Biostatistics & Population Health

Communicating NNT and absolute risk to patients

Clinical Overview and When to Suspect Innumeracy in Shared Decision-Making

— Patient quotes a drug ad ("cuts risk in half") without baseline risk

— Patient demands or refuses a screening test based on a single statistic

— Family conference invoking percentages for ICU prognosis

— Low health-literacy markers: limited education, language discordance, age >65, cognitive impairment

— ARR = control event rate − treatment event rate

— RRR = ARR / control event rate

— NNT = 1 / ARR (must specify time horizon)

— NNH = 1 / absolute risk increase of harm

— "The drug reduces stroke by 40%" with no baseline

— Patient asks "What are my chances, doc?"

— Disagreement between patient and clinician about a low-yield test

Core problem: patients (and clinicians) systematically misinterpret risk when it is presented as relative risk reduction (RRR) rather than absolute risk reduction (ARR) or number needed to treat (NNT). A "50% reduction in heart attacks" sounds transformative; the underlying ARR may be 2% → 1%, an NNT of 100 over 5 years.

When to suspect a numeracy gap:

Step 3 framing: the exam tests whether you can translate trial data into a 1-minute, plain-language conversation during an outpatient visit — statin initiation, mammography, PSA, anticoagulation for AF, aspirin for primary prevention.

Key vocabulary to master:

Why it matters clinically: informed consent is legally and ethically invalid if the patient cannot understand the magnitude of benefit/harm. Misframed risk drives overtreatment, low-value care, and downstream litigation.

Red flag stems on Step 3:

Board pearl: when an answer choice offers RRR alone, it is almost always wrong; the correct framing is ARR + NNT over a defined time horizon, paired with NNH for the most relevant adverse event. Always anchor numbers to the patient's own baseline risk (e.g., ASCVD 10-yr risk) rather than a generic trial population estimate.

Presentation Patterns and Key History

— Annual wellness visit reviewing statin initiation in primary prevention

— Counseling a 55-year-old about lung cancer screening LDCT

— AF patient weighing DOAC vs no anticoagulation (CHA₂DS₂-VASc discussion)

— Postmenopausal woman considering bisphosphonate for osteopenia

— Parent asking about HPV vaccine efficacy

— Family meeting on CPR / code status with prognostic percentages

— Health literacy screen: "How confident are you filling out medical forms?" (single-item literacy screener)

— Numeracy screen: ask patient to interpret a simple risk ("If 10 of 1000 have X…")

— Preferred decision style: paternalistic, shared, or informed (patient-led)

— Values clarification: fear of stroke vs fear of bleeding; longevity vs quality

— Prior experiences: family member harmed by a drug, distrust of system

— Source of the patient's statistic: TV ad, friend, social media — identify and correct

— Need for certified medical interpreter (not family) when consenting to risk-based therapy

— Numeric concepts (percent, probability) vary cross-culturally — use frequencies ("3 out of 100") rather than decimals

Typical encounter types where NNT/ARR communication is tested:

History elements that shape framing:

Cultural and language history:

Time horizon question: always ask, "Over what period are we deciding?" A 5-year NNT for statins differs sharply from a lifetime NNT.

Step 3 management: in an outpatient stem, when a patient says "I don't want the statin — it only helps a tiny number of people," the right next step is not to override or capitulate but to elicit their understanding, share ARR/NNT with a pictograph, and check teach-back. This is shared decision-making (SDM), the testable answer over "prescribe anyway" or "respect refusal without discussion."

Board pearl: documenting teach-back ("Patient restated that ~1 in 100 will avoid a heart attack over 5 years") is the gold-standard chart entry and the favored Step 3 answer.

Physical Exam Findings (and Communication Assessment when relevant)

— Understanding: can patient restate the condition and the proposed intervention?

— Appreciation: does patient apply the information to their own situation?

— Reasoning: can patient compare options logically?

— Choice: can patient communicate a stable preference?

— "Just so I know I explained it well, can you tell me in your own words what taking this medication would mean for your heart-attack risk?"

— Re-teach gaps; do not ask "Do you understand?" (yes-bias)

— Confuses percent with percentage points ("20% is the same as 20")

— Cannot estimate which is larger: 1/10 vs 1/100

— Anchors on a single anecdote ("My uncle had a bleed")

— Magnifies rare harms, minimizes common benefits (availability heuristic)

— Pictographs / icon arrays (100 or 1000 figures, colored to show events)

— Bar graphs comparing treated vs untreated arms

— Mayo Statin Choice decision aid, Option Grid, AHRQ SHARE tools

— Avoid pie charts (poor at small differences) and isolated percentages

Numeracy and risk communication has no physical exam per se, but Step 3 tests structured assessment of the patient's comprehension and decisional capacity — the "exam" of shared decision-making.

