Sources of data for a decision analysis problem

Sources of data	Comments
Randomized controlled trials (RCTs)	Excellent source of data on effects of therapies. Often the study requirements (entry and exclusion criteria) will make the results of the trial applicable to only a narrow range of patients.
Cohort studies	Useful for risk factor determination, natural history.
Administrative databases	Excellent sources for broad population-based estimates of disease; often more accurate as estimate of effectiveness for various procedures than RCTs, which, in general, estimate efficacy. However, it is difficult to be sure that confounding variables are controlled for.
Electronic health records (EHRs)	Provides additional detail above what administrative databases are able to provide. Very useful for determining the distribution of a particular variable in different diseases.
Meta-analyses/systematic reviews	Provides summary measures over several different studies for a particular parameter. Generally restricted to RCTs, so often global population data are not well-represented.
Expert opinions	Often plagued by biases and dependent upon the method used to assess opinion. Most closely resembles the methods used to make clinical decisions in current clinical practice (use of consultants, etc); therefore, the use of expert opinion has face validity.
Investigator estimates	A last resort, if there are no other good sources of data. When combined with appropriate sensitivity analysis, can be a useful first start.

Graphic 70046 Version 2.0