Understanding number needed to treat (NNT): A practical guide for anaesthesia and critical care clinicians

INTRODUCTION The number needed to treat (NNT) is a metric in clinical research that provides a direct measurement of the absolute effectiveness of an intervention. A commentary describing the imperative of reporting NNT in anaesthesia research was published in the Indian Journal of Anaesthesia (IJA) by Ganesh et al.1 This editorial is a follow-up on this commentary to emphasise the importance of translating research findings into meaningful clinical decisions for clinicians. While P values and confidence intervals dominate statistical reporting,2 NNT provides a more intuitive measure of clinical effectiveness. This editorial provides a step-by-step approach to calculating and interpreting NNT in both randomised controlled trials (RCTs) and observational studies, with hypothetical examples from anaesthesia and critical care practice. UNDERSTANDING THE 2 × 2 TABLE: FOUNDATION OF NNT CALCULATIONS The 2 × 2 contingency table forms the cornerstone of NNT calculations. This simple yet powerful tool organises data to compare outcomes between treatment and control groups Table 1.Table 1: Structure of the 2 × 2 tableKEY RATE CALCULATIONS Absolute risk in treated = Experimental event rate (EER) = a/(a + b); Absolute risk in controls = Control event rate (CER) = c/(c + d); Absolute risk reduction (ARR) = ARcontrol− ARtreated = CER − EER Number needed to treat (NNT) = 1/ARR STEP-BY-STEP NNT CALCULATION Example from RCT Example 1: Postoperative nausea and vomiting (PONV) prevention. Clinical scenario: A randomised trial comparing ondansetron versus placebo for PONV prevention in 400 patients undergoing laparoscopic cholecystectomy. RESULTS Ondansetron group (n = 200): 30 patients experienced PONV Placebo group (n = 200): 80 patients experienced PONV. Step 1: Construct the 2 × 2 table Table 2.Table 2: 2 x 2 table for PONV exampleStep 2: Calculate event rates: EER (ondansetron) = 30/200 = 0.15 (15%) CER (placebo) = 80/200 = 0.40 (40%). Step 3: Calculate ARR: ARR = CER − EER = 0.40 − 0.15 = 0.25 (25%) Step 4: Calculate NNT: NNT = 1/ARR = 1/0.25 = 4 Interpretation: For every four patients treated with ondansetron instead of a placebo, one case of PONV is prevented. Example from observational study Example 2: Intensive Care Unit (ICU) mortality and early mobilisation. Clinical scenario: A cohort study examining the association between early mobilisation and ICU mortality in 800 mechanically ventilated patients. Steps 1 and 2: 2 × 2 table of results and effect measures Table 3.Table 3: 2 x 2 table of ICU mortality exampleStep 3: Calculate ARR: ARR = CER − EER = 0.25 − 0.15 = 0.10 (10%) Step 4: Calculate NNT: NNT = 1/ARR = 1/0.10 = 10 Interpretation: For every 10 patients who receive early mobilisation instead of standard care, one death is prevented. Understanding risk vs odds: Fundamental concepts It is crucial to understand the distinction between risk and odds, as these concepts are often confused in clinical practice. The key is in the denominator Table 4. The terms absolute risk and risk are often used interchangeably. The prefix “absolute” is used to denote direct probability or frequency of events in a group, and it is imperative to avoid confusion with “relative risk,” which is a ratio of absolute risks.Table 4: Difference between risk, odds, and relative measuresMathematical relationship between risk and odds If Risk = R, then: Odds = R/(1 − R) and Risk = Odds/(1 + Odds) Importance of baseline risk When the baseline risk or CER is high, as in the PONV example described above (40%), there is a wide difference between the relative risk (RR) and odds ratio (OR) (RR = 0.38 vs OR = 0.27), whereas in the ICU mortality example (25% CER), the difference between RR and OR is much narrower (RR = 0.60 vs OR = 0.529). In general, the OR always appears more extreme than RR and is not a good measure when the baseline risk (CER) is high. In the case of a rare outcome (i.e., the prevalence of the outcome is ≤10%), the OR will approximate the RR. For rare outcomes, the OR and RR are similar in magnitude. As the incidence (baseline risk) of the outcome increases, the OR diverges further from the RR and becomes more extreme. This is important for interpretation, especially in case-control studies where only ORs are estimated directly.3 Importance of absolute risk, absolute risk difference/reduction, and NNT While relative measures (RR and OR) provide the strength of association, absolute risk