Baade PD, English DR, Youl PH, McPherson M, Elwood JM, Aitken JF. The Relationship Between Melanoma Thickness and Time to Diagnosis in a Large Population-Based Study. Arch Dermatol. 2006;142(11):1422-1427. doi:10.1001/archderm.142.11.1422
To examine the relationship between melanoma thickness and reported time from first recognition and from first physician contact to the diagnosis of invasive melanoma.
Telephone survey of patients recently diagnosed as having melanoma, combined with relevant pathological data (including melanoma thickness and morphologic structure) from the population-based Queensland Cancer Registry. A test-retest study (n = 176) was also conducted.
Population-based study in Queensland.
Residents of Queensland (n = 3772) who had been diagnosed as having invasive melanoma between January 1, 2000, and December 31, 2003.
Main Outcome Measures
Prepresentation time (time between first noticing a suspicious spot and the first physician visit), postpresentation time (time between the first physician visit and diagnosis), and total time to diagnosis (time from initial detection of the melanoma to diagnosis).
With 1 exception, we found no significant association between melanoma thickness and reported time to diagnosis for all melanomas combined, superficial spreading melanomas, or nodular melanomas. The exception was a positive association between melanoma thickness and postpresentation delay of physician-detected nodular melanomas. The reliability study gave intraclass correlation coefficients of 0.85 to 0.90 for the measures of intervals.
This large study demonstrates no clear relationship between the melanoma thickness when diagnosed and the time from first recognition of changes or from first physician examination to diagnosis. This may be because of varying biological characteristics of melanomas, as well as methodological limitations of retrospective studies when trying to measure this complex association.
Prognosis from melanoma is dependent on thickness at diagnosis. In Australia, 10-year survival decreases from 98% for lesions less than 0.76 mm thick to 53% for lesions greater than 3 mm thick.1 The outcome for people with distant metastasis is poor, with fewer than 5% of patients surviving for longer than 5 years.2
Public education campaigns encouraging early detection and treatment of melanoma are based on the assumption that melanoma is a slowly progressive disease and that earlier detection will mean less advanced disease at diagnosis. Time to diagnosis can be prolonged if the patient does not consult promptly with the physician after noticing a suspicious spot (prepresentation time) or if there are delays between the initial physician consultation and definitive diagnosis (postpresentation time).
However, a simple association of longer time to diagnosis and increased melanoma thickness has not been generally shown in the results of recent studies. Some studies reported an inverse relationship between prepresentation delay and thickness (in that people with thick melanoma had shorter times to diagnosis),3- 5 or there was no significant relationship.6- 8 Only 1 study9 has reported that increased delay on the part of the patient resulted in increased thickness; however, that study was limited to nodular melanomas. Several additional studies10- 12 reported no significant association between postpresentation delay and tumor thickness.
Possible explanations for these inconsistent results are that most studies were limited by small samples, low response rates, inaccurate measurement tools, or nonrepresentative sampling methods. To further explore the relationship between time before melanoma diagnosis and melanoma thickness at diagnosis, this article reports data from a large population-based study conducted in Queensland, Australia, which has the highest reported incidence of melanoma in the world.13,14
A full description of the methods of this study has been published elsewhere.15 The relevant information is summarized briefly herein.
Queensland residents aged 20 to 75 years who were diagnosed as having histologically confirmed first primary invasive cutaneous melanoma between January 1, 2000, and December 31, 2003, were eligible for this study. For sampling efficiency and cost, all participants with thick melanoma (>0.75 mm) and a random 60% sample of those with thinner melanoma (≤0.75 mm) were included. Patients with metastatic disease, a previous melanoma, or acral lentiginous or noncutaneous melanoma were excluded. After permission was obtained from the treating physician, patients were invited by letter to participate. Ethical clearance for this study was obtained from the University of Queensland Ethical Review Committee.
Of 4839 eligible patients, physician consent was obtained for 4510 (93.2%). Of these, 3887 patients (80.3% of the total eligible sample) agreed to participate, with 3772 (77.9% of the total eligible sample) completing an interview. The median time between diagnosis and interview was 5 months (range, 1-26 months).
Data were collected from respondents via a telephone interview conducted by trained interviewers using a computer-assisted telephone interview system. Interviewers were blinded to specific tumor thickness and to histologic subtypes. This interview collected information on sociodemographics, site of melanoma, and how and when the melanoma was detected. Patients were categorized as to whether their melanoma was first noticed by themselves (or another lay person) or by their physician. Interview data were combined with pathological data in the Queensland Cancer Registry, including melanoma thickness, histologic type, and level of invasion.
