Expert opinion on predictive models to early identify older patients in need of palliative care: A Delphi study

Introduction

Access to palliative care (PC) has traditionally focused on patients facing their last days or weeks of life, a view which is at odds with the emphasis on early integration advocated by the World Health Organization (WHO) and also supported by a growing number of studies and policies.

One of the most important difficulties for integrated early PC care is the identification of patients in need, especially among older patients with non-malignant diseases. In this sense, InAdvance project (ref.: 825750) aims at developing stratification models to support the identification of older patients with complex chronic conditions who could benefit from early PC interventions.

The objective of this study is to explore the views of experts on the use of electronic routine data (ERD) from medical records for the early identification of patients in need of PC and for supporting clinical decision-making.

Methods

A two-round, online consensus Delphi study is being performed with an expert panel (n = 30) of researchers and front-line clinicians in PC or chronic care from 5 countries (Spain, United Kingdom, Greece, Netherlands and Belgium).

Results 

Round 1 highlighted that ERD was considered very useful by panellists as a population health management tool to improve PC (63.33%). However, the need of clinical validation and mistrust in terms of dearth of evidence were identified as critical aspects of the use of ERD on healthcare sector. The sample considered predicting 12-month mortality risk and predicting accumulation of health deficiencies as the most appropriate models to anticipate decline trajectories of chronic patients. But the use of automatic stratification tools was considered as preferable in combination with the expert opinion of healthcare staff.

Discussions

These findings will contribute to developing PC predictive models, selecting key clinical variables, geriatric syndromes, symptoms and health impacts as early indicators of decline trajectories to be included into those models, as well as to identifying the approach to follow for a feasible and usable use of these models by clinicians.

Conclusions

In this international Delphi study experts reached consensus on that a model predicting accumulation of health deficiencies is the most preferable and more useful ERD-based strategy to identify chronic patients that may benefit from early PC.

Lessons learned

The use of the Delphi methodology in the development of PC predictive models may facilitate international synergies and scaling-up and knowledge transfer opportunities about the potential use of Artificial Intelligence on the PC field.

Limitations

No widely accepted benchmark does exist within the research community of what constitutes an adequate level of consensus. However, researchers have established clear criteria for consensus (median ratings, IQR).

Suggestions for future research  

To explore the potentials of the analysis of very large datasets from routine registers, on poblational approach strategies aimed at early PC identification needs.

To explore opportunities when using automatic predictive models on older patients with moderate to severe chronic illnesses on clinical staff: improving prognostic certainty, saving time, as a decision-support systems for referral to integrated care pathways.