Table 3.

Cox proportional hazards models in a subgroup analysis.

Model #	Dependent variable	Independent variable	HR	HR, 95% CI	P independent variable	Log rank test (robust) p-value	Wald scorep-value
1	Intracranial lesion	Log10-S100B	4.01	1.78–9.02	<0.001	0.03	<0.001
2	ICH	Log10-S100B	7.05	2.54–19.56	<0.001	0.03	<0.001
3	Ischemia	Log10-S100B	3.72	0.99–14.05	0.052	0.11	0.052

Patients included in the subgroup analyses had all undergone a CT scan following a S100B peak. Similarly, to Table 2, three different models with different dependent variables are shown. In all analyses, log10-transformed S100B was the independent variable. Overall, Model #1–2 were significant, which was assessed using the Robust Log Rank Test and the Wald Score, since these do not assume independence of clustered observations. Since Model #3 was not significant, the results of this model are non-interpretable. S100B emanated as a significant predictor in Model #1–2. The interpretation of the HR in the case of a continuous variable, e.g., S100B, is that a one-unit increase was associated with a 4 times increased risk for any intracranial lesion, and a 7 times increased risk for ICH. CI, confidence interval; ICH, intracranial hemorrhage; HR, hazard ratio.