GDM Diagnostic Decision Tree

Universal vs risk-based screening for gestational diabetes in India

R for HTA (Basics) — Workshop 2026

RRC-HTA, AIIMS Bhopal | HTAIn, DHR

The Clinical Question

Gestational diabetes mellitus (GDM) affects 10–14% of pregnancies in India.

Much higher than Western countries.

Undetected GDM increases risks of: - Macrosomia and birth injuries - Neonatal hypoglycaemia - Pre-eclampsia - Future type 2 diabetes in mother

The HTA Question

Should India implement universal screening for GDM using a fasting 75g OGTT, or is risk-based screening more cost-effective?

This is a classic diagnostic decision tree problem.

The kind of analysis you might currently do in Excel or TreeAge — but now in R.

Three Strategies Compared

  1. Universal screening
    • All pregnant women get OGTT at 24–28 weeks
  1. Risk-based screening
    • Only high-risk women (age >25, BMI >25, family history) get OGTT
  1. No formal screening
    • GDM detected only when symptomatic (status quo in many settings)

Key Parameters

Parameter Value
GDM prevalence 12%
High-risk women 60%
OGTT sensitivity 90%
OGTT specificity 85%
Screening cost ₹250
Treatment cost ₹8,000
Adverse outcome (if treated) 10%
Adverse outcome (if missed) 35%
Figure 1

All sourced from Indian studies where available.

The Decision Tree Structure

Code 
library(DiagrammeR)

grViz("
digraph gdm_tree {
  graph [rankdir=LR, bgcolor='transparent', fontname='Helvetica', nodesep=0.5]
  node [fontname='Helvetica', fontsize=10]

  D [label='Screening\nStrategy', shape=square, style=filled, fillcolor='#4e79a7', fontcolor='white', width=1.2]

  S1 [label='Universal\nScreening', shape=box, style='filled,rounded', fillcolor='#59a14f', fontcolor='white']
  S2 [label='Risk-Based\nScreening', shape=box, style='filled,rounded', fillcolor='#f28e2b', fontcolor='white']
  S3 [label='No Formal\nScreening', shape=box, style='filled,rounded', fillcolor='#e15759', fontcolor='white']

  C1 [label='GDM+\n(12%)', shape=circle, style=filled, fillcolor='#d4e6f1', width=0.8, fontsize=9]
  C2 [label='GDM-\n(88%)', shape=circle, style=filled, fillcolor='#d5f5e3', width=0.8, fontsize=9]

  TP [label='Test +\nTreat', shape=box, style='filled,rounded', fillcolor='#aed6f1', fontsize=9]
  FN [label='Test -\nMissed', shape=box, style='filled,rounded', fillcolor='#f5b7b1', fontsize=9]

  D -> S1
  D -> S2
  D -> S3

  S1 -> C1
  S1 -> C2
  C1 -> TP
  C1 -> FN
}
")
Figure 2: GDM screening decision tree

Strategy 1: Universal Screening

Every woman gets an OGTT. Those who test positive receive treatment.

Test results in 10,000 women:

  • True positives (GDM detected): 12% × 90% = 10.8% of cohort
  • False negatives (GDM missed): 12% × 10% = 1.2% of cohort
  • False positives (wrongly flagged): 88% × 15% = 13.2% of cohort
  • True negatives (correctly cleared): 88% × 85% = 74.8% of cohort

Cost Calculation: Universal

Code 
# Screening cost: everyone gets OGTT
cost_screening_universal <- n_cohort * cost_ogtt

# Treatment costs: true positives + false positives
cost_treatment_universal <- (true_positive + false_positive) * cost_gdm_treatment

# Adverse outcomes
adverse_outcomes_universal <- (true_positive * p_adverse_gdm_treated) +
                              (false_negative * p_adverse_gdm_untreated) +
                              (false_positive * p_adverse_no_gdm) +
                              (true_negative * p_adverse_no_gdm)

cost_outcomes_universal <- (adverse_outcomes_universal * cost_adverse) +
                           ((n_cohort - adverse_outcomes_universal) * cost_no_adverse)

total_cost_universal <- cost_screening_universal + cost_treatment_universal +
                        cost_outcomes_universal

QALY Calculation: Universal

Code 
# QALYs by outcome group
qaly_tp <- true_positive * ((1 - p_adverse_gdm_treated) * utility_gdm_treated +
                             p_adverse_gdm_treated * utility_adverse)

qaly_fn <- false_negative * ((1 - p_adverse_gdm_untreated) * utility_no_adverse +
                              p_adverse_gdm_untreated * utility_adverse)

qaly_fp <- false_positive * ((1 - p_adverse_no_gdm) * utility_gdm_treated +
                              p_adverse_no_gdm * utility_adverse)

qaly_tn <- true_negative * ((1 - p_adverse_no_gdm) * utility_no_adverse +
                             p_adverse_no_gdm * utility_adverse)

total_qaly_universal <- qaly_tp + qaly_fn + qaly_fp + qaly_tn

Strategy 2: Risk-Based Screening

Only high-risk women are offered OGTT. Low-risk women are NOT screened → GDM undetected in ~60% of cases.

