SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Most education campaigns for melanoma (MM) detection in the general population have used the “ABCD” algorithm, although recognition of objects in the real life is based on a holistic image recognition rather than on analytic criteria. The objective was to compare analytic (ABCD) and cognitive (photographs) strategies for teaching self-recognition of MM. A prospective 4-arm stratified randomized trial in 255 individuals compared 3 realistic educative interventions by leaflets: 1) ABCD algorithm (“ABCD”), 2) a set of photographs chosen to stimulate recognition of MM among benign pigmented lesions (“Cog”), 3) photographs + explanations (“Cog-Ex” arm) and 4) no intervention (“NI”). A 40-slides test was performed before intervention (T0), 1 week after (T1) and after induction of anxiety (T2). In the “ABCD” arm, sensitivity slightly improved (80 to 83.8%, p = 0.04), but specificity dropped from 65.1 to 56.3% (p < 0.001), with no benefit in accuracy as compared to “NI”. In “Cog”arm, there was no change in sensitivity, but a strong increase in specificity (65.9 to 81.1%, p < 0.001) and accuracy (42.1 to 53.1%, p < 0.001). “Cog-ex” resulted in similar although lower benefit. Under stress (T2), there was a dramatic loss of specificity and accuracy in “ABCD”arm (65.1 to 44.1%, p < 0.001 and 40.8% to 35.8%, p ≤ 0.001) without higher gain in sensitivity, while sensitivity and accuracy increased (p < 0.005) after “Cog” leaflet, without decreasing specificity. Finally, the “ABCD” message alone does not seem efficacious and is even worse in the context of anxiety, whereas a quick look at a few photographs is sufficient to improve the ability of the laymen to recognize a MM just by optimizing their spontaneous image recognition capacities. Education by photographs is a realistic strategy which should replace or complete “ABCD” message in the campaigns for self-detection of MM. © 2005 Wiley-Liss, Inc.

The campaigns for self-examination of the skin and early diagnosis of melanoma are certainly one of the most cost-effective methods to improve melanoma prognosis in the general population1, 2 since patients are responsible for most of the delay in melanoma diagnosis.3 A major condition for the efficacy of cancer detection campaigns is to provide the population with a reasonably accurate detection tool. As to melanoma, the tool is an educational message that helps people to recognize the pigmented lesions that require medical advice on their own skin as well as on the skin of their relatives. A message responsible for a too-low sensitivity in melanoma detection would not be acceptable since no benefit on mortality could be expected from a late diagnosis. Conversely, owing to the very small proportion of melanoma among benign pigmented nevi, a too-low specificity would result in a huge and costly number of useless consultations. The choice of the best educative message is thus crucial. Most campaigns over the world have used morphological information based on analytic algorithms such as the ABCD rule4 ± E(evolution) or other scoring systems.5, 6 These algorithms have never been prospectively tested in the general population, but they may have a low specificity in the hands of laymen since many seborrheic keratoses and atypical nevi often fulfill the ABCD criteria. Furthermore, this criteria-based identification is somewhat artificial and does not correspond to the natural process by which we learn to recognize any object class around us. We can identify objects, although no teacher has provided us with any list of criteria. We unconsciously build our own recognition pattern from the images of each object class, which thereafter permits a spontaneous holistic process of identification by associating the object with its pattern.7 A recent study highlighted that the diagnostic accuracy of dermatologists in pigmented lesions relied much more on such an holistic assessment of their morphology than on the application of a combination of criteria.8 These data suggest that this spontaneous process of image identification available in each human being could be also used in melanoma recognition by laymen, provided that they have the opportunity to see enough images of melanoma and benign pigmented nevi to build their own recognition pattern. Whether such a cognitive training with photographs could be more efficacious to teach melanoma detection in the general population than a message based on the ABCD algorithm is a crucial question for the future campaigns for skin self-examination. We thus designed a randomized study that compares the efficacy of 3 different educational strategies, in conditions as close as possible to real public campaigns: cognitive training with photographs only, explanations based on descriptive criteria and combination of the 2.

Methods

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Objectives

The main objective was to compare the impact of 3 realistic educative interventions on the ability of normal people to recognize what is likely to be a melanoma: an explanatory strategy based on an analytic algorithm of criteria ABCD, a purely cognitive strategy based on photographs of benign and malignant pigmented lesions without any explanations (“Cog”) and a strategy combining a cognitive approach and explanations (“Cog-Ex”). The secondary objective was to evaluate how much anxiety usually associated with any message on cancer could impact on the effect of these educative strategies.

Design

The design was a prospective 4-arm randomized intervention study, stratified on age, sex and education level with 3 intervention arms “ABCD”, “Cog” and Cog-ex and a no-intervention arm (NI) as a control.

