HS 326 cheat sheet

Randomization allocates participants by chance to reduce selection bias and balance known and unknown confounders across arms.

Blinding levels: single (participant), double (participant plus clinician or assessor), triple (adds data analyst); blinding reduces performance and detection bias.

Allocation concealment hides the upcoming assignment from the person enrolling participants; it differs from blinding and prevents selection bias.

Superiority vs noninferiority: noninferiority trials require a prespecified margin and both intention-to-treat and per-protocol analyses to interpret results.

Crossover designs need washout periods and assume no carryover effect; within-subject comparison reduces sample size but is vulnerable to period effects.

Cluster randomized trials require adjustment for intracluster correlation; design effect = 1 + (m − 1) × ICC, where m is cluster size.

Phase I establishes safety and dosing, Phase II preliminary efficacy, Phase III comparative efficacy, Phase IV post-marketing surveillance.

PICOT frames a researchable question: Population, Intervention, Comparator, Outcome, Time frame.

A confounder is associated with both exposure and outcome and is not on the causal pathway; effect modification is a differing effect across strata.

Levels of evidence (typical hierarchy): systematic reviews of RCTs > RCTs > cohort > case-control > cross-sectional > case series > expert opinion.

GRADE rates certainty as high, moderate, low, or very low, downgrading for risk of bias, inconsistency, indirectness, imprecision, and publication bias.

Risk of bias tools: Cochrane RoB 2 for randomized trials, ROBINS-I for nonrandomized studies, AMSTAR 2 for systematic reviews, QUADAS-2 for diagnostic accuracy.

Intention-to-treat analysis preserves randomization and estimates real-world effect; per-protocol analysis estimates efficacy under adherence and can bias results.

Attrition above roughly 20% threatens validity; report loss to follow-up by arm and compare baseline characteristics of completers and dropouts.

A funnel plot asymmetry suggests small-study effects or publication bias; asymmetry can also arise from true heterogeneity.

Absolute risk reduction is more clinically interpretable than relative risk reduction; a large relative reduction can correspond to a tiny absolute benefit.

Surrogate endpoints (e.g., LDL cholesterol, HbA1c) require validated linkage to patient-important outcomes before guiding practice.

Sensitivity = TP/(TP+FN); specificity = TN/(TN+FP); both are properties of the test, not the population.

Positive predictive value = TP/(TP+FP); PPV rises with prevalence, negative predictive value falls as prevalence rises.

Likelihood ratio positive = sensitivity/(1 − specificity); LR negative = (1 − sensitivity)/specificity; LR > 10 or < 0.1 meaningfully shifts post-test probability.

Number needed to treat = 1/absolute risk reduction; NNT increases as the event rate falls.

p-value is the probability of data at least as extreme as observed, given the null hypothesis is true; it is not the probability the null is true.

Type I error (alpha) is a false positive; Type II error (beta) is a false negative; power = 1 − beta, conventionally 0.80.

95% confidence interval corresponds to alpha 0.05; a CI excluding the null value corresponds to p < 0.05 for a two-sided test.

Standard deviation describes spread of individual values; standard error of the mean = SD/√n describes precision of the sample mean.

Odds ratio approximates relative risk only when the outcome is rare (roughly <10%); for common outcomes, report risk ratios or risk differences.

Incidence measures new cases per population at risk per time; prevalence measures all existing cases at a point or period and equals incidence × duration for steady states.

Attack rate = new cases/people at risk during an outbreak; case fatality rate = deaths from disease/confirmed cases.

Cohort studies yield incidence and relative risk directly; case-control studies yield odds ratios and are efficient for rare outcomes.

Bias types: selection, information (recall, misclassification), confounding; ecologic fallacy applies group-level associations to individuals.

Herd immunity threshold ≈ 1 − 1/R0; for R0 = 5 the threshold is 80%.

Screening test criteria (WHO-style): important health problem, accepted treatment, recognizable latent stage, suitable test, cost-effective.

Lead-time bias inflates survival from earlier detection without extending life; length-time bias favors detection of slower-progressing disease.

Age-adjusted rates permit comparison across populations with different age structures; crude rates reflect both risk and population composition.

EBM steps: formulate question, acquire evidence, appraise validity and relevance, apply to patient, evaluate performance.

Pre-test probability combined with LR yields post-test probability; convert probability to odds, multiply by LR, convert back.

Therapeutic threshold is the probability above which benefits of treatment outweigh harms; testing is useful when results cross that threshold.

Shared decision-making integrates best evidence, clinician expertise, and patient values; decision aids improve knowledge and reduce decisional conflict.

Clinical practice guidelines should report evidence grading, conflicts of interest, and applicability; they are not a substitute for patient-specific judgment.

Choosing Wisely-style recommendations target low-value care; overdiagnosis and overtreatment are harms of indiscriminate testing.

A clinically meaningful difference is the smallest benefit patients would consider important; statistical significance does not equal clinical importance.

Belmont principles: respect for persons (informed consent), beneficence and nonmaleficence (risk-benefit), justice (fair selection).

Informed consent elements: purpose, procedures, risks, benefits, alternatives, confidentiality, voluntariness, and right to withdraw.

Institutional review board approval precedes human-subjects research; exempt and expedited categories differ from full board review.

Declaration of Helsinki and the Common Rule govern human-subjects protections; the FDA regulates drugs and devices in the United States.

Vulnerable populations (children, prisoners, pregnant women, cognitively impaired) require additional safeguards and assent where applicable.

Research misconduct includes fabrication, falsification, and plagiarism; it excludes honest error and differences of opinion.

Authorship requires substantial contribution, drafting or critical revision, final approval, and accountability; ghost and gift authorship are violations.

Conflict-of-interest disclosure covers financial and nonfinancial interests; registration of trials (e.g., ClinicalTrials.gov) reduces selective reporting.

Absolute risk formats (e.g., 2 in 100 vs 1 in 100) are understood better than relative risk statements alone.

Teach-back confirms patient comprehension; plain-language materials should target roughly sixth- to eighth-grade reading level.

SBAR (Situation, Background, Assessment, Recommendation) structures interprofessional handoffs.

Health literacy universal precautions assume all patients may struggle; use pictures, numbers, and limit to three key messages.

Numeracy framing effects: gain-framed messages increase prevention uptake, loss-framed messages can increase detection behaviors.

Interprofessional collaboration competencies include roles and responsibilities, teamwork, communication, and values/ethics.

Evidence implementation uses audit and feedback, reminders, and champions; passive dissemination alone rarely changes practice.