We searched the Cochrane Library, PubMed, Web of Science, Scopus, Embase, and ScienceDirect databases for randomized controlled trials on the effect of rTMS on sleep quality in patients with more than mild depressive mood, published in full text from the time of the construction of the library to September 2023, using a combination of subject headings and free-text headings via a computer network. The language is in English only. The English search terms mainly included “Transcranial Magnetic Stimulation,” “Depression,” “Sleep quality,” and “Dyssomnias.”

The EndNote X9 software was used to remove duplicates from the search, and then two reviewers independently read the titles and abstracts of the articles to establish eligibility for inclusion. Studies that failed to meet the inclusion criteria were not reviewed further. Those that could not be excluded were retrieved, and two reviewers (YL and XZ) assessed the whole text. The authors were contacted via e-mail when data validation or more information was required. Disagreements or ambiguities were resolved through discussion with a third reviewer (SH).

The inclusion criteria were as follows:

The exclusion criteria were as follows:

The Hamilton Depression Scale, Beck’s Depression Inventory, Depression Self-Rating Scale, and Pittsburgh Sleep Quality (sleep quality was measured by the Pittsburg Sleep Quality Scale) were used.

The following data were obtained from the included studies: information on the first author, year of publication, basic information of the experimental subjects, type of intervention, stimulation time, stimulation parameters, follow-up, country, duration of the experiment, the Depression Index, and the Pittsburgh Sleep Quality Index.

The quality of the included literature was also evaluated independently by two investigators using the Cochrane Collaboration’s risk-of-bias guidelines. In the case of a disagreement, a third investigator was called in to discuss the issue together until a consensus was reached. The evaluation measures included: 1) random sequence generation; 2) allocation blinding; 3) blinding of participants and personnel; 4) blinding of outcome assessors; 2) incomplete ending data; 6) selective reporting of results; and 7) other offsets. A trial was classified as having a “low risk of bias” (“+”) if it was judged to have a low risk of bias for all domains for this result. If it was assessed to raise some concerns in at least one area for this result, but not in any other, then it was categorized as having a “high risk of bias” (“−”). A trial was categorized as having an “uncertain risk of bias” (“?”) if it was considered to have concerns for several domains in a manner that considerably decreases confidence in the results. Discrepancies in the evaluation of the two assessors (YL and ZZ) were resolved through discussion with a third reviewer (WL).

All of the outcome indicators were the continuous variables in this study, and the mean difference (MD) or the standardized mean difference (SMD) with 95% confidence interval (95%CI) was used as the effect size to analyze these studies. MD was used when the studies were measured in the same manner or in the same units; otherwise, SMD was used. The meta-analysis produced effect sizes, which are statistically standardized representations of the quantitative findings of each study. These were calculated based on the mean pre–post change in the treatment group minus the mean pre–post change in the comparison group, divided by the pooled pretest standard deviation. Meta-analysis was performed using RevMan 5.4 and Stata 17.0 software. Heterogeneity was quantified using the I ² statistic, and a fixed effects model was used when there was no statistically significant difference in the heterogeneity test (I ² < 50%, p > 0.05); otherwise, a random effects model was used. Subgroup analyses and sensitivity analyses were performed to examine the sources of heterogeneity for the outcome indicators where heterogeneity existed. Publication bias was assessed using funnel plots and Egger’s asymmetry test for the outcome indicators with >10 included studies. All statistical significance levels were set at α = 0.05.

No experimental studies on humans and animals were conducted in this study, only a meta-analysis of original studies. Ethics committee approval was therefore not required.

A total of 339 pieces of literature were obtained through the database search, with 186 of these remaining after the removal of duplicates using EndNote. After the initial screening by reading the titles and abstracts of the studies, three researchers excluded 159 irrelevant papers based on the inclusion and exclusion criteria. Of the remaining 27 studies, 19 studies that did not meet the inclusion criteria were excluded after reading the full text and rescreening. An additional three studies found in the references that met the inclusion criteria were included. A total of 11 studies were finally included in the meta-analysis. A flowchart of the literature screening process is shown in Figure 1 .

