Galileo Risk Stratification Framework
The Galileo Risk Stratification Framework (GRSF) can be used to identify target populations for complex care and population health.
Many innovative models of care can succeed only with a well-selected target population. Providing systematic care rests on the following premise: proactive interventions today can reduce the likelihood of adverse events such as preventable hospitalizations and ED visits that are simultaneously costly and represent poor health outcomes. In order to achieve this goal, proactive management must be focused on patients that: 1) are likely to respond to the treatments and services offered by the program, and 2) have a likelihood of adverse events that can justify the up-front costs. An effective risk stratification methodology must prioritize patients according to both these requirements.
See these tables in the Population Health Model:
- Historical risk factors for model development: riskfactor
- Current risk level for each person: risklevel
- Current risk profiles for each person: riskprofile
See also:
Risk Factors Computed by GRSF
Galileo’s risk stratification model computes a broad portfolio of risk factors that are used to divide a population into cohorts of patients with a similar risk profile. Galileo’s library of risk factors includes indicators that measure a patient’s chronic disease load, recent utilization, medication therapies, clinical indicators of disease severity, and social determinants of health. The specific risk factors used to determine a patient’s risk level can be adjusted based on the capabilities and goals of the care team that will use the risk stratification model. The library of risk factors that are available to the model have been developed to address the following themes:
Realtime Availability. Galileo computes the risk stratification model on a weekly or nightly basis in order to re-prioritize patients according to recent changes in health status. To facilitate this near-realtime approach, risk factors must be able to detect acute events, medication changes, and clinical indicators that have occurred recently. Preferred risk factors can detect activity using a combination of historical claims, insurance authorizations, inpatient ADT, and EMR data. An estimate of a patient’s prospective risk level is considerably more accurate if it relies on an up-to-date picture of a patient’s current health status.
Actionable. Risk factors are preferred if they suggest a care management activity within the capabilities of the care team. For example, a risk factor that considers recent adverse events in combination with overall disease load can hint at a management approach. The primary diagnosis associated with the adverse event provides useful information about what may be driving recent changes in a patient’s health status within the context of a patient’s overall chronic disease. In contrast, a risk factor that detects “more than $100k costs in the prior year” is less suggestive of the next steps that should be considered for the patient.
Correlation with “Targeted Events”. Risk factors are selected based on how well they predict specific adverse events rather than overall utilization. The “targeted events” used to select risk factors should represent utilization that is expected to “respond” to the care team’s planned interventions; that is, these are the events that should be less likely given an effective care program. For example, conditions such as congestive heart failure and medication therapies such as diuretics that are correlated with preventable hospitalizations represent preferred risk factors. In contrast, conditions that suggest a future high cost without a high likelihood of a preventable adverse event are considered less useful.
The definition of “targeted events” that is used to tune the model should be selected based on the capabilities of the care team and the key process indicators that the team intends to use to measure success.
Transparency. It is expected that the recommended risk level will be used to augment the subjective clinical judgment of the care team that uses it. With this end in mind, risk factors are preferred if they can present a clinician with the specific criteria (i.e. significant chronic conditions, dates and primary diagnosis associated with relevant prior utilization, medication therapies, and the values of critical clinical indicators) that are driving the recommended risk level. Given the reasons for a risk level assignment, a clinician can quickly evaluate the quality and appropriateness of the risk factors used by the model. The potent combination of an algorithm that can prioritize a complex data set with the subjective judgment of a skilled clinician offers the best means to accurately assess a patient’s future risk.
