EDEN: Towards a Computational Framework

to Align Incentives in Healthy Aging

Wasu Mekniran

1,2 a

and Tobias Kowatsch

3,4 b

Centre for Digital Health Interventions (CDHI), Department of Management, Technology, and Economics,

ETH Zurich, Switzerland

CDHI, Institute of Technology Management, University of St. Gallen, Switzerland

Institute for Implementation Science in Health Care, University of Zurich, Switzerland

School of Medicine, University of St. Gallen, Switzerland

Keywords: Healthy Aging, Incentive, Natural Language Processing, Retrieval-Augmented Generation, Large-Language

Models, Network Analysis.

Abstract: Incentive misalignment among healthcare stakeholders poses significant barriers to promoting healthy aging,

hindering efforts to mitigate the burden of long-term care. Despite extensive research in public health,

incentive gaps persist, as static implementation guidelines often fail to accommodate dynamic and conflicting

incentives. This study introduces and evaluates EDEN (eden.ethz.ch), a computational framework designed

to dynamically map stakeholder incentives using a Retrieval-Augmented Generation pipeline. A comparative

study using a health insurer use case evaluates alternative incentive analyses; qualitative content analysis,

large language models, and EDEN. The evaluation assesses their ability to identify and address incentive gaps.

Preliminary findings demonstrate the EDEN's ability to map incentives and highlight misalignment compared

to alternative approaches. These findings demonstrate how EDEN can offer evidence-based strategies for key

healthcare stakeholders, such as health insurers, based on retrieval features to align incentives in healthy aging.

1 INTRODUCTION

The healthy aging initiatives face significant

challenges in fostering collaboration among diverse

stakeholders, including health insurers, government

agencies, healthcare providers, and local

communities (WHO, 2020). These stakeholders often

operate with distinct and sometimes conflicting goals,

leading to fragmented efforts that undermine the

scalability and effectiveness of long-term initiatives

(Mekniran et al., 2024; Fried et al., 2022). For

instance, health insurers' focus on short-term cost

containment often conflicts with healthcare providers'

long-term goal of improving population health

outcomes. Such misaligned incentives, coupled with

limited mechanisms for coordination and macro-

system integration, impede the development of

cohesive and sustainable healthcare strategies.

Despite the recognized importance of

collaboration and alignment, significant gaps remain

https://orcid.org/0000-0001-5184-0438

https://orcid.org/0000-0001-5939-4145

in understanding how stakeholder incentives align

and integrate within the ecosystem (Berwick et al.,

2008). Current efforts, such as static best-practice

checklists, often fail to account for the dynamic and

evolving nature of stakeholder priorities. For health

insurers, these gaps are particularly evident in their

regulated care provision (Mekniran, Kramer, et al.,

2024). Furthermore, the lack of systematic tools to

identify value propositions, analyze alignment

patterns, and generate actionable insights leaves

unexplored potential collaborations and unaddressed

incentive gaps.

This workshop paper introduces EDEN, a

computational framework that augments the Cross-

Industry Standard Process for Data Mining (CRISP-

DM) framework (Chapman, 2000) by integrating

artificial intelligence (AI) techniques, including

Natural Language Processing (NLP), Large

Language Models (LLMs) along with Retrieval-

Augmented Generation (RAG). EDEN adapts

Mekniran, W. and Kowatsch, T.

EDEN: Towards a Computational Framework to Align Incentives in Healthy Aging.

DOI: 10.5220/0013359800003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 2: HEALTHINF, pages 1067-1076

ISBN: 978-989-758-731-3; ISSN: 2184-4305

1067

CRISP-DM to the dynamic complexities of

healthcare ecosystems, making it capable of

systematically mapping stakeholder incentives,

identifying misalignments, and uncovering

actionable collaboration opportunities. Through the

analysis of unstructured data, such as value

propositions from health insurers and healthcare

providers, EDEN dynamically generates user-

specific evidence-based recommendations. To this

end, the study addresses the following research

questions:

1. How effectively does EDEN identify thematic

clusters in value propositions?

2. How effectively does EDEN generate

recommendations for collaboration?

Next, we present the theoretical foundation of EDEN,

outline its approach, discuss the results of

comparative analysis, and conclude with implications

for the proposed tool and its potential development.

