Types of studies

All included studies were observational. Our search didn't retrieve randomised controlled trials (RCTs), quasi‐RCTs, cluster‐RCTs, or controlled clinical trials (CCTs) of proton versus photon therapy. Given the small number of included studies and small study size, we did not report outcomes according to the tumour location. These criteria were set as the type of surgery and radiotherapy (RT; see Characteristics of included studies). We stated whether the included studies had incomplete or obvious confounders (see Assessment of risk of bias in included studies). We did not restrict the inclusion criteria to studies with at least three years follow‐up, although it has been reported that half of chordoma recurrences occur within the first three years (Weber 2016). Instead, we reported the follow‐up period for each study.

Types of participants

We included adult participants, aged 18 years or older, with pathologically‐confirmed primary chordoma, who were irradiated with a curative intent with protons or high‐energy (megavoltage) photons in the form of fractionated RT, SRS (stereotactic radiosurgery), SBRT (stereotactic body radiotherapy), or IMRT (intensity modulated radiation therapy). In the curative setting, a 60 Cobalt Gray Equivalent (CGE) dose or higher is typically used to achieve local control, and therefore, we restricted inclusion to people treated to at least this dose (Terahara 1999; Igaki 2004; Noel 2005). We included people who had undergone biopsy, subtotal, or gross resection.

We excluded people treated with brachytherapy.

All studies included participants who met the inclusion criteria, so we used the subset of data that was of interest to this Cochrane Review.

Types of interventions

Types of outcome measures

Electronic searches

We searched the following electronic databases:

• Cochrane Central Register of Controlled Trials (CENTRAL; 2021, Issue 4) in the Cochrane Library (searched 29 April 2021; Appendix 1);

• MEDLINE Ovid (1946 to week 16, 2021; Appendix 2);

• Embase Ovid (1980 to week 16, 2021; Appendix 3).

For databases other than MEDLINE, we adapted the search strategy accordingly.

Searching other resources

We searched the following electronic trial registries and databases to 29 April 2021.

• ClinicalTrials.gov (http://clinicaltrials.gov);

• International Clinical Trials Registry (http://apps.who.int/trialsearch);

• Current Controlled Trials (www.controlled-trials.com);

• National Comprehensive Cancer Network (NCCN) Guidelines (www.nccn.org/professionals/physician_gls/f_guidelines.asp);

• Turning Research into Practice (TRIP; www.tripdatabase.com);

• National Guideline Clearinghouse (www.guideline.gov).

We handsearched the following resources.

• Reference lists of relevant articles to identify additional publications;

• Conference proceedings of the:

  • American Society for Radiotherapy and Oncology (ASTRO; 1990 to April 2021);

  • European Society for Radiotherapy and Oncology (ESTRO; 1990 to April 2021);

  • Particle Therapy Co‐Operative Group (PTCOG; 1990 to April 2021);

  • North American Skull Base Society (NASBS; 1990 to April 2021).

Selection of studies

Four review authors (IS, DT, EL, SD) independently screened the title and abstract of all articles identified as a result of search, for inclusion. We obtained the full‐text articles of those identified as potentially eligible for inclusion, and the available review authors (IS, SD) independently assessed the full‐text articles. We resolved any disagreements by discussion or by consulting a third review author (DT). No potential conflict of interest existed (e.g. no authorship of a potentially included study). We listed all studies excluded after full‐text assessment and their reason(s) for exclusion in a Characteristics of excluded studies table. We illustrated the study selection progress in a PRISMA diagram (Figure 1).

Figure 1

---

---

Study flow diagram.

Data extraction and management

We extracted data on the following: publication data (first author's last name, year of publication, institution), study design, sample size, participants' features (mean age, sex ratio, race, country of the population studied, tumour location, type of surgery, the definition of extent of resection), details of interventions (indication for RT, RT technique, time interval of participant accrual, total dose, dose per fraction, target definition), clinical outcomes (outcome definitions, LC, OS, PFS rates, type of toxicity, toxicity rates, length of follow‐up). We considered studies in which at least part of the RT course used particles as particle studies.

For eligible studies, two review authors (IS and SD) independently extracted data using the data extraction form. If more than one publication referred to the same study, or there was more than one paper from the same institution, we used the most recent publication, or that with the most complete reporting. . We resolved any disagreements by discussion or consulted a third review author (EL or DT). We considered studies to have sufficient data if at least one of the listed primary outcomes, and at least one of the selected time points were reported.

None of the included studies required translation as they were all in English.

Assessment of risk of bias in included studies

We based the risk of bias evaluation on the data provided in the included studies. We did not retrieve any randomized controlled trials; all the included studies were observational. Hence, we used the Risk of Bias In Non‐randomized Studies of Interventions (ROBINS‐I) assessment tool (Sterne 2016). We assessed the risk of bias according to the following domains.

• Pre‐intervention bias: due to confounding based on referral pattern and departmental protocol eligibility criteria as well as bias in selection of participants into the study

• At‐intervention bias: in classification of the intervention

• Post‐intervention bias:

  • due to deviations from the intended intervention

  • due to missing data

  • in measurement of outcomes

  • in selection of the reported results

Each of the seven domains of bias contain signalling questions to facilitate judgements of risk of bias. The full signalling question and response framework for each outcome is provided in Sterne 2016.

