Background

Milliman & Robertson, Inc. has provided actuarial consulting services for the health insurance industry for over 50 years. Our primary role has been advising traditional insurers, and later HMOs and providers on the various aspects of health insurance risks, costs, pricing, etc. Because of the short-term nature of most health insurance contracts, and the volatility of health insurance trends, we began some research in the mid-1980s to try to better predict short-term health insurance trends.

Our objective was not to collect insured trend data, because of various problems with accurate enrollment data, and the effects of benefit changes, underwriting and other contractual provisions over time. These factors can cause the trends experienced by different insurers to vary widely. Rather our objective was to try to identify the underlying average forces of health cost trends that affect everyone. Then we can try to understand the differences from carrier to carrier based upon their unique circumstances.

We have been tracking the underlying medical care cost trends for the non-Medicare population with a history extending back to 1974. We developed a model using various survey and government data with proxies for unmeasured components and weighted to reflect a typical Major Medical Plan.

Since the actual data sources reflect the total population and our focus is private insurance, we have backed out Medicare experience to the extent possible. It was not possible to remove Medicaid or uninsured data because of the lack of timely data sources. As such, the models are representative of non-Medicare population data including insured and uninsured and both direct and third party payments. We have the capability of separately reflecting the leveraging impact of standard deductibles and out-of-pocket limits and other benefit design provisions to estimate this impact on third-party payers and insurers. Our model is called the Health Cost Index ™ (HCI).

Since our model contains Medicaid and uninsured data, it probably understates, on average, insured employee/employer trends. In addition, since we exclude Medicare eligible we also exclude the drug costs for these retirees that would otherwise increase average measured trends on groups with retiree drug coverage. And, since we are measuring the full non-Medicare population, we do not reflect the impact of adverse selection, which nearly every insured group insurer suffers because of the multiple benefit options available. In fact, each of the separate benefit options could experience higher trends than the overall average. Each group and each carrier suffers its own unique level of adverse selection depending upon the rate of enrollment turnover and growth, competition, benefit differences, and underwriting effectiveness. In addition, provider contracting and management directly impacts each carrier’s trends, with those increasing their discounts at a faster rate, or more effectively improving their utilization controls experiencing more moderate trends. Our models are intended to measure the force of trend reflecting the average impact of increasing discounts and managed care effectiveness, but excluding the effects of adverse selection. The history of the HCI with no deductible leverage and our forecast through 2002 is shown in Chart 1. This shows a series of cyclical ups and downs, with trends much higher in the 1970s and early 1980s when underlying inflation was also much higher.

In addition to developing a concurrent model of underlying medical care trends, our objective was also to develop forecasting techniques using leading indicator or economic scenario relationships to project future trends. These models are constructed using software that detects trend variables and other interventions on an automatic basic. The final variables that comprise the statistical models of our medical trend history and resulting forecasts are: CPI (excluding energy), Real Personal Income (with multiple lags generally of 3 to 5 years), and a variable to reflect the relative increase of Medicare vs. Non-Medicare hospital inpatient per diems. These variables have consistently shown a strong relationship over the nearly 15 years we have been doing this modeling. These variables explain the annual growth in our medical care trend model, along with a constant 3.5% per year of excess growth (trend variable) from inception of our data. This is roughly consistent with the average excess Medical GDP vs. Total GDP growth over this period. However, in the early 1990s, our models began to overpredict future trends. After some time our software was able to identify a paradigm shift and to develop a reduction in the annual "excess growth" from about 3.5% down to about 1% beginning in the late 1980s. The other model variables remained relatively unchanged. We attributed this reduction to the "managed care effect" since it happened concurrently with the rapid "managed care" expansion at that time. This reduction seems to have remained in effect at least through the late 1990s.

Recently however, we have started to accumulate anecdotal evidence that the "managed care backlash" may be exerting significant upward pressure on health care trends. We have warned our subscribers of this possibility over the past two years. This effect may be greater on carriers and in geographic areas where the managed care controls have been more aggressively implemented over the past decade, and appears to be affecting both the pricing and utilization side. This change could represent a paradigm shift where past statistical and economic relationships are not good predictors of future trends. This is similar to the situation around 1990, when the "managed care effect" first dampened the level of excess medical care growth. However, we believe the changes are not really a paradigm shift but rather some temporary outliers or spike in trends that are likely to be of a short duration.

We do not expect that the full "managed care effect" that has held trends down for more than a decade will be eliminated, but we have, in fact, subjectively reversed about 25% of this effect in our current forecast models as a way to represent our short-term expectations of higher trends. Theoretically, however, if the entire impact of the reduced trends during the 1990s were to be unwound in a short period of time, actual trends could jump dramatically for a few years. We do not see this as a likely scenario. Rather, our most likely scenario is a step up in trends in 2000 and probably 2001, with a return to the 1990s reduced trend environment (perhaps slightly higher) by 2002 or 2003. Thus, we expect trends to return to about GDP plus 1% or so, but subject to the fluctuations and lags of our other model variables. We expect that the long-term effect of the backlash won’t be significant, mainly because our Hospital Efficiency Index™ still indicates significant inefficiencies in the hospital system. Although some hospitals are relatively efficient in managing care and costs, many others are quite inefficient. But overcoming these inefficiencies will require a more cooperative effort between payers and providers. Right now the economy is robust and individuals seem to be placing more emphasis on freedom of choice than cost, since they have been shielded from some of the recent cost increases because of the tight labor market. However, as the economy softens, the cost of health care will once again become a significant factor.