2 RELATED WORK

2.1 Collaboration for Healthy Aging

Promoting healthy aging relies on multi-stakeholder

collaboration, as highlighted by frameworks such as

the World Health Organization's Decade of Healthy

Ageing, which prioritizes fostering age-positive

attitudes, empowering older adults in supportive

communities, delivering person-centered care, and

ensuring access to long-term care (WHO, 2020). The

Global Roadmap for Healthy Longevity advocates for

inclusive social infrastructure to combat ageism,

improve digital access, and integrate health and

lifelong learning to support equitable care and active

participation for older adults (Dzau & Jenkins, 2019;

Fried et al., 2022; National Academy of Medicine,

2022; Wong et al., 2023). Similarly, Cox and

Faragher emphasize prioritizing aging biology,

fostering interdisciplinary research networks, and

scaling aging initiatives through targeted funding and

awareness campaigns (Cox & Faragher, 2022).

Although these frameworks highlight the

importance of collaboration, they often rely on static

principles that do not account for dynamic

stakeholder incentives. Altpeter et al. emphasize the

participation of non-traditional partners,

strengthening healthcare linkages, and adapting local

programs to better serve vulnerable populations

(Altpeter et al., 2014). Bonnes et al. advocate for

longevity clinics that integrate public health with

research collaborations, offering early detection and

lifestyle interventions to improve clinical outcomes

(Bonnes et al., 2024).

2.2 Iterative Analytics Process

EDEN, an acronym for Emerging Business Models in

Digital Health for Healthy Longevity, adopts the

CRISP-DM framework to address the complexities of

stakeholder analysis in healthcare. CRISP-DM,

recognized for its iterative and cyclical structure, is

widely adopted for developing data-driven solutions

due to its flexibility and domain-agnostic design,

making it particularly suited to dynamic and complex

healthcare systems such as healthy aging ecosystems

(Larose, 2015). EDEN enhances the CRISP-DM

framework by integrating NLP using the Natural

Language Tool Kit (NLTK) package (Bird et al.,

2009), LLMs, and RAG. These techniques enable

EDEN to derive actionable insights by modelling and

contextualizing stakeholder value propositions. The

adapted phases of CRISP-DM within EDEN are

shown in Figure 1.

Figure 1: EDEN's iterative analytics process, extends the

CRISP-DM framework (Chapman, 2000) by integrating

NLP, LLMs, and RAG, making CRISP-DM more suited to

evolving priorities of healthcare stakeholders.

(1) Business Understanding phase identifies and

analyzes stakeholder incentives in preventive care

through qualitative methods, including systematic

literature reviews, market research, and expert

interviews. These inputs establish a foundational

understanding of stakeholder goals and challenges

(Giger et al., 2024; Mekniran, Giger, et al., 2024;

Mekniran, Kramer, et al., 2024; Mekniran &

Kowatsch, 2023).

(2) Data Understanding phase explores structured

and unstructured datasets, including organizational

metrics from financial database, to identify value

Scale-IT-up 2025 - Workshop on Scaling Up Care for Older Adults

1068

propositions and stakeholder priorities in a healthy

aging ecosystem (Mekniran, Giger, et al., 2024).

(3) Data Preparation phase involves text

preprocessing steps such as tokenization,

lemmatization, and vectorization, enabling thematic

and relational analyses of stakeholder information

(Bird et al., 2009).

(4) The Modeling phase applies topic modeling to

uncover thematic clusters and cosine similarity to

measure alignment. LLMs contextualize stakeholder

propositions, with RAG enhancing organizational

data integration and network analysis to uncover

evidence-based collaboration patterns (Silge &

Robinson, 2017).

(5) Evaluation phase assesses the model's

accuracy and utility using metrics such as similarity

scores and network centrality, ensuring robust

representation of stakeholder relationships.

(6) Deployment phase presents computational

results through interactive network graphs,

highlighting collaboration opportunities and enabling

users to input value propositions for real-time

alignment assessments. Cloud computing ensures

scalability and real-time accessibility, supporting

larger datasets and complex stakeholder ecosystems.