Setting

The setting of the study was people working in “Eurocopter EADS-Company”, the largest company in the Marseille area covering a large panel of the general adult population between 18 and 65 in terms of age, sex, and education level.

People were not initially aware of the objectives of the study. They volunteered to a proposal from their company medical service to have some information on environment and health. The volunteers were attributed to 12 stratification classes on the basis of age (3 groups) sex (2 groups) and education (2 groups), and the individuals from each stratification group were randomly allocated to one of the 4 arms: “ABCD”, “Cog”, “Cog-Ex” and “NI”.

Interventions

The different interventions had to be realistic in the context of a public campaign: little time available and message understandable by everyone. As a consequence, 3 different short leaflets (4 pages) were designed and were given to the individuals from each intervention group for only 10 min, without other comments.

The “ABCD” leaflet clearly described the ABCD algorithm on a melanoma photograph (Fig 1a). The “Cog” leaflet included only photographs of nevi, seborrheic keratoses and early melanomas (Fig. 1b) to provide images as patterns to stimulate the spontaneous holistic recognition of melanoma. Twenty-seven selected photographs, covering the most frequent different aspects of pigmented lesions, were presented by sets of 3 sharing some common overall morphological features (1 benign and morphologically regular, 1 benign and more irregular and 1 melanoma). The only mentions on this leaflet were the label “Good” for the most regular lesions, “Also good” for the more atypical and “Cancer. Danger” for the melanomas. Finally, the “Cog-Ex” leaflet (Fig. 1c) provided the same photographs of melanoma, nevi and seborrheic keratoses but presented with explanations pointing out the differences by the ABCD criteria.

thumbnail image

Figure 1. The 3 different leaflets. (a) “ABCD leaflet” describes the ABCD algorithm. (b) “Cog” leaflet includes 27 photographs presented by sets of 3, sharing some common overall morphological features: 1 benign and morphologically regular, 1 benign and morphologically more irregular and 1 melanoma with a simple comment. (“Ces grains de beauté sont bénins” + These moles are benign: “Bon” + good, “toujours bon” + also good, “ceux-ci sont des cancers” + these ones are cancers). (c) “cog-ex leaflet” includes the same 27 photographs presented by sets of 9, together with a clear description of the ABCD algorithm.

Download figure to PowerPoint

Outcome measures

Outcome measures were sensitivity (TP/TP+FN) (with T+True, F+False, P+Positive and N+ Negative), specificity (TN/TN+FP) and accuracy (TP/TP+FP+FN) in the recognition of melanoma. They were assessed by a photographic testing that included 40 slides of pigmented lesions with 10 melanoma, and 30 nevi and seborrheic keratoses. These pictures were chosen to fulfill a set of realistic conditions: 1) melanomas had to be early ones, 2) nevi had to be representative of the usual nevi we deal with, including one third of so-called atypical nevi, and 3/ the lesions could not be ambiguous for a trained dermatologists. The set of slides had been evaluated by 2 dermatologists, and the photographs that had not been correctly identified were removed and replaced.

The same photographic testing was successively submitted 3 times in each group: a reference test (T0) was performed before any intervention, the same test (T1) was performed again 1 week after intervention to assess the impact of the intervention. T0 and T1 were presented as follow: “According to you, is this lesion more probably a mole, or a cancer (at least suspicious enough to be a cancer which should be immediately referred to a doctor)”. In the 3 intervention arms (“ABC”, “Cog” and “Cog-Ex”), the test (T2) was performed again 1 hr after T1 but presented in a way that induced more anxiety: “Knowing that you musn't miss a cancer which could kill you or one of your relatives, review thoroughly these pictures to identify which ones could be cancers”.

Statistical analysis

In order to assess and to detect a difference of 3 correct responses between 2 interventions, with a standard deviation of 5 and a power of 80%, the sample size calculation by group was 60. Seventy individuals were thus proposed for each group. For each intervention, the effects were compared by a Wilcoxon test. The effects between the different interventions on sensitivity, specificity and accuracy were compared by multiple comparisons adjusted for multiplicity after a Kruskall Wallis test. The significance level was 5%.

Results

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Comparability of the groups

“ABCD”, “Cog”, “Cog-ex” and “NI” groups included 61, 65, 60 and 69 individuals, respectively. All the volunteers accepting the test T0 completed the study. The groups were well-balanced with no significant difference for sex, age or socio-occupational level. Before intervention, sensitivity, specificity and accuracy at T0 did not significantly differ between groups (p = 0.08, p = 0.72, and p = 0.66, respectively).

