The systematic review was reported based on Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines[10]. The protocol for this systematic review is available in PROSPERO (CRD42020207243).
Search strategy:
We searched the PubMed, Embase, Web of Science, The Cochrane Library, CNKI, WanFang and CBM databases from the inception dates to July 14, 2021, using head and neck cancer, metronomic chemotherapy and relevant words as key words. An additional file showed in more detail [see Additional file 1]. Clinical.gov was also searched to ensure all the related studies published or ongoing being included. There were no publication date restrictions, and languages were restricted into Chinese and English.
Inclusion and exclusion:
Population:
Adult patients older than 18 years and diagnosed as head and neck cancer were included. Cancers of head and neck are further categorized in the area of oral cavity, pharynx (including oropharynx and hypopharynx), larynx, paranasal sinuses and nasal cavity, salivary glands[11]. Pathologic biopsy is a confirm diagnosis of cancer. Cancers of the brain, the eye, the esophagus, and the thyroid gland, as well as those of the scalp, skin, muscles, and bones of the head and neck, which are not usually classified as head and neck cancers, were excluded from the study[1].
Intervention:
Patients received metronomic chemotherapy were defined as intervention group, whether single agent or combined agents. Therapies except metronomic chemotherapy were all regarded as control treatments, including other chemotherapies, radiotherapy or without treatment.
Type of study:
We selected literatures related to head and neck cancer with metronomic chemotherapy. Studies published or ongoing, randomized or non-randomized design were all included. It was different from our strategy registered in PROSPERO, because randomized control trails were too less. Considering the possible bias resulting from the limited literatures included, we modified our inclusion criteria of study type by including cohort studies, single arm studies and retrospective studies. Records meeting the following conditions were excluded: 1. Trails not targeted to treat, such as determining the optimal dose or qualitative studies; 2. Literatures like comments or letters[12]. If there was any disagreement, we would consult a third member (CW). The primary studies included in systematic reviews and meta-analysis were reviewed, selected and extracted using criteria above.
Risk of bias assessment:
The methodological quality of randomized control trials was assessed by 2 researchers (TZ and J.B. W) independently based on Cochrane Risk-of-Bias criteria, which graded each quality item as low risk, high risk, and unclear risk[13]. The items used to evaluate bias including the random sequence generation, allocation concealment, blinding of participants and personnel, incomplete outcome data, blinding of outcome assessor, selective reporting and other bias[13]. The trials were graded as low quality, high quality, or moderate quality based on the following criteria: 1. Trials were considered low quality if either randomization or allocation concealment was assessed as high risk of bias, regardless of the risk of other items; 2. Trials were considered high quality when both randomization and allocation concealment were assessed as a low risk of bias, and all other items were assessed as low or unclear risk of bias; 3. Trials were considered moderate quality if they did not meet criteria of high or low risk[14].
The methodological quality of cohort studies was assessed by The Newcastle-Ottawa Scale (NOS). It awarded each item a maximum of one star in the selection and outcome categories while a maximum of two stars for comparability[15]. We considered one star as one point to quantify the outcomes for comparison.
The methodological quality of non-randomized control trials nor cohort studies was assessed using 11 items checklist recommended by The Agency for Healthcare Research and Quality (AHRQ)[16]. Each quality items were identified as yes, no, or unclear. We considered to score items answered “yes” for one point, and items answered “no” or “unclear” for zero. The quality of studies was graded as low quality for score of 0–3, moderate quality for score from 4 to 7 and high quality for score from 8 to 11.
Data extraction:
The following information from each study was extracted by two researchers (XL and TZ) independently using a predefined electronic data extraction form: lead author, publication year, country of origin, participant characteristics (number, age, tumor location), study design, treatments (regimen including agents, dose, course/procedure, single or combination agent) and outcomes referred above. A third author (LY) checked extracted data for accuracy. Disagreements were resolved by a group consensus. If the trials had more than 2 group, we only extracted the information interested[14].
Primary and secondary outcomes:
The primary outcome was overall survival (OS), which was measured from the date of randomization or research beginning to the date of death or last follow-up. Some studies gave the overall survival rate, defined as the number of patients who were still alive when last follow-up, usually 3 or 6 months.
The secondary outcomes included progression-free survival (PFS), disease-free survival (DFS), the incidence of adverse events, toxicity, progression, and so on. PFS was measured from the date of randomization or research beginning to the date of progression. DFS was defined as time interval from date of registration till date of recurrence or death of any causes. If trials gave both intention-to-treat (ITT) and per-protocol results, we only extracted ITT results for analyses.
Data synthesis:
Studies that met the inclusion criteria were synthetized using a narrative description because of the heterogeneity in cancer location, intervention (agents, dose, course of different regimens), duration of follow-up and different effect index. Besides, we performed a meta-analysis in Stata SE15 to calculate the hazard ratio (HR) for survival and HR for PFS, and 95% CIs using the Mantel-Haenzel statistical method.
The fixed-effects model was selected first, and statistical heterogeneity between data was evaluated using I2 statistic. If the heterogeneity was not accepted, then using random-effects model. Subgroup analysis was performed by different study designs, or different regimens. Sensitivity analysis was performed by excluding researches affected the stability of outcomes.