3 METHODS

This study uses a comparative analysis to assess the

effectiveness of EDEN in generating actionable

recommendations within the healthy aging ecosystem

(Sharma & Kaur, 2017; Verma et al., 2016). Using

value propositions from user input and from previous

longevity landscape study (Mekniran, Giger, et al.,

2024), our comparative analysis compares (A)

qualitative content analysis conducted by an author

W.M. based on existing guidelines to code

organization segment, target customers, and products

and services (Fried et al., 2022; Hsieh & Shannon,

2005; WHO, 2020), (B) standalone LLM using GPT-

4o model (OpenAI et al., 2024), and (C) proposed

EDEN's CRISP-DM methodology, which integrates

LLMs GPT-3.5-turbo with NLP techniques such as

topic modeling and similarity computation, see Figure

2. The study examines the contextual relevance,

granularity, and actionability of the insights generated,

providing an initial requirement of EDEN's capability

in addressing incentive misalignments.

EDEN integrates NLP with an adapted RAG

framework to analyze unstructured stakeholder value

propositions and provide actionable

recommendations. NLP techniques, such as topic

modeling and Term Frequency-Inverse Document

Frequency (TF-IDF) vectorization, are employed to

identify thematic clusters, extract key incentives, and

map similarities among stakeholders (Jelodar et al.,

2019; Valdez et al., 2018). These foundational

processes enable a structured analysis of the value

proposition and collaboration opportunities within

the healthy aging ecosystem, supporting strategic

decision-making (Frow & Payne, 2011; Murtaza &

Ikram, 2010).

The integration of RAG in EDEN addresses key

limitations of traditional generative AI systems, such

as hallucinations (Béchard & Ayala, 2024; Gao et al.,

2023; Huang & Huang, 2024), by dynamically

retrieving context-specific data, including thematic

clusters and similarity scores derived from network

analysis, and incorporating it into the prompt context

for LLMs. This bridging of retrieval and generation

ensures that output is evidence-based, actionable, and

aligned with domain-specific knowledge (Li et al.,

2024). RAG further enhances contextualization,

scales to dynamic datasets, and adapts to the complex

dynamics of the healthy aging ecosystem. Figure 2.

illustrates the EDEN workflow and integration of

these components.

Figure 2: A comparative analysis evaluates (A) content

analysis, (B) standalone LLM, and (C) EDEN.

3.1 Data Preprocessing

The dataset, comprising stakeholders' value

propositions (e.g., diagnostics, wellness, monitoring)

and organizational details (e.g., funding amount,

location, founded year), was sourced from a previous

study (Mekniran, Giger, et al., 2024). Using Python's

Pandas package (McKinney, 2010), this dataset was

structured into a data frame for further analysis.

EDEN: Towards a Computational Framework to Align Incentives in Healthy Aging

1069

Preprocessing steps were applied to standardize and

clean the textual data, ensuring its suitability for

downstream analyses:

1. Stopword Removal: Eliminating commonly used

words with minimal semantic value.

2. Lemmatization: Reducing words to their root

forms to ensure linguistic consistency.

3. Tokenization: Splitting text into individual

words to facilitate granular analysis.

Next, the preprocessed data was vectorized using the

NLTK package (Bird et al., 2009), creating numerical

representations that could be analyzed using thematic

clustering and network analysis. Clean data plays a

crucial role in this step, as its high quality ensures that

relevant relationships are effectively captured and

modeled (Bird et al., 2009), see Code Snippet 1S.

3.2 Topic Modeling

To analyze textual data, we employed Term

Frequency-Inverse Document Frequency (TF-IDF) to

construct a document-term matrix. This numerical

representation assigned weights to terms based on

their contextual importance, balancing term

frequency within individual documents and term

rarity across the entire dataset (Bengfort et al., 2018).

This approach ensured that the matrix emphasized the

most relevant phrases for specific stakeholders,

capturing contextually significant terms.

Non-negative Matrix Factorization (NMF) was

applied to the document-term matrix using the Scikit-

learn package (Pedregosa et al., 2011), following

Code Snippet 1S. NMF decomposed the high-

dimensional document-term matrix into two non-

negative, lower-dimensional matrices, identifying

thematic clusters such as "digital innovation" and

"community engagement." Its non-negativity

constraint ensured that the resulting clusters were

concise, interpretable, and suitable for the smaller

datasets used in this study, see Code Snippet 1S. This

decomposition is mathematically represented as (1):

  

(1)

Where:

V: Original document-term matrix

W: Basis matrix representing the thematic clusters

H: Coefficient matrix indicating the document-topic

associations.

