Demographic characteristics
Following the flow chart (Figure 1), 53 cases and 144 cases were enrolled into the favorable outcome group and unfavorable outcome group respectively. No significant differences were observed in aspects of age, gender, BMI, diabetes, hypertension, hyperlipidemia, diastolic pressure, location and side of hematoma, residual hematoma volume, operation time and type between the two groups (Table 1).
Table 1
Clinical Summary of Patients’ Data
Characteristics | Favorable outcome group (N=53) | Unfavorable outcome group (N=144) | P |
Age(year) | 63.98±7.99 | 61.28±11.94 | 0.071 |
Sex | | | 0.506 |
Male | 36(67.9%) | 89(61.8%) | |
Female | 17(32.1%) | 55(38.2%) | |
Body mass index(kg/m2) | 25.47±2.69 | 26.10±4.25 | 0.227 |
Diabetes (n, %) | 39(73.6%) | 123(85.4%) | 0.061 |
Hypertension (n, %) | 49(92.5%) | 138(95.8%) | 0.463 |
Hyperlipidemia (n, %) | 6(11.3%) | 13(9.0%) | 0.786 |
SBP (mmHg) | 152.92±28.65 | 170.29±32.24 | 0.001 |
DBP (mmHg) | 88.17±15.73 | 93.54±18.95 | 0.067 |
GCS | 10.47±5.01 | 6.87±3.62 | <0.001 |
Location | | | 0.066 |
Thalamus, Basal ganglia | 23(43.4%) | 89(61.8%) | |
Frontal, temporal, parietal,occipital lobe | 25(47.2%) | 47(32.6%) | |
Cerebellum | 5(9.4%) | 6(4.2%) | |
Brainstem | 0 | 2(1.4%) | |
Affected side | | | 0.525 |
Left | 30(56.6%) | 74(51.4%) | |
Right | 23(43.4%) | 70(48.6%) | |
Midline shift (mm) | 4.11±3.71 | 7.68±5.03 | <0.001 |
Hematoma volume (ml) | 43.29±27.63 | 60.74±36.34 | 0.002 |
Residual Hematoma volume (ml) | 9.52±10.38 | 11.19±14.29 | 0.453 |
Operation type | | | |
Craniotomy surgery | 41(77.4%) | 104(72.7%) | |
Minimally invasive surgery | 12(22.6%) | 40(27.8%) | |
Time to operation room (h) | 26.18±21.95 | 31.38±48.56 | 0.304 |
Time to operation room group | | | <0.001 |
<=12h | 11(20.75%) | 97(67.36%) | |
>12h, <36h | 34(64.15%) | 16(11.11%) | |
>=36h | 8(15.09%) | 31(21.53%) | |
Operation time (min) | 152.26±56.84 | 164.64±71.15 | 0.256 |
In the favorable outcome group, the mean value of SBP was 152.92±28.65 mmHg, which was significantly higher than that in unfavorable outcome group (p=0.001). The mean value of GCS was 10.47±5.01 for those who suffered favorable outcome, which was significantly higher than that in the unfavorable outcome group (p<0.001). In favorable outcome group, the mean value of midline shift and hematoma volume was 4.11±3.71 mm and 43.29±27.63 ml respectively, which was significantly lower than that in the unfavorable outcome group (p<0.01). For the time to operation room (TOR), there was no significant difference between two groups, while we divided the cases into three groups based on the TOR. We found that the cases who underwent surgery at >12h and <36h in the favorable outcome group were more than that in the unfavorable outcome group (p<0.001).
The linear relationship between time to operation room and outcome
The smoothing spline was used to analyze the relationship between time to operation room and outcome after adjusting of GCS; Sex; Age; BMI; Diabetes; Hypertension; DBP; SBP; Midline shift; Hematoma Volume; Hyperlipidemia; Location; Affected side; Residual Hematoma volume; Operation type and time. A linear relationship between time to operation room and outcome was shown in Figure 2 (p<0.001). The red points and blue points expressed the fitting spline and the 95% confidence intervals respectively. When the time to operation room was less than 21h, the occurrence of unfavorable outcome decreased with the time going (OR=0.8, p < 0.001). When the time to operation room was more than 21h, the occurrence of unfavorable outcome increased with the time going (OR=1.3, p < 0.001) (Table2).
