Eligibility criteria

In this network meta-analysis, we will include participants diagnosed with chronic cancer pain, according to the criteria defined by the International Association for the Study of Pain (IASP) [1]. While it is generally accepted that chronic pain is defined as pain lasting longer than three months, the precise concept of time is not stressed in cancer pain because cancer patients may experience rapid disease progression and shortened survival times. Therefore, we will define chronic cancer pain is a kind of chronic pain caused by the primary cancer itself or metastases (chronic cancer pain) or its treatment (chronic post-cancer treatment pain) [1].

We will include randomized controlled trials (RCTs) that feature a control group for efficacy comparison. The control group can receive a placebo, conventional treatment, or any intervention different from the primary treatment method. If a study lacks a control group, it must demonstrate that the target intervention has higher efficacy compared to other interventions applied in the sample.

The studies must employ a randomized controlled trial design to assess the effectiveness of both pharmacological and non-pharmacological treatments for chronic cancer pain patients. Non-pharmacological interventions must not impose any pharmacological effects on participants.

Participants must be adult males or females, at least 18 years old, experiencing pain caused by cancer. Both race and gender will be considered independent variables in the research.

Exclusion criteria

We will exclude randomized controlled trials (RCTs) where participants primary health condition is not chronic cancer pain or where the studies do not retain complete study data. Studies will be excluded if they do not meet the following criteria: they do not have a control group that received a different treatment from the one being studied; they do not demonstrate substantial efficacy of the target therapy compared to the other therapies included in the study; they include individuals who are awaiting a biopsy or a diagnosis; they conduct pain research specifically focused on diseases other than cancer; they have too broad inclusion criteria that include participants who do not meet the requirements of this study.

Interventions

In this study, non-pharmacological intervention will refer to the implementation of management measures that do not rely on the administration of nutritional supplements or medicines. We have established a predetermined set of intervention measures (S4 File) derived from previously published RCTs, systematic reviews, and clinical practice guidelines. The intervention measures we will be interested in are the following: pharmacological therapies [8] including nonsteroidal anti-inflammatory drugs (NSAIDs) such as aspirin, acetaminophen, diclofenac, indomethacin, quinolone, ibuprofen, naproxen, flurbiprofen esters, piroxicam, meloxicam, paclitaxel, and celecoxib; weak opioids such as codeine, oxycodone, and tramadol; and strong opioids such as morphine hydrochloride, morphine sulfate, remifentanil, pethidine, buprenorphine, and oxycodone. Non-pharmacological therapies include acupuncture [24–27], cognitive behavioral strategy [28], mindfulness and meditation [29–31], psychological therapy [32], spiritual intervention [33], electrotherapy [34,35], hypnosis [36,37], massage therapy [24,38,39], music therapy [28,40,41], telephone-based interventions [42], yoga, Tai Chi, and Qigong [28,43], and complementary therapies [44]. Each intervention mentioned above controls for completely different confounding factors, so we will create three different control groups that include “placebo & sham”, “no intervention & waiting list”, and “standard drug treatment”. To ensure that various interventions do not violate the assumption of transitivity for the network meta-analysis (NMA), we will define the “standard drug treatment” as following the three-step pain ladder characterized by nonsteroidal anti-inflammatory drugs, weak opioids, and strong opioids. In this network meta-analysis, comparisons of interventions will be made with each other. All control groups in the studies we will include will be assigned to these three groups.

Outcomes

The results will be categorized into primary and secondary outcomes. We will include studies that assessed at least one of the following indicators:

Study selection and data extraction

After the literature search, the results will be exported to the reference manager Endnote X9 software, where any duplicate studies will be removed. Two independent reviewers, W.H. and X.L., will then examine the titles and abstracts and evaluate any potential full-text material. Studies that meet the eligibility criteria will be included in the review. If necessary, the authors will be contacted via email for clarification, with emails sent every Monday and Friday for up to a month. If the authors fail to respond, the study will be excluded, and the reasons will be documented in the PRISMA flowchart. Any disagreements between the two reviewers will be resolved by a third reviewer, Y.R. We will use Excel 2023 to record the data of the included studies and Review Manager 5.3 to conduct a comprehensive analysis of the extracted data. The PRISMA flowchart is available in the attached file (S1_Fig).

For data extraction, two independent reviewers (W.H. and X.L.) will use a pre-designed sheet, with any disagreements resolved by Y.R. The extracted data will include information on study methods, such as country, first author, study design (parallel group, crossover, or cluster-randomized control trial), number of research centers, and year of publication. Participant details will cover sample size, average age, gender, duration of illness, baseline pain score, comorbities, medication use, diagnostic criteria, inclusion and exclusion criteria, randomization, medication discontinuation in each group, and whether the data were analyzed according to the intention-to-treat principle. For the intervention, we will record the name, dosage, intervenor, duration, frequency, follow-up period, and main characteristics. Results will include primary outcomes such as pain intensity, total effective rate of treatment, onset of analgesia, and quality of life, as well as secondary outcomes like adverse reactions. We will also note the measurement tools and times, along with the language of publication and any contact with authors for further information.

