Vitamin D Supplementation and Fractures in Adults: A Systematic Umbrella Review of Meta-Analyses of Controlled Trials.

CONTEXT: The growing number of systematic reviews/meta-analyses (SR/MAs) on vitamin D (± calcium) for fracture prevention has led to contradictory guidelines.
OBJECTIVE: This umbrella review aims to assess the quality and explore the reasons for the discrepancy of SR/MAs of trials on vitamin D supplementation for fracture risk reduction in adults.
METHODS: We searched 4 databases (2010-2020), Epistemonikos, and references of included SRs/MAs, and we contacted experts in the field. We used A MeaSurement Tool to Assess systematic Reviews 2 (AMSTAR-2) for quality assessment. We compared results and investigated reasons for discordance using matrices and subgroup analyses (PROSPERO registration: CRD42019129540). We included 13 SR/MAs on vitamin D and calcium (Ca/D) and 19 SR/MAs on vitamin D alone, compared to placebo/control.
RESULTS: Only 2 from 10 SRs/MAs on Ca/D were of moderate quality. Ca/D reduced the risk of hip fractures in 8 of 12 SRs/MAs (relative risk [RR] 0.61-0.84), and any fractures in 7 of 11 SR/MAs (RR 0.74-0.95). No fracture risk reduction was noted in SRs/MAs exclusively evaluating community-dwelling individuals or in those on vitamin D alone compared to placebo/control. Discordance in results between SRs/MAs stems from inclusion of different trials, related to search periods and eligibility criteria, and varying methodology (using intention to treat, per-protocol, or complete case analysis from individual trials).
CONCLUSION: Ca/D reduces the risk of hip and any fractures, possibly driven by findings from institutionalized individuals. Individual participant data meta-analyses of patients on Ca/D with sufficient follow-up periods, and subgroup analyses, would unravel determinants for a beneficial response to supplementation.

Vitamin D plays a critical role in musculoskeletal health through its effects on mineral homeostasis and bone metabolism (1, 2). Vitamin D deficiency is common among older individuals because of reduced cholecalciferol synthesis in the skin, reduced intestinal calcium (Ca) absorption, and changes in lifestyle favoring lower exposure to ultraviolet radiation (3). This deficiency may contribute to the observed steep rise in the risk of fractures with older age, particularly at the hip (4). This is of particular importance given that the estimated 1-year mortality following hip fracture is around 30% (5).

The benefit of vitamin D supplementation on fracture prevention has been extensively assessed, with an exponential rise in the number of systematic reviews/meta-analyses (SRs/MAs) reporting discordant conclusions (6). A recent review in 2017 identified more than 40 international vitamin D guidelines with highly variable recommendations (7). There is still conflicting evidence with regards to the extent of vitamin D’s benefit in fracture prevention, the target population likely to benefit the most, the desirable serum 25-hydroxyvitamin D (25[OH]D) concentration, the optimal vitamin D dose, and the need for coadministration of Ca (7-9).

We therefore conducted an umbrella review of SRs/MAs of vitamin D supplementation randomized clinical trials (RCTs) evaluating fracture risk reduction in adults, to assess the quality of each SR/MA, explore similarities and differences between them, investigate the reasons for any discrepancy, and formulate a reliable conclusion on the topic. Such an approach would bring clarity to much confusion, paving the path for the formulation of informed population-tailored conclusions regarding the benefit of vitamin D supplementation in preventing fractures.

Methods and Literature Search

We followed the guidance provided by Pollock et al (10) to develop the protocol for this umbrella review. We registered the protocol on PROSPERO (CRD42019129540) (11). We included SRs/MAs of RCTs of adults aged 18 years and older evaluating the risk of hip and any fracture, with vitamin D supplementation, alone or in combination with Ca, compared with placebo/control.

We searched MEDLINE, PubMed, Embase, Cochrane, and Epistemonikos databases, without language restriction, from January 1, 2010 until October 23, 2020, with the help of a medical librarian (A.F.), to cover the reviews published following vitamin D guidelines issued by 2 major societies, the Institute of Medicine in 2009 (9) and the Endocrine Society in 2011 (8).We used Medical Subject Headings terms and keywords relevant to vitamin D, fractures, meta-analysis, systematic review, and randomized controlled trials. We reviewed the citations of included SRs/MAs and of narrative reviews on the topic, and contacted experts in the field (Supplementary data 1) (12).

Pairs of reviewers (C.G. and F.K., D.B. and S.A., Y.J. and M.A., A.B. and S.A.) completed title/abstract and full-text screening, and data abstraction in duplicate and independently, after a calibration exercise. Disagreement was resolved through discussion with a content expert (G.E.H.F. or M.C.).

We collected information on study population, intervention(s), comparator(s), and outcome(s), in addition to the methodology of each SR/MA. We constructed, for each of hip and any fractures, figures displaying the effect size estimates derived from each SR/MA, 95% CIs, and degree of heterogeneity using I2. We constructed matrices to compare and contrast the SR/MAs’ respective results, including RCTs and respective data, and quality ratings. We calculated the “corrected covered area” (CCA) for each of hip and any fracture, as suggested by Pieper et al (13), using the following formula: CCA = (N – r)/[rc – r], where N is the total number of included publications (including double counting) as displayed in the matrix for all the SR/MA, r is the number of RCTs included at least once in the matrix, and c is the number of SR/MAs. We also collected data on subgroup analyses by baseline serum 25(OH)D concentration (prespecified), and on other subgroups as available in each SR/MA.

We assessed the quality of SR/MAs independently in triplicate using A MeaSurement Tool to Assess systematic Reviews 2 (AMSTAR-2) (14).

