Clinimetrics and the DEPTh Model: Choosing the Right Clinical Metrics for Research

📌 Why We Measure: The Role of Clinimetrics

Clinical research isn’t just about collecting data—it’s about asking, “How do we know what works, what causes what, and what helps whom?” To answer this, we need clinimetrics:

Clinimetrics = clinical + metricsQuantitative, statistically valid, and clinically meaningful ways to measure:

X (determinant)
Y (endpoint)
The X→Y relationship

Metrics are not just math—they reflect clinical logic.

🔍 DEPTh Model: The Clinical Logic Backbone

Before choosing your metrics, always define your DEPTh type:

Type	Key Clinical Question
Therapeutic	Does intervention X improve outcome Y?
Etiognostic	Is X a risk factor or cause of Y?
Prognostic	What will happen to a patient with Z?
Diagnostic	Is test X accurate in detecting Y?
Methodologic	How should we measure/model this problem?

This classification leads to specific measurement logic.

💊 Therapeutic Research: Measuring Treatment Effects

Goal: Quantify how much treatment X changes outcome YDesign: Usually RCT or quasi-experimentalKey Measures: Mean difference, Risk ratio, Rate ratio, Hazard ratio

1. Using Means

When Y is continuous (e.g., BP, pain score):

Compare mean values post-treatment
Metric: Mean difference, optionally Mean Ratio

Example:Compare standard vs heated nebulization for asthma:

Mean PEFR increase: 28 L/min vs 18 L/min
Mean difference = +10 L/min

2. Using Risk (Proportion)

When Y is binary (e.g., treatment success):

Metric: Risk Ratio, Risk Difference, NNT

Example:

Control group: 40% failed
Intervention group: 10% failed
RR = 0.25, RD = −30%, NNT = 1 / 0.30 = ~4

3. Using Rate (Person-Time)

When follow-up varies or repeated exposures exist:

Metric: Rate Ratio, Rate Difference

Example:

Rubbing skin reduces pressure sore incidence
Control: 205/1000 days; Treatment: 41/1000 days
Rate Ratio = 0.2, Rate Difference = −164/1000 days

4. Using Survival (Time-to-Event)

When time until an event matters:

Metric: Median survival time, Restricted mean time, Hazard Ratio (HR)

Example:

Swallow rehab post-stroke:
- Median time to safe swallow: 5 vs 11 days
- Median difference = 6 days

⚠️ Etiognostic Research: Identifying Causal Factors

Goal: Does X increase the chance of Y occurring?

Cohort-Based

Metric: Risk Ratio, Risk Difference
Example: Glove breach during surgery increases SSI
- Risk: 7.5% vs 3.9% → RR = 1.92

Case-Control-Based

Metric: Odds Ratio only
Example: Family history and thyroid cancer risk
- OR = 7.19 for those with family history

🧠 Causal Tip:RR = 1 or RD = 0 = no effectOR = 1 = no association

⏳ Prognostic Research: Predicting Future Outcomes

Goal: What’s likely to happen next for a patient?

Subtypes per PROGRESS group:

Fundamental – describe outcome trends
Prognostic Factor – identify what predicts outcome
Prediction Models – risk calculators, AUROC
Stratified Medicine – which subgroups benefit most

Example: Predicting stroke or bleeding in AF

X: Body weight < 50 kg
Y: Time to ischemic stroke
Metric: Hazard Ratio, Median survival, Rate Ratio

Prediction Model Example:

Stroke patients → predict ICH after tPA
Metric: AUROC, Odds Ratio
Good model AUROC > 0.80

🧪 Diagnostic Research: Measuring Test Accuracy

Goal: How well does test X detect condition Y?

Subtypes:

Accuracy – sensitivity/specificity
Added Value – does the test add info beyond clinical judgment?
Prediction – no gold standard, only probability
Intervention – using test changes outcomes

Accuracy Example:

H. pylori stool test
- Sens = 80%, Spec = 95%, PPV = 89%, NPV = 91%

Added Value:

AUROC increases from 0.71 (clinical info) to 0.75 (info + test)

Diagnostic Intervention:

Alvorado score pre-alerts in ER
- Time to diagnosis ↓ 6.9 hrs
- Rupture rate ↓ 70% → Risk difference

🧠 Core Summary Table: Metrics by DEPTh Type

Metric	Therapeutic	Etiognostic	Prognostic	Diagnostic
Mean Difference	✅	❌	✅	❌
Risk Ratio / Difference	✅	✅ (cohort)	✅	❌
Odds Ratio	✅	✅ (case-ctrl)	✅	✅
Rate / Rate Ratio	✅	✅	✅	❌
Hazard Ratio	✅	✅	✅	❌
Median/Mean Survival	✅	❌	✅	❌
Sens/Spec, LR+, AUROC	❌	❌	❌	✅
NNT	✅	❌	❌	❌

✅ Key Takeaways

Use the DEPTh model to classify your clinical question—this determines the metric.
Always match the data type (X & Y) and research purpose to the right measure.
Don’t blindly use p-values—focus on clinically meaningful differences.
Know when to use Risk, Rate, Odds, or Hazards.
Understand the design base (cohort, case-control, RCT) before picking metrics.