- Email: [email protected]
Incidence and prediction of nonmelanoma skin cancer post-renal transplantation: A prospective study in Queensland, Australia
Transplantation
Incidence and Prediction of Nonmelanoma Skin Cancer Post–Renal Transplantation: A Prospective Study in Queensland, Australia Robert P. Carroll, MB, ChB, Helen M. Ramsay, MD, Anthony A. Fryer, MRCPath, Carmel M. Hawley, MD, David L. Nicol, MD, and Paul N. Harden, MD ● Background: Nonmelanoma skin cancer (NMSC) is a significant clinical problem after renal transplantation, particularly in areas of high UV light exposure. A single-center prospective study of a population of Queensland renal transplant recipients was performed with the aims of: (1) establishing NMSC incidence and tumor accrual post–renal transplantation, and (2) developing a clinically derived predictive index to identify transplant recipients at greatest risk. Methods: Three hundred ten of 398 transplant recipients (78%) who underwent baseline assessment between July 1999 and April 2000 were reassessed a mean of 18 ⴞ 3.5 (SD) months later. A structured interview, full skin examination, biopsy of suspicious lesions, and review of medical and pathological records were used to determine the number and types of NMSC arising between the two assessments. Incidence (percentage of the population developing NMSC per year) and tumor accrual (number of tumors per person per year) were calculated. A clinically derived predictive index was generated using stepwise logistic regression models. Results: Overall NMSC incidence was 28.1% and increased with duration of immunosuppression therapy: 18.8%, 24.8%, 33.3%, and 47.1% at less than 5, 5 to 10, 10 to 20, and greater than 20 years of immunosuppression therapy, respectively. Mean NMSC accrual was 1.85 ⴞ 3.84 tumors/person/y, increasing to 3.35 ⴞ 4.29 tumors/person/y after 20 years of immunosuppression therapy. Renal transplant recipients were stratified into categories of high and low NMSC risk by using predictive indices. Conclusion: Clinically derived predictive indices can allow NMSC risk stratification of an Australian transplant population and may provide an evidence-based and cost-effective approach to developing a targeted clinical NMSC surveillance program. Am J Kidney Dis 41:676-683. © 2003 by the National Kidney Foundation, Inc. INDEX WORDS: Australia; immunosuppression; incidence; kidney transplantation; nonmelanoma skin cancer (NMSC); predictive index.
N
ONMELANOMA skin cancer (NMSC) is a significant clinical problem after renal transplantation, with a predominance of squamous cell carcinomas (SCCs).1-4 Tumors behave more aggressively, develop at an earlier age, and are more likely to metastasize.5 Fair-skinned renal allograft recipients residing in Queensland, Australia, have the highest NMSC risk in the world, with cumulative incidences previously reported as 45% and 70% at 11 and 20 years posttransplantation, respectively.6 We previously From the Departments of Renal Medicine and Dermatology; Clinical Biochemistry Research Group, School of Postgraduate Medicine, Keele University, North Staffordshire Hospital, Stoke-on-Trent, Staffordshire, UK; and the Renal Transplant Unit, Princess Alexandra Hospital, Brisbane, Queensland, Australia. Received May 23, 2002; accepted in revised form September 26, 2002. Supported in part by the St John Ambulance Travelling Fellowship in Transplantation. Address reprint requests to Paul N. Harden, MD, Oxford Kidney Unit, The Churchill Hospital, Oxford OX3 7LJ, UK. E-mail: [email protected] © 2003 by the National Kidney Foundation, Inc. 0272-6386/03/4103-0017$30.00/0 doi:10.1053/ajkd.2003.50130 676
showed that in this region, 32.6% of the transplant population will have experienced at least one NMSC after 5 years of immunosuppression therapy.7 There are no prospectively gathered data regarding NMSC incidence (total number of patients affected by NMSC per year) or tumor accrual (number of tumors per patient per year) in this population; however, previous studies have relied on retrospective, cross-sectional, or registry data. There currently is a limited structured program for skin cancer surveillance after renal transplantation in Queensland. To facilitate the development of such strategies, an accurate picture of the scale of the clinical problem is required in terms of both incidence and tumor accrual. Accordingly, a prospective study was performed to establish accurate data on incidence and tumor accrual in a population of Queensland renal transplant recipients. Furthermore, a set of clinical risk factors was used to develop a predictive index that may allow the identification of recipients at greatest risk. Together, these data provide for the first time an evidenced-based means to develop targeted skin surveillance and determine whether current
American Journal of Kidney Diseases, Vol 41, No 3 (March), 2003: pp 676-683
PREDICTION OF SKIN CANCER POSTTRANSPLANT
American Society of Transplantation (AST) guidelines regarding the outpatient surveillance of renal transplant recipients are applicable to this population.8 PATIENTS AND METHODS Baseline assessment of 398 of 435 adult renal transplant recipients (91.5%) attending the Princess Alexandra Hospital (PAH), Brisbane, Australia, was performed between July 1999 and April 2000. This consisted of a single-observer interview, full skin examination, biopsy of suspicious lesions, and review of medical and/or pathological records to determine the number, types, and risk factors for NMSC posttransplantation. This study, including demographic details of the study population, has been reported in detail previously.7,9 Three hundred ten of 398 transplant recipients (78%) were reassessed a mean of 18 ⫾ 3.5 (SD) months later by a second observer blinded to previous dermatological records. Eighty-eight transplant recipients were not available for a second interview (3 patients withheld consent, 10 patients had died, 35 patients had been transferred for follow-up at peripheral nephrological centers, and 40 patients were unavailable for review). This assessment consisted of a structured interview incorporating a second full skin examination. Potentially malignant lesions or lesions with atypical features were referred to the Dermatology Department at the PAH for histological confirmation of diagnosis and treatment. When dermatological management occurred outside the PAH, details were obtained from the attending practitioner using a standard proforma. Clinically diagnosed (eg, those treated by ablative methods without previous biopsy) and histologically confirmed tumors were recorded separately. Medical and pathological records since first assessment were reexamined. The second observer received initial training from the dermatologist who performed the baseline assessment to ensure consistency in terms of diagnostic criteria and documentation. Clinical characteristics and maintenance immunosuppression regimens are listed in Table 1. Demographic information and clinical risk factors were gathered by structured interview and case note review at baseline assessment. Briefly, data gathered included age, sex, smoking history, skin type I through VI (original Fitzpatrick classification: skin type I corresponds to fair skin, which always burns and never tans, through skin type VI, which is permanently deeply pigmented), and natural hair and eye color at the age of 21 years. UV light exposure was estimated using a number of parameters, including occupational exposure, calculated cumulative sun exposure based on average hours spent outdoors throughout life, sunbathing habits (assigned a sunbathing score), and recalled childhood sunburn episodes. Data also were gathered on country of birth, residence in temperate/tropical/subtropical climates, and chemical carcinogen exposure (tobacco smoke and arsenic). Incidence (percentage of the population developing NMSC per year) was calculated by dividing the total number of patients developing NMSC per year by the total number of patients reexamined (n ⫽ 310). Mean tumor accrual (number of tumors per person per year) was calculated by
677 Table 1. Patient Characteristics in 310 Renal Transplant Recipients Included in the Incidence Analysis Patient Characteristics
Caucasian Male sex No. of allografts 1 2 3 Donor (current allograft) Cadaveric Live related Live unrelated Immunosuppression regimen Triple therapy* Cyclosporine/azathioprine* Cyclosporine/prednisolone* Azathioprine/prednisolone* Others* Cyclosporine* (mg/d) Azathioprine* (mg/d) Prednisolone* (mg/d) Age at 1st transplantation stratified by duration of immunosuppression (y) ⬍5 (n ⫽ 103) 5-10 (n ⫽ 73) 10-20 (n ⫽ 106) ⬎20 (n ⫽ 28) Prospective skin cancer surveillance stratified by duration of immunosuppression (y) ⬍5 (n ⫽ 103) 5-10 (n ⫽ 73) 10-20 (n ⫽ 106) ⬎20 (n ⫽ 28)
284 (91.6) 191 (61.2) 267 (86.1) 32 (10.3) 11 (3.5) 261 (86.7) 38 (12.6) 2 (0.7) 126 (44.3) 34 (12.0) 26 (9.2) 42 (14.8) 56 (19.7) 225 (150-300) 87.5 (0-125) 7 (5-10)
47.2 ⫾ 12.9 42.7 ⫾ 15.4 39.1 ⫾ 12.8 30.1 ⫾ 11.1
1.54 ⫾ 0.2 1.53 ⫾ 0.3 1.53 ⫾ 0.2 1.53 ⫾ 0.3
NOTE. Values expressed as number (percent); median (interquartile range); or mean ⫾ SD. *Immunosuppression data at 1 year posttransplantation available on 284 patients. Other immunosuppressants used included tacrolimus, sirolimus, and mycophenylate.
dividing the sum of the number of tumors per year for each transplant recipient by the total number of recipients examined at second interview (n ⫽ 310). The PAH Research Ethics Committee approved this study, and all participants provided written informed consent.
