Feinstein’s Framework for Diagnostic Research in Clinical Epidemiology: Understanding Bias, Confounding, and Study Architecture

Diagnostic Research Development: Integrating Feinstein’s Framework and Modern Diagnostic Subtypes (DEPTh Model)

Introduction

In clinical epidemiology, diagnostic research is commonly understood through two complementary conceptual frameworks:

1. Feinstein’s Framework

2. Modern Diagnostic Research Subtypes in the DEPTh model

These frameworks describe different dimensions of diagnostic research.

• Feinstein’s Framework explains the developmental stages of diagnostic knowledge.

• DEPTh Diagnostic Subtypes describe the analytic types of studies conducted once a diagnostic tool or model exists.

Therefore:

• Feinstein’s Framework → describes the sequential development of diagnostic knowledge (Phases I–IV).

• DEPTh Diagnostic Subtypes → describe the methodological forms of analysis when a diagnostic test, score, or model is being evaluated.

The most important conceptual point is:

The five diagnostic subtypes in the DEPTh model are generally reserved for later phases of research (Phases II–IV), when investigators are actively evaluating a specific diagnostic test, score, model, or diagnostic strategy.

In contrast, Phase I research is exploratory, aimed at identifying potential diagnostic clues rather than evaluating a formal diagnostic tool.

---

1. Feinstein’s Framework: Phases of Diagnostic Research

Alvan R. Feinstein proposed that diagnostic knowledge develops progressively through four phases, each addressing a different research question.

These phases represent the natural evolution of diagnostic discovery, moving from observation to clinical impact.

---

Phase I – Exploratory Diagnostic Clues

Core Question

Do patients with the target disorder have different clinical signs, laboratory findings, or characteristics compared with individuals without the disorder?

Phase I represents the initial exploratory stage of diagnostic research. The purpose is to identify potential indicators of disease that may later become diagnostic tests or predictors.

At this stage, researchers are not evaluating a diagnostic tool. Instead, they are looking for possible diagnostic clues, which may include:

• clinical symptoms

• physical examination findings

• laboratory biomarkers

• imaging characteristics

---

Forms of Phase I Studies

Phase I research can take two major forms.

1. Descriptive Diagnostic Exploration

Some Phase I studies are purely descriptive, focusing on describing the characteristics of patients with a particular disease.

Example:

A study describing the clinical presentation of early Lyme disease may report:

• frequency of fever

• frequency of rash

• joint pain prevalence

• laboratory abnormalities

The goal is simply to characterize the disease and identify potential diagnostic indicators.

These studies may report:

• prevalence of symptoms

• distribution of laboratory values

• demographic patterns

but they do not attempt to evaluate diagnostic performance.

---

2. Exploratory Comparative Studies

Other Phase I studies involve simple comparisons between groups to detect potential differences.

Example:

Researchers may compare patients with bacterial meningitis and viral meningitis to examine whether the following differ:

• CSF glucose levels

• CSF protein levels

• CSF lactate levels

Statistical tests may be applied, but the goal remains hypothesis generation, not test validation.

---

What Phase I Does NOT Do

Phase I studies do not formally evaluate diagnostic tests.

Therefore, they do not report diagnostic performance metrics, such as:

• Sensitivity

• Specificity

• Positive predictive value

• Negative predictive value

• AUROC

Instead, Phase I identifies candidate diagnostic variables that may later become:

• diagnostic biomarkers

• predictors in a clinical score

• components of a diagnostic test

These variables are then investigated further in later phases.

---

Phase II – Diagnostic Association

Core Question

Are patients with certain test results more likely to have the target disorder than patients with other test results?

In Phase II, researchers begin to examine whether the candidate markers identified in Phase I are associated with disease presence.

At this stage, investigators start evaluating the diagnostic usefulness of a specific variable or test.

Example:

Suppose Phase I research suggested that biomarker X increases in sepsis.

A Phase II study may investigate whether:

• higher levels of biomarker X increase the probability of sepsis.

Researchers may begin calculating measures such as:

• sensitivity

• specificity

• likelihood ratios

However, the population studied may still be somewhat selective or not yet fully representative of real clinical practice.

---

Phase III – Diagnostic Accuracy in Clinical Populations

Core Question

Among patients in whom it is clinically reasonable to suspect the disorder, can the test distinguish those with and without the disease?

Phase III studies evaluate the diagnostic accuracy of a test in the intended clinical population.

This stage represents the classical diagnostic accuracy study.

A key feature is that each participant receives:

• the index test (the diagnostic test being evaluated)

• the reference standard (the best available method for determining the true disease status)

Researchers then calculate diagnostic performance metrics such as:

• Sensitivity

• Specificity

• Predictive values

• Likelihood ratios

• AUROC

Example:

A study evaluating whether rapid antigen testing accurately diagnoses influenza compared with PCR.

