Table 2.
Comparison of head acceleration event (HAE) exposure across six data cleaning methods
| Method | Impact number | Impact rate (95% CI) | Compared method | Impact rate ratio (95% CI) | ICC (95% CI) |
|---|---|---|---|---|---|
|
1 (Uncleaned) |
15719 | 67.75 (59.52-77.13) | 2 | 2.36 (2.29-2.43)* | 0.632 (0.381-0.786)* |
| 3 | 25.23 (23.29-27.33)* | 0.008 (− 0.205-0.242) | |||
| 4 | 37.43 (33.97-41.23)* | 0.004 (− 0.207-0.238) | |||
| 5 | 35.40 (32.22-38.91)* | 0.010 (− 0.202-0.244) | |||
| 6 | 96.44 (82.65-112.53)* | 0.001 (− 0.208-0.234) | |||
|
2 (time-windowed HAEs only) |
6666 | 28.73 (25.21-32.75) | 3 | 10.70 (9.86-11.62)* | 0.012 (− 0.205-0.249) |
| 4 | 15.87 (14.38-17.52)* | 0.009 (− 0.205-0.243) | |||
| 5 | 15.01 (13.64-16.53)* | 0.024 (− 0.191-0.259) | |||
| 6 | 40.90 (35.01-47.77* | 0.002 (− 0.208-0.234) | |||
|
3 (algorithm-classified HAEs without time windowing) |
623 | 2.69 (2.31-3.12) | 4 | 1.48 (1.31-1.68)* | 0.836 (0.612-0.921)* |
| 5 | 1.40 (1.24-1.58)* | 0.424 (0.175-0.623)* | |||
| 6 | 3.82 (3.22-4.54)* | 0.284 (− 0.002-0.525)* | |||
|
4 (algorithm-classified HAEs with time windowing) |
420 | 1.81 (1.54-2.13) | 5 | 0.95 (0.83-1.08) | 0.505 (0.271-0.684)* |
| 6 | 2.58 (2.15-3.09)* | 0.476 (0.142-0.693)* | |||
|
5 (video verification) |
444 | 1.91 (1.63-2.24) | 6 | 2.72 (2.28-3.26)* | 0.386 (0.003-0.645* |
|
6 (HAEs confirmed by both proprietary classification algorithm and video verification) |
163 | 0.70 (0.58-0.86) | |||
Impact count and impact rate per athlete exposure are presented for each data cleaning method. Impact rate ratios (Impact rate for method A divided by impact rate for method B and intraclass correlation coefficients (ICCs) with 95% confidence intervals CI were calculated to assess agreement between methods. ICCs were interpreted as follows: < 0.5 = poor agreement, 0.5–0.75 = moderate agreement, 0.75–0.90 = good agreement, > 0.90 = excellent agreement. *indicates a significant difference was observed. HAE head acceleration event