Photochemical & Photobiological Sciences

Climate change and human skin cancer

Jan C. van der Leun^a, Rubén D. Piacentini^bc and Frank R. de Gruijl^d
aEcofys Netherlands, PO Box 8408, NL-3503, RK, Utrecht, The Netherlands. E-mail: j.vanderleun@ecofys.nl
^bInstituto de Física Rosario, CONICET–Universidad Nacional de Rosario, 27 de Febrero 210bis, 2000, Rosario, Argentina. E-mail: ruben@ifir.edu.ar
^cFacultad de Ciencias Exactas, Ingeniería y Agrimensura, Universidad Nacional de Rosario, Rosario, Argentina
^dDermatology, Leiden University Medical Center/S2P, PO Box 9600, NL-2300, RC, Leiden, The Netherlands. E-mail: f.r.de_gruijl@lumc.nl

As part of an inventory of potential interactions between effects of ozone depletion and climate change, a possible effect of ambient temperature on sun-induced skin cancers was suggested. Mouse experiments had shown that increased room temperature enhanced ultraviolet (UV) radiation-induced carcinogenesis; the effective UV dose was increased by 3–7% per °C. The present investigation was aimed at studying a possible temperature effect on human skin cancer. Existing data on the incidence of human skin cancer were analyzed, as available from two special surveys of non-melanoma skin cancer in the United States. The incidence of non-melanoma skin cancer in the ten regions surveyed not only correlated significantly with the ambient UV dose but also with the average daily maximum temperature in summer. For squamous cell carcinoma the incidence was higher by 5.5% (SE 1.6%) per °C and for basal cell carcinoma by 2.9% (SE 1.4%) per °C. These values correspond to an increase of the effective UV dose by about 2% per °C. Although the precise nature of this correlation with temperature requires further studies, it can be concluded that the temperature rises coming with climate change can indeed amplify the induction of non-melanoma skin cancers by UV radiation in human populations.

Higher temperatures accompanying climate change¹ may lead, among many other effects, to increasing incidence of skin cancer in human populations. We suggested this on the basis of old experimental findings in mice that had been exposed to ultraviolet radiation at different room temperatures.² There the same UV exposures induced more skin cancers in a room at higher temperature.^3,4 Our step from mice to humans was based on the premise of a fundamental similarity of UV carcinogenesis in the two species.^5,6 However, there are also important quantitative differences in the carcinogenic processes in mice and man.^6,7 For a good quantitative prediction of the suspected effect, data are needed on the influence of temperature on carcinogenesis in human populations.

Determining the influence of temperature from data on skin cancer incidence in human populations residing in various regions is not easy. The UV radiation is clearly the predominant factor and temperature at most a secondary one. A warm environment by itself does not cause skin cancer and UV radiation can do it at any environmental temperature. Moreover, ambient UV irradiance and temperature are globally climatologically correlated. Going in the direction of the equator UV irradiance and temperature both tend to increase. This general correlation makes it difficult to separate out the possible contribution of temperature. One solution proposed is to compare incidences in populations living at different altitudes. With increasing altitude, the UV irradiance increases while the temperature usually decreases. Such a decoupling of temperature and UV irradiation may give the additional information needed to extract the quantitative influence of temperature. We are planning to carry out such a purposely designed study, but it will take years before the results become available.

Therefore we tried to derive information on the influence of temperature from already available data, in spite of the difficulties mentioned. One of the best collections of data on the incidence of human skin cancer was produced in two National Skin Cancer Surveys in the United States.⁸ Skin cancer data were actively collected in 1971–72 and 1977–78 for non-Hispanic white Caucasians living in ten regions spread over the USA, covering large latitudinal (17.5°) and longitudinal (39.3°) intervals. Active collection of data produces more reliable results than the standard medical registrations, which are not always complete, usually by lacking registration of the most common cancer: basal cell carcinoma. The publication on the two surveys⁸ also had data on the UV irradiance in these regions and we added data on temperature that were available from meteorological sources. Our analysis showed that going in the direction of the equator, summer temperature, UV irradiance and skin cancer incidence on average indeed all increased, but there were also appreciable deviations. These deviations were exploited in our present analysis to distinguish between the UV and temperature effects.

