Analyses were done for several different types of measures. First, data from HCFA(now known as CMS) were used to describe counties with regard to changes in Medicare+Choice offerings. Next, plan-level analysis was done based on InterStudy data, with HCFA(now known as CMS) matches taken as Medicare+Choice plans, to investigate plan attributes associated with 1998 Medicare+Choice participation and withdrawals from the program in 1999. County-level analysis was then done to identify county-level attributes associated with these two outcomes. This analysis is designed to match the InterStudy plan analysis, so differs from the initial, HCFA(now known as CMS)-based description of program changes. The plan and county information were then combined to disentangle the effects of plan and county characteristics on the county-level plan decision to participate in 1998 and to withdraw in 1999. The final results describe metropolitan statistical areas (MSAs) as Medicare+Choice areas. As mentioned before, these results are descriptive in nature, so provide the groundwork for more sophisticated modeling of plan behavior and the role of policy factors in participation and withdrawal decisions.
Service Area Changes, 1998-1999. This first analysis of changes in plan Medicare+Choice service areas, withdrawals, and new offerings is based on categorizing counties with regard to changes in Medicare+Choice plan offerings. Seven categories of counties are considered, based on the availability of plans in September of 1998 and how that changed (or not) by April 1999:
|Definition of County Type for Analysis of Medicare+Choice Plan Offerings|
|County type||Number of Plans 9/98||Relationship between number plans 4/99 and 9/98||Change in plans?|
|No plans, no change||0||Equal||No|
|Gained all plans||0||More||Yes|
|Lost all plans||>0||Less||Yes|
So, for example, the Balanced Change group differs from the No Change by having had a change in the specific plans offered, although both had the same number of plans available in each time period.
This ignores any sub-county changes - if a plan's service area in 1998 included only a small portion of a county and this was withdrawn in 1999 then the entire county is counted as having lost a plan. As a result, the numbers of beneficiaries affected by changes in plan offerings are likely overstated. This analysis is based solely on information from HCFA(now known as CMS).
According to this categorization, two-thirds of counties had no plans and no change (Table 2). These counties contained about 25 percent of the nation's Medicare beneficiaries as of December 1998. An equal share of beneficiaries live in the 559 counties that had no change in plans offered between the two time periods, so overall about half of beneficiaries were unaffected by changes in plan participation decisions and service area changes. Conversely, nearly 40 percent of beneficiaries live in counties that had a net loss in plan offerings. These 271 counties account for nearly two-thirds of risk plan enrollees.
Nearly 2 percent of beneficiaries live in counties that lost all plans between September 1998 and April 1999. These 84 counties included 1,687 plan enrollees.
Those counties that lost all plans typically went from one to zero plans, while those that gained all plans went from zero to one plan (Table 3). Other groups of counties that lost plans - those with a net loss in plan offerings and those with balanced change - still have more plans on average than all other types of counties. This could be evidence that some markets have become (over)saturated with plans while others still have unmet demand for Medicare managed care and so will accommodate new entrants.
At this simple level of analysis, changes in plan offerings at the county level do not obviously correlate with payment rates (Table 3). Although those counties that lost all plans have relatively low payment rates, those counties that went from none to some plans have even lower rates on average. Similarly, those with a net loss in plans have higher rates than those with net gains.
Finally, plan changes at the county level are not obviously associated with general local market economic measures such as the unemployment rate or local wages (Table 4). Areas with net loss of plans had unemployment rates similar to those that had net gains. The wage index was highest in counties that had a net loss of plans and relatively low in counties that gained all plans or that lost all plans. A county-level demographic index, based on the factors historically used by HCFA(now known as CMS) to risk-adjust plan payments, is lowest in the counties that had no plans in both periods and highest in those with net gain in plans or with balanced change. This last finding may reflect how plans consider not only payment rates but also (perceived or real) risk status of local populations in their decisions. It may also relate to inadequacies in the risk adjustment process that will be phased out in 2000 which may be helpful to understand, especially if it would help anticipate plans' likely response to introduction of the new risk adjustment process next year.
To consider these different possibilities, plan-level, county-level, and plan-county analyses were conducted, based on service areas and other plan characteristics as described by InterStudy.
Plan Participation in 1998. Simple descriptive statistics suggest that, compared to all commercial plans, those participating in Medicare+Choice in 1998 were more likely to be federally qualified, have FEHBP enrollees, be part of a national chain, larger, be older, be less physician-intensive, be less hospital-intensive, and have expanded service areas in the past several years (Table 5).
These simple means are difficult to interpret, however, because some of these factors - such as age and size - are likely to be associated with one another. Simple ordinary least-squares regressions were run to analyze the relationships among these factors (Table 6). These results show that, at significance levels of 5 percent or better, there are some important differences between those plans that participated and those that did not. In particular, those that participated in Medicare+Choice in 1998 are:
- more likely to be federally qualified,
- less likely to have Medicaid patients,
- large, and
- not physician-intensive.
The estimates from this linear-probability model are reliable with regard to sign and significance, but their magnitude should be interpreted with caution. It explains nearly one-third of the variation in plan Medicare+Choice participation.
Plan Withdrawal in 1999. Among the plans that participated in 1998, about 35 withdrew from the program in 1999 (Table 5). Based on mean characteristics, these plans appear to be more likely to be IPA model plans, to be for-profit firms, to have Medicaid enrollees, to not have FEHBP enrollees, to be in a national chain, to be younger, to be physician and hospital intensive, to be less likely to have been involved in a corporate split in the past few years, to have had service area changes in the past few years, and to have non-typical enrollment growth (up or down). Again, however, the relationships between these many characteristics make it hard to interpret these simple means. Based on a linear-probability model, only affiliation with a national chain is associated with a significant difference in dropout probability - those in chains were more likely to withdraw from Medicare+Choice in 1999 (Table 6). Chain affiliation is not associated with a significant difference in joining Medicare+Choice in the first place but, once they've joined, these plans appear somewhat more likely to have dropped out in 1999. Once all of the factors are accounted for, other differences between plans by dropout status are insignificant.
