Modeling the effects of melanoma education on visual detection: a gradient shift analysis.
Promoting Early Detection
Despite being the most lethal fonn of skin cancer, melanoma is particularly amenable to treatment if detected early (e.g., Geller et al. 2011; Kasparian et al. 2009; Rigel and Carucci 2000). Unfortunately, no consensus currently exists on best practices for early detection (e.g., Goodson and Grossman 2008; Pollitt et al. 2009). The primary method for initial identification of melanoma is via visual inspection, which is then followed by skin biopsy to allow for histological examination.
Although dermatological specialists are skilled at visually detecting melanomas, the same cannot necessarily be said for at-risk individuals who constitute the first line of defense in early detection and who are encouraged, in melanoma education efforts by groups such as the American Academy of Dermatology and the American Cancer Society, to engage in regular skin self-examination (SSE). Although many at-risk individuals engage in some form of SSE (Miller et al. 1996; Weinstock et al. 1999), the U.S. Preventive Services Task Force (2009) report on screening for skin cancer concluded that its efficacy is unclear. Studies suggest that visual detection by laypersons is inconsistent at best (e.g., Branstrom et al. 2002; Hamidi et al. 2010; Liu et al. 2005; Miles and Meehan 1995; Yagerman and Marghoob 2013), and although roughly half of new melanoma cases are first detected by patients (e.g., McGuire et al. 2011), melanomas detected by patients tend to be more advanced than those detected by medical professionals (Brady et al. 2000; Epstein et al. 1999; Lamerson et al. 2012; McPherson et al. 2006; Richard et al. 2000; Titus et al. 2012; Schwartz et al. 2002).
Based on the preceding, there would seem to be considerable value in improving our understanding of the detection process among laypersons. At a behavioral level, three steps would appear to be involved: (1) initiating and executing SSE, (2) visually detecting melanoma symptoms during SSE, and (3) seeking medical attention to confirm the need for intervention. Step 2 is a particularly promising target for improving early detection because an individual who cannot discriminate melanoma from a normal lesion (e.g., a benign mole) has little reason to engage in SSE (Step 1) and may fail to seek medical attention (Step 3) for genuinely problematic lesions or may seek attention when it is not needed.
Melanoma Detection as Behavior Under Stimulus Control
Our own interest lies is in examining the potential role of behavioral processes in visual detection of melanoma symptoms, particularly as they relate to early diagnosis. The process of distinguishing between normal and symptomatic lesions can be described in terms of stimulus discrimination (Dalianis et al. 2011), in which the presence or absence of certain stimulus features leads to differential responses as a result of previous experience with those stimuli. As such, it seems the efficacy of SSE likely depends on the extent to which prior training enhances a patient's tendency to distinguish between normal and symptomatic lesions.
Where patient training is concerned, the medical community relies heavily on educational interventions in which patients are told what melanoma symptoms look like. Friedman et al. (1985) developed a widely employed mnemonic device to help with educating patients about the visual characteristics of melanoma that correlate with malignance, referred to as the ABCDs: asymmetry of shape, border irregularity, color variegation (uneven pigmentation), and diameter greater than 6 mm. The ABCD mnemonic, and images comparing normal and highly symptomatic lesions, form the backbone of public service websites (e.g., see http://www.skincancer.org/skincancer-information/melanoma), printed materials intended for display or distribution in physician offices (e.g., http:// www.melanoma.org/understand-melanoma/resource-library/ educational-programs/melanoma-awareness-toolkit), and best practice recommendations for patient education (e.g., McWhirter and Hoffman-Goetz 2013). However, research examining training of the ABCD symptoms using text descriptions alone have been found not to promote early symptom detection (McWhirter and Hoffman-Goetz 2013).
Because stimulus discrimination depends on more than simply encountering stimuli (Dinsmoor 1995a, b; Guttman and Kalish 1956), the ability of patients to discriminate symptoms is likely to depend in part on the manner in which visual images are used to educate them. Melanoma education interventions may not be designed as exercises in operant discrimination learning per se, but the stimulus comparisons they present do mirror one key aspect of discrimination learning--namely the functioning of dissimilar stimuli as S+ and S(stimuli signaling reinforcement and nonreinforcement, respectively). An S+ that evokes a certain response (exhibits stimulus control) will normally engender similar behavior in the presence of stimuli that share features with it as compared to stimuli that are more disparate from it. The extent to which this spread of effect occurs can be assessed by examining responding to other stimuli along a given dimension of that S+ (e.g., symptom severity of lesions), thereby producing a generalization gradient, with highest responding at the S+ and decreases in responding as stimuli become less similar. When learning to discriminate between S+ and S-, the stimulus control that would otherwise be exerted naturally by the S+ is often altered along the shared dimension(s) of the two stimuli, resulting in a shift in responding away from the S+ toward stimuli dissimilar to the S- (Hanson 1959). This phenomenon is known as gradient shift (e.g., Ghirlanda and Enquist 2003).
Relating this concept to melanoma education, the ABCD mnemonic (Friedman et al. 1985) describes stimulus features typical of malignance with the anticipation that patients who notice an abnormal lesion will contact a physician in order to receive potentially lifesaving treatment. In the language of stimulus control, an image of a highly symptomatic lesion, as presented on a website or in a brochure, is (presumably) expected to become S+ for the response of see a doctor, with the "reinforcer" of avoiding death or living longer, and the effect of this training is expected to generalize to other symptomatic lesions. While preparing this article, we consulted 15 melanoma websites, as identified in by a popular Internet search engine, and found that all adopted this general approach. For example: "Look for the ABCD signs of melanoma, and if you see one or more, make an appointment with a physician immediately" (http://www.skincancer.org/skincancer-information/melanoma). If a patient being educated about melanoma also is presented with an image of an asymptomatic lesion, that image (presumably) becomes S for the response of see a doctor.