The ten regions where incidence data for skin cancer were collected in the National Skin Cancer Surveys mentioned are given in Table 1. UV irradiances were measured during the 1977–78 Survey with Robertson–Berger meters. These instruments spectral responses resembled the action spectrum of induction of sunburn (i.e., skin redness or erythema), and thus the readings represented biologically effective UV irradiances.^9–11 The action spectrum for UV carcinogenesis¹² is also similar to the spectral response of the Robertson–Berger meter. The readings were given in terms of annual UV-counts, or annual RB-counts.⁸

Extensive temperature data for many measuring-stations are available on the web. For each of the regions considered in the present work, we took temperature data (www.worldclimate.com) from about ten stations and took the average. Typically the temperatures were already given as averages over 30 years, which for our purpose is fine, as the formation of skin cancer usually takes several decades. We selected the average of the maximum daily temperatures for the months of June, July and August. This choice was made because the UV radiation that induces the skin cancer peaks in summer months, near the middle of the day. It is not known whether temperature during the exposure is important, or also the temperature some time after the exposure, either by synergistic interaction with UV radiation in damaging cells (specifically, their DNA) or by influencing biological/enzymatic reactions following the exposure, such as repair processes. Both possibilities are taken into consideration by using the maximum daily temperatures in summer. The summer temperatures we used agree excellently (squared correlation coefficient R² = 0.98; p < 0.0001) with those from another source (http://cdo.ncdc.noaa.gov/cgi-bin/climatenormals/climatenormals.pl). It suits our analysis that the annual UV-counts mentioned are also strongly dominated by the UV radiation during the summer months. The annual UV-counts correlate closely (R² = 0.96; p < 0.0001) with the annual mean of erythemal UV irradiances at noon (i.e., mean UV index) (TOMS data from Nimbus 7 satellite, 1979–80).¹³

The set of environmental data collected in this way is displayed in Table 1. Table 2 summarizes the skin cancer data collected in the ten regions.

Region	Period	BCC males	BCC females	SCC males	SCC females
Seattle	77–78	209.9 (6.6)	125.2 (4.5)	46.6 (3.2)	16.1 (1.6)
Minneapolis/St.Paul	77–78b	213.1 (5.5)	144.0 (3.9)	36.6 (2.3)	11.8 (1.1)
Detroit	77–78	142.1 (3.1)	97.0 (2.3)	30.0 (1.5)	11.0 (0.8)
Iowa	71–72	c154.0 (9.2)	d90.3 (9.0)	e50.8 (2.7)	e14.3 (1.2)
Utah	77–78	327.4 (8.5)	197.7 (6.0)	123.1 (5.3)	45.9 (2.9)
San Francisco/Oakland	77–78b	239.0 (4.4)	145.1 (3.0)	56.3 (2.2)	18.4 (1.1)
New Mexico	77–78	571.7 (25.7)	339.8 (17.0)	180.2 (14.6)	63.4 (7.3)
Atlanta	77–78	423.2 (10.3)	228.5 (6.3)	131.0 (6.0)	52.6 (3.0)
Dallas	71–72	c491.8 (29.5)	d269.9 (24.3)	e144.7 (6.5)	e54.4 (3.4)
New Orleans	77–78	409.5 (12.0)	215.4 (7.3)	153.4 (7.4)	48.8 (3.4)
a Incidences of BCC (basal cell carcinoma) and SCC (squamous cell carcinoma) in terms of annual age-standardized rates per 100000 of the white population (SE in brackets).8  b Also included in 1971–1972 survey.  c 71–72 data increased by 25 (±5)% according to geometric mean rise in Minneapolis and San Francisco.  d 71–72 data increased by 32 (±9)% according to geometric mean rise in Minneapolis and San Francisco.  e Between 71–72 and 77–78 practically no differences in SCC incidences in Minneapolis and San Francisco.

