Risk Ratios and Odds Ratios: Understanding and Calculating

Published June 20, 2026  |  Last updated August 10, 2026

A cardiologist reviewing a new anticoagulant finds two numbers side by side in the published literature. A cohort study reports a risk ratio of 0.65 for stroke. A case-control study on the same drug class reports an odds ratio of 0.58 for the identical outcome. Both numbers describe a protective effect, yet they come from different arithmetic, different study designs, and different assumptions about how common the outcome actually is. Confusing the two, or reporting one as though it were the other, ranks among the statistical errors peer reviewers flag most often in a methods or results section.

This guide separates the two measures cleanly. Two fully worked examples, computed independently here rather than copied from a textbook, show exactly how a risk ratio and an odds ratio can come out different on the identical dataset, and why that gap widens or shrinks depending on how common the outcome is. A later section works through when a manuscript should report a risk ratio, when only an odds ratio is defensible, and what the epidemiological literature actually says about the much-repeated rare-disease rule of thumb, which turns out to be less simple than most textbooks let on.

Key Takeaways

• A risk ratio divides the cumulative risk of an event in an exposed group by the risk in an unexposed group, and it requires knowing the true size of both groups from the start, the situation a cohort study or a randomized trial provides.
• An odds ratio divides the odds of an event in one group by the odds in another, a calculation that stays valid even when a researcher, not nature, decided how many cases and how many controls to enrol.
• In a real 189-birth cohort dataset from Baystate Medical Center, maternal smoking carried a risk ratio of 1.61 for low birth weight against an odds ratio of 2.02 on the identical numbers, a concrete illustration of how far the two measures can separate once an outcome is not rare.
• Symons and Moore’s 2002 analysis in the Journal of Clinical Epidemiology found that the gap between risk ratio, hazard ratio, and odds ratio grows with longer follow-up and with risks further from 1, not from outcome prevalence alone.
• A 2008 audit of 150 published case-control studies found that only 4 discussed the rare-disease assumption explicitly, despite the correct reading of the reported odds ratio depending on it in most of the sample.

What Sets a Risk Ratio Apart From an Odds Ratio

Risk Ratio Defined

A risk ratio, also called relative risk, divides the cumulative risk of an event in an exposed group by the cumulative risk in an unexposed group over the same fixed period. Cumulative risk is plain arithmetic, the number of people who experience the event divided by the total number who started at risk. Getting that denominator right, the true number who started at risk, is only possible in a design that enrols people first and follows them forward, a cohort study or a randomized trial.

Odds Ratio Defined

An odds ratio divides the odds of an event in one group by the odds in another, where odds itself means the probability of an event divided by the probability of the event not happening, a different quantity from risk. Jerome Cornfield’s 1951 paper in the Journal of the National Cancer Institute demonstrated the property that makes the odds ratio so useful outside cohort studies. An odds ratio calculated from cases and controls comes out identical whether the calculation starts from exposure and asks about outcome, or starts from outcome and asks about exposure. That symmetry is exactly what a case-control study needs, because a case-control study samples on the outcome by design, gathering a fixed number of cases and a separately fixed number of controls rather than following a whole population forward.

Calculating a Risk Ratio From Real Cohort Data

Rather than reuse a textbook’s rounded, hypothetical numbers, the risk ratio below comes from an actual clinical dataset, 189 births recorded at Baystate Medical Center in Springfield, Massachusetts and published by Hosmer and Lemeshow as a standard teaching example in Applied Logistic Regression. Cross-tabulating maternal smoking status against low birth weight, defined as under 2,500 grams, produces the table below, computed directly for this guide and cross-checked in R.

Risk in the smoking group works out to 30 divided by 74, or 40.5%. Risk in the non-smoking group works out to 29 divided by 115, or 25.2%. Dividing the first figure by the second gives a risk ratio of 1.61 (95% CI 1.06 to 2.44), meaning mothers who smoked during this pregnancy cohort faced 61% higher risk of delivering a low-birth-weight infant than mothers who did not. A chi-square test on the same table returns χ² = 4.92 (p = 0.026), confirming the association is unlikely to reflect chance alone, a calculation covered in full in Editverse’s guide to chi-square tests.