Comparison of the impact of the different educativestrategies (Fig. 2)

“ABCD” method slightly improved sensitivity from 80 to 83.8% (p + 0.04) but more clearly decreased specificity from 65.1 to 56.3% (p < 0.001), resulting in a slight reduction of accuracy from 40.8 to 38.2% (p + 0.04), but the before-after changes (T1−T0) in “ABCD” group did not significantly differ from those obtained in “NI” group.

thumbnail image

Figure 2. Sensitivity, specificity and accuracy in melanoma recognition before education (T0), 1 week after education (T1), after education + stress (T2), for the 3 intervention arms, i.e. information with ABCD algorithm, cognitive education with photograph (Cog) and mixed intervention (Cog-ex).

Download figure to PowerPoint

“Cog” method did not impact significantly on sensitivity but significantly and strongly improved specificity from 65.9 to 81.1% (p < 0.001) and accuracy from 42.1 to 53.9% (p < 0.001). The before-after improvements (T1−T0) of specificity and accuracy were significantly better than the one observed in “NI” (p < 0.001)and “ABCD” (p < 0.001). Although they tended to be better than in “Cog-ex”, the difference was not significant.

“Cog-ex” method did not significantly impact on sensitivity but significantly improved specificity from 63.4 to 75.1% (p < 0.001) and accuracy from 42.6 to 48.9% (p = 0.003) these before-after intervention improvements (T1−T0) were significantly better than in “NI” and “ABCD” groups (p < 0.001).

Comparison of the impact of the educative strategies, under stressing conditions (Fig. 2).

When the efficacy of the 3 educative strategies were assessed under more stressing conditions (T2), there was a marked increase of sensitivity (+7.7% mean increase in the 3 intervention groups, p < 0.001), and decrease of specificity (−11.9% mean decrease in the 3 intervention groups p ≤ 0.004), as compared to nonstressing conditions (T1).

Under these conditions, the differences between the impact of the 3 types of intervention became more clear-cut. After “ABCD” training, the loss of specificity was dramatic in conditions of stress (T2−T0), from 65.1 to 44.1% (p < 0.001), resulting in a decrease of accuracy from 40.8% to 35.8% (p ≤ 0.001), respectively. This effect on specificity and accuracy was significantly worse than in “NI” and “Cog” groups (p ≤ 0.005), although “ABCD” training did not provide a significantly higher gain in sensitivity than other strategies. Conversely, after “Cog” training, the sensitivity and accuracy increased under stressing conditions from 83.1% and 42,1% to 89.8 % and 49,6%, respectively (p < 0.005), without deleterious impact on the specificity, which even tended to increase from 65.9 to 69,7% (p = 0.085). This benefit (T2−T0) in specificity and accuracy was significantly better than in the “ABCD” group (p < 0.001). “Cog-ex” intervention, assessed under stressing conditions, resulted in a nonsignificant increase of sensitivity and accuracy and decrease in specificity, as compared to T0.

Discussion

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Our study demonstrates that a very short cognitive education with photographs is more efficacious to improve the ability of the laymen to recognize a melanoma among benign pigmented lesions than a written information based the ABCD algorithm. Since these data were obtained under realistic conditions in a sample of the general population, they suggest that this cognitive strategy should be used in public campaigns for the early detection of melanoma. For the laymen, photograph training increases the accuracy of recognition, whereas the ABCD algorithm decreases its specificity, without any substantial increase of sensitivity. Adding explanations about ABCD to photographs does not add any benefit and even tends to lower the efficacy of a purely cognitive education. Furthermore, when some stress and anxiety is associated to the message, which is common in cancer campaigns, the specificity of the ABCD algorithm alone decreases to a point that would really affect the cost-effectiveness of melanoma detection campaigns.

The superiority of a fast education with photographs alone over an information based on ABCD algorithm is not really surprising. First, photographs refer to intuitive image recognition by humans, which is an immediate holistic process, used in everyday life. Recognizing an image is associating this image with a memory of a model or a pattern, which we have built from a collection of previous images of this object stored in our visual experience.7 Once this pattern has been given an identity, we can recognize the object and name it. Two studies have already compared the effects of photographs on melanoma recognition. Although performed in artificial conditions, both concluded that there was a superiority of photographs over written information.9, 10 Similarly, dermatologists in their daily practice have been shown to base their diagnostic opinion on this holistic image recognition process and not on an algorithmic strategy.8 Second, the discriminative properties of this ABCD strategy in the hands of laymen in an everyday situation are doubtful. Indeed, a huge number of seborrheic keratoses or atypical nevi fulfill most of the ABCD criteria, whereas nodular melanomas, for example, do not always respond to these criteria. The accuracy of the ABCD algorithm, which has been recently reviewed,11 has only been evaluated in a small sample of melanomas,12 in a retrospective fashion in hospitals13, 14 and in an artificial spectrophotometric study.15 ABCD information might be useful for training primary health care agents,11 but our results show that they provide a much lower accuracy in the laymen than a stimulation of spontaneous recognition process by uncommented photographs.

