Vitamin D Supplementation and Mortality in the Elderly: A Review of a Meta‐Analysis of Randomized Controlled Trials Using the REB Method Authors: Natalia Kostakis, Jean‐Christophe Lega, Remy Boussageon Published: 2026 ABSTRACT ABSTRACTBackgroundThe results of randomized controlled trials (RCTs) regarding the effect of vitamin D supplementation on mortality reduction are conflicting.ObjectiveThe aim of this study was to assess the level of evidence through a sensitivity analysis of the meta‐analysis by Shah et al. (2024) using the Rebuilt the Evidence Base (REB) method.MethodsThe main inclusion criterion was a double‐blind, placebo‐controlled RCT design. The exclusion criteria were trials combining vitamin D with other nutrients and unpublished RCTs. The main outcome was all‐cause mortality. A bias analysis was performed, followed by a hypothetico‐deductive analysis to determine the confirmatory or exploratory nature of each RCT. The REB method was used to interpret the results and assess the level of evidence.ResultsAmong the 25 RCTs included in the meta‐analysis by Shah et al., the analysis was performed on 8 RCTs at low risk of bias including 30 320 participants. Seven RCTs raised some concerns, and two RCTs were at high risk of bias according to ROB2. Bias arose from the randomization process, deviations from intended interventions, missing outcome data, and selection of the reported results. The results were statistically significant, in favor of vitamin D supplementation: RR = 0.96 (95% CI [0.92, 1.00]), p = 0.04. No RCT was conclusive.ConclusionAccording to the REB method, there is an evidence signal regarding the benefit of vitamin D supplementation in the population aged 75 and older, which remains to be confirmed.Keywords: all‐cause mortality, meta‐analysis, RCTs, REB method, vitamin D Abbreviations confidence interval Fragility Index hazard ratio intention‐to‐treat international unit Jean‐Christophe Lega modified intention‐to‐treat Mini‐Mental State Examination Natalia Kostakis odds ratio participants, intervention, comparison, outcome Preferred Reporting Items for Systematic Reviews and Meta‐Analyses Rémy Boussageon randomized controlled trial rebuild the evidence base review manager Risk of Bias Version 2 risk ratio Introduction In July 2024, by consensus of a panel of experts, the American Society of Endocrinology established guidelines on supplementation and doses of vitamin D in different populations and age groups [1]. The guidelines provide a weak recommendation (“suggests”) for vitamin D supplementation among the population aged over 75, based on “its potential to lower the risk of mortality” (recommendation 6, page 1911). These guidelines are supported by the meta‐analysis of the Endocrine Society addressing 14 questions on age categories, populations, events (mortality, cardiovascular events, risk of fractures, etc.) [2]. One of them focuses on the effect on mortality in individuals aged 75 years and older: RR = 0.96 (95% CI [0.93, 1.00]). However, the confidence interval is close to the null value. In France, the current recommendations align with those of the American Society. A daily dose of 1000 IU of vitamin D3 is suggested for individuals aged 75 and older [3]. Yet an Australian randomized controlled trial (RCT) including 21 315 participants over 60 concluded that there was no benefit of a 60 000 IU monthly supplementation of vitamin D3 on mortality versus placebo [4]: HR = 1.04 (95% CI [0.93, 1.18]). A subgroup analysis of participants aged over 75 showed no statistically significant effect: HR = 1.00 (95% CI [0.87, 1.15]). Moreover, a meta‐analysis of 50 RCTs assessing the effect of vitamin D versus placebo on all‐cause mortality among adults concluded that there was no statistically significant effect: RR = 0.98 (95% CI [0.96, 1.02]) [5]. These conflicting results require a rigorous assessment of the level of evidence regarding the efficacy of vitamin D supplementation in reducing mortality. Even a small effect would provide a strong rationale to justify supplementation in this specific population. This study is a sensitivity analysis of the meta‐analysis published by the American Endocrine Society [1], using the “rebuilt the evidence base” (REB) method, which is a rigorous method for interpreting results [6]. Methods The study protocol was not preregistered. This study is a reanalysis of an existing meta‐analysis, and the analytical framework is inherently defined by the REB methodology. Thus, the fact that the protocol was not preregistered does not