Figure 1a illustrates the preceding description of melanoma education in terms of stimulus control. Here, S+ is a highly symptomatic lesion and S- is an asymptomatic lesion. Between them is a continuum of symptom severity. The bold, dashed line in the panel shows the general pattern of responding that is expected: stimuli physically similar to S+ are likely to occassion behavior and those that are dissimilar are unlikely to do so. However, following explicit training with S-, gradient shift occurs, displacing responding in the direction opposite the S-. The bold, solid line in the panel illustrates this outcome. Given that this is the underlying structure of stimuli presented in many melanoma education efforts, gradient shift effects could influence the efficacy of SSE. Presumably, melanoma education interventions present patients with highly symptomatic lesions because these are easy to recognize (discrimination learning is promoted when S+ and S- are very different; e.g., Terrace 1963). The goal, however, is for patients to notice changes in the condition of their skin as soon as possible after they begin to appear (Abbasi et al. 2004), and thus, ideally, patients will detect symptoms that are more subtle than shown in the example lesions of melanoma education efforts. In the language of stimulus control, this means responding to stimuli that are only marginally similar to S+ as if they were S+. Unfortunately, such a gradient shift effect would be expected to reduce the tendency for mildly symptomatic lesions to be responded to as symptomatic (see Fig. 1a).
Goals of the Present Investigation
The present investigation was designed to employ formal generalization testing as a means of modeling the potential effects of melanoma education efforts on the tendency to label subtly symptomatic lesions as symptomatic versus asymptomatic. Of particular interest was the potential gradient shift effect shown in Fig. la. Participants in a Standard Education group first were taught to distinguish between images of two skin lesions, one asymptomatic (S-) and one symptomatic (S+ ), just as seen in commonly employed melanoma education efforts. A generalization test then assessed responding to images of lesions of varying symptom severity, with the primary interest in whether, as expected based on gradient-shift effects, mildly symptomatic lesions tended to be categorized as asymptomatic. For this purpose the frame of reference was a control group in which participants learned to respond appropriately to the same S+ without being exposed to an S-. This group was expected to produce an "unshifted" generalization gradient.
A third (Reversed Education) group was included in the experimental design to permit examination of a plausible approach to melanoma education that we have not seen employed. For this group, the S+ and S- stimuli described above for the primary group were reversed, as summarized in Fig. lb. Imagine a melanoma education intervention in which patients are shown an asymptomatic lesion and told, "When you see this, your skin is healthy, and you can go about your normal business." In this case, the asymptomatic lesion is expected to control (serve as S+ for) a variety of responses other than seeing a physician. A symptomatic lesion, by contrast, would be S- for the same constellation of "non-help-seeking" responses (which, presumably, leaves help-seeking as prepotent). While the distinction between this scenario and that of Fig. la may seem to be one of semantics, in stimulus control terms it could be practically important because of the potential gradient shift illustrated in Fig. lb. Here, that shift translates into a reduced tendency to treat mildly symptomatic lesions as asymptomatic--something that would be helpful in promoting early melanoma detection.
[FIGURE 1 OMITTED]
Participants were 24 undergraduate students (19 female, 5 male, 18 to 23 years of age) who were recruited from an introductory behavioral science course. For participating, they earned 1 % of extra credit added to the final course grade. Participants were randomly assigned to the three groups (each n=8) that are described below.
Settings, Apparatus, and Stimuli
Data were collected in a computer lab (9 m by 6 m) containing approximately 20 Dell Optiplex 360 computers equipped with Coby CV185 headphones. A computer mouse was placed within reach, and the keyboard was absent. Dell P1911 17" wide-aspect flat-panel monitors were equipped with screen protectors to minimize the ability of adjacent participants to view others' screens. Datasheets were placed on the table directly in front of the monitor along with a ballpoint pen to record responses.
Stimuli were created using the same morphing software (Morpheus Photo Morpher) as reported in Dalianis et al. (2011). We obtained two benchmark digital images from a melanoma-education website. One showed an asymptomatic lesion (i.e., a benign mole) and the other showed a lesion that was highly symptomatic of melanoma. The two benchmark images were loaded into the morphing program, which then created a series of intermediate images that combined features of each original image progressing in equal intervals from one to the next. The program generated 98 intermediate images, resulting in a progression of images from asymptomatic to highly symptomatic; we will refer to these stimuli as SI to SI00. For the experiment we used 15 stimuli from the progression, including Stimulus 1 and 14 additional images at interval steps of 7 (thus, S8, S15, S22, S29, S36, S43, S50, S57, S64, S71, S78, S85, S92, and S99). Because the benchmark images are copyright protected, the actual images used in the study cannot be reproduced here.
The morphing software used a linear algorithm, such that a constant amount of stimulus change distinguished adjacent images in the symptom-severity progression (e.g., the difference between SI and S2 was equivalent to the difference between S64 and S65). Whether the degree of psychophysical distance between stimuli also was linear is not known, although Dalianis et al. (2011) presented evidence roughly consistent with this assumption.
Stimuli were presented using Microsoft PowerPoint. On the computer screen the stimuli ranged in diameter from approximately 4.5 mm (SI) to approximately 7.5 mm (S99) and appeared in the center of the screen on a flesh-colored background (RGB code=213 R, 172 G, 129 B). At no time during the experiment were the stimuli described as skin lesions or as representing cancer. In this way we hoped to examine control by physical features of the stimuli unencumbered by any peculiar associations participants might have with cancer.