Capacity assessment (Appelbaum criteria):

Teach-back technique (the bedside "exam"):

Observed innumeracy clues:

Decisional aids — the validated tools:

Key distinction: decisional capacity is a clinical judgment any physician can make; competence is a legal determination by a court. A patient with low numeracy can still have capacity if, with adequate explanation, they meet Appelbaum criteria — innumeracy alone is not grounds to override autonomy.

CCS pearl: in a CCS case, ordering "patient education, decision aid, teach-back" is appropriate before consent for elective interventions (e.g., elective CEA, screening colonoscopy, anticoagulation). It buys time on the clock and is rarely wrong as an early action.

Diagnostic Workup — Calculating ARR, RRR, and NNT from Trial Data

— CER (control event rate) = events in control / total in control

— EER (experimental event rate) = events in treated / total in treated

— ARR = CER − EER

— RRR = ARR / CER = 1 − (EER/CER)

— NNT = 1 / ARR (round up to nearest whole patient)

— ARI / NNH = same math, for adverse events

— CER = 2.8% MI over 2 yr; EER = 1.4%

— ARR = 1.4% (0.014); RRR = 50%; NNT = 1/0.014 ≈ 72 over 2 years

— Communicated: "Out of 100 people like you taking this for 2 years, about 1–2 will avoid a heart attack."

— Stroke CER ≈ 1.6%/yr; EER ≈ 1.27%/yr → ARR ≈ 0.33%/yr → NNT ≈ 303/yr

— Major bleed ARR ≈ 0.96%/yr favoring apixaban → NNH (vs warfarin) actually favorable

The 2×2 framework every Step 3 candidate must own:

Worked example — statin primary prevention (JUPITER-like):

Worked example — apixaban in AF (ARISTOTLE-like):

Always specify the time horizon — an NNT of 50 over 10 years ≠ NNT of 50 over 1 year.

Confidence intervals matter: if the 95% CI for ARR crosses 0, the NNT is not interpretable (could be NNH); report as "no statistically significant benefit." A point-estimate NNT with a wide CI (e.g., 25 to ∞) should be communicated as uncertain.

Composite endpoints inflate apparent benefit — break them down ("most of the benefit was in revascularization, not death") before quoting NNT.

Board pearl: when a question gives CER and EER, the answer is virtually always ARR or NNT, not RRR. Memorize: NNT = 100 / ARR(%). An NNT of 100 = ARR of 1%.

Diagnostic Workup — Identifying Misleading Statistical Framing

— Relative-only reporting: "cuts risk by 40%" without baseline → demand absolute numbers

— Surrogate endpoints: LDL drop, BP drop, tumor shrinkage ≠ mortality benefit; NNT for surrogates is not equivalent to NNT for outcomes

— Composite endpoints dominated by soft components (revascularization, hospitalization)

— Per-protocol vs intention-to-treat: PP exaggerates benefit

— Subgroup analyses: post-hoc subgroups are hypothesis-generating, not actionable

— Lead-time and length-time bias in screening — apparent survival ↑ without mortality ↓

— 5-year survival is inflated by lead time; use disease-specific mortality

— Stage-shift does not prove screening reduces death

— Overdiagnosis (PSA, thyroid, DCIS) means NNT to prevent one death may be enormous and NNH for harms (biopsy, incontinence, anxiety) is small

— Biennial mammography ages 50–74: NNT ≈ 1000 over 10 years to prevent 1 breast cancer death

— ~150 false positives, ~20 biopsies, ~1–2 overdiagnoses per 1000 — these go in the conversation

— ERSPC: NNT ≈ 781 invited to screen to prevent 1 prostate cancer death over 13 yr; NNH for ED/incontinence dwarfs benefit

Common framing manipulations on Step 3 stems and in patient-facing media:

Screening-specific innumeracy traps:

Mammography example (USPSTF-style framing):

PSA example:

Number needed to screen (NNS) ≠ NNT — NNS includes the entire screened population, most of whom never had disease.

Key distinction: NNT applies to treatment of an established condition or a defined-risk population; NNS applies to population screening and is almost always far larger. Conflating them is a classic distractor.

Board pearl: if a stem cites a dramatic-sounding screening benefit with only relative risk or 5-year survival, the correct critique is lead-time bias and absence of mortality benefit — the test wants you to refuse to recommend the test based on that data alone.