Calculating an Odds Ratio From a Real Case-Control Study

A cohort with a known denominator is not always available. Tuyns, Péquignot, and Jensen’s 1977 case-control study in Ille-et-Vilaine, France, enrolled 200 men with a new diagnosis of esophageal cancer as cases and a separate group of population controls, then asked each participant about past alcohol and tobacco consumption. Collapsing the reported alcohol intake into a high group (80 or more grams per day) and a low group (under 80 grams per day) and cross-tabulating against case or control status, computed directly for this guide, produces the table below, drawn from 200 cases and 775 controls.

Among cases, the odds of high alcohol exposure run 96 to 104. Among controls, that same odds figure runs 109 to 666. Dividing the exposure odds in cases by the exposure odds in controls, using the cross-product formula OR = (a × d) / (b × c), gives an odds ratio of 5.64 (95% CI 4.00 to 7.95). Heavy drinkers in this sample carried well over five times the odds of esophageal cancer compared with lighter drinkers, an association strong enough that Pearson’s chi-square with Yates’ continuity correction returns χ² = 108.2 (p < 0.0001).

Dividing the case count by the case-plus-control total in each alcohol group, 96 divided by 205 for high alcohol against 104 divided by 770 for low alcohol, produces a number that looks like a risk ratio, 3.47 on this data. That number is not a risk ratio, and reporting it as one would misrepresent the study. Tuyns and colleagues did not sample every heavy and light drinker in the source population. They sampled a fixed set of cases and a separately fixed set of controls, so the case-to-control ratio in the table reflects the researchers’ sampling decisions, not the true prevalence of esophageal cancer among heavy and light drinkers in Ille-et-Vilaine. Only the odds ratio survives that kind of sampling intact, which is precisely the property Cornfield’s 1951 paper established and why case-control methodology depends on it.

Why an Odds Ratio and a Risk Ratio Diverge on the Same Data

Textbooks often reduce the relationship between these two measures to a single line, the odds ratio approximates the risk ratio when the outcome is rare, typically under 10%. That rule is a reasonable starting heuristic, and it explains why the birth-weight example above showed real separation. A 25.2% baseline risk is nowhere close to rare, and the calculated odds ratio of 2.02 sits well above the risk ratio of 1.61 on the same table. But the rare-disease heuristic understates how the two measures actually behave.

Symons and Moore’s 2002 analysis in the Journal of Clinical Epidemiology modeled the divergence between the risk ratio, the hazard rate ratio, and the odds ratio directly, and found the gap grows with three factors multiplied together, the length of follow-up, the average event rate over that follow-up, and how far the true risk sits from 1 in either direction. Risks above roughly 2.5, they found, drive substantial divergence on their own, a factor separate from and in addition to how rare the outcome is at baseline. Pearce’s 1994 paper in the International Journal of Epidemiology adds a further layer of nuance specific to case-control studies. The odds ratio computed from a case-control table estimates three genuinely different underlying quantities depending on how controls were sampled. Controls drawn from person-time at risk throughout the study yield an odds ratio that estimates a rate ratio directly. Controls drawn from the population at risk at baseline yield an odds ratio that estimates a risk ratio directly. Controls drawn only from survivors at the end of follow-up yield the narrower, traditional odds ratio, the one that needs the rare-disease condition to approximate a risk ratio at all.

Greenland, Thomas, and Morgenstern’s 1986 critique in the American Journal of Epidemiology reached a related conclusion from a different angle, cautioning that simply including more prevalent cases in a control group does not, by itself, make an odds ratio behave like an unbiased risk ratio estimate. Rodrigues and Kirkwood’s 1990 review in the same journal walked through how the right sampling scheme lets a case-control study estimate relative risk or relative rate directly, without leaning on rarity as a crutch at all. None of this makes the classic rare-disease rule of thumb wrong exactly, it makes it one useful case inside a larger, more precise picture that depends on how a study actually selected its controls, not just on how common the outcome happens to be.

