TAM Example – Hypertension & Diabetes
Cost Analytics
Modus Health Group’s data and analytics are valuable to top healthcare organizations for benchmarking treatment costs and improving patient care. With the proper information, it is possible for healthcare risk-bearers to predict future trends and get ahead of consumer demand. Data that provides real time insights to anticipate the impact of changes in healthcare services and policy allows health organizations to determine crucial information about their patients such as average length of stay, population comorbidities, demographics and general risk.
Modus Health – Predictive Analytics
In the healthcare sector, risk adjustment and risk assessment are used as means to calculate payments for healthcare stakeholders based on the relative health of the at-risk population. Risk adjustment is necessary in order to ensure that risk-bearing entities are fairly compensated for the risks that they carry.
Risk
Cost
Outcome
Risk adjustment is usually done on a budget-neutral basis; this means that regardless of the proportion of high or low-risk patients enrolled, the net effect on total payments is unchanged. The degree of risk adjustment is determined via risk assessment, the method of examining the extent to which an individual’s risk deviates from the average population. Risk assessment is typically determined by an individual’s age, gender, recent illnesses and other biological factors.
Healthcare risk models will gather any relevant medical data about an at-risk population and use it to generate a numerical value for each patient. This value is then used to determine a “risk score” for the patient so that a weighted average can be calculated. This model informs the risk-bearer of the relative risk of one population to another, and the ratio of the average health spending for all individuals in the examined population. Next, we will examine how our model uses data from a one-year period to identify medical conditions and determine risk score:
| Indication | Share |
|---|---|
| Chronic Lower Respiratory | 8.03% |
| Diabetes | 18.27% |
| Hypertension | 31.50% |
| Kidney Disease | 30.74% |
| Obesity | 2.58% |
| Osteoporosis | 1.70% |
| Pain | 7.18% |
| ZIP Code | Share |
|---|---|
| Chronic Lower Respiratory | 4.82% |
| Diabetes | 18.37% |
| Hypertension | 39.16% |
| Kidney Disease | 2.11% |
| Obesity | 3.31% |
| Osteoporosis | 1.20% |
| Pain | 31.02% |
Our model uses data from any available time period to identify medical conditions and determine risk score. For example in the 77055 ZIP of Houston, Texas, 18.37% of doctor visits are for diabetes and 39.16% of visits are for hypertension. In comparison, Houston’s total case numbers are estimated to be 18.27% for diabetes and 31.50% for hypertension, showing a .1% and 7.6% difference. Out of a 77055 ZIP population of 43,691 people with a 104:100 male to female ratio, men are more likely than women to have hypertension (46.81% of cases vs. 33.51% of cases). This differs significantly in contrast to the total population where 31.36% of hypertension cases are for men (15.45% difference vs. total) and 31.62% are for women (1.89% difference vs. total). When it comes to diabetes, local women are slightly more likely to be affected, accounting for 19.37% of cases as opposed to 17.02% for men. In regards to the total Houston population, 17.97% of cases are for women and 18.85% are for men, demonstrating a slight 1.7% and 1.83% deviation.
From this data we can determine that amongst the local population, men possess a higher risk for hypertension than women, and that men and women are of similar risk for diabetes. For the total population, men and women seem to experience cases of diabetes and hypertension in near equal amounts. However, when it comes to comorbidities, gender is not the only factor that is worth considering in our data. For example, in regards to age, our data shows that 22% of doctor visits among 49-year olds are due to diabetes and 44.44% are due to hypertension. For the total Houston population of 49-year olds, those numbers are instead 16.55% for diabetes and 15.70% for hypertension. Out of all of our tracked conditions (kidney disease, chronic lower respiratory disease, obesity, osteoporosis, pain, diabetes etc.) men and women of 49 years of age are at highest risk for having hypertension in the 77055 ZIP, at rates 28.74% higher than the general Houston population.
