AI-POWERED ANALYSIS
Validation of a Screening Score Model to Predict the Development of Retinopathy of Prematurity
By Johanes E. Siswanto, Asri C. Adisasmita, Sudarto Ronoatmodjo, Boromeus A. Daniswara, Peter H. Dijk, Arend F. Bos, Pieter J.J. Sauer
Executive Impact Summary
This research pioneers a critical solution for preventable childhood blindness in low- and middle-income countries (LMICs) by developing and validating two pragmatic, risk-based AI screening models for Retinopathy of Prematurity (ROP). Tailored to Indonesian neonatal data, these models (FiO2-based and SpO2-based) provide a robust, locally validated approach that significantly enhances early ROP detection compared to traditional methods. For healthcare enterprises, implementing such localized, data-driven screening tools promises substantial improvements in patient outcomes, optimized resource allocation in underserved regions, and a significant reduction in long-term care costs associated with ROP-induced blindness.
0%
Combined Sensitivity
0%
Combined Specificity
0%
Positive Predictive Value
0%
Negative Predictive Value
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
The Challenge of ROP in LMICs
Retinopathy of Prematurity (ROP) is a leading cause of preventable blindness worldwide, particularly affecting preterm infants in low- and middle-income countries (LMICs). The increasing survival of very preterm infants in these regions, coupled with limited capacity for comprehensive screening, has led to a "third epidemic" of ROP. Current screening protocols, largely derived from high-income settings, are often not directly applicable to LMIC care pathways, necessitating tailored approaches.
0.00
Post-Test Probability of ROP with Positive Screen
This highlights the significant improvement in identifying high-risk infants for ROP with the validated screening models.
Developing Pragmatic Screening Models
This study developed two pragmatic risk-based screening models (Model A: FiO2-based and Model B: SpO2-based) using multicenter Indonesian neonatal data. These models incorporated significant predictors such as intrauterine growth restriction, oxygen exposure, exchange transfusion, and socioeconomic status. The methodology involved a retrospective cohort study design, bivariable assessment, multivariable logistic regression, and backward elimination for predictor selection.
Enterprise Process Flow
Identify Candidate Predictors
→
Develop FiO2-based Model (A)
→
Develop SpO2-based Model (B)
→
Internal Validation
→
External Validation
→
Bedside Clinical Application
Validation and Robust Performance
Both models demonstrated moderate discrimination during internal validation (AUC 0.719-0.732, sensitivity 77-86%, specificity 44-58%). External validation in an independent cohort confirmed robust performance, with the combined rule (positive if either Model A or Model B was positive) achieving particularly strong results, offering a practical enhancement to existing screening criteria in resource-limited settings.
| Metric | Model A (FiO2-based) | Model B (SpO2-based) | Combined Rule |
|---|---|---|---|
| Sensitivity | 76% | 38% | 84% |
| Specificity | 41% | 76% | 81% |
| Positive Predictive Value (PPV) | 48% | 53% | 76% |
| Negative Predictive Value (NPV) | 71% | 63% | 87% |
Practical Clinical Application
The study presents a bedside operational score form for clinical use, alongside a Fagan nomogram to illustrate the shift in ROP probability. A pre-test probability of ROP at 0.42 increased to 0.76 after a positive screen and decreased to 0.13 after a negative result. This provides clinicians with clear, actionable insights to optimize ROP case finding, especially where specialist resources are limited.
Case Study: Optimized ROP Screening in a Resource-Limited NICU
A neonatal intensive care unit in rural Indonesia faced challenges in timely ROP screening due to limited ophthalmologist availability. Implementing the new combined FiO2- and SpO2-based screening score significantly improved their detection rate.
For infants with a pre-test ROP probability of 42%, a positive screening score now raised the probability to 76%, allowing for immediate prioritization and referral. Conversely, a negative score reduced the probability to 13%, reducing unnecessary specialist visits and optimizing resource allocation. This data-driven approach enabled more efficient case identification and earlier intervention, dramatically improving patient outcomes.
Strategic Implications for Global Health
These locally validated risk-based scores serve as a pragmatic complement to existing gestational age and birth weight criteria, offering a more nuanced and effective screening approach for LMICs. The findings underscore the potential for context-specific AI tools to address global health inequities. Future directions involve integrating these clinical scores with emerging technologies like smartphone-based video capture and AI interpretation, alongside telemedicine, to further expand screening capacity and reduce avoidable childhood blindness at scale.
For healthcare enterprises operating in LMICs, adopting and localizing such AI-powered diagnostic tools presents a clear pathway to fulfilling social responsibility mandates while simultaneously achieving greater operational efficiency and improved patient care standards at scale.
Calculate Your Potential ROI
Estimate the potential savings and reclaimed hours by implementing AI-powered solutions in your enterprise, tailored to your industry and operational specifics.
Annual Cost Savings
$0
Annual Hours Reclaimed
0
Your AI Implementation Roadmap
A typical journey to deploying enterprise AI solutions. Each phase is tailored to your unique organizational needs and the specific AI model being integrated.
Data Integration & Model Setup
Securely integrate existing patient data, ensuring compliance and data privacy. Configure the ROP screening models (FiO2-based and SpO2-based) for your specific clinical environment and infrastructure, establishing data pipelines for real-time input.
Pilot Testing & Local Validation
Conduct a controlled pilot study within a selected neonatal unit. Validate the model's performance against local clinical outcomes, gathering feedback from clinicians and refining the algorithm for optimal accuracy and usability.
Staff Training & System Rollout
Provide comprehensive training for medical staff on using the new AI screening tool and interpreting its scores. Implement the system across all relevant neonatal units, ensuring seamless integration into daily clinical workflows and EMRs.
Continuous Monitoring & Refinement
Establish ongoing monitoring of the model's performance and clinical impact. Implement a feedback loop for continuous improvement, adapting to new data or evolving clinical guidelines to maintain high predictive accuracy and relevance.
Ready to Transform Your Healthcare Operations?
Connect with our AI specialists to discuss how these validated screening models, or other bespoke AI solutions, can be tailored and implemented within your enterprise to achieve superior patient outcomes and operational efficiency.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat