Biostatistics & Population Health

Number needed to treat and number needed to harm

Clinical Overview and When to Suspect Misuse of NNT/NNH

— Formula: NNT = 1 / ARR, where ARR = absolute risk reduction = control event rate (CER) − experimental event rate (EER).

— Always round up to the nearest whole patient (you cannot treat a fraction of a person to get a benefit).

— A trial or vignette gives you percentages or event counts in two arms and asks how many patients you must treat to prevent one event.

— A question contrasts a drug with a relative risk reduction (RRR) that sounds impressive (e.g., "30% reduction") but the baseline risk is tiny — NNT exposes that the absolute benefit is small.

— Shared decision-making vignettes where the patient asks, "How likely is this to actually help me?"

Number needed to treat (NNT) is the number of patients who must receive an intervention, instead of the comparator, for one additional patient to benefit over a defined time horizon.

Number needed to harm (NNH) is the analogous figure for adverse events: NNH = 1 / ARI, where ARI = absolute risk increase = EER − CER for the harm outcome.

When to "suspect" you need NNT/NNH on the exam:

Step 3 framing: NNT/NNH translate population-level trial data into individual counseling, the core of value-based and patient-centered care.

Board pearl: If a question gives you only RRR, you cannot compute NNT without the baseline (control) event rate. The single most common trap is being handed RRR + nothing else and being asked for NNT — the answer choice will be "cannot be calculated."

Key reflex: see two event rates → compute ARR → invert → that is your NNT. See an adverse event rate difference → invert → that is your NNH.

A low NNT (e.g., 5) = highly effective therapy; a high NNT (e.g., 200) = marginal benefit, weigh against cost and NNH.

Presentation Patterns and Key History

— "A new trial shows..." — followed by event rates in two arms; the question asks NNT.

— "Your patient read about a drug..." — patient counseling using absolute, not relative, terms.

— "Compared to placebo, drug X reduced MI by 25%..." — testing whether you notice this is RRR, not ARR.

— "Of 1,000 patients treated, how many extra strokes..." — reverse direction, asking you to multiply by 1/NNT or use ARR × N.

— Control event rate (CER) — risk in the untreated/placebo group.

— Experimental event rate (EER) — risk in the treated group.

— Time horizon — NNT is meaningless without a duration ("NNT of 50 over 5 years" is very different from "NNT of 50 over 6 months").

— Outcome definition — composite vs hard endpoint changes interpretation.

— Therapies with narrow therapeutic index (warfarin, chemo, immunosuppressants).

— Primary prevention in low-risk patients — baseline event rate is small, so NNT balloons while NNH stays fixed.

— Elderly or frail patients where harms accrue faster than benefits realize.

Typical Step 3 stems present NNT/NNH inside a journal-club, clinic-counseling, or quality-improvement scenario rather than a sick patient.

Recognizable stem flavors:

Key data points to extract from any vignette:

History cues that NNH dominates the discussion:

Key distinction: RRR is unitless and scale-free (it hides baseline risk), while ARR/NNT/NNH are anchored to absolute risk and therefore reflect what the individual patient actually experiences. Boards test whether you preferentially use ARR-based metrics for counseling, and RRR for describing biologic effect size.

Document in the chart: baseline risk, NNT, NNH, time frame, and that the patient understood absolute (not relative) numbers — this is also the medico-legal expectation for informed consent.

Physical Exam Findings — Reading the 2x2 Table

— Rows = exposure/treatment (Treated vs Control).

— Columns = outcome (Event vs No event).

— Cells: a = treated with event; b = treated without; c = control with event; d = control without.

— EER = a / (a+b)

— CER = c / (c+d)

— ARR = CER − EER (for a beneficial outcome reduction)

— RRR = ARR / CER = 1 − (EER/CER)

— NNT = 1 / ARR (round up)

— Relative risk (RR) = EER / CER

— If EER < CER → treatment helps → report NNT.

— If EER > CER → treatment harms → report NNH.

— If EER ≈ CER → ARR ≈ 0 → NNT → ∞ → no clinically meaningful effect regardless of p-value.

The "exam" of NNT/NNH is the 2×2 contingency table. Master its layout:

Derived metrics:

For harms, swap "event" to "adverse event"; the algebra is identical but you compute ARI and NNH.

Quick "hemodynamic" read of a table:

Board pearl: A statistically significant result with a huge NNT (e.g., 800) can still be clinically trivial — large trials detect tiny absolute effects. Conversely, a small NNT with wide CI crossing infinity may be statistically unstable. Always inspect the 95% CI around NNT; if it crosses infinity (i.e., the CI of ARR crosses zero), the NNT is not statistically significant.

Watch for per-protocol vs intention-to-treat denominators — ITT is the boards' preferred analysis and gives more conservative (larger) NNT estimates.

Step 3 management: When presented with raw counts, draw the 2×2 first, compute CER and EER, then subtract. Do not try to invert RRR to NNT mentally without baseline risk — it is the #1 source of arithmetic errors.

Diagnostic Workup — Computing NNT and NNH from Trial Data

— 1. Identify CER (control event rate, as decimal).

— 2. Identify EER (experimental event rate, as decimal).

— 3. ARR = CER − EER.

— 4. NNT = 1 / ARR, then round up.

— CER = 0.08, EER = 0.06, ARR = 0.02, NNT = 1/0.02 = 50 over 5 years.

— RRR = 0.02/0.08 = 25% — sounds impressive but the absolute benefit is only 2 per 100.

— ARI = 0.008, NNH = 1/0.008 = 125 over 5 years.

— Comparison: NNT 50 vs NNH 125 → benefit outweighs harm at the population level, but the patient must value MI prevention more than the myopathy risk.

— Percent → decimal (8% = 0.08) before subtracting.

— RRR known + baseline risk known → EER = CER × (1 − RRR) → then standard formula.

— Odds ratio (OR) is not directly convertible to NNT without baseline risk; OR approximates RR only when outcomes are rare (<10%).

Step-by-step computation algorithm (use this exact order on the exam):

Worked example: Statin trial, 5-year MI rate 8% placebo vs 6% statin.

Worked NNH example: same trial, myopathy in 1% statin vs 0.2% placebo.

Conversions you must know cold:

CCS pearl: When ordering a "shared decision-making" intervention in a CCS case (e.g., counseling about anticoagulation in AFib), think CHA₂DS₂-VASc → annual stroke risk → multiply by drug RRR → derive personalized NNT. This is the operationalized form of individualized risk-based prescribing that Step 3 rewards.

Pitfall: NNT computed from subgroup analysis of an underpowered trial is fragile; treat with skepticism unless prespecified.

Diagnostic Workup — Confidence Intervals, Time Horizons, and Survival Data

— Report as: "NNT 25 (95% CI 15 to 80)" — a tight CI signals a reliable estimate.

— If the ARR CI crosses zero: NNT is not statistically significant; do not quote a point estimate as definitive.

— Comparing NNTs across trials requires matching durations — beware boards stems that compare a 6-month trial to a 5-year trial.

— NNT can be derived from Kaplan-Meier curves: 1 / (KM_control − KM_treated) at a chosen time point.

— Hazard ratios (HR) alone, like RRR, cannot yield NNT without baseline absolute risk.

— LHH > 1 favors treatment (more likely to help than harm).

— LHH < 1 favors withholding.

— Example: NNT 50, NNH 125 → LHH = 2.5 → patient 2.5× more likely to be helped than harmed.

95% CI for NNT = 1 / (95% CI bounds of ARR). Because ARR can include zero (no effect), the NNT CI can include infinity, and even cross into "number needed to harm" territory if the lower bound is negative.

Time horizon is mandatory. "NNT = 50" over 1 year ≠ NNT = 50 over 10 years.

Survival/time-to-event data:

Composite endpoints inflate event rates, lowering NNT artificially — scrutinize whether the components are of comparable patient importance (death + hospitalization + lab abnormality is a classic offender).

Likelihood of being helped vs harmed (LHH) = NNH / NNT.

Board pearl: A trial reporting only HR or RRR with a small p-value but no absolute numbers is a red flag for spin — Step 3 expects you to demand ARR/NNT before recommending therapy.

Key distinction: Statistical significance (p < 0.05) addresses whether an effect exists; NNT/NNH address whether it matters. A drug can be highly significant yet clinically useless if NNT = 1,000.

Risk Stratification — Interpreting NNT/NNH at the Bedside

— NNT 1–10: highly effective (e.g., PPI for duodenal ulcer healing, NNT ~2; antibiotics for strep pharyngitis to prevent rheumatic fever in endemic areas, low NNT).

— NNT 10–50: clinically valuable (e.g., statins for secondary prevention over 5 years).

— NNT 50–200: marginal benefit; weigh harms, cost, patient preference (statins for primary prevention in low-risk adults).

— NNT > 200: usually only justified for catastrophic, irreversible outcomes (e.g., screening for rare lethal cancer).

— Same drug, same RRR → higher-risk patient = lower NNT = more benefit.

— Example: aspirin for primary prevention in low-risk 40-year-old (NNT in thousands) vs secondary prevention post-MI (NNT ~50). RRR is similar; baseline risk differs by orders of magnitude.

— NNH for serious bleed on anticoagulation is roughly constant across populations.

— NNT for stroke prevention with anticoagulation falls as CHA₂DS₂-VASc rises.

— Therefore the decision threshold is a CHA₂DS₂-VASc at which NNT < NNH (LHH > 1) — typically ≥2.

NNT magnitude is context-dependent; there is no universal "good" NNT, but rough benchmarks help:

Baseline risk drives NNT more than drug potency:

This is why guidelines tie therapy initiation to risk calculators (ASCVD pooled cohort, CHA₂DS₂-VASc, FRAX) — they identify patients where NNT becomes favorable.

Step 3 management: When the stem asks "should this patient be on a statin," compute (or recognize) 10-year ASCVD risk, then mentally apply the ~25% RRR → if absolute risk reduction is ≥1% over 10 years (NNT ≤100), guidelines support therapy.

Harm asymmetry:

Always discuss NNT and NNH together in counseling; quoting only NNT misleads the patient.

Pharmacotherapy — Applying NNT/NNH to Common Drug Decisions

— Aspirin, secondary prevention post-MI (~2 yrs): NNT ~50 for vascular events; NNH for major GI bleed ~250 → LHH ~5, strongly favors treatment.

— Aspirin, primary prevention, low-risk adult: NNT often >1,000; NNH for major bleed ~250 → LHH <1, net harm — basis for 2022 USPSTF recommendation against routine primary-prevention aspirin in most adults.

— Statins, secondary prevention (5 yrs): NNT ~30 for major vascular event; NNH for myopathy small, diabetes ~250.

— Warfarin/DOAC in AFib, CHA₂DS₂-VASc ≥2: NNT ~25–50/yr for stroke; NNH for major bleed ~50–100 → LHH typically favorable.

— Antihypertensives in stage 1 HTN, low CV risk: NNT for CV event over 5 yrs ~100+; modest absolute benefit drives shared decision-making.

— PPI for NSAID gastroprotection in high-risk patients: NNT ~10 for serious GI event.

— Identify the patient's baseline risk class (primary vs secondary prevention; risk-stratified).

— Match to expected NNT.

— Compare to NNH for the relevant adverse event (bleeding, falls, drug interaction).

— Recommend therapy only when LHH > 1 and the patient values the prevented outcome more than the harm.

High-yield NNT/NNH pairings the boards revisit (approximate, illustrative):

How to use these in vignettes:

Board pearl: USPSTF "Grade A/B" recommendations generally reflect favorable NNT/NNH balances in defined populations; "Grade D" recommendations (e.g., PSA in men >70, beta-carotene supplementation) reflect NNH ≥ NNT — memorize the population definitions.

Document the conversation: patient was informed of absolute benefit (1 in X) and absolute harm (1 in Y) — this is the gold-standard chart note for prevention prescribing.

Procedures and Screening — NNT/NNH for Interventions and Tests

— Mammography (women 50–74, biennial): NNS ~700–1,300 over 10 yrs to prevent one breast cancer death; NNH includes false positives (1 in 2 over 10 yrs of screening), unnecessary biopsies, and overdiagnosis (~1 in 5 cancers detected).

— Colonoscopy q10y starting 45: NNS ~150–300 to prevent one CRC death over a lifetime; NNH for perforation ~1,000.

— Low-dose CT lung cancer screening (high-risk smokers): NNS ~320 to prevent one lung cancer death; high false-positive rate, NNH for invasive workup notable.

— AAA screening, men 65–75 ever-smokers: NNS ~500 to prevent one AAA-related death.

— PCI for STEMI (vs fibrinolysis): NNT ~50 for death; NNT ~20 for reinfarction at 30 days.

— CABG vs medical therapy in left main disease: NNT for survival ~5–10 over 5 yrs in high-risk anatomy.

— Bariatric surgery for severe obesity with comorbidity: NNT for diabetes remission low (~3–5).

Screening test NNT framing — for screening, NNT becomes "number needed to screen" (NNS) to prevent one death:

Procedural NNTs:

Overdiagnosis trap: low NNS for "case detection" ≠ low NNS to prevent death or meaningful morbidity. Boards reward distinguishing disease-specific mortality endpoints from intermediate ones.

Key distinction: Lead-time bias and length-time bias can inflate apparent screening benefit without changing NNT for true outcomes — always tether NNS to a hard endpoint, ideally all-cause mortality (the most bias-resistant).

CCS pearl: When ordering screening in a CCS ambulatory case, the order set should reflect USPSTF Grade A/B (favorable NNT/NNH) interventions; avoid Grade D (PSA in elderly, hormone replacement for prevention) — these increment harm without commensurate benefit and can be scored against you.

Special Populations — Elderly and Renal/Hepatic Impairment

— Competing mortality truncates the time horizon over which benefits accrue. A statin for primary prevention with NNT 100 over 10 years is useless in a patient with <10-year life expectancy from comorbidity.

— Higher baseline risk of the target outcome (CV, stroke) lowers NNT — favors treatment.

— Higher baseline risk of harm (bleeding, falls, polypharmacy interactions) lowers NNH — favors withholding.

— Anticoagulation in elderly AFib: NNT for stroke drops sharply with age (favorable), but NNH for intracranial bleed also drops (unfavorable). Most data still favor anticoagulation up to advanced age unless HAS-BLED is very high or falls are frequent and severe.

— Statins in primary prevention >75: NNT rises (less time to benefit); guidelines call for shared decision-making, no automatic prescribing.

— Tight glycemic control in elderly: NNT for microvascular benefit requires ~8–10 years to manifest; NNH for hypoglycemia is immediate — relax A1c targets to 7.5–8% in frail elderly.

— Renal: many drugs accumulate (DOACs, metformin, NSAIDs) → NNH falls (more harm per patient treated) without changing NNT.

— Hepatic: warfarin, statins, acetaminophen → similar pattern.

Elderly patients shift the NNT/NNH balance in three ways:

Net effect varies by condition; the boards' expectation is individualized:

Renal/hepatic impairment alters NNH disproportionately:

Step 3 management: For any geriatric prescribing decision, estimate life expectancy first (e.g., using ePrognosis or comorbidity-based judgment); if the time-to-benefit of an intervention exceeds expected survival, NNT is effectively infinite for that patient — deprescribe.

Deprescribing as a positive action: removing a drug whose NNH now exceeds NNT is good medicine, not therapeutic nihilism. Document the rationale.

Special Populations — Pregnancy, Pediatrics, and Underrepresented Groups

— Most RCTs exclude pregnant patients, so NNT/NNH estimates are extrapolated from observational data — wider uncertainty.

— Examples with reasonable NNT data: low-dose aspirin for preeclampsia prevention in high-risk patients, NNT ~50 to prevent one case; minimal NNH.

— Antihypertensives in chronic HTN in pregnancy (CHAP trial): NNT ~14 for adverse pregnancy outcome composite at BP threshold 140/90.

— Baseline event rates for many adult-targeted conditions are tiny → NNT enormous → most prevention strategies untenable.

— Vaccination NNT can be highly favorable in high-incidence diseases (measles, pertussis) and less so in low-incidence ones, but NNH is typically very small (rare serious AEs), keeping LHH very high.

— Subgroup NNT can differ when baseline risk differs (e.g., ASCVD risk higher in Black adults at given risk-factor profile → lower NNT for statins).

— Beware risk calculators not validated in the patient's group; transported NNT estimates may mislead.

— Some agents (clopidogrel in CYP2C19 poor metabolizers, warfarin VKORC1/CYP2C9 variants) have population-average NNT that mask wide individual variation — emerging precision-medicine framing tests on Step 3.

Pregnancy:

Pediatrics:

Race, ethnicity, sex:

Pharmacogenomics:

Board pearl: When trial enrollment underrepresents women, elderly, or non-White participants, applying the pooled NNT to those groups is an external-validity (generalizability) limitation — always flag this in shared decision-making.

Key distinction: Efficacy NNT (from RCTs, ideal conditions) is usually smaller than effectiveness NNT (real-world adherence, comorbidity, monitoring gaps) — counsel patients with the larger number.

Complications and Adverse Outcomes of Misusing NNT/NNH

— Overtreatment: quoting RRR ("cuts your risk by 30%") without ARR exaggerates benefit, leading patients to accept therapies whose true NNT is hundreds.

— Undertreatment: focusing solely on NNH (e.g., "1 in 100 bleed") without contextual NNT may scare patients away from net-beneficial therapy.

— Therapeutic inertia / nihilism: clinicians who fixate on large NNTs may withhold guideline-recommended therapy in high-risk patients.

— Mass deployment of interventions with marginal NNT consumes resources without proportional benefit — relevant in value-based care, accountable care organizations, and formulary decisions.

— Cascade harms: a screening test with favorable NNS for detection but unfavorable NNH for downstream invasive workup (false-positive biopsies, anxiety) — classic in PSA, low-yield imaging.

— Framing effect: relative risk framing increases enthusiasm; absolute framing dampens it — boards test recognition of framing manipulation.

— Base-rate neglect: ignoring baseline risk and reasoning solely from RRR.

— Affect heuristic: dread of a feared outcome (cancer) overrides numeric reasoning about NNS.

Clinical complications from miscommunicated risk:

Population-level harms:

Cognitive biases that amplify these errors:

Step 3 management: In counseling, lead with absolute numbers ("Of 100 people like you, X benefit and Y are harmed over Z years") and show pictograms when available — these reduce framing bias and increase comprehension, especially in low-numeracy patients.

Quality-improvement angle: institutional pay-for-performance metrics that reward intervention rates (e.g., statin prescription rates) without risk stratification can drive treatment of patients with unfavorable individual NNT — a known unintended consequence.

When to Escalate — Demanding Better Data Before Acting

— Only surrogate endpoints are reported (LDL, A1c, BP) with no patient-oriented evidence that matters (POEM) such as MI, stroke, death.

— Composite endpoints are driven entirely by the softest component (hospitalization, revascularization) rather than the hardest (death).

— Wide CI around ARR crossing zero — NNT is not stable.

— Industry-sponsored trial with selective reporting, no prespecified analysis plan, or unpublished outcomes.

— Per-protocol analysis only, without ITT.

— Short follow-up that cannot capture long-term harms (NNH undercounted).

— Multiple trials show conflicting NNT estimates → meta-analysis NNT preferred.

— Newly approved drug with accelerated FDA approval on surrogate endpoints — wait for confirmatory outcome data.

— Off-label use where NNT/NNH in the target indication is unknown.

— NNT and NNH are close (LHH near 1).

— Outcomes prevented and outcomes caused differ in patient-perceived severity (e.g., preventing stroke vs causing GI bleed).

Triage decisions about a trial's NNT/NNH should escalate to "do not adopt yet" when:

Trigger a consultation with clinical pharmacy, evidence-based medicine resources, or specialty guidelines when:

At the patient level, escalate counseling time / shared decision-making aids when:

CCS pearl: In a CCS scenario, ordering a "shared decision-making consult" or providing patient education materials is a creditable action when the NNT/NNH tradeoff is genuinely close — the test rewards explicit acknowledgment of preference-sensitive care.

Board pearl: "Statistically significant" is necessary but not sufficient; the higher bar is clinically significant, which is what NNT/NNH operationalize. Demand both.

Key Differentials — Related Effect-Size Metrics

— Absolute risk reduction (ARR): the difference itself, in percentage points. NNT = 1/ARR.

— Absolute risk increase (ARI): same algebra, for harms. NNH = 1/ARI.

— Attributable risk (AR) in epidemiology: equivalent to ARR conceptually, applied to exposures.

— Population attributable risk (PAR): ARR × prevalence of exposure → public-health framing.

— Relative risk (RR) = EER/CER.

— Relative risk reduction (RRR) = 1 − RR.

— Hazard ratio (HR): time-to-event analog of RR.

— Odds ratio (OR): ratio of odds; approximates RR only when event rate <10%.

Other absolute-effect metrics often confused with NNT/NNH:

Relative metrics that cannot become NNT without baseline risk:

Number needed to screen (NNS): NNT analog for screening tests, anchored to mortality or hard morbidity over a defined horizon.

Likelihood of being helped vs harmed (LHH) = NNH/NNT: a single ratio capturing the benefit-harm balance.

Minimal clinically important difference (MCID): smallest change in an outcome perceived as meaningful by the patient — orthogonal to NNT (MCID asks "is it big enough?", NNT asks "how many to achieve it?").

Key distinction: RRR is constant across baseline risks; ARR (and thus NNT) varies with baseline risk. This is why a single drug has many NNTs depending on the population.

Common boards trap: question gives OR from a case-control study and asks for NNT. Answer: NNT cannot be derived from case-control data because baseline risk is not measurable in that design — case-control yields ORs only.

Key Differentials — Other Study Design and Bias Issues

— Selection bias: trial enrolls lower-risk patients than your clinic population → real-world NNT larger than published.

— Attrition / loss to follow-up: differential dropout inflates or deflates apparent ARR; ITT analysis mitigates partially.

— Detection bias / unblinded outcome assessment: especially affects soft endpoints (pain scores, hospitalization).

— Publication bias: negative trials unpublished → meta-analyzed NNT smaller (more favorable) than truth; address with funnel plots, trial registries.

— Non-inferiority trial: NNT framing irrelevant; the goal is bounded equivalence, not superiority.

— Cluster randomization: effective sample size smaller; CIs wider; NNT estimates less precise.

— Surrogate endpoint trial: NNT for surrogate ≠ NNT for patient-important outcome.

— Confounding by indication makes "treated" patients sicker; raw event-rate comparison underestimates benefit. Use propensity scores or adjusted estimates before computing pseudo-NNT.

— Immortal-time bias in retrospective cohorts inflates treatment benefit.

Sources of error that distort apparent NNT/NNH:

Trial design features that change NNT interpretation:

Confounders unique to observational studies:

Board pearl: A meta-analysis that pools heterogeneous trials may produce a pooled NNT that applies to no single real population; check I² statistic and clinical heterogeneity before quoting the pooled number.

Key distinction: RCT-derived NNT addresses efficacy under ideal conditions; pragmatic-trial or registry NNT addresses effectiveness in usual care — the latter is more transportable to your patient but rarely available.

When the boards present an "encouraging new study," check design > size > effect size > NNT > harms — in that order.

Secondary Prevention and Long-Term NNT-Based Prescribing

— Post-MI: aspirin, P2Y12 inhibitor, high-intensity statin, beta-blocker, ACEi/ARB — composite NNT for each in single digits to low tens.

— Post-stroke (non-cardioembolic): antiplatelet + statin + BP control → NNT ~30 over 5 yrs for recurrent stroke.

— Post-AFib stroke: anticoagulation NNT ~12 over 1 yr for recurrent stroke in many cohorts.

— Post-HFrEF hospitalization: GDMT (ARNI/ACEi, beta-blocker, MRA, SGLT2i) — each component NNT 20–30 over 1–2 yrs for death or rehospitalization; stacked effect is even more compelling.

— Absolute benefit ("of 100 patients like you, X will avoid another heart attack over 2 years").

— Absolute harm ("Y will have a serious bleed").

— The time-to-benefit (some drugs take months to manifest survival benefit).

— That stopping the drug returns the patient to baseline risk rapidly for some agents (e.g., antiplatelets).

Secondary prevention almost always has more favorable NNT than primary prevention because baseline risk is higher.

Discharge medications after major events — typical NNT for hard outcomes over 1–2 years:

Discharge counseling should explicitly state:

Step 3 management: Build the discharge order set around therapies with NNT < NNH for the relevant time horizon; deprescribe agents that no longer have a favorable balance (e.g., aspirin for primary prevention in a patient now on anticoagulation — net harm).

Adherence reality check: real-world adherence ~50% at 1 year → effectiveness NNT is roughly double efficacy NNT. Address this with simplified regimens, combination pills, and follow-up cadence.

Follow-Up, Monitoring, and Counseling

— Short-NNH drugs (anticoagulants, immunosuppressants): early follow-up within 2–4 weeks for adverse effects, then risk-tiered.

— Long-time-to-benefit drugs (statins, antihypertensives): emphasize adherence at 4–12 weeks; reassess risk and absolute benefit annually.

— Statins: liver enzymes only if symptoms; CK only if myalgia — routine surveillance does not lower NNH meaningfully.

— Anticoagulants: renal function, CBC, signs of bleeding; reassess fall risk and concomitant antiplatelets at each visit.

— Antihypertensives: BP, K⁺, Cr at 2–4 weeks after initiation/titration.

— Major clinical event changes baseline risk (new MI, stroke, AFib).

— New comorbidity (renal failure, dementia, malignancy with limited prognosis).

— Aging into a new risk stratum or out of a beneficial life-expectancy window.

— New trial data alter the assumed effect size.

— Pictograms (icon arrays of 100 figures) reliably outperform numeric framing for low-numeracy patients.

— Decision aids validated for statin therapy, anticoagulation in AFib, PSA screening, mammography in 40s.

— Teach-back: have the patient restate the absolute benefit and harm in their own words.

Follow-up cadence after starting an intervention should match the time horizon of NNT/NNH:

Monitoring parameters tied to NNH reduction:

Reassessment triggers — recompute personalized NNT/NNH when:

Counseling tools:

CCS pearl: Document in the CCS note: "Discussed absolute benefit (NNT X over Y years) and absolute harm (NNH Z), patient understood and elected to proceed/defer." This is the canonical structured note for prevention decisions.

Build long-term plans around periodic reassessment, not "set and forget" prescribing.

Ethical, Legal, and Patient Safety Considerations

— Concrete Step 3 example: a clinician prescribes a primary-prevention statin to a low-risk 45-year-old by quoting "25% risk reduction" without ARR. Patient develops myopathy. The malpractice/ethics critique is that the patient was not informed the absolute benefit was <1% over 10 years (NNT >100) while the absolute harm was comparable — leading to a consent that the patient would likely not have given if numerate.

— At hospital discharge, medication reconciliation must include reassessing NNT/NNH — patients are commonly continued on hospital-initiated drugs (PPIs, BP meds, anticoagulants) whose ambulatory NNT/NNH balance is unfavorable.

— Failure to deprescribe is a patient safety event in older adults (Beers criteria overlap).

— Risk calculators (and therefore NNTs) calibrated on majority populations may misestimate risk in minorities, leading to under- or over-treatment.

— Acknowledge this limitation in chart documentation.

— Vaccine NNT/NNH discussions are subject to truthful risk communication; misrepresenting harm (either inflating or minimizing) can violate public health duties and erode trust.

Informed consent legally requires disclosure of material risks and benefits in terms the patient can understand. Quoting only RRR ("reduces your risk by 30%") without absolute numbers is increasingly viewed as inadequate consent in elective and preventive interventions, because it systematically biases patients toward accepting therapy.

Transitions of care:

Equity and justice:

Mandatory reporting / public-health framing:

Board pearl: The Step 3 ethically defensible counseling script names the time horizon, absolute benefit, absolute harm, and alternatives including no treatment — all four are required for valid consent in preventive care.

Quality and safety: institutional dashboards that reward prescribing rates without risk stratification can incentivize patient-level NNT-unfavorable care — escalate as a QI concern.

High-Yield Associations and Rapid-Fire Clinical Facts

NNT = 1 / ARR; NNH = 1 / ARI; always round up; always state the time horizon.

Cannot compute NNT from RRR alone, OR alone, HR alone, or case-control data — baseline risk is mandatory.

NNT varies inversely with baseline risk; same drug, different NNTs by population.

LHH = NNH / NNT; >1 favors treatment, <1 favors withholding.

Wide CI around ARR that crosses zero → NNT not statistically significant.

Composite endpoints can artificially lower NNT — scrutinize component severity.

Surrogate endpoints generate NNTs that may not translate to patient-oriented outcomes.

Primary prevention with aspirin in low-risk adults: NNT > NNH → USPSTF discourages.

Secondary prevention (post-MI, post-stroke, post-AFib stroke): NNT typically << NNH → strongly recommended.