Three dates form the basis of this analysis. First, participants were asked, “What date was it when you/they/the physician (as appropriate, according to who first noticed the melanoma) first believed something might be wrong with that spot?” (time T1). Second, those patients whose melanoma was not detected by a physician were then asked, “What date did you first see a physician about this spot?” (time T2). The value of T2 was set equal to T1 for physician-detected melanomas. The third date was the date of definitive diagnosis (ie, excision) (time T3) as reported on the pathology form and as confirmed by the patient at interview.
We calculated 3 measures of time to diagnosis. Prepresentation time was the time (in months) between first noticing a suspicious spot and the first physician visit (T2 minus T1) and included only patients whose melanoma was not detected by a physician. Postpresentation time was calculated for all patients as the time (in months) between the first physician visit and diagnosis (T3 minus T2). Total time to diagnosis was the time (in months) from initial detection of the melanoma to diagnosis (T3 minus T1 for patient-detected lesions and T3 minus T2 for physician-detected lesions).
In acknowledging that the inaccurate recall of dates by respondents may compromise the validity of this study, we conducted a reliability study by reinterviewing 176 of the original respondents 1 to 3 months after the first interview. The intraclass correlation coefficients for prepresentation time, postpresentation time, and total time were 0.90 (95% confidence interval [CI], 0.87-0.93), 0.85 (95% CI, 0.81-0.89), and 0.85 (95% CI, 0.80-0.89), respectively. We also constructed Bland-Altman plots (data not shown) for the 2 dates in question, which showed no evidence of a systematic positive or negative bias in the difference of the dates according to how long ago the events occurred.
Skin self-examination was assessed by asking, “During the 3 years before you first believed something was wrong, had you or someone else who is not a physician deliberately checked all or almost all of your whole body for the early signs of skin cancer?” Similarly, clinical skin examinations was determined by asking, “During the 3 years before you first believed something was wrong, had a physician deliberately checked all or almost all of your whole body for the early signs of skin cancer?” The validity of self-reports of clinical skin examination was quantified in a separate study.16
Lesions were determined to be nonvisible (to the patient) if they were on the scalp, back, buttocks, back of the neck, ears and shoulders, back of the legs, and soles of the feet. All other melanomas were considered to be on visible sites of the body. Similar definitions have been used previously.17- 19 Respondents were also asked whether the lesion was first detected incidentally or as part of a deliberate skin examination.
Respondents were asked whether they had a previous diagnosis of nonmelanoma skin cancer (yes or no), how their skin generally reacts to the sun (always burns and never tans, usually burns and sometimes tans, sometimes burns and usually tans, or never burns and always tans), and the number of moles they have according to a visual diagram provided before the interview (none, few, some, or many). Basic demographics included in the analyses were sex, age (collapsed into 10-year age groups), and educational achievement (primary, grades 7-10, or grade ≥11).
The proportion of thin melanomas was weighted. This was necessary to account for the undersampling (60%) of thin melanomas in this study, in contrast to the complete enumeration of thick melanomas.
To assess the relationship between melanoma thickness and time to diagnosis, we used multiple linear regression analysis, with melanoma thickness (log transformed) as the outcome measure and time to diagnosis as an independent variable. The choice of factors to include as potential confounders was initially based on those that had statistically significant (P<.01) bivariate associations with melanoma thickness (Table 1). We also looked at the Spearman rank correlations (data not shown) between each of those variables and the 3 intervals (prepresentation time, postpresentation time, and total time to diagnosis). Of those variables in Table 1 that were statistically significant, only educational achievement, presence of moles, and histologic type had no statistically significant associations (P<.01) with any of the 3 intervals. We decided to keep histologic type in the model because of its strong association with melanoma thickness. Therefore, the estimates were adjusted for sex, age group, histologic type, method of detection, physician screen in the previous 3 years, skin reaction to summer sun, and who first noticed the melanoma.
Separate regression models were used for each interval. To investigate the potential effect of long intervals on the model, we fitted corresponding models limited to intervals of 12 months or less. We examined the association between melanoma thickness and time to diagnosis separately for superficial spreading melanomas and for nodular melanomas. Where relevant, we considered the associations separately for patient-presented lesions, physician-detected lesions, and all lesions combined.
