John P. Cookson, F.S.A. Peter K. Reilly, F.S.A.
In our research report entitled Modeling and Forecasting Health Care
Consumption we explored the economic and other factors driving National Health Expenditures
(NHEs). As part of this research, we produced forecasts of future NHEs. Since publishing
our report, the Health Care Financing Administration (CMS) has released a major
revision of historical NHEs and two additional years (1992-1993) of information.
The revisions to the NHE series gives us an opportunity to update our original report.
This includes: an update of the research report using this additional information
to test how the original forecasts performed, another look at the question of how
the increasing market penetration of managed care has affected NHEs and, finally,
an extension of forecasts through 1996.
In November 1994 CMS released revisions to their prior estimates
of NHEs. These revisions were extensive and extended back into the 1970's. Also,
some modest changes were made as far back as the late 1960's. In this release, CMS
has included their first estimates of 1992 and 1993 NHEs. Our original research
report was based on data through 1991 which was released in late 1992. The changes
in the historical series are illustrated in Figure 1 showing the annual trend in
NHEs.
The major revisions in the series occurred in the years 1984, 1985
and 1991. The rate of increase in NHEs was revised significantly higher for 1984-1985,
and significantly lower for 1991. The revisions stemmed from definitional changes
in occupational coding (the SIC-80 codes) that resulted in the shifting of some
services between expenditure categories. In the past, CMS attempted to adjust for
the past deficiencies of these codes. However, with these revisions, CMS believes
the earlier estimates were replaced with more accurate measures. See Table I for
a summary of the revisions.
| Source of Revision |
Comments |
| 1987 changes in SIC-80 codes fully implemented |
Resulted in significant increases in professional services
expenditures, particularly in 1984 & 1985 |
| 1994 revision relied more on 1990 census data than
on IRS records |
|
| Changes in the treatment of donations and disproportionate
share payments to hospitals |
Reduced hospital expenditures by over $6 billion from
prior estimates |
| 1987 National Medical Expenditure Survey data fully
reflected |
Only partly reflected in 1992 release. Fully implemented
in current historical estimates. |
In addition to the significant increases in NHEs shown in the 1984-1985
time period, a major trend reduction was shown for the 1991 period. Where the 1991
over 1990 increase had previously been over 11%, the new estimates are now under
8.5%. In our previous report we indicated that the 1991 trend could be revised downward.
These historical changes are important for two reasons. First, the
revisions can change our estimates of the historical relationships of important
explanatory variables. Second, variables formerly not significant may now be significant
in the revised models. This could be due to both the historical changes and the
addition of two years of data.
In our May 1994 research report, we developed econometric models explaining
historical increases in NHEs and generating forecasts based on these observed relationships.
Among the variables tested and found to be statistically significant were real personal
income lagged for several years, consumer price increases (CPI), physician supply
and percent out-of-pocket health expenditures. Among those tested, but not included
(not significant) in our final models, were demographic changes in the population,
specialist composition in the physician supply and the market penetration of HMOs.
The following table compares the former NHE data and model forecasts
(1992-1993) with the actual results shown in the current NHE estimates.
| Year |
Original NHE (billions) (1992 estimates) |
Growth Rate |
RevisedNHE (billions) (1994 estimates) |
Growth Rate |
| 1990 |
$675.0 |
11.7% |
$696.6 |
11.7% |
| 1991 |
$751.8 |
11.4% |
$755.6 |
8.5% |
| 1992 |
$818.1 |
8.8%* |
$820.3 |
8.6% |
| 1993 |
$889.5 |
8.7%* |
$884.2 |
7.8% |
* Our model forecasts from the research report Modelling and Forecasting
Health Care Consumption, are based on CMS data through 1991.
Note that 1990 NHEs were revised to be $21.6 billion higher than previously
estimated. However, this large upward revision for 1990 does not fully carry through
to 1991, thus significantly reducing the trends for 1991. Our forecast of a 1992
growth rate of 8.8% was very close to the most recent estimate of 8.6%. Our forecasted
1993 growth rate of 8.7% is 0.9% higher than the current estimate of 7.8%. This
is well within the bounds of expected statistical fluctuations based on the historical
error terms from our model and could also be affected by the historical revisions
that would have produced different forecasts. The most recent 1992 and 1993 CMS
estimates are also potentially subject to significant revisions.
