#### INJURIES IN LONG-TERM CARE

Worker injury rates in the long-term care (LTC) industry are well-known to exceed those of private industry generally and are the highest, even among the already high health care–related industry classes. Tracked by the US Department of Labor's Bureau of Labor Statistics, the recordable illness and injury rate per 100 full-time employees in 2009 was 3.6 in private industry, but in LTC was more than twice that at 8.4 per 100 full-time employees (Bureau of Labor Statistics; http://www.bls.gov/news.release/osh.t01.htm). This injury excess has been observable for decades and is indisputably driven by injuries related to the manual lifting and transfers of LTC residents.^{1},^{2}

In a 1997 report, the National Institute for Occupational Safety and Health (NIOSH), part of the Centers for Disease Control and Prevention, issued a critical review of the epidemiologic evidence for work-related musculoskeletal disorders (MSD) of neck, upper extremity, and low back and found that “The highest incidence rates of work-related injuries from over-exertion occur among workers in nursing and personal care facilities.”^{3}^{(Section 1, p 3)}

Performing a similar review of the evidence several years later, a report titled “Musculoskeletal Disorders and the Workplace: Low Back and Upper Extremities,” the Institute of Medicine found “a rich and consistent pattern of evidence that supports a relationship between the workplace and the occurrence of MSDs of the low back and upper extremities.”^{4}^{(p6)}

#### COSTS OF LTC INJURIES

The workers' compensation experience of the LTC industry is, in turn, negatively influenced by these high injury rates. Data supplied by the National Council on Compensation Insurance (NCCI), which collects data on insured facilities in 37 states, have reported a frequency of 3.62 workers' compensation claims per one million dollars of payroll in LTC as an industry class in 2005–2007. This compares to a rate of 1.11 per million dollars of payroll in all industries and 2.20 per million dollars payroll for police officers and firefighters.

The dollar loss per claim in the nursing home industry is less than that for some other industries such as mining or fire service. Nonetheless, the frequency of claims together with severity yields insurance premium costs for long-term care ($4.24 per $100 payroll) that approach those of other “high hazard” sectors named earlier. These costs then siphon dollars otherwise available for resident care, additional staffing, and benefits.

#### CAUSE OF LTC INJURIES

Caring for residents in long-term care facilities often involves manual lifting, transferring (eg, from bed to chair), and repositioning of residents in bed. One study estimated that during an average day shift of 8 hours, a nursing assistant performs more than 20 lifts or transfers of residents.^{5} Another study showed that the manual lifting and repositioning of patients and frail elderly residents frequently exceeds the physical lifting capacity of most caregivers.^{6} Factors that contribute to the hazard of lifting or transferring residents include the size and weight of the resident, the ability of the resident to bear some of their weight, and the resident's cognitive ability to cooperate with the caregiver.^{7}

#### REDUCTION IN LTC INJURIES

For over a decade, available evidence has demonstrated the efficacy of mechanical lift equipment in preventing health care worker MSDs. Learning from other industries that applied lessons from ergonomics, the science of fitting the job to the physical capabilities of the worker, health care too began applying risk reduction methods to address its skyrocketing worker injury rates from patient lifting. Examining force, repetition, and awkward postures—the primary risk factors for MSDs—in light of existing manual lifting standards made it clear long ago that a one-person and even a two-person lift of a frail patient or resident surpassed injury thresholds calculated by ergonomics engineers.^{8},^{9} Several intervention studies have since shown that using mechanical lifts to assist frail patients clearly decreased worker injury from MSDs and often lessened compensation and other costs to facilities implementing the lifts.^{10},^{11}

Given this injury experience in the LTC industry and having observed declines in injury rates when process innovations were introduced in other industries, NCCI researchers, in collaboration with academic partners, assessed the impact of safe lifting programs on workers' compensation costs in LTC facilities nationwide.

Although the implementation of safe lift programs often emphasizes the availability of mechanical lift equipment in LTC facilities, the mere presence of lifts is not enough to ensure that they are used and used correctly. Thus, the study also assessed programmatic and work practice elements, such as worker training and attitudes toward using the lift equipment, to assess the impact of a comprehensive safe lift program.

