Prediagnostic Serum Vitamin D, Vitamin D Binding Protein Isoforms, and Cancer Survival

Abstract

Background

Higher circulating vitamin D has been associated with improved overall cancer survival, but data for organ-specific cancers are mixed.

Methods

We examined the association between prediagnostic serum 25-hydroxyvitamin D [25(OH)D], the recognized biomarker of vitamin D status, and cancer survival in 4038 men and women diagnosed with 1 of 11 malignancies during 22 years of follow-up (median = 15.6 years) within the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. Multivariable-adjusted proportional hazards regression estimated hazard ratios (HR) and 95% confidence intervals (CI) for associations between baseline 25(OH)D concentration and subsequent cancer survival; we also stratified on the common vitamin D binding protein isoforms (Gc1f, Gc1s, and Gc2) defined by two single-nucleotide polymorphisms (rs7041 and rs4588) in the vitamin D binding protein gene GC. All P values were 2-sided.

Results

Higher 25(OH)D concentrations were associated with greater overall cancer survival (HR for cancer mortality = 0.83, 95% CI = 0.70 to 0.98 for highest vs lowest quintile; P_trend = .05) and lung cancer survival (HR = 0.63, 95% CI = 0.44 to 0.90; P_trend = .03). These associations were limited to cases expressing the Gc2 isoform (HR = 0.38 for Gc2-2, 95% CI = 0.14 to 1.05 for highest vs lowest quintile; P_trend = .02; and HR = 0.30 for Gc1-2/Gc2-2 combined, 95% CI = 0.16 to 0.56; P_trend < .001 for overall and lung cancer, respectively).

Conclusions

Higher circulating 25(OH)D was associated with improved overall and lung cancer survival. As this was especially evident among cases with the genetically determined Gc2 isoform of vitamin D binding protein, such individuals may gain a cancer survival advantage by maintaining higher 25(OH)D blood concentrations.

Higher concentrations of prediagnostic circulating vitamin D have been prospectively associated with improved overall cancer survival in most (1-9), but not all (10,11), studies, although fewer investigations have examined organ-specific cancers. Although vitamin D is well-known for its influence on bone health (12), compelling evidence from animal and laboratory studies demonstrate anticancer properties for the active form of vitamin D (1,25-dihydroxyvitamin D or 1,25(OH)₂D), including increasing apoptosis and cellular differentiation, anti-inflammation, anti-angiogenesis, and antiproliferation (12-15). These properties could also account for a role of vitamin D in increasing cancer survival and inhibiting progression (ie, not only initiation) (16), as has been demonstrated in cell culture experiments (17). Recent studies found that associations between circulating vitamin D and risk of colorectal adenoma, colorectal cancer, and colorectal cancer mortality (18-20) differ based on the inherited vitamin D binding protein (DBP) isoform. DBP, also known as group-specific component (Gc), is the major transport protein of vitamin D and its metabolites (21). We used serum 25-hydroxyvitamin D [25(OH)D], the recognized biomarker of vitamin D status, to examine the association between prediagnostic circulating vitamin D and overall and organ-specific cancer survival within the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO), including stratifying by genetically defined DBP isoform.

Methods

Study Population

The present study is a secondary analysis including 4038 cancer cases from previously published PLCO nested case-control studies of serum vitamin D for the following 11 cancer sites: breast (n = 785), prostate (one study of 748 White cases and another of 226 Black cases), colorectum (n = 475), lung (n = 450), bladder (n = 365), hematopoietic cancer (n = 291), pancreas (n = 229), kidney (n = 158), endometrium (n = 144), upper gastrointestinal tract (n = 93), and ovary (n = 74) (22-33). See the Supplementary Methods (available online) for details of the PLCO Trial and cancer case identification. Details on how sociodemographic and risk factor characteristics information were collected are also in the Supplementary Methods (available online). Written informed consent was obtained from each participant. The trial was approved by the institutional review boards of the National Cancer Institute and the 10 PLCO centers (23).

Vitamin D Status

Baseline circulating 25(OH)D was measured using radioimmunoassay (34,35), chemiluminescence immunoassay (36), or liquid chromatography and mass spectrometry (24,37). Supplementary Table 1 (available online) summarizes each nested case-control study including sample size, laboratory, assay type, assay coefficients of variation based on masked quality control samples, and assay-specific cut points (discussed below). Further descriptions can be found in the individual study publications (22-33,38). Circulating DBP was available for 1801 individuals and assayed as described (24,25,27,39).

