Epidemiology: 14) Control Selection: Primary Study Base

Front 1

Fixed accrual and unmatched controls

Back 1

Common to assume all persons in the study base have the same study window - analgous to fixed/defined followup design

Front 2

5 - Procedures for using a primary study base

Back 2

1) Identify every member of the study base. 2) Identify each of the M1 cases. 3) Randomly sample k*M1 controls from non-cases (k is number of controls per case). 4) Ascertain covariates. 5)Ordinary logistic regression

Front 3

Why Match

Back 3

It can improve statistical efficiency and control for confounding

Front 4

Dangers of matching

Back 4

Over-matchind and unmatched analysis can reduce odds or statistical efficiency

Front 5

4 steps to category/frequency matching

Back 5

1) Study base divided into mutually exclusive/exhaustive strata. 2) For each case a stratus of the case is identified. 3) k controls are selected by random sampling (each non-case used once). 4) Repeated for all cases

Front 6

Individual Matching

Back 6

each case matched on factors associated with difficult-to-define factors. Use conditional logistic regression

Front 7

overmatching

Back 7

factors that are effects of exposure on on the causal pathway should NOT be used for matching

Front 8

4 advantages of matching

Back 8

1) Control confounders, more efficient than in analysis. 2) Statistical efficiency if true confounder (5 - 15% variance reduction) 3) Reduce information bias. 4)Feasible way to identify controls

Front 9

5 disadvantaged to matching

Back 9

1) cost. 2) exclusion of unmatches cases. 3) Cannot study effect of matching variable. 4) Reduced stat efficiency if exposure variation is reduced.
5) More comples stats analysis

Front 10

Variable Accrural

Back 10

used for varaibale time window or accrural period. AKA - Incidence, Incidence density sampling, risk set and risk set sampling

Front 11

5 Procedures for variable accrural

Back 11

1) Identify every member of study base. 2) Identify To (begninning) and T1 (end) of study eligibility. For cases, index date ends eligibility. 3) Process the cases in order of the index dates. 4) Exposure, covariates and effect modifiers are measured at index date. 5) Conditional logisitic regression

Front 12

3 steps to process cases in variable accurual

Back 12

1) At date Tx - identify all persons who are still eligible (risk set). 2) Randomly select k controls from risk set. 3) Set index date of controls to Tx

Front 13

Unique componant to variable accrural

Back 13

A subject who was a control can become a case. The same control can be selected for more than one case.

Front 14

Number of controls

Back 14

Depending on cost. Usuall 4 - 5 per case. Do power calculations.