Number Needed to Treat & Number Needed to Harm (NNT & NNH)

01Core Definitions and Formulas

Number Needed to Treat (NNT) answers the clinical question: "How many patients do I need to treat with this intervention, over a defined time period, to prevent one additional bad outcome?" It converts the abstract concept of probability into a patient-centered number that clinicians can use at the bedside and in shared decision-making.

The foundation of NNT is the Absolute Risk Reduction (ARR), which is the arithmetic difference in event rates between the control group and the treatment group. If the control group has a higher event rate for a bad outcome and the treatment lowers it, the ARR tells you the actual reduction in risk attributable to the intervention.

$$ARR=CER-EERARR\; =\; CER\; -\; EER$$

Where $CERCER$ = Control Event Rate (proportion of patients with the outcome in the control/placebo group) and $EEREER$ = Experimental Event Rate (proportion of patients with the outcome in the treatment group).

$$NNT=\frac{1}{ARR}=\frac{1}{CER-EER}NNT\; =\; \backslash frac\{1\}\{ARR\}\; =\; \backslash frac\{1\}\{CER\; -\; EER\}$$

Number Needed to Harm (NNH) is the mirror calculation applied to adverse events. It asks: "How many patients need to receive this treatment before one additional patient experiences a harmful side effect compared to control?"

The foundation of NNH is the Absolute Risk Increase (ARI), which is the excess rate of the adverse event in the treatment arm versus control.

$$ARI=EE{R}_{harm}-CE{R}_{harm}ARI\; =\; EER\_\{harm\}\; -\; CER\_\{harm\}$$

$$NNH=\frac{1}{ARI}=\frac{1}{EE{R}_{harm}-CE{R}_{harm}}NNH\; =\; \backslash frac\{1\}\{ARI\}\; =\; \backslash frac\{1\}\{EER\_\{harm\}\; -\; CER\_\{harm\}\}$$

Critical rule: NNT and NNH are always rounded UP to the next whole number in clinical practice. You cannot treat a fraction of a patient. An NNT of 3.2 becomes 4. This is the only measure in biostatistics where rounding is always upward regardless of the decimal.

02Calculation Method

The standard approach to computing NNT and NNH follows from a 2x2 contingency table. Understanding how to extract rates from this table is the most commonly tested skill.

From this table:

$$EER=\frac{a}{a+b}EER\; =\; \backslash frac\{a\}\{a+b\}$$

$$CER=\frac{c}{c+d}CER\; =\; \backslash frac\{c\}\{c+d\}$$

$$ARR=CER-EER=\frac{c}{c+d}-\frac{a}{a+b}ARR\; =\; CER\; -\; EER\; =\; \backslash frac\{c\}\{c+d\}\; -\; \backslash frac\{a\}\{a+b\}$$

$$NNT=\frac{1}{ARR}NNT\; =\; \backslash frac\{1\}\{ARR\}$$

Worked calculation:
Suppose a trial of Drug X for stroke prevention enrolls 200 patients in each arm. In the control group, 40 out of 200 patients have a stroke. In the treatment group, 20 out of 200 patients have a stroke.

$$CER=\frac{40}{200}=0.20(20\mathrm{\%})CER\; =\; \backslash frac\{40\}\{200\}\; =\; 0.20\; \backslash quad\; (20\backslash \%)$$

$$EER=\frac{20}{200}=0.10(10\mathrm{\%})EER\; =\; \backslash frac\{20\}\{200\}\; =\; 0.10\; \backslash quad\; (10\backslash \%)$$

$$ARR=0.20-0.10=0.10(10\mathrm{\%})ARR\; =\; 0.20\; -\; 0.10\; =\; 0.10\; \backslash quad\; (10\backslash \%)$$

$$NNT=\frac{1}{0.10}=10NNT\; =\; \backslash frac\{1\}\{0.10\}\; =\; 10$$

Interpretation: You need to treat 10 patients with Drug X to prevent one additional stroke over the study period.

