|
Single repair feature
|
First film is betterb (progression) (n = 143, 36%)
|
Second film is better (repair) (n = 254, 64%)
|
Odds ratio to detect repair
|
True positive rate (sensitivity)
|
False positive rate (1 – specificity)
|
Positive likelihood ratio
|
Negative likelihood ratio
|
|---|
|
Filling-in of erosions
|
112 (33%)*
|
225 (67%)**
|
2.2
|
0.89
|
0.78
|
1.1
|
0.50
|
|
Cortication
|
84 (30%)
|
192 (70%)
|
2.2
|
0.76
|
0.59
|
1.3
|
0.58
|
|
Sclerosis
|
68 (31%)
|
149 (69%)
|
1.6
|
0.59
|
0.47
|
1.1
|
0.78
|
|
Remodelling
|
32 (24%)
|
97 (76%)
|
2.1
|
0.38
|
0.22
|
1.7
|
0.79
|
|
Trabeculation
|
41 (37%)
|
78 (63%)
|
1.1
|
0.20
|
0.21
|
1.0
|
1.0
|
|
Reconstitution of a normal joint
|
41 (53%)
|
37 (47%)
|
0.43
|
0.15
|
0.29
|
0.5
|
1.2
|
|
Any of the above features of repair
|
130 (36%)
|
234 (64%)
|
1.2
|
0.92
|
0.91
|
1.0
|
0.89
|
- aTest performance in Exercise I of putative features of repair in relation to progression and repair as indicated by inferred assignment (first film is better versus second film is better). Of the total number of 397 observations, 143 (36%) were judged as showing progression and 254 (64%) as showing repair without taking into account specific features of repair. Adding information on features of repair only marginally influences the discrimination between progression and repair.
- bDesignation of first or second film based on the true sequence. Numbers indicate the numbers of observations in which a given single repair feature was recorded as present. Percentages indicate the positive predictive value of a specific repair feature for a progression* or repair** classification. For example, 'filling-in' was observed 325 times: 103 times (32%) in cases with progression and 222 times (68%) in cases of repair. The positive likelihood ratio is the quotient of the true positive rate divided by the false positive rate (for example, 'filling-in' in cases of repair) and the false positive rate (for example, 'filling-in' in cases of progression). The negative likelihood ratio is the quotient of the false negative rate (no 'filling-in' in cases of repair) divided by the true negative rate (no 'filling-in' in cases of progression). In order to be of diagnostic value, the positive likelihood ratio should be high (for example, > 4) and the negative likelihood ratio should be low (for example, < 0.3).
