TABLE 4.

Comparison of the accuracy, including confidence intervals, of two predictive models of poor outcomes in oral language, reading fluency and reading comprehension

Prediction model	Outcome at 12	Criterion #	Sensitivity	Specificity	Positive predictive value	Negative predictive value	AUC (95% CI)
Number of risk factors	Oral language	≥3	0.47 (0.30–0.65)	0.81 (0.73–0.87)	0.37 (0.27–0.49)	0.86 (0.82–0.90)	0.75*a (0.66–0.84)
	Reading fluency	≥3	0.47 (0.30–0.65)	0.81 (0.74–0.863)	0.33 (0.23–0.44)	0.88 (0.85–0.91)	0.74*b (0.65–0.83)
	Reading comprehension	≥3	0.61 (0.43–0.76)	0.84 (0.78–0.90)	0.48 (0.37–59)	0.90 (0.86–0.93)	0.78*c (0.70–0.86)
Continuous measures of risk factors	Oral language	0.370	0.50 (0.32–0.68)	0.89 (0.83–0.94)	0.53 (0.39–0.67)	0.88 (0.84–0.91)	0.84*a (0.66–0.84)
	Reading fluency	0.339	0.53 (0.35–0.70)	0.91 (0.859–0.95)	0.55 (0.40–0.68)	0.91 (0.87–0.93)	0.77*b (0.68–0.87)
	Reading comprehension	0.340	0.55 (0.38–0.71)	0.89 (0.84–0.94)	0.55 (0.42–0.68)	0.89 (0.85–0.92)	0.83*c (0.75–0.90)

^# For the second model, the criterion is the critical value of the logistic residual ‘estimated probability of positive outcome; values ≥ the criterion are predictions of poor outcome.

^*All AUC statistics had asymptotic significance < 0.001. The AUC values for each pair of models (categorical vs. continuous risk) for the three outcomes were formally compared using a critical ratio z value. These indicated no significant differences: ^a z = 0.32 p = .626; ^bz = 0.10, p = .540; ^cz = 0.18, p = .571.