Important confounders of interest in this Cochrane Review include the following:

• Histology (conventional, chondroid, de‐differentiated, poorly differentiated)

• Extent of resection

• Location (skull base, spine)

• Type of surgery (biopsy, complete, incomplete resection)

• RT total dose

• Timing of RT (prior to surgery, early after surgery (< 90 days), at relapse or progression (> 90 days after surgery)

We resolved any discrepancies about results by discussion and consensus agreement. We included all assessments in a risk of bias graph, and a risk of bias summary figure, in addition to risk of bias tables (Figure 2; Figure 3).

Figure 2

---

---

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 3

---

---

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Measures of treatment effect

Unit of analysis issues

We planed to take into account the level at which randomization occurred. We planed to extract the following information for cluster‐RCTs (Higgins 2011).

• The number of clusters (or groups) randomised to each intervention group

• The outcome data ignoring the cluster design for the total number of individuals

• An estimate of the intracluster (or intraclass) correlation coefficient (ICC)

However, we did not include cluster‐randomised trials, or trials in which participants received more than one intervention

Dealing with missing data

We did not impute for any missing data. We contacted the study authors to request more information. However, we were unable to obtain more information or missing data from study authors, and we conducted available‐case analysis.

Assessment of heterogeneity

We accounted for heterogeneity by visually inspecting the forest plots to detect overlapping confidence intervals, using the Chi² test and the I² statistic (Higgins 2011). The I² statistic represents the percentage of the overall variability due to between‐study (or interstudy) variability, as opposed to within‐study (or intrastudy) variability (Higgins 2011). We interpreted the I² value as follows: 0% to 40% might not be important; 30% to 60% may represent moderate heterogeneity; 50% to 90% may represent substantial heterogeneity; and 75% to 100% may represent considerable heterogeneity (Deeks 2001).

Assessment of reporting biases

We intended to investigate potential publication bias by using funnel plots, but we were unable to, because < 10 studies met the inclusion criteria of the review.

Data synthesis

We analysed the data using Review Manager 5 (Review Manager 2014). Chordoma of skull base and of the spinal axis were not analysed separately due to the small number of studies and small study sizes. We were unable to extract the HRs and their variances for survival‐type data from the articles, as they were not reported. We either estimated the log HR (experimental relative to control) by (O – E)/V, which has a standard error 1/√V, where O is the observed number of events in the experimental intervention, E is the log‐rank expected number of events in the experimental intervention, O – E is the log‐rank statistic, and V is the variance of the log‐rank statistic (Higgins 2011), and computed the log HR from individual patient data (Parmar 1998), from three studies (Park 2006; Takahashi 2009; Jagersberg 2017), or we depicted the log HR and standard error of log HR from the Kaplan‐Meier curve, using an Excel spreadsheet by Tierney 2007 for one study (Colli 2001).

Regarding treatment‐related toxicity, we used the RR as a measure of association, and fixed a higher type I error (α = 0.10, two‐sided (Shadish 2002)).

We meta‐analysed the HRs and RRs by using the generic inverse‐variance method. We used a random‐effects model, as studies were clinically and statistically heterogenous. (DerSimonian 1986).

Subgroup analysis and investigation of heterogeneity

There were too few included studies that reported sufficient details, so we were unable to conduct subgroup analyses for categorical covariates, or meta‐regression for categorical and continuous covariates, according to the following:

• Histology (conventional, chondroid, de‐differentiated, poorly differentiated)

• Location (skull base, spine).

• Age (continuous)

• Gender (male, female)

• Type of surgery (biopsy, complete, incomplete resection)

• RT total dose (continuous) and fractionation (dose per fraction, continuous)

• RT target volume (treatment margin used, if any, in centimetres, continuous)

• Timing of RT (prior to surgery, early after surgery (< 90 days), at relapse or progression (> 90 days after surgery)

Sensitivity analysis

We intended to explore reasons for study heterogeneity, and if necessary, carry out sensitivity analysis by omitting studies rated as high risk of bias' However, we only included a small number of studies, which did not enable us to conduct a sensitivity analysis.

Summary of findings and assessment of the certainty of the evidence

We presented the overall certainty of the evidence for each outcome according to the GRADE approach, which takes into account issues related to internal validity (risk of bias, inconsistency, imprecision, publication bias), and external validity, such as directness of results (Langendam 2013). We created a summary of findings table, based on the methods described in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), and using GRADEpro GDT software (GRADEpro GDT). We used the GRADE checklist and GRADE Working Group certainty of evidence definitions (Meader 2014). We downgraded the evidence from high certainty by one level for serious (or by two for very serious) concerns for study limitations.

• High‐certainty: we are very confident that the true effect lies close to that of the estimate of the effect.

• Moderate‐certainty: we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

• Low‐certainty: our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect.

• Very low‐certainty: we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Limited available evidence from a small number of included studies didn't permit us to present a table specific to each location.