JKJeroen Keppens

Jeroen Keppens

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Jeroen Keppens has a total of 109 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Recorded mental health recovery narratives for people with mental health problems and informal carers: the NEON research programme including 3 RCTs

    by Mike Slade, Stefan Rennick-Egglestone, Yasmin Ali, Chris Newby, Caroline Yeo, James Roe, Ada Hui, Ashleigh Charles, Laurie Hare-Duke, Luke Paterson, et al.

    Background: Personal narratives describing recovery from mental health problems are widely available to the public. We developed theory on the characteristics and impact of recovery narratives, developed curation procedures for the NEON Collection of 659 recovery narratives and developed and evaluated the NEON Intervention, a theory-informed web application providing access to the NEON Collection. Objectives: To evaluate the effectiveness and cost-effectiveness of the NEON Intervention as compared to usual care and whether this varies by prior health service usage. Design: Three pragmatic parallel-group randomised controlled trials of the NEON Intervention. Intervention arm participants received immediate access. Control arm participants received access after a 52-week follow-up. The effectiveness analysis was a linear regression model of outcome at 52 weeks. The cost-effectiveness analysis compared the incremental cost-effectiveness ratio to the £20,000–30,000 threshold defined in the National Institute for Health and Care Excellence reference case. All analyses were intention-to-treat and baseline-adjusted, with multiple imputation for missing data. Setting: England. Participants: All trials recruited people who were aged 18+ years, resident in England, capable of accessing or being supported to access the internet, able to understand written and spoken English and capable of providing online informed consent. NEON Trial participants also had experience of mental health-related distress in the last 6 months, and psychosis in the previous 5 years. NEON-O (i.e. non-psychosis) Trial participants also had experience of mental health-related distress in the last 6 months, but with no psychosis in the previous 5 years. People identifying as informal carers for people affected by mental health problems but not eligible for the NEON Trial or NEON-O Trial were recruited to the NEON-C feasibility trial. All inclusion criteria were self-rated. Recruitment was from March 2020 to March 2021, through public communications by the central study team, and the work of clinical support officers at 11 secondary care research sites. Interventions: The NEON Intervention has four narrative access mechanisms: theory-informed algorithmic recommendation, random selection, self-selection by narrative category and return to impactful narratives. Participants used the NEON Intervention as much as they wished. Main outcome measures: Primary outcome: quality of life (Manchester Short Assessment). Secondary outcomes: distress, hope, self-efficacy, meaning in life and health status. Results: For the NEON-O (i.e. non-psychosis) Trial, we found a significant baseline-adjusted difference of 0.13 (95% confidence interval 0.01 to 0.26, p = 0.041) in the Manchester Short Assessment score between intervention and control, and a significant baseline-adjusted difference of 0.22 (95% confidence interval 0.05 to 0.40, p = 0.014) in the presence subscale of the Meaning in Life Questionnaire. The incremental cost-effectiveness ratio was £12,526 per quality-adjusted life-year, lower than a threshold of £30,000 per quality-adjusted life-year used for health service commissioning in England. For participants who had used specialist mental health services at baseline, the intervention appeared to reduce cost, although confidence intervals were wide and results were not statistically significant (–£98, 95% credible interval –£606 to £309). It also improved quality-adjusted life-years (0.0165, 95% credible interval 0.0057 to 0.0273) per participant. Hence, for this subgroup of participants, it dominated usual care. For the NEON Trial, no significant baseline-adjusted differences in outcome were found. An incremental cost-effectiveness ratio of £110,501 was found for the NEON Intervention. A subgroup analysis provided preliminary evidence for greater cost-effectiveness for current mental health service users, with an incremental cost-effectiveness ratio of £35,013. The NEON-C Trial showed acceptability and feasibility for informal carers. It recommended integration of carer narratives and creation of an online carer community. Limitations: Participants were recruited during a period in which movement and social interaction were widely affected by the COVID-19 pandemic, with the potential to influence generalisability. For the NEON-O Trial, we had an unrepresentative proportion of female-gendered participants (79.3%). Therefore, our NEON-O Trial findings cannot be generalised. Conclusions: This research programme has shown promising findings from the testing of the NEON Intervention. There is further research to do before implementation can be suggested. Future work: The NEON Intervention should be evaluated through a randomised controlled trial with people experiencing psychosis and using mental health services. The NEON Intervention should be refined to suit the needs of carers and then evaluated through a randomised controlled trial. The NEON-O Trial should be repeated with narrower mental health populations (e.g. mood disorders, eating disorders) to refine knowledge on effectiveness and cost-effectiveness. This may include refining the narrative collection used with these populations. If the NEON Intervention is implemented on a larger scale for people with non-psychosis mental health problems, then studies should be conducted to monitor benefits, continuously assess safety and documentation implementation processes. Future studies should consider alternative forms for presenting recovery narratives, including through multilanguage or multiculture support, and addressing digital exclusion by providing access through widely available technologies, such as smartphones and text messaging. Longitudinal designs are needed to document the short-term, medium-term and long-term impacts of recovery narratives. Trial registration: This trial is registered as NEON Trial ISRCTN11152837, NEON-O Trial ISRCTN63197153 and NEON-C Trial ISRCTN76355273. Funding: This award was funded by the National Institute for Health and Care Research (NIHR) Programme Grants for Applied Research programme (NIHR award ref.: RP-PG-0615-20016) and is published in full in Programme Grants for Applied Research; Vol. 13, No. 9. See the NIHR Funding and Awards website for further award information.


