2.1 Study design

Based on a non-systematic literature review of key publications on implementation of ACP (Jimenez et al., 2018; Lovell & Yates, 2014; Lund et al., 2015; Rietjens et al., 2017), relevant themes were identified and used to create open-ended questions for the first round of a modified online Delphi study. We consulted an expert panel over the course of three rounds between April and August 2021 using the online survey tool Unipark. The first round served to generate ideas for potential elements of a care pathway. To decide on inclusion or exclusion, elements were consecutively presented as written statements for numerical ratings (Round 2) and rankings (Round 3). Relevant results of the previous round were presented to the panel as a basis for decision-making. The panel was offered the opportunity to provide free-text comments to supplement and contextualise their ratings or elaborate their opinion, ultimately informing the research steering group's decision process. At the end of each round, panel members were asked to enter their email addresses if they were willing to participate in subsequent rounds of the study. The survey link was accessible for 3 weeks for the first round and 2 weeks for the following rounds.

Before the individual rounds were activated, the survey was checked for comprehensibility and user-friendliness by lay people and researchers who neither belonged to the core research team nor had in-depth knowledge on the topic of ACP.

Our research was conducted and is reported according to standards for Delphi studies proposed by Jünger et al. (2017).

2.2 Sampling and recruitment

The panel consisted of a convenience sample of German-speaking stakeholders including professionals with academic or practical expertise in ACP and a ‘public and patient involvement group’. Fifty-four experts with freely available contact details who spoke at the first German ACP congress (DiV-BVP, 2020) in Cologne in 2020 were contacted. We further reached out to 15 facilitators, trainers and academics in the field of ACP known to the research team and 69 cancer patients who previously participated in ACP research. Forty-four lead contacts for support groups for cancer patients were asked to forward the invitation to their members. Potential panel members were invited to participate by responding directly to a link provided in an email invitation. Additionally, we invited patients and carers through social media, more precisely via the public Instagram profile Let's talk about Krebs [@letstalkaboutkrebs] (2021) and open-access German Facebook groups for cancer patients and/or caregivers.

2.3 Data collection

In the first Delphi round, participants wrote free-text comments responding to explicit questions covering a range of aspects concerning the implementation of ACP for cancer patients in the hospital setting. In the following round, participants were asked to provide a rating on each element's usefulness according to a nine-level Likert rating scale (1 = not useful, 9 = very useful). If the results from Round 2 revealed conflicting alternatives or ambivalent answers, these were presented for ranking in the third round. Decision criteria for ratings and rankings, that is, adoption or rejection of an item, depended on participants' own definition. Based on the fact that not every member of our heterogeneous panel was familiar with all hospital processes, the answer option ‘beyond my expertise’ was also selectable.

Anonymity of the participants was maintained throughout all rounds, contact information could not be linked to the answers.

2.4 Analysis

For each round, we summarised demographic characteristics and personal information of panellists using counts (percentages) and means (standard deviations) as appropriate. A.P. and B.P. performed an inductive content analysis of responses extracted from Round 1 using MAXQDA 2020. In a first step, the two authors independently clustered responses of 20 data sets representing relevant ideas to codes through line-by-line review. Ensuring a structured approach, each code received a ‘code definition’. The response passages assigned to the codes were compared both within individual responses and across participants, and overarching themes were identified and combined into a code structure which was expanded in an iterative process to all data as new aspects arose. Based on these results, individual elements of the care pathway could then be converted into statements. To enhance rigour and confirmability, team meetings served to review and redefine statements before presenting them to the panel in the following rounds.

Ratings derived in the second round were analysed according to the method used by the Organization for Economic Cooperation and Development Health Care Quality Indicators project (Mattke et al., 2006). Owing to skewed data, we used the median as a measure of central tendency with ratings of 7 to 9 indicating support for an item, 4 to 6 ambiguity and 1 to 3 rejection.

2.5 Ethics

Ethical approval was obtained from the institutional review board of the Philipps-University Marburg (ID-No.: 200/20). The study was conducted in accordance with the Declaration of Helsinki, Good Clinical Practice guidelines and the EU data collection directive.

