Exploring the conflict between oral health and patient autonomy: A feasibility study using epistemic network analysis

Abstract

Medical ethics has relied heavily on theoretical argumentation; however, addressing complex, real-world ethical issues necessitates contextual understanding and empirical research. Qualitative methods are appropriate for exploring normative and phenomenological perspectives; such data are usually coded to explore patterns therein, which can be represented in quantitative models as well. Our objective was to showcase and address the utility of epistemic network analysis (ENA) in refining existing theory. We conducted interviews with dentists and patients on ethical challenges of patient autonomy in dentistry. Utilizing a codebook of ethical constructs based on theory, we amended the codebook with guided-inductive coding, and after coding the dataset deductively, we visualized the relative co-occurrence of codes and compared groupings of data. We refined existing ethical constructs and identified novel ones to create ENA models. Using these models, we demonstrated that examining the relative co-occurrence computed for various groupings yields richer insights into the data. However, ENA has its limitations, such as creating codes for different groups, defining those groups, or not gaining meaning from codes in isolation. Operationalizing theory and creating ENA models enables the researcher to identify points of consensus and divergence in stakeholder narratives and leverage various perspectives to inform, validate, or refine theory.

Introduction

Everyday medical practice is grounded in Codes of Professional Conduct (CPC), which are guided and justified by ethical principles, the aspirational goals of the profession. Besides the CPC, fundamental ethical principles also undergird advisory opinions (interpretations that apply the CPC to specific situations), which highly influence medical decision-making and how a council might interpret the CPC in a disciplinary proceeding. General medicine CPC rests on four fundamental ethical principles: Patient autonomy (duty to treat the patient according to the patient's desires; involving patients in treatment decisions in a meaningful way), nonmaleficence (duty to cause no harm), beneficence (duty to promote the patient's well-being) and justice (duty to distribute resources fairly). ¹

Different medical fields have adopted and tailored them to their practice. For example, the American Dental Association (ADA) added veracity (duty to communicate truthfully) to the four stated principles. ² The ADA's CPC emphasizes that these five ethical principles are not isolated, they can interact with each other, compete with each other for priority, in combination justify a given element of the CPC, and at times need to be balanced against each other. ² For example, patient autonomy may come into conflict with beneficence or nonmaleficence in dentistry due to the correlation between satisfaction with orofacial appearance and psychosocial well-being,^3–5 exemplified by the popularity of cosmetic interventions.^6–8 Moreover, patient requests often influence decisions regarding tooth extractions.^9–11

Theory aids in resolving these tensions and developing guidelines for consideration in specific (clinical) situations. The most commonly employed theories in bioethics use ethical principles for justification; however, this often results in conflicting moral conclusions. ¹² Ideally, ethical theory is informed by society (as urged by the ADA), grounded in the lived experiences and mental models of those comprising society. The ADA highlights the importance of an ongoing dialogue between key stakeholders, particularly between the medical profession and patients, to ensure that ethical principles, and by extension CPC and advisory opinions, remain responsive to and reflective of lived realities.^2,13 Theories are grounded in particular epistemologies and ontologies; they identify constructs of interest, which can guide empirical research, for example, regarding stakeholder opinions.

In dental ethics, the most prevalent theory engaging with fundamental ethical principles is the Central Practice Values (CPVs) theory by Ozar et al., ¹⁴ and Beauchamp and Childress, ¹ which identifies six values: The patient's life and general health, The patient's oral health, Patient autonomy, The dentist's preferred patterns of practice, Esthetic values and Efficiency in the use of resources. The authors conceptualize these six central values as partially analogous with the four fundamental ethical principles. Considering the ethical dilemmas mentioned earlier (performing medically not indicated cosmetic interventions and tooth extraction on patient request), patient autonomy would conflict with oral health, which instantiates the principles of beneficence and nonmaleficence. Ozar and co-authors present these values within a universally applicable, static hierarchy. However, this framework was developed from within the dental profession and does not incorporate the patient's perspective. As Rule and Veatch ¹⁵ have noted, it also does not represent a consensus within the profession itself, but rather a theoretical model proposed by its authors.

