Pilot study of an automated one-year mortality prediction tool to trigger Advance Care Planning

Rationale: If we had an accurate, automated, computer-based tool to identify patients with a limited prognosis, we could use this tool to trigger ACP and EOL interventions more appropriately and reliably. A recent Canadian study reported a highly accurate tool for predicting mortality using only retrospective administrative data.

Hypothesis: With small changes, this mortality prediction tool could be used prospectively at the time of hospital admission to identify patients with a limited prognosis and send automatic notifications to remind the medical team to address ACP or EOL needs.

Objectives: This is a pilot study. We aim to study the feasibility of implementing this tool, and the impact of the tool on physicians, other healthcare practitioners, patients and family members. We will also study a subset of patients to see whether the introduction of electronic notifications is associated with an increase in documented discussions about ACP or goals of care.

Research plan: In this project, we will develop an automated mortality prediction tool based on the HOMR score (pre-pilot, four months); study the feasibility of calculating the modified HOMR prospectively at the time of admission and use qualitative methods to determine how to convey prognostic information to the admitting team (phase one, four months); and study the impact of using the tool to identify patients at risk of death, from the perspective of all stakeholders (phase two, four months).

Key Findings

• We created a computer-based tool to identify people who are at high risk of death in the coming year.
• Patients, family members and healthcare providers said the tool was acceptable.
• The tool promoted communication, helped to avoid aggressive treatments, and engaged palliative care experts in the care planning.

Evidence gap addressed by study/Why this study was needed

One obstacle to improving end-of-life care is the failure of clinicians to recognize people who are nearing the end-of-life. There are many ways to improve end-of-life care, but many cannot be used until we know who would benefit from them. An automated tool that correctly identifies people nearing the EOL could improve the use and benefits of almost all palliative intervention.

Suggestions on how these findings could impact older adults living with frailty and/or their family caregivers and how this might be measured

1. The tool is ideally meant as the “trigger” for interventions that improve EOL care, such as comfort medications or advance care planning. We plan to study the tool in more detail for this purpose.
2. Although our tool appears to help change care, most patients identified by the tool did not receive any of the EOL interventions. It is possible that a different notification would be more effective for changing care.
3. This tool only works for people admitted to a hospital. Most Canadians are admitted in the final year of life, which means that this tool could identify many dying Canadians. However, other tools are needed to identify dying patients in other settings (e.g. outpatient clinics, long-term care, primary care).

Brief comment on type of study in lay terms/plain language

• We built a computerized tool that uses data collected at the time of admission to identify people at risk of death in the near future.
• For patients at high risk, we sent notices to their physician reminding them to assess and address unmet palliative needs.
• We interviewed patients and physicians to see how they used the tool and whether they found it useful.
• We looked at the medical records to see whether care changed for patients after we started sending notices.

Key Findings

• We generated a computerized tool to identify inpatients at elevated risk of mortality at the time of admission in two different hospitals using different health information systems.
• The tool was reliable and acceptable to patients, family members and healthcare providers alike.
• Our results suggest that the tool may be effective for promoting communication, avoiding aggressive treatments, and engaging palliative care experts in the management of Canadians with frailty and serious illness.

Evidence gap addressed by study/Why this study was needed

One important obstacle to improving end-of-life (EOL) care is the failure of clinicians to reliably identify those who are approaching the EOL, particularly for those with frailty and non-cancer illness. An accurate and automated mortality prediction tool could enhance the real-world effectiveness of any palliative intervention and facilitate research and quality improvement by reliably identifying people who might benefit and prompting staff to initiate interventions or address unmet palliative needs. It could also help close the EOL care gap that exists between those dying of cancer and those dying of non-cancer illness.

Suggestions on how administrators or policy makers could use the findings

1. We developed a simple, scalable, reliable and automated tool that appears to be effective for identifying frail and seriously ill inpatients at risk of mortality and improving their EOL care. Identifying this population is the critical first step in any quality improvement initiative focused on improving EOL care.
2. The tool is versatile and can be adapted to trigger other interventions deemed appropriate for patients nearing the EOL. It is not intended as a standalone intervention but is ideally suited to enhance the real-world effectiveness of any intervention that currently relies on providers to remember to initiate them.

Brief comment on type of study in lay terms/plain language

• We built a computerized tool that uses data collected at the time of admission to identify people at risk of death in the near future.
• For patients at high risk, we sent notices to their physician reminding them to assess and address unmet palliative needs.
• We interviewed patients and physicians to see how they used the tool and whether they found it useful.
• We looked at the medical records to see whether care changed for patients after we started sending notices.

Key Findings

• We generated a computerized tool to identify inpatients at elevated risk of mortality at the time of admission in two different hospitals using different health information systems.
• The tool was reliable and acceptable to patients, family members and healthcare providers alike.
• Our results suggest that the tool may be effective for promoting communication, avoiding aggressive treatments, and engaging palliative care experts in the management of Canadians with frailty and serious illness.

Evidence gap addressed by study/Why this study was needed

One important obstacle to improving end-of-life (EOL) care is the failure of clinicians to reliably identify those who are approaching the EOL. There are promising interventions for improving the care delivered to patients nearing the EOL. However, the patient identification and initiation of the intervention in these studies was performed by research staff who do not exist in the real-world clinical environment. An accurate and automated mortality prediction tool could enhance the real-world effectiveness of any palliative intervention and facilitate research and quality improvement by reliably identifying people who might benefit and prompting staff to initiate interventions or address unmet palliative needs.

Brief overview of the methodology

We developed and piloted a computerized application that uses a modified version of the Hospital One-Year Mortality Risk score (HOMR) to automatically identify inpatients at risk of one-year mortality using administrative data collected at the time of admission. For patients at elevated mortality risk, automated notices were sent to the admitting medical teams reminding them to assess and address unmet palliative needs. We used both qualitative and quantitative means to assess the tool in terms of feasibility, acceptability, effect and effectiveness. The study consisted of two phases: Phase 1 used the application invisibly, verifying that the tool worked correctly, and conducted qualitative interviews to gain the perspective of patients, family members, and healthcare providers about the idea of an automated prediction tool. In Phase 2, we sent notifications to admitting teams, after which we reviewed medical records to determine whether the notifications were associated with differences in care provided.

Potential impact of findings on clinical practice/patient care and how this impact might be measured

1. We identified a simple, scalable, reliable and automated tool appears to be effective for identifying frail and seriously ill inpatients at risk of mortality and improving their EOL care.
2. The tool is versatile and can be adapted to trigger other interventions deemed appropriate for patients nearing the EOL. It is not intended as a standalone intervention but is ideally suited to knowledge translation- enhancing the real-world effectiveness of any intervention that currently relies on providers to remember to initiate them.

Remaining knowledge/research gaps

1. The tool is not intended primarily as a standalone intervention, but it is ideally suited to knowledge translation for other EOL interventions.
2. Although the tool appears effective for changing care, most patients who triggered a notification did not receive an EOL intervention. It is possible that a different notification would be more effective for changing care.
3. This tool only works for inpatients. Most Canadians are admitted in the final year of life, which means that this tool could identify many dying Canadians. However, a more comprehensive prospective surveillance system would include tools designed to identify dying patients in other settings (e.g. outpatient clinics, long-term care, primary care).

CFN Webinar (January 23, 2019): Pilot study of an automated one-year mortality prediction tool to trigger Advance Care Planning