Now suppose that same Drug X causes liver toxicity in 15 of 200 treated patients vs. 5 of 200 in the control group:

$$ARI=\frac{15}{200}-\frac{5}{200}=0.075-0.025=0.05(5\mathrm{\%})ARI\; =\; \backslash frac\{15\}\{200\}\; -\; \backslash frac\{5\}\{200\}\; =\; 0.075\; -\; 0.025\; =\; 0.05\; \backslash quad\; (5\backslash \%)$$

$$NNH=\frac{1}{0.05}=20NNH\; =\; \backslash frac\{1\}\{0.05\}\; =\; 20$$

Interpretation: For every 20 patients treated with Drug X, one additional patient will develop liver toxicity.

03Clinical Interpretation and Decision-Making

Interpreting NNT

Interpreting NNH

Likelihood of Being Helped vs. Harmed (LHH)

When both NNT and NNH are available from the same trial, they can be combined into a single trade-off ratio called the Likelihood of Being Helped vs. Harmed (LHH).

$$LHH=\frac{NNH}{NNT}=\frac{ARR}{ARI}LHH\; =\; \backslash frac\{NNH\}\{NNT\}\; =\; \backslash frac\{ARR\}\{ARI\}$$

An LHH greater than 1 means the treatment is more likely to help than to harm. A higher LHH favors the intervention.

Using the Drug X example above:

$$LHH=\frac{20}{10}=2LHH\; =\; \backslash frac\{20\}\{10\}\; =\; 2$$

This means a patient is 2 times more likely to benefit (stroke prevention) than to be harmed (liver toxicity) by Drug X. Whether this ratio is acceptable depends on the clinical severity of each outcome, which is a judgment call that no formula alone can settle.

Two important caveats about LHH that exam writers test:

1. The benefit and harm being compared must be of similar clinical importance. An LHH of 5 is misleading if the benefit is mild symptom relief while the harm is death.

2. NNT and NNH must be measured over the same time horizon for the ratio to be valid.

04Relationship to Other Measures of Effect

Understanding how NNT relates to Relative Risk Reduction (RRR) is one of the most heavily tested concepts in biostatistics. Test writers exploit the gap between absolute and relative measures to construct misleading answer choices.

ARR vs. RRR

The relationship between all four measures:

$$ARR=CER\times RRRARR\; =\; CER\; \backslash times\; RRR$$

$$NNT=\frac{1}{CER\times RRR}NNT\; =\; \backslash frac\{1\}\{CER\; \backslash times\; RRR\}$$

This equation reveals a critical insight: NNT depends on the baseline risk (CER). The same RRR applied to a high-risk population yields a lower (better) NNT than when applied to a low-risk population.

Example: A statin reduces cardiovascular events by 25% (RRR = 0.25) in all populations.

• In a high-risk patient (CER = 20%): $NNT=\frac{1}{0.20\times 0.25}=\frac{1}{0.05}=20NNT\; =\; \backslash frac\{1\}\{0.20\; \backslash times\; 0.25\}\; =\; \backslash frac\{1\}\{0.05\}\; =\; 20$

• In a low-risk patient (CER = 2%): $NNT=\frac{1}{0.02\times 0.25}=\frac{1}{0.005}=200NNT\; =\; \backslash frac\{1\}\{0.02\; \backslash times\; 0.25\}\; =\; \backslash frac\{1\}\{0.005\}\; =\; 200$

Same drug, same RRR, but wildly different NNTs. This is why exam questions that give only the RRR and ask about treatment benefit are testing whether you can recognize that absolute measures (ARR and NNT) are needed for clinical decision-making, not relative measures alone.

05Worked Exam-Style Examples

Example A: Straightforward NNT Calculation

Vignette: A randomized controlled trial enrolls 1,000 patients with hypertension into two groups: 500 receive a new antihypertensive and 500 receive placebo. Over 5 years, 50 patients in the placebo group develop a stroke compared to 30 in the treatment group. What is the NNT?

Solution:

$$CER=\frac{50}{500}=0.10CER\; =\; \backslash frac\{50\}\{500\}\; =\; 0.10$$

$$EER=\frac{30}{500}=0.06EER\; =\; \backslash frac\{30\}\{500\}\; =\; 0.06$$

$$ARR=0.10-0.06=0.04ARR\; =\; 0.10\; -\; 0.06\; =\; 0.04$$

$$NNT=\frac{1}{0.04}=25NNT\; =\; \backslash frac\{1\}\{0.04\}\; =\; 25$$

Answer: 25 patients must be treated for 5 years to prevent one stroke.