We have updated the forecast models, re-estimating the parameters
to reflect the historical changes in the NHE series and the two additional data
points (years). These changes in the series leave open the possibility that variables
previously tested and rejected may now prove to be significant.
The original models tested whether managed care has had an impact
on NHEs, by using Health Maintenance Organization (HMO) market penetration as a
proxy for the impact of managed care. This proxy may be less than perfect if the
effectiveness of managed care has also changed over time. In addition, the impact
of other managed care initiatives (e.g. Preferred Provider Organizations, Utilization
Review) is not directly measured by this variable. However, the impact of non-HMO
based managed care could be indirectly picked up if it roughly coincides with the
impact of the HMO market penetration.
HMOs have historically controlled costs in two ways, 1) through aggressive
provider contracting and 2) through financial incentives and other direct measures
to control the utilization and the setting for delivery of care. If aggressive provider
contracting merely shifts costs to other payers with total cost unchanged, this
would not have an impact on aggregate healthcare consumption. Furthermore, some
historical cost control methods, such as utilization review, have been found to
have mixed effectiveness. If these attempts at managing costs had affected NHEs,
it was not apparent in our earlier statistical models.
Recently, HMOs have become more aggressive in implementing methods
of managing care. Under one approach, they are attempting to change provider practice
patterns through applying guidelines or protocols. Such guidelines, for example,
prescribe the day by day treatment for each inpatient admission for uncomplicated
cases. These guidelines assume that proper provisions for post-hospital care have
been made to accommodate early discharge. The result is that facilities are utilized
more efficiently and care is delivered in the most cost effective manner. Milliman
and Robertson has been publishing its own Healthcare Management Guidelines since
1989.
It is also possible that managed care effectiveness increases as HMO
market share increases. This could result as both providers and consumers become
more accustomed to these techniques over time. Also the HMOs negotiating leverage
over providers may increase as market share increases. However, this could lead
to a scenario where HMOs are merely successful at cost shifting until their market
dominance allows them to force change on the healthcare delivery system. Some evidence
on the impact of high HMO market share has been published by James C. Robinson,
Ph.D. in a study entitled HMO Market Penetration and Hospital Cost Inflation. This
study showed that higher HMO penetration corresponds to lower increases in hospital
costs.
Due to low market share in the past, it may only be recently that
HMOs have had sufficient impact on NHEs to be statistically measurable. Growing
market penetration, and increasing effectiveness at changing provider practice patterns
may be increasing the HMO's overall impact. To test this hypothesis, the HMO penetration
variable was again included in our models, and produced a statistically significant
relationship.
The following table summarizes the model results from our May 1994
report (based on 1992 CMS data) and the new model results based on the new 1994
NHE series.
Comparison of Milliman & Robertson, Inc. Original and Revised
NHE Forecast Models
| Original Model* |
Revised Model |
|||
| Variable |
Coefficient |
T-Statistic |
Coefficient |
T-Statistic |
| Constant |
.033 |
2.75 |
NA |
NA |
| Autoregressive Order 1 Term |
.64 |
4.97 |
NA |
NA |
| CPI-W All Items |
.48 |
8.58 |
.28 |
5.7 |
| Out of pocket payment percentage |
-.19 |
-2.81 |
-.14 |
-14.4 |
| Physicians per capita |
2.33 |
13.08 |
3.08 |
16.5 |
| Real personal income - lagged one year |
NA |
NA |
.38 |
5.1 |
| Real personal income - lagged two years |
NA |
NA |
.15 |
2.0 |
| Real personal income - lagged three years |
.17 |
2.07 |
NA |
NA |
| Real personal income - lagged four years |
.39 |
5.30 |
.63 |
11.0 |
| Real personal income - lagged five years |
.33 |
4.40 |
NA |
NA |
| HMO Penetration |
NA |
NA |
-1.14 |
-7.8 |
* Original model refers to results published in the May 1994 research
report entitled Modeling and Forecasting Health Care Consumption.
| Model Statistics |
Old Model |
New Model |
| Number of Residuals |
30 |
30 |
| Degrees of Freedom |
22 |
23 |
| R2 |
.999921 |
.999942 |
| Sum of squares of residuals |
.0018796 |
.0013827 |
The significant factor to note in the new model is the statistical
significance of the HMO penetration variable. The coefficient implies that for every
1.0% increase in the HMO penetration variable NHE's declined by 1.14%. This is a
much larger impact than we would have anticipated. A few possible explanations exist
for this. First, the HMO penetration variable may be picking up the impact of non-HMO
managed care, such as PPOs and indemnity plans implementing utilization review (UR)
or second, it may be accounting for a sentinel effect on the rest of the market
from the HMO's actions. Furthermore, market penetration is likely to inadequately
describe the dynamics of how HMOs are affecting healthcare consumption over time.
