Multivariate Analysis of Patterns of Informal and Formal Caregiving among Privately Insured and Non-Privately Insured Disabled Elders Living in the Community. B. Use of Medicare Financed Home Health Care Services


While we could not discern payment sources for each particular home care service for the two samples, we were able to identify the payment source for formal care as a whole, and thereby compare the distribution of payment sources for home care services. As a starting point, it is important to remember that virtually all individuals in the Insured Panel use formal services and that the majority (more than 70%) do not pay anything out-of-pocket for these services.

In the analyses that follow, we focus on the subset of individuals who use formal (paid) services. As shown in Figure 2, Medicare is not a particularly important payment source for the privately insured, in contrast to the non-privately insured, where roughly 30% of disabled elders use Medicare as a payment source for home care services.14 For every privately insured individual accessing Medicare, there are five non-privately insured, disabled elders doing the same.

  FIGURE 2: Payment Sources for Home Care Services by Insurance Status  
Bar Chart: Private Pay Only -- Privately Insured (83%), Non-Privately Insured (41%); Private Pay & Medicare -- Privately Insured (6%), Non-Privately Insured (14%); Medicare Only -- Non-Privately Insured (15%); Medicaid Only -- Non-Privately Insured (6%); Other -- Privately Insured (2%), Non-Privately Insured (7%); Unsure -- Privately Insured (9%), Non-Privately Insured (17%).
SOURCE: 1999 National Claimant Study (n=586) and 1994 NLTCS data (n=1357).

There may be a number of reasons why the privately insured access Medicare services less frequently than do the non-privately insured. First, because LTC insurance contracts include coordination of benefits clauses, the overlap between receipt of Medicare services and receipt of LTC insurance benefits ought to be minimal. Second, given that people are paying premiums for policies that cover home care, there may be less incentive and general awareness about the ability to access Medicare benefits for home health care services. Finally, the health and medical status of the privately insured may vary significantly from the non-privately insured, and there may be less need or demand for the skilled post-acute services for which Medicare has been a traditional payer. In order to test this last observation, we show the differences between insured and non-insured individual users and non-users of Medicare funded home care services.

  TABLE 2. The Characteristics Associated with Medicare Use Among Individuals who Use Formal Services by Insurance Status  
Characteristics Privately Insured Non-Privately Insured
  Do Not Use  
  Do Not Use  
Average Age 78 79 82* 83
Female 47%** 67% 71% 73%
ADL Limitations 4.6*** 3.7 3.6* 3.3
IADL Limitations 7.4 7.3 7.0 6.7
Perceived Health to be Poor 47%** 29% 48%*** 37%
Cognitive Impairment 50% 35% 34%*** 51%
Presence of Informal Caregiver   81% 78% 90% 90%
Living Alone 19% 29% 29% 35%
Income $28,000*** $41,000 $15,000 $16,000
Number of Medical Diagnoses .75** .57 1.2*** .9
ANOVA F-statistic significance level: * .10; ** .05; *** .10.

Table 2 shows that perceived health status, the number of ADL limitations, and the number of medical diagnoses are all related to use of Medicare home health, across both samples. Within the Insured Panel, having a lower income and being female are also characteristics that are associated with use of Medicare funded services. For the non-privately insured population, the cognitively impaired are less likely to access Medicare than are the cognitively intact. Because many of these characteristics may be related one to the other, they were entered into a logistic equation so that the independent effect of each could be adequately captured. In some cases, variables were transformed so as to aid in the interpretation of results. Table 3 summarizes results.

As shown, four variables are significantly related to the probability of using Medicare home health care services. These include having three or more ADL impairments, a self-health assessment of poor, having LTC insurance, and income. The odds ratio is presented in the third column of the table and is labeled Exp (B). The odds ratio can be used to interpret the magnitude of the effect of each variable on the probability of using Medicare home health. For example, the odds of using Medicare home health are 1.86 times greater for those with at least three ADL limitations. Also, individuals who assess their health to be poor are 1.78 times more likely to access Medicare services than are those who assess their health to be fair, good or excellent. As income increases, the probability of accessing Medicare declines.