Example B: NNH Calculation from Adverse Event Data

Vignette: In the same trial, 10 patients in the treatment group develop angioedema compared to 0 in the placebo group. What is the NNH?

Solution:

$$ARI=\frac{10}{500}-\frac{0}{500}=0.02ARI\; =\; \backslash frac\{10\}\{500\}\; -\; \backslash frac\{0\}\{500\}\; =\; 0.02$$

$$NNH=\frac{1}{0.02}=50NNH\; =\; \backslash frac\{1\}\{0.02\}\; =\; 50$$

Answer: For every 50 patients treated, one additional patient develops angioedema.

Example C: The RRR vs. ARR Trap

Vignette: A pharmaceutical company reports that their new drug reduces the risk of a rare disease by 50% (RRR). The baseline incidence of the disease is 1 in 10,000 (0.01%). Should this treatment be adopted for the general population?

Solution:

$$ARR=CER\times RRR=0.0001\times 0.50=0.00005ARR\; =\; CER\; \backslash times\; RRR\; =\; 0.0001\; \backslash times\; 0.50\; =\; 0.00005$$

$$NNT=\frac{1}{0.00005}=\mathrm{20,000}NNT\; =\; \backslash frac\{1\}\{0.00005\}\; =\; 20\{,\}000$$

Answer: You would need to treat 20,000 patients to prevent a single case. Despite a 50% RRR (which sounds impressive), the absolute benefit is negligible because the baseline risk is extremely low. This is the classic scenario where reporting only the RRR inflates the perceived treatment effect.

Example D: LHH-Based Decision

Vignette: Using Examples A and B above, calculate the LHH and determine whether the benefit outweighs the harm.

$$LHH=\frac{NNH}{NNT}=\frac{50}{25}=2LHH\; =\; \backslash frac\{NNH\}\{NNT\}\; =\; \backslash frac\{50\}\{25\}\; =\; 2$$

The patient is 2 times more likely to be helped (stroke prevented) than harmed (angioedema). Given the severity difference between stroke and angioedema, this treatment is worth considering, though both outcomes must be discussed with the patient.

06Traps and High-Yield Pearls

The single most common way students lose points on NNT/NNH questions is confusing Relative Risk Reduction with Absolute Risk Reduction. An exam stem will present a drug that "reduces mortality by 50%" and offer both $NNT=2NNT\; =\; 2$ (calculated from RRR) and $NNT=200NNT\; =\; 200$ (calculated from ARR) as answer choices. The correct NNT is always derived from the ARR. When a question provides only the RRR, you must first convert it using $ARR=CER\times RRRARR\; =\; CER\; \backslash times\; RRR$ before inverting to get NNT. If a question presents only a relative measure and no baseline risk, the NNT cannot be computed, and that itself may be the tested concept.

The second trap involves forgetting to round up. An ARR of 0.07 gives $NNT=\frac{1}{0.07}=14.28NNT\; =\; \backslash frac\{1\}\{0.07\}\; =\; 14.28$. The answer is 15, not 14. Every NNT and NNH is always rounded to the next higher integer.

The third trap involves ignoring the time horizon. An NNT of 20 over 5 years is a fundamentally different claim from an NNT of 20 over 30 days. A bare NNT without its time frame is incomplete, and an answer choice that omits the study duration should raise suspicion.

Fourth, students often forget that NNT changes with baseline risk. The same treatment will have a much lower (better) NNT in a high-risk population than in a low-risk population. This is tested when a vignette asks you to apply a treatment studied in high-risk patients to a low-risk screening population and then asks for the NNT in the new context.

Finally, regarding Indonesian competency standards: the SKDI (Standar Kompetensi Dokter Indonesia, currently the 2019 revision under KKI regulation) includes evidence-based medicine as a core competency for all medical graduates. Graduates are expected to critically appraise therapeutic studies, including the ability to calculate and interpret absolute measures of treatment effect like ARR, NNT, and NNH. There are no country-level regulatory variations in how these formulas are defined or computed; they follow universal biostatistical methodology.

The core competency being tested across all NNT/NNH questions is your ability to distinguish absolute from relative measures of treatment effect, compute NNT correctly from raw trial data, and apply it to clinical decision-making by integrating baseline risk, time horizon, and the trade-off between benefit and harm.