For continuous variables such as pain intensity, total effective rate of therapy, onset time of pain alleviation, and quality of life, we will collect the mean, standard deviation (SD), and sample size for each group. For discrete data, we will calculate the variance, confidence interval, and sample size for each group. Data will be collected at three time points: 24 hours, 24 hours to 4 weeks, and 4–12 weeks after the intervention. If multiple time points are reported, the time point closest to the end of the intervention will be prioritized. If intervention scores are not available, we will extract the mean changes from baseline and their SDs. Additionally, we will record the number of individuals who completed therapy, those who abruptly discontinued treatment, and those who responded effectively to treatment. If the number of withdrawals is not reported, we will use the figures for the number of effective treatments and the total number of participants to compute the overall treatment efficacy rate.

In cases where the SD is not provided, we will estimate it using available data such as the range value, interquartile range, p-value, standard error, or confidence interval. If the necessary data cannot be inferred, we will contact the study authors via email, sending up to three emails with a seven-day interval between each. If the authors do not respond, the study will be included in the review but excluded from the quantitative analysis. For crossover randomized controlled trials (RCTs), we will only include data from the initial randomization period.

Risk-of-bias in the included studies

Two independent trained reviewers (W.H. and X.L.) will evaluate the included trials’ methodological quality. The following 6 domains will be assessed: (1) bias resulting from the randomization process,(production of randomized allocation sequences, use of allocation concealment) (2) bias due to deviations from intended interventions,(blinding of patients and treating physicians) (3) bias due to missing outcome data,(blinding of the outcome indicator evaluator) (4) bias in the measurement of the outcome,(completeness of outcome data, number and causes of dropouts in various intervention groups, differences in subjects with and without outcome data, handling of missing values) and (5) bias in the selection of the reported result (selective reporting of outcome measures). (6) Other biases: funding sources, and baseline balance between the intervention group and control group.

In cases of disagreement between the two reviewers, an authoritative third reviewer (Y.R.) will make the final decision.

The procedures, steps, team, and tasks of the network meta-analysis review to which this protocol refers are shown in S2_Fig.

S2_Fig Picture chart of the protocol for the network meta-analysis steps. Created at 2024 Biorender, online software, Hangzhou, China.

Data synthesis and analysis

We will use ADDIS v1.16.8 software for meta-analysis and heterogeneity assessment. For dichotomous data, we will analyze using odds ratios (OR); for continuous variables, we will use mean differences (MD). A 95% confidence interval (CI) will be used to estimate the range of values. The Q test will be used to detect heterogeneity among studies, and the I² statistic will be used to assess the degree of heterogeneity. If I² ≥ 50% and p ≤ 0.10, a random-effects model will be used for analysis. Conversely, if I² < 50% and p > 0.10, a fixed-effects model will be used.

We will use the GeMTC package in R software to conduct Bayesian network meta-analysis (NMA) of the selected randomized controlled trials (RCTs). The node-splitting model (NM) will be used to check the consistency of the evidence network. If the observed differences are not statistically significant (P > 0.05), a consistency model (CM) will be used for the NMA; if the differences are statistically significant (P < 0.05), an inconsistency model (IM) will be used. We will run 50,000 simulation iterations and 10 burn-in iterations, and use the potential scale reduction factor (PSRF) to assess model convergence. When the PSRF value is close to or equal to 1, it indicates strong model convergence and reliable consistency model results. The surface under the cumulative ranking curve (SUCRA) will be calculated to determine the effectiveness of each intervention. SUCRA values range from 0 to 1, with a value of 1 indicating complete effectiveness and a value of 0 indicating no effectiveness. Interventions will be ranked based on their SUCRA values.

For dichotomous outcomes, we will use odds ratios (OR) to measure the relative effects between different interventions. For continuous outcomes, we will use standardized mean differences (SMD). We will use 95% confidence intervals (CI) to represent the estimated range of effect sizes. If the 95% CI does not include the null value (e.g., 1 for OR, 0 for SMD), the difference will be considered statistically significant. Additionally, we will use a p-value <0.05 as the criterion for statistical significance.

We will use the CINeMA framework to assess the credibility of the NMA results. CINeMA extends the GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) approach by considering six domains to evaluate the certainty of evidence: within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and incoherence. Evaluating the certainty of evidence through these methods will enhance the transparency, reproducibility, and reliability of the results. Two authors (W.H. and X.L.) will independently judge the quality of evidence (high, moderate, low, or very low), and any disagreements will be resolved by a third reviewer (Y.R.).

Risk of bias across studies

To assess the presence of publication bias or small sample effects, a corrective funnel plot will be created for outcome measures having a literature volume of 10 or above.

We will use the CINeMA framework to assess the credibility of the NMA results. CINeMA extends the GRADE (Grading of Recommendations, Assessment, Development, and Evaluations) approach by considering six domains to evaluate the certainty of evidence: within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and incoherence. Evaluating the certainty of evidence through these methods will enhance the transparency, reproducibility, and reliability of the results. Two authors (W.H. and X.L.) will independently judge the quality of evidence (high, moderate, low, or very low), and any disagreements will be resolved by a third reviewer (Y.R.).

Publication bias assessment

We will use a calibrated funnel plot analysis to assess publication bias. If the funnel plot is asymmetrical, we will further investigate potential sources of publication bias.