An academic librarian systematically designed and executed comprehensive, database-specific search strategies for seven major biomedical databases: Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, Web of Science, EMBASE, Chinese National Knowledge Infrastructure (CNKI), VIP Database for Chinese Technical Periodicals, and Wanfang Data. Our systematic search encompassed all available records from each database's inception through October 28, 2024 and updated to August 03, 2025, without imposing language or publication status limitations. We also searched the previous systematic reviews and screened the reference lists and the studies included (Supplementary Table 1).

Pairs of reviewers (HZY, LXX, CZY, ZWY) independently screened titles, abstracts, and subsequently, the full texts of potentially eligible articles using standardized, pre-tested forms. The data extraction form was structured around the PICOS framework, covering participant characteristics (Population), detailed descriptions of the interventions and comparators (Intervention/Comparison), study design (Study), outcome measures (Outcome), along with risk of bias assessments and records of adverse effects (see Supplementary File 1).

Disagreements primarily concerned the applicability of the interventions or the certainty of outcome reporting in the full-text articles assessed. All disagreements were referred to the arbitrator (YDN). The arbitrator made the final decision by referring to the predetermined inclusion criteria outlined in the PICOS framework and based on the original article text.

We included trials that met the following criteria: (1) enrolled patients diagnosed with primary dysmenorrhea; (2) randomized participants to receive localized superficial heat therapy, defined as the application of any device or substance (e.g., electric heating pads, adhesive abdominal warmers, far-infrared belts, or moxibustion) aimed at transferring thermal energy continuously to the body, vs. a control (no treatment, placebo, or NSAIDs); (3) evaluated outcomes either in the immediate term (≤24 h) or analgesic effect or over the longer term (≥3 months) for repeated-use efficacy; and (4) reported measures of pain intensity or safety endpoints.

Four reviewers (YDN, LYY, HZY, CZY) extracted data from each eligible trial sequentially, ensuring they faced away from each other during the process. We gathered information on study characteristics, including author name, year of publication, study location, funding source, sample size, and length of follow-up, as well as intervention characteristics and all patient-important outcomes.

In cases where a study reported outcomes at multiple time points, we selected the most commonly reported follow-up period among the eligible trials. To account for within-person variability, we abstracted change scores from baseline; end scores were used only when change scores were not available. Additionally, when multiple instruments or questionnaires were employed to measure a common outcome (such as pain), we abstracted data solely for the most frequently used instrument across the eligible studies.

Three reviewers (HZY, LYY, CZY) independently assessed the risk of bias using a modified Cochrane Risk of Bias Tool 1.0 (14, 15). The tool assessed the following domains: random sequence generation; allocation concealment; blinding of study participants, healthcare providers, and outcome assessors; incomplete outcome data (≥20% missing data was considered high risk of bias); and other potential sources of bias. For each item, responses were scored as “definitely or probably yes” (low risk of bias) or “definitely or probably no” (high risk of bias). Disagreements between reviewers were resolved through discussion and, if necessary, by third-party adjudication (see Supplementary Table 2).

For dichotomous outcomes, we calculated the relative risk (RR) and its corresponding 95% confidence interval (CI). For continuous outcomes, we calculated the weighted mean difference (WMD) and its corresponding 95%CI after we converted all the pain intensity data to the 10 cm visual analog scale (VAS) for pain (16). 1.5 cm was considered the minimal clinical important difference (MID) of pain intensity (17). We calculated the modeled risk difference (RD) value for comparisons to make the results easier to be understood.

We used a DerSimonian-Laird random effects model for all meta-analyses. Data were analyzed with STATA software version 17 (Stata Corp, College Station, TX, USA).

We evaluated the certainty of evidence for all outcomes using the GRADE framework (Grading of Recommendations, Assessment, Development, and Evaluations) (18). Evidence from randomized controlled trials (RCTs) is initially rated as high certainty but was subject to downgrading by one or more levels (to moderate, low, or very low) following assessment across five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias—the latter evaluated through visual inspection of funnel plot asymmetry where ≥10 studies contributed to a meta-analysis. We defined it imprecise when 95% CIs of pain intensity contained either half MID (0.75 cm) or 0 cm, and 95% CIs of adverse events included no difference (RR = 1).

We screened 2,733 citations, identifying 57 eligible trials (19–75) involving 5,359 participants (search flow shown in Figure 1). The median of the mean ages reported across the 53 trials (19–25, 27–47, 49–57, 59–74) that provided age data was 22.3 years. Among the 44 trials (19–24, 27, 28, 30–34, 37–44, 47, 50–55, 57–64, 66, 68–74) reporting the duration of primary dysmenorrhea, the median of the mean durations was 44 months. Of the studies reporting location, 54 were conducted in Asia (19–28, 30–44, 46–74), two in Europe (29, 75), and one in South America (45). Twenty-eight trials (24–26, 29–34, 38, 40, 42, 45, 46, 48, 49, 52, 53, 55, 56, 59, 61, 64, 66–68, 73, 74) compared heat therapy with a control group [three (26, 29, 45) assessing short-term effects], and 24 trials (21–23, 27, 29, 35, 37, 39, 41, 44, 47, 50, 51, 54, 57, 58, 60, 62, 65, 69–72, 75) compared heat therapy with NSAIDs [two (29, 60) assessing short-term effects] (see Table 1).

