Health Medical Research Clinical research methodology, biostatistics, epidemiology, and evidence-based medicine
哲学与伦理 Study Design Hierarchy (Evidence Pyramid)
Systematic Reviews & Meta-analyses (strongest)Randomized Controlled Trials (RCTs) Cohort Studies (prospective > retrospective)Case-Control Studies Cross-Sectional Studies Case Reports / Expert Opinion (weakest)
Study Types
RCT Design
Parallel : treatment vs control, simultaneousCrossover : each subject gets both, washout period betweenCluster : randomize groups (hospitals, clinics), not individualsNon-inferiority : prove new treatment is no worse than standardAdaptive : modify design mid-trial based on interim results
Observational
快速安装
Health Medical Research npx skills add RomulanAI/tal-shiar
星标 0
更新时间 2026年4月9日
职业
Prospective cohort : follow exposed/unexposed forward in time → relative riskRetrospective cohort : use historical records → same measures, less controlCase-control : start with cases (disease) and controls → odds ratioCross-sectional : snapshot at one point in time → prevalence, associations
Key Measures Measure Formula Use Relative Risk (RR) P(event|exposed) / P(event|unexposed) Cohort studies Odds Ratio (OR) (ad) / (b c) from 2x2 table Case-control studies Hazard Ratio (HR) Instantaneous rate ratio Survival analysis NNT 1 / ARR (absolute risk reduction) Clinical decision-making NNH 1 / ARI (absolute risk increase) Harm assessment Sensitivity TP / (TP + FN) Diagnostic test (rule out) Specificity TN / (TN + FP) Diagnostic test (rule in) PPV/NPV Depends on prevalence Post-test probability AUC-ROC Area under ROC curve Discriminative ability
Biostatistics
Sample Size # Two-group comparison (proportions)
library(pwr)
pwr.2p.test(h = ES.h(p1 = 0.30, p2 = 0.20), # effect size
sig.level = 0.05, power = 0.80)
# Two-group comparison (means)
pwr.t.test(d = 0.5, sig.level = 0.05, power = 0.80, type = "two.sample")
Survival Analysis library(survival)
library(survminer)
# Kaplan-Meier
km <- survfit(Surv(time, event) ~ treatment, data = df)
ggsurvplot(km, pval = TRUE, risk.table = TRUE,
xlab = "Months", ylab = "Survival Probability")
# Cox Proportional Hazards
cox <- coxph(Surv(time, event) ~ treatment + age + stage, data = df)
summary(cox) # HRs with CIs
cox.zph(cox) # test PH assumption
library(meta)
# Fixed/random effects meta-analysis
m <- metagen(TE = log_OR, seTE = se_log_OR, studlab = study,
data = studies, sm = "OR", random = TRUE)
forest(m) # forest plot
funnel(m) # publication bias
metabias(m) # Egger's test
# Python alternative
import pymare
dataset = pymare.Dataset(y=effect_sizes, v=variances, names=study_names)
results = pymare.estimators.DerSimonianLaird().fit_dataset(dataset)
Reporting Guidelines Study Type Guideline Checklist RCT CONSORT 25 items + flow diagram Observational (cohort/case-control) STROBE 22 items Systematic Review PRISMA 27 items + flow diagram Diagnostic Accuracy STARD 30 items Quality Improvement SQUIRE 18 items Case Reports CARE 13 items
Literature Search
PubMed / MEDLINE # Boolean search
(("diabetes mellitus"[MeSH]) AND ("metformin"[MeSH]) AND ("randomized controlled trial"[pt]))
# Filters: humans, English, last 5 years, RCT
Systematic Review Workflow
Define PICO (Population, Intervention, Comparison, Outcome)
Search PubMed, Embase, Cochrane, Web of Science
Screen titles/abstracts → full text review
Extract data (standardized form)
Assess risk of bias (Cochrane RoB 2.0 or Newcastle-Ottawa)
Synthesize (narrative or meta-analysis)
GRADE certainty of evidence
Clinical Coding System Purpose Example ICD-10/11 Diagnosis classification E11.9 = T2DM without complications SNOMED CT Clinical terminology Concept hierarchy with relationships CPT Procedures (US billing) 99213 = office visit, established LOINC Lab/clinical observations 2345-7 = glucose in serum ATC Drug classification A10BA02 = metformin
Bias Checklist Bias What It Is Mitigation Selection Non-random group assignment Randomization, matching Information Measurement differs by group Blinding, standardized instruments Confounding Third variable causes both Adjustment, stratification, matching Attrition Differential dropout ITT analysis, sensitivity analysis Publication Positive results published more Funnel plot, register protocols Lead-time Early detection ≠ longer survival Use mortality, not survival from diagnosis Immortal time Misclassified person-time Proper time-zero, landmark analysis
Critical Appraisal Questions
Was the study question clearly defined (PICO)?
Was the study design appropriate for the question?
Was selection bias minimized?
Were outcomes measured validly and reliably?
Were confounders identified and controlled?
Was follow-up adequate?
Are the results clinically meaningful (not just statistically significant)?
Are the results generalizable to my patient population?
02
Study Types
Health Medical Research | Skills Pool