Both maximum summer temperature and annual UV-counts showed significant trends with geographic latitude: −0.5 ± 0.2 °C/°latitude (p = 0.027) and −3.6 ± 0.6 %UV/°latitude (p < 0.001), respectively (with SE following the ± symbol). But the data points deviated substantially from the regression lines (R² = 0.48 and 0.80, respectively) (data not shown). Despite these common trends with latitude, there was no significant correlation between summer temperature and annual UV-counts (0.09 ± 0.05 °C/%UV, p = 0.12) (data not shown). This lack of correlation between temperature and UV radiation should facilitate identification of a temperature effect, separate from the dominant UV effect.

The incidences of basal cell carcinoma and squamous cell carcinoma are depicted in Fig. 1 in a double log-plot against annual UV-counts.

	Fig. 1 Double log plots of incidences versus deviation of annual UV dose from the geometric mean (140.0 × 104 RB-counts): (a) SCC incidences for males and females, (b) BCC incidences for males and females. The error bars represent SE in ln(incidence) (i.e. relative errors from Table 2). The regression lines were fitted to the male (solid line) and female (dashed line) data simultaneously by eqn (1) using the Statistical Package for Social Sciences version 11.0, with inverted squared relative errors from Table 1 as weights (i.e. a Poisson regression). For SCC: a = 2.94 ± 0.42, b = 3.31 ± 0.13 for women and b is 1.04 ± 0.16 higher for men. For BCC: a = 1.64 ± 0.26, b = 5.08 ± 0.08 for women and 0.54 ± 0.11 higher for men. These fitted parameters all differ significantly from 0 (p < 0.001).

The incidences of SCC and BCC (Fig. 1) are clearly related to the annual UV-counts, as found earlier,^7,14 and the males and females data points run parallel. Incidences for both genders were fitted simultaneously to UV-counts for SCC and BCC separately according to the equation:

ln(incidence) = aln(UV) + b

(1)

As can be seen in Fig. 1, there remained considerable residuals between the fitted lines and the data points (R² = 0.84 for SCC and 0.80 for BCC). These residuals are now plotted versus T, the difference between the average maximum summer temperature for a region and the mean of those temperatures over the ten regions (28.5 °C), as is seen in Fig. 2. The fitted lines give:

residual ln(incidence) = aT + b

(2)

where a represents the fractional increase in incidence per °C. Thus, we find that SCC rises by (4.4 ± 1.5)% per °C and BCC by (2.5 ± 1.2)% per °C. The increase in SCC differs significantly from 0 (p = 0.008), but that in BCC is of borderline significance (p = 0.054).

	Fig. 2 The residuals in ln(incidence) for males and females versus T (the deviation in T from the mean, 28.5 ° C): (a) for SCC, (b) for BCC. The solid line depicts the results of the weighted least squares regression (eqn (2)). The error bars represent SE in ln(incidence). For SCC a = 0.044 ± 0.015 (p = 0.008), and for BCC a = 0.025 ± 0.012 (p = 0.054); b for either SCC or BCC does not differ significantly from 0.

Fig. 2 shows that the West Coast data points from Seattle (upper left point) and San Francisco (lower left point) with the lowest summer temperatures are of great importance to the regression, especially San Francisco with its relatively low SCC incidences and smaller errors appears to act as a leverage point in Fig. 2(a).