Risk Stratification — Tailoring NNT to the Individual Patient

— Low-risk 45-yo, 10-yr ASCVD = 5%, RRR ≈ 10% → ARR = 0.5% → NNT = 200

— Same RRR in 10-yr ASCVD = 20% → ARR = 2% → NNT = 50

— Bleeding NNH is roughly constant in absolute terms → in low-risk patients, NNH < NNT → net harm. This is why USPSTF (2022) downgraded aspirin for primary prevention.

— PCE / PREVENT for ASCVD

— CHA₂DS₂-VASc for AF stroke; HAS-BLED for bleeding

— FRAX for fracture; Gail / Tyrer-Cuzick for breast cancer

— Lung cancer: PLCOm2012 or USPSTF criteria

— Statins: ASCVD ≥7.5% → discuss; ≥20% → recommend

— Anticoagulation: CHA₂DS₂-VASc ≥2 (men) / ≥3 (women) → benefit usually exceeds harm

— When NNT/NNH ratio ≤ 1, intervention favored; when >1, reconsider

The single most important concept: NNT is not a fixed number — it varies with the patient's baseline absolute risk. The same RRR applied to a low-risk and high-risk patient yields wildly different NNTs.

Worked example — aspirin for primary prevention:

Risk calculators to anchor the conversation:

Decision threshold logic:

Step 3 management: for a 52-yo with LDL 145 and 10-yr ASCVD 6%, the correct action is shared decision-making with personalized ARR/NNT, not automatic statin or automatic deferral. Show the patient: "Out of 100 people like you over 10 years, about 6 will have a heart attack or stroke without treatment, and about 4–5 with a statin — so 1–2 avoid an event."

Time horizon counseling: lifetime risk reframes low 10-yr risk (young patient with familial dyslipidemia) — use lifetime NNT to justify earlier therapy.

Board pearl: the same drug can have an NNT of 30 or 300 depending on the patient — Step 3 rewards individualized framing, not trial-level numbers quoted verbatim.

Pharmacotherapy of Communication — First-Line "Regimen" for Risk Discussion

— Seek patient participation

— Help explore and compare options

— Assess values and preferences

— Reach a decision together

— Evaluate the decision over time

— Ask what the patient already knows/believes

— Tell in plain language with absolute numbers + visuals

— Ask them to teach back

— Use frequencies, not percentages ("3 out of 100" beats "3%")

— Use a constant denominator when comparing options ("out of 100" in both arms)

— Present both benefits and harms, side-by-side

— Use pictographs with random distribution of "event" icons

— Avoid decimals <1 ("0.3%") — convert to "3 in 1000"

— Give time horizon every time

— State uncertainty ("our best estimate is…")

— RRR alone, vague qualifiers ("significantly lower"), and statistical jargon (p-values, hazard ratios) at the bedside

— Loaded framing ("Most doctors recommend…") that undermines autonomy

— Anchoring on a single number without range/CI

— "Over 10 years, about 6 of 100 people like you will have a heart attack or stroke. With a daily statin, that drops to about 4 of 100 — so 2 of 100 are helped, 94 of 100 would have been fine anyway, and 4 of 100 still have an event despite the pill. The most common downside is muscle aches, in about 5 of 100."

Step 3 favors a structured SDM script. Memorize the AHRQ SHARE approach:

Or the simpler "Ask-Tell-Ask":

Numeric communication rules (evidence-based best practices):

Avoid:

Sample script for statin:

CCS pearl: order "counsel patient, shared decision-making, decision aid" before prescribing a preventive medication in an ambulatory CCS case — this counts as a substantive action and is rarely penalized.

Procedures — Decision Aids, Visual Tools, and Documentation

— Mayo Clinic Statin Choice — interactive pictograph, personalizes ASCVD risk

— Option Grid — one-page side-by-side comparison

— AHRQ SHARE toolkit, Ottawa Personal Decision Guide

— PREVENT/ASCVD Risk Estimator Plus with built-in NNT visualization

— NNT.com — clinician reference for pooled NNTs

1. Icon arrays / pictographs with 100 or 1000 figures, random placement

2. Bar charts with identical scales

3. Numerical frequencies ("3 in 100")

4. Percentages

5. Relative risk statements (worst — avoid)

— Quote procedure-specific NNT (e.g., asymptomatic CEA: NNT ≈ 20 over 5 yr to prevent 1 stroke; perioperative stroke/MI/death ≈ 2–3%)

— Quote NNH for the specific complication being signed for

— Document patient-specific risk (STS score, EuroSCORE, NSQIP) not generic averages