Results

The search strategy retrieved 14 357 citations. We reviewed 1375 full-text papers. We identified 81 SRs/MAs of RCTs on vitamin D supplementation, 25 of which assessed fractures (n = 21) (15-35) or fall-related fractures (n = 4) (36-39), all in the English language. We included SRs/MAs on vitamin D with concomitant Ca supplementation (Ca/D) (13 SR/MAs) (15-17, 19, 22-24, 28-31, 37, 39), or vitamin D alone (19 SR/MAs) (15, 16, 19-21, 23-28, 30-32, 34-37, 39), compared to placebo/control; 9 SR/MAs included other comparisons (15, 17, 18, 20, 21, 33, 37-39) that were beyond the focus of this manuscript (Fig. 1).

Figure 1.

Flow diagram of the included studies. List of excluded articles is available on request. These will be included in separate manuscripts.

Results on Calcium and Vitamin D Supplementation vs Placebo/Control

We identified 12 SRs/MAs assessing hip fracture risk and 11 assessing any fracture risk, with Ca/D supplementation compared with placebo/control, including one individual participant data meta-analysis (IPD-MA) (23). The mean age of participants ranged between 62 and 85 years. The vitamin D dose was 400 to 800 IU/day, and 6 SR/MAs included 1 trial providing a high dose of 300 000 IU once (15, 16, 19, 24, 31, 37). The Ca dose was 500 to 1200 mg/day. The included trials extended from 1 to 7 years. Five SRs/MAs reported on baseline 25(OH)D concentration (20.9-83.8 nmol/L) (15, 19, 24, 31, 37), and the achieved concentration (15-112.3 nmol/L) (19) (Table 1).

Table 1.

Characteristics of populations and interventions in the included systematic reviews/meta-analyses on vitamin D and concomitant calcium supplementation compared to placebo/control

Characteristics of populations and interventions	No. (%) or range
Mean age, y	62-85.2
Sex
• Men and women	9 (69)
• Women only	2 (15)
• Not reported	2 (15)
Residency status
• Community-dwelling only	3 (23)
• Institutionalized only	0 (0)
• Both	8 (62)
• Not reported	2 (15)
Mean serum 25(OH)D concentration at baseline, nmol/L	11.9-84
• Reported	5 (38)
• Not reported	8 (62)
Mean serum 25(OH)D concentration at follow-up, nmol/L	15.0-112
• Reporteda	1 (8)
• Not reported	12 (92)
Vitamin D dose frequency
• Daily, IU/d	400-1600
• Once, IU	300 000
Calcium dose, mg/d	500-1200
• Daily
Vitamin D route of administration
• Oral supplementation only	1 (8)
• Oral and parenteral supplementation	4 (31)
• Administration mode not mentioned	8 (62)
Individual trial duration
• SR/MA including ≥ 1 trial < 12 mo	0 (0)
• SR/MA including all trials ≥ 12 mo	9 (69)
• Not reported	4 (31)
Fracture siteb
• Hip	12
• Total/Any/Other	11
• Vertebral	6
• Nonvertebral	5
SR/MAs with trials on individuals at high risk of fracturec	9

Abbreviations: 25(OH)D, 25-hydroxyvitamin D; IU, international units; SR/MA, systematic review and meta-analysis.

Yao et al (30) reported change rather than follow up 25-hydroxyvitamin D concentration.

Some MAs may include more than one fracture site.

Previous fracture or osteopenia/osteoporosis.

Hip fractures

Most SRs/MAs had the same direction, but were discrepant both in magnitude of effect and statistical significance. Eight of 12 SRs/MAs reported a significant reduction in hip fracture risk when combining institutionalized and community-dwelling individuals (15, 16, 19, 22, 28-30, 39) (Fig. 2A). Only one Cochrane SR/MA, by Avenell et al (15), was of moderate quality, and was the only one evaluating the quality of evidence, reported as high, with a 16% relative risk reduction (RRR) in hip fracture, with Ca/D supplementation compared with placebo/control. The other 7 SR/MAs, of critically low quality, showed a 16% to 39% RRR in hip fracture (16, 19, 22, 28-30, 39). Two other SR/MAs (17, 23), including 1 IPD-MA (23), and the 2 SRs/MAs that exclusively evaluated RCTs in community-dwelling individuals (24, 31), failed to show any significant fracture reduction. They were of low (31) to critically low quality (17, 23, 24) (Supplementary data 2) (12).

Figure 2.

Effect size estimates and 95% CI for A, hip, and B, any fracture risk with vitamin D and calcium supplementation vs placebo/control. Quality assessment using A MeaSurement Tool to Assess systematic Reviews 2 (AMSTAR-2) tool described between parenthesis: A, moderate quality; B, low quality; and C, critically low quality. HR, hazard ratio; OR, odds ratio; RR, risk ratio. Meta-analyses that included institutionalized trials. Network meta-analyses. Not available.

Nine SRs/MAs on hip fracture were included in the matrix because they provided the needed information (Table 2). The population consisted of older women and/or men, with (40-44) or without a history of fracture (45-49). The Women’s Health Initiative (WHI) trial (N = 36 282, Ca/D dose per day 1000 mg/400 IU) (48) and the RECORD trial (N = 5292, Ca/D dose per day 1000 mg/800 IU) (41) contributed with the largest weights for the findings in the community. The 2 trials by Chapuy et al (45, 46) were the only trials on institutionalized individuals included in all SRs/MAs (N = 3847, Ca/D dose per day 1200 mg/800 IU) (see Table 2).

Table 2.

