Data:
The study was conducted by merging survey and secondary data sources for the year 2017-2018. EHR implementation data was collected via mail surveys sent to Directors of Nursing (DON) in high Medicaid nursing homes in the US. To ensure a higher response rate, we followed a modified approach to Dillman’s Total Design Method: three rounds of surveys with post-card reminders and follow-up phone calls from November 2017 through March 2018 [32]. All mailings included a link to the online survey. An incentive payment of $25 was provided to survey respondents.
We constructed the analytical sample as follows: 85% or higher Medicaid census, and consistent with the prior approaches, we excluded nursing homes with 10% or higher private pay and/or 8% or higher Medicare (Mor et al. 2004). Our final sample size was 1050; we received 391 survey completes for a response rate of 37%. We believe that our survey response rate was analogous to studies of this kind.
The survey data was merged with the following secondary data sources: Brown University’s Long-Term Care Focus (LTCFocus), Area Health Resource File (AHRF), and the Centers for Medicare and Medicaid Services’ (CMS) Medicare cost reports. LTCFocus provided nursing home organizational, demographic, and market data. AHRF was utilized to obtain demographic and market data at the county level. Medicare cost reports provided data on nursing home financial performance. The primary and secondary data were merged using nursing home provider ID and year.
Variables:
Dependent variable:
Total (profit) margin is a measure of overall financial performance and is defined as net income divided by total revenue. Net income is calculated as the difference between total revenue and total expenses of the nursing home.
Independent variables:
Mail surveys on EHR implementation assessed five functionalities: 1. administrative 2. documentation 3. order entry 4. results viewing 5. clinical tools. Administrative functions included processes and reporting such as scheduling systems and clinical task assignments. Documentation included health information and data such as resident demographics and medical history. Order entry had order management information including medication order entry. Results viewing had data on routing, managing, and presenting test results to clinical personnel for review. Clinical tool had decision support system and telemonitoring/telehealth data. The items included in each functionality and their summary statistics are provided in Appendix A. Each item had four response options (0= not available, 1= paper only, 2= paper and electronic, 3= fully electronic). The composite EHR implementation score was the average of administrative (6 items), documentation (9 items), order entry (2 items), results viewing (4 items), and clinical tools (2 items) (Appendix A).
Control variables:
Control variables were identified based upon the factors that may affect the financial performance of a nursing home[1, 10] and include the following: organizational-level (size, occupancy rate, chain affiliation, ownership, payer mix, use of nurse practitioners/physician assistants, nurse staffing, acuity Index, and proportion of racial/ethnic minorities), and market/county-level factors (competition/Herfindahl- Hirschman Index (HHI), Medicare Advantage (MA) market penetration, per capita income, unemployment rate, location, poverty level, education level, and percent of population 65 and older).
In terms of organizational factors, size captured the total number of beds within the nursing home. Occupancy rate was the percentage of occupied nursing home beds. Chain affiliation reflected whether the nursing home was part of a chain (0= freestanding; 1= chain affiliated). Ownership identified whether a nursing home was for-profit (0=for-profit), not-for-profit (1= not-for-profit) or government- owned (3= government-run). Payer mix was the proportion of the residents covered by Medicaid, Medicare, or private pay. Whether nurse practitioners/physician assistants were present at the nursing facility was coded as Yes (=1) and No (=0). Nurse staffing reflects four measures: Registered Nurse (RN) staffing mix, RN hours per resident day (PRD), licensed practical nurse (LPN) hours PRD certified nursing assistant (CNA) hours PRD. The acuity index is an average measure of the resident’s level of care needed. This measure was based on the number of residents needing assistance with various activities including mobility and activities of daily living (ADL). Proportion of racial/ethnic minorities was the proportion of nursing home residents who were Black, Hispanic, and other race/ethnicity.
In terms of market/county-level factors, market competition is measured as the sum of the squared of the market shares (based on beds) for nursing homes in a county. HHI is a continuous variable that ranges from 0 to 1 with lower values associated with higher competition – an HHI score close to zero would represent perfect competition. MA market penetration was calculated as the proportion of all Medicare beneficiaries in the county who were enrolled in a MA plan. Per capita income is a measure of the average income of individuals in a county. Unemployment rate was the percentage of individuals in the county who were unemployed. The location variable was coded as urban and rural where the urban was the reference category. Poverty level is percentage of persons in the county in poverty. Education level is the percentage of persons in the county aged 25 or above with a high school diploma or more. Percentage of population age 65 and older is calculated at the county level with total county population as the denominator.
Data analysis:
To adjust for potential non-response bias, we included propensity score weights in the regression analysis [33]. The propensity score weights were calculated as the inverse of the propensity scores for nursing homes that participated in the survey. To estimate the propensity score, we used a logistic regression model where we regressed respondence status (respondent=1, non-respondent=0) on the following variables: size, ownership status, chain affiliation, payer mix, acuity index, occupancy rate, race/ethnicity, RN staffing mix, RN hours per resident day, LPN hours per resident day, CNA hours per resident day, Medicare Advantage market penetration, per capita income, poverty, unemployment, education, competition (HHI), location, and percent of individuals over 65. Then we calculated the inverse of the propensity score, the propensity score weight to include in the models.
Multivariable linear regression was used to model the relationship between EHR implementation and financial performance, adjusting for both organizational and market level variables. Two models were run: first, the association of total (average) EHR functionalities with total margin; and second, the association of individual EHR functionalities with total margin. Stata 16 was utilized for data management and analysis, and statistical tests were evaluated at the 0.05 level of significance. The study was approved by the Institutional Review Board of the University of Alabama at Birmingham (IRB-140828005) and informed consent was taken from all the survey participants.[34]