Arguably, as do principles, values interact in complex ways, and this ‘competition, combination, and balancing’ ² may take on different forms in various contexts, under certain conditions and in particular mental models (e.g. dentist versus patient). Identifying patterns in how values or principles interact is informative for both the CPC and advisory opinions. Similarly, in ethical decision-making, insight can arise from exploring how values interact and cohere, rather than evaluating them in isolation. ¹⁶ We assume such interactions constitute a dynamic intra- and inter-personal system that evolves over time and manifests differently across conditions (e.g. granting harmful patient requests is unethical, as the patient may regret it if their value system changes ¹⁷ ). Mental models comprised of the interplay between constructs can be mapped in various ways, and when the manifestation of constructs is still not well-known, qualitative research is needed.

Qualitative data are usually processed by annotating them with representations of constructs of interest (codes). Empirical findings may enrich and/or validate theory in fine-tuning constructs of interest and understanding the relationships between those.

Our objective was to examine the feasibility of a method aiming to refine construct definitions and elucidate relationships between constructs as a means to inform theory, which could, in turn, contribute to the development of CPC and advisory opinions for dentistry. To achieve this objective, we apply the CPVs to qualitative data to map the mental models of two stakeholder groups (dentists and patients) involved in the ethical conflict between patient autonomy and oral health. We aimed to: (1) supplement ontology (i.e. refine construct definitions, offer novel constructs of interest), (2) model the relationships between constructs (via the interaction of codes) and (3) explore different conditions under which these relationships may exhibit different patterns. The following is a worked example along the three subobjectives, serving as a feasibility study. We begin with introducing the data and analytical process, then discuss the affordances and limitations of this method within the context of relevant literature.

Methods

To enhance transparency, the study was preregistered. The preregistration is available at: https://osf.io/exunf. The tabularized dataset containing applied codes and segmentation (sans data) is available at: https://osf.io/b4a72

Sampling and data collection

We sampled from two populations: dentists and patients living in Hungary, employing non-proportional quota sampling. In the patient subsample, quotas were based on sex and age, as previous literature has found these crucial factors in the lay perception of dental aesthetics^18,19 and moral judgements ²⁰ ; practitioners were stratified based on sex ²⁰ and leadership experience at private clinics, university departments, or professional organizations.

Patient participants were included if they had experienced a situation in which their request did not align with the corresponding medical diagnosis or legally accepted clinical practices (e.g. seeking treatment for purely aesthetic reasons, refusing the extraction of an asymptomatic tooth, or requesting care from non-licensed practitioners). Exclusion criteria for patients were legal incompetence (e.g. under 18 years of age) or holding qualifications related to dental care. Dentist participants were eligible if they had successfully completed accredited specialty training, had at least 10 years of professional experience, and had worked in both public and private sectors. Dentists who did not meet all of these criteria were excluded. Dentists were recruited using publicly available contact information; snowball sampling was employed to identify additional dentist and patient participants. Eligible patients were recruited following the completion of their treatment. Sample sizes were determined by theoretical saturation and were considered sufficient for demonstrating proof-of-concept. Data were collected through semi-structured interviews and a self-developed sociodemographic survey (birth date, sex assigned at birth, education, residence, and dentist's workplace and experience). The interview guide covered ethical values in dentistry, a case on mass tooth extraction, personal experiences with the ethical dilemma, views on cosmetic dentistry and informed consent and trust in dentists. Interviews were recorded, transcribed verbatim and anonymized.

The dentist subsample consisted of 14 participants (8 males, 6 females), with a mean age of 48.5 (range: 35–74). Among them, 8 were university department heads, 6 were private clinic heads, 11 held leadership roles in professional organizations, and 4 did not hold any leadership positions. The patient subsample included 10 participants (4 males, 6 females), with a mean age of 45.5 (range: 28–78).

Code development followed a guided-inductive approach. Two coders, working independently, conducted sentence-level coding on 10% of the data using the Interface for Reproducible Open Coding Kit (iROCK) (https://i.rock.science). Initial coding was structured around the CPVs, with the addition of an ‘Other’ code to capture aspects of decision-making that were not encompassed by the CPVs. ¹⁴ Coders were given the flexibility to revise the codebook to ensure code definitions were grounded in the data and reflective of the mental models represented within both subsamples. The coders then triangulated their results and developed a tentative code structure, which was tested deductively on another 10% of the data and triangulated again. Following further rounds of testing and triangulation, the coders assessed their coding through inter-rater reliability and subsequently refined code definitions through discussion of discrepancies and their underlying rationale until a final codebook was established. The codebook based on the CPVs is presented in Table 1. The coders deductively coded the entire dataset independently.