Using the nmf.fit() function, the algorithm

iteratively adjusted W and H to learn patterns within

the data, forming a basis for identifying alignment

patterns and collaboration opportunities among

stakeholders. These thematic clusters served as a

foundation to measure stakeholder alignment.

3.3 Network Analysis

To assess thematic alignment, cosine similarity was

calculated for NMF-transformed topic vectors,

quantifying the degree of alignment between

stakeholder value prepositions (Bengfort et al., 2018),

following Code Snippet 2S. The nodes in the resulting

network graph represented stakeholders, while the

edges constructed an undirected network connecting

stakeholders whose similarity scores exceeded a

predefined threshold (e.g., the 50th percentile).

Cosine similarity, a widely used metric in text

analysis, quantifies thematic alignment by measuring

the cosine of the angle between two vectors  and 

as in (2) (Silge & Robinson, 2017).





 

























 











3.4 Adapted RAG for

Recommendations Engine

Within EDEN, the RAG framework retrieves curated

data and network analysis outputs, which are integrated

into LLM prompts that include the user's input value

proposition, identified similar organizations, and

suggestions for collaboration and strategy refinement

(OpenAI et al., 2024). By combining retrieval with

thematic analysis and LLM-driven recommendations,

EDEN provides a systematic, data-driven approach to

address incentive misalignments and foster

partnerships, see Code Snippet 2S. Furthermore,

cloud-based deployment enhances scalability and

accessibility, enabling EDEN to process larger datasets

and adapt to more intricate stakeholder ecosystems

while supporting real-time, iterative strategy

refinement (Khurana, 2014).

4 RESULTS

This section evaluates the effectiveness of EDEN in

addressing research questions by comparing its

ability to identify thematic clusters (RQ1) and

generate actionable collaboration recommendations

(RQ2) with alternative approaches: qualitative

content analysis and standalone LLM. The evaluation

is based on a given health insurer's value proposition;

see user input in Code Snippet 3S.

Scale-IT-up 2025 - Workshop on Scaling Up Care for Older Adults

1070

4.1 How Effectively Does EDEN

Identify Thematic Clusters in Value

Propositions? (RQ1)

Thematic clusters were identified using (A)

qualitative content analysis, (B) standalone LLM, and

analysis based on existing guidelines (Fried et al.,

2022) and manually coding each sentence of the user-

provided text. In contrast, the standalone LLM

generated thematic clusters based on its contextual

interpretation of the user's input. EDEN employed a

topic modeling technique, independent of external

contextual information, to categorize the user's input

systematically.

(A) Content Analysis

1. Guidelines for the well-being of older adults: 'A

companion for the next stage of your life' At our

health insurance, we understand that the transition to

retirement requires more than just financial

preparation.'

2. Coordination of care delivery: 'Our digital platform

is all about helping you make the new phase of your

life healthy and enriching.'

3. Evidence-based personalized healthcare: 'Advice

from experts Benefit from personalised advice from

our health coaches tailored to your needs as a retiree.'

'We offer comprehensive support for a fulfilling life

in retirement. We combine content and offers on

health topics that are specifically designed to meet the

interests of those approaching retirement.'

(B) Standalone LLM

1. Holistic support for retirement transition:

emphasize the need for preparation and support

during the transition to retirement life, addressing

physical, mental, and social health needs.

2. Health and wellness focus, personalized guidance:

enabling a fulfilling and enriching lifestyle by

offering tools and resources that cater to retirees'

interests and aspirations.

3. Expert-driven services: providing expert guidance

and advice customized to individual needs for healthy

aging.

1. Care, health, patient, insurance, platform, virtual,

support, academic, service, management

2. Research, aging, funding, innovation, foster,

community, institute, collaboration, program, public

3. Skin, data, tool, app, recommendation, acne,

therapy, analysis, personalized, artificial

4. Cell, cellular, therapy, disease, human, medicine,

allogeneic, mogrify, technology, type

5. Test, age, blood, dna, epigenetic, based, epiage, kit,

insight, methylation

6. Healthcare, network, health, within, management,

medical, developing, online, provider, behvavioral

7. Longevity, fund, company, offer, series, venture,

medicine, investment, disease, therapeutic

8. Life, science, health, healthy, people, digital,

solution, integrating, capability, goal

EDEN's topic modeling generated 8 clusters,

each containing 10 topics derived from the user input,

demonstrating comparable performance to alternative

methods by identifying similar overarching themes.