— Patient's baseline risk + source (calculator name, inputs)

— Absolute benefit (ARR, NNT) and absolute harm (ARI, NNH) over stated time horizon

— Decision aid used (name + version)

— Teach-back confirmed

— Patient's stated preference and rationale

— Plan and follow-up

Validated decision aids (Step 3 may reference by name or function):

Visual format hierarchy (best to worst for risk perception):

Procedure-specific consent (e.g., elective PCI, CEA, hip replacement):

Documentation template (auditable, medico-legally protective):

Health-system context: CMS reimburses SDM visits for lung cancer screening LDCT (G0296) and ICD implantation — the visit and documentation of NNT/NNH discussion are billable and required for coverage.

Board pearl: for lung cancer LDCT screening, a documented SDM visit using a decision aid is a CMS coverage requirement — failing to document this can render the screening study non-reimbursable and is a tested Step 3 health-systems point.

Special Populations — Elderly and Cognitive/Sensory Impairment

— Declining working memory and processing speed

— Higher prevalence of low health literacy (~60% age 75+)

— Sensory deficits (vision, hearing) impair pictograph and verbal comprehension

— Polypharmacy makes NNH conversations more important than NNT

— Higher baseline risk → often lower NNT for treatment (statins, anticoagulation)

— But competing mortality can erase long-horizon benefit — a 5-year NNT means little if life expectancy is 3 years

— Use time-to-benefit (TTB) literature: e.g., statins in primary prevention need ~2.5 yr to show benefit; tight glycemic control ~8 yr; mammography ~10 yr

— Large-print pictographs; ensure hearing aids in

— Slow pace, single concept per minute

— Include caregiver/proxy with patient permission

— Use life-expectancy calculators (ePrognosis) to bound the time horizon

— Reframe: "Given your overall health, treatment X is likely to help over the next [X] years"

— "Continuing this pill at your age may give you a 1-in-500 chance of help but 1-in-50 chance of dizziness/fall"

— Tested on Step 3 in geriatric polypharmacy stems

— MCI ≠ incapacity; assess decision-specific capacity for each high-stakes choice

— If capacity lacking → previously expressed wishes, advance directive, surrogate per state hierarchy

Older adults face compounded innumeracy risk:

NNT shifts in elderly:

Practical adjustments:

Deprescribing conversations are NNT/NNH discussions in reverse:

Capacity in cognitive impairment:

Step 3 management: in an 84-yo with dementia and AF, CHA₂DS₂-VASc=6, HAS-BLED=4, life expectancy ~2 yr — the answer is rarely automatic anticoagulation. It is a goals-of-care conversation with surrogate, framing NNT vs NNH within remaining life expectancy. Reflexive DOAC initiation is usually the wrong choice.

Special Populations — Pregnancy, Pediatrics, Low Literacy, LEP

— Risk communication must include maternal AND fetal outcomes — two NNTs/NNHs

— Example: aspirin 81 mg from 12 wk in high-risk pregnancy → NNT ≈ 50 to prevent preeclampsia; minimal fetal harm

— Avoid framing that pressures (autonomy is paramount in obstetric decisions)

— Counsel on vaccine NNT/NNH (Tdap, influenza, RSV, COVID) — pregnant patients are a high-misinformation group

— Communicate to parents AND, when age-appropriate, the adolescent (assent)

— Vaccines are the archetype: HPV vaccine NNT ≈ 100s to prevent cervical cancer over decades — frame as lifetime risk reduction

— Address parental misperception of vaccine NNH (denominator illiteracy drives anti-vaccine sentiment)

— Certified medical interpreter required for any informed-consent or risk-discussion encounter — using family is a patient safety event and Joint Commission violation

— Translate decision aids into the patient's primary language; many AHRQ tools are multilingual

— Universal precautions: assume everyone needs plain language

— 5th–6th grade reading level for written materials

— Chunk-and-check: deliver one concept, teach-back, then next

— Use stories sparingly — anecdotes are powerful but biased

— Numeracy correlates with education and SES, not intelligence

— Same patient may have high verbal literacy but low numeracy — screen separately

Pregnancy:

Pediatrics:

Limited English Proficiency (LEP):

Low health literacy (any demographic):

Numeracy disparities:

Board pearl: when a Step 3 stem describes a parent refusing HPV vaccine citing risk, the correct answer is explore concerns, share absolute benefits/harms, schedule follow-up — not dismissal, not coercion, not reporting. Persistent vaccine counseling over multiple visits has a documented effect on uptake.

