Abstract
A low 25-hydroxyvitamin D3 (25(OH)D3) serum concentration at melanoma diagnosis might be associated with worse survival. We prospectively studied the prognostic value of 25(OH)D3 at diagnosis and during follow-up.
MelanCohort is a cohort of invasive melanoma patients. Serum 25(OH)D3 was measured by mass spectrometry and standardized on month of blood drawn, age, sex, and body mass index (BMI). Role of 25(OH)D3 levels and risk of relapse was analyzed in a Cox proportional hazards model adjusting for age, sex, BMI, and American Joint Committee on Cancer (AJCC) stage. All statistical tests were two-sided.
One thousand one hundred seventy-one patients were included. 25(OH)D3 levels at diagnosis (median = 49.0 nmol/L) were inversely correlated with prognostic factors such as AJCC stage (P < .001 Kruskal-Wallis), Breslow’s thickness (P < .001 Spearman correlation), and ulceration (P < .001 Kruskal-Wallis), but not with risk of relapse. Changes in 25(OH)D3 levels during follow-up were associated with worse prognosis: With a third quartile Q3 of average change per year (-0.30 to 4.60 nmol/L/Y) used as reference, hazard ratios for the first, second, and fourth quarters were 1.94 (95% confidence interval [CI] = 1.36 to 2.76), 1.23 (95% CI = 0.85 to 1.78), and 1.61 (95% CI = 1.14 to 2.28), respectively. In sensitivity analyses, no changes were observed either by AJCC stage, number of 25(OH)D3 measures performed, or by 25(OH)D3 level at baseline. No evidence of reverse causation was identified. Analyses performed on overall survival yielded similar results.
We show that 25(OH)D3 variation during follow-up is an independent melanoma prognostic marker, but not its level at diagnosis. Previously reported associations between low 25(OH)D3 level at diagnosis and poor prognosis seem to be due to insufficient adjustment for prognostic factors.
The pro-hormone vitamin D (vitD) has pleiotropic actions on human cells, including some that are relevant to cancer, notably on cell growth, differentiation or apoptosis, or on the immune system. VitD is mainly synthetized in the skin during ultraviolet exposure or, to a lesser extent, is ingested from some foods or supplements. It is mainly transported to target cells after hydroxylation to 25 hydroxy-vitaminD3 (25(OH)D3), which is believed to represent its reservoir form. 25(OH)D3 acts after transformation in target cells into active compounds, the main being 1–25(OH)2D3, through interaction with the vitD receptor (1).
Several prospective population-based studies have shown higher 25(OH)D3 serum levels statistically significantly associated with decreased risk of later diagnosis of breast (2), colon (3), renal (4), hepatocellular (5), or tobacco-related cancers (6). However, randomized intervention studies of vitD supplementation failed to demonstrate any reduction of cancer development (7), suggesting a reverse causation association between low 25(OH)D3 serum levels and presence of cancer.
Several cohort studies have shown statistically significant associations between low 25(OH)D3 serum levels at diagnosis and presence of poor prognosis criteria, decreased cancer-specific or overall survival (OS) in patients with colon (8), breast (2,9), prostate (10), bladder (11), or all cancers (7,12). However, when multivariate adjustments with other prognostic criteria, cancer therapy, or body mass index (BMI) were made, the association vanished, as for breast cancer (13). Few studies have investigated the prognostic value of variations of 25(OH)D3 serum level during follow-up of cancer patients. Postdiagnosis vitD intake or high 25(OH)D3 serum levels were not associated with any improvement of all-cause or disease-specific mortality in prospective cohorts of colon (14), breast (13), or prostate (15) cancer patients. Similarly, no impact of vitD supplementation was demonstrated on prostate-specific antigen level in black men with prostate cancer (16).
Cutaneous melanoma is a severe cancer once spread to distant organs, with main prognostic factors being Breslow’s thickness, ulceration, and mitotic rate of the primary lesion and status of the draining lymph nodes (17). In vitro studies have shown growth inhibition of malignant melanoma cell lines by 1,25(OH)2D3 (18). In patients with newly diagnosed melanoma, low 25(OH)D3 serum levels were associated with worse prognostic factors, such as Breslow’s thickness (19), higher American Joint Committee on Cancer (AJCC) melanoma stage (20), and worse survival independent of Breslow’s thickness in one prospective study (21).
Studies on 25(OH)D3 and cancer occurrence or prognosis have often been flawed by various biases, such as lack or incomplete handling of factors that control serum level, like seasonality of blood drawing, body mass index (BMI), age, and sex. In addition, no study has ever evaluated the prognostic role of 25(OH)D3 serum level measured during the follow-up of melanoma patients. We aimed, using a prospective cohort of melanoma patients with multiple assessment of 25(OH)D3 serum levels and where most factors that control 25(OH)D3 serum levels were recorded: 1) to confirm the prognostic value of 25(OH)D3 serum levels collected at diagnosis and 2) for the first time, to evaluate whether changes during follow-up of 25(OH)D3 serum levels could be associated with modifications of disease-free survival (DFS) or OS.