Table 1.

Codebook based on the central practice values with examples from our dataset.

Code label	Central practice value	Definition	Example
Well-being	The patient's life and general health	Physical, mental or social well-being	A treatment outcome boosting the patient's confidence
Medical indication	The patient's oral health	Appropriate and pain-free function of the oral cavity and surrounding tissues	Extracting a carious wisdom tooth that could potentially cause further issues
Patient autonomy	Patient autonomy	Treatment decisions made based on patient's own beliefs, goals and values; Patient's right to choose their dentist	Dentist accepting patient's request without attempting to further persuade the patient
Practice patterns	Dentist's preferred patterns of practice	Dentist's preferences for equipment, medication, materials, setting, treatment strategy; Decisions based on dentist's habits, experience, skills and philosophy (e.g. to extract teeth as a last resort)	No example was identified
Esthetics	Esthetic values	Aesthetic standards of individuals or the wider community	Patient requesting replacing amalgam fillings due to their colour
Efficiency	Efficiency in the use of professional resources	Allocating resources (dental expertise, dentist's capacity, physical resources) with the aim of maintaining accessibility to oral care for other patients and society at large	No example was identified
Other	Not applicable	Further aspects related to the dilemma between patient autonomy and oral health	Patient stating that the choice of dentist is a matter of confidence

Employed modelling tool

We operationalized the relationship between constructs (CPVs) as the co-occurrence of codes (representations of CPVs) within a given sentence and the preceding and/or following sentence in an interview. We examined these patterns at the level of individuals, subsamples and sociodemographic groups. While such co-occurrences can be tabulated in a matrix, we chose to visualize them as networks to enable both visual inspection and more complex structural analysis. We employed epistemic network analysis (ENA), which generates networks where nodes represent codes and edges represent the frequency of code co-occurrences. As the nodes are positioned in the projection space identically for all networks within a model, individual or mean networks (of groups of data providers) can be compared based on connection strengths.

From the tabularized dataset, ENA (https://www.epistemicnetwork.org) constructs a matrix of pairwise code co-occurrences within specified data segments represented as vectors in high-dimensional space. Since these vectors contain different amounts of data, they are normalized (divided by their own length), thus computing the relative frequency of co-occurrence. Vectors were subjected to a dimensional reduction procedure (means rotation and singular value decomposition) and projected as networks into a two-dimensional space, along an X and Y axis. As the space is constructed from patterns of code co-occurrences, the location of network nodes is meaningful: Nodes in close proximity exhibit similar co-occurrence patterns, more distal nodes, dissimilar ones. Differing connection strengths (quantified in edge weights) represent the relative frequency of co-occurrence between unique pairs of codes and can be used to compare networks. ENA offers another, coordinated representation of the data: Plotted points. These are points representing the location of each unit of analysis (e.g. individuals or groups of participants) in the projection space. This visualization also depicts the 95% confidence interval of the group means. Since the visualizations are coordinated, node and plotted point locations mutually inform each other. Any two networks can be compared in a difference plot (subtracted network), where ENA subtracts the weight of each connection in one network from the corresponding connections in the other. For a more detailed account of ENA model construction, please see:^21–24

In subsequent sections, code labels are capitalized and italicized, while participant narratives are italicized in quotation marks. The narratives, originally in Hungarian, were translated by the first author.

Results

Supplementing ontology

With the guided-inductive coding process, we aimed to operationalize existing constructs (six CPVs), and with the code ‘Other’ leave conceptual space for any novel constructs to arise. As we included the patient subsample data in code development, we enabled another crucial stakeholder perspective to mold theory. In the following, we present an example of a refined construct definition and a list of novel constructs with one example elaborated in detail.

Within the CPV framework, Esthetics is defined by Ozar et al. ¹⁴ from a dentist perspective as the dentist's interpretation of society's aesthetic standards, with the responsibility to guide patients toward these norms to support their social well-being. However, across narratives, we did not encounter any suggestion that aesthetics represents a social consensus or that it is the dentist's role to interpret or enforce such standards. Therefore, we revised the definition to encompass aesthetic judgments made by society or any individual.

