📖 About NCD-CIE & Evidence

Methodology, validation, and scientific foundation

What is NCD-CIE?

NCD-CIE (Non-Communicable Disease Causal Inference Engine) is a causal knowledge graph-based clinical decision support system for predicting and preventing non-communicable diseases. It combines evidence from epidemiological studies, randomized controlled trials, and observational data into a unified causal model with 107 weighted edges connecting 51 clinical variables across 8 domains.

Unlike traditional risk scores (Framingham, QRISK3, SCORE2), NCD-CIE models causal pathways between risk factors and disease endpoints, enabling what-if intervention simulation using Pearl's do-calculus framework.

Pearl's Ladder of Causation

👁️

Level 1: Association

P(disease | risk factor)

Patients with high LDL have more heart attacks

Traditional risk scores

🔬

Level 2: Intervention

P(disease | do(intervention))

If we prescribe statins, will heart attack risk decrease?

NCD-CIE What-If Simulator

🤔

Level 3: Counterfactual

P(disease_y₀ | do(x₁), observe(y₁))

Would the patient have had a heart attack if they had taken statins?

Future research direction

Methodology

Risk Scoring (Logistic-Link)

R_d = σ(β₀_d + Σ w_i · z_i)

Where σ is the logistic sigmoid, z_i are z-score standardized biomarkers, and w_i are causal edge weights (log-odds ratios).

What-If Cascade (do-calculus approximation)

δ_k = Σ W_{(v_p,v_k)} · (x_p^INT − x_p) · γ^depth

Topological propagation with γ=0.7 attenuation factor, maximum depth d_max=3. This ensures effects diminish along longer causal chains while maintaining biological plausibility.

Composite NCD Risk

R_NCD = 1 − (1−R_CVD)(1−R_T2DM)(1−R_CKD)

Evidence Grade System

Grade AStrong

Multiple RCTs, meta-analyses, or Mendelian randomization studies

58 edges

Grade BModerate

At least one RCT, large prospective cohorts

42 edges

Grade CLimited

Observational studies, cross-sectional, biological plausibility

5 edges

Grade DEmerging

Expert consensus, preliminary findings

2 edges

Validation Results

AUC-ROC (SCORE2)

0.704

AUC-ROC (D'Agostino)

0.721

Calibration Slope

0.91

Brier Score

0.118

NHANES Concordance

r = 0.91

RCT Alignments

6/6

Nodes

Causal Edges

107

Comparison with Existing Risk Scores

Feature	NCD-CIE	Framingham	QRISK3	SCORE2
Disease Coverage	CVD+T2DM+CKD+NAFLD	CVD only	CVD only	CVD only
Causal Modeling	✅ Full DAG	❌ Regression	❌ Regression	❌ Regression
What-If Simulation	✅ do-calculus	❌	❌	❌
Medication Effects	✅ 6 drug classes	❌	Partial	❌
Variables	51	8	21	6
Evidence Grading	✅ A-D grades	❌	❌	❌
Uncertainty	✅ 95% CI	❌	Partial	❌

Landmark RCT Validation

4S Trial

Statin → LDL↓ → CVD↓

Aligned ✅

UKPDS

HbA1c control → T2DM complications↓

Aligned ✅

SPRINT

Intensive BP → CVD↓

Aligned ✅

EMPA-REG

SGLT2i → HF↓, CKD↓

Aligned ✅

DPP

Lifestyle → T2DM↓

Aligned ✅

HOPE

ACEi → CVD↓, CKD↓

Aligned ✅

⚠️ Clinical Disclaimer

NCD-Care+ is a clinical decision support tool and does not replace clinical judgment. Risk predictions are based on population-level data and may not accurately reflect individual risk. All clinical decisions should be made in consultation with a qualified healthcare professional. The NCD-CIE model is validated for research purposes; clinical deployment requires institutional review.