1. CASE CONTROL
   1. Start with OUTCOME variable (cases v/s controls)
   2. Trace back what risk factors they were exposed in PAST
   3. Quick, cheaper, few subjects
   4. Good for : RARE DISEASES ; LONG DURATION betn exposure and outcome or several exposures
   5. Good for DIAGNOSIS AND ETIOLOGY questions
   6. PROBLEMS: RECALL BIAS, NOT good for RARE EXPOSURES
2. COHORT
   1. Start with RISK EXPOSURE (exp v/s non-exp)
   2. F/U to see who has the OUTCOME
   3. Takes long, expensive, more subjects
   4. Good for: RARE EXPOSURES; temporal relatn, multiple outcomes
   5. GOod for: ETIOLOGY, HARM, PROGNOSIS, INCIDENCE ESTIMATE
   6. PROBLEMS: SELECTION BIAS, Confounding, diff. blinding, no randamization, NOT good for RARE OUTCOMES
3. CROSS-SEC SURVEY
   1. PREVALANCE ESTIMATE
   2. Cheaper, easier but needs large no.
   3. BIASES
      1. Volunteer bias/ Social desirability / Deviation/ Faking good and faking bad / Yes-sayers and No-sayers / +ve and -ve skew / end aversion or Central tendency bias
4. OBSERVATIONAL STUDIES
   1. DESCRIPTIVE
      1. SURVEY
      2. CASE REPORT
         1. ~hypothesis generation
      3. CASE SERIES
      4. ECOLOGICAL STUDY
         1. Study population or community
      5. QUALITATIVE STUDY
   2. ANALYTICAL
      1. CASE-CONTROL
      2. COHORT
      3. CROSS-SECTIONAL SURVEY
5. EXPERIMENTAL STUDIES
   1. OPEN
      1. no control group, cheap, easy
      2. The sample size of the trial is subject to achieving a desired effect size at the predefined alpha and beta conditions
      3. open-plan or closed-plan
   2. CONTROLLED
   3. PRAGMATIC
      1. reflective of everyday clinical practice, effectiveness
      2. diff to control, dif. to blind, drop-outs
   4. CROSSOVER
      1. Good for Rx of RARE DISEASES, Pt are their own controls (so study with small actual no. of pt)
      2. Problem: order effects, historical controls, carry-over effects
      3. PAIRED DATA are compared
   5. N OF 1
      1. Competing Rx are randomized to diff. time periods for each trial subject (usually 1-30 subjects)
      2. Pt to id subjective benefit, so good for CHRONIC SYMPTOMS
      3. CONFOUNDING: 1. Pathologies need to be relatively stable over time. 2. Rx should not directly alter disease process itself.
      4. HAWTHORNE AND CARRY OVER EFFECTS ARE COMMON
   6. CLUSTER RCTs
      1. Randomize Rx centres (e.g. GP practices)
      2. GOOD: when blinding clinicians is impractical (med v/s surgical Rx), any single centre can't switch betn Rx
      3. e.g. comparing diff. psychoRx, health edu or promotion prog
   7. RCT
   8. FACTORIAL DESIGN
      1. RCT design where two or more Rx FACTORS are compared simultaneously (e.g. 2 factors -->2! comparisons)
      2. Only relative comparisons can be made, unless a placebo gp is included