The procedure was based on those of many previous studies involving human participants. Specifically, participants were exposed to both S+ and S- before indicating whether a range of generalization-test stimuli were or were not S+, and were given verbal feedback in the form of written messages rather than the tangible consequences typically employed with nonhumans (e.g., Bizo and McMahon 2007; Derenne 2010; Spetch et al. 2004). Importantly, many generalization effects that are considered standard in animal laboratory experiments have been widely replicated using such procedures.
Up to 12 participants completed the study simultaneously in the computer lab, but they worked individually without conferring with one another. Participants were told to read instructions on the screen (reproduced below), to the wear headphones during the duration of the study, and to record responses on the sheet in front of them.
The procedure included a training phase and a generalization-test phase, with groups differing only in terms of stimuli encountered during training. Participants in one group received training in which S+ was more symptomatic than S-. This Standard Education group represented typical practices in melanoma education. A second group received training in which S+ was less symptomatic than S-. This group, referred to here as the Reversed Education group, represented the aforementioned hypothetical scenario in which melanoma education emphasizes the appearance of normal lesions rather than those showing melanoma symptoms. Participants in a Control group received training only with a symptomatic S+ (no S-) and were expected to produce unshifted generalization gradients.
Table 1 summarizes the training stimuli. To allow for controlled examination of the training effects, the stimulus midway on the generated continuum of symptom severity (S50) was used as S+ for all groups. Had the procedure exactly paralleled the examples described in Fig. 1, one of the most extreme stimuli (S99 for the Standard Education group and S1 for the Reversed Education group) would serve as S+, but this would result in a restricted range of stimuli for measurement purposes. For example, with S99 as S+, no stimuli more symptomatic than S+ would exist for use in measuring the extent of any resulting gradient shift. Using S50 provided a range of test stimuli that were both more and less symptomatic than S+ for all groups. This arrangement retained the key feature of images used in melanoma education efforts by contrasting an asymptomatic lesion with a symptomatic lesion.
Both training and testing trials were presented in a successive-discrimination format, which was selected because it captures key features of how individuals evaluate changes in their own moles. It is possible to look at one mole at a time and decide whether it or not it shows troubling changes from how it looked previously. Under natural conditions, it is not possible to look simultaneously at the same mole under different levels of symptomatic progression over time.
Training Participants initiated the study by mouse-clicking a button labeled "Start" once they had finished reading these instructions:
You will be shown series of pictures depicting a shape. When this shape appears, study it carefully. You will have to remember what the shape looks like. After a short delay, you will be shown a series of pictures of the same shape. Some of the shapes that you see will be the same as the original picture; others will be different. The original picture and the new picture will be only slightly different, but there is a difference in the shape. You will have to indicate whether you think a given image is the original one or not. A beep will sound on your headphones and a "Record" message will appear on the screen when it is time to answer. You will mark your responses by checking either the "Yes" or "No" box on the datasheet in front of you. At first you will be told whether your choices are correct by clicking on the response you recorded. Later, there will not be any feedback. Try to be as accurate as you can. You will occasionally be given a 60-s break. Please remain seated and facing the screen. A timer will be displayed. The images will begin appearing as soon as the timer runs out. Click on the "Start" button to start the experiment.
Training began with S+ presented on screen along with a label identifying it as the "original" stimulus. After 10 s, the screen went briefly blank and training trials began. On each trial, a single stimulus was presented and remained onscreen throughout the trial. At the onset of the trial the query, "Is this the same as the original?" appeared. After 3 s, a 1-s tone (projected through the headphones) and the instruction, "Record" prompted the recording of a response by checking a "Yes" box or a "No" box on the answer sheet. After a further 3 s, "Yes" and "No" buttons appeared on the screen. Participants were instructed to mouse-click in order to register the same response as on the answer sheet. Mouse-clicking one of the buttons was required to advance to the next trial.
Following a correct response to S+, the word "Correct!" appeared in green for 2 s, accompanied by a chime sound played into the headphones. Following an incorrect response to S+, the word "Incorrect" appeared in red for 2 s, with no accompanying sound. For training trials presenting S-, the same feedback was provided except that the chime did not play following correct responses to the S-.
Following feedback, a 5-s intertrial interval (ITI) occurred, during which the screen turned black. At the end of the ITI, white text appeared indicating that clicking on the message would initiate the next trial.
Training included 20 trials. For the Control group, all training trials presented S+. For the two experimental groups, training trials presented either S+ or S- in semirandom order (no more than three consecutive trials of one stimulus type). Following training, participants were given a 60-s break, told that they would no longer be using the computer mouse, and instructed to place the mouse to the side.
Generalization Test In the generalization test, which was identical for the three groups, trials were structured similarly to those in training, with two exceptions. First, no feedback was presented. Thus, "Yes" and "No" buttons did not appear on screen, and responses were only recorded on paper. Second, mouse clicks were not required to advance to the next trial following the 5-s ITI.
During the generalization test, all 15 stimuli from the array were presented in randomized order, once within each trial block. The test comprised nine trial blocks, totaling 135 trials. Individual trial blocks were not signaled to the participants. Breaks of 60 s took place after the third and sixth blocks, during which a countdown timer appeared on screen to indicate when the test would resume.
During training, participants made few errors (0-2 incorrect answers) and all responded correctly on at least 9 of the last 10 training trials, showing mastery of training. For the Control group, 6 of 8 participants completed training without errors. In both the Standard Education and Reversed Education groups, 3 participants each completed training without error. The distribution of errors differed somewhat between the groups receiving discrimination training, with more participants in the Standard Education group responding incorrectly to S+ (S50) than to S- (S1), and more participants in the Reversed Education group responding incorrectly to S- (S99) than to S+ (see Table 2). That is, participants in the Standard Education group were more likely to record "No" to the S+, indicating that it was not the original stimulus, and those in the Reversed Education group were more likely record "Yes" to the S-, indicating that the highly symptomatic stimulus was the original stimulus. No participants in either group made errors to both types of stimuli.