Generation of Predictive Indices Only histologically confirmed tumors were included in statistical analyses. Predictive indices were generated using stepwise logistic regression models (Stata Statistical Software, version 7; Stata Corp, College Station, TX) with a probability cutoff value for inclusion in the models of P of 0.1 or less. In stepwise analyses, an additional 22 patients who were unavailable for clinical examination who developed histologically proven NMSC since baseline also were
678
CARROLL ET AL
included. These additional patients were not included in the calculation of incidence figures. Of the resulting 332 patients included in stepwise analyses, 26 patients of nonCaucasian origin (skin types V and VI) were excluded because of constitutive protection against skin cancer. The remaining 306 patients were randomly assigned (using the random number generator function in Stata) into two groups in a 2:1 ratio: an exploratory cohort (n ⫽ 204) and a confirmatory cohort (n ⫽ 102). Our strategy was to generate predictive indices using the exploratory group and test these in the confirmatory cohort. The following potential predictors were included in the stepwise models: age at transplantation, current age, duration of immunosuppression, male sex, smoking history (ever versus never and pack-years), possible/probable arsenic exposure, birth in a hot climate, years in an outdoor occupation, occasional/frequent childhood sunburn, sunbathing score, cumulative sun exposure, skin type I, blue/hazel eyes, red hair, previous history of SCC, basal cell carcinoma (BCC), solar (actinic) keratoses (AK), keratoacanthoma (KA), in situ SCC (Bowen’s disease [BD], considered separately), viral warts, and number of transplants. Identification of optimal scores for the predictive indices was determined by plotting sensitivity and specificity against probability cutoff. Comparison of models was performed using receiver operating characteristic curves. Separate indices were generated for risk for BCC, SCC, and total NMSC (including BD and KA).
RESULTS
Table 2 lists numbers of patients and tumors identified between the first and second assessments. Of 131 patients who developed NMSC during the follow-up period, 77 patients (58.8%) developed more than one tumor type. The ratio of SCC (invasive plus in situ) to BCC was 3.1:1. Importantly, 129 of 850 NMSCs (15.2%) occurred on the relatively photo-protected skin of the trunk and proximal limbs. Three new cases of melanoma also were identified at the second assessment. An additional 106 NMSCs were diagnosed on clinical criteria alone, composed of 42 BCCs, 23 SCCs, 20 BDs, and 17 KAs. These Table 2. Histologically Confirmed NMSCs Identified During the 18-Month Follow-Up Period
BCC Invasive SCC In situ SCC (BD) KA Miscellaneous* Total NMSCs
No. of Patients (%)
No. of Tumors
76/310 (24.5) 89/310 (28.7) 76/310 (24.5) 29/310 (9.4) 3/310 (1.0) 131/310 (42.3)
197 373 241 36 3 850
*One each of vaginal intraepithelial neoplasm III, liposarcoma, and spindle cell fibroxanthoma.
tumors were excluded from all subsequent statistical comparisons. Between the first and second assessments, 12 patients had developed NMSC for the first time. Five patients required nine reconstructive plastic surgical procedures to repair defects after excision of NMSC. Three patients developed metastatic SCC after primary treatment; 2 of these patients underwent axillary lymph node dissection and local radiotherapy, and 1 patient had hepatic metastasis. One patient died of metastatic melanoma. Overall NMSC incidence was 28.1% (21.4% for women and 31.8% for men) and increased with duration of immunosuppression (Fig 1). Considering individual tumor types, the incidence of invasive SCC was 19.1%; BD, 16.3%; BCC, 16.3%; and KA, 6.2%. Mean NMSC accrual during the 18-month follow-up was 1.85 ⫾ 3.84 (SD) tumors/patient/y. Mean accruals for invasive SCC, BD, BCC, and KA were 0.82 ⫾ 2.23, 0.52 ⫾ 1.37, 0.42 ⫾ 1.08, and 0.08 ⫾ 0.30 tumors/patient/y, respectively. As with incidence, accrual increased with duration of immunosuppression (Fig 2). Generation of Predictive Indices BCC predictive index. Using the exploratory cohort, the stepwise model used to predict BCC risk within the follow-up period selected previous history of BCC (P ⬍ 0.001; odds ratio [OR], 6.3), previous history of AK (P ⫽ 0.011; OR, 8.0), birth in a hot climate (P ⫽ 0.005; OR, 23.5), male sex (P ⫽ 0.002; OR, 4.5), and previous history of BD (P ⫽ 0.019; OR, 3.1) as the significant set of predictors. Table 3 shows the predictive index equation generated. The sensitivity and specificity against probability cutoff plots allowed generation of the score from the predictive index that best differentiated between BCC-positive and BCC-negative cases. Using this optimum score cutoff, sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy of the predictive index were calculated (Table 3). OR, 95% confidence intervals (CIs), and significance level using this cutoff are also listed in Table 3. This predictive index then was applied to the confirmatory cohort, and results obtained are listed in Table 3. As expected, data in the confir-
PREDICTION OF SKIN CANCER POSTTRANSPLANT
679
Fig 1. Incidence (percentage of population per year) of NMSC and duration of immunosuppression in 310 renal transplant recipients. (F) BCC; (}) SCC; (Œ) BD; (■) KA; (䊐) ALL.
matory cohort are not as good at predicting BCC risk as those of the exploratory cohort, particularly with reference to sensitivity and positive predictive value. However, this suggests that although the index will indicate screening of patients who will not develop subsequent BCC, most of those who will develop tumors will be included in the screened set. SCC predictive index. In the exploratory group, the model used to predict SCC risk within
the follow-up period selected previous history of SCC (P ⬍ 0.001; OR, 17.8), blue/hazel eyes (P ⫽ 0.040; OR, 3.4), ever smoker (P ⫽ 0.040; OR, 2.8), male sex (P ⫽ 0.003; OR, 4.6), previous history of KA (P ⫽ 0.071; OR, 2.7), and previous history of BD (P ⫽ 0.075; OR, 2.6) as the significant set of predictors. The predictive index–generated equation is listed in Table 3. The cutoff value used generated a particularly large OR: 42.8 (95% CI, 17.1 to 110.7; P ⬍
Fig 2. Mean tumor accrual against duration of immunosuppression for 310 patients. (F) BCC; (}) SCC; (Œ) BD; (■) KA; (䊐) ALL.
680
CARROLL ET AL Table 3.