Comparison of individual trials included in traditional systematic reviews/meta-analyses and reasons for exclusion of trials by some systematic reviews/meta-analyses investigating hip fracture risk with calcium and vitamin D supplementation

	Bergman et al (17) 2010	Avenell et al (15) 2014	Bolland et al (19) 2014	Weaver et al (29) 2016	Zhao et al (31) 2017	Barrionuevo et al (16) 2019	Hu et al (24) 2019	Yao et al (30) 2019	Eleni and Panagiotis (22) 2020
Pooled RR	0.72 (0.32-1.40)	0.84 (0.74-0.96)	0.84 (0.74-0.96)	0.61 (0.46-0.82)		0.81 (0.71-0.93)		0.84 (0.72-0.97)	0.61 (0.40-0.92)
Institutionalized RR		0.75 (0.62-0.92)c						0.69 (0.53-0.90)d
Community RR		0.91 (0.77-1.09)			1.09 (0.85-1.39)		1.15 (0.81-1.62)e	0.92 (0.77-1.10)
AMSTAR Quality Assessment	Critically low	Moderate	Critically low	Critically low	Low	Critically low	Critically low	Critically low	Critically low
Included trials
Chapuy et al (45) 1992	NA 80/1387 110/1403 NA	38.24% 137/1634 178/1636 0.77 (0.62-0.95)	38% 137/1634 178/1636 NA	NA 21/877 37/888 0.57 (0.34-0.97)	Excluded institutionalized trials	42.77% NA NA 0.77 (0.62-0.95)	Excluded institutionalized trials	25.5% 80/1634 110/1636 0.72 (0.53-0.96)	15.2% 21/1387 37/1403 0.57 (0.34-0.98)
Chapuy et al (46) 2002	NA 27/393 21/190 NA	6.02% 27/389 21/194 0.64 (0.37-1.10)	6% 27/389 21/194 NA	NA 27/393 21/190 0.62 (0.36-1.07)	Excluded institutionalized trials	6.50% NA NA 0.64 (0.37-1.10)	Excluded institutionalized trials	5.5% 27/393 21/190 0.58 (0.31-1.08)	15.0% 27/393 21/190 0.62 (0.36-1.07)
Dawson-Hughes et al (47) 1997	Excluded trials including men	0.31% 0/187 1/202 0.36 (0.01-8.78)	0.2% 0/187 1/202 NA	NA 0/170 1/148 0.29 (0.01-7.08)	0.6% 0/187 1/202 0.36 (0.01-8.78)	Excluded trials including men	NA 0/187 1/202 NA	Excluded trials with < 500 participants	1.5% 0/187 1/202 0.36 (0.01-8.78)
Larsen et al (43) 2004	NA	“No treatment group received vitamin D & calcium alone”f	NA	NA	“no treatment group received vitamin D and calcium alone”	NA	NA	“Not randomized (cluster randomized factorial design)”	18.6% 87/4957 114/2116 0.33 (0.25-0.43)
Avenell et al (40) 2004	Excluded trials including men	0.21% 1/35 1/35 1.0 (0.07-15.36)	0.2% 1/35 1/35 NA	“Doesn’t qualify as an RCT of Calcium and vitamin D”	0.8% 1/35 1/35 1.0 (0.07-15.36)	Excluded trials including men	NA 1/35 1/35 NA	Excluded trials with < 500 participants	“It assessed the open trial design and not explicitly the combination of Ca & vitamin D”
Harwood et al (NoNOF) (42) 2004	Excluded trials on post–hip fracture	0.29% 1/75 1/37 0.49 (0.03-7.67)	0.2% 1/75 1/37 NA	No data on hip fracture per SR/MA definition	0.8% 1/75 1/37 0.49 (0.03-7.67)	0.26% NA NA 0.49 (0.03-7.67)	NA 1/75 1/37 NA	Excluded trials with < 500 participants	NA
Grant et al (RECORD) (41) 2005	Included in sensitivity analysis	8.73% 46/1306 41/1332 1.14 (0.76-1.73)	10% 46/1306 41/1332 NA	No data on hip fracture per SR/MA definition	35.2% 46/1306 41/1332 1.14 (0.76-1.73)	Excluded trials including men	NA 46/1306 41/1332 NA	12% 46/1306 41/1332 1.15 (0.75-1.76)	NA
Porthouse et al (44) 2005	NA 8/1321 17/1993 NA	2.91% 8/1321 17/1993 0.71 (0.31-1.64)	2% 8/1321 17/1993 NA	NA 5/607 2/602 2.48 (0.48-12.78)g	8.6% 8/1321 17/1993 0.71 (0.31-1.64)	2.74% NA NA 0.71 (0.31-1.64)	NA 8/1321 17/1993 NA	3.4% 8/1321 17/1993 0.72 (0.32-1.61)	11.0% 8/1321 17/1993 0.71 (0.31-1.64)
Jackson et al (WHI) (48) 2006	Excluded interventions incorporating hormone therapy	42.86% 175/18176 199/18106 0.88 (0.72-1.07)	42% 175/18176 199/18106 NA	Excluded because used WHI post hoc analysis by Prentice et al (50) instead	52% 70/4015 61/3957 1.13 (0.8-1.59)	47.06% NA NA 0.88 (0.72-1.07)	NA 70/4015 61/3957 NA	52.7% 175/18176 199/18106 0.87 (0.71-1.07)	19.4% 175/18176 199/18106 0.88 (0.72-1.07)
Salovaara et al (OSTPRE-FPS) (49) 2010	Beyond search period	0.43% 4/1586 2/1609 2.03 (0.37-11.06)	0.6% 4/1718 2/1714 NA	NA 4/1586 2/1609 2.03 (0.37-11.06)	2.1% 4/1718 2/1714 2.0 (0.37-10.88)	0.67% NA NA 2.03 (0.37-11.06)	NA 4/1718 2/1714 NA	0.9% 4/1586 2/1609 1.98 (0.40-9.81)	4.5% 4/1586 2/1609 2.03 (0.37-11.06)
Prentice et al (50) 2013	Beyond search period	NA	NA	NA 19/7530 35/7406 0.53 (0.31-0.93)	NA	Post hoc analysis of WHI trial	NA	NA	14.8% 19/7530 35/7801 0.56 (0.32-0.98)

The trials by Chapuy et al (45, 46) are the only ones conducted in institutionalized populations. All the other trials are conducted in community-dwelling populations. Individual rows within cells are filled with the following: Percentage is weight of each trial taken from the forest plot of the corresponding SR/MA, number of fractures/total number of participants in the treatment group, number of fractures/total number of participants in control group, and effect size estimate of the trial taken from the forest plot of the corresponding SR/MA.