difference (ARD) or the absolute risk reduction (ARR) give the actual probability of experiencing the outcome. This distinction is crucial for clinical decision-making. Example: Consider two scenarios with identical relative effects: Scenario A: CER = 40%, EER = 25%, ARR = 15%, NNT = 7, RR = 0.625 Scenario B: CER = 4%, EER = 2.5%, ARR = 1.5%, NNT = 67, RR = 0.625. Both have the same RR reduction (37.5%), but the absolute benefit and clinical impact differ dramatically. Examining the above examples, we can see that when the RR is constant, if the CER is low (low baseline risk, low-risk population) and the ARR is high, the NNT is very small. If the CER is high (high-risk population) and the ARR is low, the NNT is high. This tells us that a treatment with a consistent relative effect (e.g., a constant RR or OR) will have a lower NNT in populations with higher baseline risk and a higher NNT in low-risk populations, because absolute risk reductions scale directly with baseline risk. This underscores the importance of considering baseline risk when translating relative measures (RR/OR) into absolute patient benefits (NNT), as NNT can vary widely across patient populations even if the relative effect is unchanged. Why both absolute and relative measures matter? In clinical research and practice, the distinction between absolute and relative measures is not merely statistical—it is fundamental to clear, honest communication and patient-centred care. RR helps clinicians and patients understand “how much does a treatment reduce risk?”—for example, “Ondansetron reduces your nausea risk by 62.5%.” Absolute risk addresses the more personal question, “What is my actual probability of the outcome?” such as, “Your risk of nausea drops from 40% to 15%.” The NNT translates these effects into tangible terms: “How many patients need treatment for one to benefit?” In our PONV example, “For every four patients treated, one case of nausea is prevented.” Each of these measures provides a different lens on clinical impact, and together they offer a more complete picture than any single statistic alone Table 2. Calculating NNT for different types of categorical outcomes The NNT can be calculated from summary measures presented for different types of categorical outcomes. These have been summarised in Table 5.Table 5: NNT for different categorical outcomesCalculating NNT from ORs: NNT from ORs In many studies—especially case-control designs or those reporting logistic regression results—the OR is reported rather than the absolute risks. However, clinicians often want to understand the absolute impact in terms of the NNT. To estimate NNT from OR, you also need to know the baseline risk (CER)4: NNT=1CER×1−OR1−CER+(CER×OR) Suppose a study reports an OR of 0.5 for an intervention and the CER is 20% (0.20): Calculate the experimental (treated) event rate (EER) from OR and CER:EER=OR×CER1−CER+(OR×CER)EER=0.5×0.21−0.2+(0.5×0.2)=0.10.8+0.1=0.10.9≈0.111Calculate ARR:ARR=CER−EER=0.20−0.111=0.089Calculate NNT:NNT=1ARR=10.089≈11.2 Thus, approximately 11 patients need to be treated to prevent one additional event. Calculating NNT for time-to-event outcomes To calculate the NNT for time-to-event outcomes, such as those analysed with Kaplan-Meier survival curves, one must compare survival probabilities at a specific time point between treatment and control groups. This approach recognises that the absolute benefit of a therapy is best understood not only by relative measures such as hazard ratios but also by quantifying how many patients actually benefit over a clinically relevant period. The method is straightforward: At the time point of interest (e.g., 1 year), determine the survival probability in the control group, Scontrol (t), and in the treatment group, Streatment (t), as estimated from the Kaplan-Meier curves. The NNT at that time, NNT(t), is calculated as the reciprocal of the absolute difference between these survival probabilities5,6: NNT(t)=1Scontrol(t)−Streatment(t) In our simulated example, suppose the 1-year survival was 89% in the treatment group and 76% in the control group. The absolute survival benefit at one year is, therefore, 13%. The NNT at 1 year is simply the reciprocal of this difference: NNT(1 yr)=10.89−0.76=7.7. This result tells us that, on average, treating eight patients leads to one additional survivor at 1 year compared to standard care. This method enables clinicians and researchers to convey the impact of an intervention in intuitive, patient-centred terms, making the abstract