All of the participants made at least one affirmative response during generalization trials, suggesting some degree of generalization, and recorded a response for each trial throughout the experiment, providing a full data set for analysis.
Visual Inspection of Generalization Gradients To provide a global summary of generalization outcomes, the frequency of affirmative responses--a response of "Yes" to the question "Is this the same as the original?"--made by each individual in a group at each stimulus value was averaged across all group participants. Figure 2 shows the resulting average gradients. For the control group, "Yes" responses were most frequent to S+ and became less frequent as the stimuli diverged from S+. There was a slight bias for responding "Yes" to stimuli displaying higher levels of symptom severity (to the right of S+). Average gradients for both the Standard Education and Reversed Education groups showed shifting away from their respective S-. For the Standard Education group, "Yes" responses were most frequent to S57 and S64. For the Reversed Education group "Yes" responses were most frequent to S43.
Figures 3, 4 and 5 depict individual gradients, most of which show the same general pattern as the average gradients. For the Control group, gradients tended to approximate left-right symmetry, with peaks showing little systematic deviation from S+. For the Standard Education and Reversed Education groups combined, 13 of 16 gradients showed visually apparent shifting away from S-. In examining whether responding during training influenced responding during the generalization test, there did not appear to be any consistent effect. However, the only participant in the Standard Education group who responded incorrectly to S- also had the most pronounced shift of any participant (see Fig. 4, panel 6).
[FIGURE 2 OMITTED]
Quantitative Analyses of Gradient Shift To corroborate the visually apparent patterns of Figs. 2,3,4 and 5, we conducted statistical analysis of gradient means (Hanson 1959; Galizio 1985) to quantify gradient shift. To determine gradient mean for each participant, we examined the portion of the gradient to the side of S+ toward which a shift was expected to occur. Thus, gradient means were obtained, for responses to the right of S+ for the Standard Education group and to the left of S+ for the Reversed Education group, by multiplying the number of "Yes" responses to each stimulus by the stimulus value, summing the products, and dividing by the total number of responses. Results were compared to the relevant gradient means for the Control group using one-tailed Mann-Whitney U tests, with both the Standard Education and Reversed Education groups differing significantly from Control (U= 10.50, p<.05, and U= 15.50, p<.05, respectively).
[FIGURE 3 OMITTED]
Previous research has found decreased responding over the course of generalization tests conducted under conditions of extinction (e.g., Blough 1972; Guttman and Kalish 1956). Affirmative responses to stimuli during each trial block are denoted in the individual graphs of Figs. 3, 4 and 5 by the gray blocks scaled to the right y-axis. Response frequencies varied by participant and across trial blocks. A two-way repeated measures ANOVA of response frequency by trial block found a main effect of time, F(16, 168)=6.91 , p<.05, but not group, F(2, 21)=0.61, p>.05, indicating that affirmative responding generally decreased across trial blocks.
To determine whether the decrease in responding across time occurred differentially across stimuli, we conducted a two-way repeated measures ANOVA of gradient means. For this analysis, each participant's responses during the generalization test were separated into three sets of three trial blocks (Blocks 1-3, 4-6, and 7-9), and gradient means were calculated as before except that responses to all 15 stimuli were included. It was necessary to analyze trial blocks in sets as there were instances in which no affirmative responses were recorded within a single block. No statistically significant change in gradient mean was observed across time for any group, F(2,42)=2.42, p>.05, indicating that, despite changes in overall frequency, the distribution of "Yes" responses among the stimuli remained fairly consistent for each participant throughout the generalization test.
[FIGURE 4 OMITTED]
The purpose of this study was to assess whether gradient shift could have an impact on the melanoma symptom-detection repertoires of people who have been taught to distinguish between clearly symptomatic and asymptomatic lesions. After participants received training with designated stimuli, generalization was assessed with respect to stimuli representing skin lesions on a continuum of symptom severity. Participants trained with the moderately symptomatic S+ only (Control group) showed largely unshifted gradients (albeit with a slight bias in responding toward more symptomatic stimuli). That is, they showed the classic pattern of designating stimuli as "like S+" only to the degree that they were similar to S+. Participants whose discrimination training included an S- stimulus generally displayed gradient shift. Compared to what was seen in the Control group, these groups showed differing tendencies to "report" subtly symptomatic lesions.
[FIGURE 5 OMITTED]
Our findings can be appreciated on two levels. At the more general level, although we employed stimuli very different from what has been used in most studies of stimulus generalization and employed procedures that differ from those of animal operant-laboratory experiments, we replicated the general finding of gradient shift, thereby attesting to the robustness and generality of this effect (Ghirlanda and Enquist 2003; Honig and Urcuioli 1981). Second, and more central to the purpose of the present investigation, this early-stage translational experiment has possible implications for melanoma education efforts in the everyday world, in particular suggesting that traditional melanoma education interventions may discourage detection of very subtle symptoms, an outcome that would run counter to the early-detection emphasis of all melanoma education efforts.
Gradient Shift and Detection of Subtle Melanoma Symptoms
Commonly in melanoma education efforts, at-risk patients are shown images of symptomatic and asymptomatic lesions and told to seek medical help if they find a symptomatic lesion on their own skin, similar to the example scenario depicted in Fig. 1a. Our data suggest that due to gradient shift effects this practice could have adverse effects on early detection. That is, compared to our Control group, participants in the Standard Education group were less likely to "report" subtly symptomatic lesions (e.g., Fig. 2). Based on the present analysis and a general understanding of stimulus control principles, a relatively simple improvement over present practices might be to present atrisk individuals with examples of melanoma (S+) only, thereby avoiding the potentially problematic gradient shift seen in our results.