Abbreviations: Ca, calcium; D, vitamin D; MA, meta-analysis; NA, not available; RCT, randomized clinical trial; RR, relative risk; SR, systematic review; WHI, Women’s Health Initiative.

MAs not included in the matrix: DIPART, 2010: individual participant data without details about individual RCTs (23); Murad et al, 2012: Network MA without details about Ca/D vs placebo (28); Tricco et al, 2017: Network MA without details about individual RCTs (39).

Reasons for exclusion of some trials by SRs/MAs either as exactly reported or derived from inclusion/exclusion criteria.

“The test for subgroup differences was not significant (P = .15).”

“P for heterogeneity: .07”; P value for interaction term was not provided.

This RR is for high D (≥ 800 IU/day) + high Ca (≥ 800 mg/day) vs placebo. RR for low D (< 800 IU/day) + high Ca (≥ 800 mg/day) is 1.13 (0.78-1.63). RR for low D (< 800 IU/day) + low Ca (< 800 mg/day) is 0.36 (0.01-8.81).

Avenell excluded Larsen because participants in each of the 3 treatment clusters received 1 or more cointerventions designed to reduce falls (medication review, environmental hazard, and health assessment, and osteoporosis/fall prevention leaflets) but the control group received no intervention. No treatment group received D and Ca alone.

CI for “equally allocated” subgroup.

There was a substantial overlap in the RCTs conducted in community-dwelling individuals, included in the SRs/MAs by Avenell and colleagues and Bolland et al (15, 19), Zhao et al and Hu et al (24, 31), Yao et al and Barrionuevo and colleagues (16, 30), thus explaining their very similar effect size estimates. Conversely, the overlap varied across other SRs/MAs. One study (Porthouse et al 2005) (44) was included in all SRs/MAs, whereas another one (Larsen et al 2004) (43) was included only in the most recent one (22). This is explained by differences in either the search period or the eligibility criteria. Some SRs/MAs included RCTs with a relatively large sample size (> 500 [30] or > 1000 participants [23]), or exclusively from the community (23, 24, 31), or exclusively women (16, 17), or excluded specific trials or trial arms if the intervention was a single high vitamin D dose (22, 27), Ca only, vitamin D only (15, 19, 24, 30, 31), or hormone replacement therapy (17) (Supplementary data 3) (12). One SR/MA included duplicate data from the WHI trial (22) (primary [48] and post hoc analyses [50]). The CCA for SRs/MAs on hip fracture was 57%.

For the same RCTs, the derived effect size estimates varied across SRs/MAs. For instance, for Chapuy 1992 (45), the sample size in the intervention and the control varied, 877 to 1634, and 888 to 1636, respectively. Similarly, the number of hip fractures considered in the analysis ranged between 21 and 137 in the intervention arm, and 37 to 178 in the control arm. Therefore, the effect size was also variable at 0.57 to 0.77 (see Table 2). Such findings are related to the data abstraction methods, targeting intention to treat (17, 22), per-protocol analyses (29), complete case analyses (15, 16, 19, 30), or excluding deaths (15, 16, 22, 29, 30). Furthermore, the methods used for pooling data were also heterogeneous; while the majority used a traditional meta-analysis, few conducted a network (16, 24, 27, 28, 39) or an IPD-MA (23) (see Supplementary data 3) (12).

Subgroup Analyses

Residency status as a predictor of response was scrutinized in 2 SRs/MAs. Although the CI of the effect size estimate in the subgroup of institutionalized individuals in the Cochrane SR/MA (15) did not cross 1.0, subgroup effect by residency was not significant, with a P interaction of .15 (see Table 2). Similarly, another SR/MA reported no significant heterogeneity across residence subgroups (15, 30), without providing an interaction P value (see Table 2). The Cochrane SR/MA reported an overall 16% RRR in hip fractures (15). The absolute risk reduction therefore varied depending on the baseline risk. It was 1 of 1000 (0/1000-2/1000) for a lower-risk population derived from the community (baseline hip fracture risk of 8/1000), while it was 9/1000 (2/1000-14/1000) for a higher-risk population derived from institutional residence (baseline risk of 54/1000) (15). Age 80 years and older and a baseline serum 25(OH)D concentration of 50 nmol/L or less approached significance, as modifiers. Conversely, subgroup analyses based on sex (31), osteoporotic fracture history (15, 30), and vitamin D dose (23, 31) did not (Table 3).

Table 3.

Subgroup analyses of vitamin D with calcium vs placebo on risk of hip and any fracture

