Odds ratio vs risk ratio in nursing research is an important distinction when you interpret a study’s findings. Both measures compare groups, but they describe different quantities. Confusing odds with probability can make an association sound much larger than it is, especially when the outcome is common.
This guide explains the calculations, study-design choices and reporting language through invented nursing examples. It focuses on choosing and interpreting these two measures. For model-building advice, use our separate nursing regression analysis guide.
Odds ratio vs risk ratio in nursing research: quick answer
A risk ratio compares the probability of an outcome between two groups over a defined observation period. An odds ratio compares the odds of the outcome, where odds equal the number with the outcome divided by the number without it.
Both measures have a null value of one. Values above or below one indicate the direction of the defined comparison. However, an odds ratio of two does not generally mean that an outcome is twice as likely. Always identify the measure, outcome and reference group.
The appropriate choice depends on the design and analysis. Cohort studies and trials can often estimate risks directly. In a conventional case-control study, the sampling process fixes or selects the numbers of cases and controls, so the sample does not usually provide the population risks needed for a direct risk ratio.
Understand probability before comparing groups
Suppose 20 of 100 patients experience an outcome during a specified follow-up period. The observed risk is 20/100, or 0.20. The odds are 20/80, or 0.25, because 80 patients do not experience the outcome.
The denominators differ. Risk divides events by all people at risk, whereas odds divide events by non-events. Both quantities are valid, but they need different words in your results and discussion chapters.
| Quantity | Calculation | Example with 20 events among 100 people |
|---|---|---|
| Risk | Events ÷ total at risk | 20/100 = 0.20 |
| Odds | Events ÷ non-events | 20/80 = 0.25 |
| Risk ratio | Risk in one group ÷ risk in reference group | Requires the second group’s risk |
| Odds ratio | Odds in one group ÷ odds in reference group | Requires the second group’s odds |
A risk must also have a meaningful time frame. Readmission within 30 days is a different outcome from readmission at any time during a year. The measure should connect the outcome definition, eligible population and follow-up period.
Worked nursing example: a common outcome
These numbers are invented for teaching and do not describe a real study. Suppose a nurse-led follow-up programme is compared with usual care. Each group contains 100 independently observed patients, and the outcome is completion of a planned review within two weeks.
| Group | Review completed | Not completed | Total |
|---|---|---|---|
| Programme | 60 | 40 | 100 |
| Usual care | 40 | 60 | 100 |
The programme risk is 60/100 = 0.60. The usual-care risk is 40/100 = 0.40. Therefore, the risk ratio is 0.60/0.40 = 1.50. The observed completion probability is 50% higher relative to the usual-care probability.
The programme odds are 60/40 = 1.50. The usual-care odds are 40/60, approximately 0.667. Therefore, the odds ratio is 1.50/0.667 = 2.25. The programme has 2.25 times the observed odds of completion.
These are different numerical summaries of the same table. Saying that patients were “2.25 times as likely” to complete a review would incorrectly describe the odds ratio as a risk ratio. The actual risk ratio in this example is 1.50.
Add the absolute difference
The risk difference is 60% − 40% = 20 percentage points. This gives another useful description of the sample: 20 additional completions per 100 patients were observed in the programme group compared with usual care.
Be careful with the word “additional”. In this teaching table it describes the observed comparison. Whether the programme caused those extra completions depends on the design, allocation and other influences.
Report the group frequencies alongside the relative measure. Readers can then see how the result relates to the baseline probability. A ratio without its starting risk can leave the practical scale unclear.
Why the measures can look similar for rare outcomes
Consider another invented table, this time with two events among 100 patients in one group and one event among 100 in the other. The risk ratio is 0.02/0.01 = 2.00. The odds ratio is (2/98)/(1/99), approximately 2.02.
Here the measures are close because the event is uncommon in both groups. When probabilities are small, the non-event denominator is close to the total denominator. The odds and risk calculations consequently produce similar values.
The CDC’s measures-of-association guidance discusses this relationship and the use of odds ratios in case-control studies. It does not justify treating every reported odds ratio as if it were a risk ratio (CDC, n.d.).
Do not rely on a single universal percentage threshold for “rare”. Check the event frequency in the relevant population and groups. A rare-outcome approximation is a contextual judgement, rather than permission to ignore the measure label.
Match the measure to the study design
In a cohort study, people are followed from a defined starting point and outcomes are observed. If suitable denominators and follow-up information are available, the risks in exposure groups can be compared. A randomised trial may similarly provide group risks over a planned follow-up period.