Risk Ratio or Odds Ratio? Matching the Measure to the Study Design

Study design decides which measure is even calculable before any statistics enter the picture. A cohort study or a randomized trial enrolls a defined population, exposed and unexposed, or treated and untreated, and follows every participant forward for a fixed period, which means both the numerator and the true denominator for a cumulative risk are directly observed. The Physicians’ Health Study illustrates this cleanly on real data. A randomized trial that followed 22,071 male physicians for an average of 60.2 months found a 44% reduction in myocardial infarction risk on low-dose aspirin, a risk ratio of 0.56 (95% CI 0.45 to 0.70), calculated from incidence rates of 254.8 per 100,000 person-years on aspirin against 439.7 per 100,000 person-years on placebo. That is the kind of clean, forward-looking design a risk ratio is built for.

A case-control study inverts the starting point entirely, and Editverse’s companion guide to cohort and case-control study design covers the sampling mechanics in more depth than fits here. Knol and colleagues audited 150 case-control studies published across leading medicine and epidemiology journals between 2001 and 2007 and found 125 of the 150 (83%) enrolled incident cases. Of those, 105 (84%) reported an odds ratio that could legitimately be read as a rate ratio given how controls were actually sampled. A further 17 studies (14%) could interpret their odds ratio as a genuine risk ratio, though 16 of those still needed the rare-disease assumption to make that reading hold. Only 4 of the full 150 studies discussed the rare-disease assumption explicitly at all, despite it silently underpinning most of the reported interpretations.

That gap between what a published odds ratio technically estimates and what a reader assumes it means shows up throughout related modeling choices too. Editverse’s guide to the Cox proportional hazards model and Editverse’s guide to logistic regression in medical diagnostics both address this from the modeling side, since a hazard ratio from a Cox model and an odds ratio from logistic regression carry their own, closely related interpretation traps.

Confidence Intervals for Risk Ratios and Odds Ratios

A point estimate alone, 1.61 or 2.02 or 5.64, never tells the full story. Both worked examples above were computed with a 95% confidence interval using the standard log-transformation method, since risk ratios and odds ratios are bounded at zero and skewed on their natural scale, which makes the logarithm the scale on which a normal approximation actually behaves well. The birth-weight risk ratio of 1.61 carried a 95% confidence interval of 1.06 to 2.44, an interval that stays entirely above 1 but only barely, since the lower bound sits close enough to 1 that a modestly different sample could have crossed it. The esophageal cancer odds ratio of 5.64 carried a much tighter interval, 4.00 to 7.95, reflecting the larger sample and the stronger underlying association. Whenever a confidence interval for either measure crosses 1, the data are consistent with no association at all, regardless of how far the point estimate itself sits from 1.

Common Mistakes When Interpreting Risk Ratios and Odds Ratios

• Reporting an odds ratio as though it were a risk ratio. An odds ratio of 2.0 does not mean the outcome is twice as likely in an absolute sense once the baseline risk climbs much past 10%, and describing it that way overstates the true effect to a reader skimming the abstract.
• Computing a risk ratio directly from a case-control table. The esophageal cancer example above shows exactly why this fails. The case-to-control ratio reflects a sampling decision, not the underlying prevalence, so any “risk” calculated from those raw counts is an artifact of study design rather than a real epidemiological quantity.
• Treating a marker’s odds ratio as proof of good diagnostic performance. Pepe and colleagues’ 2004 paper in the American Journal of Epidemiology found that a marker with an odds ratio of 3, calibrated to a 10% false-positive rate among controls, correctly flags only 25% of true cases as positive, a level of discrimination most clinicians would consider disappointing despite the seemingly respectable odds ratio.
• Assuming the rare-disease rule always applies. Whether an odds ratio approximates a risk ratio depends on how controls were sampled at least as much as it depends on outcome prevalence, as the section above shows.
• Skipping the confidence interval. A risk ratio or odds ratio reported without its interval hides exactly how much sampling uncertainty surrounds the number, information a reviewer will ask for if it is missing from a submitted manuscript.

Reporting a Risk Ratio or Odds Ratio for Journal Submission

Most health-science journals expect the same core elements every time a risk ratio or odds ratio appears in a results section, the point estimate, the 95% confidence interval, and the study design that made the calculation valid in the first place. Naming the design matters because a reader cannot otherwise tell whether “odds ratio 2.0” came from a case-control study, where it needs the qualifications discussed throughout this guide, or from a cross-sectional or cohort analysis using logistic regression as a modeling convenience rather than a design requirement. Editverse’s overview of regression models in medical research covers how logistic, Poisson, and Cox regression each produce their own version of a ratio measure, and getting the underlying study design right in the methods section is what lets a results table full of ratios actually mean what it claims to mean.

