Table 6.

Table 6 Prediction of Failure (C grade) at SHIS

Failure Prediction Model
Omnibus Test	χ2	df	p	Nagelkerke R2
∗ Significance for Spearman Rho Coefficient (Distribution of probability of C is non-normal).
Statistic	12.70	2	0.002	0.31

Logistic Regression Equation Statistics
Source	B	SE	Wald	df	p	OR	OR 95% CI
							Lower	Upper
UGPA	−1.739	1.069	2.65	1	0.104	0.176	0.022	1.429
URMS	2.292	0.878	6.82	1	0.009	9.90	1.77	55.31
Constant	3.149	3.329	0.90	1	0.344	23.32	-	-

Classification Tables (Failure Decision Threshold = 0.15.)
Model Development (n = 65)				Model Cross Validation (n = 49)
Failure	Predicted			Failure	Predicted
Observed	No C	C	sp % Se %	Observed	No C	C	sp % Se %
No C	46	9	83.6	No C	37	9	80.4
C	4	6	60.0	C	0	3	100.0
NPV, PPV, tot. %	92.0	40.0	80.0	NPV, PPV, tot. %	100.0	25.0	81.6

Predicted Versus Observed Failure Measures of Association
Association	R	p	r2	Association	R	p	r2
Prob. C versus obs C	0.50	0.001 ∗	0.25	Prob. C versus obs C	0.55	0.008 ∗	0.30
Pred. C versus obs C	0.37	0.002	0.14	Pred. C versus obs C	0.45	0.001	0.20
OR pred C/not pred C 95% CI		7.67 (1.79.32.78)		OR pred C/not pred C		Undef.

CI = confidence interval; NPV = negative predictive value; OR = odds ratio; PPV = positive predictive value; SE = standard error; SHIS = School of Health Information Science, UT; UGPA = undergraduate grade point average; URMS = underrepresented minority status;

Note: B = Logistic Regression Coefficient; SE = Standard Error; Wald = Wald statistic; OR = odds ratio; df = degrees of freedom. The row headed NPV/PPV % shows, as percents, negative predictive values in columns labeled “No C” and positive predictive values in columns labeled “C”. Columns headed with sp % displayed above Se % show, as percents, specificity in the row labeled “No C”, above sensitivity in the row labeled “C”, above overall accuracy of prediction in the row labeled NPV/PPV %' Prob. C = predicted probability of “C” grade; Obs C = observed binary occurrence of “C” grade; pred C = predicted binary occurrence of “C” grade.