Complications and Adverse Outcomes of Poor Risk Communication

— Overtreatment: patients consent to interventions whose NNT vastly exceeds NNH (e.g., PSA screening in low-risk men → biopsy complications, overtreatment)

— Undertreatment: patients refuse beneficial therapy because RRR sounded too small or NNH was overemphasized

— Decisional regret: post-hoc dissatisfaction, linked to inadequate pre-decision information

— Anxiety / "VOMIT" (victims of modern imaging technology): incidentaloma cascades from low-yield screening

— Non-adherence: patients who don't understand benefit stop the drug

— Low-value care consumes resources and exposes patients to harm

— Malpractice exposure: inadequate informed consent is the most common procedural malpractice claim

— Health inequities: numeracy gaps amplify existing disparities

— PSA without SDM → unnecessary biopsies → sepsis, ED, incontinence

— CT for low-risk chest pain → incidentaloma → invasive workup

— Genetic testing without pretest counseling → uninterpretable VUS → patient distress

— CPR consent based on TV-show survival rates (~70%) vs real in-hospital ~17%, out-of-hospital ~10% → DNR decisions distorted

— Framing effect: "90% survive" feels better than "10% die" — same number

— Anchoring on first statistic heard

— Availability bias from recent vivid case

— Optimism bias ("won't happen to me")

— Default bias — patients accept the default (use defaults ethically)

Patient-level harms:

System-level harms:

Specific high-yield examples of communication failure:

Bias amplifiers:

Key distinction: decisional regret ≠ bad outcome regret. A patient who experiences a known complication after adequate SDM has a bad outcome; one who says "I never would have agreed if I'd known" has inadequate informed consent — the latter is a defect in process and the testable failure mode.

When to Escalate — Ethics Consult, Risk Management, Surrogate Decision-Making

— Suspected lack of capacity despite clear explanation → formal capacity assessment ± psychiatry consult

— Persistent disagreement between patient and surrogates, or among surrogates → ethics consultation

— Refusal of life-saving therapy in a context suggesting impaired understanding (delirium, untreated depression, intoxication) → treat reversible causes, reassess

— Suspected coercion (family pressure, financial abuse) → social work, adult protective services

— Cultural or religious conflict where standard SDM doesn't fit → chaplaincy, cultural liaison

— Pediatric refusal of recommended treatment with risk of significant harm → ethics, possibly child protective services / court order

1. Court-appointed guardian

2. Healthcare power of attorney / proxy

3. Spouse

4. Adult children (majority)

5. Parents

6. Adult siblings

7. Close friend

— Substituted judgment (preferred when patient's wishes known): "What would the patient choose?"

— Best interest standard: used when patient's wishes unknown

— Surrogates should also receive NNT/NNH framing for proposed interventions

— Jehovah's Witness refusing transfusion: capacity-assured adult — autonomy prevails; minor — court order possible

— Pregnant patient refusing intervention — autonomy generally prevails (Step 3 favors maternal autonomy)

Indications to escalate beyond bedside conversation:

Surrogate decision-making hierarchy (varies by state but typical order):

Substituted judgment vs best interest:

Special situations:

CCS pearl: in a case with ambiguous capacity or family conflict, ordering "ethics consultation" and "social work consult" is rarely wrong and reflects appropriate escalation. Do not unilaterally override a patient's stated preference without due process.

Key Differentials — Same-Category Statistical Measures Often Confused

— NNT = treat to prevent 1 event; NNH = treat to cause 1 harm

— Both require same time horizon and same population

— NNS includes all screened, most without disease; always larger than NNT

— ARR is the only measure that scales with baseline risk; RRR is roughly constant across baseline risks for a given drug

— In rare outcomes, OR ≈ RR; in common outcomes, OR overestimates RR

— OR is non-intuitive for patients — convert to ARR

— Time-to-event measure from Cox models; not directly ARR

— Convert via baseline survival to absolute terms

— AR = ARR in exposed; PAR = population-level burden

— Sens/spec are test properties; PPV/NPV depend on prevalence — critical for screening counseling (low-prevalence disease → low PPV even with good test)

— Example: D-dimer in low-pretest-probability PE has high NPV; in high-pretest, low PPV

— Survival inflated by lead/length-time bias in screening contexts

— Mortality is the unbiased outcome

— Median is robust to outliers; mean skewed by long survivors

— Always disaggregate; benefit may be driven by softest endpoint

NNT vs NNH:

NNT vs NNS (number needed to screen):

ARR vs RRR:

Absolute risk vs relative risk vs odds ratio:

Hazard ratio (HR):

Attributable risk vs population attributable risk:

Sensitivity/specificity vs PPV/NPV:

5-year survival vs disease-specific mortality:

Median survival vs mean vs hazard:

Composite vs individual endpoints:

Board pearl: when a stem reports a screening test's "85% sensitivity, 95% specificity" and asks how to counsel a patient, the next step is to compute or estimate PPV using local prevalence — a positive test in a low-prevalence setting is more likely false than true, and that's what the patient needs to hear.