Often in generalization experiments S+ and S- are quite different, defining two extremes on some stimulus dimension. Following this example, in melanoma education one might choose to show at-risk patient highly symptomatic melanoma symptoms. However, using Fig. la as a reference, the further to the right of the symptom-severity continuum that S+ lies, the less generalization can be expected to occur to subtly symptomatic stimuli. In other words, if patients are taught to recognize relatively advanced melanoma, this may prepare them to recognize moderately advanced melanoma, but perhaps not small changes in lesions that should be the focus of early detection. It may be more helpful to provide example of moderately symptomatic lesions, much as we did in the present study, because the melanoma-reporting repertoire would therefore generalize to more subtly-symptomatic lesions. (1)
A possible alternative approach to melanoma education might be based on the procedures employed with our Reversed Education group. This counterintuitive approach would focus on the behavior class of "acting normally" rather than the behavior of "reporting cancer," using an asymptomatic lesion as S+ and a symptomatic lesion as S-. Educational media might stress that when skin lesions look like the asymptomatic example, "You can relax. You have no troubling symptoms. There is no need to seek medical help. Spend your time doing other things." As Fig. 2 (bottom) showed, this approach might lead to more sensitive detection of early-stage melanoma symptoms.
To be clear, our Reversed Education group raises interesting questions about behavioral processes as they may play out in melanoma education interventions, but it was not designed as a strict model for those interventions. For experimental convenience we used a moderately symptomatic lesion as S+, whereas in actual Reversed Education efforts, S+ might be completely asymptomatic. Alternatively, both symptomatic and asymptomatic lesions could be shown, with equal emphasis placed on the appropriate responses to both, rather than, as in typical current practices, emphasizing only the action to be taken when symptoms appear. This might be expected to essentially sum the gradient shifts seen in our Standard Education and Reversed Education groups, resulting in an unshifted generalization gradient peaking at the optimal range for accurate early detection. Empirical evaluation of these procedures would be necessary to determine which, if any, of these suggestions would benefit melanoma education.
Limitations and Next Steps
A possible point of concern in the present investigation is whether the Control group really provides a valid frame of reference for evaluating gradient shifts in the other groups. The generalization gradients of most participants in the Control group were not completely symmetrical but rather showed minor shifting toward the more-symptomatic end of the continuum of experimental stimuli. As Ghirlanda and Enquist (2003) have noted, however, modestly biased responding during generalization tests (asymmetric responding around S+ in absence of training with S-) is a common finding. For example, biases toward one end of the stimulus array also were seen in Derenne's (2010; Derenne et al. 2008) studies in which the stimuli represented degrees of facial symmetry and female waist-to-hip ratio. Such effects can result from a variety of factors (e.g., they may indicate that discriminability among stimuli does not change uniformly along the continuum employed for experimental purposes) but do not invalidate the present results, because the gradient shifts of both experimental groups represented significant changes from the pattern shown by the Control group. Thus, gradient shifts tended to exceed what would be expected to result strictly from features of the stimuli themselves.
What can be stated unequivocally is the degree of generalization was a function of the symptom-severity progression shown in the stimuli. Impossible to determine from the present results is what specific melanoma symptoms controlled detection responses. As is typical of melanoma, the highly symptomatic image used to create our stimulus progression showed each of the "ABCD" symptoms (asymmetry of shape, border irregularity, color variegation, and diameter greater than 6 mm). Any one of these might have been the basis of discrimination (during training) and generalization (during the test), and there is no guarantee that our results would have been the same had a different benchmark lesion been employed. An obvious next step in this line of research would be to replicate using a variety of symptomatic stimuli. We note, however, that most of the melanoma educational media that we have seen present between one and a few symptomatic lesions, so the problem of limited exemplars in our study applies equally to many, if not most, melanoma education efforts. Another important future direction would be to conduct studies to examine the relative impact on detection of each of the major symptom types (cf. Bloom et al. 1982).
Several factors may limit the generality of these results to actual melanoma education efforts. The first possible limitation is that, although the benchmark images used in creating stimuli were obtained from a medically informed website, the intermediate stimuli representing melanoma symptoms were not checked for face validity by a dermatologist. It would be valuable to have dermatologist input into stimulus construction in future investigations. We note, however, that some melanoma education efforts, such as that on the Melanoma Education Foundation's website (http://www.skincheck.org), employ morphing technology similar to ours to produce a range of intermediate images.
A second possible limitation is that we took care not to describe the stimuli of the present study as skin lesions or to mention cancer at any time, and we established and measured stimulus control over a generic behavior (reporting whether an image was the same as a comparator), rather than medically relevant responses. In these cases, it is possible that individuals would respond differently to similar stimuli when cancer is the topic of conversation (e.g., Girardi et al. 2006). We suggest, however, that the relevant stimulus control, including that exerted by stimuli that may be associated with melanoma, is generic in the sense that it reflects a common set of principles (e.g., gradient shift may occur when S+ and S- represent a common stimulus dimension; Ghirlanda and Enquist 2003). A "cancer" context may provide special motivation to engage in SSE, but during an examination one must still be able to tell the difference between symptomatic and asymptomatic lesions. Altering the question to be more directly related to melanoma detection--such as by having individuals indicate whether they should see a doctor, whether they are "fine," or rating the severity of a lesion--would be an important step toward greater ecological validity.
