Nursing Home Ownership Trends and Their Impact on Quality of Care. METHODS

08/01/2009

Layout of the Master Database. The final database is designed primarily around the OSCAR observations, which originally has a single entry for each survey of each facility. In the final version, we expand the OSCAR such that every entity that has a substantial (at least 5 percent) stake of either the ownership or management (or both) has its own entry for each survey, giving a facility-survey-entity structure. Unlike in the original OMT tables, however, only the final owners are described here--in other words, the GP would not show up, but instead each partner would have its own entry. Each observation includes several of the most important OSCAR variables, which are shown in later tables, as well as the name and category information for the entities just described. Furthermore, we include the total percentage of the management and ownership that these entities control, and the number of levels deep that such control exists--from 1 to 7, as denoted in the Hierarchy description of the Data section. Finally, for each facility survey we generate variables explaining the average complexity level (number of layers deep), whether the management and ownership companies are controlled by the same people, and the type of real estate arrangement that exists.

File Preparation. OSCAR file preparation primarily consisted of abridging the dataset by limiting it to Texas facilities in operation from 2000 forward. Some additional variables were generated from existing data. These included occupancy levels, a count of the number of g-level deficiencies, a sum of the total number of facilities owned by a particular chain in any given year, and nurse staffing based on full-time equivalents (FTE) per resident. Finally, we manually cleaned the names of the chains, which often varied substantially among data sources and entry.

For the Facility Ownership (OMT Table 2), Facility Management (OMT Table 3) and Hierarchy (OMT Table 4) files, substantial data cleaning was necessary due to data entry errors. For example, many ownership incidences had durations of -1, 0 or 1 day, often indicating a placeholder entry while other changes were being made; all such entries had to be deleted and the corresponding true entries manually corrected. Duplicate entries were deleted, and missing entries were often generated based on a complex series of decision rules.

The remaining OMT files required little internal cleaning, although much preparation was still required to allow for the proper merging of these files to the OSCAR. Even though Medicare IDs were provided in the OMT Facility Provider Numbers file (Table 1), this field alone contributed to a mere 50 percent merge rate. A later file provided by the OMT office gave some additional help, but ultimately we had to attempt an unreliable text-field merge based on address information. We attempted four separate merges (in decreasing order of priority) based on combinations of facility name, address and zip code: all three, just name and address, just address and zip, and just name and zip. These processes yielded a 90 percent merge rate overall.

Dataset Construction. As mentioned earlier, OSCAR data were the core of the master data file. To the base OSCAR file, we added the edited versions of OMT Table 2 and Table 3, such that the top-level owners and managers of each facility were entered. By design, each of these top-level entities, which ranged in frequency from one to six, was marked by a separate entry. The merge was done through the facility ID information described in the above section. Overall, the merge had an approximately 90 percent success rate. The homes that did not merge correctly did not significantly differ from those that did and were discarded from further analysis.

Merging the Hierarchy file was a complex exercise, as its original form had no direct way of linking a top-level entity to its final controlling entities--each entry, as described above, only links two adjacent ownership levels (1 to 2, 4 to 5, etc.). As such, our first step was to iteratively expand each top-level entry from the previous step to include all of its subsequent owners; this was done through a loop command that merged based on entity ID. Ultimately, the process created one entry for each final branch of the ownership tree. Because our master database required only the top-level and final-level entities, all connecting steps were removed for simplicity. Finally, since many individuals owned pieces of multiple higher-level ownership entities, ownership stakes were summed for each top-level/final-level combination, while those with cumulative shares less than 5 percent were ultimately dropped. Obvious outliers and errors (those with stakes >125 percent) were tagged and disregarded from subsequent analyses. These steps were done separately, but identically, for both ownership and management.

The next merge involved adding the real estate information based on the OMT facility IDs. There was near complete merge success for these two datasets since both came from the OMT office. Nonetheless, the target file was longitudinal while the real estate file was cross-sectional, thus causing a loss of some information in the process. Therefore, each facility had a single real estate entity based on the most recent information.

The final two merges consisted of adding coding information from the Type Codes and Owner files. The former, merged by the codes themselves, helped to classify the entities according to their role in the organization, as explained above. The latter, merged through the entity IDs, gave provider names and basic information about all the top-level and final-level entities contained within. Both of these processes had 100 percent success due to the identical origins of the merging and target files.

A final variable was generated to express the relationship between the ownership and managing entities. Three classifications were possible: Self-Managed, meaning that no managing company was hired to run the facility; Separate Owner, implying the ownership and management entities were separate; or Same Owner, meaning the owner and manager could reasonably be thought of as the same or connected parties. This final designation was given if at least one entity had a 10 percent or greater stake in both aspects of the nursing home.

View full report

Preview
Download

"TXNHown.pdf" (pdf, 3.91Mb)

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®