Identification of multimorbid patients with impactable risk profiles using artificial intelligence, the IM-PACT method
A relatively small group of patients with complex and long-term needs (CLN) for health services account for a large proportion of resource use in our health services. Roughly speaking, both Norwegian and international studies show that 1% of patients account for about 20% of consumption in the specialist health service, 5%, account for 50% of consumption, and 10% account for 66% of consumption. This group is characterized by the fact that they are older, have many concomitant long-term diagnoses, have a poorer quality of life, and have a high risk of death. Many authors have documented how health services are poorly adapted to people with CLN, and show how this contributes to both higher costs and poor quality of services for this group. Once a person has entered the group that needs the most health services, it is often difficult to reverse the process. There is therefore a need to use predictive tools to identify people in the risk groups who may benefit from preventive measures.
Generally, identifying these patients for treatment is based on risk prediction models. Several studies have shown that it is possible to correctly identify high-risk (high risk of death and complications) and high-need patients (high care utilization). There are also many studies that show how case management interventions can be useful in reducing risks and improving outcomes, but the results from these studies are inconsistent and difficult to reproduce. We believe this is due to a gap in our understanding: For this patient group, we do not know which interventions fit which sub-groups. Said in other words: being at 'high risk' is not the only requirement for enrolment into intensified care programs. Patients must also stand to benefit from available interventions. By this new way of thinking, patients must be 'impactible' for the preventative care, treatment, or care pathway.
Impactibility can be defined as the degree to which different subpopulations will benefit from a range of interventions. Identification of high risk is only half of the equation since impact may be highest before high-risk is detectable by usual methods. This project proposes a novel impactibility model IM-PACT, based on artificial intelligence, which will formally define clinical subgroups of patients with CLNs whose risk profiles can be impacted by one of the treatments shown to have a beneficial effect on this group of patients, The Patient Centred Care Team (PACT).
This research project uses artificial intelligence (AI) methods to analyze data that has been collected over six years in the PACT cohort study to provide an improved and efficient way to identify multimorbid patients with impactable risk profiles (IM-PACT). The proposed IM-PACT process is based on identifying patients whose risk profiles can be impacted by the treatment. The results have great potential to improve care and reduce costs for elderly patients with complex needs, and this ultimately leads to better health services for the population.