Projects
Mind – suicide prevention helpline
Together with Mind, one of Sweden’s largest suicide prevention helplines, we analyse helpline conversations – both chats and calls – using natural language processing. The aim is to better understand how suicidal crises unfold over time and which support strategies help, and to use these insights to improve crisis support.
Help-seekers remain anonymous, and the conversations are handled in a secure workflow approved by the Swedish Ethical Review Authority.
SOS Alarm – suicidal crises in emergency calls
Together with SOS Alarm and researchers in psychiatry at Lund University, we study how suicidal crises are communicated, identified and managed in calls to Sweden’s emergency number 112.
The nationwide study links emergency call data with national health registers. We contribute AI-based language analysis to identify communication patterns associated with later suicide and self-harm, and to improve the identification of suicide-related emergency calls.
Precision mental health
Sund Psykologi comprises three clinics in Malmö and Lund, providing therapy to more than 1000 patients/year.
Minding Health (founded by Oscar Kjell) provide language-based assessments in clinical settings.
PTSD and the World Trade Center Health Program
Together with psychiatry researchers at the World Trade Center Health Program, we study World Trade Center responders – a cohort followed for over twenty years since 9/11.
The research is multimodal: we combine how responders describe their lives and health in audio and video recorded clinical interviews and written responses with DNA and medical records. We develop AI-based assessments of posttraumatic stress disorder (PTSD) and of mental and physical health, and evaluate them with models pre-registered before testing in prospective clinical data, against clinical diagnoses and healthcare costs.
The LEADING Guideline
Accurate assessments, such as for depression, are essential but face many challenges. Similar assessment methods across psychology, psychiatry, and medicine involve experts reviewing several sources of longitudinal information to achieve best-estimate assessments. However, the quality of these assessments is often difficult to evaluate due to poor reporting of the assessment methods. Referring to an assessment as a best-estimate assessment (and sometimes even a gold standard) while vaguely reporting the assessment method is alarming.
To tackle this gap, we have developed the LEADING reporting guideline to support researchers in reporting research that aims to achieve best-estimate assessments.
Sequential Evaluation with Model Pre-registration
We developed the Sequential Evaluation with Model Pre-registration (SEMP), a study design procedure that combines good scientific practices (i.e., pre-registration) with robust artificial intelligence (AI) model development practices (i.e., data preprocessing, hyper-parameter tuning, and out-of-sample test).
Mental health assessments dashboard
How do clinicians assess depression in practice, and what do they think about standardized methods and AI? This interactive dashboard presents results from our survey of clinicians in the United States, the United Kingdom, the Netherlands, Sweden and other countries. You can explore which data collection methods (e.g., clinical interviews, rating scales) and data combination methods clinicians use, and filter the results by country, profession and clinical setting.
Implicit Motives
Implicit motives are subconscious needs driving behavior, dating back over 100 years. They have been resource-intensive to assess as it is typically done by manual coding of stories to ambiguous pictures. We work with automating this process for the big three needs, affiliation, achievement and power as well as for autonomy.
Well-being, balance and harmony in life
What makes a good life? Our well-being research goes beyond satisfaction with life to include harmony and balance – feeling that the different parts of one’s life fit together, and living in balance with other people and one’s surroundings.
We developed the Harmony in Life Scale, showed that people’s own descriptions of their well-being analysed with AI can approach the theoretical upper limits of rating scales, and explored how people across 154 nations understand balance and harmony. We also study who flourishes and who does not – for example in Sweden as part of the Global Flourishing Study – and what common measures such as the Cantril Ladder actually capture.
The Harmony in Life Scale: Translations and validations
The Harmony in Life Scale (HILS) consists of 5 items and has also been validated as a concise 3-item version, alongside a 3-item version of the Satisfaction with Life Scale (SWLS). The scale has been translated and validated in several languages, including Hindi, Turkish, Setswana and Swedish.
