| Prospective data collection (no recall bias) | Can be difficult to change study protocol, e.g. retrospectively add data, change outcome |
| Hypothesis driven | Still does not prove causation |
| Essential for intervention studies | Recruitment of representative population |
| Longitudinal analysis can reduce sample size | Drop out |
| Analysis can minimise effect of drop out | Often only considers one exposure |
| Able to consider different exposures at different times | Cost |
| Follow up for >10 years often necessary |
| Lack of gold standard outcome |