Methods
Study Population
MelanCohort has been described previously (22). Briefly, a cohort of adults diagnosed with skin melanoma was set up from great Paris university hospitals (APHP) between September 2003 and December 2008 and followed up prospectively. Case patients had confirmed melanoma by central pathology review and were included within three months after curative surgery of primary melanoma or of invaded nodes or within one month after diagnosis of distant metastasis. This study was registered in ClinicalTrials.gov database (NCT00839410). All patients gave written informed consent, and ethical committee CPP IDF 8 approved the study. Follow-up visits were recorded prospectively every two visits required by the French Guidelines (twice yearly for stage I patients, 4 times yearly for stage II-III) (23). Melanoma prognostic criteria were collected prospectively (Breslow thickness, ulceration, nodal status, mitotic rate, age, sex, location of the primary, AJCC stage). At inclusion and during prescheduled follow-up visits, blood was drawn and serum samples were aliquoted and stored at -80ºC in centralized biobank. Using central APHP Diagnosis Related Groups system, we estimated that roughly 50% of melanoma patients presenting to APHP during the study period were included in MelanCohort.
The follow-up was conducted until June 2013 for this study, allowing OS to be scarcely influenced by new efficient treatments of metastatic melanoma such as vemurafenib, which was launched in French pharmacies in February 2013. This date constituted the right censoring date for the follow-up. Detailed information on follow-up was collected, such as the last contact date, living status, and date of death if any. Living status was confirmed by cross-checking individual status in the national register for cause of death (CepiDC, http://www.cepidc.inserm.fr). Information on relapse was also collected prospectively.
For this study, we analyzed patients with only one invasive melanoma. 25(OH)D3 serum level was measured, blinded to clinical information, by mass spectrometry coupled to high-performance liquid chromatography, with a precision of 5%, and expressed as nmol/L (see the Supplementary Methods, available online).
Statistical Methods
To enable comparisons of survival according to serum level of 25(OH)D3 (initial value at diagnosis and values recorded during follow-up), we produced standardized 25(OH)D3 levels adjusted for month of sampling, age, sex, and BMI, using a generalized linear model restricted to patients for whom 25(OH)D3 was measured within six months after primary melanoma diagnosis (Figure 1; Supplementary Table 1, available online). Because of seasonal variations of 25(OH)D3, sensitivity analyses were conducted by repeating this standardization restricted to patients for whom the 25(OH)D3 was measured within 90 or 30 days after primary melanoma diagnosis.
First 25(OH)D3 measurements performed within 180 days after diagnosis on melanoma patients participating to MelanCohort A) before and B) after standardization on age, sex, body mass index, and month of blood sampling. The line corresponds to the Loess curve (locally weighted scatterplot smoothing) fitted to 25(OH)D3 measurements and date of blood sampling, and the corresponding 95% confidence limits. This Loess curve was estimated for display purpose only.
For each individual, the yearly change of 25(OH)D3 level was computed based on a linear regression of the standardized 25(OH)D3 levels upon the years following diagnosis. The linear trend could not be computed for patients with only one 25(OH)D3 measure performed, for whom reverse causation bias was likely (death before having at least two measurements). A sensitivity analysis was conducted restricted to individuals with at least four measurements. To evaluate the assumption of linearity of the yearly trend of 25(OH)D3, the Student’s t test was performed on the residuals of the linear regressions testing departure from zero.
When continuous variables were converted into categorical data, the quartiles of the distribution of the variable were used as a cutpoint. For the trend of 25(OH)D3, an additional variable was computed based on every 10% of the distribution. The association between two variables was tested by nonparametric methods: Kruskal-Wallis for a continuous variable and a categorical variable, Spearman correlation for two continuous variables; chi-square test was used for categorical variables.
The prognostic role of 25(OH)D3 level at inclusion and change along time was analyzed in a Cox proportional hazards model adjusting for age, sex, BMI, and AJCC stage. Proportional hazards hypothesis was verified by investigating Schoenfeld residuals. Primary and secondary endpoints were DFS and OS, respectively. DFS was defined as the time from first melanoma pathological diagnosis to first progression (occurrence of regional or first distant metastasis) or death. For stage IV patients, only date of death was used for DFS. OS was defined as the time from melanoma diagnosis to death. The SAS software version 9.3 for PC (SAS Institute, Cary, NC) was used.
All tests of statistical significance were two-sided. Power calculations have been performed considering a 0.7 hazard ratio (HR) against DFS or OS for the upper 25-OHD3 quartile vs the three lowest quartiles. A minimal follow-up time of three years was estimated mandatory to gain 80% power with a two-sided type I error of 5%. An objective hazard ratio of 0.70 seemed adequate as in the Newton-Bishop study (21) the hazard ratios for risk of relapse were 0.70 (95% CI = 0.42 to 1.14) and 0.57 (95% CI = 0.33 to 0.97) for the upper two 25(OH)D3 level tertiles, respectively. This project required follow-up of all MelanCohort patients until at least June 2011, and 4094 serum samples were expected. A P value of less than .05 was considered statistically significant.