Instead of fitting the residuals in ln(incidence) to maximum summer temperature after regression on ln(annual UV dose), both variables, ln(UV) and T, should be fitted simultaneously to ln(incidence), according to the equation

ln(incidence) = aln(UV) + bT + c

(3)

For SCC we then find BAF = a = 2.38 ± 0.37, and a (5.5 ± 1.6)% increase in incidence per °C (i.e., b = 0.055 ± 0.016, and c = 3.30 ± 0.11 for women and c is 1.04 ± 0.13 higher for men). For BCC we find BAF = a = 1.44 ± 0.25 and a 2.9 ± 1.4% increase in incidence per °C (i.e., b = 0.029 ± 0.014, and c = 5.12 ± 0.07 for women and c is 0.53 ± 0.10 higher for men). These fitted coefficients all differ significantly from zero (p < 0.001, except for b, with p = 0.004 for SCC and p = 0.049 for BCC). By inclusion of T, the residuals are markedly reduced in comparison to fitting eqn (1), but they remain considerable, especially for BCC (with T included we get R² = 0.91 for SCC and 0.85 for BCC).

Ambient UV radiation is clearly a dominant factor in skin carcinoma risk, as reflected in highly significant regressions (Fig. 1). Nevertheless, ambient UV radiation explains only about 80% of the variance in the incidence, and the large remaining variance is most likely due to other risk modifying factors, such as genetic background (minimized by restricting the data to fair-skinned Caucasians), exposure behavior (sun seeking or not) and possibly ambient temperature. Mainly owing to the substantial summer temperature differences between the West coast and other parts of the USA with comparable ambient UV loads, we found evidence that some of the remaining variance in incidence may be attributed to temperature, with a (5.5 ± 1.6)%/°C increase in SCC incidence, and a (2.9 ± 1.4)%/°C increase in BCC incidence (raising the explained part of the variances from 84 to 91 and from 80 to 85%, respectively). These figures would imply, for instance that with a long-term increase of summer temperatures by 2–4%, we could end up with substantial increases in incidence of skin carcinomas, which have already risen to very high numbers over the last century in white Caucasians (>1 million cases a year in the USA).¹⁵

We found that taking into account a temperature effect slightly reduced the dependency of the incidences on ambient UV radiation, i.e., the BAF for SCC sank from 2.9 to 2.4, and for BCC from 1.6 to 1.4% increase per % increase in annual UV dose (comparing a from eqn (1) to eqn (3)). This reduction reflects that UV radiation and temperature are to some extent correlated, although not significantly in the present data set.

The incentive to investigate whether ambient temperature had an effect on incidences of skin carcinomas stemmed from mouse experiments already performed in the 40s² and 60s³ of the previous century. We estimated from these data that for each °C above 20 °C the impact of UV irradiation was raised to the equivalent of raising the UV dose by 3–7% at constant temperature.² For the human populations investigated here, we find an equivalent impact of about a 2% increase in effective UV dose per °C (the ratio of a over b from eqn (3), which comes out similar for SCC and BCC with (2.3 ± 0.8%) and (2.0 ± 1.0%) per °C, respectively, with a weighted average of 2.2% per °C); i.e. a temperature effect in the same direction and of the same order of magnitude as in the mouse experiments.

Evidently, the human data are more variable and confounded than the data obtained from the mouse experiments. Especially, human exposure behavior may introduce random or systematic modifications that could mask any underlying physiological temperature effect, or could even cause a temperature effect. In a relatively cold area, people may go outdoors more at higher temperatures, but the opposite may be more likely in a relatively warm area. Although we appear to have found a temperature effect similar to what was found in mice, further studies are needed to disentangle possible behavioral from physiological effects.

A better understanding of the mechanism of the physiological temperature effect, such as observed in the mouse experiments on UV carcinogenesis, should guide future studies in defining the proper metric for the temperature effect: is it the average temperature during and/or after exposure, or do extremes weigh more heavily?

In conclusion, the present study found evidence of an effect of environmental temperature on human skin cancer incidence; whether the effect is physiological in nature or not, it does suggest a possibly substantial effect of the temperature changes coming with climate change. For a more precise quantitative determination of this effect, studies will be needed that are purposely directed at exploiting data from regions with unusual combinations of ambient temperature and UV irradiance.