AI for Targeted CKD Screening: Mapping Vulnerable Populations for Early Intervention

Chronic Kidney Disease (CKD) affects approximately 10% of the global
population, with significant underdetection, particularly among vulnerable populations. These
groups face greater disparities in CKD incidence and progression due to poverty, limited
healthcare access, and environmental risks. Yet, these same vulnerabilities often prevent them
from being screened, compounding health inequities. By leveraging AI to create a data-driven
CKD Population Vulnerability Index and an interactive dashboard, our solution empowers
researchers and public health officials to pinpoint geographic areas where targeted screenings
would bring the greatest impact, enabling more equitable and cost-effective interventions.

1. Maria Aroca, PhD
January 22, 2024

2. Introduction
CKD is Underdetected
•
•
Chronic kidney disease (CKD) affects
about 10% of the global population, with
significant underdetection, particularly
in low-income countries (Rovin et al. 2024,
Bello et al. 2024).
Social Determinants of Health (SDoH) can
impact CKD diagnosis and outcomes
(Norton et al. 2016, Hall 2017).
The costs of non-action
•
•
Untreated CKD leads to advanced-stage
interventions, driving up healthcare
costs.
For instance, the UK estimated kidney
disease costs at £7 billion, rising to £13.9
billion by 2033 if left unaddressed, with
over 10,000 preventable deaths (Kidney
Research UK 2023).
Screening Challenges
•
•
The cost-effectiveness of CKD
screening in the general population
remains debated.
Targeted screening (Yeo et al.
2024) and social needs screening
(Painter et al. 2024) shown to
improve cost-effectiveness.
CKD: A Silent Epidemic with a disproportionate impact on
vulnerable populations
Global Incidence (A) and prevalence (B) of treated kidney failure. Source: Bello et al. 2024

3. The Problem When Vulnerability Blocks Access to Intervention

4. Competitors Landscape
•
•
•
•
•
•
Conventional Approaches:
Broad CKD screenings that are cost prohibitive. Targeted screenings with a focus on individual risk factors.
Ad-hoc, researcher-led screenings.
Healthcare facility-based detection, which misses underserved and remote populations.
Limitations of Existing Approaches:
Rarely integrate SDoH or environmental factors to guide screenings.
Screening efforts often fail to address underlying socioeconomic barriers.
Notably, research and interventions on health equity are expanding (Crews & Novis 2019, Musso et al. 2024,
Google’s Health Equity Research Initiative)
How can we identify and
prioritize populations for
CKD screening in a way that:
☑
☑
☑
☑
Maximizes cost-effectiveness
Focuses on the root causes—poverty, healthcare inaccessibility, and
environmental risks—that exacerbate CKD disparities
Provides a systematic, data-driven approach to allocate scarce resources
Equips stakeholders (e.g., government officials) with actionable insights
to evaluate and focus efforts on high-risk areas for intervention
Limitations of Conventional CKD Screening Approaches

5. Our Vision AI-Driven CKD Population Screening
Starting in the Colombian Caribbean
region, where disparities are pronounced,
we aim to validate our approach and
refine our tools. This region serves as a
testbed to demonstrate the model's
effectiveness, with the goal of expanding
to other underserved areas nationally
and globally, addressing CKD disparities
on a broader scale.
Big Change Begins Local
With the CKD Population Vulnerability
Geospatial Mapping Tool, we transform
the vulnerability index into actionable
insights for stakeholders. The tool
facilitates the identification of priority
areas for screening and can be used to
support advocacy efforts, ensuring the
right interventions reach the right
populations.
Precision Public Health
AI techniques enable us to analyze a
broad range of environmental and
socioeconomic factors influencing CKD
prevalence and synthesize these into a
single, interpretable population
vulnerability index. This index pinpoints
high-risk geographic areas, guiding cost-
effective strategies where they’re needed
most.
AI-Guided CKD Screening

