| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||
Correspondence: Address correspondence to Norman V. Carroll, PhD, School of Pharmacy, Virginia Commonwealth University, 410 North 12th Street, Box 980533, Richmond, VA 23298-0533. E-mail: nvcarroll{at}vcu.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Key Words: Costs • Long-term care • Falls • Economics
Because they result in substantial morbidity for nursing home residents, it is likely that falls result in significant economic costs for LTCFs. A number of studies have estimated the costs of falls in various settings and for various outcomes. Carroll, Slattum, and Cox (2005) used data from the Medical Expenditure Panel Survey to estimate a mean cost per faller of $2,591 (2002 dollars). This estimate was based on a community-dwelling sample of individuals and included all types of fall injuries—from minor lacerations to hip fractures—treated across a range of health care settings (e.g., physicians' offices, emergency rooms [ERs], inpatient hospitals, outpatient clinics). Englander, Hodson, and Terregrossa (1996) estimated direct medical costs of fall-related injury based on population estimates and average charges for health care services. Their estimate was $6,215 per injured faller in 1994 dollars. As in Carroll and colleagues' study, this estimate incorporated care provided for all types of fall-related injury and in all types of settings.
Several studies have developed cost estimates based on the severity of the fall-related injury. A 1998 study estimated the cost of falls for community-dwelling elders in Connecticut (Rizzo et al., 1998). The estimates of fall-related costs were $2,500 (1996 dollars) for elders experiencing one noninjurious fall, $11,900 for those experiencing two or more noninjurious falls, and $19,440 for those experiencing one or more injurious falls. Brainsky and colleagues (1997) estimated that community-dwelling elders experienced direct medical and nonmedical costs of between $16,322 and $18,727 (1993 dollars) in the year following a hip fracture. Baron, Barrett, and Berger (1996) estimated the direct medical costs to Medicare for 10 common types of fractures using Medicare claims data. They estimated costs per person ranging from $2,564 for a wrist fracture to $15,294 for a hip fracture for the year following the fracture.
Two studies estimated fall-related costs by the setting in which fall-related injuries were treated. Using a case-control design, Finkelstein, Chen, Miller, Corso, and Stevens (2005) estimated fall-related costs of $22,260 (2000 dollars) for fallers who had been hospitalized, $3,890 for fallers who had visited the ER but not been hospitalized, and $5,040 for fallers who had received office-based or outpatient care but had been neither hospitalized nor visited the ER. They estimated slightly lower costs—$20,920, $3,230, and $4,200, respectively—using a case-crossover design. Roudsari, Ebel, Corso, Molinari, and Koepsell (2005) used data for a sample of Medicare patients from the MarketScan database to estimate mean costs of $17,483 (2004 dollars) for a fall-related hospitalization, $236 for an ER visit, and $412 for an outpatient visit.
In the only LTCF-specific cost study that we were able to locate, Sorenson and colleagues (2006) estimated average fall-related costs of $1,892 for the typical-case fall and a range of $700 for the best-case scenario to $12,817 for the worst-case scenario. The authors based these estimates on the judgment of an expert panel. We were unable to find any well-designed, empirical studies of the costs of falls for residents of LTCFs.
A number of measures have been proposed, and many implemented, to decrease the incidence of falls. Most of these require significant expenditures. Without good estimates of the costs of falls, it is not possible to determine the cost-effectiveness of such measures. Furthermore, government, managed care, and long-term-care decision makers need information on the relative costs of various diseases and conditions to guide their resource allocation decisions.