	Hip fracture	Any fracture
By age, y	Yao et al, 2019 (30): < 80: 0.92 (0.77-1.10) ≥ 80: 0.69 (0.53-0.90) P (interaction): NA P (heterogeneity): .07	Yao et al, 2019 (30): < 80: 0.96 (0.90-1.02) ≥ 80: 0.76 (0.62-0.92) P (interaction): NA P (heterogeneity): .02
By sex	Zhao et al, 2017 (31): Women only: 1.07 (0.79-1.46) Both: 1.12 (0.75-1.68) P (interaction): .86	Zhao et al, 2017 (31): Women only: 0.88 (0.71-1.08) Both: 0.83 (0.48-1.42) P (interaction): .84
By history of osteoporotic fracture	Zhao et al, 2017 (31): Yes: 1.12 (0.75-1.68) Othera: 1.07 (0.79-1.46) P (interaction): .86	Zhao et al, 2017 (31): Yes: 0.92 (0.76-1.11) Othera: 0.88 (0.71-1.09) P (interaction): .78
	Avenell et al, 2014 (15): Yes: 1.02 (0.71, 1.47) No: 0.82 (0.71, 0.94) P (interaction): .26	Avenell et al, 2014 (15): Yes: 0.93 (0.79-1.10) No: 0.95 (0.90-1.00) P (interaction): .84
By baseline serum 25(OH)D concentration, nmol/L	Zhao et al, 2017 (31): ≥50: 0.36 (0.01-8.78) < 50: 1.14 (0.88-1.48) P (interaction): .48	Zhao et al, 2017 (31): ≥50: 1.87 (0.32-11.02) < 50: 0.90 (0.76-1.05) P (interaction): .42
	Bolland et al, 2014 (19): ≥ 50: 0.36 (0.02-8.79) < 50: 0.84 (0.74-0.97) P (interaction): .82	Bolland et al, 2014 (19): ≥ 50: 1.50 (1.02-2.19) < 50: 0.75 (0.43-1.31) P (interaction): .24
By D dose	Zhao et al (with Ca), 2017 (31): ≥ 1 g and ≥ 800 IU: 1.06 (0.74-1.51) Another dose: 1.12 (0.80-1.57) P (interaction): .83	Zhao et al (with Ca), 2017 (31): ≥ 1g and ≥ 800 IU: 0.90 (0.78-1.04) Another dose: 1.15 (0.28-4.74) P (interaction): .74
	DIPART, 2010 (23): 10 μg: 0.74 (0.60-0.91) 20 μg: 1.30 (0.88-1.91) P (interaction): NA	DIPART, 2010 (23): 10 μg: 0.91 (0.85-0.99) 20 μg: 0.95 (0.80, 1.14) P (interaction): NA
Other subgroup analyses	Yao et al, 2019 (30): region, open-label, Ca dose (1000 mg-1200 mg), treatment difference in 25(OH)D (< 50 nmol/L-≥ 50 nmol/L) Bolland et al, 2014 (19): achieved 25(OH)D (nmol/L), duration of trial Zhao et al, 2017 (31): Ca dose (< 1 g-≥ 1 g), dietary Ca intake (< 900 mg-≥ 900 mg), dose frequency Weaver et al, 2016 (29): use of personal supplements, adherence to assigned study pills in WHI

Units listed as reported; to convert from nmol/L to ng/mL, divide by 2.496.

Abbreviations: 25(OH)D, 25-hydroxyvitamin D; C, community-dwelling individuals; Ca, calcium; D, vitamin D; I, institutionalized; NA, not available; WHI, Women’s Health Initiative.

Other includes no fracture, partial fracture, or missing fracture data.

Any Fracture

Eleven SRs/MAs assessed the benefit of Ca/D on any fractures (15, 17, 19, 22-24, 29-31, 37, 39). They had the same direction but were discrepant both in magnitude of effect and statistical significance. Seven SRs/MAs reported a significant reduction in any fracture risk in analyses combining institutionalized and community-dwelling individuals (15, 19, 22, 23, 29, 30, 39) (Fig. 2B). Only 2 SRs/MAs, by the Cochrane Group, were of moderate quality (15, 37). The first on fall-related fracture included exclusively RCTs from the community and showed no effect (37). The second combined institutionalized and community-dwelling individuals and was the only one evaluating the quality of the evidence. It reported a high quality of evidence for a 5% RRR in any fractures (15). The remaining 6 SRs/MAs, of critically low quality, reported an RRR ranging between 6% and 29% (19, 22, 23, 29, 30, 39), one of which was a network MA (39) and one an IPD-MA (23). Three other SRs/MAs on community-dwelling individuals were of critically low quality and did not show evidence of fracture reduction (17, 24, 31).

Ten SRs/MAs were included in the matrix (Table 4). The population consisted mostly of older women and/or men with (40-44, 49, 51) or without (45-48, 52) a history of fracture. The RCTs with the largest weights were the community-based WHI and RECORD trials (41, 48). The 2 trials by Chapuy et al (45, 46) were the only ones on institutionalized individuals included in SRs/MAs. The overlap of included RCTs from the community was variable. As reported for the hip fracture, the variability in the inclusion of RCTs and the discrepancy in their individual effect size estimates across SR/MAs are explained by the difference in the search period, eligibility criteria, and analysis methods. The CCA for any fracture was 39%.

Table 4.

Comparison of individual trials included in traditional systematic reviews/meta-analyses and reasons for exclusion of trials by some systematic reviews/meta-analyses investigating any fracture risk with calcium and vitamin D supplementation