A third possible limitation is that our participants were not specifically recruited on the basis of current risk for developing melanoma. However, greater use of indoor tanning beds among college-aged women has resulted in heightened risk in this population (Boniol et al. 2012; Guy et al. 2013). Replicating the present study with participants who are at special risk of developing melanoma would be essential for ensuring that training programs informed by behavioral processes are useful for the intended population.
Despite the fact that early melanoma identification is linked to improved survival rates (e.g., Geller et al. 2011; Kasparian et al. 2009; Rigel and Carucci 2000), existing technology for promoting early detection appears to be fairly ineffective (e.g., Goodson and Grossman 2008; McWhirter and Hoffman-Goetz 2013; Pollitt et al. 2009). There is clearly a need for new perspectives on this problem. We noted previously that some approximation of discrimination training appears to be embedded in many melanoma education interventions. Yet, these interventions do not appear to be grounded in stimulus control principles. Our study represents a first attempt to understand how discrimination training, conducted in certain ways, might affect subsequent visual detection of melanoma. Based on the clarity and obvious potential relevance to everyday melanoma detection of our results, we suggest that stimulus control methods and principles constitute a potentially fertile source of guidance for translational research on this important topic.
Our findings are subject to many caveats and therefore insufficient to guide the immediate development of new melanoma education technologies. They do, however, provide encouragement about the value of a long-term translational research program examining stimulus control processes in melanoma symptom recognition. Pursued to its logical conclusion, such a program can inform interventions, as evidenced by the success of MammaCare, a technology for teaching breast self-examination that was developed by integrating stimulus-control principles and research with medical expertise and materials engineering. In its consumer-ready form, the technology can be implemented in as little as 15 minutes and, compared to traditional approaches to teaching self-examination, has been shown to cost-effectively improve the detection of small breast lumps (e.g., Fletcher et al. 1990; O'Malley 1993; Saunders et al. 1986). Importantly, the early development of MammaCare included laboratory studies in which well-understood stimulus control principles were examined in the context of stimuli with clear relevance to breast cancer (e.g., Adams et al. 1976). It is in this preliminary sense that we suggest our results be considered as informative to melanoma symptom detection.
Published online: 10 December 2014
Acknowledgments Portions of this study were completed in partial fulfillment of the first author's doctoral degree requirements at the University of Kansas. The authors express sincere gratitude to Kathryn Saunders, David Jarmolowicz, Steven Fawcett, and Kimberly Engelman for their comments on the design and analysis of this study.
This investigation was supported by the University of Kansas New Faculty General Research Fund allocation #2302290 and General Research Fund allocation #2301722.
Abbasi, N. R., Shaw, H. M., Rigel, D. S., Friedman, R. J., McCarthy, W. H., Osman, I., ... Polsky, D. (2004). Early diagnosis of cutaneous melanoma. Journal of the American Medical Association, 292, 2771-2776. doi:10.1001/jama.292.22.2771.
Adams, C. K., Hall, D. C., Pennypacker, H. S., Goldstein, M. K., Hench, L. L., Madden, M. C., ... Catania, A. C. (1976). Lump detection in simulated human breasts. Perception & Psychophysics, 20, 163-167. doi: 10.3758/bf03198595.
Bizo, L. A., & McMahon, C. V. (2007). Temporal generalization and peak shift in humans. Learning & Behavior, 35, 123-130. doi:10. 3758/bfD3193047.
Bloom, H. S., Criswell, E. L., Pennypacker, H. S., Catania, A. C., & Adams, C. K. (1982). Major stimulus dimensions determining detection of simulated breast lesions. Perception & Psychophysics, 32, 251-260. doi: 10.3758/bf03206229.
Blough, P. M. (1972). Wavelength generalization and discrimination in the pigeon. Perception & Psychophysics, 12, 342-348.
Boniol, M., Autier, P., Boyle, P., & Gandini, S. (2012). Cutaneous melanoma attributable to sunbed use: Systematic review and metaanalysis. British Medical Journal, 345, e4757. doi: 10.1136/ bmj.e4757.
Brady, M. S., Oliveria, S. A., Christos, P. J., Berwick, M., Coit, D. G., Katz, J., & Halpem, A. C. (2000). Patterns of detection in patients with cutaneous melanoma. Cancer, 89,342-347. doi: 10.1002/10970142(20000715)89:2<342:: AID-CNCR19>3.0.CO;2-P.
Branstrom, R., Hedblad, M., Krakau, I., & Ullen, H. (2002). Laypersons' perceptual discrimination of pigmented skin lesions. Journal of American Academy of Dermatology, 46, 667-673. doi: 10.1067/ mjd.2002.120463.
Dalianis, E. A., Critchfield, T. S., Howard, N. L., Jordan, J. S., & Derenne, A. (2011). Using morphed images to study visual detection of cutaneous melanoma symptom evolution. The Psychological Record, 61, 341-362.
Derenne, A. (2010). Shifts in postdiscrimination gradients within a stimulus dimension based on bilateral facial symmetry. Journal of the Experimental Analysis of Behavior, 93, 485-494. doi: 10.1901/jeab. 2010.93-485.
Derenne, A., Breitstein, M., & Cicha, R. J. (2008). Shifts in postdiscrimination gradients within a stimulus dimension based on female waist-to-hip ratios. The Psychological Record, 58, 51-60.
Dinsmoor, J. A. (1995a). Stimulus control: Part I. The Behavior Analyst, 18, 51-68.
Dinsmoor, J. A. (1995b). Stimulus control: Part II. The Behavior Analyst, 18, 253-269.
Epstein, D. S., Lange, J. R., Gruber, S. B., Mofid, M., & Koch, S. E. (1999). Is physician detection associated with thinner melanomas? Journal of the American Medical Association, 281, 640-643. doi: 10.1001/jama.281.7.640.
