6. Secondary analyses for alternative definitions of the target condition.
| Subgroup | Datasets | Participants (cases) |
Diagnostic odds ratio (DOR) (95% CI) |
Sensitivity (95% CI %) |
Specificity (95% CI %) |
Relative DOR (RDOR) (95% CI) |
P value (RDOR) | P valuea (hierarchical summary receiver‐operator curves (HSROC) models) |
| Differences between in‐person and image‐based evaluations | ||||||||
| Detection of invasive melanoma alone | ||||||||
| In‐person | 7 | 6857 (208) | 62.4 (17.6 to 222) | 86% (68 to 94) |
91% (81 to 96) |
4.21 (0.62 to 28.6) | 0.13 | 0.27 |
| Image‐based | 5 | 599 (150) | 14.8 (3.56 to 61.9) | 76% (50 to 91) |
83% (62 to 93) |
‐ | ‐ | ‐ |
| Detection of any skin lesion requiring excision | ||||||||
| In‐person | 7 | 8091 (2187) | 20.5 (7.11 to 59.3) | 81% (68 to 90) |
81% (56 to 93) |
1.70 (0.24 to 12.3) | 0.55 | 0.87 |
| Image‐based | 3 | 547 (138) | 11.9 (2.22 to 65.3) | 75% (49 to 90) |
79% (38 to 96) |
‐ | ‐ | ‐ |
aLikelihood ratio test assessing differences in both accuracy and threshold.