Key Differentials — Other-Category Pitfalls in Risk Communication

— Framing effect: mortality framing → risk-averse; survival framing → risk-seeking

— Loss aversion: losses weighted ~2× gains — "lose 5 years" hits harder than "gain 5 years"

— Default bias: opt-in vs opt-out radically changes uptake (organ donation)

— Anchoring: first number sets reference point

— Availability heuristic: recent vivid event distorts probability

— Innumeracy itself: confusing percent change with percentage points

— Base-rate neglect: ignoring prevalence when interpreting tests

— Classic Step 3 stem: "Test 99% sensitive/specific for disease with 0.1% prevalence — what's the chance a positive truly has disease?" → answer ~9% (PPV)

— Paternalistic clinician + autonomy-preferring patient → conflict

— Patient wants direction; clinician overloads with numbers → decisional paralysis

— More than 3–4 numbers per visit → comprehension drops

— Written summary mitigates

— Some cultures favor family-based decisions; check preference rather than assume

— Pharma DTC ads (relative-only), social media, AI chatbots without context

— Address directly: "Where did you see that number? Let's look at it together."

— Trial sponsors emphasize RRR; clinician's job is to translate to ARR/NNT independently

Behavioral economics traps (each a likely distractor):

Bayesian reasoning failures:

Communication style mismatches:

Cognitive load issues:

Cultural framing:

Misinformation sources:

Conflict of interest:

Key distinction: mortality framing ("10% die") and survival framing ("90% live") describe the same statistic but produce different decisions — the ethical antidote is to present both frames routinely, sometimes called "balanced framing," and this is the Step 3-favored approach over either frame alone.

Secondary Prevention / Longitudinal Plan for Embedding Risk Communication

— Risks change over time (age, new comorbidity, new evidence)

— Re-discuss statin, anticoagulation, screening at annual visits or with major status change

— Document re-discussion in problem list

— Diabetes: A1c target individualized via NNT for microvascular benefit vs NNH for hypoglycemia (ADA: tighter in younger/healthier, looser in elderly/frail)

— HTN: SPRINT-style intensive control NNT vs syncope/AKI NNH

— AF: annual reassessment of CHA₂DS₂-VASc and HAS-BLED

— Osteoporosis: drug holiday after 5 yr bisphosphonate — NNT for further fracture prevention falls

— Mammography ≥75: USPSTF "insufficient evidence" → SDM based on life expectancy and values

— Colonoscopy 76–85: individualize; >85 generally stop

— These are NNT-shrinks-with-shortening-time-horizon conversations

— Post-MI: dual antiplatelet NNT for recurrent MI vs NNH for bleeding — frame at discharge AND at 1-yr follow-up (DAPT duration decision)

— Anticoagulation post-PE: 3 mo vs extended — NNT for recurrence falls, NNH for bleed accumulates

— EHR-embedded risk calculators with NNT outputs

— Patient portal access to personalized risk summaries

— Pre-visit decision aids ("homework") improve in-visit efficiency

Make SDM a recurring, not one-off, process:

Embed in chronic disease management:

Screening cessation discussions:

Discharge / care-transition framing:

Health-systems tools:

Step 3 management: at the annual wellness visit, re-run the ASCVD calculator, re-present updated NNT, and document SDM — this is the value-based-care answer and aligns with CMS quality measures (MIPS, HEDIS preventive care metrics). Don't simply renew the statin without re-engaging the patient.

Follow-Up, Monitoring, and Counseling Cadence

— Short interval (2–4 weeks) after initiating a preventive therapy: tolerability, adherence, address regret

— 3 months: lab monitoring as indicated (e.g., statin → ALT/CK only if symptomatic, lipid panel for response)

— Annually: re-evaluate risk, re-discuss continuation

— Statin: LDL response, muscle symptoms, new diabetes (NNH ~1/250)

— Anticoagulant: bleeding, renal function (DOAC dosing), annual reassessment

— Aspirin primary prevention: GI bleed, ICH — if NNH exceeds NNT, deprescribe

— SGLT2/GLP-1: GU infection, euglycemic DKA (SGLT2); GI side effects (GLP-1)

— Decisional regret scale, satisfaction with decision — useful in QI

— Validated literacy tools: REALM, TOFHLA, single-item screener

— "Are you still taking X? How is it going? Any side effects? Do you remember why we started it?"

