Table 3B

Summarised results of the included non-RCT DA studies

StudySummaryIntervention or group
(number of lesions)
Outcome measures
Healthcare practitioner diagnosisExpert reviewSens/SpecPPV/NPVOthers
Non-RCT DA studies
Ahmadi et al 27 Retrospective cross-sectional study of medical files from 3 general practices.(580)67 malignant, 75 premalignant, 399 non-suspicious, 39 unknown. 16.7% of patients referred.151 lesions confirmed by HP/dermatology: 37 BCC, 4 MMs, 1 lentigo maligna, 20 unknown.PPV: benign lesions 85.7%, premalignant lesions 18.2%, malignant lesions 53.8%,
BCCs 53.3%, melanoma 25%.
Tools used: dermoscopy 8.4%, experienced PCP advice 1.4%, biopsy 1.9%, excision 10.3%
Bourne et al 28 Sequential design DA study. 4 PCPs used new BLINCK dermoscopy algorithm on 50 lesion images, compared with same PCPs using 3-point checklist, Menzies and clinical assessment on same 50 images.BLINCK (200)Found 33/36 MMs.
Biopsied 131/200 lesions.
Images of 50 lesions used: 1 invasive MM (0.52 mm), 8 in situ.Sens 90.8%
Spec 50%
DA 65.5%
Number-needed-to-excise 6
3PCL (200)Found 19/36 MMs.
Biopsied 105/200 lesions.
Sens 59.4%
Spec 42.2%
DA 48.3%
Number-needed-to-excise 11
Menzies (200)Found 16/36 MMs.
Biopsied 71/200 lesions.
Sens 54.7%
Spec 69%
DA 63.9%
Number-needed-to-excise 13
Clinical (200)Found 9/36 MMs.
Biopsied 74/200 lesions.
Sens 52.6%
Spec 74.8%
DA 65%
Number-needed-to-excise 22
Menzies et al 29 Sequential design DA study comparing the performance of SolarScan with that of clinicians with varying dermatology experience on 78 images of PSLs.Dermatologist (78)10.5Diagnosing MM:
Sens 81%, spec 60%, PPV 30, NPV 94
Decision to excise:
Sens 79%, spec 60%, PPV 29, NPV 93
PCP (78)
8Diagnosing MM:
Sens 62%, spec 63%, PPV 26, NPV 89
Decision to excise:
Sens 62%, spec 61%, PPV 25, NPV 89
SolarScan (78)11.1Diagnosing MM: sens 85%, spec 65%, PPV 32, NPV 96
Decision to excise: sens 85%, spec 65, PPV 32, NPV 96
SolarScan’s sensitivity was higher than PCPs but NS.
Pagnanelli et al 30 Sequential design DA study to assess if internet-based course suitable to teach dermoscopy to 16 clinicians with minimal dermoscopy experience looking at 20 images of PSLs.Control: pretraining (20)20 PSLs in test set: 6 MMs, 14 non-MMs.Pattern analysis:
Sens 67.7%, spec 76.3%
DA 49.8
k-intraobserver agreement 0.42
ABCD:
Sens 58%, spec 73.4%
DA 38.8
k-intraobserver agreement 0.31
7-point checklist:
Sens 100%, spec 67.5%
DA 60.2
k-intraobserver agreement 0.58
Menzies method:
Sens 80.4%, spec 72%
DA 53.5
k-intraobserver agreement 0.50
Intervention: post-training (20)Pattern analysis:
Sens 82%, spec 78.5%
DA 60.1
k-intraobserver agreement 0.58
ABCD:
Sens 78.4%, spec 79.6%
DA 56.6
k-intraobserver agreement 0.55
7-point checklist:
Sens 100%, spec 69.8%
DA 64.5
k-intraobserver agreement 0.61
Menzies method:
Sens 93.4%, spec 76%
DA 62.8
k-intraobserver agreement 0.66
Rogers et al 31 Sequential design DA study examining performance of new TADA when used by 120 clinicians of various specialties looking at 50 PSL images.(50)5641 lesion evaluations performed: 3034 malignant, 2607 non-suspicious.50 lesion images in test set, 23 benign, 27 malignant. Sens and spec calculated for malignant lesions.Dermatologists:
Sens 94.8%, spec 78.5%
Non-dermatologists:
Sens 93.7%, spec 72.1%
>1 year dermoscopy experience:
Sens 95.4%, spec 77.3%
<1 year dermoscopy experience:
Sens 91.3%, spec 74.2%
Rogers et al 32 Sequential design DA study comparing performance of new TADA with existing dermoscopy algorithms when used by 120 clinicians of various specialties.(50)No data for 3-point checklist and AC rule.
TADA: 5646 lesions evaluated.