As for novel constructs, Table 2 contains our disaggregated Other code with constructs to be potentially incorporated into theory.

Table 2.

Disaggregated other codes.

Code label	Definition	Example
Beneficence	Desire to fulfil ‘what is in best interest’ of the patient; Desire to preserve the patient's health or bodily integrity	Dentist persuading the patient with an abscess to give consent to a tooth extraction
Professional autonomy	Considering the dentist's personal set of values, practice patterns; The dentist's personal responsibilities	Dentist rejecting patient request due to personal convictions
Laws and rules	Choosing a course of action in accordance with the laws governing dental practice; Including reference to quackery	Dentist carrying out a patient request if they sign a form
Feasibility	Weighing the constraints of a desired treatment outcome; What is realistic; What is modern or outdated	Patient acknowledging that modern medicine provides opportunities to treat teeth instead of extracting teeth
Informing	Decision made based on access to information; Informing patients without the intent to convince them; Patient desire to be fully informed	Patient rejecting an intervention based on possible risks outlined by the dentists
Patient comfort	Minimizing the number of sessions, time spent at the dental office, fear (of the treatment, not of dentists in general), or inconvenience of the intervention	Patient experiencing emotional distress caused by an intervention
Finance	Decision based on financial considerations	Patient assuming any treatment is possible if one can afford it
Plurality	Acknowledgement of multiple moral systems; Including: Dentist and patient viewpoints, cultural differences	Dentist stating that if they decline a patient's request, another dentist will likely agree to perform the treatment
Needs	Ethicality of an intervention differs on the specific circumstance in which it occurs	Dentist fulfilling aesthetic requests for a bride's wedding
Health attitude	Evaluating the feasibility of a treatment based on oral hygiene or regular check-ups	Dentist recommending a different treatment strategy to a patient with insufficient oral hygiene
Prestige	Confidence towards the dentist or medicine as a whole; Public perception of dentistry and medicine	Patient stating that the choice of dentist is a matter of confidence
Triangulation	Weighing others’ opinions	Patient requesting an intervention because it was promoted or undergone by a celebrity
Personal experience	Decisions informed by prior events where participants were directly involved or had observed others	Patient expressing greater trust in younger dentists after experiencing issues with a filling placed by an older dentist
Minimally invasive	Opting for the alternative that involves less damage to tissues or structures	Dentist prioritizing preventive interventions

Minimally invasive is a code we developed to denote the decisions made with the goal to maintain the integrity of oral structures to the greatest extent possible, in contrast to improving function or aesthetic appearance, as a patient elaborated when asked about attempting to save a tooth with a questionable prognosis: ‘Saving my teeth is more important than the number of interventions [i.e. tooth extraction is a last resort]’. Within the context of dentistry, this concept illustrates a practical application of nonmaleficence, a principle articulated by Beauchamp and Childress, ¹ not directly present in the CPV.^,14

Modelling relationships between constructs

The CPVs are structured as a hierarchy, which implies that it is composed of isolated constructs and adopts a unidimensional approach to the relationships among them, namely a ranking of perceived importance. We employed ENA to explore relationships via code co-occurrences. Figure 1 displays the mean network of the dentist and patient subsamples. In both networks, Esthetics emerged as a dominant code, strongly interacting with other codes, such as Medical indication for both dentists and patients, and Well-being for patients. The co-occurrence between Esthetics and Medical indication typically referred to procedures that improve both function and appearance, particularly when the initial complaint was aesthetic; as a dentist stated: ‘I simply observed that I have probably never had a patient, who did not need any oral hygienic interventions, and if this intervention is removing calculus, then removing stains whitens teeth in a way that makes patients satisfied’. In contrast, the connection between Esthetics and Well-being framed orofacial aesthetics in terms of its contribution to the overall quality of life and psychosocial health, as articulated by a patient: ‘Obviously, very bad aesthetics can cause very bad mental health’. Thus, examining Esthetics in connection with other codes revealed its function guiding clinical decision-making, rather than merely affirming its general importance.

Figure 1.

Mean epistemic networks of dentists (above, in red) and patients (below, in blue). Nodes (black circles) denote the codes; node size indicates the relative frequency of the code co-occurring with other codes. Edge (line) thickness and saturation correspond to the relative frequency of co-occurrence between each pair of codes.