DBP Isoforms

GC single-nucleotide polymorphisms rs7041 (Asp432Glu) and rs4588 (Thr436Lys) were extracted from existing genotype data (Supplementary Methods, available online). Based on combinations of rs7041 and rs4588, respectively (40), the DBP isoforms were defined as: Gc1s, Glu and Thr; Gc1f, Asp and Thr; and Gc2, Asp and Lys. These are also presented for both DNA strands as Gc1s-Gc1s, Gc1s-Gc1f, Gc1f-Gc1f, Gc1s-Gc2, Gc1f-Gc2, and Gc2-Gc2 (Supplementary Table 2, available online) and then simplified by combining Gc1s and Gc1f: Gc1-1, Gc1-2, and Gc2-2.

Statistical Analysis

Survival time began on the date of cancer diagnosis and ended on the date of death or the censor date. For the all-cancer model, deaths were defined as any cancer death. For the organ-specific models, cancer death was defined as death attributed to that specific cancer site only. All other individuals with any other cause of death, including death from other cancers, were considered cancer survivors for the specific organ-site models and censored at their death date; those who were alive were censored at the end of follow-up (December 31, 2015). Analyses where other causes of death were considered as competing risks did not materially alter the findings (data not shown).

Demographic and other factors for those who died of cancer vs cancer survivors were compared using Wilcoxon rank sum tests and χ² tests (for continuous and categorical variables, respectively). Hazard ratios (HR) and 95% confidence intervals (CI) for cancer mortality among those who developed cancer in PLCO, by 25(OH)D levels, were calculated using Cox proportional hazards regression. The hazard proportionality assumption was tested by modeling the interaction of follow-up time (log) with vitamin D quintile, and no evidence of a violation was found. Because different laboratories and methods were used for the vitamin D assays in the nested case-control studies, separate 25(OH)D quantiles were created by laboratory and assay (n = 5) and by sex (n = 2). In addition, because of known seasonal variation in 25(OH)D concentrations, quantiles were also created separately for winter (December-May) and summer (June-November) seasons, as previously defined for PLCO (25) (see Supplementary Table 1, available online). For the all-cancers analysis as well as for the most common cancers (eg, breast, lung, and prostate), vitamin D was categorized into quintiles, whereas for organ sites with fewer cases (eg, pancreas, kidney, and endometrium), tertiles were selected based on the number of cases, deaths, and survivors, to produce more stable risk estimates. Tests for linear trend were conducted by coding each category either 1-5 (for quintiles) or 1-3 (for tertiles) and treating these as continuous variables.

Multivariable-adjusted models for all cancers combined included the following factors potentially associated with survival: age at diagnosis, sex, race, body mass index (BMI), smoking status, physical activity, history of diabetes, family history of cancer, calendar year of diagnosis, cancer stage, cancer grade, and cancer site. Further adjustment for age at blood collection did not alter the results (<10% change in HRs). Organ-specific models were not adjusted for cancer site, or for sex or race for single sex or race groupings. Persons with missing covariates were retained in separate categories for the relevant factor. The lung cancer model was additionally adjusted for cigar, pipe, and pack-years of smoking. Using cigarettes per day and years of smoking rather than pack-years did not alter the lung cancer results, nor did adding any of these variables to the all-cancer model. Parity, duration of oral contraceptive use, and menopausal status and duration of menopausal hormone use (all measured at baseline) were added to the breast, endometrial, and ovarian cancer models and curative surgery was added to the pancreatic cancer models, but these adjustments did not change the interpretation of the study results and, hence, were not retained in the final models.

We explored whether the association of circulating 25(OH)D with all-cancer survival varied by season of blood collection, latitude, sex, smoking status, BMI, age at diagnosis, year of diagnosis, time elapsed from blood draw to cancer diagnosis, and cancer stage and grade (for the latter two, cases diagnosed with cancer within 2 years of the baseline blood draw were excluded). The log-likelihood test was used to statistically evaluate effect modification by comparing models with and without a cross-product term of the vitamin D quantiles and the effect modifier.

We also conducted analyses stratifying circulating 25(OH)D by the GC isoform in the all-cancer model as well as among the 4 cancer sites with the largest case numbers (ie, breast, prostate, colorectum, and lung). Statistical analyses were conducted using SAS version 9.4 (SAS Institute, Inc, Cary, NC), and all P values were 2-sided, with a P value less than .05 considered statistically significant.