Results
For 31 individuals, no adequate serum sample was available. Overall, 1171 individuals (579 men and 592 women, mean age 54.2 years, interquartile range [IQR] = 41–67) with at least one 25(OH)D3 measurement were included, and 1008 had a first 25(OH)D3 level measured within six months after melanoma diagnosis. Median follow-up was overall 4.5 years. Forty-one individuals were lost to follow-up. Average BMI was 25.2kg/m2 (IQR = 22.0–27.3, 11 individuals with missing information). Overall, 3728 25(OH)D3 serum level measurements were available (mean 3.25/individual, range = 1–13, IQR = 1–4). Sentinel lymph node biopsy was performed in 408 patients, with clear nodal invasion in 102.
Table 1 shows descriptive statistics of patients included and corresponding standardized 25(OH)D3 levels at baseline. Median 25(OH)D3 at diagnosis was 49.0 nmol/L (IQR = 35.7–64.9). The median delay between diagnosis and first 25(OH)D3 measurement was 39 days (IQR = 24–77). 25(OH)D3 levels at diagnosis were inversely associated with melanoma prognostic factors, such as AJCC stage (P < .001 Kruskal-Wallis test), Breslow’s thickness in mm (P < .001 Spearman correlation), and ulceration (P < .001 Kruskal-Wallis test).
Descriptive statistics of main prognosis factors and 25(OH)D3 level at baseline in 1171 MelanCohort melanoma patients
| Prognostic factor | Number of cases | Median 25(OH)D3 level at baseline* (in nmol/L) (IQR) |
|---|---|---|
| Age, y | ||
| <50 | 453 | 47.91 (36.71-62.28) |
| 50–69 | 485 | 51.35 (36.96-68.05) |
| 70+ | 233 | 46.11 (31.92-61.03) |
| Sex | ||
| Men | 579 | 49.42 (36.58-64.36) |
| Women | 592 | 48.13 (35.42-66.36) |
| AJCC | ||
| IA | 425 | 52.07 (39.11-66.79) |
| IB | 218 | 52.27 (38.37-66.57) |
| IIA | 124 | 48.69 (35.66-70.20) |
| IIB | 85 | 45.59 (37.73-69.03) |
| IIC | 39 | 41.15 (28.83-59.43) |
| IIIA | 61 | 49.64 (35.42-59.90) |
| IIIB | 88 | 44.11 (33.77-60.11) |
| IIIC | 58 | 40.74 (30.36-60.61) |
| IV | 70 | 35.42 (25.63-48.69) |
| Missing | 3 | 42.44 (36.07-48.81) |
| Breslow, mm | ||
| 0–0.5 | 196 | 52.20 (39.17-66.35) |
| 0.5–1 | 255 | 51.72 (38.52-67.71) |
| 1–2 | 325 | 49.07 (37.45-66.04) |
| 2–3 | 145 | 46.50 (34.69-61.28) |
| 3–4 | 69 | 47.50 (35.89-64.36) |
| 4+ | 136 | 41.58 (30.15-59.43) |
| Missing | 45 | 42.58 (29.89-61.00) |
| Ulceration | ||
| No | 781 | 50.63 (37.69-66.36) |
| Yes | 225 | 45.14 (31.34-61.00) |
| Missing | 165 | 43.57 (33.77-58.60) |
| Prognostic factor | Number of cases | Median 25(OH)D3 level at baseline* (in nmol/L) (IQR) |
|---|---|---|
| Age, y | ||
| <50 | 453 | 47.91 (36.71-62.28) |
| 50–69 | 485 | 51.35 (36.96-68.05) |
| 70+ | 233 | 46.11 (31.92-61.03) |
| Sex | ||
| Men | 579 | 49.42 (36.58-64.36) |
| Women | 592 | 48.13 (35.42-66.36) |
| AJCC | ||
| IA | 425 | 52.07 (39.11-66.79) |
| IB | 218 | 52.27 (38.37-66.57) |
| IIA | 124 | 48.69 (35.66-70.20) |
| IIB | 85 | 45.59 (37.73-69.03) |
| IIC | 39 | 41.15 (28.83-59.43) |
| IIIA | 61 | 49.64 (35.42-59.90) |
| IIIB | 88 | 44.11 (33.77-60.11) |
| IIIC | 58 | 40.74 (30.36-60.61) |
| IV | 70 | 35.42 (25.63-48.69) |
| Missing | 3 | 42.44 (36.07-48.81) |
| Breslow, mm | ||
| 0–0.5 | 196 | 52.20 (39.17-66.35) |
| 0.5–1 | 255 | 51.72 (38.52-67.71) |
| 1–2 | 325 | 49.07 (37.45-66.04) |
| 2–3 | 145 | 46.50 (34.69-61.28) |
| 3–4 | 69 | 47.50 (35.89-64.36) |
| 4+ | 136 | 41.58 (30.15-59.43) |
| Missing | 45 | 42.58 (29.89-61.00) |
| Ulceration | ||
| No | 781 | 50.63 (37.69-66.36) |
| Yes | 225 | 45.14 (31.34-61.00) |
| Missing | 165 | 43.57 (33.77-58.60) |
* 25(OH)D3 levels were standardized on age, sex, body mass index, and month of blood sampling. IQR = interquartile range.