	Bergman et al (17) 2010	Gillespie et al (37) 2012	Avenell et al (15) 2014	Bolland et al (19) 2014	Weaver et al (29) 2016	Zhao et al (31) 2017	Hu et al (24) 2019	Yao et al (30) 2019	Eleni and Panagiotis (22) 2020
Pooled RR	0.95 (0.46-1.85)		0.95 (0.9-0.99)	0.92 (0.85-0.99)	0.86 (0.75-0.98)			0.94 (0.89-0.99)	0.74 (0.58-0.94)
Institutionalized RR			0.85 (0.74-0.98)c		0.67 (0.52-0.88)d			0.76 (0.62-0.92)e
Community RR		0.83 (0.59 and 1.16)	0.96 (0.91-1.01)		0.95 (0.85-1.06)	0.90 (0.78-1.04)	0.89 (0.74-1.06)f	0.96 (0.90-1.02)
AMSTAR Quality assessment	Critically low	Moderate	Moderate	Critically low	Critically low	Low	Critically low	Critically low	Critically low
Total fx definition	Any fx excluding hip and NV fx	Fall-related fx at any site	Any fx excluding hip, V, and NV fx	Any fx excluding hip fx	Any fx excluding hip fx	Any fx excluding hip, V, & NV fx	Any fx excluding hip and V fx	Any fx including hip fxg	Any fx excluding hip and wrist fx
Trials
Chapuy (45) 1992	NA 80/1387 115/1403 NA	Excluded institutionalized trials	10.67% 255/1634 308/1636 0.83 (0.71-0.96)	19% 255/1634 308/1636 NA	NA 66/877 97/888 0.69 (0.51-0.93)	Excluded institutionalized trials	Excluded institutionalized trials	7.0 160/1634 215/1636 0.72 (0.58-0.89)	11.1% 80/1387 110/1403 0.74 (0.56-0.97)
Chapuy (46) 2002	NA 70/393 30/190 NA	Excluded institutionalized trials	1.57% 69/389 34/194 1.01 (0.7-1.47)	4% 69/389 34/194 NA	No data on any fracture as per SR/MA definition	Excluded institutionalized trials	Excluded institutionalized trials	1.6 70/393 35/190 0.96 (0.61-1.51)	7.9% 27/393 21/190 0.62 (0.36-1.07)
Inkovaara (83) 1983	NA	Excluded trials not reporting fall outcome	“Unclear whether the data represent fractures or participants with fractures”	NA	NA	0.2% 0/46 3/42 0.13 (0.01-2.46)	NA	Excluded trials with total fx events < 10	NA
Dawson-Hughes (47) 1997	Excluded trials including men	Excluded trials not reporting fall outcome	0.87% 11/187 26/202 0.46 (0.23-0.9)	1% 11/187 26/202 NA	NA 11/170 26/148 0.37 (0.19-0.72)	No data on any fracture as per SR/MA definition	No data on any fracture as per SR/MA definition	Excluded trials with < 500 participants	6.5% 11/187 26/202 0.46 (0.23-0.90)
Larsen et al (43) 2004	NA	“The outcome is only ‘severe’ falls leading to acute hospital admission”	“no treatment group received vitamin D & calcium alone”	“It’s a cluster randomized controlled trial”	NA	“No treatment group received vitamin D and calcium alone”	NA	“Not randomized (cluster randomized factorial design)”	12.3% 217/4957 270/2116 0.34 (0.29-0.41)
Avenell et al (40) 2004	Excluded trials including men	Excluded trials not reporting on fall outcome	0.14% 2/35 4/35 0.5 (0.10-2.56)	0.3% 3/35 5/35 NA	“Doesn’t qualify as an RCT of Calcium and vitamin D”	0.8% 2/35 4/35 0.5(0.10,2.56)	NA 2/35 4/35 NA	Excluded trials with < 500 participants	“It assessed the open trial design and not explicitly the combination of Ca & vitamin D”
Harwood et al (NoNOF) (42) 2004	Excluded trials on post–hip fracture	1.5% NA NA 0.5 (0.16-1.56)	0.23% 6/75 5/37 0.59 (0.19-1.81)	0.5% 6/75 5/37 NA	NA 3/29 5/35 0.72 (0.19-2.78)	1.6% 6/75 5/37 0.59 (0.19-1.81)	NA 6/75 5/37 NA	Excluded trials with < 500 participants	2.5% 3/39 5/37 0.57(0.15,2.22)
Grant et al (RECORD) (41) 2005	Excluded trials including men	19.96% N = 5292h 1.02 (0.87-1.19)	6.14% 165/1306 179/1332 0.94 (0.77-1.15)	14% 179/1306 192/1332 NA	NA 179/1306 192/1332 0.95 (0.79-1.15)	52.6% 165/1306 179/1332 0.94 (0.77-1.15)	NA 165/1306 179/1332 NA	6.6% 179/1306 192/1332 0.94 (0.76-1.17)	12.6% 387/2649 377/2643 1.02 (0.90-1.17)
Porthouse et al (44) 2005	NA 29/1321 38/1993 NA	10.14% N = 3314h 1.01 (0.71-1.44)	2.52% 58/1321 91/1993 0.96 (0.7-1.33)	5% 58/1321 91/1993 NA	NA 24/607 22/602 1.08 (0.61-1.91)i	19.8% 58/1321 91/1993 0.96 (0.7-1.33)	NA 58/1321 91/1993 NA	2.8% 58/1321 91/1993 0.96 (0.69-1.34)	10.6% 58/1321 91/1993 0.96 (0.70-1.33)
Jackson et al (WHI) (48) 2006	Excluded interventions incorporating hormone therapy	Excluded trials not reporting fall outcome	74.96% 2102/18176 2158/18106 0.97 (0.92-1.03)	49% 1921/18176 1961/18106 NA	Used post hoc analysis by Prentice et al (50) instead	No data on any fracture per SR/MA definition	NA	78.3% 2102/18176 2158/18106 0.97 (0.91-1.03)	13.0% 2102/18176 2158/18106 0.97 (0.92-1.03)
Bischoff-Ferrari et al (84) 2006	NA	3.75% N = 3890h 0.46 (0.23-0.91)	NA	NA	NA	NA	NA	NA	NA
Bolton-Smith et al (85) 2007	NA	Excluded trials not reporting fall outcome	0.07% 2/62 2/61 0.98 (0.14-6.76)	0.2% 2/62 2/61 NA	NA	No data on any fracture per SR/MA definition	NA	Excluded trials with < 500 participants	Excluded because “confounded by the use of vitamin K1 in the therapeutic regimen”
Salovaara et al (49) 2010	Beyond search period	11.63% N = 3195h 0.89 (0.65-1.21)	2.82% 71/1586 82/1609 0.88 (0.64-1.2)	6% 78/1718 94/1714 NA	NA 78/1586 94/1609 0.84 (0.63-1.13)	23.9% 78/1718 94/1714 0.83 (0.62-1.11)	NA 78/1718 94/1714 NA	3.7% 86/1586 103/1609 0.84 (0.63-1.13)	11.0% 78/1586 94/1609 0.84 (0.63-1.13)
Prentice et al (50) 2013	Beyond search period	Beyond search period	NA	NA	NA 405/7530 458/7801 0.92 (0.80-1.04)	NA	NA	NA	12.6% 405/7530 458/7801 0.92 (0.80-1.04)
Liu et al (52) 2015	Beyond search period	Beyond search period	Beyond search period	Beyond search period	NA	0.4% 1/50 2/48 0.48 (0.04-5.12)	NA 1/50 2/48 NA	Excluded trials with < 500 participants	NA
Xue et al (51) 2017	Beyond search period	Beyond search period	Beyond search period	Beyond search period	Beyond search period	0.7% 3/139 2/173 1.87 (0.32-11.02)	NA 3/139 2/173 NA	Excluded trials with < 500 participants	NA

The trials by Chapuy et al (45, 46) are the only ones conducted in institutionalized populations. All the other trials are conducted in community-dwelling populations. Individual cells are filled with the following: Percentage is weight of each trial, number of fractures/treatment group, number of fractures/control group, and effect size estimate.

