertise in antimicrobial resistance (AMR) and health, particularly with leading experts from MIT. Additionally, connecting with key decision-makers from major hospitals and healthcare centers in Latin America (LATAM) that we have not yet contacted would be beneficial. Facilitating this contact through The Trinity Challenge to map more hospitals would greatly assist in expanding our solution's reach and impact.