Towards an international taxonomy of integrated primary care: a Delphi consensus approach

Two separate modified Delphi studies were conducted to: 1) investigate the appropriateness of the key features and, 2) test the categorisation across the domains of the RMIC. The modified Delphi methodology is a research technique designed to obtain opinions from experts through the use of a self-administrated questionnaire (Round 1) and of a physical meeting of experts (Round 2) to discuss the ratings of Round 1 [7]. Given the polymorphous nature of integrated care [4], the physical meetings enabled the experts to clarify each other’s perspectives on integrated care [8].

For each international Delphi study, a purposive sampling strategy was used to identify experts with experience in practice, science or policy in the deployment of integrated service models in a primary care setting. Potential experts were nominated using the following criteria: 1) scientific (performing research) experience and/or 2) practical (working in a professional or service organisation) experience regarding the organisation of integrated primary care. We attempted to balance the number of potential experts from each category in order to ensure each was represented in both studies. We did not include experts with an explicit policy or insurance background or patients in order to minimalize conflict of interest during the face-to-face meeting of the second Delphi round [6]. For example, the presence of policymaker or health insures could influence the (strategic) behaviour of the practice experts, as they are (financially) dependent of these stakeholders for the continuity of their practices. Experts who met the selection criteria were e-mailed an invitation with information on the research aims and details of the Delphi study. Only those who agreed to participate in both the initial online self-administrated questionnaire and the second face-to-face meeting were included. Experts who already participated within one of the previous Delphi studies were excluded.

The first international Delphi study was conducted to test the appropriateness of the 59 features at an international level in order to augment the research we conducted on the appropriateness of the features in the previous national Delphi study. Thirty-seven international experts from an a priori list of participants of the World Congress on Integrated Care (2013) in Singapore [9] were approached by e-mail and invited to participate. During Round 1, the experts received a link to an online questionnaire. They were asked to rate the appropriateness of each feature to achieve integrated primary care on a nine-point Likert-scale, ranging from 1 (completely irrelevant) to 9 (extremely relevant). Experts were asked to comment on any of the features, to suggest possible rephrasing and to highlight any features that may have been missed in the initial list. Two e-mail reminders were sent to non-responders.

In Round 2, a face-to-face meeting of the expert panel took place during the World Congress on Integrated Care. The meeting was chaired by one member of the research team (HV). The goal of the meeting was to discuss the results of Round 1 and, after the discussion, to reassess the value of the appropriateness of the features to achieve integrated primary care. Based on the results of Round 1, a summary report was provided to the experts at the meeting with the following key feedback information for each feature: 1) respondents’ own ratings, 2) median agreement rating, 3) summary of qualitative comments, and 4) whether a consensus was achieved at Round 1. Because of time, during the second round, we decided to only discuss the features that were not considered appropriate in the first round.

To begin the discussion over a disputed feature value, panel members who had rated the disputed feature with either its highest and lowest scores were asked to clarify their considerations. Next, a short discussion among all group members took place to explore if differences were due to real disagreement or to a misunderstanding or misinterpretation of the feature [8]. Finally, the experts were asked to individually rate the feature once again on their summary report. During the discussion, notes were taken by two observers.

A second international Delphi study was conducted to: 1) refine the descriptions of the domains of the RMIC and 2) categorise the features under one of the domains of the RMIC. Only key features that were considered appropriate within one of the previous Delphi studies (national and international) were included in the final Delphi study. Thirty-six experts from an a priori list of participants at the 14^th International Conference on Integrated Care (2014) in Brussels [9, 10] were invited by e-mail. In Round 1, the experts received a link to an online questionnaire and were asked to categorise each feature into one of the domains of the taxonomy: 1) clinical integration, 2) professional integration, 3) organisational integration, 4) system integration, 5) functional integration, 6) normative integration, and 7) person-focused and population-based care. Two e-mail reminders were sent to non-responders.

In the second round, a face-to-face expert panel meeting took place. A member of the research team (MB) chaired this meeting and facilitated the panel discussion. Based on the results of Round 1, a summary report was provided to the experts with the following key feedback information: 1) respondents’ own categorisation of each feature, 2) whether consensus was achieved regarding the categorisation in Round 1, and 3) a summary of qualitative comments. First, an iterative group discussion was conducted about the descriptions of the seven domains of the draft taxonomy. Second, the features that did not result in a consensus regarding their categorisation were discussed. Next, a short discussion among all group members took place. Finally, the experts were asked to once again individually categorise the features on which no consensus was reached in the first round. During the discussion, notes were taken by two observers.