Generation and Validation of Predictive Indices Using Stepwise Logistic Regression Models in the Exploratory and Confirmatory Cohorts BCC
Predictive index ⫽
Exploratory cohort Predictive index score Proportion with tumor Sensitivity (%) Specificity (%) Positive predictive value (%) Negative predictive value (%) Accuracy (%) OR 95% CI P Confirmatory cohort Predictive index score Proportion with tumor Sensitivity (%) Specificity (%) Positive predictive value (%) Negative predictive value (%) Accuracy (%) OR 95% CI P
(B * 1.79) ⫹ (I * 1.16) ⫹ (A * 1.66) ⫹ (M * 1.49) ⫹ (H * 1.66)
SCC
(S * 2.60) ⫹ (I * 1.09) ⫹ (T * 1.02) ⫹ (M * 1.67) ⫹ (E * 1.00) ⫹ (K * 0.97)
NMSC*
(S * 1.21) ⫹ (I * 1.44) ⫹ (B * 0.92) ⫹ (M * 1.32) ⫹ (E * 0.69) ⫹ (A * 0.95) ⫹ (H * 1.60)
⬍8.3 ⱖ8.3 ⬍5.4 ⱖ5.4 ⬍4.8 ⱖ4.8 10/133 (7.5%) 43/64 (67.2%) 10/130 (7.7%) 57/73 (78.1%) 19/111 (17.1%) 75/86 (87.2%) 81.1 85.1 79.8 85.4 88.2 89.3 67.2 78.1 87.2 92.5
92.3
82.9
84.3 25.2 10.3-63.7 ⬍0.0001
87.2 42.8 17.1-110.7 ⬍0.0001
84.8 33.0 13.9-80.6 ⬍0.0001
⬍8.3 ⱖ8.3 ⬍5.4 ⱖ5.4 ⬍4.8 ⱖ4.8 12/72 (16.7%) 16/29 (55.2%) 8/72 (11.4%) 23/32 (71.9%) 15/58 (25.9%) 35/43 (81.4%) 57.1 74.2 70.0 82.2 87.3 84.3 55.2 71.9 81.4 83.3
88.6
74.1
75.2 6.2 2.1-17.9 0.0001
83.3 19.8 6.1-66.3 ⬍0.0001
77.2 12.5 4.4-37.6 ⬍0.0001
Abbreviations: A, preexisting solar keratoses (1 if any present, 0 if absent); B, preexisting BCC (1 if any present, 0 if absent); C, sunburning during childhood (1, if occasional/frequent, 0 if never/rare); E, eye color (1 if blue/hazel, 0 if brown/green); H, birth in hot climate (1 if hot country, 0 if temperate, etc); I, preexisting BD (1 if any present, 0 if absent); K, preexisting KA (1 if any present, 0 if absent); M, sex (1 if male, 0 if female); S, preexisting SCC (1 if any present, 0 if absent); T, smoking status (1 if ever, 0 if never). *NMSC ⫽ BCC ⫹ SCC ⫹ BD ⫹ KA combined.
0.0001). In the confirmatory cohort, the predictive index also was very accurate at identifying high-risk patients (Table 3), with high values for both sensitivity and specificity obtained. NMSC predictive index. A further predictive index was generated given that total risk for NMSC, rather than that for individual tumor types, will determine surveillance strategy. In the exploratory cohort, the model used to predict total NMSC risk within the follow-up period selected previous history of SCC (P ⫽ 0.012; OR, 3.5), previous history of BCC (P ⫽ 0.039; OR, 2.6), male sex (P ⫽ 0.004; OR, 3.8), birth in a hot climate (P ⫽ 0.031; OR, 5.3), blue/hazel
eyes (P ⫽ 0.042; OR, 2.6), previous history of AK (P ⫽ 0.055; OR, 2.8), and previous history of BD (P ⫽ 0.007; OR, 4.0) as the significant set of predictors. This generated the equation: Predictive index ⫽ 共S * 1.21兲 ⫹ 共I * 1.44兲 ⫹ 共B * 0.92兲 ⫹ 共M * 1.32兲 ⫹ 共E * 0.69兲 ⫹ 共A * 0.95兲 ⫹ 共H * 1.60兲 where S is preexisting SCC, I is preexisting BD, B is preexisting BCC, M is sex, E is eye color, A is preexisting AK, and H is birth in a hot climate, as listed in Table 3. This index then was applied to the confirmatory cohort (Table 3).
PREDICTION OF SKIN CANCER POSTTRANSPLANT Table 4.
Score
681
Association of Score From the Simplified Predictive Index for Total NMSC With Tumor Risk and Accrual Proportion With Tumor
Exploratory cohort 0 0/2 (0.0%) 1 0/4 (0.0%) 2 0/9 (0.0%) 3 6/40 (15.0%) 4 7/31 (22.6%) 5 7/26 (26.9%) 6 22/30 (73.3%) 7 26/28 (92.9%) 8 26/27 (96.3%) Confirmatory cohort 0 0/1 (0.0%) 1 0/3 (0.0%) 2 2/10 (20.0%) 3 2/16 (12.5%) 4 3/14 (21.4%) 5 10/16 (62.5%) 6 10/15 (66.7%) 7 16/19 (84.2%) 8 7/7 (100.0%)
Mean NMSC/y
Mean BCC/y
Mean SCC/y
OR
95% CI
0.00 0.00 0.00 0.18 0.33 0.31 1.81 2.65 9.07
0.00 0.00 0.00 0.05 0.17 0.03 0.22 0.42 2.00
0.00 0.00 0.00 0.07 0.04 0.12 0.90 1.49 4.27
} }1.0 } } 2.4 3.0 22.5 106.2 212.3
— — — — 0.7-7.9 0.9-10.1 7.0-72.5 20.0-563.7 24.2-1859.2
0.00 0.00 0.24 0.10 0.12 1.01 1.44 2.39 3.90
0.00 0.00 0.11 0.03 0.09 0.18 0.41 0.34 1.15
0.00 0.00 0.00 0.03 0.00 0.59 0.80 1.04 1.69
} }1.0 } } 1.8 10.8 13.0 34.7 †
— — — — 0.3-9.3 2.5-46.7 2.9-58.7 6.9-175.4 †
P
Reference*
0.155 0.075 ⬍0.001 ⬍0.001 ⬍0.001
Reference*
0.498 0.001 0.001 ⬍0.001 †
NOTE. Scores derived from the simplified equation: Predictive index ⫽ A ⫹ B ⫹ E ⫹ (2 * H) ⫹ I ⫹ M ⫹ S, where A is preexisting solar keratoses (1 if any present, 0 if absent), B is preexisting BCC (1 if any present, 0 if absent), E is eye color (1 if blue/hazel, 0 if brown/green), H is birth in hot climate (1 if hot country, 0 if temperate, etc), I is preexisting BD (1 if any present, 0 if absent), M is sex (1 if male, 0 if female), and S is preexisting SCC (1 if any present, 0 if absent). NMSC includes SCC, BCC, BD, and KA. *Reference category is groups 0, 1, 2, and 3 combined. †OR could not be generated because of the absence of cases without tumors in this group.
