Overall search findings
A total of 25,637 records were retrieved, 25,605 from the seven electronic databases and 32 from manual searching. After removed the duplicate records and obviously irrelevant records, 39 records remained. Read the full text then 18 records were excluded. The 18 excluded studies as well as the reasons for exclusion were listed in Additional file 4. A total of 21 eligible RCTs were included for meta-analysis by reading the full text of the potentially relevant records. The flow chart for the identification of studies was presented in Fig. 1.
Characteristics of included studies
These studies were originated from nine countries and were published between 1998 and 2019. The sample sizes of included studies ranged from 24 to 2214. A total of 7571 participants were included in systematic review, of which 3812 were in the intervention group and 3759 were in the control group. They were recruited from homes, communities or hospitals. The participants of six studies [27–29, 35–37] were MCI patients and the other fifteen studies [25, 34, 38–50] were elderly adults without cognitive impairment. Some of these participants were accompanied with one or more chronic diseases, such as mental disorders, stroke, diabetes mellitus, transient ischemic attack or hypertension. The baseline levels of serum Hcy of participants included were high (ranged from 9.7 ~ 20.6 µmol/L). The mean age of participants ranged from 60.0 to 83 in the intervention group and 60.0 to 82 in the control group. The treatment duration ranged from 1 month to 3.4 years, eleven studies [27, 36, 37, 39, 40, 45–50] ≤ 12 months and ten studies [25, 28, 29, 34, 35, 38, 41–44] > 12 months. The detail characteristics of 21 included studies were presented in Table 1.
Quality assessment of included studies (risk of bias)
The risk of bias of included studies were summarized in Fig. 2. Overall, the selection bias of the 21 studies included was reasonable: 13/21 of the studies included were assessed to have a low risk of bias for random sequence generation; 14/21 of the studies included were assessed to have a low risk of bias for allocation concealment. The study conducted by Fei Ma which published in 2019[27] had a high risk of bias for blinding of assessors, participants and providers. Similarly, another study conducted by Fei Ma which published in 2017[29] had a high risk of bias for blinding of assessors. All the studies included had a low risk of bias for incomplete outcome data. The six studies included [25, 34–36, 41, 43] were assessed to have a low risk of bias for selective reporting, and the protocols of the rest studies included were unavailable. The quality assessment results of risk of bias were presented in Additional file 5.
Outcome evaluation
Four outcome measures of cognitive function (global cognitive function, information processing speed, episodic memory, executive function) and the levels of serum Hcy were used to evaluate the preventive efficacy of vitamin B supplement in MCI patients or elderly adults without cognitive impairment. We presented the detailed results of meta-analysis as follows.
Global cognitive function
We pooled the data extracted from 15 RCTs [25, 27–29, 34–36, 39, 41–44, 46, 48, 49] and found a substantial heterogeneity (heterogeneity test: Chi-square = 173.45, df = 14, P < 0.01, I-square = 92%). The intervention group achieved significant preventive efficacy on global cognitive function (SMD: 0.36; 95% CI: 0.18 to 0.54, P < 0.01) by using a random-effects model (see Fig. 3).
The sensitivity analysis revealed the pooled SMD as well as its 95%CI of global cognitive function was robust. The obvious overlap of the 95%CI of short treatment duration (≤ 12 months) contrast long treatment duration (> 12months) and the test for subgroup differences (P = 0.67) indicated that there was no interaction between the pooled SMD and treatment duration. On the contrary, the negligible overlap of the 95%CI of MCI patients contrast elderly adults without cognitive impairment and the test for subgroup differences (P = 0.03) indicated that there was interaction between the pooled SMD and participants. The detailed results of subgroup analysis were presented in Table 2.
Table 2
The results of subgroup analysis
Outcome measures | Subgroup | The Num of studies | Pooled estimate [SMD/MD (95%CI)] | P value | I2(%) | Test for subgroup differences |
global cognitive function | Treatment duration | ≤ 12 months | 7 | 0.35(-0.15, 0.84) | P = 0.17 | 87 | P = 0.67 |
> 12 months | 8 | 0.23(0.07, 0.40) | P = 0.006 | 91 | |
participants | MCI patients | 5 | 0.82(0.20, 1.45) | P = 0.01 | 95 | P = 0.03* |
EAWCI | 10 | 0.13(0.03, 0.23) | P = 0.01 | 66 | |
information processing speed | Treatment duration | ≤ 12 months | 5 | -0.19(-0.38, 0.01) | P = 0.06 | 39 | P = 0.007* |
> 12 months | 4 | 0.16(0.01, 0.31) | P = 0.04 | 64 | |
participants | MCI patients | 1 | 0.04(-0.28, 0.36) | P = 0.79 | NA | P = 0.89 |
EAWCI | 9 | 0.07(-0.13, 0.27) | P = 0.50 | 81 | |
episodic memory | Treatment duration | ≤ 12 months | 7 | -0.09(-0.27, 0.09) | P = 0.34 | 40 | P = 0.09 |
> 12 months | 6 | 0.08(0.02, 0.13) | P = 0.01 | 0 | |