Fletcher, S. W., O'Malley, M. S., Earp, J. L., Morgan, T. W., Lin, S., & Degnan, D. (1990). How best to teach women breast self-examination: A randomized controlled trial. Annals of Internal Medicine, 112, 772-779. doi: 10.7326/0003-4819-112-10-772.
Friedman, R. J., Rigel, D. S., & Kopf, A. W. (1985). Early detection of malignant melanoma: The role of physician examination and self-examination of the skin. CA: A Cancer Journal for Clinicians, 35, 130-151. doi: 10.3322/canjclin.35.3.130.
Galizio, M. (1985). Human peak shift: Analysis of the effects of three-stimulus discrimination training. Learning & Motivation, 16, 478-494. doi: 10.1016/0023-9690(85)90028-1.
Geller, A. C., Swetter, S. M., Oliveria, S., Dusza, S., & Halpern, A. C. (2011). Reducing mortality in individuals at high risk for advanced melanoma through education and screening. Journal of the American Academy of Dermatology, 65, S87.el-S87.e9. doi:10. 1016/j.jaad.2011.05.045.
Ghirlanda, S., & Enquist, M. (2003). A century of generalization. Animal Behaviour, 66, 15-36. doi:10.1006/anbe.2003.2174.
Girardi, S., Gaudy, C., Gouvernet, J., Teston, J., Richard, M. A., & Grob, J. J. (2006). Superiority of a cognitive education with photographs over ABCD criteria in the education of the general population to the early detection of melanoma: A randomized study. International Journal of Cancer, 118, 2276-2280. doi: 10. 1002/ijc.21351.
Goodson, A. G., & Grossman, D. (2008). Strategies for early melanoma detection: Approaches to the patient with nevi. Journal of the American Academy of Dermatology, 60, 719-735. doi: 10.1016/j. jaad.2008.10.065.
Guttman, N., & Kalish, H. I. (1956). Discriminability and stimulus generalization. Journal of Experimental Psychology, 51, 79-88. doi:10.1037/h0046219.
Guy, G. P., Berkowitz, Z., Watson, M., Holman, D. M., & Richardson, L. C. (2013). Indoor tanning among young non-Hispanic white females. JAMA Internal Medicine, 773(20), 1920-1922. doi: 10. 1001/jamaintemmed.2013.10013.
Hamidi, R., Peng, D., & Cockburn, M. (2010). Efficacy of skin self-examination for the early detection of melanoma. International Journal of Dermatology, 49, 126-134. doi: 10.1111/j.1365-4632. 2009.04268.x.
Hanson, H. M. (1959). Effects of discrimination training on stimulus generalization. Journal of Experimental Psychology, 58, 321-334. doi: 10.1037/h0042606.
Honig, W. K., & Urcuioli, P. J. (1981). The legacy of Guttman and Kalish (1956): 25 years of research on stimulus generalization. Journal of the Experimental Analysis of Behavior, 36, 405 445. doi:10.1901/ jeab. 1981.36-405.
Howlader, N., Noone, A. M., Krapcho, M., Garshell, J., Miller, D., Altekruse, S. F., ... Cronin, K. A. (eds) (2013). SEER Cancer Statistics Review, 1975-2011. Retrieved from National Cancer Institute website http://seer.cancer.gov/csr/1975_2011/.
Karakousis, G. C., & Czemiecki, B. J. (2011). Diagnosis of melanoma. PET Clinics, 6, 1-8. doi: 10.1016/j.cpet.2011.02.001.
Kasparian, N. A., McLoone, J. K., & Meiser, B. (2009). Skin cancer-related prevention and screening behaviors: A review of the literature. Journal of Behavioral Medicine, 32, 406-428. doi: 10.1007/ s10865-009-9219-2.
Lamerson, C. L., Eaton, K., Sax, J. L., & Kashani-Sabet, M. (2012). Comparing melanoma invasiveness in dermatologist-versus patient-detected lesions: A retrospective chart review. Journal of Skin Cancer, 2012, 1-5. doi: 10.1155/2012/187963.
Liu, W., Hill, D., Gibbs, A. F., Tempany, M., Howe, C., Borland, R., ... Kelly, J. W. (2005). What features do patients notice that help to distinguish between benign pigmented lesions and melanomas?: The ABCD(E) rule versus the seven-point checklist. Melanoma Research, 15, 549-554. doi: 10.1097/00008390-200512000-00011.
McGuire, S. T., Secrest, A. M., Andrulonis, R., & Ferris, L. K. (2011). Surveillance of patients for early detection of melanoma: Patterns in dermatologist vs patient discovery. Archives of Dermatology, 147, 673-678. doi:10.1001/archdermatol.2011.135.
McPherson, M., Elwood, M., English, D. R., Baade, P. D., Youl, P. H., & Aitken, J. F. (2006). Presentation and detection of invasive melanoma in a high-risk population. Journal of the American Academy of Dermatology, 54, 783-792. doi: 10. 1016/j.jaad.2005.08.065.
McWhirter, J. E., & Hoffman-Goetz, L. (2013). Visual images for skin self-examination and melanoma detection: A review of published studies. Journal of the American Academy of Dermatology, 69, 47-55. doi: 10.1016/j.jaad.2013.01.031.
Miles, F., & Meehan, J. W. (1995). Visual discrimination of pigmented skin lesions. Health Psychology, 14, 171-177. doi: 10.1037/02786220.127.116.11.