3.1 Participants

We conducted a three-round Delphi study involving a 41-member expert panel including 12 cancer patients, four carers and 25 professionals. Of these initial respondents, 70.7% completed Round 2 and 51.2% Round 3. In Rounds 2 and 3, consensus was reached with professionals who had an average of 14.4 (SD ± 12.6) to 14.6 (SD ± 5.4) years of experience in hospital work; 88.2% to 92.3% of them actively engaged in ACP discussions in the month prior to the survey with a mean of 3.7 (SD ± 6.2) to 4.8 (SD ± 5.4) discussions per month. Further participant characteristics for each round are provided in Table 1.

3.2 Consensus building

The qualitative analysis of Round 1 identified 10 essential domains for hospital-based ACP for cancer patients: prerequisites, organisation and coordination, identification and referral, provision of information, information sources, family involvement, ACP discussion, documentation, update and quality assurance. These domains were covered by 177 key statements representing potential elements of a care pathway. In the second round, 153 of these elements (86.4%) received support according to median ratings. After assessment of these elements by the research steering group, 10 were merged under two existing more generally phrased statements that also reflected them in content. Seven elements were presented again in Round 3 to clarify priorities between competing or opposing concepts. In the context of other supported statements, one element was considered to be out of the hospital's sphere of influence and was thus not integrated into the pathway. Ten elements (5.6%) were rated below the cut-off of 4 resulting in exclusion. Median ratings reflected ambiguity for 14 elements (7.9%). Four were excluded due to a competing element scoring a median greater than or equal to 7. Two elements could be directly integrated into the care pathway as they represented a natural consequence of other supported elements. Lastly, six were put to the vote again in Round 3 to receive a final decision on inclusion or exclusion. In order to better explore relationships between individual elements presented in Round 2, some new statements were drafted and presented for voting in Round 3. In this following round, 21 elements were ranked by the panel using nine categorical questions, of which the highest ranked could ultimately be included in the care pathway. The final care pathway, therefore, encompasses 148 elements.

In principle, the voluntary nature of participation was identified as a basic but necessary prerequisite. To ensure voluntary use of services, panellists offered several strategies that were incorporated into the care pathway. Furthermore, panellists considered it obligatory to secure funding without financial contribution by the patients or families. The final care pathway is shown in Figure 1a–c. For a full list of the included elements please see the Supporting Information.

4.1 Strengths and limitations

Despite sceptics of the Delphi technique arguing that its reliability and validity have not been confirmed (Hasson & Keeney, 2011; Keeney et al., 2001), the study design has been proven successful in earlier consensus research on ACP (Mohan et al., 2021; Sinclair et al., 2016; Sudore et al., 2017; Sudore et al., 2018). The choice of a modified Delphi method resulted from decisive advantages for the development of a care pathway. First, ACP for cancer patients, particularly in the hospital setting, is not routine in Germany, meaning that the direct experience of the experts in this area can certainly be considered limited. A key strength of our study lies in the structured consensus process involving professionals with clinical experience in hospitals and practical experience with ACP, so that a realistic assessment of the feasibility of the individual pathway elements can be assumed.

Second, for the purpose of our study, the traditional Delphi approach underwent some modifications. Instead of supplying participants with traditional hard copies of the survey, it was opted for a web-based survey to receive consensus on a care pathway. This methodology did not only accelerate and facilitate data collection and analysis but also increased dissemination covering a geographically spread sample. Most importantly, however, it ensured participant safety during the COVID-19 pandemic.

Third, a high level of agreement for the individual elements indicates that our pathway may be suitable for different hospital settings and patient groups.

However, our modified Delphi study also faces some noteworthy limitations. First, response rates for online questionnaires are known to be notoriously low (Dobrow et al., 2008; Koo & Skinner, 2005; Petrovčič et al., 2016) which may explain limited participation and high attrition in our sample.

Second, the representativeness of the sample could be impaired by the chosen design possibly entailing accessibility issues, for example, due to lack of internet access, poor internet skills, visual impairment, illiteracy or low educational level of representatives of patients and public.

Third, by incorporating nonprofessional people's perspectives into the development process, patients' views in particular, greater meaningfulness of the pathway could be achieved. Nevertheless, despite contacting different ‘patient and public involvement groups’, the ratio of patients and carers over professionals is lower as hoped, with a dominating proportion of physicians. While experienced professionals may have a good insight into patients' needs and clinical processes, there is a small risk of a normative perspective in the evaluation of the presented pathway elements. It is therefore essential to verify how the care pathway performs in clinical practice.

Lastly, our pathway was designed for the German healthcare system, so that general applicability is not necessarily given. Thus, further studies would be important to show how an implementation could be accomplished in other countries.