2.1. Research Question
Is magnesium supplementation taken by mouth, in comparison to placebo, effective and safe for use by older adults with insomnia?
2.2. Search and Selection Strategy
A SR of primary studies was conducted in MEDLINE (1946 to October Week 1 2020), EMBASE (1947 to October 16, 2020) and Allied and Complementary Medicine (AMED) (1985 to October 2020) up to 18 October 2020. The electronic search was developed in MEDLINE, then translated to EMBASE and AMED (chosen for their recognized pharmacological and CAM focuses). Two additional databases, PROSPERO and Cochrane Library were hand-searched using the key term “magnesium”; zero reviews were found relevant to the research question. More details of search terms, rationale for inclusion of each database and exported strategies are available in Additional File 1 and 2.
The search strategy used both key text words and indexed standardized subject terms (MESH). To balance comprehensiveness and relevance in this relatively unstudied topic, three concepts were identified: insomnia, older age and magnesium. A broad set of search terms was gathered for the first two concepts and then cross-referenced with existing systematic reviews on similar topics (10,11). Magnesium supplementation was identified to be the most specific concept; to increase the sensitivity of the search, the general term magnesium was used (instead of magnesium supplementation or intake). The key words and MESH terms were combined with OR Boolean operators and the three concepts were combined using the AND Boolean operator. Finally, a validated randomized control trial (RCT) search filter was applied (12). Additionally, reference lists of included papers were hand-searched for records.
Citations were imported into a reference manager, EndNote (2018) (13), where duplicates were removed. The remaining records were uploaded to Covidence (2019) (14), a SR web platform where both authors screened titles and abstracts before proceeding to full text retrieval and eligibility assessment.
2.3. Eligibility Criteria
Bibliographic records screened were eligible for inclusion if:
- Studies were RCTs of parallel-group or cross-over design. (Type of Study)
- The mean age of study participants was greater or equal to 55 years old. (Population)
- Participants were diagnosed with insomnia by standardized measure (e.g. validated questionnaire), clinician evaluation/test (e.g. sleep laboratory), or self-reported sleep diary. (Population)
- Studies evaluated oral magnesium supplementation of any dose, frequency, duration or formulation in comparison to placebo or no treatment. (Intervention & Comparison)
- Outcomes were not used to determine eligibility to minimize selection bias by the sole author. However, the author decided that the symptoms of insomnia (difficulties initiating or maintaining sleep or affecting quality of life) would guide the clinically relevant outcomes of this SR. Outcome domains: (1) sleep quality, measured by sleep questionnaires and (2) sleep quantity, measured by sleep parameters. A third outcome domain of (3) presence of adverse events was chosen.
Studies combining magnesium supplementation with another intervention (co-intervention) were excluded. Non-English studies were excluded due to resource constraints.
2.4. Data Extraction
The authors JM and TP extracted the following data using a modified version of Cochrane’s (2020) template data collection form for intervention reviews of RCTs (Additional File 3): (1) General information (location, ethics, funding sources, conflicts of interest); (2) Study methods (aim, design of RCT, start/end dates, methods of randomization, allocation concealment and blinding); (3) Participants (inclusion/exclusion criteria, setting, method of recruitment, number per group, characteristics including age and sex, comorbidities, method of diagnosis); (4) Intervention and comparison (full description of magnesium supplementation regimen, duration of treatment, full description of control treatment, washout period for cross-over RCTs), (5) Follow-up (length, withdrawal rate and reasons); (5) Outcome data (description, definitions, time points measured/reported, measurement tools); (6) Results (effect estimates and precision per group, subgroups); (7): Analysis data (intention-to-treat or per-protocol analysis, comparability at baseline, statistical techniques); (7) key author conclusions. Discordance in data collection were adjudicated and consensus was reached for each discordant item.
2.5. Assessment of Risk of Bias
The authors carried out risk of bias assessments for included studies using the RoB 2.0 tools (see example in Additional File 4) for individually randomized parallel-group and cross-over trials and included the following bias domains: randomization/allocation process, deviation from intended intervention, missing outcome data, outcome measurement and selective outcome reporting (15,16). Visualization of RoB 2.0 was produced using robvis (17). Similarly, discordance was dealt with by adjudication amongst the authors until consensus was reached by following the appropriate algorithms.
2.6. Data Synthesis
Measures of effect were summarised using mean differences (post-intervention or change-from-baseline where appropriate) with standard deviations for continuous outcomes (no outcomes were dichotomized). When interventions were similar, and outcome measures were reported in the same scale, a meta-analysis was used to synthesise results and increase precision using random-effects model allowing for heterogeneity in the included studies. Heterogeneity was assessed qualitatively (by comparing studies for variability amongst populations, interventions, or designs) and statistically (with the chi-squared test and I2 statistic). Sensitivity analyses and funnel plots were not conducted due to insufficient number of studies. If a meta-analysis was not appropriate, and outcome data was not consistent across studies, vote counting based on direction of effect (not statistical significance) was employed as a synthesis method to complement the narrative synthesis. A modified effects direction plot was used to visualize vote counting for outcome domains, accounting for sample size and study quality (18,19). Quality of evidence across outcomes was assessed using the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) approach, which incorporates the RoB 2.0 tool(20). GRADE defines the certainty of evidence as the extent to which we can be confident that our results are representative of the true value of interest (20)17. Additional File 5 lists the five GRADE domains and provides an example of the approach. This systematic review protocol was not registered, but follows PRISMA guidelines (Additional File 6).