  • Open Access

    A Survey of Cognitive Distortion Detection and Classification in NLP

    by Archie Sage, Jeroen Keppens, Helen Yannakoudakis

    Archie Sage, Jeroen Keppens, Helen Yannakoudakis. Findings of the Association for Computational Linguistics: EMNLP 2025. 2025.


  • Open Access

    Effectiveness and cost-effectiveness of online recorded recovery narratives in improving quality of life for people with psychosis experience (NEON Trial): a pragmatic randomised controlled trial

    by Mike Slade, Stefan Rennick-Egglestone, Clare Robinson, Chris Newby, Rachel A. Elliott, Yasmin Ali, Caroline Yeo, Tony Glover, Sean P. Gavan, Luke Paterson, et al.


  • Open Access

    The Implementation of Recommender Systems for Mental Health Recovery Narratives: Evaluation of Use and Performance

    by Emily Slade, Stefan Rennick-Egglestone, Fiona Ng, Yasuhiro Kotera, Joy Llewellyn-Beardsley, Chris Newby, Tony Glover, Jeroen Keppens, Mike Slade

    Background: Recommender systems help narrow down a large range of items to a smaller, personalized set. NarraGive is a first-in-field hybrid recommender system for mental health recovery narratives, recommending narratives based on their content and narrator characteristics (using content-based filtering) and on narratives beneficially impacting other similar users (using collaborative filtering). NarraGive is integrated into the Narrative Experiences Online (NEON) intervention, a web application providing access to the NEON Collection of recovery narratives. Objective: This study aims to analyze the 3 recommender system algorithms used in NarraGive to inform future interventions using recommender systems for lived experience narratives. Methods: Using a recently published framework for evaluating recommender systems to structure the analysis, we compared the content-based filtering algorithm and collaborative filtering algorithms by evaluating the accuracy (how close the predicted ratings are to the true ratings), precision (the proportion of the recommended narratives that are relevant), diversity (how diverse the recommended narratives are), coverage (the proportion of all available narratives that can be recommended), and unfairness (whether the algorithms produce less accurate predictions for disadvantaged participants) across gender and ethnicity. We used data from all participants in 2 parallel-group, waitlist control clinical trials of the NEON intervention (NEON trial: N=739; NEON for other [eg, nonpsychosis] mental health problems [NEON-O] trial: N=1023). Both trials included people with self-reported mental health problems who had and had not used statutory mental health services. In addition, NEON trial participants had experienced self-reported psychosis in the previous 5 years. Our evaluation used a database of Likert-scale narrative ratings provided by trial participants in response to validated narrative feedback questions. Results: Participants from the NEON and NEON-O trials provided 2288 and 1896 narrative ratings, respectively. Each rated narrative had a median of 3 ratings and 2 ratings, respectively. For the NEON trial, the content-based filtering algorithm performed better for coverage; the collaborative filtering algorithms performed better for accuracy, diversity, and unfairness across both gender and ethnicity; and neither algorithm performed better for precision. For the NEON-O trial, the content-based filtering algorithm did not perform better on any metric; the collaborative filtering algorithms performed better on accuracy and unfairness across both gender and ethnicity; and neither algorithm performed better for precision, diversity, or coverage. Conclusions: Clinical population may be associated with recommender system performance. Recommender systems are susceptible to a wide range of undesirable biases. Approaches to mitigating these include providing enough initial data for the recommender system (to prevent overfitting), ensuring that items can be accessed outside the recommender system (to prevent a feedback loop between accessed items and recommended items), and encouraging participants to provide feedback on every narrative they interact with (to prevent participants from only providing feedback when they have strong opinions).