Descriptive statistics of main prognosis factors and 25(OH)D3 level at baseline in 1171 MelanCohort melanoma patients
| Prognostic factor | Number of cases | Median 25(OH)D3 level at baseline* (in nmol/L) (IQR) |
|---|---|---|
| Age, y | ||
| <50 | 453 | 47.91 (36.71-62.28) |
| 50–69 | 485 | 51.35 (36.96-68.05) |
| 70+ | 233 | 46.11 (31.92-61.03) |
| Sex | ||
| Men | 579 | 49.42 (36.58-64.36) |
| Women | 592 | 48.13 (35.42-66.36) |
| AJCC | ||
| IA | 425 | 52.07 (39.11-66.79) |
| IB | 218 | 52.27 (38.37-66.57) |
| IIA | 124 | 48.69 (35.66-70.20) |
| IIB | 85 | 45.59 (37.73-69.03) |
| IIC | 39 | 41.15 (28.83-59.43) |
| IIIA | 61 | 49.64 (35.42-59.90) |
| IIIB | 88 | 44.11 (33.77-60.11) |
| IIIC | 58 | 40.74 (30.36-60.61) |
| IV | 70 | 35.42 (25.63-48.69) |
| Missing | 3 | 42.44 (36.07-48.81) |
| Breslow, mm | ||
| 0–0.5 | 196 | 52.20 (39.17-66.35) |
| 0.5–1 | 255 | 51.72 (38.52-67.71) |
| 1–2 | 325 | 49.07 (37.45-66.04) |
| 2–3 | 145 | 46.50 (34.69-61.28) |
| 3–4 | 69 | 47.50 (35.89-64.36) |
| 4+ | 136 | 41.58 (30.15-59.43) |
| Missing | 45 | 42.58 (29.89-61.00) |
| Ulceration | ||
| No | 781 | 50.63 (37.69-66.36) |
| Yes | 225 | 45.14 (31.34-61.00) |
| Missing | 165 | 43.57 (33.77-58.60) |
| Prognostic factor | Number of cases | Median 25(OH)D3 level at baseline* (in nmol/L) (IQR) |
|---|---|---|
| Age, y | ||
| <50 | 453 | 47.91 (36.71-62.28) |
| 50–69 | 485 | 51.35 (36.96-68.05) |
| 70+ | 233 | 46.11 (31.92-61.03) |
| Sex | ||
| Men | 579 | 49.42 (36.58-64.36) |
| Women | 592 | 48.13 (35.42-66.36) |
| AJCC | ||
| IA | 425 | 52.07 (39.11-66.79) |
| IB | 218 | 52.27 (38.37-66.57) |
| IIA | 124 | 48.69 (35.66-70.20) |
| IIB | 85 | 45.59 (37.73-69.03) |
| IIC | 39 | 41.15 (28.83-59.43) |
| IIIA | 61 | 49.64 (35.42-59.90) |
| IIIB | 88 | 44.11 (33.77-60.11) |
| IIIC | 58 | 40.74 (30.36-60.61) |
| IV | 70 | 35.42 (25.63-48.69) |
| Missing | 3 | 42.44 (36.07-48.81) |
| Breslow, mm | ||
| 0–0.5 | 196 | 52.20 (39.17-66.35) |
| 0.5–1 | 255 | 51.72 (38.52-67.71) |
| 1–2 | 325 | 49.07 (37.45-66.04) |
| 2–3 | 145 | 46.50 (34.69-61.28) |
| 3–4 | 69 | 47.50 (35.89-64.36) |
| 4+ | 136 | 41.58 (30.15-59.43) |
| Missing | 45 | 42.58 (29.89-61.00) |
| Ulceration | ||
| No | 781 | 50.63 (37.69-66.36) |
| Yes | 225 | 45.14 (31.34-61.00) |
| Missing | 165 | 43.57 (33.77-58.60) |
* 25(OH)D3 levels were standardized on age, sex, body mass index, and month of blood sampling. IQR = interquartile range.
During follow-up, 411 individuals experienced relapse, and 303 died. Table 2 presents the Cox model on risk of relapse, adjusting on age, sex, BMI, AJCC stage, and 25(OH)D3 level at diagnosis. When adjusting for known melanoma prognostic factors, the standardized baseline 25(OH)D3 serum level was not associated with prognosis.