Abbreviations: AMSTAR-2, A MeaSurement Tool to Assess systematic Reviews 2; Ca, calcium; D, vitamin D; fx, fracture; MA, meta-analysis; NA, not available; NV, nonvertebral; RCT, randomized clinical trial; RR, relative risk; SR, systematic review; V, vertebral; WHI, Women’s Health Initiative.

MAs not included in matrix: DIPART, 2010: individual participant data without details about individual RCTs (23); Tricco et al, 2017: Network MA without details about individual RCTs (39).

Reasons for exclusion of some trials by systematic reviews/meta-analyses either as exactly reported or derived from inclusion/exclusion criteria.

“The test for subgroup differences was not significant (P = .13).”

Not a subgroup analysis; community and institutionalized were each analyzed separately.

“P for heterogeneity: 0.02.”

This RR is for high D (≥ 800 IU/day) + high Ca (≥ 800 mg/day) vs placebo. RR for low D (< 800 IU/day) + high Ca (≥ 800 mg/day) is 0.48 (0.04-5.16). RR for high D (≥ 800 IU/day) + low Ca (< 800 mg/day) is 1.87 (0.31-11.13).

Any fracture is defined as an fx occurring at any site, but if an RCT reported only hip fx, these were also counted as any fx.

Only total number available (number of events per treatment/control is not available).

CI for equally allocated subgroup.

Two SRs/MAs conducted subgroup analysis by residency (15, 30). Although there was a trend for a higher effect in institutionalized individuals, none of the SRs/MAs found a subgroup effect (15, 30). The Cochrane SR/MA reported a 5% RRR in any fracture and an absolute risk reduction varying from 1 of 1000 for a lower baseline risk population (of 26/1000), to 4 of 1000 for a higher baseline risk population (of 74/1000) (15). Weaver et al (29) conducted 2 primary analyses according to residency and the effect size CIs were overlapping. There was a subgroup effect by age in one SR/MA; P heterogeneity of .02, P of interaction not reported (30). Other subgroup analyses did not show any significant effect (see Table 4).

Results of Systematic Review/Meta-Analysis on Vitamin D Alone vs Placebo

Out of the 19 SRs/MAs on vitamin D alone vs placebo/control, only 3 were of moderate quality, published by the Cochrane Group (15, 36, 37), 15 evaluated the effect on hip fractures (15, 16, 19, 20, 23-28, 30-32, 35, 39), 13 on any fractures (15, 19, 21, 23, 24, 27, 30-32, 34, 36, 37, 39), and 2 on fall-related fracture (36, 37). They were all concordant in direction and statistical significance. They showed a trend for an increased risk of hip (RR 1.10-1.30) and any fracture (RR 1.01-1.09), but none reached statistical significance.

Discussion

In this umbrella review, we synthesize, compare, and contrast methodological quality and findings, and investigate reasons for discordance of SRs/MAs of RCTs evaluating the efficacy of vitamin D supplementation, with or without Ca supplementation, on fracture risk. Importantly, the majority of the identified SRs/MAs were rated of low to critically low quality, secondary to the lack of or insufficient data in one or more of the critical AMSTAR-2 domains, specifically limitations in the search strategy, the lack of a list for excluded trials, the absence of an appropriate bias assessment or accounting for it in the interpretation of results, and the lack of investigation of publication bias. Several of these items were added to the AMSTAR-2 tool, but were not available in AMSTAR-1, which was available when many of the included SRs/MAs were conducted and/or published (14).

Most of the included SRs/MAs showed a protective effect of Ca/D compared with placebo/control. The discordance in results stemmed from differences in the magnitude of the effect size and statistical significance, both for hip and any fracture. SRs/MAs that exclusively considered RCTs in community-dwelling individuals failed to show any significant reductions in fractures. We identified only one SR/MA of moderate quality, pooling data from institutionalized and community-dwelling individuals, and that evaluated the quality of evidence (15). It reported high-quality evidence of a small RRR in hip fracture by 16% and any fracture by 5% with vitamin D dose of 400 to 800 IU/day, coadministered with Ca, compared with placebo/control. Fracture risk reduction in institutionalized individuals was derived from 2 RCTs, included in all SRs/MAs except those exclusively targeting the community (31, 38). These 2 RCTs enrolled older women (mean age 85 years) living in nursing homes, with mean serum 25(OH)D concentration 22.5 to 40 nmol/L, randomly assigned to daily Ca/D 1200 mg/800 IU or placebo, for 18 to 24 months (45, 46). Each of these RCTs demonstrated a beneficial effect of Ca/D in reducing the risk of hip and nonvertebral fractures (45, 46).

Subgroup analyses by residency status were performed in 2 SRs/MAs. The first revealed an RR of 0.75 (0.62-0.92) (15), and the second of 0.69 (0.53-0.90) (30), in institutionalized individuals, but the interaction term was not significant for the former, thus ruling out an effect of residency on risk reduction, and was not provided for the latter. The lack of evidence for a subgroup effect by residency may be due to the imprecise effect size estimates, with large CIs, despite a large number of trials and sample sizes (2000-3500 [41, 44, 53], and > 36 000 participants in the WHI trial [48]), explained by the very low baseline fracture risk (< 1%) in this population (47-50). Indeed, the neutral results of subgroup analyses could imply a lack of effect, or it may also be affected by the evaluation of subgroup effect within vs between trials, and prespecifying subgroup analysis before the conduct of the SR/MA, among others (54).