Table 2
Threshold effect analysis of time to operation room (TOR) on unfavorable outcome using piece-wise linear regression
Inflection points of TOR | Crude OR (95%CI) p-value | *Adjusted OR (95%CI) p-value |
<21h | 0.8 (0.7, 0.8) <0.001 | 0.8 (0.7, 0.9) <0.001 |
≥21h | 1.0 (1.0, 1.1) <0.001 | 1.3 (1.2, 1.5) <0.001 |
Crude: no adjustment. |
*Adjusted: adjusted for GCS; Sex; Age; BMI; Diabetes; Hypertension; DBP; SBP; Midline shift; Hematoma Volume; Hyperlipidemia; Location; Affected side; Residual Hematoma volume; Operation type; Operation time |
Interaction and stratified analyses
The results of the stratified analyses of the association between TOR and outcome are presented in Supplementary table1 and Figure 3. The association between TOR and outcome remained generally consistent across several clinically relevant characteristics. The stratified analysis demonstrated that TOR (<21h) was associated with decreased risk of unfavorable outcome group in both age groups (<65 years: OR 0.5; ≥65 years: OR 0.2), sex (male: OR 0.3; female: OR 0.5), BMI (<2.5 kg/m2: OR 0.2; ≥25 kg/m2: OR 0.6) and diabetes (NO: OR 0.3; YES: OR 0.2), and GCS (<8: OR 0.5; ≥8: OR 0.7), location of hematoma (Thalamus, Basal ganglia: OR 0.5; Frontal, temporal, parietal,occipital lobe: OR 0.3; Cerebellum: OR 0.1), affected side of hematoma (left: OR 0.3; right: OR 0.5), midline shift (<6.5mm: OR 0.3; ≥6.5mm: OR 0.6) and Residual Hematoma volume (<6ml: OR 0.4; ≥6ml: OR 0.4), Operation time (<155min: OR 0.2; ≥155min: OR 0.6). The positive association between TOR (<21h) and the occurrence of unfavorable outcome group was observed in participants without hyperlipidemia (OR: 1.1) than in those with hyperlipidemia (OR: 0.2; P-interaction = 0.0407) (Figure 3), while the interaction analysis showed there was no significant difference (P-interaction = 0.2355). A significantly stronger positive association between TOR (<21h) and the occurrence of unfavorable outcome group was observed in participants with hematoma volume ≥ 48.6 ml (OR: 1.6) than in those with hematoma volume < 48.6 ml (OR: 0.2; P-interaction = 0.0099) (Figure 3). Moreover, a significantly stronger positive association between TOR (<21h) and the occurrence of unfavorable outcome group was observed in participants with minimally invasive surgery (OR: 1.1) than in those with craniotomy surgery (OR: 0.2; P-interaction = 0.0407) (Figure 3).