Number needed to screen (NNS) for mammography ~700–1,300; for colonoscopy ~150–300; for LDCT lung ~320; all-cause mortality endpoints preferred.

Time-to-benefit matters in elderly: if life expectancy < time-to-benefit, intervention NNT is effectively infinite — deprescribe.

Framing effect: relative-risk framing inflates perceived benefit; absolute-risk framing is ethically preferred for consent.

Pictograms (icon arrays) improve risk communication, especially in low-numeracy patients.

Per-protocol analysis flatters NNT; ITT is the boards' preferred conservative estimate.

Pooled NNT from meta-analysis may not apply to any single population — check heterogeneity (I²).

Deprescribing when NNH > NNT is good medicine, not nihilism.

Board pearl: When forced to choose among ARR, RRR, RR, OR, and NNT for counseling, pick NNT (or ARR); for describing biologic effect size in a paper, RRR/HR is acceptable.

Board Question Stem Patterns

— "In a trial, 6% of placebo and 4% of drug X patients had MI at 5 years. What is the NNT?" → ARR = 0.02 → NNT = 50.

— "Drug Y reduces stroke by 40%. What is the NNT?" → Answer: cannot be determined; need baseline risk.

— "Major bleeding occurred in 3% of anticoagulated and 1% of control patients. NNH?" → ARI = 0.02 → NNH = 50.

— Given NNT 25, NNH 100 → LHH = 4 → treatment favored; counsel and proceed.

— "Drug Z has a 30% RRR. NNT in high-risk patients was 20 (baseline 15%). What is NNT in low-risk (baseline 3%) patients?" → New ARR = 0.30 × 0.03 = 0.009 → NNT ≈ 112.

— Patient asks about benefit; correct response uses absolute terms and pictograms, not RRR.

— Given mortality reduction and screening cohort size, derive NNS; recognize lead-time and overdiagnosis caveats.

— Trial reports p < 0.001 but ARR 0.1%; the right takeaway is NNT = 1,000, likely not clinically meaningful.

— 88-year-old with limited life expectancy on primary-prevention statin; recognize NNT > life expectancy and discontinue.

— Asks NNT from OR; correct answer notes baseline risk unobtainable.

Pattern 1 — Compute NNT from raw rates:

Pattern 2 — RRR alone trap:

Pattern 3 — NNH calculation:

Pattern 4 — LHH judgment:

Pattern 5 — Population-shift question:

Pattern 6 — Counseling/communication:

Pattern 7 — Screening NNS:

Pattern 8 — Statistical significance vs clinical significance:

Pattern 9 — Deprescribing in elderly:

Pattern 10 — Case-control trap:

Step 3 management: Reflexively (1) extract event rates, (2) compute ARR, (3) invert for NNT, (4) compare to NNH, (5) frame in absolute terms for the patient. This algorithm answers ~90% of NNT/NNH stems.

One-Line Recap

High-yield recap bullets:

NNT (1/ARR) and NNH (1/ARI) translate trial data into the absolute, patient-anchored numbers that drive ethical informed consent, guideline-based prescribing, and shared decision-making — and they cannot be derived from RRR, HR, OR, or case-control data alone.

Formulas: NNT = 1/ARR, NNH = 1/ARI, LHH = NNH/NNT; always round NNT/NNH up; always quote a time horizon.

Baseline risk is the driver: the same drug yields very different NNTs in primary vs secondary prevention; high-risk patients have lower (more favorable) NNT. Risk calculators (ASCVD, CHA₂DS₂-VASc, FRAX) exist to identify patients whose NNT crosses the treatment threshold.

Communication standard: counsel with absolute numbers, pictograms, time-to-benefit, NNT, and NNH together — relative-risk-only counseling biases consent and is increasingly seen as inadequate.

Deprescribe when time-to-benefit exceeds life expectancy, when NNH now outweighs NNT, or when a competing therapy has rendered the original prescription net-harmful (e.g., aspirin in a patient now anticoagulated).

Board reflexes: RRR + no baseline → cannot compute; statistically significant + huge NNT → clinically trivial; favorable NNS for detection ≠ favorable NNS for mortality; case-control studies cannot yield NNT.

CCS pearl: Document NNT, NNH, time horizon, alternatives, and patient understanding for every preventive prescribing decision — this is the structured note the exam (and real-world peer review) expects.