The risk of bias assessment for the 57 included trials is summarized in Supplementary Table 2. Random sequence generation was adequately reported in 36 trials (63%) (21, 27–29, 31–35, 39–42, 44, 45, 47, 49–53, 55–58, 60, 62, 64–69, 71, 72, 74), suggesting a low risk of selection bias for this domain in these studies. However, allocation concealment was implemented in only 17 trials (30%) (21, 27–29, 31–35, 39–42, 45, 52, 55, 74), potentially compromising 1.5 cm was considered the minimal clinica integrity. Only 3 trials (5%) (45, 52, 74) blinded participants, and 3 (5%) (28, 40, 45) blinded healthcare providers. This high risk of performance bias means that the expectation of receiving a therapeutic intervention (heat) could have influenced participants' reporting of pain relief. Similarly, blinding of outcome assessors and data analysts was reported in only 6 trials (11%) (28, 33, 35, 40, 41, 52), constituting a significant source of detection bias for the subjective outcome of self-reported pain. Importantly, no trials had ≥20% missing data, which minimizes bias from incomplete outcomes and strengthens the robustness of the pooled analysis.

Compared to no treatment, heat therapy reduces pain in primary dysmenorrhea with comparable safety. When compared to NSAIDs, heat therapy demonstrates minimal difference in pain intensity but is probably associated with fewer adverse events. These treatment outcomes remain consistent across both short-term (24-h) and long-term (3-month) assessments.

We have identified two systematic reviews in the literature addressing heat therapy for primary dysmenorrhea (76, 77); however, we excluded 6 RCTs for specific reasons. The first meta-analysis (76) included three RCTs on thermotherapy. One trial was excluded due to a lack of baseline data (78). The other two (79, 80), with treatment durations of 1 and 2 months, were also excluded. The second meta-analysis (77) included six RCTs (29, 78, 80–83), only one of which was included in our analysis (29). The other five trials were excluded due to the absence of extractable outcome measures (78, 80, 81), unavailable resources (82), or non-compliant interventions (83). A detailed breakdown is provided in Supplementary Table 3.

Earlier systematic reviews offered valuable preliminary insights by suggesting heat therapy might be effective and potentially comparable to analgesic medication. However, their conclusions were notably constrained: Jo and Lee's analysis, while indicating superiority over placebo, was limited to only 6 RCTs (77); Igwea et al. identified merely 3 heat therapy trials, were unable to perform a direct comparative meta-analysis, and ultimately highlighted the need for more robust evidence (76).

Our study comprehensively addresses these limitations through key advancements: a markedly expanded evidence base (57 RCTs) enabling more precise and generalizable treatment estimates; broader intervention diversity encompassing microwave therapy, electromagnetic wave therapy, moxibustion, and hot packs beyond previous narrow focus; and demonstration of consistent therapeutic benefits across both immediate (24-h) and sustained (3-month) timeframes—a previously unexamined dimension. Methodologically, the systematic application of the GRADE framework provides rigorous evidence certainty assessment, thereby substantiating prior hypotheses and establishing a more reliable foundation for positioning heat therapy as a viable non-pharmacological treatment for primary dysmenorrhea.

This review has several strengths. We predefined the MIDs to visually demonstrate between-group differences and evaluate the clinical significance of the findings. The calculation of MID-derived risk differences (RDs) further enhanced the interpretation of the clinical feasibility of treatment effects. Moreover, our study overcomes the limitations of previous analyses by comprehensively incorporating diverse thermotherapy modalities and leveraging a robust sample size, thereby yielding more precise and generalizable estimates.

Nevertheless, several limitations warrant consideration. The methodological quality of many included trials was compromised by inadequate randomization and concealment of allocation. Furthermore, significant heterogeneity was observed in some analyses, likely stemming from clinical diversity in patient populations and variations in heat therapy protocols. Finally, the predominance of studies conducted in Asian populations may limit the generalizability of our findings to other regions.

Our findings provide evidence for informing a stepped-care approach to managing primary dysmenorrhea. During acute episodes, local heat therapy using modalities such as hot water bottles or self-heating patches can provide immediate pain relief comparable to NSAIDs, with a superior safety profile. This offers an ideal first-line option for patients who cannot or prefer not to use medication. During the intermenstrual period, regular application of heat therapies like moxibustion or infrared therapy can serve as an effective preventive measure. Long-term adherence may reduce the frequency and intensity of pain episodes and decrease reliance on analgesic medications. For patients with severe pain, a “heat therapy-first, medication-as-supplement” combination strategy could be considered—employing heat therapy both preventively and during acute phases, reserving short-term NSAID use only for peak pain levels to optimize both efficacy and safety.

From a research perspective, the application of MIDs in our meta-analysis offers a concrete method for evaluating the clinical significance of future findings. However, the promising results are constrained by the low certainty of evidence and prevalent risk of bias in existing studies.

Future research should therefore prioritize high-quality, adequately powered RCTs that are specifically designed to overcome these limitations. We recommend that future trials: (1) calculate sample sizes based on the established MIDs for pain scales to ensure sufficient statistical power; (2) predefine and consistently apply a standardized heat intervention protocol (specifying temperature, application site, duration, and treatment frequency) to reduce heterogeneity; (3) adhere to the CONSORT reporting guidelines, providing clear descriptions of randomization, allocation concealment, and blinding methods; and (4) systematically record and report all adverse events to better establish the long-term safety profile of repeated heat application. Such rigorously generated evidence is crucial to confirm these findings and establish clear, evidence-based clinical guidelines.