In education campaigns directed toward early detection of melanoma by self-examination in the community, both sensitivity and specificity are needed. The 3 types of interventions that we have tested had a quite different impact. The small gain in sensitivity with ABCD was counterbalanced by a drop in specificity and thus a loss of accuracy. Conversely the cognitive strategies (“Cog”) mainly improved specificity and thus accuracy. The complementary explanations (“Cog-ex”) did not add any advantage over simple photographs. Sensitivity is obviously crucial since people must be able to recognize early most of lesions that require a consultation for fear of a possible melanoma. The sensitivity benefit may be considered as insufficient, whatever the method, but 2 factors have to be taken into account. First, there might be some ceiling effect of any intervention in an already well-informed population. Indeed, the awareness of the study population from the South of France about melanoma was already very high, with a basal sensitivity over 80% for a panel of early melanomas. Second, our study was focused on the best way to teach a direct morphological recognition of melanoma, but other useful indirect alert messages based on surveillance, i.e. “a recent change” or “E” criteria,16 and comparison, i.e. the “ugly duckling sign”17 could not be investigated in our study. They may however increase the sensitivity in the recognition of melanoma in real campaigns, if combined with a cognitive education with photographs. Moreover, “a recent change” and “ugly duckling sign” may work in a number of melanomas that do not look like usual ones and would thus escape a pure cognitive education. Despite the importance of sensitivity, specificity cannot be neglected in the self-detection of melanoma by the general population. With an annual risk of transformation of nevi in melanoma estimated under 1/200,000,18 and a similar number of other pigmented lesions such as seborrheic keratoses especially in elderly individuals, a very low specificity strategy (around and even under 50%) such as the ABCD algorithm may induce millions of useless consultations with several damageable effects: saturation of the medical system, huge cost and finally demotivation. Indeed, anyone repeatedly asking for consultation for ABCD lesions and who is repetitively answered by the doctor that the lesion is benign is likely to renounce asking for new advice. Finally, a cognitive strategy based on photographs, which provide both sensitivity and specificity over 80%, is an acceptable tool to promote self-detection in the general population.

These results were obtained under conditions that can be considered as realistic for a public campaign. People had only 10 min to look at a leaflet, which is probably compatible with the short time devoted by the public to an health education message on any media, such as TV, newspaper, leaflet or posters. The limited number of photographs on the “cog” leaflet is acceptable for a large-scale use. People who participated in the study, although not really representative of the general population, included a panel of age, sex and education, which is realistic as a target of a campaign for health education. However, interesting groups such as males over 65 were not included in this panel. Owing to the particularly poor prognosis of melanoma in this group,19 it may be worth a specific study. The slides used to assess the impact of the different strategies can also be considered as realistic since they included small early melanomas and a number of so-called atypical nevi, rather than large ugly melanomas and perfectly regular nevi. The successive tests being conducted with the same set of slides, the nonintervention group was used to control for a possible influence of the first view of the slides on the result of the next test. The effect on sensitivity in the ABCD group being similar to NI group, one could suspect that a previous knowledge of ABCD in the general population may have lead to underestimate the actual impact of ABCD. However, the knowledge about ABCD in NI was probably low since ABCD algorithm has not really been used as such in France for any campaign. The assessment of sensitivity and specificity using slides of pigmented lesions certainly differs from what would happen at the visual detection of tumors on the skin in the real life, but the difference between the respective impacts of the “ABCD” and cognitive strategies are likely to be in the same direction.

Our data also underline how much the way the educative message is presented can influence the accuracy of people. Simply by increasing psychological pressure before the test (T2), sensitivity increased by 6 to 9% whatever the arm, but with a concomitant fall in specificity, which was even dramatic after “ABCD” leaflet. This anxiety-induced increase in sensitivity, which raises up to 89.7% with a cognitive training, can be positively utilized in small high-risk subgroups, such as people with personal or familial history of melanoma, in which the loss of specificity is not a major problem. Conversely, the stress-induced decrease of specificity can be very counterproductive for the campaigns in the general population. Indeed, our data clearly show that a campaign based on a low specificity algorithm such as ABCD, presented in a way that could induce fear or anxiety, may paradoxically result in worse effects than no campaign at all, with an excess of millions of useless consultations for only a few more melanoma detected.

Although the impact of a campaign in the general population can only be assessed in the real scale, our study provides strong evidence that a campaign using a fast cognitive education based on photographs is feasible and could be much more effective than a campaign based on a written information about “ABCD” criteria. The combination of this cognitive strategy with an information about 2 other alert signs, i.e. a “recent change in a nevus ” and the “ugly duckling sign” is likely to be one of the most appropriate message for such campaigns.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Thanks to Eurocopter Company and to Dr. A. Poudevigne, and Dr. A. Tramier for their active contribution in the organization of this study.

References

  1. Top of page
  2. Abstract
  3. Methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References