This model was simplified further for clinical use by reducing the equation to: Predictive index ⫽ A ⫹ B ⫹ E ⫹ 共2 * H兲 ⫹ I ⫹ M ⫹ S 共Table 4兲 Using a cutoff score of five or less versus less than five gave parameters similar to the more detailed equation (sensitivity, 78.7%; specificity, 89.3%; positive predictive value, 87.1%; negative predictive value, 82.1%; accuracy, 84.3%; OR, 30.9; 95% CI, 13.1 to 75.1; P ⬍ 0.0001; Table 3). However, in both the exploratory and confirmatory cohorts, the probability of developing an additional NMSC increased dramatically with increasing score (Table 4). Using these data, it was possible to calculate mean accrual for BCC, SCC, and total NMSC (tumors per year) for each score. This showed that mean numbers of BCCs and SCCs per year increased more rapidly in patients with a score greater than five. Patients with the maximum score showed very
large accrual rates, particularly for SCC. Table 4 also shows the relative increase in risk compared with patients with the minimum score (zero) in the exploratory group. Those with a score of seven showed a 106-fold increased risk compared with those with a score of zero, whereas those with the maximum score (eight) were more than 200 times more likely to develop an additional NMSC. Similar results were obtained when the simplified score was applied to the confirmatory cohort (Table 4). DISCUSSION
There were 5,235 transplant recipients alive with a functioning renal allograft in Australia at risk for skin cancer in December 2000 (Australia and New Zealand Dialysis and Transplantation Registry [ANZDATA]). To our knowledge, this is the first time prospectively gathered data on NMSC incidence and accrual after renal transplantation in an Australian population has been reported. We describe an annual incidence rate of
682
NMSC in Queensland of 28.1%, increasing to greater than 45% in those who underwent transplantation for more than 20 years. Cumulative NMSC incidences of 45% and 70% at 11 and 20 years posttransplantation found in previous studies have been retrospectively derived from ANZDATA.6 It is important to recognize that there are significant limitations in the completeness of reporting of NMSCs to ANZDATA and other cancer registries because of the large numbers of cases, treatment of cases without histological diagnosis, and fragmentation of patient care.7,10-12 In the same population, we previously showed that 28.4% of Caucasian transplant recipients developing NMSC since transplantation were not documented by ANZDATA.9 Furthermore, cumulative incidence is a population-based figure, and alone, it provides no indication of number of individuals in that population affected per year. For the first time, we generated incidence data that, unlike cumulative incidence, provide a dynamic measure of the scale of the problem to be dealt with on an annual basis by units managing renal transplant recipients in Queensland. However, we recognize that a potential limitation is that these data are derived from two observational dates and ideally require confirmation during subsequent years. Individual burden of disease also has been quantified with a mean tumor accrual rate of 1.85. These data are unavailable from ANZDATA because this registry records only the first episode of BCC and SCC posttransplantation. Distribution of accrual in this study is skewed; most have an accrual of less than 3 tumors/y (mean, 4.34 tumors/y; median, 2.43 tumors/y). However, a significant minority (n ⫽ 36) had accrual in excess of 5 tumors/y, and some patients showed very large accrual rates (maximum NMSC accrual, 24.5 tumors/y). Assessment of tumor accrual therefore provides an important indicator of individual morbidity and allows the identification of a subset of patients who require intensive surveillance. The AST recommends that all renal transplant recipients in North America be examined annually for skin cancer,8 where the reported prevalence of NMSC post–renal transplantation is 18%.13 In Queensland, the cumulative incidence is 51.8%, and currently, skin cancer surveillance
CARROLL ET AL
is limited, with most high-risk patients seen in dermatology clinics.9 Similarly, in northern Europe, only 19% of units following up renal transplant recipients offer annual full skin examination for skin cancer.14 It is not clear whether the AST guidelines can be applied to areas of high (Australia) or low (northern Europe) exposure to UV light. We previously used clinical risk factors to predict tumor risk in individuals after renal transplantation in a UK population.15 We have now used a similar method to generate a predictive index applicable to Queensland renal transplant recipients. The high incidence of NMSC in this population has allowed us to randomly assign the patient group into exploratory and confirmatory cohorts. This has enabled predictive indices generated from the exploratory population to be tested in the confirmatory group, providing evidence of the reproducibility of these scoring systems. Although previous studies from Australia have examined clinical associations with skin cancer risk,6,16,17 these have been limited in method by the use of retrospective data sets and cancer registry data, allowing consideration of only selected risk factors. Positive associations previously identified include increased age at transplantation, increased latitude of transplant center, male sex,17 fair skin types, blue eyes, increasing age,16 and pretransplantation skin cancer.6 Notably, this is the first study that phenotypically characterized a population of Australian transplant recipients and systematically examined all potential clinical risk factors in relation to NMSC incidence and accrual. Importantly, we show that overall NMSC risk can be simply predicted using a combination of clinical factors (previous history of AK, BCC, invasive or in situ SCC, birth in a hot climate, blue/hazel eyes, and male sex). As shown in Figs 1 and 2, duration of immunosuppression therapy also is associated strongly with increasing incidence of NMSC, particularly SCC. Although individually, this is a strong predictor of NMSC risk, it correlated highly with previous history of SCC and therefore was displaced from the predictive indices by this variable. The simplified NMSC index allows stratification of renal transplant recipients into categories of high and low NMSC risk. Moreover, the same index can be applied to determine risk for multiple tumors in an individual patient. Absolute score in the high-risk category allows
PREDICTION OF SKIN CANCER POSTTRANSPLANT
further categorization of risk for tumor accrual, which would allow provision of a more clinically and cost-effective management strategy for multiple NMSCs. All statistical analyses were performed using data on histologically confirmed tumors only. However, an additional 106 tumors were diagnosed on clinical grounds only. To assess possible biasing of results caused by exclusion of these tumors, we also repeated stepwise models including these tumors. In each of the three stepwise models, the same variables were selected as the best set of predictors, with the exception of birth in a hot climate in the BCC index, although even this was only just below the threshold for inclusion in the model (P ⫽ 0.107). We believe the application of simplistic clinically derived predictive indices can allow NMSC risk stratification in an Australian transplant population. Furthermore, these models could be used effectively to counsel patients about individual risk for NMSC when considering the risk-benefit ratio of future potential transplantation. Although data obtained from the confirmatory cohort provide some validation of the indices, prospective evaluation, ideally in a separate cohort, is required to fully assess the reliability of models described here. If validated, use of the indices could be readily extrapolated to determine frequency of surveillance required according to individual risk. This may provide an evidence-based and cost-effective approach to the implementation of a clinical NMSC surveillance program. REFERENCES 1. Penn I: Cancers in renal transplant recipients. Adv Ren Replace Ther 7:147-156, 2000 2. Hardie IR: Skin cancer in transplant recipients. Transplant Rev 9:1-16, 1995 3. Glover MJ, Niranjan N, Kwan JTC, Leigh IM: Nonmelanoma skin cancer in renal transplant recipients: The extent of the problem and a strategy for management. Br J Plast Surg 47:86-89, 1994
683
4. Ramsay HM, Fryer AA, Smith AG, Harden PN: Clinical risk factors for non-melanoma skin cancer following renal transplantation. Am J Kidney Dis 36:167-176, 2000 5. Euvrard S, Kanitakis C, Pouteil-Nobel F, et al: Aggressive squamous cell carcinomas in organ transplant recipients. Transplant Proc 27:1767-1768, 1995 6. Bouwes Bavinck JN, Hardie DR, Green A, et al: The risk of skin cancer in renal transplant recipients in Queensland, Australia: A follow-up study. Transplantation 61:715721, 1996 7. Ramsay HM, Fryer AA, Hawley CM, Smith AG, Nicol D, Harden PN: A cross sectional study of non-melanoma skin cancer risk in a high risk Australian transplant population. Br J Dermatol 145:S17, 2001 (abstr, suppl 59) 8. Kasiske BL, Vazquez MA, Harmon WE, Brown RS, Danovitch GM, Gaston RS: Recommendations for the outpatient surveillance of renal transplant recipients. J Am Soc Nephrol 11:S1-S86, 2000 (suppl) 9. Ramsay HM, Fryer AA, Hawley CM, Smith AG, Nicol DL, Harden PN: Non-melanoma skin cancer risk in the Queensland renal transplant population. Br J Dermatol 147: 950-956, 2002 10. Armstrong BK, Kricker A: Skin cancer. Dermatol Clin 13:583-593, 1995 11. Australia’s Health 1998: The Sixth Biennial Health Report of the Australian Institute of Health and Welfare. Canberra, Australia, Australian Institute of Health and Welfare, 1998, p 86 12. Goodwin RG, Roberts DL: Skin cancer registration in the UK. Br J Dermatol 145:S17-S18, 2001 (suppl 59) 13. Penn I: Tumours after renal and cardiac transplantation. Hematol Oncol Clin North Am 7:431-445, 1993 14. Harden PN, Reece SM, Fryer AA, Smith AG, Ramsay HM: Skin cancer surveillance in renal transplant recipients: Questionnaire survey of current UK practice. BMJ 323:600-601, 2001 15. Harden PN, Fryer AA, Reece SM, Smith AG, Ramsay HM: Annual incidence and predicted risk of nonmelanoma skin cancer in renal transplant recipients. Transplant Proc 33:1302-1304, 2001 16. Kelly GE, Mahony JF, Sheil AGR, Meikle WD, Tiller DS, Horvarth J: Risk factors for skin carcinogenesis in immunosuppressed kidney transplant recipients. Clin Transplant 1:271-277, 1987 17. Roeger LS, Sheil AGR, Disney APS, Matthew TH, Amiss N: Risk factors associated with the development of squamous cell carcinomas in immunosuppressed renal transplant recipients. Clin Transpl 6: 202-211, 1992
Incidence and Prediction of Nonmelanoma Skin Cancer Post–Renal Transplantation: A Prospective Study in Queensland, Australia Robert P. Carroll, MB, ChB, Helen M. Ramsay, MD, Anthony A. Fryer, MRCPath, Carmel M. Hawley, MD, David L. Nicol, MD, and Paul N. Harden, MD ● Background: Nonmelanoma skin cancer (NMSC) is a significant clinical problem after renal transplantation, particularly in areas of high UV light exposure. A single-center prospective study of a population of Queensland renal transplant recipients was performed with the aims of: (1) establishing NMSC incidence and tumor accrual post–renal transplantation, and (2) developing a clinically derived predictive index to identify transplant recipients at greatest risk. Methods: Three hundred ten of 398 transplant recipients (78%) who underwent baseline assessment between July 1999 and April 2000 were reassessed a mean of 18 ⴞ 3.5 (SD) months later. A structured interview, full skin examination, biopsy of suspicious lesions, and review of medical and pathological records were used to determine the number and types of NMSC arising between the two assessments. Incidence (percentage of the population developing NMSC per year) and tumor accrual (number of tumors per person per year) were calculated. A clinically derived predictive index was generated using stepwise logistic regression models. Results: Overall NMSC incidence was 28.1% and increased with duration of immunosuppression therapy: 18.8%, 24.8%, 33.3%, and 47.1% at less than 5, 5 to 10, 10 to 20, and greater than 20 years of immunosuppression therapy, respectively. Mean NMSC accrual was 1.85 ⴞ 3.84 tumors/person/y, increasing to 3.35 ⴞ 4.29 tumors/person/y after 20 years of immunosuppression therapy. Renal transplant recipients were stratified into categories of high and low NMSC risk by using predictive indices. Conclusion: Clinically derived predictive indices can allow NMSC risk stratification of an Australian transplant population and may provide an evidence-based and cost-effective approach to developing a targeted clinical NMSC surveillance program. Am J Kidney Dis 41:676-683. © 2003 by the National Kidney Foundation, Inc. INDEX WORDS: Australia; immunosuppression; incidence; kidney transplantation; nonmelanoma skin cancer (NMSC); predictive index.