Based on the results of the previous and current Delphi studies, a final version of the taxonomy was constructed. The research team synthesised the results and comments provided by the experts to produce a final taxonomy of features. To be initially included, features had to meet the following sequential eligibility criteria: 1) features had to be considered appropriate in Delphi study 1 as well as the national Delphi study and 2) a consensus had to be reached regarding categorisation based on the results of Delphi study 2 (see Fig. 2). Thereafter, three authors (PV, MB and IB) independently assessed the compiled taxonomy of features. To ensure a comprehensive analysis, the authors checked if each domain contained sufficient key features and iterative revisions were made. Features that were identical or nearly identical were aggregated. All authors gave feedback on the final taxonomy to refine the descriptions of the key features. PV summarised the feedback and the taxonomy accordingly.

Criteria of the RAND UCLA appropriateness method were used to analyse the data from the previous national Delphi study and the current Delphi study 1 [8]. We categorized the overall panel median as follows: 1–3 as inappropriate, 4–6 as equivocal and 7–9 as appropriate. Agreement signified that ≥ 70 % of panellists’ ratings were within the same 3-point region (that is, 1–3, 4–6, or 7–9) as the observed median. A feature was defined as “appropriate” with an overall panel median score of ≥ 7 and a level of agreement of ≥ 70 % within the 3-point region 7–9. A panel median of 4–6 or median with consensus of ≤ 70 % within the same 3-point region was defined as “equivocal”. A feature with a panel median of 1–3 and a level of agreement of ≥ 70 % within the 3-point region 1–3, was defined as “inappropriate”. The decision rules used in the national Delphi study and international Delphi study 1 are shown in Table 1. During Delphi study 1, PV and HV tabulated and discussed qualitative responses after Round 1, and circulated the results to the experts for Round 2. For international Delphi study 2, agreement signified that ≥ 60 % of panellists categorised a feature within the same domain. During Delphi study 2, PV and MB tabulated and discussed qualitative responses after Round 1 and circulated the results to the participants for Round 2. Quantitative analysis was done using Statistical Package for the Social Sciences (SPSS) version 21 for Windows (IBM Statistics).

This study has conformed to the principles embodied in the Declaration of Helsinki. All experts approved their participation by an informed consent. As this study does not involve patients or study subjects, according to the Dutch Medical Research in Human Subjects Act (WMO), this is exempt from ethical approval in The Netherlands.

After the first round, the international experts of Delphi study 1 considered 25 of the 59 proposed features of the taxonomy appropriate. (The overall panel median was between 7 and 9 with a consensus ≥ 70 % within this 3-point region, see Table 3, columns 3 and 4). Thirty-four features were rated as equivocal (overall panel median between 4 and 6 or median with consensus ≤ 70 % within the same 3-point region). Furthermore, the experts suggested five new features during this round (see Table 3, column 1): incentive systems (no. 43), community participation (no. 44), universal health coverage (no. 45), single point of access (no. 46) and alignment of regulatory frameworks (no. 47). After reviewing the results, the research team initially categorised the newly added features within the domain of system integration of the draft taxonomy. This addition led to a list of 64 key features.

In the second round, the 34 equivocal features were discussed during the face-to-face meeting. This resulted in an extra 17 features rated as appropriate, see Table 3, columns 5–7. The other 17 features remained equivocal after the second round. With regard to the five newly added features, three were rated as appropriate (no. 43, no. 44 and no. 47) and the remaining two (no. 45 and no. 46) as equivocal. To summarise, 45 features were considered appropriate (25 in the first round and 20 in the second round), while 19 features were considered equivocal (see Table 3, column 7).

Fifty features were included in the international Delphi study 2, because they were considered appropriate within the previous national Delphi study (see Table 3, column 2) or international Delphi study 1 (see Table 3, column 7). In the first round of international Delphi study 2, the panel members agreed (≥60 %) on the categorisation of 22 of the 50 features under one of the seven domains of the taxonomy (see Table 3, column 8).

In the second round, the panel members first discussed the descriptions of the seven domains of the taxonomy. The comments on the descriptions of the domains of the taxonomy and changes made in response to them are summarised in Table 4. Subsequently, the panel members discussed the categorisation of the remaining 28 features that were equivocal after the first round. This resulted in an additional 16 features that experts reached an agreement on regarding the categorisation under one of the seven domains of the taxonomy. The remaining 12 (28 minus 16) features were not agreed upon with regard to their categorisation (see Table 3, column 9).

