Skip to main content

Table 2 Comparisons of SROC curves for prediction of WS based on direct comparative studies

From: Accuracy of respiratory muscle assessments to predict weaning outcomes: a systematic review and comparative meta-analysis

Assessment comparison

Studies

(n)

Patients

(n)

 − 2Log Likelihood value of the fitted models

Likelihood ratio assessments

RDOR (95% CI)

Varied model (V)

Fixed shape model (FS)

Fixed accuracy model (FA)

Model comparison

Chi-square (df = 1)

p value

DE versus PImax

8

377

190.4

193.5

194.8

FS versus V

3.1

0.08

1.51 (0.39–5.84)

8

377

FA versus FS

1.3

0.25

DTF versus DE

28

2081

649.3

650.9

664.5

FS versus V

1.6

0.20

1.84 (1.25–2.70)

28

2081

FA versus FS

13.6

 < 0.01

DE versus Tdiee

5

258

110.4

111.3

111.3

FS versus V

0.9

0.34

0.74 (0.14–3.80)

5

258

FA versus FS

0

1

Pimax versus P0.1

4

265

83

84.3

84.7

FS versus V

1.3

0.25

0.59 (0.02–18.80)

4

265

FA versus FS

0.4

0.53

  1. Direct comparisons were made only using data from studies which compared each pair of assessments on the same patients. To compare SROC curves, the following HSROC models were compared: Model 1 (“Varied”, V) which includes covariates to allow accuracy, threshold and shape to vary by assessment; Model 2 (“Fixed shape”, FS) from which the covariate term for shape was removed, to assume that the SROC curves under comparison have the same shapes; Model 3 (“Fixed accuracy”, FA) from which also the covariate term for accuracy was removed to assume that the SROC curves under comparison have the same accuracy. Significance level: p < 0.05. Values of RDOR were calculated as relative diagnostic odds ratio of the first assessment compared to the second, as indicated by the order specified in the “Assessment comparison” column. A value of the RDOR higher or lower than 1 indicates that the first assessment has higher or lower accuracy than the second one. Confidence intervals of RDOR not containing 1 indicate significant higher or lower accuracy. Comparisons of the SROC curves were not performed for DTF versus Tdiee and DTF versus Tdiei because of the irregular shape of one of the curves and for DE versus Tdiei because model FS could not be fitted. Significant p-values (p<0.05) and RDOR 95%CI are highlighted in bold. Abbreviations: CI Confidence interval, DE Diaphragm excursion, DTF Diaphragm thickening fraction, PImax Maximal inspiratory pressure, P0.1 Airway occlusion pressure, Tdiee Diaphragm thickness at end-expiration, df Degrees of freedom which are equal to the difference in the number of parameters between the models that are compared, WS Weaning success. Bold value indicates statistically significant values
Back to article page

Critical Care

ISSN: 1364-8535