N
ONMELANOMA skin cancer (NMSC) is a significant clinical problem after renal transplantation, with a predominance of squamous cell carcinomas (SCCs).1-4 Tumors behave more aggressively, develop at an earlier age, and are more likely to metastasize.5 Fair-skinned renal allograft recipients residing in Queensland, Australia, have the highest NMSC risk in the world, with cumulative incidences previously reported as 45% and 70% at 11 and 20 years posttransplantation, respectively.6 We previously From the Departments of Renal Medicine and Dermatology; Clinical Biochemistry Research Group, School of Postgraduate Medicine, Keele University, North Staffordshire Hospital, Stoke-on-Trent, Staffordshire, UK; and the Renal Transplant Unit, Princess Alexandra Hospital, Brisbane, Queensland, Australia. Received May 23, 2002; accepted in revised form September 26, 2002. Supported in part by the St John Ambulance Travelling Fellowship in Transplantation. Address reprint requests to Paul N. Harden, MD, Oxford Kidney Unit, The Churchill Hospital, Oxford OX3 7LJ, UK. E-mail: [email protected] © 2003 by the National Kidney Foundation, Inc. 0272-6386/03/4103-0017$30.00/0 doi:10.1053/ajkd.2003.50130 676
showed that in this region, 32.6% of the transplant population will have experienced at least one NMSC after 5 years of immunosuppression therapy.7 There are no prospectively gathered data regarding NMSC incidence (total number of patients affected by NMSC per year) or tumor accrual (number of tumors per patient per year) in this population; however, previous studies have relied on retrospective, cross-sectional, or registry data. There currently is a limited structured program for skin cancer surveillance after renal transplantation in Queensland. To facilitate the development of such strategies, an accurate picture of the scale of the clinical problem is required in terms of both incidence and tumor accrual. Accordingly, a prospective study was performed to establish accurate data on incidence and tumor accrual in a population of Queensland renal transplant recipients. Furthermore, a set of clinical risk factors was used to develop a predictive index that may allow the identification of recipients at greatest risk. Together, these data provide for the first time an evidenced-based means to develop targeted skin surveillance and determine whether current
American Journal of Kidney Diseases, Vol 41, No 3 (March), 2003: pp 676-683
PREDICTION OF SKIN CANCER POSTTRANSPLANT
American Society of Transplantation (AST) guidelines regarding the outpatient surveillance of renal transplant recipients are applicable to this population.8 PATIENTS AND METHODS Baseline assessment of 398 of 435 adult renal transplant recipients (91.5%) attending the Princess Alexandra Hospital (PAH), Brisbane, Australia, was performed between July 1999 and April 2000. This consisted of a single-observer interview, full skin examination, biopsy of suspicious lesions, and review of medical and/or pathological records to determine the number, types, and risk factors for NMSC posttransplantation. This study, including demographic details of the study population, has been reported in detail previously.7,9 Three hundred ten of 398 transplant recipients (78%) were reassessed a mean of 18 ⫾ 3.5 (SD) months later by a second observer blinded to previous dermatological records. Eighty-eight transplant recipients were not available for a second interview (3 patients withheld consent, 10 patients had died, 35 patients had been transferred for follow-up at peripheral nephrological centers, and 40 patients were unavailable for review). This assessment consisted of a structured interview incorporating a second full skin examination. Potentially malignant lesions or lesions with atypical features were referred to the Dermatology Department at the PAH for histological confirmation of diagnosis and treatment. When dermatological management occurred outside the PAH, details were obtained from the attending practitioner using a standard proforma. Clinically diagnosed (eg, those treated by ablative methods without previous biopsy) and histologically confirmed tumors were recorded separately. Medical and pathological records since first assessment were reexamined. The second observer received initial training from the dermatologist who performed the baseline assessment to ensure consistency in terms of diagnostic criteria and documentation. Clinical characteristics and maintenance immunosuppression regimens are listed in Table 1. Demographic information and clinical risk factors were gathered by structured interview and case note review at baseline assessment. Briefly, data gathered included age, sex, smoking history, skin type I through VI (original Fitzpatrick classification: skin type I corresponds to fair skin, which always burns and never tans, through skin type VI, which is permanently deeply pigmented), and natural hair and eye color at the age of 21 years. UV light exposure was estimated using a number of parameters, including occupational exposure, calculated cumulative sun exposure based on average hours spent outdoors throughout life, sunbathing habits (assigned a sunbathing score), and recalled childhood sunburn episodes. Data also were gathered on country of birth, residence in temperate/tropical/subtropical climates, and chemical carcinogen exposure (tobacco smoke and arsenic). Incidence (percentage of the population developing NMSC per year) was calculated by dividing the total number of patients developing NMSC per year by the total number of patients reexamined (n ⫽ 310). Mean tumor accrual (number of tumors per person per year) was calculated by
677 Table 1. Patient Characteristics in 310 Renal Transplant Recipients Included in the Incidence Analysis Patient Characteristics
Caucasian Male sex No. of allografts 1 2 3 Donor (current allograft) Cadaveric Live related Live unrelated Immunosuppression regimen Triple therapy* Cyclosporine/azathioprine* Cyclosporine/prednisolone* Azathioprine/prednisolone* Others* Cyclosporine* (mg/d) Azathioprine* (mg/d) Prednisolone* (mg/d) Age at 1st transplantation stratified by duration of immunosuppression (y) ⬍5 (n ⫽ 103) 5-10 (n ⫽ 73) 10-20 (n ⫽ 106) ⬎20 (n ⫽ 28) Prospective skin cancer surveillance stratified by duration of immunosuppression (y) ⬍5 (n ⫽ 103) 5-10 (n ⫽ 73) 10-20 (n ⫽ 106) ⬎20 (n ⫽ 28)
284 (91.6) 191 (61.2) 267 (86.1) 32 (10.3) 11 (3.5) 261 (86.7) 38 (12.6) 2 (0.7) 126 (44.3) 34 (12.0) 26 (9.2) 42 (14.8) 56 (19.7) 225 (150-300) 87.5 (0-125) 7 (5-10)
47.2 ⫾ 12.9 42.7 ⫾ 15.4 39.1 ⫾ 12.8 30.1 ⫾ 11.1
1.54 ⫾ 0.2 1.53 ⫾ 0.3 1.53 ⫾ 0.2 1.53 ⫾ 0.3
NOTE. Values expressed as number (percent); median (interquartile range); or mean ⫾ SD. *Immunosuppression data at 1 year posttransplantation available on 284 patients. Other immunosuppressants used included tacrolimus, sirolimus, and mycophenylate.
dividing the sum of the number of tumors per year for each transplant recipient by the total number of recipients examined at second interview (n ⫽ 310). The PAH Research Ethics Committee approved this study, and all participants provided written informed consent.