Cox proportional hazards model on the risk of relapse (410 events) in 1171 melanoma patients participating to MelanCohort in France
| Parameter | Number of cases | HR* (95 % CI) |
|---|---|---|
| Age, y (continuous per individual year) | 1171 | 1.02 (1.01 to 1.03) |
| Sex | ||
| Male | 579 | 1 (Ref) |
| Female | 592 | 0.83 (0.68 to 1.02) |
| BMI (continuous per unit of BMI) | 1171 | 1.02 (0.99 to 1.04) |
| AJCC stage | ||
| IA | 425 | 1 (Ref) |
| IB | 218 | 2.61 (1.70 to 4.03) |
| IIA | 124 | 3.86 (2.46 to 6.05) |
| IIB | 85 | 7.46 (4.81 to 11.58) |
| IIC | 39 | 10.52 (6.29 to 17.58) |
| IIIA | 61 | 7.06 (4.33 to 11.49) |
| IIIB | 88 | 7.69 (5.02 to 11.76) |
| IIIC | 58 | 13.70 (8.71 to 21.55) |
| IV | 70 | 10.30 (6.67 to 15.91) |
| 25(OH)D3 baseline†, nmol/L | ||
| <37.15 | 312 | 0.95 (0.71 to 1.27) |
| 37.15;49.77 | 288 | 1.04 (0.77 to 1.40) |
| 49.77;66.05 | 273 | 1 (Ref) |
| 66.05+ | 274 | 0.99 (0.72 to 1.36) |
| Missing | 24 | 2.25 (0.98 to 5.17) |
| Change in 25(OH)D3† level during follow-up (in nmol/L/year) | ||
| <-5.25 | 214 | 1.94 (1.36 to 2.76) |
| -5.25;-0.30 | 214 | 1.23 (0.85 to 1.78) |
| -0.30;4.60 | 214 | 1 (Ref) |
| 4.60+ | 214 | 1.61 (1.14 to 2.28) |
| Only one test | 291 | 2.11 (1.50 to 2.95) |
| Missing | 24 | ‡ |
| Parameter | Number of cases | HR* (95 % CI) |
|---|---|---|
| Age, y (continuous per individual year) | 1171 | 1.02 (1.01 to 1.03) |
| Sex | ||
| Male | 579 | 1 (Ref) |
| Female | 592 | 0.83 (0.68 to 1.02) |
| BMI (continuous per unit of BMI) | 1171 | 1.02 (0.99 to 1.04) |
| AJCC stage | ||
| IA | 425 | 1 (Ref) |
| IB | 218 | 2.61 (1.70 to 4.03) |
| IIA | 124 | 3.86 (2.46 to 6.05) |
| IIB | 85 | 7.46 (4.81 to 11.58) |
| IIC | 39 | 10.52 (6.29 to 17.58) |
| IIIA | 61 | 7.06 (4.33 to 11.49) |
| IIIB | 88 | 7.69 (5.02 to 11.76) |
| IIIC | 58 | 13.70 (8.71 to 21.55) |
| IV | 70 | 10.30 (6.67 to 15.91) |
| 25(OH)D3 baseline†, nmol/L | ||
| <37.15 | 312 | 0.95 (0.71 to 1.27) |
| 37.15;49.77 | 288 | 1.04 (0.77 to 1.40) |
| 49.77;66.05 | 273 | 1 (Ref) |
| 66.05+ | 274 | 0.99 (0.72 to 1.36) |
| Missing | 24 | 2.25 (0.98 to 5.17) |
| Change in 25(OH)D3† level during follow-up (in nmol/L/year) | ||
| <-5.25 | 214 | 1.94 (1.36 to 2.76) |
| -5.25;-0.30 | 214 | 1.23 (0.85 to 1.78) |
| -0.30;4.60 | 214 | 1 (Ref) |
| 4.60+ | 214 | 1.61 (1.14 to 2.28) |
| Only one test | 291 | 2.11 (1.50 to 2.95) |
| Missing | 24 | ‡ |
* Hazard ratios from a Cox proportional hazards model adjusting on age, sex, body mass index, American Joint Committee on Cancer stage, 25(OH)D3 at baseline and change in 25(OH)D3 during follow-up. AJCC = American Joint Committee on Cancer; BMI = body mass index; CI = confidence interval; HR = hazard ratio.
† 25(OH)D3 levels were standardized on age, sex, BMI, and month of blood sampling.
‡ Same as 25(OH)D3 at baseline.