Results

Of the study participants, 41.6% were female, and 58.4% were male. The racial and ethnic distribution was as follows: American Indian (0.2%); Asian (2.0%); Black, non-Hispanic (9.0%); Hispanic (0.7%); Pacific Islander (0.2%); and White, non-Hispanic (87.9%). Median age at blood collection was 64 years and at cancer diagnosis was 69 years, with a median time from blood draw to diagnosis of 4.4 years. Cancer survivors compared with nonsurvivors were more likely to be female, Black individuals, never-smokers, younger at blood collection and cancer diagnosis, and diagnosed in the winter, during the earlier years of the trial, and with earlier stage and lower-grade cancers. Unadjusted median baseline 25(OH)D concentrations did not differ, however (Table 1). Those who were current smokers or Black individuals, or who had higher BMI, less physical activity, or diabetes tended to have lower vitamin D concentrations (Supplementary Table 3, available online).

Table 1.

Selected characteristics at the time of blood draw, by cancer survivor status in the PLCO Trial

Characteristic	Died of cancer (n = 1416)	Cancer survivorsa(n = 2622)	P
Median serum 25(OH)D (10th-90th), nmol/L	57.7 (31.1-90.1)	57.7 (31.7-86.9)	.78b
Median age at blood collection (10th-90th), y	65 (58-72)	63 (56-71)	<.001b
Median age at diagnosis (10th-90th), y	70 (62-78)	68 (60-76)	<.001b
Sex, No. (%)			<.001c
Female	513 (36.2)	1168 (44.6)
Male	903 (63.8)	1454 (55.5)
Race, No. (%)			.03c
American Indian	1 (0.1)	5 (0.2)
Asian	31 (2.2)	51 (2.0)
Black, non-Hispanic	101 (7.1)	264 (10.1)
Hispanic	12 (0.9)	18 (0.7)
Pacific Islander	1 (0.1)	6 (0.2)
White, non-Hispanic	1270 (89.7)	2278 (86.9)
Median BMI (10th-90th), kg/m2	26.7 (22.1-33.3)	26.7 (22.0-33.5)	.90b
Smoking status, No. (%)			<.001c
Never	461 (32.6)	1166 (44.5)
Former	723 (51.1)	1208 (46.1)
Current	232 (16.4)	247 (9.4)
Physical activity, No. (%), h/wkd			.38c
None or <1	461 (32.6)	785 (29.9)
1-2	359 (25.4)	686 (26.2)
≥3	504 (35.6)	966 (36.8)
Missing	92 (6.5)	185 (7.1)
History of diabetes, No. (% yes)	126 (8.9)	220 (8.4)	.81c
Family history of cancer, No. (% yes)e	807 (57.0)	1517 (57.9)	.80c
Calendar year of diagnosis, No. (%)			<.001c
1994-1998	255 (18.0)	609 (23.2)
1999-2003	720 (50.9)	1428 (54.5)
2004-2008	405 (28.6)	514 (19.6)
2009-2013	36 (2.5)	71 (2.7)
Winter blood collection (December-May), No. (%)	666 (47.0)	1320 (50.3)	.045c
Latitude of study center, No. (%)			.92c
<34oN	92 (6.5)	176 (6.7)
34oN to <42oN	682 (48.2)	1273 (48.6)
≥42oN	642 (45.3)	1173 (44.7)
Stage at cancer diagnosis, No. (%)			<.001c
1	136 (9.6)	718 (27.4)
2	217 (15.3)	1013 (38.6)
3	210 (14.8)	216 (8.2)
4	260 (18.4)	31 (1.2)
Missing	593 (41.9)	644 (24.6)
Grade at cancer diagnosis No. (%)			<.001c
1	87 (6.1)	440 (16.8)
2	410 (29.0)	1311 (50.0)
3	397 (28.0)	497 (19.0)
4	37 (2.6)	35 (1.3)
Missing	485 (34.3)	339 (12.9)

^aCancer survivors are participants who were alive at the end of follow-up or, if deceased, whose cause of death was not cancer. 25(OH)D = 25-hydroxyvitamin D; BMI = body mass index; PLCO = Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial.

^b P values based on 2-sided Wilcoxon rank sum tests (for continuous variables).

^c P values based on 2-sided χ² tests (for categorical variables).