#### METHODS

A list of US Centers for Medicare and Medicaid (CMS)–certified LTC facilities was obtained (N∼7500), and NCCI matched the facilities on that list to a database containing claims information that yielded 656 facilities matched. Excluded from the match were most facilities that were members of chains, since claims experiences for individual facilities in a chain were not always stored separately, making linkages to a specific facility's safe lift program impossible. This match of less than 10% of the facilities listed by CMS was due in part because NCCI does not collect data in several large states. However, when we examined matched facilities to determine generalizability of the results, they were found to be reassuringly similar in average size (∼100 beds per facility) and in geographic location to the LTC facility distribution nationwide (Table 1).

The original intent of this study was to compare facilities with and without safe lift equipment, but the survey results indicated that by the end of the survey period, close to 95% of facilities had powered mechanical lifts. Therefore, the focus shifted from whether or not facilities had safe lift equipment to the implementation of a comprehensive safe lift program.

Accordingly, we chose to limit this analysis to facilities with safe lift programs in place for more than 3 years. We also included other aspects of the design and implementation of the safe lift program, such as worker training, as a determinant of injury reduction.^{12} To this end, directors of nursing (DONs) at LTC facilities were surveyed to determine the status of their safe lift programs. Names of administrators and DONs in those facilities were obtained from searches of state databases matching them to the facilities list obtained from CMS.

After the approval of institutional review board was obtained, a Web-based questionnaire was designed and tested, and nurse telephone interviewers were trained to conduct surveys for DONs who preferred a phone survey. A small number of DON respondents preferred to complete a mailed survey.

The DONs at 271 facilities completed the survey between November 5, 2007, and May 12, 2008. After data review, six surveys were excluded because of incomplete or inconsistent survey responses; thus, a total of 265 surveys were analyzed. The survey captured data over a 3-year time interval, which is of significance because while some variables were queried repeatedly (on an annual basis for the years 2005, 2006, and 2007), others were recorded only once (at the time the survey was conducted). Therefore, the closest match timewise to the survey observations were the responses given for the year 2007. The DONs were encouraged to consult facility records to provide data for the earlier time periods about which they were queried.

The DON survey information for each facility was then linked to NCCI data on facility injuries and workers' compensation costs. In this way, it was possible to test the association between workers' compensation claims and costs and safe lift programs.

##### Construction of Lifts per Resident Ratios

Observations available on an annual basis (for 2005, 2006, and 2007) from the DON survey included the number of residents of the long-term care facility and the number of powered mechanical lifts (full free standing, sit-stand, full overhead). From these observations, the ratio of total lifts to 100 residents can be constructed for each year. Also reported in the DON survey was an estimate of the number of residents who required lift assistance. From these data, ratios of lifts per resident requiring lift assistance could be determined. Overhead lifts were included in the calculations of these ratios, but will have very little impact on results because overhead lifts are not widely used by the surveyed facilities (in 2007 three facilities reported having four overhead lifts, in 2006 one facility had two overhead lifts, and in 2005 one facility had one overhead lift).

##### Other Surveyed Features of Safe Lift Programs

Other survey variables that can be used to characterize individual safe lift programs at the time of the survey reflect administrative and work practice elements of a program as well as attitudes of nursing leadership regarding the use of lift assist equipment. The survey also asked about outcome measures that could be impacted by the presence or absence of a safe lift program, such as staff retention and turnover. The elements queried about in the survey are displayed in Table 2.

##### Creating the Safe Lift Index

The use of an index is a common and typically necessary approach when working with a large number of highly correlated potential explanatory variables. Once the surveys were complete, a safe lift index was constructed for the purpose of aggregating answers from the survey questions into a single value. Factor analysis was used to select a limited set of variables derived from those displayed in Table 2 that best predicted safe lifting practices. These survey variables pertained to policies and procedures regarding powered mechanical lifts, training of certified nursing assistants in the use of powered mechanical lifts, preferences of the DONs for powered mechanical lift use, potential barriers to the use of powered mechanical lifts, and enforcement of the lift policies. Of 11 variables that go into the creation of the index, four reflect the institution's policies and procedures having to do with the safe lifts, three reflect the preferences of DONs, three reflect barriers related to physical facility impediments and resident attitudes, and one measures lift policy enforcement.