Association between variables and infection in different models
The multivariate logistic regression was used to analyze the association between SBP, GCS, midline shift, hematoma volume, time to operation room and unfavorable outcome (Table 3). Firstly, we found the tolerance was >0.1 and VIF was <10 for the predictors, suggesting no collinearity among these independent variables (Supplement
Table 3
Relationship between variables and unfavorable outcome in different models
Variables | Crude Model | | Adjust I Model | | Adjust II Model |
OR (95%CI) | p | OR (95%CI) | p | OR (95%CI) | p |
SBP | 1.0 (1.0, 1.0) | <0.001 | | 1.0 (1.0, 1.0) | <0.001 | | 1.1 (1.0, 1.3) | 0.006 |
GCS | 0.9 (0.8, 1.0) | 0.055 | | 0.9 (0.8, 1.0) | 0.062 | | 0.9 (0.8, 1.1) | 0.041 |
Midline shift | 1.3 (1.1, 1.4) | <0.001 | | 1.2 (1.1, 1.4) | 0.001 | | 1.3 (1.1, 1.6) | <0.001 |
Hematoma volume | 1.0 (1.0, 1.0) | 0.502 | | 1.0 (1.0, 1.0) | 0.856 | | 1.2 (1.0, 1.4) | 0.049 |
Time to operation room | | | | | | | | |
<12h | Reference | | | Reference | | | Reference | |
>=12h, <36h | 0.0 (0.0, 0.1) | <0.001 | | 0.0 (0.0, 0.1) | <0.001 | | 0.1 (0.0, 0.2) | <0.001 |
>=36h | 0.3 (0.1, 1.1) | 0.075 | | 0.3 (0.1, 1.1) | 0.073 | | 0.6 (0.0, 0.8) | 0.019 |
Crude model adjust for: None |
Adjust I model adjust for: sex; age |
Adjust II model adjust for: SEX; AGE; BMI; DIA; HYPE; DBP; HYPERLI; LOCATION; SIDE; REMAINS; OPERATION; OPERATION.TIME |
Table 2)
In the crude and adjust I (adjusting for age and sex) models, SBP, midline shift and time to operation room (>=12h, <36h) were all considered to be significant predictors of unfavorable outcome (Crude, OR 1.0, 1.3 and 0.0 respectively; Adjust I, OR 1.0, 1.2 and 0.0 respectively).
In the adjust II (adjusting for sex; age; BMI; Diabetes; Hypertension; DBP; Hyperlipidemia; location and side of hematoma; Residual Hematoma volume; Operation type and time) model, the SBP, GCS, midline shift, hematoma volume and time to operation room (>=12h, <36h) were all considered to be significant predictors of unfavorable outcome (OR 1.1, 0.9, 1.3, 1.2, and 0.1 respectively). The results of adjust II model showed that the occurrence of unfavorable outcome was positively related with the SBP, Midline shift and hematoma volume, and negatively related with the GCS and time to operation room (>=12h, <36h) (Table3).
Development of nomogram and clinical validation
We developed two models to predict the rate of unfavorable outcome. Model I: logit (unfavorable outcome) = -2.59256 +0.02880*SBP -0.07851*GCS +0.23582*Midline shift -0.00575*Hematoma volume -3.15442*(Time to operation, >=12h, <36h) -1.12692*(Time to operation, >=36h). Model II: logit (unfavorable outcome) = -1.67113 +0.01975*SBP -0.15566*GCS +0.18055*Midline shift -0.00368*Hematoma volume. The ROC analysis showed that the AUC of model I and model II were 0.899 and 0.812 respectively (Figure4a). The decision curves showed the model I had more benefits than the All or None scheme if the threshold probability is >15% and <100%, for the model II, the threshold probability is >45% and <100% (Figure4b). The calibration curves showed that the predicted outcome fitted well to the observed outcome in both model I and model II (p=0.129 and 0.317 respectively, Figure 5). The ROC analysis, decision curves and calibration curves showed that the model I is more reliable than model II. Consequently, we developed a nomogram (Figure6) to generate the probability of unfavorable outcome based on the model I. The AUC, accuracy, specificity, and sensitivity of the nomogram were 0.90, 0.87, 0.72, 0.93 respectively (Table4). Using the bootstrapping validation, the AUC for the nomogram was confirmed to be 0.894, which indicated favorable discrimination.
Table 4
Items | AUC | Accuracy | Specificity | Sensitivity | PLR | NLR | DOR |
Model I | 0.90 | 0.87 | 0.72 | 0.93 | 3.29 | 0.097 | 33.95 |
Model II | 0.81 | 0.74 | 0.79 | 0.72 | 3.45 | 0.359 | 9.59 |
AUC: Area Under the Curve |
PLR: positive likelihood ratio |
NLR: negative likelihood ratio |
DOR: Diagnostic Odds Ratio |