^dHours per week spent in vigorous activities.

^eFamily history of any cancer reported in first-degree relatives (including parents, full siblings, and children).

In multivariable-adjusted models, higher serum 25(OH)D was associated with improved overall cancer survival (HR for cancer mortality = 0.83, 95% CI = 0.70 to 0.98 for highest [Q5] vs lowest quintile [Q1]; P_trend = .05), with the hazard ratios in Q2 to Q4 of 0.85, 0.86, and 0.84, suggesting a threshold association whereby all hazard ratios above the referent are similarly decreased (Table 2). Exclusion of cases diagnosed within 2 years of blood collection did not alter the results. Cases with vitamin D concentrations in all categories higher than the referent had increased survival from cancers of the prostate (in White men), lung, pancreas, and kidney; however, only for lung cancer was the association statistically significant (HR for lung cancer mortality = 0.63, 95% CI = 0.44 to 0.90 for Q5 vs Q1; P_trend = .03). Endometrial cancer survival was non-statistically significantly lower for women in the highest vs lowest 25(OH)D tertile (HR for endometrial cancer mortality = 5.20, 95% CI = 0.80 to 33.88), but this was based on only 144 cases and 12 deaths. Colorectal cancer survival was inversely associated with serum vitamin D (HR for colorectal cancer mortality = 1.48, 95% CI = 0.81 to 2.69 for Q5 vs Q1, with a statistically significant trend; P_trend = .03). Vitamin D status was not associated with survival among cases of breast, prostate (in Black men), bladder, hematopoietic, upper gastrointestinal, or ovarian cancer (Table 2).

Table 2.

Association between prediagnostic serum 25(OH)D and overall and organ-specific cancer mortality in the PLCO Trial^a

Cancer site	Season-, sex-, and laboratory and assay–specific quantiles of serum 25(OH)Db					P trend c
	Q1	Q2	Q3	Q4	Q5
All cancers (n = 4038)
Number of deaths/survivors	302/516	276/534	280/525	277/530	281/517
Survival time, median, yd	10.16	10.26	10.25	10.05	10.45
Death ratee	41.04	37.06	37.66	36.87	37.51
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.85 (0.72 to 1.01)	0.86 (0.73 to 1.02)	0.84 (0.71 to 1.00)	0.83 (0.70 to 0.98)	.05
Breast (n = 785)
Number of deaths/survivors	9/151	11/144	15/143	17/140	10/145
Survival time, median, yd	13.20	12.95	13.04	13.11	12.88
Death ratee	4.43	5.53	7.60	8.81	5.13
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	1.45 (0.58 to 3.61)	1.95 (0.82 to 4.68)	1.95 (0.81 to 4.70)	0.86 (0.31 to 2.45)	.87
Prostate
White men (n = 748)
Number of deaths/survivors	17/138	13/140	14/144	14/134	11/123
Survival time, median, yd	13.83	13.71	14.05	14.54	14.68
Death ratee	8.62	6.53	6.96	6.92	5.91
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.67 (0.31 to 1.48)	0.65 (0.30 to 1.39)	0.75 (0.34 to 1.61)	0.67 (0.29 to 1.52)	.42
Black men (n = 226)
Number of deaths/survivors	6/70	8/68	5/69	—	—
Survival time, median, yd	11.45	10.08	10.43	—	—
Death ratee	7.50	10.47	6.66	—	—
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	1.97 (0.52 to 7.50)	0.83 (0.21 to 3.30)	—	—	.81
Colorectum (n = 476)
Number of deaths/survivors	28/70	26/70	32/69	29/69	26/57
Survival time, median, yd	7.72	6.51	6.89	8.15	8.38
Death ratee	34.24	35.11	41.11	35.38	36.34
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.84 (0.47 to 1.51)	1.00 (0.55 to 1.80)	1.81 (1.00 to 3.26)	1.48 (0.81 to 2.69)	.03
Lung (n = 450)
Number of deaths/survivors	76/16	64/26	70/20	68/22	66/22
Survival time, median, yd	0.58	1.37	1.06	0.95	1.08
Death ratee	416.31	191.94	249.00	256.69	250.82
Multivariable-adjusted HR (95% CI)f	1.00 (Referent)	0.70 (0.49 to 1.01)	0.79 (0.55 to 1.13)	0.71 (0.50 to 1.01)	0.63 (0.44 to 0.90)	.03
Bladder (n = 370)
Number of deaths/survivors	12/68	14/64	15/61	12/61	15/48
Survival time, median, yd	9.11	8.92	8.67	7.20	8.28
Death ratee	15.91	19.37	24.08	20.66	27.39
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	1.06 (0.44 to 2.56)	1.14 (0.47 to 2.75)	0.84 (0.32 to 2.42)	1.17 (0.48 to 2.89)	.90
Hematopoietic (n = 292)
Number of deaths/survivors	16/34	21/30	20/40	18/40	22/51
Survival time, median, yd	10.37	6.42	8.90	11.29	9.18
Death ratee	37.52	60.45	40.02	30.59	32.02
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	2.01 (1.02 to 3.95)	1.26 (0.65 to 2.46)	1.10 (0.54 to 2.32)	1.22 (0.62 to 2.43)	.85
Pancreas (n = 233)
Number of deaths/survivors	73/6	57/6	81/10	—	—
Survival time, median, yd	0.41	0.55	0.48	—	—
Death ratee	1269.62	959.85	882.83	—	—
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.84 (0.58 to 1.21)	0.79 (0.54 to 1.16)	—	—	.29
Kidney (n = 159)
Number of deaths/survivors	23/32	14/36	17/37	—	—
Survival time, median, yd	6.78	10.22	10.27	—	—
Death ratee	57.56	31.87	32.99	—	—
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.43 (0.20 to 0.94)	0.51 (0.24 to 1.07)	—	—	.09
Endometrium (n = 144)
Number of deaths/survivors	3/47	4/40	5/45	—	—
Survival time, median, yd	13.19	13.61	13.90	—	—
Death ratee	4.87	7.13	7.78	—	—
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	1.39 (0.20 to 9.75)	5.20 (0.80 to 33.88)	—	—	.09
Upper gastrointestinal tract (n = 93)
Number of deaths/survivors	23/9	19/13	24/5	—	—
Survival time, median, yd	1.08	1.13	1.35	—	—
Death ratee	229.71	140.73	230.69	—	—
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	1.20 (0.56 to 2.57)	1.33 (0.63 to 2.77)	—	—	.46
Ovary (n = 74)
Number of deaths/survivors	19/5	17/4	20/9	—	—
Survival time, median, yd	4.53	6.82	7.08	—	—
Death ratee	149.01	110.20	92.24	—	—
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.84 (0.35 to 1.93)	0.93 (0.41 to 2.09)	—	—	.87