##### Methodology for the Safe Lift Index Determination

The goal in constructing a safe lift index is to reduce the number of variables required to measure the construct “safe lifting,” choosing only those that best predict it. Candidate predictor variables for “safe lifting” were derived from the literature, from focus groups conducted with DONs and other facility personnel, and from previous work.^{13} The complete method for building the safe lift index is also given in the study of Gucer et al^{14} but is included here as well. To select the best questions for inclusion in the safe lift index, we first inspected descriptive statistics and discarded those unsuitable (eg, with no variance or very few answered the question). Then we summed the rest and correlated each with the summary score, removing those that did not correlate well (not significant at the 0.05 level) with the summary score. Finally, we conducted a factor analysis. For the factor analysis, we transformed variables into z scores, because the questions did not all have the same response options and this procedure prevented variables with more response options (greater variance) from having disproportionate influence. The factor analysis helped us determine whether we could further reduce the number of variables and whether separate factors existed within the scale. The Kaiser-Meyer Olekin statistic (0.731) indicated that the data would support factor analysis. We chose principal components analysis and varimax rotation. Four factors having an eigenvalue (the variances extracted by the factors) of 1 or higher were retained. Together these factors explain 65% of the variance of the safe lift index. Eleven variables that correlated 0.5 or more with one resulting underlying factor and less than 0.4 on other factors were retained and combined to form the index.

On inspection of the factors, we decided to use the full 11-item scale rather than the separate factors. Factor one described policies and procedures. Factor two contained one question on policies and procedures and two questions on DON preferences for lift use by weight of resident. Factor three was consistent, but contained only three questions on DON perceptions of barriers to lift use. Factor four contained only one item: stringency of certified nursing assistant punishment. Because only one question was contained on factor four, and factors one and two overlapped conceptually, we opted to use the entire 11-item scale rather than analyze separately by factor.

Reliability was measured using Cronbach α, which was 0.749. Validity was assessed by correlating the safe lift index score with a variable measuring the DONs' assessment of the status of the safe lift program, with 1 equal to “no program” and 5 equal to a “fully operational” safe lift program in which all residents designated as needing a powered mechanical lift are lifted with one. The Spearman nonparametric correlation was 0.429, *P* = 0.000.

The safe lift index was then used to predict outcomes, and because it is a single variable, fewer degrees of freedom are lost, which is important when small samples are analyzed.

##### Data on Workers' Compensation Claim Experience at the Surveyed Facilities

Survey data for each facility was matched with NCCI policy and claims data. Therefore, the study is limited to facilities where NCCI could directly match the survey responses to the workers' compensation data. For the most part, this primarily limited the facilities to ones with a single location. However, some multilocation facilities were included where the data could be isolated for the specific facility that answered the survey. Available insurance policy data include the policy effective date, class code, exposure, and premium. Available claims data include accident date, class code, indemnity and medical loss dollars, report number, part of body, cause of injury, and nature of injury.

Two outcome variables were derived from NCCI data—one for claim frequency and one for total costs. In both cases, calendar/accident year data at an annual rate were constructed from the policy period data—this was to better match the timing of the observations with the calendar year survey data. The focus is on injuries due to lifting in nursing-related class codes, which limited the data to eight specific class codes related to nursing, convalescent or retirement homes, or hospitals and the cause of injury code for lifting. Variables for 2005, 2006, and 2007 were constructed using data at first report of injury to allow consistent data use for all 3 years covered by the study.

The *frequency variable* is defined as all claims due to lifting in nursing-related class codes (medical only and lost time) per full-time equivalent worker at an annual rate. The number of full-time equivalent workers of an LTC facility is the ratio of the payroll of the facility to the respective average annual wage of nursing care facilities in the state in which the facility operates. Average wages for nursing care facilities (NAICS Code 6231) are from the Quarterly Census of Employment and Wages produced by the Bureau of Labor Statistics. For each facility, data were used for the corresponding ownership (either private or government) and state. Because data from the Quarterly Census of Employment and Wages are available only quarterly, the wages were prorated to most closely match payroll data for each policy effective period.

As mentioned, calendar/accident year frequency at an annual rate was constructed from policy period data such that it was consistent with the calendar year survey data. To achieve this consistency, exposure was adjusted to a calendar-year basis by assuming the number of full-time equivalent workers is constant for the time period, and then claims due to lifting in nursing-related class codes were counted by accident date. In cases in which continuous policies were unavailable, the numerator and denominator still covered the same time period, although this time period was now shorter than 1 year. Ratios that are based on incomplete years may serve as an approximation for the annual values because workplace injuries related to lifting are not subject to seasonality.