A conventional case-control study starts by selecting people with and without the outcome. The proportion of cases in the assembled sample reflects the sampling strategy, rather than necessarily representing the population probability of the outcome.
Therefore, dividing the number of sampled cases by all sampled participants usually does not recover population risk. An odds ratio may be estimable under the design, but its interpretation should reflect the sampling method. Some case-control sampling schemes target other population measures, so obtain design-specific advice when needed.
Cross-sectional research requires another distinction. A questionnaire or record review at one time point often measures prevalence, rather than incidence over follow-up. Terms such as prevalence ratio or prevalence odds ratio may be more accurate than calling every comparison a risk.
Use our methodology decision guide when defining the design. The choice should follow the question and data structure, rather than the measure that produces the most striking number.
What logistic regression reports
Binary logistic regression commonly reports exponentiated coefficients as odds ratios. In SPSS, these may appear under Exp(B). Interpret the output as a comparison of odds under the specified model, not automatically as a probability ratio.
For a categorical predictor, identify its reference group. An adjusted odds ratio compares the relevant category with that reference while accounting for the model’s other included predictors. The interpretation remains conditional on the specification and assumptions.
For a continuous predictor, identify the unit increase. An odds ratio per one point on a scale is different from an odds ratio per ten points. If the model uses a transformation or standardised variable, explain that coding explicitly.
Adjustment does not prove that confounding has been eliminated. Variables can be measured imperfectly or omitted, and a model can be unsuitable for the sample. Report which covariates were included and why.
Do not diagnose confounding from a changed odds ratio alone
An adjusted odds ratio can differ from an unadjusted odds ratio for reasons beyond confounding, including the mathematical property known as non-collapsibility. Therefore, a change in the estimate is not sufficient on its own to prove that a variable was a confounder.
For a dissertation, focus on a justified analysis plan and a transparent account of the model. If comparing adjusted and unadjusted results becomes central to your conclusions, obtain appropriate statistical advice.
Can a model estimate a risk ratio directly?
There are modelling approaches that target risk ratios for suitable binary outcomes, including log-binomial models and modified Poisson approaches with appropriate variance estimation. These methods have their own requirements and should be selected through statistical advice.
Do not assume that changing a menu setting transforms every odds ratio into a valid adjusted risk ratio. The outcome definition, design, link function, variance calculation and model fit all matter.
Similarly, converting an adjusted odds ratio to a risk ratio using an arbitrary baseline probability can be misleading. A simple conversion from a known table is not automatically equivalent to estimating an adjusted measure in a multivariable model.
For a small undergraduate or master’s project, clear unadjusted reporting may be more appropriate than an overcomplicated model, depending on the research question. Discuss the scope with your supervisor before promising an analysis your data cannot support.
Define the outcome and reference group
The same dataset can produce a ratio above one or below one depending on the comparison order. If the group order is reversed, the ratio becomes its reciprocal. A result of 0.5 under one definition may correspond to 2.0 under the reversed definition.
Outcome coding also matters. In the worked example, the favourable outcome is review completion. If you model non-completion instead, your narrative must change accordingly. Check the software’s event category rather than relying on memory.
A complete reporting sentence names the event, groups and time frame. For example: “The programme group had a higher observed probability of review completion within two weeks than usual care (RR 1.50).” The accompanying table supplies the 60% and 40% frequencies.
Avoid writing only that a group was “at greater risk” when the outcome is desirable. Although risk can be a neutral statistical term, ordinary readers may interpret it as harm. Use probability or completion language where it improves clarity.
Report uncertainty around either measure
An odds ratio or risk ratio without an interval can encourage overconfidence. A very large estimate from a small number of events may be highly uncertain. Include an appropriately calculated confidence interval and identify its confidence level.
The null value is one for both ratios. However, reading whether the interval includes one is only a first step. Consider the full range of plausible magnitudes and how they relate to decisions in the nursing context.
If the interval spans important benefit and important harm, describe the imprecision. Do not conclude that the groups are identical simply because a significance threshold was not crossed.
Our confidence-interval guide explains how to read that uncertainty. Always ensure the interval belongs to the estimate, analytical sample and model you are reporting.
Handle zero cells and sparse data carefully
A table with no events in one group can make a simple ratio undefined, zero or infinite, depending on the calculation. The data may be genuine, but the resulting estimate needs an appropriate method and careful interpretation.