Phase III evidence is typically required before a diagnostic test is adopted in routine clinical practice.

---

Phase IV – Diagnostic Impact on Clinical Outcomes

Core Question

Do patients who undergo the diagnostic test have better clinical outcomes than those who do not?

Even a highly accurate diagnostic test may not necessarily improve patient care.

Therefore, Phase IV research investigates whether using the test actually improves clinical outcomes.

Possible outcomes include:

• reduced mortality

• fewer complications

• faster diagnosis

• improved treatment decisions

• reduced healthcare costs

Example:

A study assessing whether point-of-care ultrasound in emergency departments reduces time to diagnosis and mortality in patients with internal bleeding.

These studies often evaluate diagnostic strategies rather than individual tests, and may involve:

• randomized diagnostic strategy trials

• comparative effectiveness studies

---

2. Modern Diagnostic Research Subtypes (DEPTh Model)

In the DEPTh framework, diagnostic research is classified according to analytic objectives.

Five major diagnostic research subtypes are recognized.

---

1. Diagnostic Accuracy Research

This subtype evaluates how accurately a diagnostic test distinguishes between patients with and without a disease.

Typical metrics include:

• Sensitivity

• Specificity

• Predictive values

• Likelihood ratios

• AUROC

Example:

Determining the sensitivity and specificity of CT pulmonary angiography for diagnosing pulmonary embolism.

---

2. Diagnostic Added-Value Research

This subtype evaluates whether a new test improves diagnostic performance when added to an existing diagnostic pathway.

Example:

Assessing whether adding a new biomarker improves the performance of a clinical model for diagnosing heart failure.

Performance improvement may be assessed using:

• change in AUROC

• Net Reclassification Improvement (NRI)

• Integrated Discrimination Improvement (IDI)

---

3. Diagnostic Prediction Research

This subtype focuses on developing multivariable diagnostic prediction models.

Multiple predictors are combined to estimate the probability of disease.

Example:

Developing a clinical score to predict appendicitis using variables such as:

• age

• abdominal pain location

• fever

• white blood cell count

Statistical approaches commonly include:

• logistic regression

• machine learning algorithms

• risk score derivation

---

4. Diagnostic Intervention Research

In this subtype, the diagnostic test itself is treated as a clinical intervention.

The focus is on whether using the test changes clinical management and improves outcomes.

Example:

Evaluating whether rapid sepsis testing reduces time to antibiotic administration and improves survival.

---

5. Dia-Prognostic Research

This subtype examines the long-term prognostic implications of diagnostic procedures.

Example:

Investigating whether colonoscopy screening reduces long-term mortality from colorectal cancer.

---

3. The Key Relationship Between Phases and Subtypes

The critical conceptual link between these two frameworks is the following:

The five diagnostic subtypes are generally applied in the later phases of diagnostic research (Phases II–IV).

This is because these phases involve active evaluation of a defined diagnostic tool, model, or strategy.

Specifically:

• Phase II–III Often involve:

  • Diagnostic accuracy studies

  • Diagnostic prediction research

  • Diagnostic added-value research

• Phase IV Often involves:

  • Diagnostic intervention research

  • Dia-prognostic research

---

Why Phase I Does Not Belong to These Subtypes

Phase I research occurs before a formal diagnostic tool exists.

Researchers are simply identifying potential diagnostic indicators through:

• descriptive studies

• exploratory comparisons

• hypothesis generation

Because no diagnostic test or model is being formally evaluated, Phase I studies do not fall into any of the five diagnostic subtypes.

---

Conceptual Flow of Diagnostic Research

Phase I
Identify diagnostic clues
(descriptive or exploratory research)

↓

Phase II–III
Evaluate diagnostic tests or models
(diagnostic accuracy, prediction, added value)

↓

Phase IV
Evaluate clinical impact
(diagnostic intervention or dia-prognostic research)

---

Conclusion

Diagnostic research can be understood through two complementary frameworks.

Feinstein’s Framework

Describes the developmental progression of diagnostic knowledge, moving from exploratory observations to evaluation of clinical impact through four phases.

DEPTh Diagnostic Subtypes

Describe the analytic types of diagnostic studies conducted when a diagnostic tool or model is being evaluated.

The key insight is that:

The five diagnostic subtypes are generally reserved for the later phases of research (Phases II–IV), where investigators actively evaluate a specific diagnostic test, score, model, or diagnostic strategy.

In contrast, Phase I research focuses on identifying potential diagnostic clues, often through descriptive or exploratory studies that generate hypotheses for later investigation.

Understanding the relationship between these two frameworks helps researchers design and interpret diagnostic studies within the broader structure of clinical epidemiology.