A framework for making predictive models useful in practice.

Area of Science:

• Health Services Research
• Clinical Informatics
• Predictive Analytics

Background:

• Predictive models can identify patients for interventions like advanced care planning (ACP).
• Healthcare delivery factors can significantly impact the effectiveness of these predictive models.
• Optimizing workflow execution is crucial for realizing the benefits of predictive analytics in healthcare.

Purpose of the Study:

• To analyze how healthcare delivery factors influence the net benefit of triggering an advanced care planning (ACP) workflow.
• To evaluate the impact of predictive model-based ACP workflow initiation on patient care.
• To compare the benefits of increasing inpatient ACP capacity versus developing outpatient ACP capabilities.

Main Methods:

• Developed a 12-month mortality predictive model using electronic health record data.
• Assessed the influence of nonclinical factors (e.g., patient capacity, discharge, outpatient availability) on ACP workflow net benefit.
• Quantified the relative benefits of inpatient versus outpatient ACP capacity expansion.

Main Results:

• Work capacity constraints and patient discharge timing can substantially decrease the net benefit of model-triggered ACP workflows.
• Implementing an outpatient ACP workflow can mitigate reductions in net benefit.
• Developing outpatient ACP capability is likely more beneficial than increasing inpatient ACP capacity, given resource limitations.

Conclusions:

• The practical utility of predictive models in healthcare hinges on analyzing their sensitivity to healthcare delivery factors.
• A framework is provided for quantifying the impact of delivery factors and capacity constraints on achieved benefits.
• Optimizing healthcare delivery workflows is essential for maximizing the value of predictive analytics in patient care.