During Delphi study 2, it appeared that our final taxonomy to describe an integrated primary care service model could be organised into three main categories: scope, type and enablers. The experts indicated that a distinction should be made between the individual (person-focused care) and population (population-based care) objectives to describe the scope of an integrated primary care initiative (see also Table 4). Furthermore, the experts indicated that the clinical, professional, organisational and system domains of our draft taxonomy could be used to describe the various types of integration processes. Finally, the qualitative comments of the experts revealed that functional and normative domains of our taxonomy are, respectively, the essential technical and cultural enablers for achieving integrated primary care. Therefore, we organised our final taxonomy into these three corresponding categories: 1) scope, 2) type, and 3) enablers of integrate primary care. Based on the comments made during Delphi study 2, the research team split the person-focused and population-based domain, resulting in a total of eight domains. A graphic representation of the final taxonomic structure is presented in Fig. 3.

The final taxonomic structure of eight domains was used to select and categorise the underlying key features. First, the features that were considered appropriate in the previous national Delphi study [6] as well as the current international Delphi study 1 were selected. This resulted in the selection of 29 features (see also Fig. 4). Second, the selected features were categorised within one of the domains of the taxonomy following the results of international Delphi study 2. Twenty-two features could be categorised within one of the eight domains. Third, the compiled taxonomy of 22 key features was reviewed by three authors (PV, MB and IB) on comprehensiveness per domain. The preliminary taxonomy did not contain any features within the system integration domain. A universal judgement was made by the three reviewers to add the features environmental climate (Table 3, no. 42) and alignment of regulatory framework (Table 3, no. 47) to the system integration domain because the categorisation reached consensus during Delphi study 1 and both features were considered appropriate during Delphi study 2. Finally, the three reviewers decided to merge the features population needs (Table 3, no.11) and population needs as binding agent (Table 3, no. 29) within the population based domain, inter-professional governance (Table 3, no. 17) and inter-organisational governance (Table 3, no. 25) within the organisational domain, and shared vision between professionals (Table 3, no. 14) and shared vision (Table 3, no. 59) within the normative domain due to similar content. The resulting taxonomy of 21 key features is shown in Table 5.

This study established an international consensus-based taxonomy to understand the integrated service models that arise in a primary care setting. The national and international Delphi studies resulted in the refinement of our previous taxonomy. The final taxonomy consists of 21 key features of an integrated primary care service model which are distributed over eight integration domains and organised into three main categories: scope (person-focused vs. population-based), type (clinical, professional, organisational and system) and enablers (functional vs. normative). The refinement of the taxonomy is a crucial step towards establishing an instrument that can measure a broad range of integrated service models.

The taxonomy contributes to a deeper understanding of the compound art of integrated care and provides direction for further field testing to identify effective components of integrated service delivery models in a primary care setting. The three main categories of the taxonomy provide a crucial differentiation to clarify and interpret practical examples of integrated care. To begin with, specifying the scope of an integrated care approach as either person-oriented or population-oriented helps to understand and describe the guiding principles and objectives of an integrated care approach. Although person-focused and population-based care could be viewed as opposite approaches, the strength of primary care philosophy is grounded in their symbiosis [2, 11]. Integrated primary care is the crucial point of tangency between public health services, which are more orientated on the population, and medical-oriented services, which are more focused on the individual [1‐3]. Consequently, there is a need to specify the balance between person-oriented and population-oriented objectives of an integrated primary care service model. In contrast to more disease specific integrated care models (which are generally more person-focused), the current taxonomy acknowledges that both scopes are needed to improve the provision of continuous, comprehensive and coordinated services in an ambulatory care setting [12].

Second, the original RMIC indicates four equally important types of integration processes: clinical, professional, organisational and system integration. However, the results of the Delphi studies indicated that the clinical, professional and organisational integration processes were the most recognised among the experts. This finding corresponds to the fact that these processes have been the prime focus of scientific research and practice [4, 13‐15]. The present study also indicates that less emphasis was placed on system integration processes. This result is in contrast with observations in the literature that societal and political influences are essential preconditions for achieving integrated primary care [3, 16, 17]. Research indicates that the development of integrated primary care is more hampered by political influences than technical influences [1]. Furthermore, the Chronic Care Model of Wagner [18] also stresses the importance of embedding integration efforts into the broader societal and political environment. Most of the experts in our study considered organisational integration processes as a systemic whole and found it difficult to differentiate between the system integration types of processes. One possible explanation might be found in the composition of our expert panels, as we did not explicitly include experts with a macro policy background. However, health care practitioners, funders and policymakers have generally a more limited scope compared to the theoretical discourses of integrated care and primary care [19]. The broad inter-sectorial system definitions of both primary care and integrated care have failed to produce practical relevance for practices and policies [6]. Moreover, actors in a public health sector generally have a broader perspective on the social-political aspects of health compared to actors with a healthcare background [20‐22]. We believe that more focus should be placed on the system environment when developing and evaluating integrated primary care service models. However, there is still a need to further clarify the domain of system integration and to explore how the different types of integration processes interact. For example, future research might investigate how local regulatory frameworks regarding integrated care influence the organisational and professional integration processes and vice versa.

