Python is a very popular coding language for doing predictive modeling and data science. We have been discussing python as part of our ongoing Predictive Analytics podcast series for the Society of Actuaries. Four episodes are available for listening now:
This is the first of a few podcasts focused on Python, a popular tool for predictive modeling and machine learning. In this episode, join Anders Larson, FSA, MAAA and Shea Parkes, FSA, MAAA as they cover some basics and talk about when Python may be a good idea to try.
Join hosts Anders Larson, FSA, MAAA, and Shea Parkes, FSA, MAAA, for the second in a series of podcasts focused on Python. After giving an introduction to the popular programming language in our previous episode, they discuss some key concepts in Python, such as its object-oriented framework, the idea of namespaces, the ability to create package sets and a few other topics.
Join hosts Anders Larson, FSA, MAAA, and Shea Parkes, FSA, MAAA, for the third in a series of podcasts focused on Python. Moving on from the foundational concepts and background from the prior two episodes, this episode moves into more practical advice for actuaries looking to get started with Python. The discussion includes pros and cons of various editing software and user interfaces. To provide some useful context, Shea also discusses the key considerations that his own team made as they implemented Python into their operations.
Join hosts Anders Larson, FSA, MAAA, and Shea Parkes, FSA, MAAA, for the fourth in a series of podcasts focused on Python. The previous episode covered how to get started with Python. This episode covers useful packages for data analysis. Python is a general purpose language at heart, so you will likely need to use a variety of packages to perform most data science tasks. Luckily, the Python ecosystem is full of feature-rich data science packages. Listen to this episode to learn about some of the most important ones.
Medicare Advantage plans have long recognized the
importance of coding to ensure accurate risk scores and clinical documentation. By contrast, Accountable Care Organizations (ACOs)
participating in the Medicare Shared Savings Program (MSSP) often focus less on
the implications of risk adjustment or assume they have little ability to affect
their risk scores. To understand why,
there are two key aspects of the MSSP to consider:
Regional adjustment: In the
early years of the MSSP, benchmarks were purely based on the ACO’s historical
experience. These benchmarks were not adjusted based on the absolute risk
scores for the ACO beneficiaries, but rather the change in risk scores over
time. Therefore, these ACOs were not
penalized if their risk scores were understated by a similar amount in each
year. This began to change in 2017 with the introduction of the regional
benchmark adjustment, in which the benchmark is adjusted for the absolute
difference between the ACO and the region average risk scores – making the
ACOs’ absolute risk scores an important factor.
However, before the Pathways to Success rule, this regional benchmark
adjustment only applied to ACOs that started a second agreement period in 2017
Demographic adjustment: Prior
to the Pathways to Success rule, the MSSP limited the risk score changes for
“continuously assigned” beneficiaries to be no more than a demographic
adjustment. Because of the way
continuously assigned beneficiaries were identified, this rule effectively
made it impossible for ACO efforts to improve coding accuracy to significantly
increase the ACO’s historical benchmark.
Given the considerations above, it is not surprising that
ACOs have historically deemphasized coding accuracy. However, the landscape is
now changing in ways that make risk scores even more important for two reasons:
the Pathways to Success rule, the
limitation on risk score changes for continuously assigned beneficiaries has
Instead, there is now a limit of 3% in the total increase in risk score
for the ACO from the historical benchmark to the performance year. This limit prevents ACOs from dramatically
increasing their benchmark strictly due to risk score increases, but it should
not deter ACOs from trying to ensure their risk scores are accurate and
the year to year risk score adjustment can be greater than the 3% cap. For
example, if the ACO’s underlying risk score decreases from baseline year 3
(BY3) to performance year 1 (PY1), then the risk score change from PY1 to PY2 could
be greater than 3%.
The figure below illustrates this possibility using a
hypothetical ACO that improves coding accuracy from PY1 to PY2.
