Etiologic Research (Etiognosis) vs Prognostic Research: Domain Logic, Time Zero, and Causal vs Predictive Questions

Mayta
Jan 14
5 min read

Abstract

Etiologic and prognostic research addresses different clinical questions because they begin from different patient domains at baseline (“time zero”). Etiologic research studies individuals who have not yet experienced the target event and asks whether an exposure contributes to the occurrence of that event, often with causal intent. Prognostic research studies individuals who have already experienced a defining event (such as diagnosis or an index clinical event) and asks what outcomes will occur next, with prediction rather than causation as the primary goal. This article clarifies the distinction using domain logic, time-zero alignment, and the causal-versus-predictive split, and provides practical rules to classify common study questions.

1. Introduction

Clinical epidemiology is easiest when the starting point is the patient at baseline. Many design and interpretation errors occur when investigators confuse etiologic language (“risk factor,” “cause”) with prognostic language (“predictor,” “course”), or when they define time zero inconsistently across patients. The most reliable way to separate etiologic from prognostic research is to classify the study domain at baseline and then determine whether the intended inference is causal or predictive.

2. The Primary Distinction: Domain at Time Zero

The fundamental difference between etiologic and prognostic research is the domain of eligible participants at baseline.

Etiologic (etiognostic) research domain: individuals who have not yet experienced the target event at time zero (they are “at risk” for the event).
Prognostic research domain: individuals who have already experienced a defining event at time zero (for example, diagnosis, ICU admission, surgery, or a first event such as myocardial infarction). These individuals are followed to observe what happens next.

This “domain rule” determines the natural design defaults (cohort entry, measurement timing), the appropriate effect measures, and how results should be interpreted.

3. Clinical Question Orientation

Because the domains differ, the clinical questions differ.

Etiologic research asks:

Why did the disease/event occur?
Does exposure X contribute to the incidence of outcome Y?

Prognostic research asks:

Given the disease/index event, what will happen next?
Which variables predict later outcomes among patients who already have the condition?

A practical restatement: etiologic research is about occurrence, while prognostic research is about course after a defined inception point.

4. Time Zero and the Inception Point

Time zero must be explicit and consistent.

In etiologic research, time zero is typically the moment a person enters the risk set (disease-free or event-free for the outcome of interest). Exposure measurement must precede outcome occurrence and should respect induction/latency periods when relevant.
In prognostic research, time zero is the inception point (for example, time of diagnosis, admission, surgery, or first event). Predictors must be measured at or before this point of prediction, and outcomes must be defined over a specified horizon (e.g., 30-day mortality, 1-year recurrence, time to progression).

Misalignment of time zero is a common fatal flaw: combining patients at different illness phases or measuring predictors after the prediction point undermines interpretability and can introduce bias.

5. Causal Intent vs Predictive Intent

Domain determines the natural question, but intent determines the analytic logic.

5.1 Etiologic research: often causal, sometimes exploratory

Etiologic research can be conducted in two paradigms:

Explanatory (causal) etiologic research: aims to estimate whether exposure X causes outcome Y; requires confounding control guided by causal structure (e.g., DAG reasoning), careful adjustment strategy, and strong attention to selection and information bias.
Exploratory (associational) etiologic research: seeks potential associations or hypothesis generation without claiming causation; confounding control is less central because causal claims are not the goal.

5.2 Prognostic research: non-causal by default

Prognostic research is designed to describe and predict outcomes among those with a condition. Its primary aim is prediction and stratification, not establishing that a factor “causes” a future outcome. Variables may be markers of severity or disease biology rather than manipulable causes. The emphasis is on defining the prediction point, measuring predictors before outcomes, and using appropriate performance concepts if building models (discrimination, calibration, validation).

5.3 Where this sits in the broader DEPTh logic

A consistent classification across clinical research types is:

Diagnostic research: predictive/non-causal orientation (classification accuracy and prediction)
Etiologic research: often causal, but can be non-causal if exploratory
Prognostic research: predictive/non-causal orientation (course prediction)
Therapeutic research: causal orientation (effects of interventions; best supported by randomized or target-trial-emulation logic)

This framing prevents the common interpretive error of reading prognostic associations as causal effects.

6. Typical Designs and Measures

6.1 Etiologic research design defaults

Preferred design: cohort (closed or dynamic), or efficient case-control variants for rare outcomes; selection and sampling should map to a coherent source population.
Common measures: risk ratio, odds ratio, incidence rate ratio, hazard ratio (time-to-event).

6.2 Prognostic research design defaults

Preferred design: inception cohort with structured follow-up from a defined prediction point; retrospective cohorts can be used if time-zero and measurement integrity are preserved.
Common measures: survival probability over time, median survival, Kaplan–Meier curves; hazard ratios may be used for prognostic factor associations but should be interpreted as predictive associations rather than causal effects.

7. One Exposure, Different Research Types

A single exposure can appear in different research types depending on domain and intent.

Example: smoking

Etiologic: among people without COPD at baseline, does smoking contribute to incident COPD? (etiologic domain; potentially causal intent)
Prognostic: among people with COPD at baseline, does smoking status predict mortality or exacerbations? (prognostic domain; predictive intent)
Therapeutic: among people with COPD, does a cessation intervention reduce mortality? (therapeutic domain; causal intent)

The variable does not define the research type; the baseline domain and question do.

8. Common Classification Errors

Using etiologic language (“risk factor,” “cause”) to describe prognostic predictors.
Mixing incident and prevalent cases in prognostic cohorts without a clear inception point.
Treating prognostic hazard ratios as causal effects without a causal design and justification.
Measuring predictors after time zero (post-baseline predictors presented as baseline predictors).
Designing an etiologic question but using a base that does not represent a valid source population (selection bias).

9. Practical Checklist for Rapid Classification

At baseline (time zero), has the patient already experienced the defining event?
- No: etiologic domain.
- Yes: prognostic domain.
Is the primary goal to estimate a causal effect of an exposure/intervention, or to predict outcomes?
- Causal effect: etiologic-explanatory or therapeutic logic.
- Prediction/course: prognostic logic.
Is time zero explicit and consistent across participants?
- If not, redesign the cohort definition before modeling.

Conclusion

Etiologic and prognostic research differ primarily by the patient domain at time zero: etiologic studies begin before the target event occurs, while prognostic studies begin after a defining event has already occurred. Etiologic research often supports causal inference when designed to control confounding and bias; prognostic research supports prediction and clinical stratification and is non-causal by default. Correct classification using domain and time-zero logic prevents design mismatches and interpretive errors and leads to appropriate analysis choices and clinically meaningful conclusions.