Code 
# High-risk (60% of cohort)
n_high_risk <- n_cohort * prop_high_risk
# Among high-risk: 18% have GDM (higher prevalence)
# Among low-risk: 3% have GDM (lower prevalence)

# Screened (high-risk only)
tp_hr <- n_high_risk * prev_gdm_high_risk * sens_ogtt
fn_hr <- n_high_risk * prev_gdm_high_risk * (1 - sens_ogtt)
fp_hr <- n_high_risk * (1 - prev_gdm_high_risk) * (1 - spec_ogtt)

# NOT screened (low-risk) → all GDM missed
n_gdm_lr <- n_cohort * prop_low_risk * prev_gdm_low_risk  # All undetected

Strategy 3: No Formal Screening

GDM detected only through clinical symptoms (~25% detection rate).

No screening cost, but many cases missed.

Result: Lowest upfront cost, but highest adverse outcome rate due to missed cases.

Results Comparison

All three strategies produce: - Costs (screening, treatment, outcome management) - QALYs (quality of life impact) - Adverse outcomes (births with complications)

Then we calculate ICERs to compare cost-effectiveness.

ICER Calculations

Using No Screening as the reference:

Code 
# Risk-Based vs No Screening
icer_risk_vs_none <- (total_cost_risk - total_cost_none) /
                     (total_qaly_risk - total_qaly_none)

# Universal vs No Screening
icer_univ_vs_none <- (total_cost_universal - total_cost_none) /
                     (total_qaly_universal - total_qaly_none)

# Universal vs Risk-Based
icer_univ_vs_risk <- (total_cost_universal - total_cost_risk) /
                     (total_qaly_universal - total_qaly_risk)

Interpreting the ICER — 4 Quadrants

A negative ICER can mean two opposite things:

ΔQALYs > 0 (better) ΔQALYs < 0 (worse)
ΔCost > 0 NE: compare to WTP DOMINATED
ΔCost < 0 DOMINANT Trade-off

Always check the signs of ΔCost and ΔQALYs separately!

Net Monetary Benefit (NMB)

NMB removes the ICER ambiguity. Positive ΔNMB = adopt.

\[\text{NMB} = \text{WTP} \times \text{QALYs} - \text{Cost}\]

Highest NMB = optimal strategy — works for any number of comparators.

No more pairwise ICER confusion with 3+ strategies.

Key Insight: Trade-offs

Universal screening: - Highest cost (everyone screened) - Lowest adverse outcome rate (most cases detected) - BUT some false positives (unnecessary treatment)

Risk-based screening: - Lower cost (fewer screened) - Intermediate outcome rate (misses low-risk cases)

No screening: - Lowest cost - Highest adverse outcome rate (clinical detection only 25%)

What You Just Did

  1. ✓ Defined clinical question using Indian epidemiological data
  2. ✓ Structured decision tree with three screening strategies
  3. ✓ Populated it with parameters from published Indian literature
  4. ✓ Calculated costs, QALYs, ICERs with 4-quadrant interpretation
  5. ✓ Computed NMB ranking across all strategies
  6. ✓ Visualised results on cost-effectiveness plane

The Power of R

In Excel: Changing GDM prevalence requires tracing through multiple cells across sheets.

In R: Change one number at the top and re-run the entire analysis.

This is where R shines for HTA.

Key Takeaway

A diagnostic decision tree is a reusable template:

  • Disease screening (GDM, diabetes, cancer)
  • Diagnostic testing (sensitivity, specificity, costs)
  • Treatment decisions (treat if positive, manage outcomes)

Every diagnostic HTA follows this structure — just with different parameters.

Bonus: rdecision Package

The same GDM model rebuilt with the rdecision package:

  • Object-oriented tree — DecisionNode, ChanceNode, LeafNode
  • Built-in PSA — Monte Carlo simulation with 1,000 iterations
  • CE plane + CEAC — decision uncertainty visualised

→ See the Session 3 Bonus page on the workshop website.