Duke-NUS Updated Multivariable Prediction Model: Identifying High-Risk COVID-19 Patients

Breakthrough in Pediatric COVID-19 Risk Prediction from Duke-NUS Medical School

In the evolving landscape of infectious disease management, Duke-NUS Medical School stands at the forefront of innovative research, particularly with its contributions to multivariable prediction models for high-risk COVID-19 patients. Although COVID-19 cases have significantly declined since the peak of the pandemic, the lessons learned remain crucial for future outbreaks. A pivotal study led by researchers affiliated with Duke-NUS, published in PLOS ONE, developed a clinical predictive model specifically for severe and critical pediatric COVID-19 infections using routinely available hospital data. This model, derived from a multinational Asian cohort, offers a practical tool for clinicians to stratify risk early in admission. 102 20

The research underscores Duke-NUS's role as a graduate medical institution bridging Duke University and Singapore's National University of Singapore (NUS), fostering clinician-scientists equipped to tackle global health challenges. During the pandemic, Singapore reported relatively low pediatric severe cases—around 14.4% in the study's cohort—but identifying the vulnerable few was essential for resource allocation in hospitals like KK Women's and Children's Hospital (KKH).

This work not only highlights the school's research prowess but also its impact on Singapore's higher education ecosystem, where medical research drives career opportunities in academia and healthcare.

The Urgent Need for Accurate Risk Stratification During the Pandemic

Singapore's response to COVID-19 was exemplary, with low overall mortality, but pediatric cases posed unique challenges. While most children experienced mild symptoms, a subset developed severe disease requiring intensive care. Data from the Pediatric Acute and Critical Care COVID-19 Registry of Asia (PACCOVRA) revealed that 14.4% of admitted children progressed to severe or critical illness, with 2.9% mortality, often linked to comorbidities. 102

Predicting who would deteriorate was vital amid surging Delta variant cases in 2021. Traditional scoring systems fell short for pediatrics, prompting Duke-NUS-linked teams to leverage real-world hospital data from Singapore, Malaysia, Indonesia, and Pakistan. This multinational approach ensured generalizability across diverse Asian populations, reflecting Singapore's multicultural context.

• Over 1,147 patients analyzed from November 2019 to 2021.
• Median age: 6 years; 85.6% mild/moderate cases.
• Key complications: pneumonia (6.8%), ARDS (2.1%).

Such models empower frontline clinicians, reducing unnecessary admissions while prioritizing high-risk cases—a balance critical in resource-constrained settings.

Duke-NUS Medical School: Pioneering Singapore's Medical Research Landscape

Established in 2005 as a collaboration between Duke University and NUS, Duke-NUS Medical School has become a beacon of research excellence in Singapore. Its MD-PhD program trains clinician-scientists, producing leaders in infectious diseases. During COVID-19, Duke-NUS achievements included early virus culturing—the third globally outside China—and national awards for pandemic contributions. 120 116

Affiliated with SingHealth, Duke-NUS researchers like Rehena Sultana from the Centre for Quantitative Medicine and Chee Fu Yung played key roles in the prediction model. The school's Emerging Infectious Diseases programme continues to lead, preparing for future threats like Nipah virus. 52

For aspiring researchers, Duke-NUS offers world-class facilities and collaborations. Explore research jobs or faculty positions to join this dynamic environment.

Methodology: Building a Robust Multivariable Model

The model was built using data from five tertiary pediatric hospitals, split into training (70%, n=802) and validation (30%, n=345) cohorts. Multivariable logistic regression via generalized linear mixed models accounted for site variability. Predictors were selected based on clinical relevance and univariate significance (p<0.2), refined through backward/forward/stepwise methods.

Step-by-step process:

• Data collection: Anonymized REDCap entries from RT-PCR confirmed cases or MIS-C.
• Outcome definition: WHO severe/critical vs. mild/moderate.
• Model fitting: Adjusted β coefficients for predicted probability logit = 4.29 + 1.87(comorbidities) + 2.18(infant) + 3.34(seizures) + 1.80(vomiting) + 1.19(fever) - 1.77(coryza) + 0.07(ANC) - 0.57(Hb) - 0.01(platelets).
• Validation: ROC curves, calibration plots, sensitivity analyses excluding MIS-C.

This pragmatic approach uses first-available labs, making it deployable in busy wards. 102

Photo by Andy Wang on Unsplash

Key Risk Factors Unveiled by the Model

The model pinpointed actionable predictors:

• Increased risk: Infant age (<1 year), comorbidities (e.g., cardiovascular), fever, vomiting, seizures, elevated absolute neutrophil count (neutrophilia signaling inflammation).
• Decreased risk: Coryza (runny nose, mild URTI), higher hemoglobin, higher platelets (better oxygenation, viral control).

Seizures had the highest odds (β=3.34), reflecting neurological involvement. 102 These align with bimodal severity patterns in infants/adults and Delta-era data.

Impressive Performance and Cross-Regional Validation

The model's discriminative power was exceptional: AUC 0.96 (95% CI 0.94-0.98) in training, 0.92 (0.86-0.97) in validation. Specificity reached 94.1% (92.1-95.7%), ideal for ruling out low-risk cases; sensitivity 53.2% (43.4-62.8%). Site-specific AUCs ranged 0.79-0.99, proving robustness. 102

Calibration was strong, with sensitivity analyses confirming stability sans MIS-C. This positions it as a reliable tool for Asian settings.

Clinical Applications and Impact in Singapore's Healthcare

In Singapore, where pediatric COVID mortality was low (2.9% in cohort), the model aids triage at KKH and beyond. High specificity minimizes overtreatment; early identification flags infants/comorbid kids for closer monitoring, oxygen, or ICU.

Post-pandemic, it informs protocols for respiratory viruses. Duke-NUS's integration with SingHealth ensures translation: models like this enhance emergency decision-making. 91

For professionals, such tools highlight demand for data-savvy clinician-scientists. Visit career advice for tips.

From Multivariable Models to AI-Driven Updates: TRACER Framework

Building on classics like the pediatric model, Duke-NUS advanced to TRACER (Transfer Learning based Real-time Adaptation for Clinical Evolving Risk), a 2025 framework for updating models amid shifts like COVID variants. 124 TRACER uses EM algorithms and Trans-Lasso to detect transitions and adapt without full retraining, boosting AUC to 0.85+ in ED admission prediction during COVID.

This 'update' mechanism exemplifies Duke-NUS's evolution, applying to COVID risk amid new strains or protocols.

Photo by Galen Crout on Unsplash

Future Outlook: Lessons for Pandemic Preparedness and Beyond

Duke-NUS continues innovating: AI for cardiac outcomes (80% accuracy post-adaptation), PathGen for pathogen surveillance. 80 These tools prepare Singapore for mpox, avian flu. Globally, they promote equitable health via Asian data.

In higher ed, Duke-NUS attracts top talent; its PhD programs yield high-impact publications. Singapore's RIE2030 invests SGD37b in R&D, boosting unis like NUS/Duke-NUS.

Career Pathways in Singapore's Medical Research Sector

Duke-NUS exemplifies opportunities: from postdocs to faculty. With COVID research accolades, grads secure roles in A*STAR, SingHealth. Primary keyword integration: multivariable prediction models demand quantitative skills.

• PhD in quantitative medicine at Duke-NUS.
• Faculty positions in infectious diseases.

Check university jobs, postdoc openings, or rate professors. For advice, see postdoc success guide.