Examples of flawed cumulative incidence proportions (%) following an analysis of data with less than five injuries in a certain state based on a relative biweekly change in running distance (categorised into four states) and relative biweekly change in running intensity (categorised into four states)
| Biweekly change in running distance (states) | |||||
| Reg>10% | Reg 10%–0% | Prog 0%–10% | Prog>10% | ||
| Biweekly change in running intensity (states) | Reg>10% | 3.8% (5) | 1.7% (0) | −18.9% (0) | 13.9% (3) |
| Reg 10%–0% | 24.2% (16) | 6.8% (17) | 44.8% (8) | 12.3% (20) | |
| Prog 0%–10% | 10.3% (13) | 16.6% (11) | 25.3% (10) | 22.3% (21) | |
| Prog>10% | 18.0% (3) | 0.1% (0) | −7.6% (0) | 9.9% (4) | |
In reality, cumulative injury incidence proportions range between 0% and 100%. However, some proportions in the example are negative because too few injuries in that state lead to biased estimated.
Number in parentheses represents number of injuries in each exposure state. Results based on a supplementary analysis of the RUNCLEVER dataset.40
Reg, regression; Prog, progression.