4.2 How Effectively Does EDEN

Generate Recommendations for

Collaboration? (RQ2)

Recommendations were generated respectively to

alternative approaches: Manual content analysis still

relied on identified guidelines and qualitative coding,

while recommendations from standalone LLM were

generated based on the model's contextual

understanding of the user's input and given dataset

(Mekniran, Giger, et al., 2024). At this stage, EDEN

coupled LLM with the computed results of network

analysis. EDEN for collaboration suggestions is

based on network insights built on similarity

computations.

(A) Content Analysis

The following recommendations reflect value

propositions derived from user-provided text, aligned

with the principles outlined in the WHO's Decade of

Healthy Ageing baseline report (WHO, 2020) and the

Global Roadmap for Healthy Longevity (Fried et al.,

2022).

EDEN: Towards a Computational Framework to Align Incentives in Healthy Aging

1071

1. Collaborate with governmental agencies: establish

partnerships to develop practical, measurable

indicators that focus on successful retirement,

including access and availability of equitable long-

term care.

2. Engage multilayered organizations: strengthen care

delivery for the aging population that consolidates

resources and expertise of retirement planning.

3. Partner with research centers: accelerate integrated

care for retirees by using clinical trials to introduce

novel interventions that address common age-related

challenges.

(B) Standalone LLM

The standalone LLM generated recommendations by

synthesizing several actionable collaboration areas

tailored to stakeholder groups (e.g., governmental

agencies, private platforms, digital health

companies); the following are extracted key points

from LLM responses that focused on collaboration.

1. Promote Age-Positive Attitudes and Combat

Ageism: Foster partnerships between advocacy

groups and NGOs to develop campaigns challenging

aging stereotypes and promoting inclusive narratives.

For example, AGE-WELL can work with age-tech

startups to develop assistive technologies for aging

populations.

2. Deliver Person-Centered and Preventive Care:

Collaborate with healthcare providers and digital

health companies to implement tailored health

interventions integrated into existing systems. For

example, InsideTracker can integrate with clinical

research networks (e.g., Max Planck Institute) to

validate their solutions.

3. Scale Digital Health Solutions: Partner with

research institutions and private platforms to adopt

AI-driven tools, wearable technology, and biomarker

insights for chronic disease prevention and

management. For example, Cedars Sinai to expand

their reach into underserved populations.

EDEN computed a similarity score between user

input "Your Idea," and government agencies was

53%, while the similarity with insurers was 52%, see

Figure 3. These scores indicate alignment with

thematic priorities such as public health and

preventive care, offering quantitative nuanced

insights.

1. Co-creating educational resources: Addressing

shared challenges in stakeholder education and

awareness.

2. Integrating health data from life science platforms:

Enabling data-driven personalization of health

services.

3. Bundling health and retirement planning services:

Combining offerings to enhance scalability and

address broader user needs.

5 DISCUSSION

This paper contributes to the Scale-IT-up 2025

workshop by providing a concrete example of how

advanced computational tools can address complex

systemic challenges in care implementation. Building

on literature that emphasizes the importance of multi-

stakeholder collaboration in healthy aging initiatives

(Cox & Faragher, 2022; National Academy of

Medicine, 2022; WHO, 2020), EDEN advances these

efforts by introducing a scalable framework capable

of dynamically mapping stakeholder incentives

beyond alternative approaches. A comparative

analysis with manual content analysis and standalone

LLM approaches underscores the advantages of

EDEN in identifying thematic groups (RQ1) and

generating personalized collaboration

recommendations (RQ2).

The evaluation demonstrates that EDEN

surpasses manual and standalone LLM approaches in

both granularity of value proposition analysis and

specificity of its similarity. For instance, it could

identify overarching themes beyond healthy aging,

such as preventive care and personalized health

services, and link actionable recommendations to

specific stakeholders with numerical ranking. By

coupling network analysis with similarity metrics,

EDEN highlights high-centrality stakeholders as

ecosystem integrators while uncovering hidden

collaboration opportunities with isolated nodes.