2056 deemed non-suspicious (1891 true negatives (92.0%), 165 false negatives (8.0%)).
3590 deemed malignant (2871 true positives (80.0%), 719 false positives (20.0%)).
50 lesion images in test set, 23 benign, 27 malignant. Sens and spec based on 40 non-PSLs.TADA:
Sens 94%, spec 75.5%
Sens for MM with TADA 94%. Spec for untrained clinicians for benign PSLs using TADA 76%–94% (beginners can be quickly trained to identify benign lesions).
AC:
Sens 88.6%, spec 78.7%
3-point checklist:
Sens 71.9%, spec 81.4%
Untrained (using TADA): Sens 93.6%, spec 69%
Trained (using TADA):
Sens 95.4%, spec 73.2%
Rosendahl et al 25 Prospective cohort study using SCARD to assess impact of dermoscopy use and subspecialisation on MM diagnosis by PCPs.Dedicated skin cancer practitionersNumber of lesions seen not recorded on SCARD.
11 992 lesions referred.
11.7% were MM.MM number needed to treat: 8.5
PCP with special interest in skin cancer1942 lesions referred.10.6% MM.MM number needed to treat: 9.4
PCPs1942 lesions referred.5.9% MM.MM number needed to treat: 17.0
High dermoscopy use17 917 lesions referred.11.2% MM.MM number needed to treat: 8.9
Medium use2657 lesions referred.9.1% MM.MM number needed to treat: 10.9
Low use1093 lesions referred.6.9% MM.MM number needed to treat: 14.6
(p<0.0001, but NS when adjusted for subspecialisation)
Secker et al 44 Sequential design DA study comparing performance of 293 PCPs in diagnosing PSLs before and after a training intervention.Pretest (clinical images, no education)
(20)
20 PSL images in test set: 3 MM, 2 BCC, 15 benign.For MM:
Sens 0.49%, spec 0.75%
% correct treatment: (1) malignant 85.87, (2) naevi 92.83, (3) benign 9.56
Post-test (clinical images with education)
(20)
Sens 0.50%, spec 0.77%(1) malignant 84.03, (2) naevi 95.61, (3) benign 7.81
Integrated
post-test (clinical and dermoscopic images with education)
(20)
Sens 0.66%, spec 0.70%
  • (1) malignant 91.74,

  • (2) naevi 92.35, (3) benign 29.35

OverallTraining improved sens and spec for all except pigmented naevi. Training improved DA for all PSLs except naevi.Increase in correct treatments for benign lesions, reduced unnecessary referrals and excisions.
Westerhoff et al 33 Intervention study assessing performance of 74 PCPs with no dermoscopy experience in diagnosing 100 PSLs using macroscopic images ± dermoscopy images before and after an educational intervention.Control:
no education
(100)
100 images of lesions used: 50 invasive MMs and 50 non-melanomas.Macroscopic images:
Pretest:
Sens 50.6%, spec 55.2%
Post-test:
Sens 53.7%, spec 51.5%
Significant improvement with training between pretest (54.6%) and post-test (62.7%) (p=0.007) on macro images. Diagnosis of MM with dermoscopy significantly better (75.9%) than macro images (62.7%)(p=0.000007). No significant difference adding dermoscopic images in diagnosing non-MM PSLs.
+Dermoscopic images:
Pretest:
Sens 52.9%, spec 58.1%
Post-test:
Sens 54.8%, spec 55.8%
Intervention:
with education
(100)
Macroscopic images:
Pretest:
Sens 54.6%, spec 53.0%
Post-test:
Sens 62.7%, spec 53.6%
+Dermoscopic images:
Pretest:
Sens 57.8%, spec 55.5%
Post-test:
Sens 75.9%, spec 57.8%
Börve et al 34 Case–control study. Smartphone TDS system in 20 primary healthcare centres compared with traditional, paper-based referral system from other primary healthcare centres.Control:
paper-based referrals
(746)
746 suspicious lesions referred.323 malignant (13 MM, 7 MM in situ, 22 SCCs, 115 BCCs, 164 AKs), 423 benign.3/4 invasive MMs given medium/low priority, 3/5 MMs in situ given low priority. Mean response time 5 days (range 0–82 days). Patients received primary treatment on single face-to-face visit in 82.2% of cases.
Intervention:
smartphone TDS referrals
(816)
816 suspicious lesions referred. 346 (42%) non-suspicious, final diagnosis benign for 343 (3 malignant lesions missed were AKs). 196 deemed malignant, 146 (74%) also malignant after dermatology/HP.229 malignant (19 MM, 16 MM in situ, 24 SCCs, 109 BCCs, 61 AKs), 587 benign.All invasive MMs prioritised correctly (high), all MM in situ at least medium priority. 22.6% more referrals given low priority. Mean response time 109 min (range 2 min to 46 hours). Waiting time for surgical treatment for MM significantly shorter (p<0.0001). Patients received primary treatment on single face-to-face visit in 93.4% of cases.