In the “No BY3 to PY1 Risk Score Change”
scenario, the ACO’s historical benchmark is limited to a 3.0% risk score change
even though coding accuracy efforts led to an 8.3% risk score increase.
In the “Negative BY3 to PY1 Risk Score Change”
scenario, the ACO’s risk score would have decreased by 2.5% relative to BY3 if
risk scores remained flat from PY1 to PY2, but the coding accuracy efforts
actually led to a 5.8% risk score increase relative to BY3. Even though the change from BY3 to any PY is
capped at 3.0%, the value of the coding accuracy efforts is 5.5% (3.0% minus -2.5%).
of Risk Scores on Historical MSSP Results
Even before the implementation of the Pathways to Success rule, coding accuracy efforts could limit risk score decreases for the continuously assigned population. Based on a review of the 2014 through 2018 MSSP Public Use Files (PUFs), we found that approximately 49% of ACOs had an aggregate risk score change from Benchmark Year 3 (BY3) to Performance Year (PY) of less than 1.000. Although the PUFs do not split out the continuously assigned and newly assigned risk scores, it is likely that many of these ACOs had a decrease in their continuously assigned risk scores, which means these ACOs could have benefitted from coding accuracy efforts to mitigate this decrease. If these ACOs had all been able to maintain an aggregate risk score change of 1.000, the average gross savings across all ACOs from 2014-2018 would have increased from approximately 1.1% to 1.9%. The graph below shows the distribution of gross savings before and after applying a 1.000 floor to the risk score change.
The impact on shared savings depends on the specific
situation of each ACO, such as their minimum savings rate, Track selection, and
quality performance. However, for
example, if we applied the sharing parameters of the ENHANCED Track and a 2.0%
minimum savings rate/minimum loss rate (MSR/MLR) to each ACO, this would have
resulted in an average increase in average net shared savings/(losses) of $51
per beneficiary per year (PBPY). For a
20,000-life ACO, this would equate to approximately $1 million.
The benchmark for most ACOs in their second or later
agreement period, and ACOs starting under the Pathways to Success rule, is
adjusted based on the ACO’s risk-adjusted expenditures relative to all Medicare
FFS beneficiaries in the ACO’s region. The
regional benchmark adjustment is based on each ACO’s risk scores for their last
benchmark year. Therefore, the risk scores in an ACO’s last benchmark year are
particularly important. For an ACO starting an agreement period under Pathways
to Success in CY 2020, the risk-adjusted regional expenditures for their next
agreement period will be based on risk scores in CY 2024, which use diagnosis
codes from CY 2023. While not
immediately relevant for active ACOs, this will have a significant impact on
the benchmark for the next agreement period.
No matter their specific situation, ACOs should ensure
their assigned beneficiaries have accurately coded risk scores that reflect
their full morbidity level. As a result
of changes to MSSP rules over the past few years, particularly the Pathways to
Success rule, risk adjustment is becoming increasingly relevant to financial
results for ACOs participating in the MSSP, and neglecting the importance of accurate
risk score coding accuracy is a risk that is not worth taking.
 Any beneficiary who was
either assigned to the ACO or received a primary care service from any of the
ACO participants in the previous year was considered “continuously assigned.”
 Due to the limitations of the PUFs, this analysis does not precisely capture the impact of risk score decreases on each ACO’s benchmark, but approximates the overall impact across all ACOs. Some ACOs with a risk score decrease may still have been subject to the demographic adjustment on the continuously assigned population, meaning they may not have benefited from a higher performance year risk score. However, it is also possible that some ACOs with a risk score increase may have had a decrease in risk score for the continuously assigned population, and therefore they could have benefited from a higher performance year risk score.
Thank you to Grant Churchill for his contributions to the analyses in this post.