Generation of Predictive Indices Only histologically confirmed tumors were included in statistical analyses. Predictive indices were generated using stepwise logistic regression models (Stata Statistical Software, version 7; Stata Corp, College Station, TX) with a probability cutoff value for inclusion in the models of P of 0.1 or less. In stepwise analyses, an additional 22 patients who were unavailable for clinical examination who developed histologically proven NMSC since baseline also were
678
CARROLL ET AL
included. These additional patients were not included in the calculation of incidence figures. Of the resulting 332 patients included in stepwise analyses, 26 patients of nonCaucasian origin (skin types V and VI) were excluded because of constitutive protection against skin cancer. The remaining 306 patients were randomly assigned (using the random number generator function in Stata) into two groups in a 2:1 ratio: an exploratory cohort (n ⫽ 204) and a confirmatory cohort (n ⫽ 102). Our strategy was to generate predictive indices using the exploratory group and test these in the confirmatory cohort. The following potential predictors were included in the stepwise models: age at transplantation, current age, duration of immunosuppression, male sex, smoking history (ever versus never and pack-years), possible/probable arsenic exposure, birth in a hot climate, years in an outdoor occupation, occasional/frequent childhood sunburn, sunbathing score, cumulative sun exposure, skin type I, blue/hazel eyes, red hair, previous history of SCC, basal cell carcinoma (BCC), solar (actinic) keratoses (AK), keratoacanthoma (KA), in situ SCC (Bowen’s disease [BD], considered separately), viral warts, and number of transplants. Identification of optimal scores for the predictive indices was determined by plotting sensitivity and specificity against probability cutoff. Comparison of models was performed using receiver operating characteristic curves. Separate indices were generated for risk for BCC, SCC, and total NMSC (including BD and KA).
RESULTS
Table 2 lists numbers of patients and tumors identified between the first and second assessments. Of 131 patients who developed NMSC during the follow-up period, 77 patients (58.8%) developed more than one tumor type. The ratio of SCC (invasive plus in situ) to BCC was 3.1:1. Importantly, 129 of 850 NMSCs (15.2%) occurred on the relatively photo-protected skin of the trunk and proximal limbs. Three new cases of melanoma also were identified at the second assessment. An additional 106 NMSCs were diagnosed on clinical criteria alone, composed of 42 BCCs, 23 SCCs, 20 BDs, and 17 KAs. These Table 2. Histologically Confirmed NMSCs Identified During the 18-Month Follow-Up Period
BCC Invasive SCC In situ SCC (BD) KA Miscellaneous* Total NMSCs
No. of Patients (%)
No. of Tumors
76/310 (24.5) 89/310 (28.7) 76/310 (24.5) 29/310 (9.4) 3/310 (1.0) 131/310 (42.3)
197 373 241 36 3 850
*One each of vaginal intraepithelial neoplasm III, liposarcoma, and spindle cell fibroxanthoma.
tumors were excluded from all subsequent statistical comparisons. Between the first and second assessments, 12 patients had developed NMSC for the first time. Five patients required nine reconstructive plastic surgical procedures to repair defects after excision of NMSC. Three patients developed metastatic SCC after primary treatment; 2 of these patients underwent axillary lymph node dissection and local radiotherapy, and 1 patient had hepatic metastasis. One patient died of metastatic melanoma. Overall NMSC incidence was 28.1% (21.4% for women and 31.8% for men) and increased with duration of immunosuppression (Fig 1). Considering individual tumor types, the incidence of invasive SCC was 19.1%; BD, 16.3%; BCC, 16.3%; and KA, 6.2%. Mean NMSC accrual during the 18-month follow-up was 1.85 ⫾ 3.84 (SD) tumors/patient/y. Mean accruals for invasive SCC, BD, BCC, and KA were 0.82 ⫾ 2.23, 0.52 ⫾ 1.37, 0.42 ⫾ 1.08, and 0.08 ⫾ 0.30 tumors/patient/y, respectively. As with incidence, accrual increased with duration of immunosuppression (Fig 2). Generation of Predictive Indices BCC predictive index. Using the exploratory cohort, the stepwise model used to predict BCC risk within the follow-up period selected previous history of BCC (P ⬍ 0.001; odds ratio [OR], 6.3), previous history of AK (P ⫽ 0.011; OR, 8.0), birth in a hot climate (P ⫽ 0.005; OR, 23.5), male sex (P ⫽ 0.002; OR, 4.5), and previous history of BD (P ⫽ 0.019; OR, 3.1) as the significant set of predictors. Table 3 shows the predictive index equation generated. The sensitivity and specificity against probability cutoff plots allowed generation of the score from the predictive index that best differentiated between BCC-positive and BCC-negative cases. Using this optimum score cutoff, sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy of the predictive index were calculated (Table 3). OR, 95% confidence intervals (CIs), and significance level using this cutoff are also listed in Table 3. This predictive index then was applied to the confirmatory cohort, and results obtained are listed in Table 3. As expected, data in the confir-
PREDICTION OF SKIN CANCER POSTTRANSPLANT
679
Fig 1. Incidence (percentage of population per year) of NMSC and duration of immunosuppression in 310 renal transplant recipients. (F) BCC; (}) SCC; (Œ) BD; (■) KA; (䊐) ALL.
matory cohort are not as good at predicting BCC risk as those of the exploratory cohort, particularly with reference to sensitivity and positive predictive value. However, this suggests that although the index will indicate screening of patients who will not develop subsequent BCC, most of those who will develop tumors will be included in the screened set. SCC predictive index. In the exploratory group, the model used to predict SCC risk within
the follow-up period selected previous history of SCC (P ⬍ 0.001; OR, 17.8), blue/hazel eyes (P ⫽ 0.040; OR, 3.4), ever smoker (P ⫽ 0.040; OR, 2.8), male sex (P ⫽ 0.003; OR, 4.6), previous history of KA (P ⫽ 0.071; OR, 2.7), and previous history of BD (P ⫽ 0.075; OR, 2.6) as the significant set of predictors. The predictive index–generated equation is listed in Table 3. The cutoff value used generated a particularly large OR: 42.8 (95% CI, 17.1 to 110.7; P ⬍
Fig 2. Mean tumor accrual against duration of immunosuppression for 310 patients. (F) BCC; (}) SCC; (Œ) BD; (■) KA; (䊐) ALL.
680
CARROLL ET AL Table 3.