6. Approx. 720,000 Colombians (1.54%) are
reported to have CKD (CAC 2022), but this is
likely a significant underrepresentation.
Recent screenings among 1,590 Afro-
descendants in the Colombian Caribbean
found 20.6% had proteinuria, with 78.4%
later confirmed to be in stages 1–3 of CKD.
Alarmingly, none of them were aware of their
condition (Aroca-Martínez et al. 2024).
At-Risk/Undiagnosed Populations
Researchers can more easily identify
communities for targeted studies, contributing to
existing research agendas, such as the study of
genetic components (e.g., APOL1 variants),
aligning with the goals of organizations like
ASOCOLNEF and SLANH.
Researchers and Associations
CKD represents a significant financial burden
in Colombia, costing between COP 8.7 and
COP 14.4 trillion annually—up to 2.7% of the
country's GDP. Late-stage CKD is particularly
expensive, with stage 5 treatment costing
over 230 times more than stage 1
(Sarmiento-Bejarano et al., 2019),
underscoring the importance of early
detection.
Public Health Officials
Historically marginalized communities—defined
by ethnicity, socioeconomic status, rurality, and
more—remain underrepresented in clinical trials.
This lack of diversity leads to treatments that
may not reflect real-world populations.
Detection among these groups could result in
more diverse recruitment.
Clinical Studies
Value Proposition and Target Market Target Market and how we add value

7. Model Structure Model Structure
•
•
•
•
Purpose: Combine insights from
population- and individual-level models into
a metric that identifies high-risk areas for
cost-effective CKD screening.
Features: Combines key socio-economic
and environmental predictors identified by
both models, including aggregated and
georeferenced factors, to build a
comprehensive risk profile.
Models: Employ advanced machine
learning techniques like Gradient Boosted
Machines for feature weighting and ranking,
geospatial models for capturing spatial
dependencies, and multi-objective
optimization to integrate predictors into a
cohesive vulnerability score. Validation
ensures alignment with real-world CKD
prevalence data.
Output: The CKD Population Vulnerability
Index scores areas from 0 to 1, highlighting
regions with high socio-economic and
environmental risk factors to guide targeted
screening.
CKD Population Vulnerability Index
•
•
•
•
•
•
•
•
•
Purpose: Assess how individual-level socio-
economic and environmental factors
influence CKD risk, disease progression, and
stage at diagnosis.
Targets: Individual CKD risk scores,
progression trajectories, late-stage
diagnosis.
Features:Demographic attributes, socio-
economic conditions, clinical data from
EHRs, and georeferenced environmental
exposures and access based on patient
locations
Models:
Logistic Regression: Baseline CKD risk
classification.
Random Forests: Identify key predictors
and non-linear relationships.
RNNs: Model disease progression with
longitudinal EHR data.
Survival Models: Predict time to late CKD
stages.
GBMs (e.g., XGBoost): Robust risk scoring
and feature insights.
Individual-Level Model
•
•
•
•
•
•
•
•
Purpose: Analyze aggregate data to identify
the most relevant socio-demographic and
environmental factors driving CKD
prevalence at the community level.
Targets: CKD prevalence rates and stage at
diagnosis at the geographic unit level.
Features: Air quality metrics (PM2.5, PM10),
income levels, healthcare access (clinic
proximity), sanitation conditions, and
education levels aggregated by geographic
unit.
Models:
GLMs: Poisson or Negative Binomial
regression for count-based prevalence.
Random Forests: Identify key predictors
and non-linear relationships.
GBMs: XGBoost or LightGBM for robust
predictions and feature interactions.
Clustering: K-means or DBSCAN to
uncover spatial CKD hotspots.
Population-Level Model

