What Is Prognostic Research? PROGRESS I: Overall Prognosis Research, PROGRESS II: Prognostic Factor Research, PROGRESS III: Prognostic Model Research, PROGRESS IV: Stratified Medicine Research

What Is Prognostic Research?

Prognostic research aims to understand what is likely to happen to a patient after diagnosis. Unlike causal studies, it doesn’t ask why something happens, but what will likely happen next.

“Given this diagnosis, what outcomes can we expect?”

This framework—PROGRESS—categorizes prognosis research into four distinct domains:

---

🟩 PROGRESS I: Overall Prognosis Research

Descriptive research for population-level insight.

Core Purpose:

To describe the average future outcomes of patients under current diagnosis and treatment contexts. This is not about testing hypotheses—but about benchmarking and tracking health outcomes over time.

Example Questions:

• What is the 5-year survival of breast cancer patients in Thailand?
• What is the median survival of metastatic lung cancer patients in your hospital?
• How do survival rates for cervical cancer differ by province?

Characteristics:

• No hypothesis testing
• No statistical comparisons
• Outputs: Kaplan-Meier curves, median survival, age-standardized survival
• Policy- and system-level implications, not individual care decisions

Design Logic:

• Cohort design (population-based or hospital registry)
• Descriptive statistics only
• Updated every few years as clinical practice evolves

🔍 Insight: This type helps governments and health systems monitor care quality, but cannot guide treatment for individuals.

---

🟨 PROGRESS II: Prognostic Factor Research

What variables are associated with better or worse outcomes?

Core Purpose:

To identify individual-level factors that are associated with future outcomes—such as survival, disease recurrence, or functional decline.

Example Questions:

• What factors predict in-hospital mortality among AECOPD patients?
• Does age affect survival in metastatic breast cancer?
• Is elevated CRP predictive of complications in dengue?

Characteristics:

• May be explanatory (causal-oriented) or exploratory (association-seeking)
• Requires multivariable analysis (e.g., Cox or logistic regression)
• Relative measures (e.g., HR, OR) must be complemented by absolute risks

Design Logic:

• Prospective cohort preferred (for timing clarity)
• Variables should be:
  • Accessible
  • Affordable
  • Interpretable
  • Reliable across settings

🔍 Insight: A prognostic factor is not always causal, but it can be clinically useful for triage, decision-making, or designing future prediction models.

---

🟦 PROGRESS III: Prognostic Model Research

Can we combine factors to predict individual-level risk?

Core Purpose:

To develop or validate a multivariable risk model (also called a clinical prediction rule or CPR) that provides a personalized risk prediction.

Example Questions:

• Can we predict 48-hour AKI in patients undergoing PCI?
• Can we build a scoring system for 6-month survival in HCC with spinal metastasis?

Workflow:

1. Select candidate predictors (literature/clinical logic)
2. Fit multivariable model (e.g., Cox regression)
3. Derive simplified score (e.g., point-based)
4. Validate (internal via bootstrapping, external in new dataset)
5. Present with score charts, nomograms, or heatmaps

Output Use:

• Decision support tools
• Risk stratification
• Personalized patient counseling

🔍 Insight: Prediction ≠ Explanation. These models don’t tell you why, but who is at risk—and by how much.

---

🟥 PROGRESS IV: Stratified Medicine Research

Who benefits most (or least) from a specific treatment?

Core Purpose:

To identify predictive factors that indicate which patients respond differently to a specific treatment.

Example Questions:

• Do EGFR+ lung cancer patients respond better to gefitinib than chemotherapy?
• Does CD-4 count modify ART treatment success?

Characteristics:

• Requires treatment × biomarker interaction logic
• Ideally performed in RCTs or target trial emulations
• Often secondary analyses of RCTs (i.e., subgroup analyses)

Design Logic:

• Identify biomarker/subgroup
• Compare treatment effects within strata
• Validate interaction effect

🔍 Insight: This is the entry point into personalized medicine—where prognosis meets therapeutic decision-making.

---

🔁 Summary Table

---

✅ Key Takeaways

• Prognostic research starts after diagnosis—it’s not about cause, but about course.
• PROGRESS I = average outcomes, PROGRESS III = individual risk prediction.
• Always define your point of prediction—that’s your cohort time zero.
• Don’t confuse prognostic factors with causal factors—logic and purpose differ.
• Validate prediction models before use—calibration + discrimination both matter.