Claims trend analysis is an important component of maintaining financial stability. These retrospective analyses may be used in analyzing cost increases to determine which types of services are showing greater than desired or expected increases. The trend analyses can also be used to determine reasons for variation of actual results from forecast or budgeted results. Maybe most importantly, these trend analyses can be used as a basis for estimation of trends in future years to enhance the accuracy of forecasts and rates.
There are a variety of methods that can be used to analyze claims trends. It is important that the actuary fully understand the method being used, its strengths and its weaknesses. Trend analysis requires much thought and is not a number-crunching exercise. Failure to analyze properly may result in trends that are not as accurate as possible.
The narrative below describes a method using moving averages in order to analyze trends.
Moving Average Trend Analysis
The Moving Average Trend Analysis uses multiple comparison of incurred claims costs per unit for experience periods to incurred claims costs per unit for other periods of the same length, usually 1, 3, 6, or 12 months. Trend analyses usually compares multiple periods to get an idea of how actual trends are changing over time. Using 12-month rolling averages would eliminate the necessity of having to apply seasonality and days adjustment factors but lacks the sensitivity to recent changes in trends that a 3-month model would show. Although more sensitive to recent trend changes, the 3-month model would be affected more by artificially low or high incurred months. For example, January 2013 has an inordinate amount of flu and respiratory disease as compared with January of previous years and previous months. Using a 3-month model would only divide this artificially high claims amount by 3 (assuming February 2013 and December 2012 were "normal" months) while the 12-month model would divide the artificially high month by 12. The 6-month model is more consistent than the 3-month model and more sensitive than the 12-month model. The answer to which model to use is dependent on the person performing the analysis and the situation surrounding the data.
Enrollment or Member Cost Units
The units by which the incurred claims will be divided can be obtained in a variety of methods. The most accurate is the use of member cost units. Member cost units reflect the anticipated claims of the given enrollees, be it members or other unit. For example, a policy with two adults has the same number of members that a policy with one adult and one child has, namely 2, but it is probably expected that the child would have less claims than the adults for most types of services. Use of a 1.1 member cost unit value for adults and a 0.7 member cost unit value for the child may reflect costs more accurately than using member counts. In certain situations the actuary may even go so far as to apply age factors to all members in order to obtain the cost units. So, for example, our 2 adult policy may have a member cost unit of 2 x 1.1 = 2.2 while the policy with 1 adult and 1 child may have a total member cost unit value of 1.1 + 0.7 = 1.8. If further detailed calculations are desired then the actuary may use age factors to obtain 4.6 member cost units (2 x 2.3) for a 2 adult policy containing 2 adults aged 62, if the assumed age factor for an adult aged 62 is 2.300..
The claims costs that appear in the numerator of the claims cost per unit are incurred claims. These incurred claims are usually estimated for each month. The actuary generally has a claims reserve model which uses a completion factor or a lag factor approach to estimating incurred claims. For the sake of this narrative we will call these incurred estimates the gross monthly incurred claims.
Unit Cost Index
Although the monthly gross incurred claims can be used as a basis for trend analysis, modifications to these claims can lead to more accurate analysis. For example, a Monthly Unit Cost Index should be set up for purposes of trend analysis, rating, underwriting, and forecasting. This will eliminate some of the uncertainty from trend analysis, namely the unit cost or fee schedule change, portion of trend.
The Unit Cost Index is a vector of factors indicating the relationship of fee schedules in a given month to a base month. For example, assume Issuer ABC uses December 2011 as its base month. Furthermore, assume 50% of its providers increase fee schedules by an average of 5% on January 1, 2012 and the other 50% increases fee schedules by 4% on July 1, 2012. The values in the Unit Cost Index would be 1.000 in December 2011, (1 + 50% of 5% =) 1.025 x 1.000 = 1.025 from January 2012 through June 2012, and (1 + 50% of 4% =) 1.020 x 1.025 = 1.045 for July 2012 through December 2012. The Index should go back as far as the earliest month of incurred claims experience being used and continue through the furthest month in the future that might need a Unit Cost Index value to obtain a monthly incurred claim estimate. For example, in April 2013 the index may need to go through December 2014 if forecasts or other projections for 2014 will be made that would incorporate the Index.
Dividing the Gross Monthly Incurred Claims by the monthly Unit Cost Index values will neutralize the effects of fee schedule changes and allow the actuary to analyze trend changes other than fee schedule changes. Once this other trend has been analyzed, the actuary can then determine the Unit Cost Index affect by comparing the weighted average Unit Cost Index values for the most recent period to the weighted average Unit Cost Index for the earlier period.
For example, if the weighted average Unit Cost Index value for October 2012 through December 2012 is 1.045 and the weighted average Unit Cost Index value for October 2011 through December 2011 is 1.000 then the Unit Cost portion of the trend from October 2011 through December 2011 to the period October 2012 through December 2012 is 1.045 / 1.000 = 1.045, or 4.5%. The "other" trend would be increased by the 4.5% trend to obtain the total trend.