^aAdjusted for age at cancer diagnosis (continuous), body mass index (continuous), smoking status (never, current, former), physical activity (none or <1 h/wk, 1-2 h/wk, ≥3 h/wk), history of diabetes (yes/no), family history of cancer (yes/no), calendar year of diagnosis (continuous), cancer stage (1-4, missing), and cancer grade (1-4, missing). The overall model is additionally adjusted for cancer site (categorical). All models are also adjusted for sex and race, except for organ-specific models if all subjects are the same sex or same race. Cancer stage variable not available for pancreas, renal, upper gastrointestinal, endometrial, or hematopoietic cancers; the P value threshold for Bonferroni correction for multiple comparisons (13 tests) = .004 (.05/13). 25(OH)D = 25-hydroxyvitamin D; CI = confidence interval; HR = hazard ratio; PLCO = Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; Q = quantile.

^bCut points for the 25(OH)D concentrations (nmol/L) were based on the distribution by season, sex, and laboratory and assay and are listed in Supplementary Table 1 (available online).

^cTests for linear trend (2-sided) were conducted by coding each category either 1-5 (for quintiles) or 1-3 (for tertiles) and treating these as continuous variables.

^dUnadjusted.

^eCrude death rate per 1000 person-years.

^fAdditionally adjusted for pack-years of smoking and for cigar and pipe use.

Overall cancer survival appeared to be increased for higher vitamin D status (Q5 vs Q1) among men (HRs for overall cancer mortality: 0.75 vs 1.04 in women, with similar results when excluding sex-specific cancers), among Black individuals (HR = 0.52 vs 0.72 in American Indian, Asian, Hispanic, Pacific Islander, and White individuals combined), and in former and current smokers (HRs = 0.76 and 0.64, respectively, vs HR = 1.09 in never-smokers); however, none of these interactions were statistically significant (Supplementary Table 4, available online).