Telehealth is the use of
electronic information and telecommunication technologies to support remote
clinical healthcare, patient and professional health-related education, public
health, and health administration. Currently millions of Americans are residing
in areas with a shortage of primary healthcare providers and often experience
delays to see a provider.[i]
Telehealth is believed to improve access to healthcare for patients living in
both rural and urban areas and ensure that patients receive the right care at a
place and time most accessible to them. In addition, telehealth is also
believed to reduce healthcare costs by:
Optimizing staff distribution and healthcare
resources across healthcare facility or system
Reducing unnecessary office and emergency room
visits and hospital admissions
Reducing financial impact on providers in case of
no-shows by patients.[ii]
Currently, 31 states and the District of Columbia have parity laws that mandate commercial insurers pay for telehealth services.[ii] Unfortunately, there are barriers to wide adoption of telehealth. For example, Medicare generally still limits coverage and payment for many telehealth services, lagging behind other payers. [iii]
We analyzed one of MedInsight’s client’s data for 2010-2017 with over 3.9 million Medicaid, Medicare, and Commercial plan members, to explore the use of telehealth services. We only explored those services recognized by federal and commercial health plans [iv] as billable for telehealth services.
Trends in telehealth use
We analyzed the data to study the utilization of
telehealth services over 8 years and its distribution across different age
groups and gender (Figure 1, Figure 2 and Figure 3). We observed that:
the proportion of telehealth visits increased over recent years, it still
remains well below one percent
the years the average cost difference between telehealth and non-telehealth
services has increased from almost nil to about $40 per visit with telehealth
being cheaper than non-telehealth services
Figure 1: Trends in telehealth use
Figure 2: Average cost trend
were found to have used telehealth more frequently as compared to males
below 19 or above 64 years old were less likely to use telehealth services as
compared to members 19 to 64 years old. This could be due to parental preference
for their children, Medicare coverage limitations, or a matter of trust on the
conventional methods amongst the older age groups.
Figure 3: Telehealth use across age groups and gender
Specialty based use of telehealth
Apart from analyzing the telehealth use by patients, we
also analyzed the use across provider specialties (Figure 4) for the year 2017.
top ten specialties with highest number of telehealth visits constituted about 87%
of the total telehealth visits
practitioners were the most likely to use telehealth amongst all provider
specialties followed by clinical psychologist and psychiatrist
Figure 4: Telehealth use by specialty
Clinical Classifications Software category based use of telehealth
Instead of analyzing data at the individual diagnosis
level, we compared the telehealth use for the year 2017 at the Clinical Classifications Software (CCS) category[v]
level, (Figure 5) which provides a method for classifying diagnoses into
clinically meaningful categories.
top ten CCS categories with highest number of telehealth visits constituted
about 82% of the total telehealth visits
out of the top ten CCS categories were related to mental health and behavioral
disorders. This is in line with the top ten specialties with maximum telehealth
telehealth constituted below 2% of all included services, the rest being
accounted by other type of visits
Figure 5: Telehealth use by CCS diagnosis category
Although the analysis was based on limited administrative data, it illustrated that telehealth is a far less expensive option in comparison to the conventional face-to-face visits. Even a one percent shift from face-to-face visits to telehealth can save millions of dollars. However, members, payers, and providers need to start embracing telehealth as effective as any other form of healthcare service delivery method.
Required cookies help make a website usable by enabling basic functions like page navigation and access to secure areas of the website. The website cannot function properly without these cookies.
Analytics & Performance Cookies
Analytics cookies are used to collect information about how visitors use our site. The information gathered does not identify any individual visitor and is aggregated. It includes the number of visitors to our site, the sites that referred them to our site and the pages that they visited on our site. We use this information to help operate our site more efficiently, to gather broad demographic information and to monitor the level of activity on our site. Performance cookies are used to enhance the performance and functionality of our services but are non-essential to their use. However, without these cookies, certain functionality like videos may become unavailable.
These cookies are used when you share information using a social media sharing button or “like” button on our sites or you link your account or engage with our content on or through a social networking site such as Facebook, Twitter or Google+. The social network will record that you have done this. This information may be linked to targeting/ advertising activities.