compromise the reliability of the study's conclusion. The reporting followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) guidelines [7]. The analysis applied the REB method [6] (Figure 1) to the results of the meta‐analysis by Shah et al. [2]. This was based on the structured method for formulating a precise research question, the PICO 6 (Population, Intervention, Comparison, Outcome) in the meta‐analysis by Shah et al. [2]. The meta‐analysis included 25 RCTs involving a population aged over 75 years (P), comparing vitamin D supplementation (I) with placebo (C), and evaluating the reduction in all‐cause mortality (O). The search in the meta‐analysis by Shah et al. [2] for our PICO was carried out in the following databases: Ovid MEDLINE (R) ALL from 1946 to March 3, 2022; EBM Reviews–Central Register of Controlled Trials in January 2022; Cochrane Database of Systematic Reviews from 2005 to March 2, 2022; and Embase from 1974 to March 3, 2022. The main inclusion criteria were RCTs or nonrandomized trials, healthy populations (except for obesity and prediabetes for the related PICO), comparison between vitamin D (regardless of dose or frequency) and no vitamin D, and trials in which vitamin D was measured before 20 weeks of pregnancy. The exclusion criteria were multiple interventions with vitamin D (excluding additional calcium supplementation), outcomes irrelevant to the research question, non‐English studies, narrative reviews, and reviews whose full text was unavailable. The methodology used by Shah et al. [2] is detailed in Table S1. REB project flowchart and interpretation table. Eligibility Criteria The inclusion criteria for our analysis using the REB method [6] were double‐blind, low‐risk RCTs according to the Risk of Bias 2 (ROB2) tool [8]. The exclusion criteria were RCTs where the combination of vitamin D and other nutrients made the measurement of the specific effect of vitamin D impossible, and unpublished RCTs making an accurate assessment of the risk of bias using the ROB2 [8] impossible. Selection of Studies The RCTs included in the meta‐analysis by Shah et al. [2] were read in full by two authors (NK and RB) to ensure they met the criteria for the REB method [6]. Data from each RCT were summarized (Table S2 and Appendix S1). Consensus was reached through discussion with the three authors, N.K., J.C.L., and R.B. Outcome Measure The primary outcome was the incidence of all‐cause mortality, which is the occurrence of death from any cause over the follow‐up period. Risk of Bias Assessment The risk of bias according to the ROB2 [8] was assessed for each RCT. For Domain 1, without a detailed description of the randomization process, the risk of bias was rated as at least “some concerns.” The researchers considered that if double blinding was maintained, the overall risk of bias was low, unless other biases were identified. For Domain 5, without protocol preregistration, the risk of bias was rated as at least “some concerns.” The researchers considered that if the “methods” section was detailed enough, the overall risk of bias judgment was low, unless other biases were identified. The Fragility Index (FI) [9] was calculated using Clincalc.com in certain limited cases such as rare events, a number of participants lost to follow‐up close to the difference between groups in the number of events, a small sample, or a p value close to 0.05. An FI between 0 and 1 confirms a high risk of bias due to missing data. Because the interpretation of bias is subjective [10], we decided to conduct a sensitivity analysis including the RCTs rated as “some concerns.” Effect Measurement Review Manager Version 5.4 (RevMan) was used to perform the statistical analysis. Relative risks and their 95% confidence intervals (95% CIs) were calculated. A result with a p value < 0.05 was considered statistically significant. Heterogeneity was assessed using tau2 and the I 2 index. A fixed‐effect model was used for the analysis, as it constitutes a valid way to test the null hypothesis of no effect. In addition, the random‐effects model requires an estimate of the between‐study variance, which is unreliable with a small number of studies [11]. However, we conducted a sensitivity analysis with a random‐effects model to test the robustness of the results. REB Analysis The REB method [6] considers the overall risk of bias judgment of a specific outcome of an RCT, heterogeneity, publication bias, the confirmatory or exploratory nature of an outcome, and the number of conclusive RCTs [6]. The meta‐analysis