Unlike static frameworks or generic guidelines,

EDEN provides actionable strategies, such as co-

creating educational resources with government

agencies and bundling health and retirement services

with health insurers, tailored to stakeholder

interrelations and thematic overlaps.

Scale-IT-up 2025 - Workshop on Scaling Up Care for Older Adults

1072

Figure 3: Visualization of stakeholders as a node and collaboration as an edge between nodes.

Limitations

Although EDEN demonstrates clear advantages,

limitations remain. Reliance on curated datasets and

NMF-based topic modeling may restrict scalability

and semantic depth. Furthermore, the lack of broader

user studies limits the generalizability of its findings.

Future iterations should extend calculation to publicly

available datasets and integrate a dynamic temporal

network analysis to improve adaptability and

scalability. Testing EDEN performance in various

healthcare systems with user studies will also validate

its practical utility and usability in varied financial

and operational scenarios, such as direct to consumer

model, licensing, and subscription model.

6 CONCLUSIONS

This study introduces EDEN as a computational

framework to align stakeholder incentives and

advance healthy aging strategies. By integrating

network analysis and RAG-powered LLMs, EDEN

provides granular, user-specific recommendations,

surpassing manual and standalone LLM approaches

in generating actionable collaboration insights.

Future work should focus on scaling EDEN's

application to larger datasets, incorporating advanced

techniques like contextual embeddings and financial

database analysis, and validating its adaptability

across diverse healthcare ecosystems.

CONFLICTS OF INTEREST

WM and TK are affiliated with the Centre for Digital

Health Interventions (CDHI), a joint initiative of the

Institute for Implementation Science in Health Care,

University of Zurich; the Department of

Management, Technology, and Economics at the

Swiss Federal Institute of Technology in Zürich; and

the Institute of Technology Management and School

of Medicine at the University of St Gallen. CDHI is

funded in part by CSS, a Swiss health insurer,

MavieNext, an Austrian healthcare provider, and

MTIP, an equity firm investing in European

Healthtech companies. TK is a co-founder of

Pathmate Technologies, a university spin-off

company that creates and delivers digital clinical

pathways, he is no longer a shareholder since 2024.

However, neither Pathmate Technologies, MTIP nor

MavieNext were involved in this research.

REFERENCES

Altpeter, M., Schneider, E. C., & Whitelaw, N. (2014).

Examining Strategies to Build and Sustain Healthy

Aging Programming Collaboratives. Health Education

EDEN: Towards a Computational Framework to Align Incentives in Healthy Aging

1073

& Behavior, 41(1_suppl), 27S-33S.

https://doi.org/10.1177/1090198114537065

Béchard, P., & Ayala, O. M. (2024). Reducing

hallucination in structured outputs via Retrieval-

Augmented Generation.

https://doi.org/10.48550/ARXIV.2404.08189

Bengfort, B., Bilbro, R., Ojeda, T., & Bengfort, B. (with

Bilbro, R., & Ojeda, T.). (2018). Applied text analysis

with python: Enabling language-aware data products

with machine learning (First edition). O’Reilly.

Berwick, D. M., Nolan, T. W., & Whittington, J. (2008).

The Triple Aim: Care, Health, And Cost. Health

Affairs, 27(3), 759–769.

https://doi.org/10.1377/hlthaff.27.3.759

Bird, S., Klein, E., & Loper, E. (2009). Natural language

processing with Python. O’Reilly.

Bonnes, S. L. R., Strauss, T., Palmer, A. K., Hurt, R. T.,

Island, L., Goshen, A., Wang, L. Y. T., Kirkland, J. L.,

Bischof, E., & Maier, A. B. (2024). Establishing

healthy longevity clinics in publicly funded hospitals.

GeroScience. https://doi.org/10.1007/s11357-024-

01132-0

Chapman, P. (2000). CRISP-DM 1.0: Step-by-step data

mining guide. https://www.semanticscholar.org/paper/

CRISP-DM-1.0%3A-Step-by-step-data-mining-guide-

Chapman/54bad20bbc7938991bf34f86dde0babfbd2d5

a72

Cox, L. S., & Faragher, R. G. A. (2022). Linking

interdisciplinary and multiscale approaches to improve

healthspan—A new UK model for collaborative

research networks in ageing biology and clinical

translation. The Lancet Healthy Longevity, 3(5), e318–

e320. https://doi.org/10.1016/S2666-7568(22)00095-2

Fried, L. P., Wong, J. E.-L., & Dzau, V. (2022). A global

roadmap to seize the opportunities of healthy longevity.

