7. Analysis by observer qualifications for detection of invasive melanoma or atypical intraepidermal melanocytic variants.
| Test | Studies |
Lesions (cases) |
DOR (95% CI) |
Specificity at 80% sensitivity (95% CI) % |
Sensitivity at 80% specificity (95% CI) % |
RDOR (95% CI) |
P value (LR) |
| Qualifications: in‐person | |||||||
| Consultant expert* | 11 | 2767 (439) | 52.4 (21.6 to 127) |
94% (84 to 98) |
91% (83 to 96) |
1.00 (comparator) |
0.33 |
| Consultant* | 10 | 8390 (1015) |
97.7 (35.6 to 268) |
97% (90 to 99) |
95% (87 to 98) |
1.86 (0.949 to 7.11) |
|
| GP | 2 | 566 (37) |
19.2 (1.6 to 226) |
82% (19 to 99) |
82% (36 to 97) |
0.37 (0.03 to 5.08) |
|
| Resident/registrara | 2 | 11137 (127) |
51.6 (2.9 to 927) |
93% (42 to 100) |
93% (42 to 100) |
Not estimable within model |
‐ |
| Mixed (secondary care)a | 1 | 309 (46) |
29.6 (13.5 to 64.8) |
88% (77 to 94) |
88% (77 to 94) |
Not estimable within model |
‐ |
| Qualifications: image based | |||||||
| Consultant expert* | 33 | 8664 (1854) |
19.4 (13.1 to28.8) |
83% (76 to 88) |
83% (77 to 88) |
1.0 (comparator) |
< 0.001 |
| Consultant* | 25 | 4589 (955) |
11.9 (7.6 to 18.6) |
74% (65 to 82) |
75% (66 to 82) |
0.61 (0.40 to 0.92) |
|
| Resident* | 5 | 927 (138) |
6.0 (2.6 to 14.0) |
59% (37 to 78) |
61% (41 to 78) |
0.31 (0.14 to 0.71) |
|
| Mixed (other) | 4 | 867 (229) |
15.1 (4.0 to 57.0) |
79% (48 to 94) |
79% (not estimable) |
0.78 (0.20 to 3.1) |
|
| GP/Primary care | 3 | 288 (55) |
1.9 (0.7 to 5.0) |
30% (12 to 57) |
34% (51 to 93) |
0.10 (0.04 to 0.25) |
|
| Mixed (secondary care)b | 4 | 399 (111) |
10.3 (3.0 to 35.3) |
72% (43 to 90) |
72% (43 to 90) |
Not estimable within model |
‐ |
| Physician assistantb | 1 | 65 (25) |
3.6 (1.1 to 11.5) |
47% (22 to 74) |
47% (22 to 74) |
Not estimable within model |
‐ |
| CI: confidence interval; DOR: diagnostic odds ratio; LR: likelihood ratio test; NR: not reported; RDOR: relative diagnostic odds ratio | |||||||
*Consultants were usually dermatologists but could also be plastic surgeons or oncologists. aIn‐person model could not be fitted including the small number of studies in these groups. Estimates for these groups are obtained from computed the DOR for the individual study, or random effects meta‐analyses of DORs where there is more than one study. Estimates at the 80% sensitivity and specificity values are computed assuming symmetric SROC curves. bImage‐based model could not be fitted including the small number of studies in these groups. Estimates for these groups are obtained from computed the DOR for the individual study, or random effects meta‐analyses of DORs where there is more than one study. Estimates at the 80% sensitivity and specificity values are computed assuming symmetric SROC curves.