Generation and Validation of Predictive Indices Using Stepwise Logistic Regression Models in the Exploratory and Confirmatory Cohorts BCC
Predictive index ⫽
Exploratory cohort Predictive index score Proportion with tumor Sensitivity (%) Specificity (%) Positive predictive value (%) Negative predictive value (%) Accuracy (%) OR 95% CI P Confirmatory cohort Predictive index score Proportion with tumor Sensitivity (%) Specificity (%) Positive predictive value (%) Negative predictive value (%) Accuracy (%) OR 95% CI P
(B * 1.79) ⫹ (I * 1.16) ⫹ (A * 1.66) ⫹ (M * 1.49) ⫹ (H * 1.66)
SCC
(S * 2.60) ⫹ (I * 1.09) ⫹ (T * 1.02) ⫹ (M * 1.67) ⫹ (E * 1.00) ⫹ (K * 0.97)
NMSC*
(S * 1.21) ⫹ (I * 1.44) ⫹ (B * 0.92) ⫹ (M * 1.32) ⫹ (E * 0.69) ⫹ (A * 0.95) ⫹ (H * 1.60)
⬍8.3 ⱖ8.3 ⬍5.4 ⱖ5.4 ⬍4.8 ⱖ4.8 10/133 (7.5%) 43/64 (67.2%) 10/130 (7.7%) 57/73 (78.1%) 19/111 (17.1%) 75/86 (87.2%) 81.1 85.1 79.8 85.4 88.2 89.3 67.2 78.1 87.2 92.5
92.3
82.9
84.3 25.2 10.3-63.7 ⬍0.0001
87.2 42.8 17.1-110.7 ⬍0.0001
84.8 33.0 13.9-80.6 ⬍0.0001
⬍8.3 ⱖ8.3 ⬍5.4 ⱖ5.4 ⬍4.8 ⱖ4.8 12/72 (16.7%) 16/29 (55.2%) 8/72 (11.4%) 23/32 (71.9%) 15/58 (25.9%) 35/43 (81.4%) 57.1 74.2 70.0 82.2 87.3 84.3 55.2 71.9 81.4 83.3
88.6
74.1
75.2 6.2 2.1-17.9 0.0001
83.3 19.8 6.1-66.3 ⬍0.0001
77.2 12.5 4.4-37.6 ⬍0.0001
Abbreviations: A, preexisting solar keratoses (1 if any present, 0 if absent); B, preexisting BCC (1 if any present, 0 if absent); C, sunburning during childhood (1, if occasional/frequent, 0 if never/rare); E, eye color (1 if blue/hazel, 0 if brown/green); H, birth in hot climate (1 if hot country, 0 if temperate, etc); I, preexisting BD (1 if any present, 0 if absent); K, preexisting KA (1 if any present, 0 if absent); M, sex (1 if male, 0 if female); S, preexisting SCC (1 if any present, 0 if absent); T, smoking status (1 if ever, 0 if never). *NMSC ⫽ BCC ⫹ SCC ⫹ BD ⫹ KA combined.
0.0001). In the confirmatory cohort, the predictive index also was very accurate at identifying high-risk patients (Table 3), with high values for both sensitivity and specificity obtained. NMSC predictive index. A further predictive index was generated given that total risk for NMSC, rather than that for individual tumor types, will determine surveillance strategy. In the exploratory cohort, the model used to predict total NMSC risk within the follow-up period selected previous history of SCC (P ⫽ 0.012; OR, 3.5), previous history of BCC (P ⫽ 0.039; OR, 2.6), male sex (P ⫽ 0.004; OR, 3.8), birth in a hot climate (P ⫽ 0.031; OR, 5.3), blue/hazel
eyes (P ⫽ 0.042; OR, 2.6), previous history of AK (P ⫽ 0.055; OR, 2.8), and previous history of BD (P ⫽ 0.007; OR, 4.0) as the significant set of predictors. This generated the equation: Predictive index ⫽ 共S * 1.21兲 ⫹ 共I * 1.44兲 ⫹ 共B * 0.92兲 ⫹ 共M * 1.32兲 ⫹ 共E * 0.69兲 ⫹ 共A * 0.95兲 ⫹ 共H * 1.60兲 where S is preexisting SCC, I is preexisting BD, B is preexisting BCC, M is sex, E is eye color, A is preexisting AK, and H is birth in a hot climate, as listed in Table 3. This index then was applied to the confirmatory cohort (Table 3).
PREDICTION OF SKIN CANCER POSTTRANSPLANT Table 4.
Score
681
Association of Score From the Simplified Predictive Index for Total NMSC With Tumor Risk and Accrual Proportion With Tumor
Exploratory cohort 0 0/2 (0.0%) 1 0/4 (0.0%) 2 0/9 (0.0%) 3 6/40 (15.0%) 4 7/31 (22.6%) 5 7/26 (26.9%) 6 22/30 (73.3%) 7 26/28 (92.9%) 8 26/27 (96.3%) Confirmatory cohort 0 0/1 (0.0%) 1 0/3 (0.0%) 2 2/10 (20.0%) 3 2/16 (12.5%) 4 3/14 (21.4%) 5 10/16 (62.5%) 6 10/15 (66.7%) 7 16/19 (84.2%) 8 7/7 (100.0%)
Mean NMSC/y
Mean BCC/y
Mean SCC/y
OR
95% CI
0.00 0.00 0.00 0.18 0.33 0.31 1.81 2.65 9.07
0.00 0.00 0.00 0.05 0.17 0.03 0.22 0.42 2.00
0.00 0.00 0.00 0.07 0.04 0.12 0.90 1.49 4.27
} }1.0 } } 2.4 3.0 22.5 106.2 212.3
— — — — 0.7-7.9 0.9-10.1 7.0-72.5 20.0-563.7 24.2-1859.2
0.00 0.00 0.24 0.10 0.12 1.01 1.44 2.39 3.90
0.00 0.00 0.11 0.03 0.09 0.18 0.41 0.34 1.15
0.00 0.00 0.00 0.03 0.00 0.59 0.80 1.04 1.69
} }1.0 } } 1.8 10.8 13.0 34.7 †
— — — — 0.3-9.3 2.5-46.7 2.9-58.7 6.9-175.4 †
P
Reference*
0.155 0.075 ⬍0.001 ⬍0.001 ⬍0.001
Reference*
0.498 0.001 0.001 ⬍0.001 †
NOTE. Scores derived from the simplified equation: Predictive index ⫽ A ⫹ B ⫹ E ⫹ (2 * H) ⫹ I ⫹ M ⫹ S, where A is preexisting solar keratoses (1 if any present, 0 if absent), B is preexisting BCC (1 if any present, 0 if absent), E is eye color (1 if blue/hazel, 0 if brown/green), H is birth in hot climate (1 if hot country, 0 if temperate, etc), I is preexisting BD (1 if any present, 0 if absent), M is sex (1 if male, 0 if female), and S is preexisting SCC (1 if any present, 0 if absent). NMSC includes SCC, BCC, BD, and KA. *Reference category is groups 0, 1, 2, and 3 combined. †OR could not be generated because of the absence of cases without tumors in this group.