Again, the use of a linear probability model results in estimates that are reliable with regard to sign and statistical significance but not magnitude. This model accounted for only 2 percent of the variance around the dropout decision.
Counties with Medicare+Choice Plans, 1998. Based on a simple comparison of means, counties with Medicare+Choice plans in 1998 appear to be fairly different from all counties in many ways (Table7). Counties with Medicare+Choice plans have different health care markets, with more commercial plans, higher admission rates, fewer hospital beds, and more physicians per capita. They have different populations, with relatively fewer people over 65, although these tend to be in higher demographic risk groups. They also have different local economic characteristics, with higher adjusted incomes and lower unemployment rates.
Most of these differences are statistically significant even when other differences are controlled for (Table 8). In particular, the chance that a county has a Medicare+Choice plan increases if:
- there are more commercial HMOs;
- the Medicare demographic risk index is higher;
- it is in an urban area;
- it has higher price-adjusted Medicare+Choice payment rate;
- the population share over 65 is high;
- hospital admission rates are low;
- there are more MDs per capita; and
- the unemployment rate is relatively high.
These factors explain nearly half of the variation in the probability that a county had a Medicare+Choice plan in 1998.
Counties Losing Plans, 1999. Simple means suggest that counties that lost plans in 1999 are fairly similar to those with plans in 1998, with a few exceptions (Table 7). For example, counties losing plans had relatively more commercial plans, were more likely to be in urban areas, had more Medicare+Choice plans, and higher Medicare+Choice enrollment rates than those counties that did not lose a plan.
A simple regression suggests that only two of these factors - number of commercial plans and number of Medicare+Choice plans - are significantly significant when other factors are held constant (Table 8). The fact that the number of commercial plans increases both the chance that a county will have Medicare+Choice plans and that it will lose plans suggests that there is some level of plan churning that may be likely to occur once there are a lot of plans in a market. Structural or behavioral models are necessary to test this hypothesis. The simple linear probability model accounts for over one-quarter of the variance around the county dropout probability.
Plan-County Participation, 1998. The inclusion of a particular county in a particular plan's Medicare+Choice service area is a function of both plan and county characteristics. To understand how the two relate, Medicare+Choice participation can be explored at the plan-county level, where plan-county pairs are determined by plan's commercial service areas. As a result, plans with a large number of counties will contribute more cases than other plans, so that means of plan-level characteristics will be weighted toward those plans with many counties in their service areas. So, for example, at the plan-county level, 52 percent were IPA model plans (Table 9), compared with 49 percent of plans at the plan level (Table 5). This implies that IPA plans have more counties in their commercial service areas than other plan types.
At the plan-county level, those in Medicare+Choice in 1998 differed from the universe in most characteristics studied (Table 9). In the regression, the county-level factors had the exact same pattern of sign and significance as they had in the county analysis above (Table 10). There were some differences, however, in the plan-level factors. Several factors that had been insignificant at the plan level are significant at the plan-county level: IPA model, presence of FEHBP enrollees, enrollees per hospital, reduction in plan service area since 1995, and dramatic change in enrollment over two years (up or down). Similarly, the presence of Medicaid patients had a significant negative effect on the probability of participation at the plan level but is insignificant in the plan-county analysis. The change in importance of whether or not a plan serves Medicaid patients in the different models suggests that this variable may have been capturing market characteristics in the plan-level models that are better controlled for in the plan-county models.
Plan-County Dropouts, 1999. At the plan-county level, dropouts look like 1998 Medicare+Choice participants, with many exceptions (Table 9). They are more similar to non-participants with regard, for example, to presence of FEHBP enrollees, enrollees per MD, and enrollees per hospital. Controlling for all other differences, factors associated with dropping out include:
- presence of Medicaid patients in plan;
- no presence of FEHBP patients in plan;
- affiliation with national chain;
- short time in operation;
- few enrollees per MD;
- having reduced service area since 1995;
- dramatic enrollment changes in past two years;
- low share of enrollees in traditional HMO product;
- non-urban location;
- low price-adjusted payment rate;
- low population share over 65; and
- high unemployment rate.
In the plan-level estimates above, only national chain affiliation had a statistically significant association with dropping out. In the county-level estimates, numbers of commercial and Medicare plans had been significant, but they are not at the plan-county level.
MSAs as Medicare+Choice Area. In addition to understanding plan participation and dropout decisions, this analysis also aims to investigate the concept of predetermined Medicare+Choice areas. Analysis of plan response to such a policy change depends on a better understanding of the determinants of plan behavior than is currently available. It is possible, however, to describe candidate areas with regard to Medicare+Choice characteristics and commercial markets. Only metropolitan statistical areas (MSAs) are considered below, although other reasonable candidate area definitions exist, such as Medicare Fee Schedule areas, modal commercial areas, and Bureau of Economic Analysis Economic Areas.
There would be 365 MSA-based areas, 240 of which had at least one Medicare+Choice plan in fall of 1998 (Table 11). According to simple means, these 240 differ from all MSAs in many ways:
- over 50 percent more Medicare beneficiaries;
- higher nominal Medicare+Choice payment rates (both at county and beneficiary level); and
- more commercial plans (11.80 v. 9.92).
Within MSAs with at least one Medicare+Choice plan, there were an average of 4.1 plans. Plans served an average of 3.2 MSAs. Among plans serving more than one MSA, enrollment was fairly evenly spread, with an average of 35 percent of enrollees in each served MSA.