Miller, D. R., Geller, A. C., Wyatt, S. W., Halpern, A., Howell, J., Cockerell, C., ... Rosenthal, L. (1996). Melanoma awareness and self-examination practices: Results of a United States survey. Journal of the American Academy of Dermatology, 34, 962-970. doi: 10.1016/s0190-9622(96)90273-x.
National Cancer Institute (2014). SEER stat fact sheets: Melanoma of the skin. Retrieved from http://seer.cancer.gov/statfacts/html/melan. html. Accessed 1 June 2014.
O'Malley, M. S. (1993). Cost effectiveness of two nurse-led programs to teach breast self-examination. American Journal of Preventive Medicine, 9(3), 139-145.
Pollitt, R. A., Geller, A. C., Brooks, D. R., Johnson, T. M., Park, E. R., & Swetter, S. M. (2009). Efficacy of skin self-examination practices for early melanoma detection. Cancer Epidemiology, Biomarkers & Prevention. 18, 3018-3023. doi: 10.1158/1055-9965.EPI-09-0310.
Richard, M. A., Grab, J. J., Avril, M. F., Delaunay, M., Gouvernet, J., Wolkenstein, P., ... Dalac, S. (2000). Delays in diagnosis and melanoma prognosis (I): The role of patients. International Journal of Cancer, 89, 271-279. doi: 10.1002/10970215(20000520)89:3<271: :aid-ijc10>3,0.co;2-7.
Rigel, D. S., & Carucci, J. A. (2000). Malignant melanoma: Prevention, early detection, and treatment in the 21st century. CA: A Cancer Journal for Clinicians, 50(4), 215-236. doi:10.3322/canjclin.50.4.215.
Saunders, K. J., Pilgrim, C. A., & Pennypacker, H. S. (1986). Increased proficiency of search in breast self-examination. Cancer, 58, 2531-- 2537.
Schwartz, J. L., Wang, T. S., Hamilton, T. A., Lowe, L., Sondak, V. K., & Johnson, T. M. (2002). Thin primary cutaneous melanomas. Cancer, 95, 1562-1568. doi:10.1002/cncr.10880.
Spetch, M., Cheng, K., & Clifford, C. (2004). Peak shift but not range effects in recognition of faces. Learning and Motivation, 35, 221-241. doi: 10.1016/j.lmot.2003.11.001.
Terrace, H. S. (1963). Discrimination learning with and without "errors". Journal of the Experimental Analysis of Behavior, 6(1), 1-27. doi: 10.1901/jeab.1963.6-1.
Titus, L. J., Clough-Gorr, K., Mackenzie, T. A., Perry, A., Spencer, S. K., Weiss, J., ... Ernstoff, M. S. (2012). Recent skin self-examination and doctor visits in relation to melanoma risk and tumour depth. British Journal of Dermatology, 168, 571-576. doi: 10.1111/bjd. 12003.
U.S. Preventive Services Task Force (2009). Screening for skin cancer. Retrieved from http://www.uspreventiveservicestaskforce.org/ uspstf/uspsskca.htm.
Weinstock, M. A., Martin, R. A., Risica, P. M., Berwick, M., Lasater, T., Rakowski, W., ... Dube, C. E. (1999). Thorough skin examination for the early detection of melanoma. American Journal of Preventive Medicine, 17, 169-175. doi: 10.1016/ S0749-3797(99)00077-X.
Yagerman, S., & Marghoob, A. (2013). Melanoma patient self-detection: A review of the efficacy of skin self-examination and patient directed educational efforts. Review of Anticancer Therapy, 1312, 1423-1431. doi: 10.1586/14737140.2013. 856272.
J. R. Miller * D. D. Reed
University of Kansas, Lawrence, KS, USA
T. S. Critchfield
Illinois State University, Normal, IL, USA
D. D. Reed ([mail])
Department of Applied Behavioral Science, University of Kansas, 4048 Dole Human Development Center, 1000 Sunnyside Avenue, Lawrence, KS 66045-7555, USA
J. R. Miller ([mail])
Department of Behavioral Psychology, Kennedy Krieger Institute, 707 N. Broadway, Baltimore, MD 21205, USA
(1) Following the generalization gradients of Fig. 2, it might be objected that using a moderately symptomatic lesion as S+ would result in patients overlooking highly symptomatic lesions (e.g., Fig. 2, right side of middle panel). Here, for two reasons, we suggest that parallels break down between our laboratory model and everyday circumstances. First, our stimuli were presented as abstract shapes, five of any experimentally provided cancer context. In the everyday world, people know about cancer and its potentially deadly consequences. Thus, unusual changes in the skin, once they are noticed, may be alarming. Second, physicians do not generally complain of patients being unable to recognize fairly advanced melanoma as unusual. These factors are likely to leave highly symptomatic lesions as easy to recognize. Of course, this proposition can be experimentally tested, in part by replicating our Standard Education procedures while describing the stimuli as related to skin cancer and modifying the response to explicitly identify possible instances of cancer.
Table 1 Stimuli as S+ and S- During Training for the Three Groups Stimulus Group S1 S50 S99 Control s+ Standard Education S- s+ Reversed Education s+ s- Table 2 Percentage of Participants in Each Group Making Errors to Stimuli During Training Errors No Group Errors SI S50 S99 Control 75 % 25 % Standard Education 38 % 13% 50% Reversed Education 38 % 25 % 38 %
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||ORIGINAL ARTICLE|
|Author:||Miller, Jonathan R.; Reed, Derek D.; Critchfield, Thomas S.|
|Publication:||The Psychological Record|
|Article Type:||Clinical report|
|Date:||Jun 1, 2015|
|Previous Article:||The effects of punishment on resurgence in laboratory rats.|
|Next Article:||Transfer-of-function and novel emergent relations using simple discrimination training procedures.|