Health care costs vary by month of the year. October is usually one of the highest months while November is usually one of the lowest months. Incurred costs per day is usually higher from January through March than from June through August but that depends on deductible levels as the higher the deductible the lower costs are in the first few months of the year.
When analyzing trends the actuary must account for this seasonality difference if trend experience periods are anything other than the same period one year to the same period in another year. For example, if 3-month moving average trends are being reviewed for October 2012 through December 2012 over the period October 2011 through December 2011 then seasonality does not generally need to be applied. If the actuary is calculating the trend from the period January 2012 through March 2012 to the period April 2012 through June 2012 then the monthly incurred claims will probably need to be adjusted for seasonality as the actual "trend" between these two periods may be more of a seasonality difference than a trend, even to the extent that claims for April through June may be less than January through March.
For this reason and since seasonality may change from year to year due to benefit changes and leveraging of costs due to deductibles and limits, we recommend that trends be analyzed from the same period one year to the same period in other years. As such we will not attempt to describe seasonality in this narrative.
Days of the Week Impact on Monthly Claims
Incurred claims for most types of services are greater on Monday through Friday than on Saturdays and Sundays. This should be taken into account when analyzing trends. For example, June 2012 has five Fridays and Saturdays along with four of Sunday through Thursday. June 2013 has five Saturdays and Sundays. Since Sundays are substantially lower in cost than Fridays, June 2013 claims would actually decrease due to the Days effect. Specifically, if the actuary sees a 5% trend from June 2012 to June 2013 then the adjusted trend would be 1.050 x 1.030 = roughly 8% (assuming a 3% Days adjustment).
The Days adjustment has more of an impact on the shorter trend experience periods. If the trend periods are 12 months then the Days impact would be minimal. A more detailed description of the Days adjustment can be found in other articles on this site.
Other Adjustments to Monthly Incurred Claims
The Gross Monthly Incurred Claims may be adjusted for one-time events or causes that no longer exist. An example is an inordinate number of large claims. The adjustment should be made with care and generally only be made when it is known with reasonable certainty that the event will not recur or, in the case of large claims, if there is a stop loss limiting liability for the large claims.
Once the Gross Monthly Incurred Claims are adjusted by the factors above to obtain Net Monthly Incurred Claims then the claims are divided by the enrollment units to obtain claims per member cost unit. This will give a Per Member Per Month (PMPM) for each month in the experience period. Net claims should also be summed for the 3-month period ending each month of the experience period for which the data is available. For example, the December 2012 entry should be the sum of claims for October 2012 through December 2012. This should also be done for the Member Cost Units. The claims column should be divided by the member cost unit column to produce 3-month rolling PMPMs for each month. Each 3-month moving average should be divided by the 3-month PMPM from the year prior. For example, the November 2012 PMPM entry (based on September 2012, October 2012, and November 2012 data) should be divided by the November 2011 PMPM entry (September 2011, October 2011, and November 2011) to produce a trend factor for the month. This should be done for all months for which the 3-month PMPM is available. This should be done for the 1-month, 6-month, and 12-month method also.
Trend Analysis Results
The steps used to complete the trend analysis depends on the goals of the analysis. For example, the trend analysis can be used to determine reasons for variation from a financial forecast. In this case the categories used in the analyses calculations should be consistent with the categories used in the forecast. Historic trends in the models above, making sure to include the Unit Cost Index trend, are compared to trends used in the forecast to determine variances. Once variances are identified then further research can be made to determine reasons for the differences.
Trend analysis is also part of a claims reserve or IBNP model. The actuary must estimate trends for the most recent month or two or three as completion factors for these months may not be reliable. In this case the actuary would look at the different trend models (1, 3, 6, and 12-month) and analyze differences in results in the models. In some situations much can be learned from the fact that models vary. For example, 3-month trends being significantly higher in recent months than 12-month trends might mean there is a late surge in trends, although it also might mean a large claim is affecting the result or that recent months are over-reserved. After this comparison is completed the actuary then obtains the trend, adjusting for the Cost Unit Index, Days adjustment, and any other change that might affect the most recent months into which the prior period is to be trended.
The method for determining trends for forecasts is similar to that for the claims reserve except that the trend is projected forward to a point in time further in the future. This means the actuary must adjust for factors that may not be present now but will affect claims in the future, such as new mandated benefits or open enrollment of new groups and members. Due to the additional uncertainty here, the actuary may tend to be slightly more conservative in estimating the future trend.
Other models may be used, such as a regression model derived from prior months of data. Rather than promote one model over another, I will say that trend analysis is not a number-crunching exercise. The actuary must fully understand the mechanics of the model used and must also understand the strengths and weaknesses of that model.
The actuary must review and abide by the Actuarial Standards of Practice in analyzing trends and projecting these trends forward. Trend-setting is an important component of forecasting, rate setting, and financial stability. Failure to trend properly may cause issues for years to come.
An Excel model replicating the 12-month Moving Average Model described above is available from Adams Actuarial LLC. If you have any questions regarding this model please contact me at firstname.lastname@example.org.