The total cost variable was constructed in a similar fashion and is defined as total medical and indemnity paid losses from the claims due to lifting in nursing-related class codes divided by exposure. Again, exposure was adjusted to a calendar-year basis, and the numerator and denominator were consistent in terms of the months included. Paid loss data were used in this study because they are actual data, whereas paid plus case data would be impacted by differences in reserving practices.

Other variables used in the analysis included data on the ownership structure (for-profit, not-for-profit, government) for each facility; this information was available from the CMS. Most facilities that responded to the survey are for-profit (165), followed by not-for-profit (78) and government (22). Statewide frequency data from NCCI were used to control for differences in workers' compensation systems by state. Statewide frequency is for all lost-time injuries at all workplaces in the state and is stated as claims per 100,000 workers.

##### The Models: A Technical Discussion

Technically, the estimated models can be described as being repeated measurement multilevel Tobit models with random effects at the unit of measurement (the institution) and fixed effects on the level of the ownership type. A Tobit model is a type of censored regression model. This approach is appropriate when there is a clustering of observed values at an upper or lower boundary. In this case, there is a lower limit of zero, reflecting the fact that several of the LTC facilities had no claims in any given year.

As discussed earlier, the outcome variables are claims frequency (defined earlier as claims due to lifting in nursing-related class codes per full-time equivalent worker at an annual rate) and total costs (defined above as total paid losses from those claims divided by payroll at an annual rate) for 2005 at first report, 2006 at first report, and 2007 at first report. The standard set of covariates included:

* Statewide frequency (to control for differences in the workers' compensation systems across states)

* The safe lift index

* The number of lifts per 100 residents

* Ownership structure (not-for-profit, for-profit, and government)

In the models, we excluded facilities that did not pass a “time check” for the year in question. In other words, if the calculated outcome variable at an annual rate for a given year was based on data for fewer than 90 days, that facility was excluded for that year.

We also included the models only those facilities that answered in the survey that their safe lift program had been operational for more than 3 years. This was to address the fact that we are including 2005 and 2006 in the models, and the variables in the safe lift index are for the point in time of the survey (November 2007 through May 2008). The rationale is that if they have had a safe lift program for more than 3 years, their answers to those questions would be more likely to apply to 2005 and 2006 than for the facilities that have had a safe lift program for less than 1 year or between 1 and 3 years.

##### Statistical Model

The frequency and total cost measures, which serve as the dependent variables, are left-censored at zero. This is because even the safest facility cannot record values of less than zero for frequency and total costs. Put differently, even if one facility is safer than a facility that routinely has zero frequency and zero total costs, the safer facility cannot do any better. The left-censoring within a standard regression approach is addressed in a Tobit model.

The data set comprises 3 years of measurement, but the panel is unbalanced. This is because not all facilities enter the data set with three measurements: some facilities are measured twice, and others are measured only once. Repeated measurement calls for random effects at the level of the facility.

##### Interpretation of Regression Coefficients

The regression coefficients allow estimates to be derived for the magnitude of the effect on claims frequencies that the various explanatory variables contribute. We calculate the percentage response of the dependent variable in response to a 1% change in the covariate for lifts per 100 residents and state frequency, to a one-standard deviation change in the covariate for the safe lift index, and to a switch from 0 to 1 for the ownership indicator variables (this effect is relative to the reference group, which is the not-for-profits). All evaluations are done on the mean of the dependent variable for institutions for which frequency or total costs are not equal to zero. (When evaluating the effect of a change in lifts per 100 residents, only institutions with nonzero values for this variable are considered.)

* For lifts per 100 residents and state frequency, we calculate an elasticity by multiplying the regression coefficient by the mean of the covariate and dividing it by the mean of the dependent variable. For lifts per 100 residents, to convert the impact of a 1% change to the impact of the addition of 1 lift, we determined that the facilities in the study with workers' compensation claims and nonzero lifts reported an average of 6.3 lifts per 100 residents. An increase of 1 lift therefore would be a 16% increase. The 16 was then applied to the estimate from the statistical analysis that a 1% increase in lifts per 100 residents is associated with a 0.3% decrease in frequency and a 0.7% decrease in total costs.