The exponential parameter estimates from the linear regression model were calculated, along with the 95% CI. For associations with categorical variables, this represents the ratio of the geometric mean of melanoma thickness relative to the reference category. For the continuous measure of delay, this represents the ratio of the geometric mean of melanoma thickness for a 1-month increment in delay. This ratio is expressed as a percentage increment or decrement.
From the test-retest reliability study, we knew that there was error associated with the intervals based on dates collected. Therefore, we used the regression calibration method described by Spiegelman and colleagues.20,21 Regression calibration is a statistical method of adjusting point estimates (and their variance) from regression models for bias due to known measurement error. We used the test-retest data to assess the within-person error for the 3 intervals and assumed that the other variables in the regression model were measured without error.
Records with missing values for the outcome or any of the explanatory variables were listwise deleted from the models. All data were analyzed using SAS version 9.01 (SAS Institute, Cary, NC).22
Nonparticipants were significantly more likely to be men (63.4% vs 57.1%, P<.01) and to have a thick melanoma (10.9% vs 7.2%, P<.01). The median thickness was 0.68 mm for participants and 0.75 mm for nonparticipants. There were no significant differences between participants and nonparticipants relative to age, site, or morphologic structure of melanoma.
For non–physician-detected lesions (n = 2891), the median prepresentation time (between when the lesion was first noticed and the first visit to the physician) was approximately 1 month (interquartile range, 0.1-3.7 months). For all lesions combined (n = 3772), the median postpresentation time (between the first physician consultation and definitive diagnosis) was about 1 week (interquartile range, 0.1-1.4 months). The median total presentation time between first noticing the lesion and definitive diagnosis for all lesions was 1.7 months (interquartile range, 0.4-6.3 months).
Bivariate associations between the range of study variables and melanoma thickness (on a log scale) are given in Table 1. As summarized in Table 2, there was a consistent lack of significant associations between melanoma thickness and the different intervals, after adjustment for sex, age group, histologic type, method of detection, physician screen in the previous 3 years, skin reaction to summer sun, and who noticed the melanoma. The only exception to this was for physician-detected nodular melanomas (n = 41). For these lesions, the thickness was almost 3% greater (2.92% [95% CI, 0.32%-5.58%] per month, P<.05) for each extra month between initial physician identification and the definitive diagnosis.
Apart from this significant association, the lack of significant associations between melanoma thickness and measured intervals held. This was true regardless of whether we limited the analysis to shorter times (≤12 months), superficial spreading melanomas, or nodular melanomas (Table 2).
In a simple model of melanoma as a slowly progressive tumor, we might expect a positive correlation between thickness and time to diagnosis, particularly when considering a single melanoma on a specific individual. This assumption underlies educational campaigns that have been conducted in Australia for some years encouraging awareness of the signs of melanoma among the medical profession and the public and rapid presentation to the physician of suspicious lesions.23
It is unlikely that the lack of association found in this study is due to limited statistical power to detect differences, because this study is by far the largest population-based study that has looked at this association. As noted previously,24 the imprecision in measuring the time between someone first noticing a skin lesion and the final histologic diagnosis is a major limitation in these types of retrospective studies. In this study, we used a consistent structured interview schedule looking at intervals before diagnosis, and we obtained a high response rate. The intraclass correlation coefficients based on reinterviews for 176 respondents were in the range of 0.85 to 0.90, which is high. There was no evidence of systematic positive or negative bias in responses. These results suggest that any systematic error associated with the collection of dates in this study is small. We found no other studies on delay or time to melanoma diagnosis that also reported test-retest reliability data for respondents' recollection about dates.
There are at least 2 reasons why a simple association between melanoma thickness and time to diagnosis would not be seen in a population-based sample. First, it is likely that melanomas progress at different rates. For example, superficial spreading melanomas typically have a longer radial growth phase compared with nodular melanomas, which typically have a faster invasion to lower levels of the dermis.25 Depth of invasion depends on vertical growth, so a lack of association with the time spent in the radial growth pattern can be expected. Some melanomas grow extremely slowly, if at all.26 There is no way to identify these different categories of lesions. Biologically aggressive lesions may be advanced at the time of first recognition of a sign or symptom and may subsequently show changes that will precipitate further intervention, leading to diagnosis. Either of these scenarios would lead to a situation in which the more biologically aggressive lesions will tend to be associated with shorter intervals to diagnosis, and this could distort any countervailing tendency for lesions to grow over time. Within histologic types, there are likely differences in the speed at which specific melanomas grow. Therefore, even within a single histologic type, combining melanomas with different (unknown) rates of growth will confound any measured association between thickness and time to diagnosis.24,27
There was a significant association between depth of invasion and time to diagnosis for only 1 subgroup, nodular melanomas first detected by a physician. Nodular melanoma has a limited or an absent radial growth phase, shows early vertical growth, and likely demonstrates visible change after first recognition,28 so the interval after first recognition is likely to be closely related to the extent of vertical growth. The stronger association seen with physician-detected lesions may be because the starting point for the observed time to diagnosis is related to an underlying biological process that is recognized by a trained observer.