Nature Aging, 1–4. https://doi.org/10.1038/s43587-

022-00332-7

Frow, P., & Payne, A. (2011). A stakeholder perspective of

the value proposition concept. European Journal of

Marketing, 45(1/2), 223–240.

https://doi.org/10.1108/03090561111095676

Gao, Y., Xiong, Y., Gao, X., Jia, K., Pan, J., Bi, Y., Dai, Y.,

Sun, J., Wang, M., & Wang, H. (2023). Retrieval-

Augmented Generation for Large Language Models: A

Survey (Version 5). arXiv.

https://doi.org/10.48550/ARXIV.2312.10997

Giger, O.-F., Pfitzer, E., Mekniran, W., Gebhardt, H.,

Fleisch, E., Jovanova, M., & Kowatsch, T. (2024).

Collaboration and Innovation Patterns in Diabetes

Ecosystems.

https://doi.org/10.1101/2024.04.25.24306351

Hsieh, H.-F., & Shannon, S. E. (2005). Three Approaches

to Qualitative Content Analysis. Qualitative Health

Research, 15(9), 1277–1288.

https://doi.org/10.1177/1049732305276687

Huang, Y., & Huang, J. (2024). A Survey on Retrieval-

Augmented Text Generation for Large Language

Models (Version 2). arXiv.

https://doi.org/10.48550/ARXIV.2404.10981

Jelodar, H., Wang, Y., Yuan, C., Feng, X., Jiang, X., Li, Y.,

& Zhao, L. (2019). Latent Dirichlet allocation (LDA)

and topic modeling: Models, applications, a survey.

Multimedia Tools and Applications, 78(11), 15169–

15211. https://doi.org/10.1007/s11042-018-6894-4

Khurana, A. (2014). Bringing Big Data Systems to the

Cloud. IEEE Cloud Computing, 1(3), 72–75.

https://doi.org/10.1109/MCC.2014.47

Larose, D. T. (2015). Data Mining and Predictive Analytics

(1st ed). John Wiley & Sons, Incorporated.

McKinney, W. (2010). Data Structures for Statistical

Computing in Python. Scipy.

https://doi.org/10.25080/Majora-92bf1922-00a

Mekniran, W., Giger, O.-F., Fleisch, E., Kowatsch, T., &

Jovanova, M. (2024). The Longevity Landscape: Value

Creation for Healthy Aging.

https://doi.org/10.1101/2024.05.28.24308017

Mekniran, W., & Kowatsch, T. (2023). Scalable Business

Models in Digital Healthy Longevity: Lessons from

Top-Funded Digital Health Companies in 2022. 609–

615. https://doi.org/10.5220/0011778400003414

Mekniran, W., Kramer, J.-N., & Kowatsch, T. (2024).

Reimagining Preventive Care and Digital Health: A

Paradigm Shift in a Health Insurance’s Role. 852–858.

https://doi.org/10.5220/0012400300003657

National Academy of Medicine. (2022). Global Roadmap

for Healthy Longevity (p. 26144). National Academies

Press. https://doi.org/10.17226/26144

OpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L.,

Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt,

J., Altman, S., Anadkat, S., Avila, R., Babuschkin, I.,

Balaji, S., Balcom, V., Baltescu, P., Bao, H., Bavarian,

M., Belgum, J., … Zoph, B. (2024). GPT-4 Technical

Report (arXiv:2303.08774). arXiv.

https://doi.org/10.48550/arXiv.2303.08774

Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V.,

Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P.,

Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., &

Cournapeau, D. (2011). Scikit-learn: Machine Learning

in Python. MACHINE LEARNING IN PYTHON.

Silge, J., & Robinson, D. (2017). Text mining with R: A

tidy approach (First edition). O’Reilly.