Greater overall cancer survival among cases with higher serum 25(OH)D was limited to those with Gc1-2 or Gc2-2 isoforms (HRs for overall cancer mortality for Q5 vs Q1: HR = 0.60, 95% CI = 0.44 to 0.82; P_trend = .002; and HR = 0.38, 95% CI = 0.14 to 1.05; P_trend = .02, respectively) as compared with Gc1-1 (HR = 0.94, 95% CI = 0.72 to 1.23; P_trend = .85, P_interaction < .001) (Table 3). The hazard ratio for highest vs lowest 25(OH)D among those with either Gc1-2 or Gc2-2 was 0.60 (95% CI = 0.45 to 0.79), whereas among the Gc1s-1s, Gc1s-1f, and Gc1f-1f isoforms, 25(OH)D was not associated with overall cancer survival (HR for Q5 vs Q1 = 0.92, 95% CI = 0.63 to 1.35; HR Q5 vs Q1 = 0.96, 95% CI = 0.58 to 1.60 and HR Q5 vs Q1 = 1.59, 95% CI = 0.64 to 3.95, respectively; Supplementary Table 5, available online). A similar pattern by isoform was observed for lung cancer survival (HR = 0.30, 95% CI = 0.16 to 0.56; P_trend < .001 for either Gc1-2 or Gc2-2) but not prostate cancer survival (Supplementary Table 6, available online). Colorectal cancer survival was non-statistically significantly reduced for highest vs lowest 25(OH)D for participants with either Gc1-2 or Gc2-2, and breast cancer models were unstable. The Gc2 isoform itself was associated with lower 25(OH)D and DBP concentrations (medians = 59.5, 56.4, and 53.9 nmol/L 25[OH]D and 4672, 4207, and 3534 nmol/L DBP for Gc1-1, Gc1-2, and Gc2-2, respectively) but was not related to overall cancer survival (HR = 1.10, 95% CI = 0.87 to 1.37).

Table 3.

Association between prediagnostic serum 25(OH)D and overall cancer mortality in the PLCO Trial, stratified by GC isoform^a

Gc Subgroups	Season-, sex-, and laboratory and assay-specific quantiles of serum 25(OH)Db					P trend c
	Q1	Q2	Q3	Q4	Q5
Gc1-1 (n = 1924)
Number of deaths/survivors	115/215	121/251	119/248	132/274	142/307
Survival time, median, yd	10.12	10.28	10.85	9.64	10.97
Death ratee	38.21	34.96	33.51	35.22	31.48
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.98 (0.75 to 1.27)	0.92 (0.70 to 1.21)	1.03 (0.79 to 1.34)	0.94 (0.72 to 1.23)	.85
Gc1-2 (n = 1312)
Number of deaths/survivors	108/203	84/178	87/175	95/168	80/134
Survival time, median, yd	10.15	11.18	10.61	10.44	9.99
Death ratee	37.91	32.96	34.70	37.99	42.69
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.72 (0.53 to 0.97)	0.77 (0.57 to 1.03)	0.66 (0.49 to 0.89)	0.60 (0.44 to 0.82)	.002
Gc2-2 (n = 275)
Number of deaths/survivors	28/53	20/35	26/46	9/34	6/18
Survival time, median, yd	11.83	11.27	10.24	13.01	12.70
Death ratee	36.15	37.70	40.22	18.55	23.27
Multivariable-adjusted HR (95% CI)	1.00 (Referent)	0.83 (0.41 to 1.65)	0.68 (0.35 to 1.34)	0.36 (0.14 to 0.91)	0.38 (0.14 to 1.05)	.02

^aAdjusted for age at cancer diagnosis (continuous), body mass index (continuous), smoking status (never, current, former), physical activity (none or <1 h/wk, 1-2 h/wk, ≥3 h/wk), history of diabetes (yes/no), family history of cancer (yes/no), calendar year of diagnosis (continuous), cancer stage (1-4, missing), cancer grade (1-4, missing), cancer site (categorical), sex, and race. P_interaction < .001. 25(OH)D = 25-hydroxyvitamin D; CI = confidence interval; HR = hazard ratio; PLCO = Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial; Q = quantile.

^bCut points for the 25(OH)D concentrations (nmol/L) were based on the distribution by season, sex, and laboratory and assay and are listed in Supplementary Table 1 (available online).

^cTests for linear trend (2-sided) were conducted by coding each quintile category as 1-5 and treating these as continuous variables.

^dUnadjusted.

^eCrude death rate per 1000 person-years.

