The group life and disability insurance sector has been slower to adopt predictive analytics than other lines of insurance. One reason for the sector’s lag is because insurers often have limited information on who they are insuring. However, there are still many ways to incorporate predictive modeling technology to improve results. Milliman consultant Jennifer Fleck provides some perspective in her article “Group insurance ‘Project Insight’.”
In the world of long-term care (LTC) insurance, making financial projections is challenging for two main reasons: a long projection horizon and complex interactions. This article by Milliman actuaries Missy Gordon and Joe Long walks through the progression from developing LTC projection assumptions using traditional methods to doing so using predictive analytics.
This article was originally published in the December 2017 issue of Long-Term Care News.
We recently used machine learning techniques to understand key drivers of Medicare Shared Savings Program (MSSP) financial performance. Of the 190-plus objective accountable care organization (ACO) features reviewed, ACO baseline efficiency proved to be the most important financial performance driver we identified. Another way of putting it is that MSSP rewarded inefficient ACOs more than ACOs that have attained efficiency.
You may be asking, “How did you measure baseline efficiency?” The chart below tells an interesting story.
We analyzed ACO baseline efficiency by reviewing ACO baseline expenditures that were unadjusted, risk-adjusted, and geographic-risk-adjusted. Risk-adjusted per capita expenditures were adjusted to account for each ACO’s average risk score and mix of entitlement categories. Geographic risk-adjusted per capita expenditures were adjusted to account for Medicare reimbursement levels in each ACO’s area.
Below are a few interesting notes:
1. Despite adjusting for risk levels, mix of entitlement categories, and reimbursement levels, there is still significant variation in baseline per capita expenditures. See the third column above for this wide range of variation.
2. The Centers for Medicare and Medicaid Services (CMS) has already made MSSP rule changes that balance the rewards between ACOs at different levels of starting efficiencies. Past financial performance in MSSP agreement period 1 may not be a strong indicator of performance in agreement period 2. ACOs should understand how these rule changes affect them.
Beyond baseline efficiency, we found that several other features were strongly associated with gross savings:
1. National fee-for-service (FFS) trends higher than local market trends
2. Location in the Southeast and south central regions
3. Low performance year expenditures for short-term inpatient admissions
4. High baseline per capita expenditures, unadjusted
5. High CMS-hierarchical condition category (HCC) risk scores
However, we also found that these features still explained less than half of the variation in gross savings across ACOs. This may indicate that ACO care management efforts are accounting for some of the remaining variation.
Milliman PRM Analytics™ (PRM), a leader in data-driven value-based healthcare support systems, today announced that Cascade Health Alliance (CHA), a coordinated care organization (CCO) serving Klamath County, Oregon, has selected the PRM Platform and its suite of cloud-based analytic and population risk management solutions to support their growing clinical integration initiatives.
“By giving us greater ability to better manage our population health, the PRM tool allows for more efficient risk stratification and management,” said Peter Waziri, CHA’s Chief Financial Officer. “Working with Milliman and PRM Analytics will help CHA to better serve our members by allowing staff deeper insights to those members’ health information.”
“We are pleased to be selected by Cascade to help them manage their risk-based populations. Milliman continues to be the industry leader in helping providers manage population risk. PRM™ represents a disruptive approach to population stratification and management. The analysis focuses on the prospective opportunity for potentially avoidable costs so the patient care team can focus on them in advance. Care management can then be focused on the patients with the greatest potential to ‘bend the cost curve’ resulting in the optimal deployment of limited care management team resources,” said Art Wilmes, FSA, MAAA, a Principal and Consulting Actuary in Milliman’s Indianapolis office.
“I have access to new ways of seeing cross-sectional data and how it all works together,” said Angela Leach, CHA’s medical informatics analyst. “Case managers will use it to get at-a-glance profiles of patients they are caring for, and the quality management department can use it to find ‘hot spots’ that may benefit from additional programs.”
The case management team at CHA can use PRM Analytics in a variety of ways, according to Diane Barr, Director of Case Management. “We can identify high-cost members, plus we can filter for diagnosis and identify members for disease management,” Barr said. “We can also look at an individual member to determine their utilization, chronic conditions and other details. The best part is we can do risk stratification and identify members that are at the highest risk for re-hospitalization or emergency department utilization. The program is easy to use and provides us with volumes of useful information.”
In this Society of Actuaries (SOA) video, Milliman’s Shea Parkes discusses his experience learning predictive modeling through Kaggle Competitions and how he has applied what he learned to his professional work as a healthcare actuary.
Predictive analytics can improve medical outcomes by identifying patients needing medical interventions. Population segmentation is a common method used to identify such patients. Individuals are grouped into cohorts to help improve the quality of their care. In this paper, Milliman’s Jordan Paulus and Nick Creten explore four common methods for population segmentation: cost cohort segmentation, condition cohort segmentation, utilization cohort segmentation, and social cohort segmentation.