Determining Risk Score:
In more detail, hypertension is tracked as either primary hypertension, hypertensive chronic kidney disease stages 1-4 or hypertensive heart disease without heart failure. 84.15% of local hypertension cases are for primary hypertension, 2.19% of cases are for hypertensive kidney disease stages 1-4 and 13.66% of cases are hypertensive heart disease without heart failure. In the total population, these values are 96.98%, .38% and 2.52% respectively. Our data shows that patients in the 77055 ZIP mirror the general Houston population by being at higher risk for primary hypertension as opposed to the other variants. With this information, if we were to determine the risk for a 49-year old man with diabetes and hypertension for the 77055 ZIP it would look something like this:
| Risk Markers for 77055 | Risk Weight |
|---|---|
| Male Age 49 | .42 |
| Type -1 Diabetes (with comorbidities) | 1.4 |
| Primary Hypertension | 3.1 |
| Total | 4.92 |
| Risk Markers for TOTAL | Risk Weight |
|---|---|
| Male Age 49 | .22 |
| Type-1 Diabetes | .8 |
| Primary Hypertension | 1.4 |
| Total | 2.42 |
Individual vs. Regional Risk Scores:
The 49-year old man’s risk score is the sum of his demographic risk weight plus the weights for his indicated conditions. In this example, if the 77055 ZIP population’s average risk weight is 2.3 and a 49-year old man has a risk score of 4.92, then his costs are expected to be 2.14 times that of his ZIP population. Furthermore, it is possible to check this individual’s risk score against the total Houston population:
Why Risk Adjustment Matters:
If the total population carries a risk score of 2.42 for diabetes and hypertension among 49-year old males, then the patient from 77055’s risk score is 1.4 times higher than the total population. In this way, risk assessment can be used to risk-adjust for payments and health treatments by tailoring to each region’s specific risk. Once a consistent number of high-risk patients in a region have been identified, the risk-bearer can then adjust their pricing accordingly. This can allow health treatment providers and insurers to lower their overall medical costs and increase their spending efficiency. Additionally, this reduces the effects of inadvertent and intentional risk selection so that risk-bearers can compete on the basis of medical efficiency, administrative efficiency and quality of service instead of on the ability to select risk.
At Modus, our health risk-assessment tool incorporates data on medical diagnoses, medical procedures, prescription drug use, age, gender, location and total health spending of each patient. According to a Society of Actuaries study, predictive models similar to ours are able to explain between 15 percent and 28 percent of variation in medical claim costs for individual patients. This is a 10 percent to 23 percent increase over using age and gender information only and scales with group size. Our risk assessment only increases in accuracy for larger groups, sometimes explaining more than 90 percent of variation for populations over 500. The scaling of accuracy is important because large groups of individuals are the most common targets for risk adjustment. This allows for risk adjustment to help providers reallocate premium income among plans in response to the health status of patients. Ultimately, this increases equity among competing plans, thus protecting plan solvency and reducing the need to avoid high-risk individuals.
Risk adjustment is absolutely necessary when multiple health plans are available to patients. When plans with high-cost-sharing requirements are offered alongside plans with low-cost sharing requirements, high-risk individuals tend to select plans with lower cost sharing. As a result, the differences in premiums may grow large between the high-cost and low-cost plans. Risk adjustment can be used to internally reallocate funds to adjust for this selection when premiums are reflective of plan-design differences but not patient bias. Medicare, Medicaid, and employer-based plans are often targets of payment adjustments reflective of patient demographics and historical diagnoses. It should be noted that after the passage of the Patient Protection and Affordable Care Act provisions related to risk adjustment were included. Healthcare plans created after March 23rd, 2010 in the individual and small group markets became subject to risk adjustment beginning in 2014. This means that states will adjust costs for plans with low-risk individuals and will provide payments to plans with enrollment of high-risk individuals.
It should be noted that the costs of constructing a predictive modeling system for low and high-risk patients are often highest the first year of implementation. The collection of both demographic and claims-based data that will be used as inputs to the model will require checking data for inaccuracy, standardizing data for consistency, storing the data on servers and processing it through the model. All of this comes at a cost; however, some of these expenses may be avoided if the data being collected is already used for other purposes. Once the start-up period for the model system has elapsed, the annual cost of maintaining the system should be considerably lower.
A well-designed risk-adjustment system is one that maximizes incentives and adequately protects risk-bearers. Balancing the tradeoffs between using available data within time constraints, maximizing the model’s predictive accuracy and minimizing the opportunity for gaming are crucial to success. At Modus Health, we aim to help you accomplish this with our data. All of our models are paired with data for product level medical tests and procedures covering all treatments and indications at the de-identified patient level. This information is gathered from every US geography captured by zip code and is sortable by age, gender and specialty.