Do not add a small number to every cell without explaining why. Continuity corrections are analytical choices with consequences, rather than neutral repairs. Software packages may use different defaults, which can affect estimates and intervals.
For an association test in a sparse independent 2 × 2 table, our Fisher’s exact test guide provides the relevant decision checks. An exact p-value still does not resolve every difficulty in estimating an effect.
Check whether observations are independent. Multiple admissions from the same patient, matching or ward clustering can require another approach. A table’s small size does not tell you whether its entries represent independent information.
Separate an association from a causal claim
A ratio quantifies a comparison under the chosen analysis. It does not automatically show that an exposure or intervention caused the outcome. Recruitment, baseline differences, measurement and follow-up can all influence the observed association.
In a non-randomised programme evaluation, patients who attend follow-up may differ from those receiving usual care. Adjustment can address some measured differences, but it cannot guarantee that every relevant influence has been removed.
Use language proportionate to the design. “Was associated with” is often more suitable for observational findings than “led to”. A randomised study still needs consideration of missing outcomes, adherence and other limitations before drawing strong conclusions.
The CDC’s analysis guidance emphasises aligning the analytical strategy with the design and planned question. Use that principle when deciding what the relative comparison can support.
Write the results and discussion clearly
A results paragraph should report the actual measure and its direction, with counts, denominators and uncertainty. Distinguish adjusted and unadjusted estimates. Explain the reference category in the table notes if space is limited.
A discussion paragraph can then connect the result with existing evidence. Before comparing ratios across papers, check whether they estimate the same outcome, population, period and measure. An odds ratio from a case-control study may not be directly comparable with a risk ratio from a cohort.
Do not describe a relative increase as an absolute percentage-point increase. In the worked example, the risk ratio of 1.50 represents a 50% relative increase, while the absolute difference is 20 percentage points. Both statements are correct when labelled accurately.
Readers also need the baseline probability. An identical relative ratio can correspond to very different absolute differences. Comparing a rare adverse outcome with a common attendance outcome requires more than placing the two ratios beside each other.
A practical ratio-reporting checklist
- Define the outcome and follow-up period.
- Identify the study design and sampling method.
- Check the event and reference categories.
- Distinguish probability from odds.
- Calculate or verify the underlying group frequencies.
- Label the measure as RR or OR accurately.
- Report an appropriate interval and confidence level.
- State whether the estimate is adjusted.
- Include absolute context where available.
- Avoid unsupported causal or equivalence claims.
A useful final check is to ask whether a reader could reconstruct your unadjusted ratio from the displayed table. If the calculation does not match, investigate coding, exclusions, weighting or a possible difference between crude and model-based estimates.
Need help interpreting your comparison?
Use nursing dissertation data analysis support to review coding, model output and reporting language. Bring anonymised tables and your study plan. The goal is to help you understand and explain your own analysis in line with university requirements.
Frequently asked questions
Is a risk ratio always better than an odds ratio?
No. The choice depends on the design and analytical target. The main requirement is to use a suitable measure and explain it accurately.
Does an odds ratio of two mean twice as likely?
Not generally. It means twice the odds under the defined comparison. The probability ratio can be substantially different when the outcome is common.
Why do case-control studies often use odds ratios?
The selected numbers of cases and controls usually do not provide the population risks. An odds ratio can be estimated under suitable case-control sampling assumptions.
What does Exp(B) usually mean in binary logistic regression?
It is the exponentiated coefficient, commonly interpreted as an odds ratio. Check the outcome coding, predictor unit and reference group.
What does a ratio of one mean?
It represents equal risk or equal odds for the defined comparison. It does not establish that every aspect of the groups is identical.
Name the measure before explaining the finding
Clear nursing reporting starts by distinguishing the probability of an outcome from its odds. Choose a measure that fits the design, identify the reference group and present uncertainty alongside absolute context. These steps prevent a valid statistical result from becoming a misleading practice claim.
References
- Centers for Disease Control and Prevention (n.d.) Principles of Epidemiology: Lesson 3, Section 5, Measures of Association. Available at: CDC guidance (Accessed: 30 September 2026).
- Dicker, R.C. (2025) Analyzing and Interpreting Data. CDC Field Epidemiology Manual. Available at: Chapter 8 (Accessed: 30 September 2026).
- Davies, H.T.O., Crombie, I.K. and Tavakoli, M. (1998) ‘When can odds ratios mislead?’, BMJ, 316, pp. 989–991. Available at: doi:10.1136/bmj.316.7136.989.