  • Open Access

    User-Centered Development of STOP (Successful Treatment for Paranoia): Material Development and Usability Testing for a Digital Therapeutic for Paranoia

    by Che-Wei Hsu, Daniel Stahl, Elias Mouchlianitis, Emmanuelle Peters, George Vamvakas, Jeroen Keppens, Miles Watson, Nora Schmidt, Pamela Jacobsen, Philip McGuire, et al.

    Background: Paranoia is a highly debilitating mental health condition. One novel intervention for paranoia is cognitive bias modification for paranoia (CBM-pa). CBM-pa comes from a class of interventions that focus on manipulating interpretation bias. Here, we aimed to develop and evaluate new therapy content for CBM-pa for later use in a self-administered digital therapeutic for paranoia called STOP (“Successful Treatment of Paranoia”). Objective: This study aimed to (1) take a user-centered approach with input from living experts, clinicians, and academics to create and evaluate paranoia-relevant item content to be used in STOP and (2) engage with living experts and the design team from a digital health care solutions company to cocreate and pilot-test the STOP mobile app prototype. Methods: We invited 18 people with living or lived experiences of paranoia to create text exemplars of personal, everyday emotionally ambiguous scenarios that could provoke paranoid thoughts. Researchers then adapted 240 suitable exemplars into corresponding intervention items in the format commonly used for CBM training and created 240 control items for the purpose of testing STOP. Each item included newly developed, visually enriching graphics content to increase the engagement and realism of the basic text scenarios. All items were then evaluated for their paranoia severity and readability by living experts (n=8) and clinicians (n=7) and for their item length by the research team. Items were evenly distributed into six 40-item sessions based on these evaluations. Finalized items were presented in the STOP mobile app, which was co-designed with a digital health care solutions company, living or lived experts, and the academic team; user acceptance was evaluated across 2 pilot tests involving living or lived experts. Results: All materials reached predefined acceptable thresholds on all rating criteria: paranoia severity (intervention items: ≥1; control items: ≤1, readability: ≥3, and length of the scenarios), and there was no systematic difference between the intervention and control group materials overall or between individual sessions within each group. For item graphics, we also found no systematic differences in users’ ratings of complexity (P=.68), attractiveness (P=.15), and interest (P=.14) between intervention and control group materials. User acceptance testing of the mobile app found that it is easy to use and navigate, interactive, and helpful. Conclusions: Material development for any new digital therapeutic requires an iterative and rigorous process of testing involving multiple contributing groups. Appropriate user-centered development can create user-friendly mobile health apps, which may improve face validity and have a greater chance of being engaging and acceptable to the target end users.


  • Open Access

    The Implementation of Recommender Systems for Mental Health Recovery Narratives: Evaluation of Use and Performance (Preprint)

    by Emily Slade, Stefan Rennick-Egglestone, Fiona Ng, Yasuhiro Kotera, Joy Llewellyn-Beardsley, Chris Newby, Tony Glover, Jeroen Keppens, Mike Slade

    Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet #Preprint #PeerReviewMe: Warning: This is a unreviewed preprint. Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn. Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period. Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author). Background: Recommender systems help narrow down a large range of items to a smaller, personalized set. NarraGive is a first-in-field hybrid recommender system for mental health recovery narratives, recommending narratives based on their content and narrator characteristics (using content-based filtering) and on narratives beneficially impacting other similar users (using collaborative filtering). NarraGive is integrated into the Narrative Experiences Online (NEON) intervention, a web application providing access to the NEON Collection of recovery narratives. Objective: This study aims to analyze the 3 recommender system algorithms used in NarraGive to inform future interventions using recommender systems for lived experience narratives. Methods: Using a recently published framework for evaluating recommender systems to structure the analysis, we compared the content-based filtering algorithm and collaborative filtering algorithms by evaluating the accuracy (how close the predicted ratings are to the true ratings), precision (the proportion of the recommended narratives that are relevant), diversity (how diverse the recommended narratives are), coverage (the proportion of all available narratives that can be recommended), and unfairness (whether the algorithms produce less accurate predictions for disadvantaged participants) across gender and ethnicity. We used data from all participants in 2 parallel-group, waitlist control clinical trials of the NEON intervention (NEON trial: N=739; NEON for other [eg, nonpsychosis] mental health problems [NEON-O] trial: N=1023). Both trials included people with self-reported mental health problems who had and had not used statutory mental health services. In addition, NEON trial participants had experienced self-reported psychosis in the previous 5 years. Our evaluation used a database of Likert-scale narrative ratings provided by trial participants in response to validated narrative feedback questions. Results: Participants from the NEON and NEON-O trials provided 2288 and 1896 narrative ratings, respectively. Each rated narrative had a median of 3 ratings and 2 ratings, respectively. For the NEON trial, the content-based filtering algorithm performed better for coverage; the collaborative filtering algorithms performed better for accuracy, diversity, and unfairness across both gender and ethnicity; and neither algorithm performed better for precision. For the NEON-O trial, the content-based filtering algorithm did not perform better on any metric; the collaborative filtering algorithms performed better on accuracy and unfairness across both gender and ethnicity; and neither algorithm performed better for precision, diversity, or coverage. Conclusions: Clinical population may be associated with recommender system performance. Recommender systems are susceptible to a wide range of undesirable biases. Approaches to mitigating these include providing enough initial data for the recommender system (to prevent overfitting), ensuring that items can be accessed outside the recommender system (to prevent a feedback loop between accessed items and recommended items), and encouraging participants to provide feedback on every narrative they interact with (to prevent participants from only providing feedback when they have strong opinions).

  • Open Access

    User-Centered Development of STOP (Successful Treatment for Paranoia): Material Development and Usability Testing for a Digital Therapeutic for Paranoia (Preprint)

    by Che-Wei Hsu, Daniel Stahl, Elias Mouchlianitis, Emmanuelle Peters, George Vamvakas, Jeroen Keppens, Miles Watson, Nora Schmidt, Pamela Jacobsen, Philip McGuire, et al.

    BACKGROUND Paranoia is a highly debilitating mental health condition. One novel intervention for paranoia is cognitive bias modification for paranoia (CBM-pa). CBM-pa comes from a class of interventions that focus on manipulating interpretation bias. Here, we aimed to develop and evaluate new therapy content for CBM-pa for later use in a self-administered digital therapeutic for paranoia called STOP (“Successful Treatment of Paranoia”). OBJECTIVE This study aimed to (1) take a user-centered approach with input from living experts, clinicians, and academics to create and evaluate paranoia-relevant item content to be used in STOP and (2) engage with living experts and the design team from a digital health care solutions company to cocreate and pilot-test the STOP mobile app prototype. METHODS We invited 18 people with living or lived experiences of paranoia to create text exemplars of personal, everyday emotionally ambiguous scenarios that could provoke paranoid thoughts. Researchers then adapted 240 suitable exemplars into corresponding intervention items in the format commonly used for CBM training and created 240 control items for the purpose of testing STOP. Each item included newly developed, visually enriching graphics content to increase the engagement and realism of the basic text scenarios. All items were then evaluated for their paranoia severity and readability by living experts (n=8) and clinicians (n=7) and for their item length by the research team. Items were evenly distributed into six 40-item sessions based on these evaluations. Finalized items were presented in the STOP mobile app, which was co-designed with a digital health care solutions company, living or lived experts, and the academic team; user acceptance was evaluated across 2 pilot tests involving living or lived experts. RESULTS All materials reached predefined acceptable thresholds on all rating criteria: paranoia severity (intervention items: ≥1; control items: ≤1, readability: ≥3, and length of the scenarios), and there was no systematic difference between the intervention and control group materials overall or between individual sessions within each group. For item graphics, we also found no systematic differences in users’ ratings of complexity (P=.68), attractiveness (P=.15), and interest (P=.14) between intervention and control group materials. User acceptance testing of the mobile app found that it is easy to use and navigate, interactive, and helpful. CONCLUSIONS Material development for any new digital therapeutic requires an iterative and rigorous process of testing involving multiple contributing groups. Appropriate user-centered development can create user-friendly mobile health apps, which may improve face validity and have a greater chance of being engaging and acceptable to the target end users.