Despite our forecast increase, we still only see short-term future trends (HCI) of just over 8%. This is just about 1% to 1.5% above the levels experienced since early 1999. Trends experienced by most employers and carriers could approach 10% or more due to adverse selection, although some will clearly experience much higher levels due to their unique circumstances.

The components of trends since 1990 and forecasted to 2002 are shown in Chart 2. Inpatient trends have generally been the lowest over this time, followed by physician trends. Hospital outpatient trends were the highest during the early 1990s, until overtaken by prescription drug trends in late 1995. These do not reflect the leverage effect due to deductibles that can easily add ½% to 1% (at current benefit levels) to overall trends. The rapid growth of prescription drug trends has caused it to rise from less than 10% of total costs in the 1980s to nearly 20% today. Drug trends experienced an acceleration beginning in 1997 coincident with the initiation of direct-to-consumer advertising. Our models predict this excess trend wearing off eventually, and drug trends appear to have peaked near 20% in late 1999, possibly with some added impact from Y2K hoarding. Our models forecast a moderation of drug trends for the next several years, but they will still remain in the teens. Inpatient hospital trends have increased from negative to moderately positive levels and are expected to increase somewhat more, relatively speaking, due to the more "kindler and gentler" approach to managing care that some major insurers appear to be moving towards. Also, some provider reimbursement contracts appear to be renewing at higher rate increases than recent years. Outpatient trends may be affected short-term by the Medicare change from cost reimbursement to prospective payment, which may pinch hospital profit margins and encourage shifting of costs in the short run. Nevertheless, this may ultimately lead to better control for commercial insurers because of the standardization of the new Medicare payment benchmarks. Physician payment increases may benefit from consolidation and a stronger bargaining position. However, we do not forecast any big increase in the near future.

Insurance premium trends tend to lag underlying medical cost trends because the future costs must be estimated several months in advance, and then guaranteed for a year. Thus, if trends change, as they did between 1996 and 1999, insurers tend to lag with their rate increases. Although it is not a perfect measure, the Employment Cost Index (ECI) for Health Insurance Premiums, an unpublished series, is a fair representation. Chart 3 shows this in comparison to our HCI. The ECI reflects premiums paid by employers, and can tend to understate trends somewhat. Benefit reductions, charging a greater share of premiums to employees, and shifting enrollment to lower cost options will all tend to reduce employers’ premium outlays. The ECI trends also reflect swings in the competitiveness of the environment. As the environment becomes less competitive, insurance rates can move up at a faster pace than assumed trends and vice versa. This can be readily seen during the 1995-1997 period when the environment was very competitive, and premium increases in the ECI were minimal. The ECI tends to lag the HCI by about 18 months. A shift of the HCI by 18 months in Chart 4 shows a much closer correspondence of these two trends. Our model of the ECI over the next couple of years using the HCI forecast and some additional variables projects premium increases above the level of our forecasted HCI trends. However, the health of the economy and competition for workers will determine whether employers will be able to avoid some of these increases through benefit reductions or premium shifts to employees.

This lag in insurance premiums has traditionally had some negative repercussions in what the industry refers to as the underwriting cycle. Thus, as premiums lag increasing trends, insurers accumulate underwriting losses and vice versa. This phenomena is due primarily to the effect of three things – a) rate guarantees, b) the competitiveness of the market (and reluctance of carriers to be the first to raise rates), and c) the generally weak forecasting ability of most carriers. The latter two of these factors would have to change in order for the industry to smooth the underwriting cycle.

Although short-term medical trends have been our focus, long-term trends have become more important since the early 1990s when employers were required to establish Financial Accounting Standards Board (FASB) liabilities for their post-retirement and post-employment health care plans. In addition, longer-term medical care trends are important to Medicare in estimating the long-term balance of this program, and to hospitals and other interested parties having long-term interests or investments in the health care economy. Having some idea of the potential long-term trends permits the development of strategies related to various investment, planning and policy issues.

When FASB liabilities were first required to be evaluated (early 1990s) trends were at very high levels. It was clear that assuming such trends would continue (which had been true since at least the 1960s), implied that health care would consume or dominate all or nearly all the GDP. Although it is important to recognize current trends in the beginning estimate of these FASB liabilities, long-term year-to-year fluctuations are not as important. The key is trying to identify reasonable long-term average (ultimate) trends and a reasonable transition plan to reach them.

Many of these ultimate assumptions developed in the early 1990s estimated that health care trends would be the same as long-term economic growth on the assumption that health care GDP would stabilize at some ultimate percentage of overall GDP. The ultimate percentage of GDP would depend upon starting trends and how quickly ultimate trends are realized. Soon after these liabilities were first established, actual trends dropped dramatically (during the mid-1990s), and generated actuarial gains on many of these liabilities. However, with the advent of more recent higher trends, some of these assumptions have come under further scrutiny. A recent Medical Trustees Advisory Panel has looked at medical care GDP out to 2073. The Panel concluded that medical care GDP growth equal to overall GDP plus 1% was not unreasonable and would produce a medical GDP share of less than 30% at the end of the period. However, the GDP plus 2% assumption quickly spiraled results to unrealistic levels. Some additional growth would presumably allow for some continuing technology growth.

It should be pointed out that medical GDP grew at an average rate of GDP plus 3% to 4% from the 1960s through the 1980s. Growth during much of the 1990s was much closer to the GDP plus 1% level. These levels are both consistent with the excess trends identified in our HCI models. This is an area that needs further research.