results of time-to-event analyses more tangible for informed decision-making. Practical nuances in interpreting the NNT When interpreting NNT, several practical considerations must be taken into account. First, the confidence interval (CI) around the NNT conveys essential information about precision; if the interval includes infinity, the result is not statistically significant, and wide intervals indicate imprecise estimates. The CI for NNT can be calculated from reciprocals of the values defining the CI for ARR. Second, baseline risk matters profoundly: a treatment with a consistent relative effect will have a lower NNT in high-risk populations and a higher NNT in low-risk groups. Third, always specify the time frame for NNT—such as “NNT = 4 to prevent one case of PONV within 24 hours,” or “NNT = 10 to prevent one death during ICU stay.” For adverse outcomes, the analogous measure is the Number Needed to Harm (NNH), calculated as the inverse of the absolute risk increase (NNH = 1/ARI). Here, ARI is the absolute risk increase. There are also common pitfalls to avoid. Misinterpreting direction can lead to errors—NNT is calculated as 1/(CER − EER) for beneficial outcomes and NNH as 1/(EER − CER) for harms. Ignoring baseline characteristics is risky; NNT calculations assume comparable groups, and confounding can distort results, especially in observational studies. Never pool NNTs from different studies directly; instead, pool absolute risk reductions and then compute NNT. Study design also shapes which measures are valid: cohort studies and RCTs allow for both RR and OR (with RR preferred for interpretation), whereas case-control studies yield only OR, which approximates RR when the outcome is rare. Finally, beware of confusing risk with odds—“the odds of survival are 80%” is incorrect; it should be “the risk of survival is 80%.” In advanced applications, especially in critical care, where multiple outcomes may be relevant, the concept of multi-outcome NNT becomes useful. For composite endpoints—such as death or major complications—separate 2 × 2 tables should be constructed for each component and for the composite outcome, ensuring clarity and precision in reporting clinical benefits and risks. CONCLUSION The NNT is an intuitive measure of treatment effectiveness that directly informs clinical decision-making. By understanding how to construct 2 × 2 tables, calculate event rates, and interpret absolute versus relative measures, anaesthesia and critical care practitioners can better evaluate research findings and communicate treatment benefits to patients and colleagues. The distinction between risk and odds, and between RR and OR, and understanding when and why OR is used (particularly in case-control studies and logistic regression) is essential for critical appraisal of research. In addition, NNT should be interpreted in conjunction with other clinical factors, including patient preferences, cost-effectiveness, and potential risks of harm. The integration of absolute measures (NNT and ARR) with relative measures (RR and OR) provides a comprehensive understanding of treatment effects, enhancing clinical decision-making. As we continue to embrace evidence-based practice, mastering these fundamental statistical tools will enhance our ability to provide optimal patient care based on rigorous scientific evidence while effectively communicating both the magnitude and clinical significance of treatment effects. Presentation at conferences/CMEs and abstract publication Nil. Study data availability De-identified data may be requested with reasonable justification from the authors (email to the corresponding author) and shall be shared. Disclosure of use of artificial intelligence (AI) generative tools The AI tools or language models (LLM) have not been utilised in the manuscript, except that software has been used for grammar corrections and references. Declaration of use of permitted tools The tools used are not copyrighted, and all tables have been made by the authors. Contribution details VG: literature search, first draft of manuscript. NS: revision of manuscript. RG: concept, revision of the manuscript for improving intellectual content. Financial support and sponsorship Nil. Conflicts of interest Dr. Rakesh Garg, who is one of the co-authors of this manuscript, is an Editor of this journal. He was not involved in any decision-making process, and an independent editor handled this manuscript. Other authors declare no conflict of interest. Supplementary material None.