Vitamin D alone does not appear to protect against fractures when compared with placebo, and this is consistent with the findings of a previous umbrella review (55). Concomitant Ca supplementation is needed (56), as the combination has a synergistic effect on intestinal Ca absorption (57), on reversal of secondary hyperparathyroidism, especially if 25(OH)D is less than 25 nmol/L (58), and on reducing body sway and fall risk (59).

The discordance in results between SRs/MAs is multifactorial (60), including the clinical question investigated, search period, eligibility criteria, intervention, cointervention, and outcome. For hip fracture outcome, few SRs/MAs aimed to assess interventions that reduce fracture risk in individuals with osteoporosis or at risk for osteoporosis or fragility fractures (16, 25, 28), while for others the population consisted of postmenopausal women and older men. Despite this variability in the population of interest, there was a significant overlap (CCA > 15% both for hip and any fracture) of the included RCTs, many of which had patients with previous fractures (15-17, 23, 24, 26, 29-31, 37). Fracture definition varied across SRs/MAs. Some SRs/MAs included proximal femoral fractures as hip fractures (15, 19, 24, 31), whereas others did not (29). “Any fracture” included fall-related fractures (36, 37), or nonvertebral fractures (15, 20), or fractures defined at specific sites (hip, spine, or wrist) (21), or fractures at any site (30), or without site specification (31). None of the SRs/MAs targeted fragility fractures per se, and few excluded nonosteoporotic fractures (skull, carpal, metacarpal, tarsal, or metatarsal bones) (61-63). There was heterogeneity in the measure of association used. While the majority of studies used RR, the preferred tool, a few studies used hazard ratio or odds ratio; the latter tends to inflate the association (64). Similarly, there was variability in the methods used for pooling of effect sizes and in the data derived from each trial (6, 60). Although all SRs/MAs used a complete case analysis, which is the appropriate analysis method to deal with missing data at the trial level in the primary analysis (65), some used intention to treat or per-protocol analyses for one or more RCTs.

There are 2 previous umbrella reviews on vitamin D supplementation. Theodoratou et al (55) included observational and interventional SR/MAs until 2013, addressing 137 health outcomes. Stubbs and colleagues (66) assessed interventions preventing falls in the community, among which was the efficacy of vitamin D supplementation on fall-related fractures. The first concluded there was no evidence, and the second commented on conflicting evidence for a fracture risk reduction. Neither delved into a critical appraisal of the quality of SRs/MAs, or importantly into sources for any discrepancy in results from various SRs/MAs, and their heterogeneity led to conflicting conclusions.

Pollock et al (67) recognized several methodologic challenges in the overview of reviews. When a meta-analysis of SR/MA is conducted, overlap between reviews is a major consideration, as it is associated with the risk of “double counting” studies included in more than one SR/MA (67). Therefore, some experts select only one SR/MA when several are eligible, such as the most recent one, or the one conducted by the Cochrane Group (67). We have quantified the overlap between the SRs/MAs included in our umbrella review. However, we did not pool their results, and therefore, we do not expect an amplification of the related bias in the interpretation of the results. We have used the AMSTAR-2 tool to evaluate the quality of the included SRs/MAs. Given that this tool implies some subjectivity, the assessment was conducted by 3 reviewers (including content experts M.C. and G.E.H.F.), and the rationale for this decision is detailed in Supplementary data 3 (12), as previously suggested (67). We did not use the GRADE approach to summarize our findings because to date there is no clear guidance on how to apply the GRADE principles in an umbrella review (67, 68). We did not assess for publication bias because this remains one of the challenges of umbrella reviews without a validated tool to assess it (67).

The limitations of our findings are in large part due to the limitation of the evidence available. Baseline vitamin D status was reported in only 14 SRs/MAs (15, 19, 21, 24-26, 30-32, 35-37), and fewer reported follow-up status (19, 21, 30, 32, 35). The large variation in 25(OH)D assays was not addressed. Several SRs/MAs on vitamin D alone compared with placebo/control included trials extending over 1 to 10 months (15, 17, 19, 21, 30-32, 35), an insufficient period to demonstrate any significant effect on fractures. For SRs/MAs on Ca/D, a narrow range of vitamin D dose was used in the included trials, and therefore a dose effect cannot be investigated. Some discrepancies in SRs/MAs might be related to human errors; however, we did not investigate this possibility because verifying trial data was beyond our aim.

To our knowledge this is the only umbrella review on a controversial topic that is exclusively focusing on SRs/MAs analyses of RCTs, and solely on fractures. Our umbrella review fulfills the Joanna Briggs Institute guidance on conducting umbrella reviews (69) (Supplementary data 4) (12). Furthermore, the quality assessment of SRs/MAs, the matrices to dissect their similarities and differences, and important subgroup analyses allow a better understanding of the evidence available and contradicting results, and provides a path forward.

There is a wide variability in the development methods of the current vitamin D guidelines (70) and in their recommended vitamin D doses. Despite the lack of solid evidence of a protective effect of vitamin D supplementation in younger adults (age 50-65 years) and noninstitutionalized individuals, the majority of the guidelines recommend vitamin D supplementation for adults (8, 9, 71-77). Only a few guidelines target older adults, those older than 60 to 70 years (78-80), or older institutionalized individuals (81, 82), or recommend a higher dose in older individuals (9, 75-77), populations in which the evidence is most convincing, namely institutionalized individuals, or most suggestive (age as a high risk).

Therefore, there is a need to improve the methodologic rigor of vitamin D guidelines in order to issue recommendations that are evidence based (70).

Conclusions

Ca/D reduces the risk of hip and any fracture in analyses combining institutionalized and community-dwelling individuals. High-risk individuals, such as those older, institutionalized, or with low vitamin D status, may benefit most, although this could not be unequivocally demonstrated in the few SRs/MAs that evaluated such predictors, most likely because of low power. An IPD-MA restricted to individuals with sufficient follow-up, rather than pooled data, will be more powerful to evaluate treatment efficacy and predictors of fracture reduction. These would include age, residency status, baseline vitamin D level, and regimen used (daily, weekly, or monthly vitamin D).