participants | MCI patients | 4 | 0.42(-0.24, 1.08) | P = 0.21 | 94 | P = 0.26 |
EAWCI | 11 | 0.04(-0.04, 0.12) | P = 0.34 | 28 | |
executive function | Treatment duration | ≤ 12 months | 4 | 0.01(-0.14, 0.17) | P = 0.87 | 0 | P = 0.21 |
> 12 months | 5 | -0.29(-0.75, 0.16) | P = 0.20 | 96 | |
participants | MCI patients | 3 | -0.33(-0.81, 0.15) | P = 0.18 | 89 | P = 0.59 |
EAWCI | 8 | -0.17(-0.51, 0.17) | P = 0.34 | 94 | |
Hcy | Treatment duration | ≤ 12 months | 7 | -4.77(-6.28, -3.27) | P < 0.01 | 78 | P = 0.68 |
> 12 months | 5 | -4.41(-5.27, -3.54) | P < 0.01 | 82 | |
participants | MCI patients | 3 | -5.45(-6.51, -4.39) | P < 0.01 | 81 | P = 0.05 |
EAWCI | 8 | -4.13(-4.93, -3.32) | P < 0.01 | 65 | |
Abbreviations: SMD: standardized mean difference; MD: mean difference; CI: confidence interval; MCI: mild cognitive impairment; EAWCI: elderly adults without cognitive impairment; Hcy: Homocysteine; NA: not applicable |
* P < 0.05 |
Information processing speed
We pooled the data extracted from 10 RCTs [34, 36, 38, 40, 41, 44–47, 50] and found a substantial heterogeneity (heterogeneity test: Chi-square = 43.23, df = 9, P < 0.01, I-square = 79%). The intervention group achieved no preventive efficacy on information processing speed (SMD: 0.06; 95% CI: -0.12 to 0.25, P = 0.49) by using a random-effects model (see Fig. 4).
The sensitivity analysis revealed the pooled SMD with its 95%CI of information processing speed was robust. The non-overlap of the 95%CI of short treatment duration (≤ 12 months) contrast long treatment duration (> 12months) and the test for subgroup differences (P = 0.007) indicated that there was interaction between the pooled SMD and treatment duration. On the contrary, the obvious overlap of the 95%CI of MCI patients contrast elderly adults without cognitive impairment and the test for subgroup differences (P = 0.89) indicated that there was no interaction between the pooled SMD and participants. The detailed results of subgroup analysis were presented in Table 2.
Episodic memory
We pooled the data extracted from 15 RCTs [28, 34–38, 40, 41, 43–45, 47–50] and found a substantial heterogeneity (heterogeneity test: Chi-square = 71.81, df = 14, P < 0.01, I-square = 81%). The intervention group achieved no preventive efficacy on episodic memory (SMD: 0.10; 95% CI: -0.04 to 0.25, P = 0.16) by using a random-effects model (see Fig. 5).
The sensitivity analysis revealed the pooled SMD with its 95%CI of episodic memory was robust. The obvious overlap of the 95%CI of short treatment duration (≤ 12 months) contrast long treatment duration (> 12months) and the test for subgroup differences (P = 0.09) indicated that there was no interaction between the pooled SMD and treatment duration. Similarly, the obvious overlap of the 95%CI of MCI patients contrast elderly adults without cognitive impairment and the test for subgroup differences (P = 0.26) indicated that there was no interaction between the pooled SMD and participants. The detailed results of subgroup analysis were presented in Table 2.
Executive function
We pooled the data extracted from 11 RCTs [28, 34–36, 38, 40, 41, 44, 45, 47, 50] and found a substantial heterogeneity (heterogeneity test: Chi-square = 139.63, df = 10, P < 0.01, I-square = 93%). The intervention group achieved no preventive efficacy on executive function (SMD: -0.21; 95% CI: -0.49 to 0.06, P = 0.13) by using a random-effects model (see Fig. 6).
The sensitivity analysis revealed the pooled SMD with its 95%CI of executive function was robust. The obvious overlap of the 95%CI of short treatment duration (≤ 12 months) contrast long treatment duration (> 12months) and the test for subgroup differences (P = 0.21) indicated that there was no interaction between the pooled SMD and treatment duration. Similarly, the obvious overlap of the 95%CI of MCI patients contrast elderly adults without cognitive impairment and the test for subgroup differences (P = 0.59) indicated that there was no interaction between the pooled SMD and participants. The detailed results of subgroup analysis were presented in Table 2.
Hcy
We pooled the data extracted from 11 RCTs [27–29, 34, 38–40, 44–46, 49] and found a substantial heterogeneity (heterogeneity test: Chi-square = 96.55, df = 10, P < 0.01, I-square = 90%). The intervention group achieved significant preventive efficacy on Hcy (MD: -4.59; 95% CI: -5.51 to -3.67, P < 0.01) by using a random-effects model (see Fig. 7).
The sensitivity analysis revealed the pooled MD with 95%CI of global cognitive function were robust. The obvious overlap of the 95%CI of short treatment duration (≤ 12 months) contrast long treatment duration (> 12months) and the test for subgroup differences (P = 0.680) indicated that there was no interaction between the pooled SMD and treatment duration. On the contrary, the non-overlap of the 95%CI of MCI patients contrast elderly adults without cognitive impairment and the test for subgroup differences (P = 0.05) indicated that there was interaction between the pooled SMD and participants. The detailed results of subgroup analysis were presented in Table 2.