8. Data Strategy Data Colection and Data-Related Chalenges
Data Colection Data-Related Chalenges
Initial focus is on the Colombian Caribbean region to address its
significant heah disparities.
Population-Level Data: Aggregated socio-demographic and
environmental data from sources like DANE, IDEAM, and SIAC,
including air quality, sanitation, income, education, and heahcare
access. Data is mapped to granular geographic levels (e.g.,
manzanas, veredas) to capture CKD prevalence patterns across
regions.
Individual-Level Data: Longitudinal data from EHRs and
RENELUP registry capturing socio-economic conditions (e.g.,
income, employment, education) and clinical indicators (e.g., CKD
stages, comorbidities). Georeferenced data links patient locations
to environmental exposures such as water quality and polution.
Pilot Phase: A pilot study wil validate the index by guiding in-situ
screenings in high-risk areas. Clinical samples, heah, and
anthropometric measures (e.g., weight, height, blood pressure, and
urine analysis) wil be colected to assess the index's utility for
targeted interventions.
Sample Size & Coverage: Undiagnosed CKD cases and limited
data in underserved areas may skew resus, requiring robust
methods to address gaps and underrepresentation.
Bias & Equity: Datasets often refsect uneven access to heahcare,
risking biased predictions if not accounted for (e.g., overfitting to
populations with more recorded diagnoses).
Privacy & Compliance: Ensuring data protection by anonymizing
sensitive patient data in line with Colombian regulations and
HIPAA standards.
Label Colection: Chalenges arise in distinguishing between truly
low-risk areas and undiagnosed regions. Creating accurate proxy
negatives or prevalence-based labels is critical to ensure valid
model predictions.

9. Implementation Strategy Implementation Strategy
Phase 1: Dashboard
Development
Build an AI-driven
vulnerability index and
dashboard to visualize CKD
risk areas, guiding targeted
interventions.
Phase 2: Training and
Adoption
Phase 3: Continuous
Improvement
Phase 4: Pilot PhasePhase 5: Expansion
Train public health officials and researchers to interpret model outputs, navigate the dashboard, and translate insights into actionable strategies.
Establish a feedback loop to refine the system’s accuracy, usability, and outputs based on real-world application.
Validate the model by conducting targeted screenings, while gathering user feedback to refine the dashboard.
Test the model’s scalability and generalizability in regions beyond the Colombian Caribbean, validating its predictions for broader implementation.

10. Key Stakeholders
Key Stakeholders, barriers and concerns
•
•
•
Value: Early detection through targeted screenings reduces late-stage CKD diagnoses, improving
outcomes and access to care.
Concerns: Privacy and stigma associated with being flagged for screenings.
Mitigation: Transparent communication on the voluntary nature of participation, robust privacy
safeguards, and collaboration with community groups to ensure trust.
•
•
•
Value: Identifies communities for targeted studies, enabling advancements in CKD research.
Concerns: Limited data representation and potential inaccuracies could impact research validity.
Mitigation: Emphasize real-world data collection, rigorous bias monitoring, transparency, and
methodological reproducibility, while partnering with clinicians and associations for validation.
•
•
•
Value: Guides cost-effective resource allocation, addressing CKD's significant financial burden and
improving population health.
Concerns: Skepticism about model reliability and equitable distribution of resources.
Mitigation: Validate predictions through pilot studies, incorporate clinician feedback, and ensure
transparency in reporting outcomes and predictive factors. Partner with local health organizations
to address gaps in underrepresented populations.
Patients
At-Risk/Undiagnosed
Populations
Researchers and
Associations
Public Health
Officials

11. Regulatory Implications
Regulatory Implications
1.
2.
3.
4.
5.
6.
7.
Data Protection Compliance: Adhere to Colombia's Law 1581 of 2012 by ensuring anonymization, secure storage, and
exclusive use of personal data for public health purposes. Align with global standards for international data sharing.
Informed Consent: Provide accessible and clear mechanisms for consent, ensuring patients understand how their data will
be used, particularly in remote areas with lower health literacy. Build public trust through transparent communication.
Transparency and Accountability: Maintain thorough documentation of model inputs, algorithms, and decision-making
processes to meet international standards, mitigate bias risks, and enhance reproducibility.
Public Health Authorization: Demonstrate scientific credibility, safety, and cost-efficiency through pilot study results,
securing validation from Colombia’s Ministry of Health or relevant authorities.
Equitable Access: Focus on prioritizing high-risk areas based on health needs while avoiding political or socioeconomic
biases. Reinforce fairness and alignment with public health objectives.
Liability and Risk Management: Implement human oversight for critical decisions and establish clear accountability
frameworks among developers, implementers, and end-users to address potential errors or misclassifications.
Ethical Compliance: Ensure the system promotes equity and autonomy, avoids exacerbating health disparities, and serves
the public good in alignment with ethical guidelines.