The purpose of this study was to estimate hospital and long-term-care costs resulting from falls in LTCFs. We did this from the perspective of a third-party payer. We estimated two types of costs: (a) the direct costs of hospitalization for fall-related injuries and (b) the additional direct costs that LTCFs incur as a result of fall-related loss of resident functioning. Costs associated with loss of function include physical and rehabilitation therapy and increased nursing and personal care. A resident who suffers a hip fracture as a result of a fall, for example, becomes significantly less mobile. This requires the facility to provide more nursing and personal care to help the resident get to the toilet, the dining room, and other places the resident needs to go in the course of a normal day. It is also likely that the facility will need to provide rehabilitation through physical or occupational therapy. The additional nursing, personal care, and rehabilitation costs may impose significant economic burdens on LTCFs.
|
Methods |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Research Design
The research design used in the study reduced a number of different types of biases. The design used a two-stage process to estimate fall-related costs. In the first stage, we calculated changes in costs from the pre- to post-period for both the falls group and the non-falls group. In the second stage, we estimated the cost of falls as the pre- to post-period change in costs for fallers minus the pre- to post-period change in costs for non-fallers. (This is referred to as the difference in changes from the pre- to post-period.)
The first stage of the estimation process reduced between-group biases by having each group serve as its own control. That is, the same participants were in the same settings in both the pre- and post-periods. This reduced biases resulting from differences between the groups—such as differences in physical environment, quality of nursing care management, facility staffing, turnover rates among nurses and nursing assistants, and differences in residents' health and functioning. As an illustration, assume that fallers were more likely to reside in facilities that provided more expensive care. A simple comparison of costs between groups in the post-period (or the pre-period) would yield falls-related cost estimates that were upwardly biased. However, a pre/post analysis would reduce these differences because fallers would be in more expensive facilities in both the pre- and post-periods and non-fallers would be in less expensive facilities in both the pre- and post-periods. Because the analysis calculates changes across periods, the higher facility expense for fallers would cancel itself out.
The study also employed matching based on propensity scores to control for baseline differences between fallers and non-fallers. The goal of propensity scoring is to balance the baseline characteristics of the different groups in an observational study. Thus, propensity scoring is another way of reducing biases that result from differences between groups.
The propensity score for each resident was the estimated probability that he or she was a faller based on his or her background characteristics. We used logistic regression to calculate propensity scores. The dependent variable in the regression was fall status. D'Agostino and D'Agostino (2007, p. 315) stated that "the focus should be on including variables in propensity score models that are unbalanced between the treated and control groups, and not necessarily be concerned specifically with whether they are related to the outcomes of interest." (In our study, the falls and non-falls groups would be analogous to the treated and control groups.) Thus, the predictors used in the logistic regression included all variables that were available from the data set and that differed between the groups. These variables included diseases measured by the MDS (see Table 1), time in the post-period, age, gender, race, pre-period hospital and LTCF reimbursement, and the type of site from which the patient was admitted to the LTCF (hospital, another LTCF, home). We matched fallers with non-fallers based on their propensity scores using a SAS greedy matching technique (Parsons, 2001). We excluded from the analysis residents without a match. This process produced a sample of fallers and a sample of non-fallers that were balanced on the measured baseline variables.
|
Assume, for example, that fall-related costs for a sample of patients amounted to $6,000 and costs related to aging over the time of the study amounted to $3,000. The pre- to post-period change in costs for fallers would be $9,000, which overstates fall-related costs by $3,000. The change in costs for non-fallers would be $3,000; the difference in the changes in costs between groups would be $6,000, which accurately estimates fall-related costs.
Assignment of Index Dates
It was necessary to assign index dates to each resident to define the pre- and post-periods. For fallers, the index date was defined as the date of the first fall recorded in the MDS after the resident's admission to the LTCF. Thus, the index date identified pre-fall and post-fall periods for each faller. Because non-fallers had no event comparable to a fall that would define pre- and post-period, it was necessary to arbitrarily assign an index date to them. The pre- and post-periods needed to be comparable in length for fallers and non-fallers to control for time-related differences in costs. For example, a resident's health status may decrease over time as a result of aging. This could increase the likelihood of hospitalization and higher LTCF costs. So, having pre- and post-periods of substantially different lengths for fallers and non-fallers could artificially introduce cost differences between the groups.