Discussion

In this analysis, using all available data from nested case-control studies of vitamin D in the PLCO Trial, higher prediagnostic serum 25(OH)D concentration was associated with improved overall and lung cancer survival. Furthermore, these associations were limited to individuals with the Gc2 DBP isoform.

Most meta-analyses of studies where blood was collected prior to cancer diagnoses indicate greater cancer survival in cases with higher circulating vitamin D concentrations (1-4), with the exception of one (10). The most recent meta-analysis showed lower cancer mortality for higher 25(OH)D (relative risk = 0.81, 95% CI = 0.71 to 0.93 for highest vs lowest category), but a statistically significant reduction was only observed in women (relative risk = 0.72, 95% CI = 0.52 to 0.98, vs 0.90, 95% CI = 0.73 to 1.12 in men) (4), which is the opposite of what we observed here. Other recent studies evaluating prediagnostic circulating 25(OH)D concentrations not included in these meta-analyses also support a beneficial association (5-9), with one exception (11).

We found statistically significantly improved lung cancer survival with higher 25(OH)D concentrations, which is supported by several studies (8,9,41) but not all (5-7,42-44). (Unless otherwise specified, all studies reviewed in this discussion are prospective and use prediagnostic vitamin D concentrations.) We also found a statistically significant trend of poorer colorectal cancer survival for cases with higher vitamin D status. This finding conflicts with the vast majority of colorectal cancer survival studies, which report greater survival (5,7-9,18,45,46) or no associations (41,42).

Prostate, kidney, and pancreatic cancer survival was non-statistically significantly increased among cases with higher 25(OH)D concentrations. Previous studies showed statistically significant greater prostate and kidney cancer survival and non-statistically significant greater pancreatic cancer survival for higher prediagnostic 25(OH)D (42,47-50); however, other studies for these sites were null (7,9,43,51). Although it has been hypothesized that low vitamin D status may contribute to increased cancer incidence and poorer survival in Black populations (52), we found no association with prostate cancer survival in Black men. Vitamin D status was not associated with survival for breast, bladder, hematopoietic, upper gastrointestinal, endometrial, or ovarian cancers. Results in prior studies for these cancer sites have been inconsistent [breast (5-9,43); bladder (42,53); hematopoietic (6,8,9,42,43); gastrointestinal (6,9,42,54,55); and ovarian (9)].

Three recent studies of colorectal adenoma, colorectal cancer, and colorectal cancer mortality found reduced risks of these outcomes with higher vitamin D only among individuals with the Gc2 isoform of DBP (18-20). We found a similar pattern for overall and lung cancer survival but a non-statistically significant reduced colorectal cancer survival among those with the Gc2 isoform. A retrospective study found statistically significantly reduced breast cancer risk for those with the Gc2 isoform (40). DBP carries 25(OH)D and 1,25(OH)₂D in circulation, and the Gc2 isoform has a lower affinity for both 25(OH)D and 1,25(OH)₂D (21). In addition, as shown in our data and prior studies, individuals with the Gc2 isoform have lower concentrations of DBP, 25(OH)D, and 1,25(OH)₂D compared with individuals with the Gc1f and Gc1s isoforms (21,56,57). It has therefore been proposed that vitamin D sufficiency thresholds may vary by Gc isoform (57), which may explain why only individuals with the Gc2 isoform appear to gain a cancer survival advantage by maintaining higher 25(OH)D concentrations. Alternatively, Gc macrophage-activating factor has been shown to inhibit tumor growth and to increase apoptosis in cell lines and has been used to treat metastatic breast cancer (40). Because fewer steps are required to convert the Gc2 isoform to Gc macrophage-activating factor compared with the Gc1 isoform, this has been hypothesized as an explanation for the reduced breast cancer risk among individuals with the Gc2 isoform (40).

We specifically examined the vitamin D–cancer survival association among women and nonsmokers, because our prior analysis in the Alpha-Tocopherol, Beta-Carotene Cancer Prevention Study cohort included only male smokers (42). Here, we found higher vitamin D status related to greater cancer survival in men and former or current smokers but not among women or never-smokers, a finding also supported by the UK Biobank cohort (9). By contrast, a meta-analysis and another recent study reported no differences in the vitamin D–survival association by sex (3,5), whereas two other meta-analyses showed increased survival only among women (1,4) and another reported no differences for cancer survival by smoking status (3).