TABLE 3. Logistic Regression Model for Use of Medicare Home Health Care
(total sample)
Variable   Coefficients     Standard  
  Exp (B)  
(0=below age 80)
-.1578 .262 .85
-.2107 .266 .81
ADL Impairments
(0=less than 3 impairments)  
.6196** .265 1.86
Self-assessed health
(0=excellent, good or fair)
.5775** .258 1.78
Cognitive Impairment
-.0177 .258 .98
LTC Insurance
-1.776*** .326 .17
Income -.0238*** .009 .98
Number of Diagnoses .2117 .138 1.24
Constant1 -.2823*** .604 ----
NOTE: The equation was tested with age and ADL status as continuous variables. The results for these and other variables did not differ but the model did not fit the data as well. Therefore, we have modeled both of these variables as dichotomous. Also note that an income and insurance interaction term was tested and found not to be significant.

Significant Level * p=.1; ** p=.05; *** p=.01; (n=418).
Nagelkerke R2=34.6%
  1. The constant is adjusted to reflect the underlying probability of Medicare use in the sample of service users -- 18.3%.

There is a negative relationship between Medicare use and insurance status. Individuals with LTC insurance are only .17 times as likely to access Medicare as are those without private insurance. Put another way, those without insurance are about six times as likely to access Medicare as are those with insurance, holding income and other variables constant.15

While the above equation clearly demonstrates that individuals with LTC insurance are less likely to access Medicare, selection effects may also be present. That is, it may be that there are unobserved differences among individuals who purchase LTC insurance that are related to the probability of using Medicare. If we don't control for selection effects, then we may incorrectly attribute lower Medicare utilization among the privately insured solely to the presence of insurance. In circumstances where the dependent variable is measured at the interval level, the Inverse Mills Ratio for insurance purchase would be entered into the equation predicting Medicare use among the insured group only. If the coefficient on the Inverse Mills Ratio was found to be significant, then this would indicate the presence of selection effects. However, the Inverse Mills Ratio can only be used in cases where the dependent variable is normally distributed. Because we are modeling a dichotomous dependent variable, i.e., whether or not an individual uses Medicare home health care, an alternative method for testing selection effects must be employed.

Identification of selection effects implies that we are trying to determine whether or not there is a "disinclination" among a privately insured group to use Medicare funded home health care services. To separate out this possible effect from the pure "insurance effect", we focus our analysis on the sub-set of individuals who do not have private insurance; that is, on disabled individuals drawn exclusively from the 1994 NLTCS dataset. For each person in this sample, we predict whether or not they "should have insurance". This is done by employing the logistic regression coefficients estimated for the insurance purchase decision developed previously (see Table 1). If the predicted probability for each individual in the 1994 NLTCS is greater than or equal to 50%, then the individual is placed into the "predicted insurance group." If the probability is less than 50%, they are placed into a "predicted non-insurance group." In essence, these two groups serve as proxies for those who do and do not have private insurance.

We then examine the rate of Medicare use between the two groups. If there is a statistically significant difference between the rate of use between the two groups, then we can be reasonably certain that there are, in fact, selection effects present. Put another way, we can estimate the magnitude of the "disinclination to use Medicare" among individuals who have characteristics associated with owning a private LTC insurance policy but do not, in fact, have one. This difference, which is based on our proxy groups, can be used to gauge the magnitude of any selection effects between the actual groups of privately and non-privately insured individuals.

Application of the logistic regression equation to the group of non-privately insured formal service users predicts that 11.0% "should have had" private LTC insurance. That is, given the characteristics of the 1994 NLTCS sample of service users, the logistic regression predicts that 11.0% of them would have private LTC insurance. The individuals predicted to have insurance are typically younger, have much higher incomes, are more educated, and are less likely to have a child living nearby or have an informal caregiver than are those predicted to not have insurance. Table 4 summarizes the rate of Medicare usage among the groups.

