Types of studies

Randomised controlled trials (RCTs). We planned to exclude quasi‐randomised and cluster‐randomised trials.

Types of participants

We planned to include patients with HPV‐positive oropharyngeal carcinoma (subsites C09 and C10 as defined by the World Health Organization classification). Cancers included were T1‐4a with or without nodal disease with no evidence of distant metastatic spread. All patients had received minimally invasive transoral surgery.

We excluded carcinoma of the oral cavity (C01‐06), nasopharynx (C11), hypopharynx (C13) and larynx (C32) (WHO 2000).

We excluded patients receiving open surgery.

Types of interventions

Types of outcome measures

We planned to analyse the following outcomes in the review, but we did not use them as a basis for including or excluding studies. Primary outcomes focus on survival whilst the secondary outcomes focus on quality of life indices that may be affected by de‐intensification of treatment.

Electronic searches

We identified published, unpublished and ongoing studies by searching the following databases from their inception:

• the Cochrane ENT Trials Register (searched via the Cochrane Register of Studies to 26 April 2018);

• the Cochrane Central Register of Controlled Trials (CENTRAL) (searched via the Cochrane Register of Studies to 26 April 2018);

• Ovid MEDLINE(R) Epub Ahead of Print, In‐Process & Other Non‐Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R) (1946 to 26 April 2018);

• Ovid EMBASE (1974 to 26 April 2018);

• LILACS, lilacs.bvsalud.org (searched 2 April 2018);

• KoreaMed (searched via Google Scholar 27 April 2018);

• Web of Knowledge, Web of Science (1945 to 26 April 2018);

• ClinicalTrials.gov (searched via the Cochrane Register of Studies to 27 April 2018);

• World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP), www.who.int/ictrp (searched to 26 April 2018).

The Information Specialist modelled subject strategies for databases on the search strategy designed for CENTRAL. Where appropriate, they were combined with subject strategy adaptations of the highly sensitive search strategy designed by Cochrane for identifying randomised controlled trials and controlled clinical trials (as described in the Cochrane Handbook for Systematic Reviews of Interventions Version 5.1.0, Box 6.4.b. (Handbook 2011). The search strategy for CENTRAL is provided in Appendix 1.

Searching other resources

We scanned the reference lists of identified publications for additional trials and contacted trial authors where necessary. In addition, the Information Specialist searched Ovid MEDLINE to retrieve existing systematic reviews relevant to this systematic review, so that we could scan their reference lists for additional trials. The Information Specialist also ran non‐systematic searches of Google Scholar to retrieve grey literature and other sources of potential trials.

Selection of studies

Two authors (JH and LM) independently screened each abstract. We independently assessed full texts against the inclusion criteria. Any conflict was resolved through discussion with a senior author.

Data extraction and management

Two authors (JH and LM) planned to independently extract data using a specifically designed data extraction form. We planned to pilot the data extraction form on several studies and adjust it as necessary prior to use. Any disagreements were resolved through consultation with a senior author. Where required we contacted study authors for clarification or missing information.

For each study we planned to record the following:

• Year of publication, country of origin and source of study funding.

• Details of the participants, including demographic characteristics and criteria for inclusion and exclusion.

• Details of the type of intervention, timing and duration (including type of surgery).

• Details of survival outcomes reported, with time intervals.

• Details of treatment‐related morbidity, categorised as acute (less than 90 days after treatment) or late (more than 90 days) and classified according to the Common Terminology Criteria for Adverse Events (CTCAE 4.03).

• Details of all other outcomes reported, including method of assessment and time intervals.

Assessment of risk of bias in included studies

Had suitable studies been identified JH and LM would have assessed the risk of bias of the included studies independently, with the following taken into consideration, as guided by theCochrane Handbook for Systematic Reviews of Interventions (Handbook 2011):

• sequence generation;

• allocation concealment;

• blinding;

• incomplete outcome data;

• selective outcome reporting; and

• other sources of bias.

We planned to use the Cochrane 'Risk of bias' tool in RevMan 5.3 (RevMan 2014), which involves describing each of these domains as reported in the trial and then assigning a judgement about the adequacy of each entry: 'low', 'high' or 'unclear' risk of bias.