  • Open Access

    Impact of receiving recorded mental health recovery narratives on quality of life in people experiencing psychosis, people experiencing other mental health problems and for informal carers: Narrative Experiences Online (NEON) study protocol for three randomised controlled trials

    by Stefan Rennick-Egglestone, Rachel Elliott, Melanie Smuk, Clare Robinson, Sylvia Bailey, Roger Smith, Jeroen Keppens, Hannah Hussain, Kristian Pollock, Pim Cuijpers, et al.

    Background: Mental health recovery narratives have been defined as first-person lived experience accounts of recovery from mental health problems which refer to events or actions over a period of time and which include elements of adversity or struggle, and also self-defined strengths, successes or survival. They are readily available in invariant recorded form, including text, audio or video. Previous studies have provided evidence that receiving recorded recovery narratives can provide benefits to recipients. This protocol describes three pragmatic trials that will be conducted by the Narrative Experiences Online (NEON) study using the NEON Intervention, a web application that delivers recorded recovery narratives to its users. The aim of the NEON Trial is to understand whether receiving online recorded recovery narratives through the NEON Intervention benefits people with experience of psychosis. The aim of the NEON-O and NEON-C trials is to evaluate the feasibility of conducting a definitive trial on the use of the NEON Intervention with people experiencing non-psychosis mental health problems and those who care for others experiencing mental health problems respectively. Methods: The NEON Trial will recruit 683 participants with experience of psychosis. The NEON-O Trial will recruit at least 100 participants with experience of non-psychosis mental health problems. The NEON-C Trial will recruit at least 100 participants with experience of caring for others who have experienced mental health problems. In all three trials, participants will be randomly allocated into one of two arms. Intervention arm participants will receive treatment as usual plus immediate access to the NEON Intervention for 1 year. Control arm participants will receive treatment as usual plus access to the NEON Intervention after 1 year. All participants will complete demographics and outcome measures at baseline, 1 week, 12 weeks and 52 weeks. For the NEON Trial, the primary outcome measure is the Manchester Short Assessment of Quality of Life at 52 weeks, and secondary outcome measures are the CORE-10, Herth Hope Index, Mental Health Confidence Scale and Meaning in Life Questionnaire. A cost-effectiveness analysis will be conducted using data collected through the EQ-5D-5 L and the Client Service Receipt Inventory. Discussion: NEON Trial analyses will establish both effectiveness and cost-effectiveness of the NEON Intervention for people with experience of psychosis, and hence inform future clinical recommendations for this population. Trial registration: All trials were prospectively registered with ISRCTN. NEON Trial: ISRCTN11152837. Registered on 13 August 2018. NEON-C Trial: ISRCTN76355273. Registered on 9 January 2020. NEON-O Trial: ISRCTN63197153. Registered on 9 January 2020.


  • Open Access

    A formalisation and prototype implementation of argumentation for statistical model selection

    by Isabel Sassoon, Sebastian Zillessen, Jeroen Keppens, Peter McBurney

    The task of data collection is becoming routine in many disciplines and this results in increased availability of data. This routinely collected data provides a valuable opportunity for analysis with a view to support evidence based decision making. In order to confidently leverage the data in support of decision making the most appropriate statistical method needs to be selected, and this can be difficult for an end user not trained in statistics. This paper outlines an application of argumentation to support the analysis of clinical data, that uses Extended Argumentation Frameworks in order to reason with the meta-level arguments derived from preference contexts relevant to the data and the analysis objective of the end user. We outline a formalisation of the argument scheme for statistical model selection, its critical questions and the structure of the knowledge base required to support the instantiation of the arguments and meta-level arguments through the use of Z notation. This paper also describes the prototype implementation of argumentation for statistical model selection based on the Z specification outlined herein.