— Re-anchor to NNT/NNH if patient's commitment is wavering

— Address misinformation surfacing between visits (social media, news, family advice)

— Lifestyle interventions (diet, exercise, smoking cessation) have better NNTs than many drugs for ASCVD prevention — communicate this explicitly

— Smoking cessation: NNT ≈ 10 to prevent 1 premature death over a lifetime — among the most powerful in medicine

Post-decision follow-up structure:

Monitoring parameters specific to NNT/NNH-driven therapies:

Patient-reported outcomes:

Counseling content at each visit:

Behavior-change support:

CCS pearl: in a CCS office case, scheduling a 2–4 week follow-up after initiating any new preventive medication (statin, antihypertensive, anticoagulant, antidepressant) and ordering relevant labs at the appropriate interval is standard — premature discharge of the case before establishing follow-up loses points. "Schedule follow-up appointment" is an orderable action.

Ethical, Legal, and Patient Safety Considerations

1. Disclosure of nature, risks, benefits, alternatives, and consequences of refusal — including absolute risks

2. Capacity to understand and decide

3. Voluntariness (free of coercion)

— Reasonable physician standard: what a similar physician would disclose

— Reasonable patient (materiality) standard: what a reasonable patient would want to know — increasingly the legal norm; mandates patient-relevant absolute numbers

— Specific risks discussed with frequencies

— Decision aid used

— Teach-back result

— Patient's stated decision and reasoning

— Witness signature where required

— Consent forms signed without conversation ("signature ≠ consent")

— Last-minute consent on the procedure table (coercion concern)

— Consent by family without patient assessment when patient has capacity

— Use of family as interpreter (LEP) — breaches confidentiality and accuracy, Joint Commission violation

— Genetic risk results that reveal non-paternity, communicable disease, or duty-to-warn third parties (Tarasoff) require careful pretest framing

— Hospital discharge without explaining new med's NNT/NNH → poor adherence, readmission

— Use teach-back at discharge; medication reconciliation; scheduled follow-up within 7–14 days — these are CMS readmission-reduction levers

— Numeracy disparities compound healthcare disparities — universal-precautions communication is an equity intervention

Informed consent is the legal backbone of risk communication. Three elements:

Standard of disclosure (varies by state):

Documentation that protects:

Failure modes that trigger litigation/safety events:

Mandatory reporting overlap:

Transition-of-care risks (classic Step 3):

Equity and justice:

Concrete Step 3 scenario: Spanish-speaking patient scheduled for elective CEA; surgeon obtains consent using the patient's bilingual adult son as interpreter the morning of surgery. Correct action: stop, obtain a certified medical interpreter, re-consent with documented absolute risk/benefit discussion (perioperative stroke ~2%, NNT 20 over 5 yr in asymptomatic disease), and confirm teach-back. Proceeding is a patient-safety event and grounds for invalid consent.

High-Yield Associations and Rapid-Fire Clinical Facts

NNT = 1 / ARR. Always specify time horizon. NNH = 1 / ARI.

RRR is misleading without baseline risk. Always demand ARR.

Pictographs > percentages > relative risks for patient comprehension.

Frequencies > percentages ("3 in 100" beats "3%").

Constant denominator when comparing options.

Teach-back is the gold-standard comprehension check — never "Do you understand?"

Certified interpreter required for LEP — family is a safety violation.

Reasonable-patient standard governs disclosure in most US states.

Capacity is decision-specific, clinician-determined; competence is court-determined.

Aspirin for primary prevention: USPSTF (2022) — individualize 40–59; do not initiate ≥60. NNT high in low-risk, NNH constant.

Statin primary prevention: ASCVD ≥7.5% triggers discussion; NNT ~50–100 over 10 yr.

Mammography: NNT ~1000 over 10 yr; substantial false-positive NNH.

PSA: NNT ~780 invited-to-screen; significant overdiagnosis NNH.

Lung CT screening: NNT ~320 to prevent 1 lung cancer death; CMS requires documented SDM visit.

HPV vaccine: large NNT but enormous lifetime benefit at low NNH.

Smoking cessation: NNT ~10 — one of the best in medicine.

CPR survival: real in-hospital ~17%, out-of-hospital ~10% — public believes ~70% (TV).

Composite endpoints inflate apparent benefit — disaggregate.