* For the safe lift index, we multiply the regression coefficient by the standard deviation of the safe lift index, add this effect to the mean of the dependent variable, then divide by the mean of the dependent variable before subtracting 1, and finally multiplying by 100 to arrive at the percentage effect.

* For the ownership indicator variables, we calculate the percentage effect in response to a one-unit change by adding the regression coefficient to the mean of the dependent variable, dividing this amount by the mean of the dependent variable before subtracting 1 and finally multiplying by 100.

The interpretation of the regression coefficients of a Tobit model requires care. Any effect that comes out of a regression coefficient has to be downweighted according to the proportion of institutions with censored observations. For instance, if an institution already has a zero frequency, then adding safe lifts will not decrease this frequency any further. So, any effect has to be multiplied by the proportion of institutions that are not censored (ie, do not have a zero frequency).

Calculations of the effects of changes in the explanatory variables on total costs are similar to those described earlier for the frequency model. Nevertheless, in addition to accounting for censored observations, in evaluating the impact of the regression coefficients on the mean for this model, we also must account for the square root transformation of the dependent variable.

#### RESULTS

##### Surveyed Facilities

The data set described here is derived from surveyed facilities that reported having a safe lift program in place for the 3-year study period of interest (2005 through 2007); after the mentioned exclusions, this totaled 317 observations from 119 facilities. Although 128 facilities reported having a safe lift program for 3 years, two facilities were excluded because of missing questionnaire data used to compute the safe lift index and seven were excluded because of failing the time check that the outcome variables are based on data for more than 3 months. Table 1 displays the number of facilities included in the models by geographic region and ownership type.

##### Lift Inventory

The histograms in Figure 1 indicate that there was a modest increase in the relative number of mechanical resident lifts over this period, as the distribution of the lifts per 100 residents gradually increased from 2005 to 2007. For instance, the share of facilities with only a limited number or no mechanical lifts fell markedly from 2005 to 2007. In 2005, 26% of facilities had no more than two lifts per 100 residents (ie, a ratio between 0 and 0.02)—this share fell to 17% in 2006 and 10% in 2007. The median ratios increased from 3.8 lifts per 100 residents in 2005 (0.038) to 5.0 in 2006 (0.050) and 5.7 in 2007 (0.057).

##### Lifts Relative to Residents Requiring Assistance

The data in Figure 1 display lifts per 100 facility residents whether they need assistance or not over each of the three study-period years. While displaying lifts per 100 residents in Figure 1 serves to show the growth in purchasing of lift assist equipment over time, a more useful variable of interest is the ratio of lifts to residents who need lifts. The NIOSH estimates that the ideal lift ratio is approximately one full lift for every eight to ten non–weight-bearing residents, which is expressed as a decimal equivalent of 0.125 to 0.100. A similar recommendation of approximately one sit-stand lift for every eight to ten partially weight-bearing residents has also been described.^{7}

Figure 2 displays data for the share of facilities in each lift ratio category (no powered lifts, below NIOSH recommendation, meet NIOSH recommendation, and above NIOSH recommendation) based on the number of full lifts in 2007 relative to the number of residents needing full lifts at the time of the survey. Overhead lifts are included in the calculation for full lifts but will have very little impact because only three surveyed facilities reported having overhead lifts in 2007. The number of sit-stand lifts in 2007 relative to the number of residents needing sit-stand lifts at the time of the survey is also shown. The figure shows that most facilities in the survey had lift ratios that exceeded the NIOSH recommendations of 0.125 (about 1 lift per 8 or so residents needing lift assistance). This figure excludes facilities without ratio information and facilities with ratios of 100% or more. The latter were considered to be outliers because they had very few patients needing lifts.

Table 3 displays the safe lift index and a more complete lift inventory of all surveyed facilities and those included in the models. Lift inventories are shown both as ratios per 100 residents and then more specifically as a ratio of the lift type per number of residents requiring that type of lift. For facilities included in the models, a mean ratio of 0.363 for full lifts and 0.312 for sit-stand lifts is observed. This translates to approximately one lift to every three residents needing lifts. The lift inventory results show in a different format what was observable in Figure 1—that the lift inventory increased across the 3-year study period, even among the facilities that did not have a safe lift program in place for the preceding 3 years.