  TABLE 4. Predicted Rate of Medicare Home Health Use among Individuals Predicted to Have or Not Have Private LTC Insurance  
Medicare Home Health
Do Not Use
Medicare Home Health
Predicted to have Insurance 31.2%* 68.8%
Predicted not to have Insurance 20.7% 79.3%
X2 = 2.696; * significant at the .10 level. (n=1,155).

As shown, the difference in the rate of Medicare use between the two groups is statistically significant at the .10 level. Only 20.7% of individuals predicted to have private insurance use Medicare compared to 31.2% of those predicted not to have insurance. This statistically significant difference suggests the presence of selection effects. Put another way, individuals who have characteristics associated with LTC insurance ownership are also less inclined to use Medicare home health care. The rate of use among those predicted to have insurance is about one-third lower than among those predicted not to have insurance.

We have clearly established that those with private insurance access Medicare funded home care services much less frequently than do those without insurance. Some of this is due to selection effects and some due to the presence of insurance. To gauge the magnitude of the two effects, we first estimate an equation for predicting Medicare usage among the non-privately insured population. This equation is then used to predict such usage among the insured population. The difference between the actual rate of Medicare home health usage and the predicted rate of use provides a basis for evaluating the initial magnitude of the insurance effect. Table 5 shows the estimated equation for Medicare use among the non-privately insured population.

  TABLE 5. Logistic Regression Model for Use of Medicare Home Health Care Among Non-Privately Insured Disabled Elders  
Variable   Coefficients     Standard  
  Exp (B)  
Age -.0342** .015 .97
.1702 .251 1.19
ADL Impairments
(0=less than 3 impairments)  
.4673** .245 1.60
Self-assessed health
(0=excellent, good or fair)
.2258 .245 1.25
Cognitive Impairment
-.2740 .250 .760
(0=no Medicaid)
.6064* .332 1.83
(0=less than $25,000)
-.1405 .281 .88
Number of Diagnoses .2591** .123 1.30
Constant1 1.6783 1.26 --------
Significance Level * p=.10; ** p=.05; *** p=.01. (n=360).
Nagelkerke R2 = 10.8%.
  1. The constant is adjusted to reflect the underlying probability of Medicare use in the Insured Panel -- 31%.

Four variables are significantly related to the probability of using Medicare among the non-privately insured disabled population. These include age, having more than three ADL impairments and the presence of multiple medical diagnoses. Also, individuals who depend on Medicaid are more likely to access Medicare funded services than are non-Medicaid eligibles.

When we use the estimated coefficients from this equation to predict Medicare usage among the Insured Panel, we derive a predicted use rate of 11.1%. This means that, given the characteristics of the Insured Panel, one would expect that 11.1% would have been accessing Medicare funded home health care services. Instead, only 5.6% did so, thus indicating a usage rate of approximately half of what would be expected. One-third of this difference is due to selection effects (see Table 4) and two-thirds due to the presence of LTC insurance.

Further analysis of the 1994 NLTCS suggests that the average cost of a Medicare reimbursed home health visit for these disabled individuals was $64.67; the average number of visits that year was 51. Thus, annual Medicare home health expenditures were $3,298 per person. By applying this information to the privately insured population and inflating the dollars to 1999 levels, we find that for every 100 privately insured claimants, Medicare saves annual costs of up to $20,647 16. Again, two-thirds of these savings are due to the insurance effect and another third are due to the propensity of these individuals to avoid accessing Medicare.


The analysis presented here suggests that individuals with LTC insurance are less likely to access Medicare home health services than are those without the insurance. Moreover, given their demographic and health characteristics, the rate of use among the privately insured is up to one-half of what would be predicted, were these individuals not to have their insurance.

The models do not do a very good job of explaining the variation in the Medicare home health care use. While more than 20 different variables were tested, in no model did more than four variables turn out to be significantly related to Medicare use. This suggests that there is wide variation in the use of this benefit and that much remains to be learned about how and why individuals -- with and without insurance -- access it.17

View full report


"multanal.pdf" (pdf, 597.33Kb)

Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®