The odds ratio (OR) is a measure of the strength of an association between an exposure (e.g., treatment, risk factor, or biomarker) and a binary outcome (e.g., event yes/no). It does not compare risks directly, but rather the ratio of the odds in two groups. Odds are defined as the ratio of the probability of an event occurring to the probability of it not occurring: \( ext{Odds}=p/(1-p)\). An OR is therefore the ratio of two odds and is frequently reported in case-control studies as well as in logistic regression.
In the daily routine of CROs and sponsors, the odds ratio is primarily encountered in safety analyses (e.g., occurrence of specific adverse events), in responder definitions (e.g., response yes/no), and in real-world analyses where events are modeled in routine data. It is crucial to interpret the OR correctly and to clearly distinguish it from relative risk and hazard ratio to prevent miscommunication with clinical teams, management, or regulatory authorities.
Definition and Calculation in the 2×2 Table
In a 2×2 table, there are \(a\) events and \(b\) non-events in the exposure group, and \(c\) events and \(d\) non-events in the comparison group. The odds in the exposure group are then \(a/b\) and in the comparison group \(c/d\). The odds ratio is calculated as:
\( ext{OR}=(a/b)/(c/d)=(a\cdot d)/(b\cdot c)\).
An OR of 1 indicates no difference between the groups. Values greater than 1 suggest that the odds for the outcome are higher in the exposure group; values less than 1 suggest lower odds. In reports, the OR is almost always provided alongside a 95% confidence interval and a p-value to allow for separate consideration of effect size and uncertainty.
Interpretation and Typical Misunderstandings
An OR of 2 means that the odds for the event in the exposure group are twice as high as in the comparison group. This is not automatically synonymous with “double the risk.” For rare events (e.g., below 10%), the OR is often close to the relative risk. For frequent events, however, the OR can be significantly larger, making the effect appear stronger than it actually is in terms of risk measures.
In practice, this means: if a stakeholder primarily wants to understand “how many additional events?”, absolute risks, risk differences, or number needed to harm/number needed to treat should be reported additionally—provided the design allows for it. If this is not possible, one should at least mention baseline rates from comparable datasets or make the limitations transparent.
Odds Ratio in Case-Control Studies and Observational Data
Case-control studies start with the outcome: cases (with the event) and controls (without the event) are recruited, and exposures are examined retrospectively. In this design, absolute risks are typically not directly estimable because the case/control ratio is determined by the design. The odds ratio is the standard measure here because it can still consistently represent the association between exposure and outcome.
In pharmacoepidemiology, case-control approaches are used to investigate rare adverse events. Typical risks of bias include confounding by indication, selection bias, and information bias. Sponsors and CROs address these through matching (e.g., by age, gender, index date), restriction, propensity score methods, and a pre-defined analysis strategy. Furthermore, the precise definition of exposure windows and outcome validation is important.
Logistic Regression: Adjusted Odds Ratios
In logistic regression, the logit—the logarithm of the odds—is modeled as a linear function of covariates. By exponentiating the regression coefficient, one obtains an odds ratio per unit of the covariate or in comparison of a category to the reference. This results in adjusted odds ratios, which describe the association while controlling for other influencing factors.
For observational data, adjustment is often crucial. Typical covariates include age, gender, comorbidities, baseline severity, concomitant medication, center effects, or calendar time. A pre-defined Statistical Analysis Plan (SAP) with rules for variable selection, handling of missing data, and sensitivity analyses is an essential quality assurance measure. For audit and inspection readiness, it should be documented how variables were derived from codes, laboratory data, or text fields, and what data cleaning procedures were performed.
Distinction from Relative Risk and Hazard Ratio
Relative risk compares probabilities directly and is often easier to interpret for randomized parallel-group studies. The hazard ratio is used in time-to-event analyses and describes the ratio of hazard rates over time. The odds ratio is particularly suitable for binary outcomes and is the central measure in certain designs (e.g., case-control). Study reports should clearly justify why a specific effect measure was used so that readers can correctly categorize the statement.
A practical tip: if a result appears very “large,” it is worth performing a plausibility check via absolute event rates. In many projects, OR-based models can additionally be tested through alternative specifications (e.g., risk models, different adjustments, sensitivity analyses) to demonstrate robustness.
FAQ: When is the odds ratio particularly suitable?
Primarily in case-control studies and in logistic regression when the outcome is binary and adjusted modeling is required.
FAQ: Why does the OR deviate from the relative risk?
Because odds and risks are different quantities. For frequent events, odds increase faster than risks, which can make the OR appear larger than the relative risk.
FAQ: How do I explain an OR clearly?
By explaining the OR in the context of baseline risks and, if possible, additionally reporting absolute event rates or risk differences.
Regulatory References (Selection)
- ICH E9 Statistical Principles for Clinical Trials – Principles for the interpretation of statistical measures
- ICH E6(R3) Good Clinical Practice – Requirements for data integrity, documentation, and traceability
- Regulation (EU) No 536/2014 – Context of clinical trials and documentation requirements