##### Safe Lift Index in Surveyed Facilities

The safe lift index values are displayed for all facilities surveyed as well as those meeting requirements to be included in the models. Note that the average safe lift index for the facilities used in the models, those reported to have a safe lift program in place for at least 3 years, had a much higher average safe lift index (

) than that of the facilities without a mature safe lift program (

).

##### Status of Lift Programs Across Facilities

As described earlier, the lift inventory did not appear highly variable between facilities, and therefore attention turned to characterizing the presence, content, and longevity of comprehensive safe lift programs as predictors of our outcomes of interest. The number of lifts per resident is included in the models but is not statistically significant when tested individually; however, it is significant and has the expected sign when tested jointly with the index measuring other aspects of the facility's safe lift program.

Table 4 describes the safe lift program status of the study facilities. This table shows that 209 DONs (or almost 79%) answered that powered lifts were routinely used in their facilities. Presuming that routine lift use was a proxy for facilities with a more mature program, these DONs were then asked the length of time that their lift program had been operational. The models considered the 128 facilities (or 48%) where the powered mechanical lift program has been operational for “more than 3 years.” After considering the previously mentioned 3-month time test and missing values, the models ultimately contained 119 facilities.

By limiting the study to only those facilities where safe lift programs have been in effect for more than 3 years, we are able to use lift program descriptors as predictors of injury outcomes over the 3-year study period, back to 2005. Thus, the models are not measuring the association between the presence or absence of lift equipment and workers' compensation injuries and costs. Instead, the models are measuring the impact of differences in the elements and implementation of safe lift programs and workers' compensation costs.

##### Model Results of Safe Lift Index Effect on Claim Frequency and Cost

##### Claim Frequency

Table 5 shows the results of the workers' compensation claim frequency model for several independent variables. The coefficients for both statewide frequency of injury and the safe lift index are highly statistically significant (*P* < 0.01) with signs as expected; that is, higher values of the safe lift index are associated with lower values for claim frequency, and facilities in states with higher statewide frequency tend to have higher frequency of injury. Although the number of safe lifts per 100 residents is not statistically significant, the sign of the coefficient is in the expected direction, and an analysis of variance on the influence of the safe lift program (ie, the joint influence of the safe lift index and number of lifts per residents) delivers statistically significant results. The ownership structure of the facility is not statistically significant, a finding that will be discussed later.

From the regression coefficients displayed in Table 5, we estimate the following:

* An increase of 1 lift per 100 residents is associated with a 5% decrease in claims frequency on average.

* A 1 standard deviation increase in the safe lift index is associated with a 49% reduction in claims frequency. If a variable is normally distributed, 70% of the variation around the mean is within one standard deviation of the mean.

* A 1% increase in statewide claims frequency is associated with a 1.9% increase in claims frequency.

* For-profit facilities have a 36% lower frequency than not-for-profits.

* Government facilities have an 11% lower claims frequency than not-for-profits.

##### Workers' Compensation Costs

Table 6 displays the findings for the total costs modeled as a function of the same independent variables. Because of a high degree of skewness in the nonzero observations of total costs, the square root is applied to this dependent variable. As with frequency, both statewide frequency and the safe lift index have the expected signs and are statistically significant. Here again, the number of safe lifts per 100 residents is not statistically significant, but an analysis of variance of the influence of the safe lift program shows that the safe lift index and the number of lifts per resident are statistically significant when tested jointly. Unlike in the injury frequency model, the ownership structure of the facility is statistically significant.

Derived from this model, changes in the explanatory variables have the following impact on total costs:

* An increase of 1 lift per 100 residents is associated with an 11% decrease in total costs on average.

* A 1 standard deviation increase in the safe lift index is associated with a 33% reduction in total costs. If a variable is normally distributed, 70% of the variation around the mean is within one standard deviation of the mean.

* A 1% increase in statewide frequency is associated with a 5.3% increase in total costs.

* For-profits have 51% lower total costs than not-for-profits.

* Government facilities have 51% lower total costs than not-for-profits.

The weak statistical evidence for the role of the ownership structure of the facilities for frequency and total costs may be related to the influence of the ownership structure being encompassed in the safe lift program. In this case then, although ownership structure may be a major determinant of the safe lift program (which manifests itself primarily in the safe lift index, as indicated by the regression results), there may be little additional effect of the ownership structure on frequency and total cost.