Sharma, A., & Kaur, B. (2017). A Research Review on

Comparative Analysis of Data Mining Tools,

Techniques and Parameters. International Journal of

Advanced Research in Computer Science, 8(7), 523–

529. https://doi.org/10.26483/ijarcs.v8i7.4255

Valdez, D., Pickett, A. C., & Goodson, P. (2018). Topic

Modeling: Latent Semantic Analysis for the Social

Sciences. Social Science Quarterly, 99(5), 1665–1679.

https://doi.org/10.1111/ssqu.12528

Wasserman, S., & Faust, K. (1994). Social Network

Analysis: Methods and Applications. Cambridge

University Press.

https://doi.org/10.1017/CBO9780511815478

WHO. (2020). Decade of healthy ageing: Baseline report.

World Health Organization.

https://apps.who.int/iris/handle/10665/338677.

Scale-IT-up 2025 - Workshop on Scaling Up Care for Older Adults

1074

APPENDIX

Code Snippet 1S: Text preprocessing and topic modeling

from nltk.corpus import stopwords

from nltk.stem.wordnet import

WordNetLemmatizer

from nltk.tokenize import word_tokenize

def preprocess(text):

tokens =

word_tokenize(text.lower())

tokens =

[WordNetLemmatizer().lemmatize(token)

for token in tokens if token.isalpha()

and token not in

set(stopwords.words('english'))]

return ' '.join(tokens)

# Topic modeling with TF-IDF and

training NMF

from sklearn.feature_extraction.text

import TfidfVectorizer

from sklearn.decomposition import NMF

# Vectorize the text data

doc_term_matrix =

TfidfVectorizer().fit_transform(df['Wha

t'].apply(preprocess))

# Apply NMF for topic modeling

nmf = NMF(n_components=8,

random_state=42).fit(doc_term_matrix)

Code Snippet 2S: Similarity computation and

recommendations engine.

import networkx as nx

import numpy as np

from sklearn.metrics.pairwise import

cosine_similarity

# Calculate cosine similarity

org_topic_matrix =

nmf.transform(doc_term_matrix)

cosine_sim =

cosine_similarity(org_topic_matrix)

# Build the graph

G = nx.Graph()

for i, seg1 in

enumerate(df['Organization']):

for j, seg2 in

enumerate(df['Organization']):

if i < j and cosine_sim[i][j] >

np.percentile(cosine_sim, 50):

G.add_edge(seg1, seg2,

weight=cosine_sim[i][j])

# Recommendations engine with RAG

import openai

# Define the prompt

collaboration_prompt = (

"Based on the identified similar

organizations and their value

propositions, provide succinct

suggestions in bullet points on how to

design a value proposition and

collaborate with these organizations. "

"Consider the following aspects: 1. Key

elements to include in the value

proposition. 2. Potential areas of

collaboration. 3. Benefits of

collaboration for both parties."

)

response =

openai.ChatCompletion.create(

model="gpt-3.5-turbo",

messages=[

{"role": "system", "content": "You

are an AI that helps identify how to

collaborate with other stakeholders to

enable healthy longevity." },

{"role": "user", "content":

collaboration_prompt + "Your value

proposition now: "+ user_input + "Your

identified similar organizations: "+

similar_scores }

)

Code Snippet 3S. User input prompt (151 words)

"A companion for the next stage of your

life

At our health insurance, we understand

that the transition to retirement

requires more than just financial

preparation. Our digital platform is

all about helping you to make the new

phase of your life healthy and

enriching.

Our service in a nutshell

Holistic preparation and health

We offer comprehensive support for a

fulfilling life in retirement. We

combine content and offers on health

topics that are specifically designed

to meet the interests of those

approaching retirement.

Advice from experts

Benefit from personalised advice from

our health coaches tailored to your

needs as a retiree.

Pilot phase and availability

Your health partner at any age

EDEN: Towards a Computational Framework to Align Incentives in Healthy Aging

1075

As your healthcare partner, we go

beyond the role of a health insurer and

is actively committed to healthy

ageing. Our service is an expression of

this commitment, with the aim of

supporting you in all phases of life."

Figure 4S: EDEN's user interface (eden.ethz.ch), users can input value propositions within 'incentivise' module to generate

network map and suggestion for collaboration. 'target' and 'scale' modules are under development to further EDEN's

computational capabilities, accessed on 12 December 2024.

Scale-IT-up 2025 - Workshop on Scaling Up Care for Older Adults

1076