4.1.1. Population

The assessment included studies with adult patients who underwent complex oesophageal interventions.

4.1.2. Volume of services

The VoS was defined as the number of performed complex oesophageal interventions per hospital, per physician, or per hospital-physician combination within a defined time period.

4.1.3. Outcomes

For the investigation, the following outcomes were examined:

• Mortality, such as
  • overall survival
  • intraoperative or perioperative mortality
  • inpatient mortality
• Morbidity, such as
  • disease-free survival
  • adverse effects of therapy, such as
    • anastomotic insufficiency
    • anastomotic stenosis
    • perioperative and postoperative bleeding
    • pulmonary complications
    • serious, life-threatening, or fatal infections
    • further serious treatment-related complications, if any
• Health-related quality of life, including activities of daily living and dependence on help from others
• Length of hospital stay

If usable data were found on other outcomes or validated quality indicators, they were permitted to be included as well.

4.1.4. Study types

Observational studies (e.g. cohort studies or case control studies) or controlled interventional studies were suitable for answering research questions 1 and 2.

For controlled interventional studies, the intervention to be examined was the specification of a minimum number of cases. Possible comparator groups were groups with a different or no specified volume.

4.1.5. Adjustment

In complex oesophageal interventions, the quality of the treatment outcome is materially influenced by individual risk factors such as the underlying disease, type of procedure, comorbidities, and complication management. Further indication-specific risk factors are also possible.

Therefore, control of relevant confounders (risk adjustment) was a prerequisite for study inclusion. Control was assumed to exist if the study analysis involved suitable statistical methods to adjust for relevant confounders in an effort to address the problem of potential structural inequalities (unfair comparisons) between hospitals or treatment providers (physicians, nurses, etc.) with high and low VoS.

Likewise, cluster effects (e.g. greater similarity of outcomes in patients within the same hospital versus patients from different hospitals due to hospital-specific characteristics) had to have been taken into consideration by means of adequate statistical methods.

4.1.6. Study duration

There were no restrictions regarding the study duration.

4.1.7. Publication period

In accordance with the commission, studies with a publication date of January 2000 or later were included in the study.

4.1.8. Transferability

To ensure the transferability of study results to the German healthcare system, studies from European countries as well as the USA, Canada, Australia, and New Zealand were eligible for inclusion.

For international studies, at least 80% of the data had to come from the above countries.

4.1.9. Tabular presentation of the criteria for study inclusion

The tables below list the criteria which had to be met by studies included in the assessment.

Table 1

Overview of inclusion and exclusion criteria for controlled interventional studies

Table 2

Overview of inclusion and exclusion criteria for observational studies

4.1.10. Inclusion of studies which do not fully meet the above criteria

In accordance with IQWiG General Methods Version 5.0, Chapter 9 [30], for inclusion criteria I1.1/I2.1 (population), I1.2 (use of a minimum number of cases), and I1.3 (comparator intervention with respect to the study’s comparator group), and/or I2.2 (VoS), and I1.9/I2.8 (transferability), it sufficed if at least 80% of included patients fulfilled these criteria. For such studies, subgroup analyses, if any, on patients who fulfilled the inclusion criteria were used. Studies in which inclusion criteria I1.1/I2.1, I1.2/I2.2, and I1.3 as well as I1.9/I2.8 were fulfilled by fewer than 80% of patients were included only if subgroup analyses were available for patients who did fulfil the inclusion criteria.

4.2.1. Focused information retrieval to search for systematic reviews

In preparation of the comprehensive information retrieval, a search for systematic reviews was conducted in the databases of MEDLINE and Cochrane Database of Systematic Reviews as well as on the websites of the National Institute for Health and Care Excellence (NICE) and the Agency for Healthcare Research and Quality (AHRQ). The search was restricted to publication dates of January 2000 or later.

The search strategies for the search in bibliographic databases are found in Appendix A. The search was conducted on 28/08/2019.

The final decision as to which systematic review(s) met the report’s inclusion criteria was taken after completing the project outline.

4.3.1. Sources of information

For the comprehensive information retrieval, a systematic search was conducted for relevant studies or documents in accordance with IQWiG General Methods Version 5.0, Chapter 8 [30]. The following primary and further information sources as well as search techniques were considered:

• Primary information sources
  • Bibliographic databases
    • MEDLINE
    • Embase
    • Cochrane Central Register of Controlled Trials
• Further information sources and search techniques
  • Use of further search techniques
    • screening of reference lists of systematic reviews found (see Section 4.2.1)
  • Requests to authors

4.3.2. Selection of relevant studies

4.4.1. Presentation of the individual studies

All information needed for the investigation was extracted from the documents regarding the included studies and entered into standardized tables. Any discrepancies found in connection with the comparison of information from different documents or from multiple data points within the same document, provided such discrepancies had the potential of considerably influencing the interpretation of results, are presented in the results section of the report.

Results were typically omitted from the investigation whenever they were based on fewer than 70% of the patients to be included in the analysis, that is, whenever more than 30% of patients were excluded from analysis.

Results were also omitted from the investigation whenever the percentage of patients excluded from analysis differed by more than 15% between groups.

Whenever the studies’ authors used several statistical models and justified their choice of a preferred model for their underlying data, the statistical model preferred by the authors was used so long as the model fulfilled the conditions defined in Section 4.1.5. Whenever several models were appropriate for the underlying data, the simpler model was used, taking into account Section 4.1.5.

4.4.2. Assessment of the informative value of results

The informative value of the results from the included observational studies was assessed on the basis of quality criteria developed especially for studies assessing volume–outcome relationships [31-33]. In terms of the informative value of results, the assessment considered the way the risk adjustment was performed, i.e. the risk factors taken into account and the sources used (administrative databases, clinical databases, medical records). Likewise, the quality of the statistical models used to examine the relationship between VoS and outcome was assessed; said quality depends on the form in which the volume attribute was entered into the analysis (continuous versus categorical data), on the consideration of cluster effects (see Section 4.1.5), and on the examination of model quality [34]. The completeness of reporting (e.g. description of analysed data and reporting of point estimates, confidence intervals, and p-values) was likewise considered an aspect impacting the informative value of results. Based on the entirety of these quality criteria, the observational studies were categorized by quality into those with high versus low informative value of results.

4.4.3. Assessment of the risk of bias

The risk of bias of the results of the included controlled interventional studies was assessed in accordance with General Methods Version 5.0, Chapter 9 [30].

4.4.4. Summary assessment of information

The results on the outcomes reported in the studies were comparatively described in the report.

Since categorical analysis is associated with a loss of information (e.g. the linearity assumption may be violated within the individual categories) and might deliver less reliable results than continuous analysis [33], results of continuous modelling were preferred over results from categorical modelling and included in the report, provided that potential non-linear relationships were adequately taken into account in continuous modelling. However, if the studies presented results exclusively from categorical analysis or only the results from categorical analysis were usable, the summary assessment relied on categorical analyses.

Where possible, beyond the comparison of results from the individual studies, suitable metaanalytical methods were used [30]. A final summary assessment of the information was performed in any case. Where possible, results reported on subgroups (e.g. intervention-specific analyses) were presented separately and summarized.