A total of 548 patients from the 11 studies (19–29) were included in this study, of which 287 patients were in the experimental group and 261 patients were in the control group. The basic characteristics of the included studies are shown in Table 1 . The average age of the patients included in the studies ranged from 33 to 61 years. Six of the studies had patients with mild depression (21, 23, 24, 26, 28, 29), four studies had patients with moderate depression (19, 20, 22, 27), and one study had patients with severe depression (25). In all of the studies, the Pittsburgh Sleep Quality Index was rated as mildly sleep-disordered. The detailed characteristics of each included study are shown in Table 1 .

Details of the intervention parameters in the included studies are shown in Table 2 . Five studies used rTMS alone (19, 21, 23, 26, 29), while six studies used rTMS in combination with therapy (20, 22, 24, 25, 27, 28). Four studies used a stimulation frequency of 1 Hz (23, 25, 28, 29), two studies used 5 Hz (19, 21), five studies used 10 Hz (20, 22, 24, 26, 27), five studies used stimulus intensities below 100% (19, 20, 22, 23, 25), and six studies had stimulus intensities above 100% (21, 24, 26–29). Six studies had stimulus locations in the left dorsolateral prefrontal cortex (L-DLPFC) (20, 21, 24, 26–28), while three studies had stimulus locations in the right DLPFC (R-DLPFC) (23, 25, 29). Interestingly, these studies all used low-frequency rTMS. One study did not report the stimulation location (22), and one study used M1 as the stimulation location (19). Seven studies used an F8 coil type (19, 21, 23–25, 28, 29), one study used a C coil (24), while three studies did not report the coil type used (20, 22, 27). Four studies used a 2-week treatment duration (19, 21, 23, 29), two studies used a 4-week treatment duration (20, 25), and five studies used more than 4 weeks of treatment (22, 24, 26–28). Four studies conducted follow-ups (19, 20, 23, 29), while seven studies did not (21, 22, 24–28). The specific parameters of the rTMS interventions are shown in Table 2 .

The methodological quality of all included studies was evaluated using the Cochrane Collaboration’s techniques for assessing bias risk. All of the included trials described randomized allocation and were assessed to have low risk in this field. In allocation blinding, six studies were classified as having a low risk (20, 22, 24, 27–29), while three studies were classified as having a high risk (19, 21, 23). Two studies were evaluated to have an uncertain risk (25, 26). In the blinding of participants and personnel, eight studies were assessed to have a low risk (19–23, 25, 26, 29), with only one study assessed as having an unclear risk (27). The rest of the studies were assessed to have a high risk. In the blinding of outcome assessment, six studies were classified to have a low risk (19–23, 26), with only one study assessed as having a high risk (28). The rest of the studies had an unclear risk. For incomplete outcome data, seven studies were classified to have a low risk (19–22, 27–29), with the rest of the studies having an unclear risk. With regard to selective reporting bias, nine studies were considered to have a low risk (19–24, 26–28), while the remaining studies were considered to have an unclear risk. With regard to other biases, five studies were assessed to have a low risk of other biases (21–23, 26, 27), four studies were assessed to have an unclear risk (19, 20, 28, 29), and two studies were assessed to have a high risk (24, 25). These results are summarized in Figure 2 .

The results of this study were moderately heterogeneous. The sensitivity analysis was carried out using the item-by-item deletion method. After deleting one study (20) through sensitivity analysis, the intervention was rTMS combined with psychotherapy. The other studies did not use rTMS combined with psychotherapy. The combined results of the other studies showed a significant reduction in heterogeneity (I ² = 28%), but there was no significant difference in the overall combined results (MD = −2.58, 95%CI = −3.17 to −1.99, p < 0.00001), demonstrating that the findings of this study are more reliable.

The funnel plots ( Figure 4 ) and the Egger’s asymmetry test for sleep quality, an outcome metric including more than 10 studies, showed no publication bias (p = 0.884).