Cox proportional hazards model on the risk of relapse (410 events) in 1171 melanoma patients participating to MelanCohort in France
| Parameter | Number of cases | HR* (95 % CI) |
|---|---|---|
| Age, y (continuous per individual year) | 1171 | 1.02 (1.01 to 1.03) |
| Sex | ||
| Male | 579 | 1 (Ref) |
| Female | 592 | 0.83 (0.68 to 1.02) |
| BMI (continuous per unit of BMI) | 1171 | 1.02 (0.99 to 1.04) |
| AJCC stage | ||
| IA | 425 | 1 (Ref) |
| IB | 218 | 2.61 (1.70 to 4.03) |
| IIA | 124 | 3.86 (2.46 to 6.05) |
| IIB | 85 | 7.46 (4.81 to 11.58) |
| IIC | 39 | 10.52 (6.29 to 17.58) |
| IIIA | 61 | 7.06 (4.33 to 11.49) |
| IIIB | 88 | 7.69 (5.02 to 11.76) |
| IIIC | 58 | 13.70 (8.71 to 21.55) |
| IV | 70 | 10.30 (6.67 to 15.91) |
| 25(OH)D3 baseline†, nmol/L | ||
| <37.15 | 312 | 0.95 (0.71 to 1.27) |
| 37.15;49.77 | 288 | 1.04 (0.77 to 1.40) |
| 49.77;66.05 | 273 | 1 (Ref) |
| 66.05+ | 274 | 0.99 (0.72 to 1.36) |
| Missing | 24 | 2.25 (0.98 to 5.17) |
| Change in 25(OH)D3† level during follow-up (in nmol/L/year) | ||
| <-5.25 | 214 | 1.94 (1.36 to 2.76) |
| -5.25;-0.30 | 214 | 1.23 (0.85 to 1.78) |
| -0.30;4.60 | 214 | 1 (Ref) |
| 4.60+ | 214 | 1.61 (1.14 to 2.28) |
| Only one test | 291 | 2.11 (1.50 to 2.95) |
| Missing | 24 | ‡ |
| Parameter | Number of cases | HR* (95 % CI) |
|---|---|---|
| Age, y (continuous per individual year) | 1171 | 1.02 (1.01 to 1.03) |
| Sex | ||
| Male | 579 | 1 (Ref) |
| Female | 592 | 0.83 (0.68 to 1.02) |
| BMI (continuous per unit of BMI) | 1171 | 1.02 (0.99 to 1.04) |
| AJCC stage | ||
| IA | 425 | 1 (Ref) |
| IB | 218 | 2.61 (1.70 to 4.03) |
| IIA | 124 | 3.86 (2.46 to 6.05) |
| IIB | 85 | 7.46 (4.81 to 11.58) |
| IIC | 39 | 10.52 (6.29 to 17.58) |
| IIIA | 61 | 7.06 (4.33 to 11.49) |
| IIIB | 88 | 7.69 (5.02 to 11.76) |
| IIIC | 58 | 13.70 (8.71 to 21.55) |
| IV | 70 | 10.30 (6.67 to 15.91) |
| 25(OH)D3 baseline†, nmol/L | ||
| <37.15 | 312 | 0.95 (0.71 to 1.27) |
| 37.15;49.77 | 288 | 1.04 (0.77 to 1.40) |
| 49.77;66.05 | 273 | 1 (Ref) |
| 66.05+ | 274 | 0.99 (0.72 to 1.36) |
| Missing | 24 | 2.25 (0.98 to 5.17) |
| Change in 25(OH)D3† level during follow-up (in nmol/L/year) | ||
| <-5.25 | 214 | 1.94 (1.36 to 2.76) |
| -5.25;-0.30 | 214 | 1.23 (0.85 to 1.78) |
| -0.30;4.60 | 214 | 1 (Ref) |
| 4.60+ | 214 | 1.61 (1.14 to 2.28) |
| Only one test | 291 | 2.11 (1.50 to 2.95) |
| Missing | 24 | ‡ |
* Hazard ratios from a Cox proportional hazards model adjusting on age, sex, body mass index, American Joint Committee on Cancer stage, 25(OH)D3 at baseline and change in 25(OH)D3 during follow-up. AJCC = American Joint Committee on Cancer; BMI = body mass index; CI = confidence interval; HR = hazard ratio.
† 25(OH)D3 levels were standardized on age, sex, BMI, and month of blood sampling.
‡ Same as 25(OH)D3 at baseline.
Overall, 856 individuals had at least two values of serum 25(OH)D3 level measured, allowing to calculate an annual trend of variation of 25(OH)D3 serum level (Supplementary Figure 1, available online). Visual inspection of the 25(OH)D3 and residuals of the predicted values at the individual level did not show departure from the linear assumption (P = .40). The annual variations of standardized 25(OH)D3 serum level over time, either increasing or decreasing, were statistically significantly associated with worse prognosis (Table 2): with a third quartile Q3 of average change per year (-0.30 to 4.60 nmol/L/Y) used as reference, hazard ratios for the first, second, and fourth quarters were 1.94 (95% confidence interval [CI] = 1.36 to -2.76), 1.23 (95% CI = 0.85 to -1.78), and 1.61 (95% CI = 1.14 to -2.28), respectively. Individuals who experienced increases above 4.60 nmol/L per year or decreases below 5.25 nmol/L per year had a worse DFS in the Kaplan Meier curve (Figure 2). Figure 3 presents the same analysis as Table 2 but with more categories of 25(OH)D3 trend. A change of less than 0.5 nmol/L per year, either increasing or decreasing, was used as reference category. The figure is U-shaped, any variation above or under 0.5 nmol/L per year being associated with worse prognosis, with risk being even greater when variations were above or below 3 nmol/L per year.
Kaplan-Meier representation of disease-free survival by quarters of 25(OH)D3 yearly trend.
Hazard ratio (HR) of risk of relapse by yearly trend in 25(OH)D3 during follow-up of melanoma patients participating to MelanCohort adjusting on age, sex, body mass index, American Joint Committee on Cancer stage, and baseline 25(OH)D3 level. The reference category is -0.5 to 0.5 nmol/L/year. Gray vertical lines correspond to quartile of the distribution of the yearly trend in 25(OH)D3. CI = confidence interval.
