Department of Clinical Research

Background

Due to well-known sociodemographic changes and the rising prevalence of multimorbidity, the demand for palliative and end-of-life care is expected to increase significantly in the coming years. This is particularly driven by the growing incidence of frailty, dementia, and chronic, progressive, and life-limiting conditions such as heart failure, lung diseases (e.g., chronic obstructive pulmonary disease [COPD]), and multimorbidity. Digital health technologies have the potential to facilitate automated detection of distress in palliative care, including symptoms like pain, breathlessness (dyspnea), panic, agitation, and delirium. Such advancements could alleviate the suffering of cognitively impaired patients who are unable to call for help, for instance, by pressing an alarm button, yet urgently require professional intervention (e.g., administration of medication for pain or breathlessness).

Aim

This observational feasibility study aims to assess if data output from a sensor system consisting of a 3-D-radar, photoplethysmographic sensors (wearables), and audio detection (microphones) is associated with patient’s distress events as identified by medical professionals during standard care.

Research Plan

We will utilize well-established sensor systems, including Qumea® and wearables, to monitor heart rate, respiratory rate, heart rate variability, body movements, postures, and audio signals, such as changes in sound intensity or voice frequency. These data will be compared against the current gold standard for detecting distress in cognitively impaired patients: assessments conducted by trained healthcare professionals, specifically specialist palliative care nurses and physicians. Identifying correlations between sensor data and professional distress assessments could facilitate future validation studies, potentially leading to the development of automated distress detection systems using sensor arrays. This advancement would significantly improve distress detection rates and reduce patient suffering.

Abstract

Background:

Palliative care aims to alleviate suffering, improve quality of life, and address distressing symptoms in people with advanced disease. Despite its clinical importance, many interventions—particularly pharmacological approaches—are supported by limited high-quality evidence, with recommendations often relying on expert consensus or long-standing routine. Clinical trials are essential to strengthen the evidence base, yet research in palliative care faces well-described barriers such as recruitment challenges, underpowered studies, and premature termination. To our knowledge, no meta-research study has systematically mapped the clinical-trial landscape in palliative care. This review intends to establish a benchmark for research activity, identify under-represented areas, and outline priorities for future investigation, including recruitment expectations and attrition estimates.

Objectives:

This systematic review provides a comprehensive overview of clinical trials conducted in palliative care over the past two decades. We aim to (1) quantify the number, scope, and characteristics of registered and published trials; (2) identify which symptoms and interventions are studied more intensively, and compare this with the known prevalence of symptoms in advanced disease; and (3) document rates and reasons for non-completion or non-publication of registered studies.

Methods:

We will search major trial registries (ClinicalTrials.gov, WHO ICTRP, EudraCT) and bibliographic databases for interventional studies in adult palliative-care populations registered between 2000 and 2020. Eligible studies include prospective interventional designs; pediatric studies and disease-modifying cancer therapies will be excluded. Two reviewers will independently screen records and extract data using piloted forms. Discrepancies between registered and published outcomes will be assessed, and reasons for non-publication will be sought by contacting investigators. Descriptive statistics will summarize study characteristics, with subgroup analyses by symptom category, intervention type, sponsor, and region. Meta-analysis of publication rates will be conducted if appropriate.

Clinical prediction models can support surgical decision-making by identifying patient profiles that are at an increased risk of poor outcomes. While artificial intelligence (AI)-enabled decision aids demonstrated potential in enhancing shared decision-making and improving patient satisfaction, there remains a significant gap in implementation. Furthermore, few studies describe stakeholders’ preferences regarding the use of clinical prediction models. The adoption of decision support tools needs to be accelerated by (i) improving their transparency and reporting standards, (ii) addressing their implementation barriers, and (iii) aligning their development and implementation with stakeholder expectations to maximize their clinical impact.

Simultaneously, the growing deployment and development of modern machine learning and AI models have sparked debates about their performance in comparison to regression models, particularly when citing issues relevant for surgical databases, including varying sample size, atypical outcome distributions (for example ceiling effects in patient-reported outcome measures), and number and strength of predictive ability of prognostic factors.

The Clinical Prediction Models for Surgical Outcomes (CPMSO) project aims to address the challenges of implementation, data, and analysis by establishing standardized guidelines, ultimately improving the reliability and adoption of AI-driven decision support tools encompassing prediction models for surgical outcomes.

Research Objective (RO) 1: To develop and validate a consensus-driven set of recommendations that support the implementation of prediction models for surgical outcomes in patients with musculoskeletal disorders.

RO2: To assess the impact of data characteristics ((i) sample size, (ii) outcome distribution and (iii) number and predictive ability of prognostic factors) on the performance of machine-learning, regression, and Bayesian models for ordinal/continuous outcomes.

METHODS. In working package (WP) 1 related to RO1, we will (1) systematically map existing evidence on the implementation of prediction models in joint surgery through a scoping review, identifying methodological challenges and factors influencing their clinical adoption. We will also (2) explore the perceptions, needs, and implementation barriers of key stakeholders—including patients, healthcare professionals, and researchers— regarding prediction models for surgical outcomes, using focus group discussions. Based on these findings, we will then (3) develop and assess the relevance of a set of consensus recommendations addressing key issues for the implementation of prediction models for surgical outcomes. In WP2 related to RO2, the apparent and bootstrap-validated overall, discrimination and calibration performance of models predicting ordinal/continuous outcomes will be compared, including machine-learning, regression, and Bayesian models. First, data simulations will be generated, for which key data characteristics will be controlled ((i) sample size, (ii) outcome distribution and (iii) number and strength of predictive ability of prognostic factors). The validity of our findings will then be assessed using four nationwide surgical databases focusing on the prediction of pain (Numeric Rating Scale [0-10]) one year after surgery. Databases will be standardized to the Observational Health Data Sciences and Informatics (OHDSI) format, facilitating the further steps of analyses, which will include handling of missing data, predictor selection, model development, and internal validation.

EXPECTED RESULTS AND IMPACT. RO1, linked to WP1, will produce key recommendations for implementing prediction models for surgical outcomes, a crucial step toward enhanced decision-making and achieving better health and socioeconomic outcomes for patients and the society. RO2, linked to WP2, will provide methodological guidance on selecting optimal models for ordinal/continuous outcomes based on key data characteristics. The findings will be published in peer-reviewed journals and presented at international scientific conferences. The CPMSO project aims also to develop a professional network of experts, and support the integration of currently developed predictive analytics tools supporting surgical decision-making in Europe.