ENA offers a complementary visualization, shown in Figure 2, where individual and group networks are displayed as plotted points. In this figure, circles represent the epistemic networks of interviewees and squares represent the mean epistemic network locations of patients and dentists (red and blue, respectively). Additionally, the visualization contains a subtracted network, highlighting the differences in edge weights in the patient and dentist mean networks. The tighter confidence interval (dashed lines around mean network locations) of the dentist subsample is indicative of more similar mental models than those of patients.

Figure 2.

Epistemic network analysis projection space displaying individual network locations for each participant (coloured circles) in the dentist (red) and patient (blue) subsamples, as well as mean network locations for both subsamples (coloured squares). Dashed lines around the means denote the 95% confidence intervals on each dimension. Subtracted graph of mean patient and dentist networks (centre); black circles represent codes, red edges (lines) and blue edges signify higher frequency of code co-occurrences in dentist and patient narratives, respectively.

The proximal location of networks vis-à-vis codes indicates a higher saturation of those connections within the networks. Codes more typical of dentist narratives included Informing, Finance, Medical indication and Minimally invasive. In contrast, codes more commonly found in patient narratives were Comfort, Personal Experience, Prestige, Feasibility, Needs and Well-being. This distinction underscores that dentist narratives drew on principles and values often used in bioethics, including the CPVs and ADA's CPC.^2,14 Conversely, patient narratives tended to emphasize what is tangible and personally relevant from their perspective: A desire to minimize inconvenience, fulfil personal needs, making decisions based on individual experiences, recognizing the prestige of medicine, and weighing technological possibilities. Patient narratives less frequently addressed topics such as cost, the dentist's obligations to ensure patient autonomy, or the expectation that the dentist should provide thorough information.

Exploring different conditions to discover patterns

Patterns of interaction may be observable between any codes, and these interactions, along with the parameters of the projection space, may change substantially when data are grouped in various ways. In the example above, we aggregated interviews by subsample; however, we can also create groups according to participant attributes. As prior literature does not address differences in how dentists with and without leadership experience in professional organizations perceive ethical dilemmas related to patient autonomy, we compared these two groups with an exploratory purpose. Figures 3 depicts the mean epistemic network of dentists with and without leadership experience in a professional organization. In the latter group, a strong connection was observed between Patient autonomy and Laws and rules, suggesting a tendency to comply with existing regulations, particularly when patient requests align with legal allowances, as in the remark concerning only treating front teeth: ‘They must confirm in a document that they accept [no treatment in the lateral zone], it is possible, as we don’t damage their health (…) we grant their request’. In contrast, the network of dentists with leadership experience exhibited stronger connections between Health attitude and Plurality, acknowledging diverse societal perspectives and distinguishing between patients who are more health-conscious and those who are not: ‘I view the patient themself, so if (…) at the age of 35, with such poor oral hygiene (…) statistics show that they will maintain this level of oral hygiene’.

Figure 3.

Mean epistemic network of dentist participants with leadership experience in a professional organization (above, in blue) and without (below, in red). Nodes (black circles) denote the codes; node size indicates the relative frequency of the code co-occurring with other codes. Edge (line) thickness and saturation correspond to the relative frequency of co-occurrence between each pair of codes.

Discussion

This feasibility study aimed to demonstrate a method by which theory can be refined via the specification and augmentation of constructs, modelling the relationship between those and exploring relevant patterns via grouping data in various ways.

Supplementing ontology

By analyzing data from two stakeholder groups using a guided-inductive approach, researchers were instructed to operationalize the existing CPVs and identify novel constructs (via the Other code) as a way to apply and refine theory.^14,25 For example, Lipworth and Little ²⁶ applied the principles of Beauchamp and Childress ¹ to pharmaceutical ethics identifying the novel principles ‘other-ness’ and ‘firm-ness’. Similarly, we included a patient population to broaden the dentist-centric theoretical scope, and our analysis refined original principles and helped in identifying new ones. We illustrated refinement of ethical principles with Esthetics, which shifted from solely the dentist's interpretation of standards to a more inclusive and individualized one. This modification acknowledges the relativity of orofacial aesthetics, as some would find unnaturally white, perfectly aligned teeth aesthetic, while in other subcultures, partial edentulism is a marker of identity. ²⁷ One of our novel constructs was Minimally invasive, which may inform theory, as preserving oral structures is a common treatment goal, ²⁸ but potential damage to oral structures may also influence granting patient requests.