In sensitivity analyses, the hazard ratio for the effect of variation of standardized 25(OH)D3 serum level remained of similar magnitude when using different initial period for standardization or restricting the sample to individuals with a short delay between diagnosis and first 25(OH)D3 measurement, (Supplementary Table 2, available online). The sensitivity analysis was also conducted by AJCC stage groups (Supplementary Table 3, available online) or by number of 25(OH)D3 measures performed. Overall, 76 individuals (6.5%) had detectable serum levels of 25(OH)D2, a source of some vitD supplementations in France, at some point during follow-up. They were on average younger (median age: 49 years for those with detectable serum level of 25(OH)D2 vs 56 for others, P = .004) and more frequently women (7.77% vs 5.18% for men, P = .07). No statistically significant association was found between having detectable 25(OH)D2 level and Breslow’s thickness (P = .24) or AJCC stage (P = .72). Serum level of 25(OH)D2 was independent of the first 25(OH)D3 serum level (P = .99) nor of any variation over time (P = .93).
Reverse causation (24) could be a likely explanation between change in 25(OH)D3 and prognosis if a rapid change in 25(OH)D3 level occurred shortly before distant metastasis development, while we assumed a linear change in 25(OH)D3 level. Hence, the analysis was replicated excluding the last 25(OH)D3 measurement if occurring within six months before the event or right censoring and provided similar results to the full analysis, without any modification of point estimates or statistical significance (data not shown). Biases could also be caused by heterogeneity in measurements, with patients with short survival having 25(OH)D3 slopes estimated with fewer tests and larger standard errors than patients with longer survival. We thus replicated the analysis, taking only the initial year of 25(OH)D3 measurement and individuals with at least one year of follow-up (Supplementary Table 4, available online) and found statistically nonsignificant results, likely because of the low power of this analysis, the reference category for the trend being based on 26 individuals and four events. To circumvent this methodological problem, we replicated the main analysis but stratified the number of tests per individual (Supplementary Table 5, available online). This analysis, for which only individuals with similar numbers of tests are compared in each stratum, resulted in similar results as the main analysis.
Analyses performed on OS yielded similar results, with no effect of baseline 25(OH)D3 on survival, but with changes of 25(OH)D3 through time associated with worse prognosis (Supplementary Table 6, available online).
Characteristics of individuals who experienced an important increase or decrease of 25(OH)D3 serum levels with time are given in Supplementary Table 7 (available online). No association was found with age or sex. These individuals had slightly worse presentation, with more advanced AJCC stage, thicker or ulcerated primary melanoma. A correlation was observed between baseline 25(OH)D3 level and variation of 25(OH)D3 with time, which appeared artifactual, as individuals with already high 25(OH)D3 level at baseline have little chance to see its increase with time.
Because our results differed markedly from those of Julia Newton-Bishop et al. (21), we applied to our data the same modeling (without the Townsend score, unrecorded in our study). When adjusting for age as continuous variable, sex, BMI, anatomical body site, and Breslow’s thickness as a continuous variable, statistically significantly improved prognosis was found with increased initial 25(OH)D3 levels (P = .04). This yielded a hazard ratio of 0.90 (95% CI = 0.82 to 0.99) for each 20 nmol/L increase. But, when adjusting for Breslow’s thickness as a categorical variable, the association with 25(OH)D3 was no longer statistically significant (HR = 0.997, 95% CI = 0.992 to 1.002 per 1 nmol/L increase, P = .19,). It became also flat when adjusting for AJCC stage (HR = 1.001, 95% CI = 0.996 to 1.006 per 1 nmol/L increase, P = .75). A further analysis stratified on AJCC stage led to flat result (HR = 1.001, 95% CI = 0.996 to 1.006, P = .8). Figure 4 shows the hazard ratio on the risk of relapse for the same Cox model using Breslow’s thickness either as a continuous variable, ie, linear parameter, or as a categorical variable to account for nonlinearity.
Hazard ratio of risk of relapse according to Breslow’s thickness (in mm) entered either as a linear (dashed line) or categorical variable (plain line). Hazard ratios are displayed on a log scale. Cox proportional hazards model was used adjusted on age, sex, body mass index, anatomical body site, and Breslow’s thickness. Vertical gray lines correspond to quartiles of the distribution of Breslow’s thickness (0.6, 1.25, 2.3mm).
Discussion
The analysis of 25(OH)D3 serum level in this prospective cohort of carefully followed-up melanoma patients enabled us to confirm an already described association between low 25(OH)D3 serum level at diagnosis and prognostic factors, such as higher Breslow’s thickness (19) or higher AJCC melanoma stage (20) and, to the best of our knowledge for the first time, to investigate the role of its variations during follow-up. As expected by the associations at diagnosis, in crude analysis, low 25(OH)D3 levels were associated with poorer DFS and OS. However, as soon as proper adjustments were used for prognostic factors such as AJCC stage, or stratifying the analysis on prognostic markers, this association totally disappeared. Thus, in our study, 25(OH)D3 serum level at diagnosis was not an independent melanoma prognostic marker.