Finally, the taxonomy assists to clarify and interpret the technical (functional) and cultural (normative) enablers in order to achieve common goals and optimal results. These functional and normative integration conditions seem to be of crucial importance to whether or not clinical, professional, organisational or system integration processes are successfully developed and sustained. Since integrated care spans across many different professional and organisational boundaries and mind-sets, it is crucial to clarify the required functional and normative prerequisites (e.g. data management, feedback, leadership) when developing and evaluating practical examples of integrated primary care.

The strength of this study is the international Delphi study approach to establish a consensus-based taxonomy. The final taxonomy is theoretically grounded on the RMIC [3], has a solid base in available literature [6] and was tested against a wide mix of expert opinions. Nevertheless, the Delphi consensus approach does not necessarily provide the “right” answer to a given problem, but should be viewed as a means to structure group communication and determine the degree of consensus between expert groups [23]. The Delphi approach in this study added substantially to the ongoing debate of defining integrated care.

A potential limitation of this study relates to the selection of experts. We attempted to be inclusive; however, not all experts who were invited were able to participate. Twenty-eight experts refused to participate in Delphi study 2, mainly because the face-to-face meetings were bound to a fixed date and time. We do not expect these rejections to have biased the results substantially because the final panel consisted of a balanced number of experts with both practice and scientific backgrounds. Moreover, the final taxonomy was based on the results from three Delphi studies. Nonetheless, the composition of the expert panels remained biased. The present taxonomy is based on professional (i.e. practical and scientific) values and preferences, while the views of other stakeholder groups (like patients, policymakers or health insurers) are also considered important in integrated care. Different stakeholder groups are likely to have different preferences [24]. This limitation can be solved when the taxonomy is tested in a local setting. All involved stakeholder groups (like patients, professionals, managers, insurers and policymakers) could be asked to comment on the relevance of the features included in the taxonomy and the relative importance per stakeholder could then be adapted accordingly [25‐27].

Ultimately, the present taxonomy represents a first step towards a common language for evaluating integrated primary care services. The variety of perspectives of the numerous actors involved in integrated care made us aware of the difficulties of developing a clear, consensus-based, non-overlapping assessment tool and a scoring scale useful for scientific research, policy and practice. Since the taxonomy is grounded on the theoretical concept of integrated primary care, the appropriateness of the taxonomy in other healthcare settings (e.g. hospital settings) should be further explored. Further research should also focus on the development of such a tool with contextualised and non-overlapping items and scoring scales. Finally, the critical challenge is to demonstrate the impact of integrated primary care models in terms of the ‘Triple Aim’ outcomes: 1) improve the individual experience of health care, 2) improve the health of populations, and 3) reduce the per capita costs of care [1]. Therefore, there is a need to further link the performance shaping features of the taxonomy with the three linked outcome measures of the Triple aim to determine the impact and to guide the continues design and redesign of integrated primary care practice.

The taxonomy is a valuable framework for patient organisations, professionals, managers, commissioners, and policymakers involved in the development of practical examples of integrated primary care. Profiling integrated primary care service models along this taxonomy makes it possible to obtain comprehensive and systematic information, and builds a common knowledge base regarding integrated primary care. Using the taxonomy to compare data across integrated care settings can promote the learning and sharing of (best) practices. Two activities involved in the development of an assessment tool are now pending: firstly, measurement instruments based on the taxonomy to generate reliable and validated quantitative scores; secondly, an agreed upon procedure that measures and incorporates the different perspectives of all the actors involved in integrated care (e.g. patients, professionals, managers, insurers and policymakers).

Once these measurement instruments and procedures are developed, we may be able to understand which interaction patterns achieve better health at a lower cost within a specific context. We plan further work to develop this assessment tool, and invite anyone interested in helping to validate the taxonomy to contact the authors.