Surrogate endpoints (LDL, BP) ≠ outcome endpoints — NNT differs.

Lead-time and length-time bias inflate screening 5-yr survival.

PPV depends on prevalence — low-prevalence positive ≈ likely false positive.

Framing effect: mortality vs survival framing flips choices — present both.

Time-to-benefit matters in elderly — statin TTB ~2.5 yr; mammography ~10 yr; tight glycemic control ~8 yr.

Decisional regret signals inadequate informed consent process.

Board pearl: if a Step 3 answer choice quotes only RRR, only 5-year survival, or only sensitivity/specificity without prevalence, it is almost certainly the wrong way to counsel the patient — pick the option that translates to absolute, time-anchored, frequency-format numbers with both benefits and harms.

Board Question Stem Patterns

Pattern 1 — "The drug reduces stroke by 40%": stem gives only RRR. Answer: request or compute ARR and NNT; counsel using absolute numbers over a defined time horizon. Distractors: prescribe based on RRR; refuse based on RRR.

Pattern 2 — Screening test in low-prevalence population: stem gives sensitivity, specificity, prevalence. Answer: compute PPV; counsel that positive result is often false; discuss confirmatory testing before invasive workup.

Pattern 3 — Patient refusing statin/anticoagulant/vaccine: stem describes a competent adult refusing recommended therapy. Answer: explore concerns, share ARR/NNT with decision aid, schedule follow-up — not override, not abandon.

Pattern 4 — Family using adolescent child as interpreter: Answer: obtain certified medical interpreter; re-consent.

Pattern 5 — Surgeon obtains consent on OR table: Answer: halt, re-consent in unhurried setting with documented risk/benefit and teach-back.

Pattern 6 — Elderly with limited life expectancy and high CHA₂DS₂-VASc: Answer: goals-of-care discussion with surrogate, individualize NNT vs NNH within remaining life expectancy, do not reflexively anticoagulate.

Pattern 7 — Patient cites TV ad ("cuts heart attacks in half"): Answer: identify framing, re-present absolute numbers, teach-back.

Pattern 8 — Family meeting on CPR: Answer: present real survival data (~17% in-hospital), correct TV-driven misperception, then ask about values.

Pattern 9 — Primary prevention aspirin in 65-yo: Answer: USPSTF — do not initiate ≥60; NNH exceeds NNT.

Pattern 10 — Mammography in 76-yo with multimorbidity: Answer: individualized SDM; consider life expectancy and TTB ~10 yr; reasonable to stop.

Pattern 11 — Composite endpoint trial: Answer: disaggregate; counsel based on patient-important outcomes (death, MI), not soft endpoints.

Pattern 12 — Capacity question: Answer: apply Appelbaum (understand, appreciate, reason, choice); innumeracy alone does not negate capacity.

CCS pearl: in CCS, after presenting a preventive option, ordering "counsel patient" + "shared decision-making" + "schedule follow-up" before prescribing is a high-value sequence; reflexive prescribing without these steps loses communication-skills points on the actual exam construct and mirrors real-world quality measures.

One-Line Recap

Effective risk communication translates trial data into patient-specific absolute risk reduction, number needed to treat, and number needed to harm — framed with frequencies and visuals over a defined time horizon, paired with teach-back — because informed consent and shared decision-making are valid only when the patient genuinely understands the magnitude of benefit and harm.

Math you must own: NNT = 1/ARR; ARR = CER − EER; NNH = 1/ARI. RRR alone is never sufficient — always anchor to the patient's baseline absolute risk.

Communication best practices: use frequencies ("3 in 100"), constant denominators, pictographs, balanced framing (both mortality and survival), time horizons, and teach-back. Avoid percentages, decimals, jargon, and relative-only statements.

Individualization is everything: the same drug yields radically different NNTs across baseline-risk strata; recompute for each patient using validated calculators (PREVENT/ASCVD, CHA₂DS₂-VASc, FRAX), and re-discuss at each annual visit and major status change.

Ethics and safety: informed consent requires disclosure (absolute numbers), capacity, and voluntariness; LEP requires certified interpreter; surrogate decisions follow state hierarchy with substituted judgment preferred; decisional regret signals process failure, not just bad outcomes.

Board pearl: on Step 3, the correct answer is almost always the one that converts relative numbers to absolute numbers, uses a decision aid, confirms understanding with teach-back, and schedules follow-up — choose the option that respects autonomy, individualizes risk, and documents the shared decision rather than the one that prescribes reflexively, defers without discussion, or relies on relative risk or surrogate endpoints alone.