For example, HMOs may have become significantly more effective recently. Finally,
to the extent that the HMO variable is collinear with any other variable in the
model, physician supply for example, then the coefficient estimates can not be interpreted
so directly.
With the reestimation of the model we have changed the lags used for
the personal income variable. In our original model we used three, four and five
year lags. The new formulation uses a one, two and four year lag. The addition of
the HMO penetration variable in conjunction with the revisions in the data could
account for this change. The various lags in personal income could account for how
the income relationships affect alternate sectors of the health care economy.
In addition, the re-estimated model now shows less sensitivity to
the CPI, but significantly greater sensitivity to real personal income and physician
supply. Physician supply is likely a proxy for technological changes in the healthcare
market place. Also, the Auto-Regressive (AR) order term, used to correct for serial
correlation in the residuals, was no longer statistically significant.
Once again we can use these model results to forecast NHEs, exploiting
the lagged relationship between economic activity and health care consumption.
This model forecasts National Health Expenditures for three years,
from 1994-1996. Since only the personal income variable has a leading indicator
relationship, a scenario must be specified for the other model variables for calendar
year 1994-1996. The following table summarizes the growth rates used.
| Year |
Trend in CPI-W |
Rate of Change in Physicians/Capita |
Change in Out-of-Pocket Payment Percent |
Growth in HMO Penetration |
Growth in Real Personal Income |
| 1993 |
2.8% |
2.0% |
-3.0% |
5.5% |
-.1% |
| 1994 |
3.0% |
2.0% |
-3.0% |
5.2% |
1.5% |
| 1995 |
3.5% |
2.0% |
-3.0% |
5.0% |
1.0% |
| 1996 |
4.0% |
2.0% |
-3.0% |
4.7% |
NA |
Our models forecast that NHE will grow by 26.8% over the three years
for an average annual rate of increase of 8.3%. NHE will grow from $884.2 billion
in 1993 to $1,121.6 billion in 1996. NHE as a percent of GDP was 13.94% in 1993.
We forecast that this will remain relatively level through 1995 and will rise to
14.35% in 1996. These estimates assume a constant rate of population growth. The
following table summarizes these results. Please note that these results are not
stated on a per capita basis.
| Year |
NHE (billions) |
Growth Rate |
NHE as a Percentage of GDP* |
| 1993 |
$884.2 |
7.8% |
13.94% |
| 1994 |
956.4 |
8.2 |
14.01 |
| 1995 |
1,024.8 |
7.2 |
14.02 |
| 1996 |
1,121.6 |
9.4 |
14.35 |
* Assumes real GDP growth of 4.5% in 1994, and 3.5% in 1995 and 1996
Figure 2 illustrates our forecast of the rate of increase in NHE (per
capita) along with the fitted values from the model.
We are expecting continued declines in the rate of increase in NHEs
during 1994 and 1995, but in 1996 higher inflation and the lagged impact of economic
growth will begin to move costs up.
Note that the forecasts of NHEs as a percent of GDP depend upon the
real GDP increase scenarios. Slower economic growth would result in a short term
increase in NHEs as a % of GDP, and faster economic growth would result in a short
term decrease. As economic growth affects future NHE growth the impact would reverse.
The historical revisions to the NHEs have significantly changed the
pattern of growth, in addition to the statistical models we have built to model
and forecast. The most significant change is that we have now found a statistically
significant impact of managed care on NHEs. Our model implies that as managed care
penetration increases, the rate of increase in NHEs slows.
The forecasts generated from our latest model show continued low growth
in NHEs in 1994 and 1995 with an increase in 1996. The acceleration in 1996 trends
is due to the lagged impact of the recovery from the 1991 recession, higher underlying
assumed inflation and slower HMO penetration growth. The net result is that NHEs
as a percent of GDP should remain stable in 1994 and 1995 and begin to rise again
in 1996.