This model was simplified further for clinical use by reducing the equation to: Predictive index ⫽ A ⫹ B ⫹ E ⫹ 共2 * H兲 ⫹ I ⫹ M ⫹ S 共Table 4兲 Using a cutoff score of five or less versus less than five gave parameters similar to the more detailed equation (sensitivity, 78.7%; specificity, 89.3%; positive predictive value, 87.1%; negative predictive value, 82.1%; accuracy, 84.3%; OR, 30.9; 95% CI, 13.1 to 75.1; P ⬍ 0.0001; Table 3). However, in both the exploratory and confirmatory cohorts, the probability of developing an additional NMSC increased dramatically with increasing score (Table 4). Using these data, it was possible to calculate mean accrual for BCC, SCC, and total NMSC (tumors per year) for each score. This showed that mean numbers of BCCs and SCCs per year increased more rapidly in patients with a score greater than five. Patients with the maximum score showed very
large accrual rates, particularly for SCC. Table 4 also shows the relative increase in risk compared with patients with the minimum score (zero) in the exploratory group. Those with a score of seven showed a 106-fold increased risk compared with those with a score of zero, whereas those with the maximum score (eight) were more than 200 times more likely to develop an additional NMSC. Similar results were obtained when the simplified score was applied to the confirmatory cohort (Table 4). DISCUSSION
There were 5,235 transplant recipients alive with a functioning renal allograft in Australia at risk for skin cancer in December 2000 (Australia and New Zealand Dialysis and Transplantation Registry [ANZDATA]). To our knowledge, this is the first time prospectively gathered data on NMSC incidence and accrual after renal transplantation in an Australian population has been reported. We describe an annual incidence rate of
682
NMSC in Queensland of 28.1%, increasing to greater than 45% in those who underwent transplantation for more than 20 years. Cumulative NMSC incidences of 45% and 70% at 11 and 20 years posttransplantation found in previous studies have been retrospectively derived from ANZDATA.6 It is important to recognize that there are significant limitations in the completeness of reporting of NMSCs to ANZDATA and other cancer registries because of the large numbers of cases, treatment of cases without histological diagnosis, and fragmentation of patient care.7,10-12 In the same population, we previously showed that 28.4% of Caucasian transplant recipients developing NMSC since transplantation were not documented by ANZDATA.9 Furthermore, cumulative incidence is a population-based figure, and alone, it provides no indication of number of individuals in that population affected per year. For the first time, we generated incidence data that, unlike cumulative incidence, provide a dynamic measure of the scale of the problem to be dealt with on an annual basis by units managing renal transplant recipients in Queensland. However, we recognize that a potential limitation is that these data are derived from two observational dates and ideally require confirmation during subsequent years. Individual burden of disease also has been quantified with a mean tumor accrual rate of 1.85. These data are unavailable from ANZDATA because this registry records only the first episode of BCC and SCC posttransplantation. Distribution of accrual in this study is skewed; most have an accrual of less than 3 tumors/y (mean, 4.34 tumors/y; median, 2.43 tumors/y). However, a significant minority (n ⫽ 36) had accrual in excess of 5 tumors/y, and some patients showed very large accrual rates (maximum NMSC accrual, 24.5 tumors/y). Assessment of tumor accrual therefore provides an important indicator of individual morbidity and allows the identification of a subset of patients who require intensive surveillance. The AST recommends that all renal transplant recipients in North America be examined annually for skin cancer,8 where the reported prevalence of NMSC post–renal transplantation is 18%.13 In Queensland, the cumulative incidence is 51.8%, and currently, skin cancer surveillance
CARROLL ET AL
is limited, with most high-risk patients seen in dermatology clinics.9 Similarly, in northern Europe, only 19% of units following up renal transplant recipients offer annual full skin examination for skin cancer.14 It is not clear whether the AST guidelines can be applied to areas of high (Australia) or low (northern Europe) exposure to UV light. We previously used clinical risk factors to predict tumor risk in individuals after renal transplantation in a UK population.15 We have now used a similar method to generate a predictive index applicable to Queensland renal transplant recipients. The high incidence of NMSC in this population has allowed us to randomly assign the patient group into exploratory and confirmatory cohorts. This has enabled predictive indices generated from the exploratory population to be tested in the confirmatory group, providing evidence of the reproducibility of these scoring systems. Although previous studies from Australia have examined clinical associations with skin cancer risk,6,16,17 these have been limited in method by the use of retrospective data sets and cancer registry data, allowing consideration of only selected risk factors. Positive associations previously identified include increased age at transplantation, increased latitude of transplant center, male sex,17 fair skin types, blue eyes, increasing age,16 and pretransplantation skin cancer.6 Notably, this is the first study that phenotypically characterized a population of Australian transplant recipients and systematically examined all potential clinical risk factors in relation to NMSC incidence and accrual. Importantly, we show that overall NMSC risk can be simply predicted using a combination of clinical factors (previous history of AK, BCC, invasive or in situ SCC, birth in a hot climate, blue/hazel eyes, and male sex). As shown in Figs 1 and 2, duration of immunosuppression therapy also is associated strongly with increasing incidence of NMSC, particularly SCC. Although individually, this is a strong predictor of NMSC risk, it correlated highly with previous history of SCC and therefore was displaced from the predictive indices by this variable. The simplified NMSC index allows stratification of renal transplant recipients into categories of high and low NMSC risk. Moreover, the same index can be applied to determine risk for multiple tumors in an individual patient. Absolute score in the high-risk category allows
PREDICTION OF SKIN CANCER POSTTRANSPLANT
further categorization of risk for tumor accrual, which would allow provision of a more clinically and cost-effective management strategy for multiple NMSCs. All statistical analyses were performed using data on histologically confirmed tumors only. However, an additional 106 tumors were diagnosed on clinical grounds only. To assess possible biasing of results caused by exclusion of these tumors, we also repeated stepwise models including these tumors. In each of the three stepwise models, the same variables were selected as the best set of predictors, with the exception of birth in a hot climate in the BCC index, although even this was only just below the threshold for inclusion in the model (P ⫽ 0.107). We believe the application of simplistic clinically derived predictive indices can allow NMSC risk stratification in an Australian transplant population. Furthermore, these models could be used effectively to counsel patients about individual risk for NMSC when considering the risk-benefit ratio of future potential transplantation. Although data obtained from the confirmatory cohort provide some validation of the indices, prospective evaluation, ideally in a separate cohort, is required to fully assess the reliability of models described here. If validated, use of the indices could be readily extrapolated to determine frequency of surveillance required according to individual risk. This may provide an evidence-based and cost-effective approach to the implementation of a clinical NMSC surveillance program. REFERENCES 1. Penn I: Cancers in renal transplant recipients. Adv Ren Replace Ther 7:147-156, 2000 2. Hardie IR: Skin cancer in transplant recipients. Transplant Rev 9:1-16, 1995 3. Glover MJ, Niranjan N, Kwan JTC, Leigh IM: Nonmelanoma skin cancer in renal transplant recipients: The extent of the problem and a strategy for management. Br J Plast Surg 47:86-89, 1994
683
4. Ramsay HM, Fryer AA, Smith AG, Harden PN: Clinical risk factors for non-melanoma skin cancer following renal transplantation. Am J Kidney Dis 36:167-176, 2000 5. Euvrard S, Kanitakis C, Pouteil-Nobel F, et al: Aggressive squamous cell carcinomas in organ transplant recipients. Transplant Proc 27:1767-1768, 1995 6. Bouwes Bavinck JN, Hardie DR, Green A, et al: The risk of skin cancer in renal transplant recipients in Queensland, Australia: A follow-up study. Transplantation 61:715721, 1996 7. Ramsay HM, Fryer AA, Hawley CM, Smith AG, Nicol D, Harden PN: A cross sectional study of non-melanoma skin cancer risk in a high risk Australian transplant population. Br J Dermatol 145:S17, 2001 (abstr, suppl 59) 8. Kasiske BL, Vazquez MA, Harmon WE, Brown RS, Danovitch GM, Gaston RS: Recommendations for the outpatient surveillance of renal transplant recipients. J Am Soc Nephrol 11:S1-S86, 2000 (suppl) 9. Ramsay HM, Fryer AA, Hawley CM, Smith AG, Nicol DL, Harden PN: Non-melanoma skin cancer risk in the Queensland renal transplant population. Br J Dermatol 147: 950-956, 2002 10. Armstrong BK, Kricker A: Skin cancer. Dermatol Clin 13:583-593, 1995 11. Australia’s Health 1998: The Sixth Biennial Health Report of the Australian Institute of Health and Welfare. Canberra, Australia, Australian Institute of Health and Welfare, 1998, p 86 12. Goodwin RG, Roberts DL: Skin cancer registration in the UK. Br J Dermatol 145:S17-S18, 2001 (suppl 59) 13. Penn I: Tumours after renal and cardiac transplantation. Hematol Oncol Clin North Am 7:431-445, 1993 14. Harden PN, Reece SM, Fryer AA, Smith AG, Ramsay HM: Skin cancer surveillance in renal transplant recipients: Questionnaire survey of current UK practice. BMJ 323:600-601, 2001 15. Harden PN, Fryer AA, Reece SM, Smith AG, Ramsay HM: Annual incidence and predicted risk of nonmelanoma skin cancer in renal transplant recipients. Transplant Proc 33:1302-1304, 2001 16. Kelly GE, Mahony JF, Sheil AGR, Meikle WD, Tiller DS, Horvarth J: Risk factors for skin carcinogenesis in immunosuppressed kidney transplant recipients. Clin Transplant 1:271-277, 1987 17. Roeger LS, Sheil AGR, Disney APS, Matthew TH, Amiss N: Risk factors associated with the development of squamous cell carcinomas in immunosuppressed renal transplant recipients. Clin Transpl 6: 202-211, 1992