NHE estimates for 1990 and later were increased significantly but
were lowered from 1989 and prior. This raised the rate of growth in the NHEs significantly
from 1985 on. A number of changes, both methodological and data source revisions,
were made to the database that could account for these changes. A discussion of
these revisions can be found in Levit, Lazenby et al.¹
The major changes that did affect the revisions of the level of NHEs
and the trends are summarized in the following table.
| Source of revision |
Source of revision |
| Prescription Drugs |
New methods and data sources that better capture expenditures
in retail outlets |
| Vision and other Durable Medical Equipment |
Definitions revised |
| Data from 1992 Census of Service Industries |
New data and historical revisions affected Physician,
dental, other professional, nursing home and home health expenditure estimates |
| Estimates of construction spending |
Census data were revised back to 1982 |
The following table compares the model forecasts with the actual results
shown in the current NHE estimates.
** These values represent our model forecasts found
in "National Health Expenditure Forecasts: 1994-1996".
Evaluating forecast accuracy is difficult when the historical figures
continue to be revised. In this case current CMS NHE estimates for 1990-1993 are
respectively 3.3%, 1.3%, 1.6% and .9% above CMS 1992 estimate. In this case comparing
forecasted rates of change (trends) is more appropriate but is still not completely
accurate, however, the impact of the data revisions is at least partially removed
from the comparison.
To date, our forecast results have been mixed, though there is relatively
few actual data points on which to evaluate our accuracy. Our forecasts of trends
based on the 1992 NHE estimates compare favorably to current CMS estimates while
the forecast of 1994 NHE trend appears to be overstated. However, the most recent
data points in the NHE series are usually subject to the most significant revisions
making any conclusive comparison difficult for some time.
In this report we have opted to replace the physician supply variable
with an excess growth deterministic trend variable for the following reasons:
In this report we have opted to replace the physician supply variable
with an excess growth deterministic trend variable for the following reasons:
In the follow up to our original research report "Modeling and
Forecasting National Health Expenditures" we made a significant revision to
our models by including HMO market penetration as a proxy for the impact of managed
care on NHE's. In doing so we recognized that this variable is an imperfect proxy,
especially if the effectiveness of managed care has also changed over time. In addition,
the impact of other managed care initiatives (e.g. PPO's, UR) is not directly measured
by this variable. We further pointed out that managed care penetration has been
increasing for years without any perceptible impact on NHE's, perhaps due to the
manner in which some HMO's were generating savings (i.e. through cost shifting and
risk selection). This created a problem in the model estimation because the proxy
used didn't reflect the recent change in the effectiveness of managed care in affecting
aggregate health care consumption.
In this report we have opted to replace the HMO penetration variable
with a deterministic trend variable reflecting the increased impact of managed care
on NHE's since 1993 for the following reasons:
We have also made changes in the lag structure for the CPI and Real
Personal Income (PI) variables. We have added a lag 1 for the CPI variable and eliminated
the one and two year lags for the PI, while adding a contemporaneous relationship
(lag 0). Finally, a dummy variable was added to account for what was identified
as a statistical outlier in the data for 1976.
The following table summarizes the model results from our prior report
and the new models based on the new NHE series.
There are a few important items to note about the revised models.
First, the CPI (both lags) now has a larger net impact in the new model, (.64 versus
.28). The personal income variable (all lags) has a smaller impact than before (.74
versus 1.16). There is a constant trend of 4.8% per year up till 1993 when it is
reduced by 2.4%.
We use this model to forecast National Health Expenditures for four
years, from 1995-1998. The Personal Income and the other variables have contemporaneous
relationships with NHE and values for future time periods must be provided in order
to produce forecasts. The following table summarizes the growth rates used.
Our models forecast that NHE will grow by 31.8% over the four years
for an annual rate of increase of 7.1%. NHE will grow from $949.4 billion in 1994
to $1,251.2 billion in 1998. NHE as a percent of GDP was 13.7% in 1994. We forecast
that this will rise to 14.3% by 1998. The following table summarizes these results.
Please note that these results are not stated on a per capita basis.
Figure 2 illustrates our forecast of the rate of increase in NHE (per
capita) along with the fitted values from the model.
We expect NHE trends to bottom out in 1995 and then increase through
1998. The increases can be attributed to the lagged impact of economic activity
and slightly higher inflation. We don't expect trends to return to historical levels
that are well in excess of economic growth unless the effect of our managed care
trend variable is only temporary. Among the factors that could cause trends to return
to significantly higher levels would be the reduction in managed care companies'
ability to control costs due to legislative changes resulting from the apparent
current backlash against managed care.
¹Levit,Lazenby et al. "National health Expenditures, 1994",
Health Care Financing Review/Spring 1996/Volume 17, Number 3.
Press here to download a Lotus 1-2-3 format spreadsheet containing the data used in these models