includes only outcomes with an overall low risk of bias according to the ROB2 tool [8]. A confirmatory outcome is prespecified in the protocol, and the analysis considers the overall alpha level and the multiplicity of tests. Otherwise, it is considered exploratory and at risk of false positives, leading only to a hypothesis. A conclusive trial has a statistically significant result for a confirmatory outcome with an overall low risk of bias [6]. Publication bias was not estimated in our analysis because the statistical power of a funnel plot is low when an insufficient number of RCTs are involved, making the estimation unreliable [12]. Results RCT Eligibility Among the 25 RCTs included in the meta‐analysis by Shah et al. [2], eight were excluded because they did not include a placebo [13, 14, 15], four because they did not measure the isolated effect of vitamin D [16, 17, 18, 19], and one because it was single‐blind [20] (Figure 2). Flowchart of the randomized controlled trials included in the analysis, PRISMA 2020. Risk of Bias Assessment Eight RCTs were at low risk of bias [4, 21, 22, 23, 24, 25, 26, 27], seven had some concerns [28, 29, 30, 31, 32, 33, 34], and two were at high risk of bias [35, 36] (Figure 3). Among the nine RCTs with some concerns or at high risk, biases were related to the randomization process in two trials [32, 36], to deviations from the intended interventions in two trials [28, 35], to missing outcome data in three trials [29, 31, 35], and to the selection of the reported results in eight trials [28, 29, 30, 31, 33, 34, 35, 36]. Risk of bias assessment according to the ROB2 tool on all‐cause mortality. The biases related to Domain 1 were due to predictable randomization based on the day [32], month, and year of birth, leading to an imbalance in baseline characteristics between groups [36]. Deviations from the intended interventions were due to a modified intention‐to‐treat analysis [35] or a single event in the study that was not analyzed [28]. Missing data biases were due to a high attrition rate (29%, Ooms et al. [35]) and a lack of sensitivity analyses, making the impact of the missing data impossible to judge [29, 31, 35]. A fragility index of 09 confirmed that three RCTs [29, 31, 35] had “some concerns.” Seven RCTs had some concerns in Domain 5 [28, 29, 30, 31, 33, 34, 35], as the outcome was not prespecified. Domain 5 of one RCT [36] was at high risk because the study protocol was not predefined, criteria were unspecified, and selective reporting of the most significant results was suspected. There are some discrepancies between our risk of bias assessment and that of Shah et al. [2, 37]. These discrepancies are reported in Appendix S3. RCT Characteristics The characteristics of the eight RCTs [4, 21, 22, 23, 24, 25, 26, 27] are presented in Table S1 (w1, w2, w7, w14, w16, w17, w19, and w23) and are described in more detail in Appendix S1. The selected RCTs were conducted in nine countries across three continents and included participants with a minimum age of 50 years. A majority of women were included in six RCTs [21, 22, 23, 24, 25, 26] and a majority of men in two RCTs [4, 27]. Doses of vitamin D included 400 [24], 800 [21], and 2000 [22, 26] international units (IU); a single bolus of 10 000 IU followed by 1000 IU/day [23]; 60 000 IU monthly [4]; and 100 000 IU every 4 months [25, 27]. Study durations varied between 2 and 6.1 years. All‐cause mortality was defined as a secondary or exploratory outcome in the majority of the trials [21, 22, 23, 24, 25, 27] and was not prespecified in one RCT [26]. It was defined as a primary and confirmatory outcome in one RCT [4]. Hypothetico‐Deductive Analysis Among the eight RCTs with a low risk of bias, none were conclusive regarding all‐cause mortality. Analysis of Results The eight RCTs with low risk [4, 21, 22, 23, 24, 25, 26, 27] had 30 320 participants, including 15 170 in the vitamin D group and 15 150 in the placebo group. Three RCTs [4, 26, 30] included larger age groups. Our analysis, as well as the meta‐analysis of Shah et al. [37], included only the 75 years and older subgroups [26, 30], or the 70 years and older subgroup [4]. Our meta‐analysis showed a statistically significant result regarding the benefit of vitamin D supplementation on all‐cause mortality with an RR = 0.96 (95% CI [0.92, 1.00], p = 0.04). There was no heterogeneity established with a tau2 and an I 2 = 0% (Figure 4). Forest plot: vitamin D supplementation on all‐cause mortality. The sensitivity analysis including RCTs with some concerns [28, 29, 30, 31, 32, 33, 34] had a total of 35 833 participants, 17 940 in the vitamin D group and 17 893 in the placebo group. It showed a statistically significant result regarding the benefit of vitamin D supplementation on all‐cause mortality with an RR = 0.96 (95% CI [0.92, 1.00], p = 0.03). There was no heterogeneity established with a tau2 and an I 2 = 0% (Figure 4). The sensitivity analysis based on a random‐effects model did not show a statistically significant result regarding the benefit of vitamin D supplementation on all‐cause mortality with an RR = 0.96 (95% CI [0.92, 1.00], p = 0.06). There was no heterogeneity detected with a tau2 and an I 2 = 0% (Figure 5). The random‐effects model shows the same point estimate of RR as the fixed‐effects model and is more conservative (the confidence interval is wider, and the p value is no longer statistically significant). Forest plot: vitamin D supplementation on all‐cause mortality using a random‐effects model. Our numbers and the ones in the meta‐analysis by Shah et al. [37] differ for two RCTs [24, 31]. These differences are detailed in Appendix S4. The funnel plot of Shah et al. [37] (page 236) did not show any asymmetry in the distribution of the studies; therefore, no publication bias was suspected. According to the REB method [6], considering the absence of conclusive RCTs and a statistically significant meta‐analysis (with the fixed‐effect model) without heterogeneity or suspected publication bias, we can conclude that there is a signal of evidence to be confirmed regarding the benefit of vitamin D supplementation on all‐cause mortality in adults aged 75 years and older. Discussion The main objective was to assess, using the REB method [6], whether the results of the meta‐analysis by Shah et al. [2], described as a “trend to reduce the risk of mortality” with a “high certainty” using the GRADE method [38], constituted conclusive evidence allowing us to recommend vitamin D supplementation in this population. The REB method [6], concluding that there is a signal of evidence to be confirmed, does not support this recommendation. Furthermore, our sensitivity analysis using a random‐effects model shows the same results with the same values for the RR, tau2, and I 2. According to the random‐effects model, there is no between‐study variance, which justifies the choice of a fixed‐effects model. However, the analysis is very sensitive to the choice of model, becoming not statistically significant with a random‐effects model (p value = 0.06). Therefore, our sensitivity analysis using the random‐effects model shows that the results are not robust, which justifies interpreting the results from the REB method as a “signal” rather than as evidence. Our results are interpreted in different ways. Shah et al. [2] consider the clinically significant result of a reduction in the relative risk of 6 per 1000 to be sufficient to recommend supplementation, which is affordable and causes few side effects. This differentiates the REB method [6] from the GRADE method [38]. Differences are observed in other domains and situations. The REB method [6], applied to the meta‐analysis by Truong et al. (2022) [39] on the benefit of cholinesterase inhibitors for mortality, which concluded that there was a moderate‐to‐high certainty of evidence according to the GRADE method [38], shows that mortality was not a confirmatory criterion in any of the included RCTs. Despite a positive meta‐analysis, the REB method [6] concludes that there is an absence of evidence. The example of the benefit of vitamin D for acute respiratory tract infections [40] was assessed as showing an absence of evidence according to the REB method [6]. Among the 27 included RCTs, 9 were confirmatory, and 1 was conclusive, with a negative meta‐analysis. This also illustrates the thinking of Bauchner and Ioannidis on how the interpretation of study results [41] retains a degree of subjectivity, potentially influenced by conflicts of interest. In an analysis of guidelines on vitamin D, these conflicts of interest are related to the recommendation for vitamin D prescriptions [42]. A point that raises questions is the risk of bias assessment. Our assessment and that of Shah et al. [2] differ for 13 of the 17 selected RCTs (Appendix S3), which clearly demonstrates the subjectivity of the ROB2 tool [10]. Among the 25 RCTs by Shah et al. [2], 10 were at low risk with a 53.47% weight [4, 14, 16, 22, 25, 26, 28, 29, 33, 34] compared to 8 in our analysis [4, 21, 22, 23, 24, 25, 26, 27]; 11 had some