The present analysis included more than 4000 cases of 11 cancer sites diagnosed prospectively within the PLCO Trial. Blood samples were collected at baseline prior to cancer diagnoses, and deaths were ascertained during up to 22 years of follow-up (median = 15.6 years). This minimized the potential for either cancer itself (eg, reverse causality) or cancer treatment to alter serum vitamin D concentrations, a limitation in studies that measure vitamin D at cancer diagnosis. We also conducted a sensitivity analysis that excluded cases diagnosed within 2 years of blood collection and observed no changes to the survival estimates. Another strength was our ability to adjust for several major potential confounding factors (eg, smoking), although the possibility of residual confounding from other exposures can never be completely eliminated. Even though all subjects included in the current analysis had been randomly assigned to the screening arm of PLCO (58), the screening tests for prostate, lung, and ovarian cancers had no material impact on cancer mortality for those sites (59-61) and therefore should not have influenced our findings. By contrast, the trial’s flexible sigmoidoscopic screening reduced both colorectal cancer incidence and distal colorectal cancer mortality (62), which could have impacted our survival findings for this malignancy, if, for example, the screen-detected colorectal cancers had different pathophysiologic profiles as compared with those that would develop in usual-screening or unscreened populations. Notably, however, excluding the screen-detected colorectal cancer cases from our analysis did not materially alter our findings. At the same time, such prospective biomarker analyses within PLCO have the theoretical advantage of not being confounded by colorectal cancer screening, as could occur in other studies where participants with high vitamin D concentrations also practice healthy behaviors such as routine cancer screenings, which could result in earlier detection and increased survival.

Our study also has some limitations. Serum 25(OH)D concentrations were assayed at several laboratories using different methods for the cancer-specific nested case-control studies. To address this limitation, we created quantiles separately by laboratory and assay method. In addition, because of the known seasonal variation in 25(OH)D concentrations, the quantiles are also seasonspecific. A limitation of this and most other observational studies is that the biomarker of interest was measured only in blood samples collected at baseline; however, 25(OH)D has been shown to be stable in blood stored up to 30 years (63,64) and to be moderately associated in blood samples taken up to 14 years apart (65-68). Although we included one nested case-control study of prostate cancer in Black men, we had small numbers of racial and ethnic minority participants across the other cancers with which to examine differences by race. This is relevant given the well-documented Black–White differences in 25(OH)D concentrations, DBP isoforms, and cancer outcomes (69,70). We were also hampered by small case numbers and therefore had limited power for several organ -specific models. Furthermore, the statically significant associations we observed could be because of multiple comparisons bias. PLCO has cancer treatment data for only prostate, lung, colorectal, and ovarian cancers. We therefore adjusted all models for calendar year of diagnosis to account for changes in cancer therapeutics over time. In addition, although we present models adjusted for and stratified by stage and grade, we lacked stage or grade data for 13.9% of the cases.

In conclusion, we observed statistically significantly improved overall and lung cancer survival among cases in the PLCO Trial with higher prediagnostic vitamin D status. Our study suggests that the benefits of higher vitamin D status in cancer survival may vary by organ site and that the improved overall cancer survival was limited to individuals with the Gc2 DBP isoform. These individuals may need higher 25(OH)D concentrations to maintain adequate vitamin D status and related beneficial biological activity. Future investigations should examine these associations in larger and more diverse populations with Gc2 isoform data.

Funding

The PLCO Trial and cohort were supported by the Intramural Research Program of the Division of Cancer Epidemiology and Genetics, and contracts from the Division of Cancer Prevention, National Cancer Institute, National Institutes of Health, Department of Health and Human Services.

Notes

Role of the funders: The design of the present study and analysis, interpretation of the data, writing of the manuscript, and decision to submit the manuscript for publication rest solely with the individual study investigators.

Disclosures: The authors report no conflicts of interest.

Author contributions: Conceptualization: SJW, DA; Statistical analysis: SJW, KY; Writing—original draft: SJW, DA; Writing—review and editing: all authors.

Acknowledgments: The authors thank the PLCO participants, the NCI study management team and staff at Information Management Services, Inc, Westat, Inc, Frederick National Laboratory for Cancer Research, Leidos Biomedical Research, Inc, and American Type Culture Collection.

Data Availability

The data underlying this specific article are available to the general scientific community upon reasonable request. Proposals should be submitted through this link: https://cdas.cancer.gov/plco/.