12. Closing Remarks
Through the integration of advanced AI
techniques and a focus on social
determinants of health, our solution
transforms CKD detection and
prevention by addressing gaps that leave
vulnerable populations underserved.
Together, we can redefine how public
health systems tackle chronic diseases,
prioritizing early intervention and
improving outcomes for those who need
it most.

13. References
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Aroca Martínez, G., Pérez Jiménez, V., Perea Rojas, D. M., Barraza Ahumada, N., Muñoz Hernandez, A., Musso, C. G., Depine, S. A., Cadena Bonfanti, A., Sarmiento Gutierrez, J. J., Rodriguez
Murgas, J. E., & Benítez, A. (2024). Evaluation of Risk Factors for Chronic Kidney Disease in Afro-descendant Populations in Health Vulnerability in the Colombian Caribbean Region (2022-
2024) [Unpublished working draft].
Bello, A. K., Okpechi, I. G., Levin, A., Ye, F., Damster, S., Arruebo, S., ... & Zaidi, D. (2024). An update on the global disparities in kidney disease burden and care across world countries and regions.
The Lancet Global Health, 12(3), e382-e395
Burgos-Calderón, R., Depine, S. Á., & Aroca-Martínez, G. (2021). Population kidney health. A new paradigm for chronic kidney disease management. International journal of environmental
research and public health, 18(13), 6786.
Crews, D. C., & Novick, T. K. (2019, May). Social determinants of CKD hotspots. In Seminars in nephrology (Vol. 39, No. 3, pp. 256-262). WB Saunders.
Kidney Research UK. (2023). Kidney disease: A UK public health emergency: The health economics of kidney disease to 2033.
Musso, C., Ricardo, A., Aroca-Martinez, G., & Chaparro, M. (2024). Chronic Kidney Disease (CKD) classification for low-resource settings: Taking into account patients’ social vulnerability.
Norton, J. M., Moxey-Mims, M. M., Eggers, P. W., Narva, A. S., Star, R. A., Kimmel, P. L., & Rodgers, G. P. (2016). Social determinants of racial disparities in CKD. Journal of the American Society of
Nephrology, 27(9), 2576-2595.
Painter, H., Parry, E., McCann, L., Lunn, A. D., & Ford, J. (2024). Social needs screening in primary care: A tool in the fight for health equity?. Public Health in Practice, 7, 100466.
Rovin, B. H., Ayoub, I. M., Chan, T. M., Liu, Z. H., Mejía-Vilet, J. M., & Floege, J. (2024). KDIGO 2024 Clinical Practice Guideline for the management of LUPUS NEPHRITIS. Kidney International,
105(1), S1-S69.
Sarmiento-Bejarano, H., Ramírez-Ramírez, C., Carrasquilla-Sotomayor, M., Alvis-Zakzuk, N. J., & Alvis-Guzmán, N. (2019). Carga económica de la enfermedad renal crónica en Colombia, 2015-
2016. Revista Salud Uninorte, 35(1), 84-100.
Yeo, S. C., Wang, H., Ang, Y. G., Lim, C. K., & Ooi, X. Y. (2024). Cost-effectiveness of screening for chronic kidney disease in the general adult population: a systematic review. Clinical Kidney
Journal, 17(1), sfad137.
*Photographs: Captured during screenings of vulnerable populations conducted by Clínica de la Costa and partners in northern Colombia. Usage approved by the Ethics Committee for this
presentation.
**Dashboard Design Concept inspired by https://beattheheat.iu.edu/hvmt/index.html
***Icons: Sourced from Freepik, Max.Icon, and Flaticon.
****Illustrations: Custom-generated using DALL-E.

14. Thank you.
AI for Targeted CKD Screening:
Mapping Vulnerable Populations for Early Intervention