A study pooling several risk ratios or odds ratios across multiple published studies runs into a related, higher-stakes version of the same issue, since mixing risk ratios from cohort studies with odds ratios from case-control studies in the same meta-analysis without adjustment can distort the pooled estimate. Editverse’s meta-analysis masterclass walks through how a forest plot handles exactly this kind of mixed-measure pooling, and Editverse’s guide to sensitivity, specificity, and predictive values covers the related diagnostic-accuracy measures a risk ratio or odds ratio often gets paired with in a results table.

Conclusion

A risk ratio and an odds ratio can both describe the same association and still hand a reader two different numbers, 1.61 against 2.02 in the birth-weight example worked through above, purely because one measure needs a true population denominator and the other does not. Study design decides which measure is available in the first place, a cohort study or randomized trial for a risk ratio, a case-control study for an odds ratio, and no amount of statistical sophistication downstream repairs a mismatch made at that first design stage. The rare-disease heuristic remains a useful starting point, but Symons and Moore’s finding that follow-up length and risk magnitude matter just as much, and Pearce’s finding that control-sampling scheme changes what the odds ratio even estimates, both show the full picture asks for more care than a single rule of thumb can carry.

Getting the choice right at the design stage, and reporting the confidence interval alongside every point estimate, separates a results section a peer reviewer accepts without query from one that draws a revision request over a measure that never should have been calculated that way to begin with.

References

1. Cornfield J. A method of estimating comparative rates from clinical data. Applications to cancer of the lung, breast, and cervix. J Natl Cancer Inst. 1951;11(6):1269-1275. doi.org/10.1093/jnci/11.6.1269
2. Tuyns AJ, Péquignot G, Jensen OM. Esophageal cancer in Ille-et-Vilaine in relation to levels of alcohol and tobacco consumption. Bull Cancer. 1977;64(1):45-60. PMID 861389
3. Hosmer DW, Lemeshow S. Applied Logistic Regression. 2nd ed. New York: Wiley; 2000.
4. Pearce N. What does the odds ratio estimate in a case-control study? Int J Epidemiol. 1994;22(6):1189-1192. PMID 8144304
5. Greenland S, Thomas DC, Morgenstern H. The rare-disease assumption revisited. A critique of “estimators of relative risk for case-control studies.” Am J Epidemiol. 1986;124(6):869-883. PMID 3776970
6. Rodrigues L, Kirkwood BR. Case-control designs in the study of common diseases, updates on the demise of the rare disease assumption and the choice of sampling scheme for controls. Int J Epidemiol. 1990;19(1):205-213. PMID 2190942
7. Noma H, Misumi M, Tanaka S. Risk ratio and risk difference estimation in case-cohort studies. J Epidemiol. 2023;33(10):508-513. PMID 35753802
8. Knol MJ, Vandenbroucke JP, Scott P, Egger M. What do case-control studies estimate? Survey of methods and assumptions in published case-control research. Am J Epidemiol. 2008;168(9):1073-1081. PMID 18794220
9. Symons MJ, Moore DT. Hazard rate ratio and prospective epidemiological studies. J Clin Epidemiol. 2002;55(9):893-899. PMID 12393077
10. Pepe MS, Janes H, Longton G, Leisenring W, Newcomb P. Limitations of the odds ratio in gauging the performance of a diagnostic, prognostic, or screening marker. Am J Epidemiol. 2004;159(9):882-890. PMID 15105181
11. Steering Committee of the Physicians’ Health Study Research Group. Final report on the aspirin component of the ongoing Physicians’ Health Study. N Engl J Med. 1989;321(3):129-135. PMID 2664509
12. Monaghan TF, Rahman SN, Agudelo CW, et al. Foundational statistical principles in medical research, a tutorial on odds ratios, relative risk, absolute risk, and number needed to treat. Int J Environ Res Public Health. 2021;18(11):5669. PMID 34070675
13. George A, Stead TS, Ganti L. What’s the risk, differentiating risk ratios, odds ratios, and hazard ratios? Cureus. 2020;12(8):e10047. PMID 32983737