Measures of treatment effect

We planned to express continuous outcomes as a mean endpoint (or change from baseline) for each group with standard deviation and number of people. For continuous outcomes measured using different (but compatible) scales we planned to express treatment effects as a standardised mean difference (SMD).

We planned to express dichotomous outcomes as a risk ratio (RR) with the number of people with the outcome and number of participants.

We planned to express time‐to‐event outcomes as a hazard ratio (HR) with standard deviation (SD).

We planned to preferentially report ordinal data as continuous, however if studies only reported data dichotomously then we would have expressed as dichotomous.

The analysis would have been on an intention‐to‐treat basis. Where useful we planned to calculate the number needed to treat to benefit/harm (NNTB/NNTH) to aid clinical interpretation of the findings.

Unit of analysis issues

We planned to use data only from individually randomised controlled trials to avoid unit of analysis issues. We excluded cluster‐randomised trials.

Dealing with missing data

Where standard deviations or hazard ratios were not reported we planned to impute these from other reported data.

We planned to contact study authors:

• where a study protocol suggested that an outcome of interest had been measured but was not reported;

• if not all data required for meta‐analysis were reported;

• if standard deviation data or hazard ratios were not available or estimable from other reported data (using the methods detailed in the Cochrane Handbook for Systematic Reviews of Interventions (Handbook 2011)).

Assessment of heterogeneity

We intended to assess clinical heterogeneity by examining the types of participants, interventions and outcomes in each study. We planned to conduct formal assessment using the Chi² test (with a significance level of α = 0.1 in view of the low power of this test) and the I² statistic (with 75% or more indicating a considerable level of inconsistency), both available in RevMan 5.3 (RevMan 2014). If meta‐analysis was performed we planned to further assess heterogeneity by inspecting the overlap of confidence intervals for the results of individual studies within a forest plot.

Assessment of reporting biases

We planned to assess reporting bias as within‐study (outcome reporting) bias and between‐study (publication) bias.

Data synthesis

We planned to extract data from the included studies and enter the data into RevMan 5.3 for statistical analysis. In the event of incomplete data, we intended to contact the study authors to obtain further information and to seek statistical advice where necessary.

Our analysis of survival and disease recurrence would have depended on the data available. We aimed to analyse the proportion surviving at one, two, three and five years as either:

• proportion surviving; or

• hazard ratios, for comparison in meta‐analysis if appropriate.

We planned to express the proportion of people with a gastrostomy tube as a risk ratio. The remaining secondary outcomes were to be restricted to assessment of validated assessment tools (where appropriate). If the data provided were in the form of means and standard deviations, we intended to display the effects on outcomes as a mean with standard deviation. If there was disparity in terms of scales we planned to express the data as a SMD with 95% confidence interval (CI). If hazard ratios were not quoted in studies, we planned to calculate them from available summary statistics such as observed events, expected events, variance, confidence intervals, P values or survival curves (Parmar 1998). If required, we planned to analyse the survival curves using the online tool: https://automeris.io/WebPlotDigitizer/.

We hoped to attempt a meta‐analysis if studies were available with similar comparisons and reporting the same outcome measures. If appropriate, we intended to calculate pooled estimates using a random‐effects model (Handbook 2011), as there is likely to be significant statistical or clinical heterogeneity (an I² value > 50%, as specified in the Cochrane Handbook for Systematic Reviews of Interventions).

Subgroup analysis and investigation of heterogeneity

We had no planned subgroup analyses.

We planned to assess heterogeneity using the methods advised in the Cochrane Handbook for Systematic Reviews of Interventions (Handbook 2011).

Sensitivity analysis

If meta‐analysis had been performed we would have used sensitivity analysis. If we included studies with high risk of bias (e.g. poor follow‐up rate) we would have re‐calculated outcomes including these studies individually to see what influence this had on our presumed treatment effect.

Where thresholds had been set for inclusion or analysis we planned to re‐analyse the data using values either side of the set threshold to assess whether our decisions had influenced the outcomes.