Grimaldi et al 35 Sequential design DA study assessing PCP diagnosis of suspicious PSLs before and after dermoscopic evaluation and accuracy of teledermatology triage system.PCP clinical
(235)
167 lesions non-suspicious, 68 suspicious.16 malignant (5 MMs), 219 benign.Dermoscopy by PCPs and then experts led to 76.5% reduction in number of surgical procedures (68 to 16).PCP clinical vs PCP dermoscopy: p<0.001, OR 0.345731
PCP
dermoscopy
(235)
206 non-suspicious, 29 suspicious (dermoscopy-corrected diagnosis in 57.3% of cases, only 1 false negative).PCP clinical vs expert dermoscopy: p<0.001, OR 0.179425
TDS
(235)
219 non-suspicious, 16 suspicious (1 false negative from 206 benign lesions).PCP dermoscopy vs expert dermoscopy: p<0.05, OR 0.518973
Livingstone and Solomon41 Prospective case series to assess cost-effectiveness, accuracy and patient satisfaction of a TDS system for non-malignant PSLs in a primary care practice.(248)248 patients that PCP would have been referred routinely to dermatology referred to TDS service. 102 needed face-to-face dermatology review. 146 advised on treatment. 3 lesions possibly malignant so referred 2-week-wait pathway.0/3 possibly malignant lesions were malignant at face-to-face review. None of other 245 lesions were malignant after review or follow-up.Waiting time for images to be taken (weeks): 0–1=27%, 1–2=45%, 2–3=7%, 3–4=0%, 4–5=7%, 5–6=7%.
Waiting time for results (weeks): 0–1=14%, 1–2=61%, 2–3=14%, 3–4=7%.
129 patients returned patient satisfaction questionnaires. 100% said TDS service was explained, 100% would recommend TDS.
Moreno-Ramirez et al 36 Sequential design DA study to assess if teledermatology with dermoscopy images would improve the current teledermatology-based triage system in referrals from a primary care centre.Control: teledermatology referrals
(61)
4 BCCs, 1 MM, 0 dysplastic naevus, 56 benign.
Referral rates 47.5% (29).
Rate of referral of true positive results 17.2% (5 true positives/29 referrals).
False positives 58.7% (17/29 referrals).
HP: 2 BCCs, 1 MM, 1 dysplastic naevus, 57 benign.Sens 1 (as 0 false negative), spec 0.65, false positive rate 0.35Clinical picture quality excellent 41%, poor 3.3%. Average diagnostic confidence 4.14/5. Agreement with histology 0.91.
Intervention: TDS referrals
(61)
2 BCCs, 1 MM, 1 dysplastic naevus, 54 benign.
Referral rates 39.3% (24) (p<0.05).
Rate of referral of true positive results 20.8% (5 true positives/24 referrals) (p<0.05).
False positives 41.7% (10/24) (p<0.05).
Sens 1 (as 0 false negatives), spec 0.78 (p<0.05), false positive rate 0.22 (p<0.05)Dermoscopic picture quality excellent 63.9%, poor 6.6%. Average diagnostic confidence 4.75/5 (p<0.05). Agreement with histology 0.94.
van der Heijden et al 37 Cohort study assessing accuracy and reliability of TDS diagnosis with images taken by PCPs compared with diagnosis at face-to-face consultations for same lesions.Control: face-to-face assessment
(76)
All 108 lesions also referred for face-to-face assessment. 76 lesions seen face-to-face by dermatology. 32 not seen as did not attend, moved away, GP did excision.Agreement face-to-face vs HP diagnosis k=0.90 (almost perfect), diagnostic agreement k=0.56–0.78 (substantial), management agreement k=0.31–0.38 (fair).
Agreement TDS vs face-to-face diagnosis k=0.55–0.73 (moderate-substantial), TDS vs face-to-face management k=0.19–0.29 (fair). Image quality: 36% bad, 36% good. TDS consultations with good image quality had better agreement, TDS vs face-to-face diagnosis (k=053–0.77, substantial), and TDS vs face-to-face management (k=0.34–0.47, fair-moderate).
Intervention: TDS referrals
(108)
108 lesions referred via TDS.HP diagnosis for 36 lesions (33%). 2 MMs and 5 non-melanoma skin cancers.Agreement TDS vs HP diagnosis k=0.41–0.63 (moderate). TDS consultations with good image quality had better agreement of TDS vs HP diagnosis (k=0.53–1.0, moderate-almost perfect).