Dataset Search Guidelines

1
Dataset Search Guidelines

• Prospective Care

2
Components

• Sample size
• Outcome Variable
• Risk Factors (data types)
• Follow up Times
• Study Design Information (should be available)
• Sample Demographics

3
Sample Size

• Minimum standard should be that you need more
  cases than the smallest number of risk factors
  you'd consider in a single model. However, this
  minimum standard might not provide any useful
  predictions.
• Datasets with only clinical variables should have
  100s of samples.
• Datasets with genomic information should ideally
  have 100s of samples, although these are scarce
  200 samples could be considered a good size.
• The keys are (1) the bigger the better and (2)
  the sample should be large enough to capture the
  clinico-genomic variation in the population.

4
Outcome Variable

• Onset time for a particular phenotype measured
  from when the first risk factors were recorded.

5
Risk Factors

• Can be a mix of categorical and continuous
  variables
• Can include demographic information
• Need description of how data is coded

6
Follow up Time

• Ideally, the follow up times (with risk factor
  measurements) should not all be the same
• Closer follow up times (for each patient) provide
  more information for prediction
• Aim for a period of follow up which covers the
  time period of interest for prediction

7
Study Design Information

• A full description of the study design should be
  available
• Focus on single population studies or studies
  which are large enough that we can use either
  cases or controls

8
Sample Demographics

• The sample should be reflective of the population
  you are trying to model
• Mix of ages, races, etc.