Introduction
Methods
Statistical analysis
Results
Baseline characteristics
Variable | All patients (n = 1625) | Development (n = 844) | Validation (n = 781) | P-value | Survivors (n = 1085) | Non-survivors (n = 540) | P-value |
---|---|---|---|---|---|---|---|
Age, years | 55 (38 to 66) | 56 (39–66) | 54 (36–66) | 0.288 | 52 (33–63) | 61 (49–80) | <0.001 |
GCS
| |||||||
3 to 6 | 828 (51) | 420 (50) | 408 (52) | 0.318 | 419 (39) | 409 (76) | <0.001 |
7 to 13 | 797 (49) | 424 (50) | 373 (48) | 666 (61) | 131 (24) | ||
APACHE II
| 22 (17 to 27) | 22 (17 to 22) | 22 (17 to 26) | 0.784 | 19 (15 to 23) | 27 (22 to 31) | <0.001 |
SAPS II
| 43 (31 to 55) | 43 (32 to 55) | 44 (31 to 55) | 0.988 | 38 (28 to 48) | 56 (45 to 63) | <0.001 |
SOFA
| 7 (5 to 10) | 7 (5 to 10) | 7 (5 to 10) | 0.744 | 7 (5 to 9) | 9 (7 to 11) | <0.001 |
Length of stay, days
| |||||||
ICU | 2 (1 to 5) | 2 (1 to 5) | 2 (1 to 5) | 0.989 | 3 (1 to 6) | 2 (1 to 4) | <0.001 |
Hospital | 6 (3 to 12) | 6 (3 to 12) | 6 (3 to 13) | 0.457 | 7 (4 to 15) | 4 (1 to 8) | <0.001 |
Mortality
| |||||||
ICU | 212 (13) | 107 (13) | 105 (13) | 0.647 | NA | 212 (39) | NA |
Hospital | 346 (21) | 173 (21) | 173 (22) | 0.416 | NA | 346 (64) | NA |
Six-month | 540 (33) | 278 (33) | 262 (34) | 0.795 | NA | 540 (100) | NA |
Age, years | Mortality, % (absolute numbers) | ||
---|---|---|---|
All patients (n = 1625) | GCS 7 to 13 (n = 797) | GCS 3 to 6 (n = 828) | |
<40 | 20 (86/438) | 2 (4/210) | 36 (82/228) |
40 to 49 | 27 (50/187) | 11 (11/97) | 43 (39/90) |
50 to 59 | 32 (116/363) | 10 (17/167) | 51 (99/196) |
60 to 69 | 42 (129/309) | 24 (36/149) | 58 (93/160) |
70 to 79 | 45 (104/232) | 31 (39/125) | 61 (65/107) |
≥80 | 57 (55/96) | 49 (24/49) | 66 (31/47) |
Hospital mortality performance
Six-month mortality performance
Performance variable | Discrimination | Calibration | Precision | ||
---|---|---|---|---|---|
AUC | 95% CI | H-L P-value | GiViTI P-value‡ | Brier score | |
Development cohort
| |||||
APACHE II | 0.81 | 0.78, 0.84 | 0.153 | NA | 0.160 |
SAPS II | 0.81 | 0.77, 0.84 | 0.343 | NA | 0.160 |
SOFA | 0.68 | 0.64, 0.72 | 0.282 | NA | 0.201 |
Adjusted SOFA* | 0.78 | 0.75, 0.81 | 0.444 | NA | 0.175 |
Reference† | 0.75 | 0.72, 0.78 | 0.144 | NA | 0.185 |
Validation cohort
| |||||
APACHE II | 0.79 | 0.75, 0.82 | 0.062 | 0.653 | 0.167 |
SAPS II | 0.80 | 0.77, 0.83 | 0.775 | 0.782 | 0.166 |
SOFA | 0.68 | 0.64, 0.72 | 0.691 | 0.710 | 0.201 |
Adjusted SOFA* | 0.79 | 0.76, 0.82 | 0.177 | 0.574 | 0.174 |
Reference† | 0.77 | 0.74, 0.80 | 0.086 | 0.072 | 0.181 |
Discussion
Key findings
Comparison with previous studies
Future implications
Limitations
Conclusion
Key messages
-
The APACHE II and SAPS II-based prediction models showed equally good prognostic performance in predicting 6-month mortality of ICU-treated patients with TBI.
-
The SOFA-based model displayed poor performance in 6-month mortality prediction. However, after the inclusion of age and the GCS, as separate variables, the performance improved significantly to match that of the APACHE II and SAPS II.
-
A simple prognostic model, including only age and GCS, also displayed fairly good prognostic performance in 6-month mortality prediction.
-
Forthcoming epidemiological studies lacking necessary data for the use of TBI-specific models may use the general ICU scoring systems APACHE II and SAPS II or the novel reference model for adequate case-mix adjustment.