Operationalizing and refining theory in this manner involved challenges; most prominently: Determining whose viewpoints are relevant to theory (i.e. specifying populations sampled from) and justifying how individual perspectives are grouped, for example, as stakeholders. Regarding the former, we assumed CPVs would be enriched by the patient perspective, as clinical decision-making involves both parties. Regarding the latter, for example, while dentists referred to patient autonomy (a value in the CPVs), patients also described their own experiences of autonomy. This requires a methodological decision of either treating these as distinct codes (e.g. separate ‘Patient autonomy’ code with different definitions for stakeholder groups) or consolidating definitions into a shared construct.

Modelling relationships between constructs

By deductively applying the final codes developed in our guided-inductive approach to the full dataset, we could examine the relationship between those. We modelled the relative frequency of co-occurrence between unique pairs of codes within designated data segments, which yielded several affordances. Firstly, we could depart from a static ranking of CPVs, which aligns with the recommendation of Beauchamp and Childress ¹ that the contribution of each principle to a clinical decision should be investigated rather than ranking them. Secondly, inspecting mean networks and confidence intervals allowed us to pinpoint areas of convergence and divergence in participant mental models. For example, we demonstrated that subsamples converged in viewing aesthetic problems as potential medical issues but pinpointed divergence in the role of Esthetics, as patients linked orofacial aesthetics to their overall well-being, while dentists considered it as a possible sign of pathology. This finding indicates that, unlike in the CPVs, where aesthetics is a low-ranking value for dentists, it is, in fact, significant, as it may be a motivator for patients to seek dental treatment. ¹⁴

This modelling method entailed several limitations. Crucially, ENA does not have hypergraph capabilities, it only computes pairwise code relationships; triads of codes (or more) cannot be modelled, curtailing the investigation of more complex code constellations. Code co-occurrences are critically determined by data segmentation, that is, how we operationalize to which segments we apply codes and in what manner we compute their co-occurrence; for a more detailed account, please see the work of Zörgő. ²¹ Albeit not solely a property of ENA, such sensitivity to methodological decisions has a defining influence on model generation. Lastly, modelling flattens data, but in the case of ENA, this was mitigated by retaining access to the original qualitative data (via the webtool) and by gaining power in exploring patterns in various groupings of data.

The model supported this by providing a proportional representation of relative code co-occurrences through node sizes and edge thicknesses. Although numerical comparisons are also possible, visual inspection is predominantly employed in ENA as it facilitates identifying salient connections when dealing with a large number of code pairs.

Exploring different conditions to discover patterns

The distinction between dentists without and with leadership experience revealed meaningful differences: the latter tended to recognize and respect societal diversity in health attitudes and adapted their approach accordingly, whereas those without leadership experience appeared to view regulations as a barrier or as patient autonomy. Grouping data for explorative purposes could be based on any collected attribute (e.g. sociodemographic characteristics), circumstance (e.g. type of procedure, psychological co-morbidity, culture-specific requests, aesthetic ideals), or time (e.g. longitudinal analysis of mental models). Neglecting to employ software for such exploratory analysis, relying on hermeneutics alone, would denote a high cognitive load and a susceptibility to (e.g. selection or confirmation) bias.

However, data aggregation requires careful consideration and a weighing of whether, for analytical purposes, individual narratives can be justifiably combined into a shared mental model (cf.: problem of ergodicity ²⁹ ).

Limitations

As with all modelling techniques, our methodological choices (e.g. in data segmentation, code development, parameterization) had a defining effect on outputs. Spatial and scope-related limitations precluded us from weighing and discussing these in our paper. Additionally, the edges of an ENA graph only represent one type of relationship between codes, which limits interpretability and further flattens the data. A solution for this issue could be using Qualitative Network Approach, ³⁰ which enables the differentiated visualization of various types of relationships (e.g. causal, temporal, structural); however, with this method, information about frequency is lost in the trade-off.

Conclusion

A significant undertaking of empirical bioethics is to actualize ethical standards with real-world dilemmas and refine theory with empirical insights. Mapping and framing the interaction among stakeholders’ ethical values offers deeper insight into their underlying mental models than analysing values in isolation. ENA served as a useful tool by facilitating the examination of relative code co-occurrence patterns of groups.