Table 7.

Ordinal Logistic Regression of 6-Month GOS-E on Age, GCS, and Quantitative Versus Qualitative CT Predictors Using Logit Link Function

	Outcome variable	Covariates	Significance	Cox and Snell R-squared	Nagelkerke R-squared
Quantitative CT model	6-month GOS-E	Componentsa 1−7 in Table 5	2×10−7	48.8%	50.1%
Qualitative CT model	6-month GOS-E	Componentsa 1−7 in Table 6	1×10−5	41.6%	42.6%

^aEach of the 7 covariates in each of the two ordinal logistic regression models above is one of the principal components shown in Tables 5 and 6.

Ordinal logistic regression of 6-month GOS-E on age, GCS, and quantitative versus qualitative computed tomography (qCT) predictors using logit link function. Thus, each covariate consists of the complete linear combination of age (in years), admission GCS, subdural hematoma volume (in cubic centimeters), subarachnoid/intraparenchymal hemorrhage volume (in cubic centimeters), epidural hemorrhage volume (in cubic centimeters), cistern effacement (ordinal variable from 1 to 3, corresponding to normal, partly-effaced, or severely-effaced cisterns), and midline shift (in millimeters), utilizing the exact coefficients for those variables as listed in Tables 5 and 6. With the use of quantitative rather than qualitative CT features, Nagelkerke R-squared improved to 50% from 43%. Thus, qCT features, age, and GCS account for approximately 50% of the variability in 6-month GOS-E score after acute head injury, compared to 43% when qualitative CT features, age, and GCS are used.

GOS-E, Extended Glasgow Outcome Score; GCS, Glasgow Coma Scale.