The second reason why a simple association between melanoma thickness and time to diagnosis would not be seen in a population-based sample is that the observed interval is from first recognition of a skin abnormality to diagnosis. The time of first recognition of an abnormality will depend on many factors and may bear no consistent relationship with any measure of the biological growth of the tumor. Let us assume that a melanoma starts at time T0, it is first noticed by the patient or physician at time T1, and it is diagnosed as a melanoma at time T3. All studies looking at thickness and time to diagnosis can only measure dates between T3 and T1 (which may be dependent on tumor-, patient-, and physician-related factors).
This inability to measure the total time to diagnosis has direct implications for the observed association between thickness and measured delay to diagnosis. If the unobserved time (T1 minus T0) was short, then a person diagnosed as having a thick lesion would be likely to have a longer observed time to diagnosis (T3 minus T1), thus resulting in a positive observed correlation between melanoma thickness and observed delay. However, if the unobserved time (T1 minus T0) was long, then the same person diagnosed as having a thick lesion may have had a much shorter observed time (T3 minus T1). This would then give a negative observed correlation between melanoma thickness and measured delay. So the ratio of unobserved delay (T1 minus T0) to observed delay (T3 minus T1) can affect whether there is a positive or negative association between thickness and measured delay, regardless of the true association. In addition, these 2 effects (positive correlation and negative correlation) could potentially cancel each other out across a research study such as this, resulting in a null overall association.
Some evidence of this hypothesis is that the only significant association between melanoma thickness and measured delay in this study was between postpresentation delay and nodular melanomas. This association is probably least affected by the aforementioned scenarios, because the postpresentation period does not include T0 to T1 and because the faster growth of nodular melanomas is probably the most consistent of all the histologic types.
There seems to be no positive association between melanoma thickness and time to diagnosis on a population basis, likely because of the wide range of biological characteristics among melanoma subtypes, combined with the inability to assess events before a patient's or physician's first recognition of an abnormality. For an individual patient, this does not imply that the time to diagnosis does not matter. Most melanomas are progressive and will metastasize in time. However, the lack of a measurable relationship between depth at diagnosis and the intervals before diagnosis in the population should caution us about overinterpretation of the observed effects of education campaigns. Increased public education about heeding the early warning signs of melanoma will produce 2 perhaps contradictory effects. First, educating those with abnormal skin lesions to seek a physician's advice promptly would tend to decrease the interval from first recognition to first physician visit (and reduce the observed time to diagnosis). Second, if the campaigns also make the public more aware of minor changes, they will be recognizing smaller lesions with less abnormality and with greater uncertainty about their significance. In this situation, it may be reasonable for patients or their physicians to wait, perhaps until further changes occur, before seeking a final diagnosis (thus increasing the observed time to diagnosis).
We need to recognize the complex relationships between tumor thickness and intervals to diagnosis, as well as the substantial limitations in these retrospective studies of delay. To quantitatively demonstrate the importance of seeking medical clarification early for suspicious skin lesions, innovative ways of dealing with these issues should be investigated.
Correspondence: Peter D. Baade, PhD, Queensland Cancer Fund, PO Box 201, Spring Hill QLD 4004, Australia (email@example.com).
Financial Disclosure: None reported.
Accepted for Publication: March 24, 2006.
Author Contributions: Dr Baade had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Baade, English, Elwood, and Aitken. Acquisition of data: Youl, McPherson, and Aitken. Analysis and interpretation of data: Baade, English, Elwood, and Aitken. Drafting of the manuscript: Baade, English, McPherson, and Aitken. Critical revision of the manuscript for important intellectual content: Baade, English, Youl, McPherson, Elwood, and Aitken. Statistical analysis: Baade and English. Obtained funding: English, Elwood, and Aitken. Administrative, technical, and material support: Youl, McPherson, Elwood, and Aitken. Study supervision: Aitken.
Funding/Support: The study was funded by grant 112600 from the National Health and Medical Research Council, Australia.