Closing the Loop
in healthcare SM
Health Systems are continually evolving their underlying systems and processes which in turn impacts the availability and quality of the data to run the learning loop. New data opens the door to new data markers. Better data coverage enables more accurate prediction and risk adjustment models. Improved timing reduces the reliance on proxy data and speeds decision making. Data sharing agreements enables the development of more accurate models, benchmarks and best practices by leveraging data outside the walls of a single organization.
The ability to proactively identify target populations is critical for effective management of value-based care contracts. Target populations are used to identify high risk patients for enrollment in intervention programs and for constructing risk-adjusted comparison groups to measure outcomes. A one-size-fits-all approach is often insufficient as prediction models may perform well on some populations and not on others. For example, an intervention to reduce readmissions for heart failure patients may not be effective for patients with language barriers, low literacy or those that have unstable housing situations.
Loopback’s Mozart™ population engine powers the learning loop. Data from diverse sources can be used to identify target populations based on risk, cost or gross margin. The tagging architecture allows data from diverse sources to be converted into usable data markers. Mozart can quickly incorporate new and diverse data sources to enhance the identification of target populations and coordinate across multiple prediction models to ensure the most appropriate model is used for each patient.
Health Systems often manage many distinct interventions to address different risk factors. The Loopback Platform can leverage available data markers to match target populations to interventions based on patient risk and capacity constraints, cost and benefit characteristics. Higher cost interventions will need to be matched with high risk patients to achieve reasonable return on investment. Machine learning can help identify sub populations that are responsive to specific interventions and those that are not responsive. Patient groups that are not benefiting can be matched to other programs that are a better fit.
Intervention programs may be as simple as notifying a provider to enroll an eligible patient to more complex multi-step workflows that span departments or organizations. The Loopback Platform provides actionable information regarding intervention program performance including program fidelity, productivity of personnel and efficiency. Patients that are eligible for a particular program may choose to opt out or drop out before a program completes. By monitoring the engagement rate for subpopulations, intervention programs can be adjusted to overcome barriers to full participation.
The relevant outcomes for target populations and intervention programs will vary based on the organizations priorities and objectives. Common outcomes include capture rates, gross margin, episode costs, utilization (hospitalizations, readmissions, ED visits) and clinical measures such as medication adherence or target lab values. To evaluate intervention programs, baseline measures can be established for each target population. To isolate the impact of an individual intervention, outcomes of an engaged population can be compared to a population with similar characteristics that did not receive the intervention. The data generated from running intervention programs can be utilized to drive continuous improvement.