These results strongly differ from those of J. Newton-Bishop et al. (21). Our analysis showed that a statistically significant association could be evidenced only when poorly accounting for prognostic factors, such as in unadjusted analysis or when using Breslow’s thickness as a continuous variable. This was further confirmed by a stratified analysis. It appeared from the comparison of both approaches (Figure 4) that Breslow’s thickness used as raw continuous variable would be systematically underadjusted for values above 2mm and overadjusted for thin melanomas. Thus, the approach used by Julia Newton-Bishop et al. likely led to biased results.
Noteworthy, our study had some methodological advantages: It followed the REMARK guidelines for tumor markers in prognostic studies (25), patient losses to follow-up were limited, serum samples were carefully handled, and 25(OH)D3 serum levels were measured by mass spectrometry, considered the most accurate method (26). Some widely used techniques of 25(OH)D3 serum level measurement are easier, but are hampered by high intermethod variability (26). Our method of standardization of 25(OH)D3 serum level used adjustments to the main factors that influence its level in the normal population (27). Although we cannot exclude that unmeasured confounders biased our results, we believe this method of standardization is accurate.
Although the baseline value of standardized 25(OH)D3 serum level was not an independent prognostic marker, its variation through time was. The use of time-dependent variables as prognostic markers is always a limitation because potential bias could arise, such as reverse causation. Our analysis did not indicate evidence of reverse causation in the analysis of changes in 25(OH)D3 serum level. Because of our careful standardization of 25(OH)D3 serum level, we think this result valuable. Noteworthy, a U-shaped association between 25(OH)D3 serum levels at diagnosis and prognosis has been shown for prostate cancer, with both high and low levels at diagnosis being associated with higher-grade disease (28).
The biological significance of our finding on the prognostic value of changes in 25(OH)D3 levels over time is unclear. A decrease in 25(OH)D3 levels upon time may be associated with inflammation, and serum concentrations of tumor necrosis factor-α or C-reactive protein have been repeatedly reported inversely correlated with 25(OH)D3 concentrations (7). Driver mutations in melanoma cells increase secretion of pro-inflammatory cytokines such as interleukin-6 and -8 and there is also evidence that ulceration of the primary lesion, a potent poor prognostic factor for melanoma, is associated with inflammation (29). Alternatively, progression is associated with poorer global health, thus limiting sun exposure, the main source of vitD. It is more difficult to understand the cause of the increase of 25(OH)D3 over time. VitD3 supplementation may be an explanation, but we did not record prospectively use of vitD supplements. However, we measured vitD2 levels, a method of vitD supplementation used in France, and failed to show that vitD2 supplementation had any impact on variation over time of 25(OH)D3 levels. Nevertheless, we cannot exclude that some patients ingested vitD3 as supplements, nor did we measure patients’ expositions to the sun. In fact, all patients were instructed to refrain from exposing their skin to the sun. Our results do not also favor the hypothesis that vitD might mediate the reported lowered risk of relapse for melanoma patients who have sunny holidays after melanoma diagnosis (30).
Because a change of 25(OH)D3 serum level over time in both directions was associated with worse prognosis in melanoma patients, it is unlikely a direct consequence of vitD biological actions. We postulate that fluctuation of 25(OH)D3 serum level reflected a global instability in the patient’s metabolism, which finally impacts 25(OH)D3 serum level by any of the multiple pathways in the vitD3 regulation (31). Although our results should be replicated, they add a new cautionary note to widespread use of vitD3 supplementation in melanoma patients in order to improve survival. In fact, vitD3 supplementation failed to prevent melanoma in two prospective cohort studies in the general population (32,33). The J. Newton-Bishop et al. study reported vitD3 supplementation was not associated with improved survival (21). Results of ongoing clinical trials on vitD supplementation to prevent relapse in high-risk melanoma are awaited (34), as well as large supplementation trials in the general population (35).
Funding
The collection of samples in the MelanCohort Study was funded by two grants from the French ministry of Health (PHRC R 2002 AOR02089, PHRC N 2006 AOM06200), “Assistance Publique - Hôpitaux de Paris” being the sponsor. This work was supported by the French National Cancer Institute (Canceropole Ile de France, France) through a grant (grant number 2010-1-PL_SHS-15) and by a Catinvest unrestricted grant.
The study sponsors had no role in the design of the study, nor in the collection, analysis, and interpretation of the data. They had also no role in the writing of the manuscript or in the decision to submit the manuscript for publication.
Amandine Canivet from the Department of Clinical Biochemistry at Saint Louis hospital provided inestimable help in the measurement of serum vitamin D2/D3. The investigators thank the Clinical Research Unit team, which worked on MelanCohort (N. Derridj, Y. Saidji, P. Martel-Samb, F. Sabri, S. Klotz, W. Teng).
This work has been presented in part at the 2014 American Society of Clinical Oncology Annual Meeting (Abstract #9057, http://meetinglibrary.asco.org/content/92147?media=vm).
We have no conflicts of interest to declare.
References