concerns [13, 17, 18, 20, 21, 23, 24, 27, 30, 32, 36] with a 44.93% weight compared to 8 in our analysis [28, 29, 30, 31, 32, 33, 34]; and 4 were at high risk [15, 19, 31, 35] with a 1.60% weight compared to 2 in our analysis [35, 36]. Considering the absence of justification for the ROB2 tool [8] judgment by Shah et al. [37], it was not possible to compare these differences in greater detail. A secondary analysis presented in the supplementary file of Shah et al. [37] ranked the RCTs by risk of bias. The meta‐analysis of 10 RCTs at low risk [4, 14, 16, 22, 25, 26, 28, 29, 33, 34] including 28 987 participants showed an RR = 0.98 (95% CI [0.93, 1.03]). This result was not reported in the conclusion and was contrary to the trend toward reduced mortality with high certainty. It makes an absence of effect very plausible. Sensitivity analyses according to the risk of bias are recommended to test the validity of results that include studies with a high risk of bias [43]. A discrepancy in results should downgrade the level of certainty. The secondary analysis [37] shows the effects of the studies with some concerns, RR = 0.95 (95% CI [0.89, 1.01]), and those at high risk, RR = 0.75 (95% CI [0.55, 1.03]). The confidence intervals tend toward a benefit of supplementation and are larger than the ones including low‐risk RCTs, RR = 0.98 (95% CI [0.93, 1.03]). Because no heterogeneity was observed (I 2 = 0%), smaller samples could explain the large CIs [44] and attest to the “small‐study effect” [45], which frequently involves studies at higher risk of bias and does not necessarily provide more information. Thus, our study shows that there is a difference in the assessment of the level of evidence for the efficacy of vitamin D in the population aged over 75 years in reducing all‐cause mortality, associated with a subjective interpretation of results depending on the consideration of the risk of bias and the exploratory or confirmatory nature of an outcome. The REB method [6] aims to define a priori what robust evidence is, through a predefined and transparent method of interpretation and evaluation of reproducible results, reducing the risk of false positives. Conclusion The REB method [6] applied to the RCTs included in the meta‐analyses of Shah et al. [2] leads to the conclusion of a signal of evidence regarding the benefit of vitamin D on all‐cause mortality in the population aged 75 and older. The weak recommendation of the American Society of Endocrinology does not seem justified. Author Contributions All the authors contributed to the design and development of the study, the analysis and interpretation of the data, the writing of this article, and its critical analysis, and they approved the submitted version. Conflicts of Interest The authors declare that they have no conflicts of interest. Supporting information Table S1: Characteristics of the meta‐analysis of Shah and al. Table S2: Characteristics of the 25 RCTs included in the meta‐analysis of Shah and al. Figure S4: Forest plot: vitamin D supplementation on all‐cause mortality, RCTs with low risk and some concerns according to ROB2. Appendix S1: Characteristics of the studies. Appendix S2: ROB2. Appendix S3: Table of differences in ROB2 assessment: meta‐analysis by Shah and al. versus our analysis. Appendix S4: Comparison between our data and those of Shah et al. Acknowledgements Open access publication funding provided by COUPERIN CY26. Contributor Information Natalia Kostakis, Email: natalia.kostakis@etu.univ-lyon1.fr. Jean‐Christophe Lega, Email: jean-christophe.lega@chu-lyon.fr. Remy Boussageon, Email: remy.boussageon@univ-lyon1.fr. Data Availability Statement Original data generated and analyzed during this study are included in this published article or in the data repositories listed in the references. References Associated Data Supplementary Materials Table S1: Characteristics of the meta‐analysis of Shah and al. Table S2: Characteristics of the 25 RCTs included in the meta‐analysis of Shah and al. Figure S4: Forest plot: vitamin D supplementation on all‐cause mortality, RCTs with low risk and some concerns according to ROB2. Appendix S1: Characteristics of the studies. Appendix S2: ROB2. Appendix S3: Table of differences in ROB2 assessment: meta‐analysis by Shah and al. versus our analysis. Appendix S4: Comparison between our data and those of Shah et al. Data Availability Statement Original data generated and analyzed during this study are included in this published article or in the data repositories listed in the references.