To investigate this hypothesis, the safe lift index is modeled as a function of the ownership structure. Because there is no time variation in the safe lift index (nor in any of the covariates explaining this dependent variable), the observations in this regression comprise only the year 2007. Table 7 shows the regression results for the safe lift index. As in Tables 5 and 6, Table 7 includes only variables indicating that a facility is for-profit or government operated; the remaining option, not-for-profit, is captured in the constant term.

Because of having only 1 year of data, the number of observations is low, which implies large standard errors for the regression coefficients. As is common in studies based on small samples, these large standard errors make it difficult to obtain statistically significant results. Fortunately, methods exist to evaluate covariates that are not statistically significant in small samples. One such approach uses the Cauchy distribution, thus exploiting its comparatively flat tails.^{15} This approach yields a 76% probability that the safe lift index in for-profit institutions is higher than the safe lift index in not-for-profits. Similarly, the probability that the safe lift index at government institutions is higher than that at not-for-profits is 69%. This translates into an odds ratio of about 3.2:1 for for-profit facilities and 2.2:1 for government facilities that the regression coefficient is greater than zero or, put differently, that the ownership variable is statistically relevant in the safe lift index model.^{16}

##### Implications of the Safe Lift Index

The statistical model indicates that the index capturing a range of characteristics of a safe lift program is clearly related to differences in workplace injuries as measured by frequency and total workers' compensation costs. Table 8 indicates the 11 variables that best predicted safe lifting practices. Variables are retained in the table if they correlate at about 0.5 or more with one resulting underlying factor and less than 0.4 on other factors. Together these factors explain 65% of the variation in the index, which is correlated with improvements in workers' compensation outcomes.

#### DISCUSSION

What do the constructs of the safe lift index and their effects on worker injury suggest is required to ensure workplace safety during resident lifting? Not surprisingly, one of the most critical components is that the institutions have a comprehensive set of policies and procedures regarding mechanical lift use. These include having procedures specifying that lifts should be used for residents not able to move around independently and specifying lift use in the residents' care plans. Training newly hired certified nursing assistants in the use of lifts and incorporating lift use in performance evaluations are other important factors.

This comprehensive approach is important because other studies have shown that just the mere presence of mechanical lifts in a facility does not assure their use^{5} and may explain the lack of worker injury rate declines observed even when equipment is available.^{17}

As the acceptance of lift assist equipment has advanced in the last decade, the fairly consistent ratio of lift assist equipment across facilities in our study may help explain why the lift ratio itself was not a major effect modifier of our outcomes of interest regarding worker injury rates and costs. Indeed, there is increasing awareness of the need for an integrated, comprehensive approach to safe lifting that would include many of the elements comprising our safe lifting index. Supporting the value of such a “multi-component patient handling” program, a recent literature review found “moderate evidence” for such an approach. This included equipment availability, worker training, and policies, which had a positive effect on musculoskeletal outcomes of workers.^{18}

Also important are the preferences of each DON. Items that correlate highly with this factor include whether two caregivers may lift a resident manually and if the DON prefers the use of powered mechanical lifts when moving residents from bed to chair and vice versa. A facility's safe lifting culture is clearly set by the preferences and beliefs of the DON. Barriers to lift use also emerge as an important factor, which explains some of the variation in lift use. These include physical barriers such as the difficulty of using lifts in the residents' bathroom. Other barriers identified, although not physical, include whether residents are concerned about falling during a lift and lift maintenance issues. Finally, enforcement of the lift policy is also a factor.

#### CONCLUSION

We have shown that an increased emphasis on safe lift programs at LTC facilities is associated with fewer workplace injuries and lower workers' compensation costs. More precisely, higher scores on the safe lift index, a measure of safe lift program elements and policies, are associated with both lower injury claims frequency and total costs.

A facility's commitment to effectively implement a safe lift program appears to be largely dependent on policy preferences and lift use perceptions of the DON. Indeed, 8 of the 11 variables making up the safe lift index reflect the rigor of policies, the role of training, and the stringency of the facility's enforcement of its policies—all subject to the DON's influence.

From our survey, we also see that merely having powered lift equipment, while necessary for a comprehensive safe lift program, is not sufficient to achieve the beneficial influence on worker injury we observed here.