We developed a procedure that set non-fallers' times in the pre- and post-periods equal to fallers' times in the comparable periods. We accomplished this by defining the index date for non-fallers as the date of the fifth MDS measurement provided. If no fifth measurement was made, we defined the index date as the date of the fourth MDS measurement. If no fourth measurement was provided, we used the third. This process continued until an index date was assigned. This process resulted in pre- and post-periods of approximately equal lengths for fallers and non-fallers.
Sample of Residents
The sample consisted of residents of a large, national, multifacility long-term-care chain who were institutionalized between January 1, 2002, and October 30, 2004; who were 65 years of age or older; and who were institutionalized for long-term care. We defined long-term care to include only residents who had been in the facility for a minimum of 6 months. Although this definition resulted in the exclusion of residents who were admitted for long-term care but who died or were transferred before residing in the facilities for 6 months, it ensured that the sample did not include residents admitted for short-term rehabilitative care and, consequently, included only residents in the facilities for long-term care.
The analysis included only residents for whom 75 or more days of data were available in the pre-period and at least 1 day was available in the post-period. The requirement of a minimum of 75 days of data in the pre-period was based on two considerations. First, longer time periods would provide more reliable estimates of pre-period costs. Second, a substantially longer period would have resulted in a much smaller sample.
Furthermore, we used only data for up to 105 days before and 375 days after the index date. This approximated a period of 3 months before and 12 months after the index date. We used the larger figures (105 and 375 days) because LTCFs may not record events such as falls and hospitalizations until several days after they occur. Using the longer periods made it more likely that we would measure all significant events in the 3- and 12-month time periods. We restricted the sample to a maximum of 105 days of data before the index date to provide a more consistent sample; that is, all residents in the sample would have roughly 3 months of data in the pre-period. We restricted it to a maximum of 375 days of data after the index date because the study's primary interest was fall-related costs in the 12 months following a fall.
We excluded from the sample any resident who had suffered a fall or fracture in the 180 days immediately preceding the index date. It was possible for non-fallers to have both falls and fractures in the 180 days before the index date, because the criteria for being a faller was having experienced a fall while in the LTCF. Those admitted as a result of falls and fractures suffered before admission, but who had not fallen since being admitted, were classified as non-fallers.
The sample also excluded those residents who did not have a propensity score match. Figure 1 summarizes the procedure through which we developed the analysis sample.
|
We used the following formula to estimate RUG reimbursement for each resident in both the pre- and post-index date periods:
|
|
If an MDS observation did not include a RUG classification, we based the reimbursement for the time period on the resident's average daily RUG reimbursement rate over his or her entire stay at the facility. We then estimated LTCF costs per resident per year (PRPY) for each period by dividing RUG reimbursement as calculated above by the number of days the resident spent in the period and multiplying by 365.
To estimate hospitalization costs, we first calculated the number of hospitalizations per resident in each period. We converted these figures to hospitalizations PRPY by dividing by the number of days the resident spent in the period and multiplying by 365. We multiplied hospitalizations PRPY by the average reimbursement for a fall-related hospitalization to estimate hospitalization costs PRPY for each period.
Two recent studies have provided estimates of the mean reimbursement for a fall-related hospitalization. Roudsari and colleagues (2005) used data from the MarketScan database to estimate a mean reimbursement for a fall-related hospitalization of $17,483 in 2004. Adjusting this estimate to 2003 dollars with the Medical Consumer Price Index yielded an estimate of $16,746. This estimate was based on a sample of 534 patients who were 65 years or older and who had Medicare supplemental insurance that was paid by their employers.
Carroll and colleagues (2005) used data from the Medical Expenditure Panel Survey to estimate a mean fall-related hospital reimbursement of $12,300 in 2002 dollars. Adjusting this figure with the Medical Consumer Price Index for 2003 yielded an estimate of $12,792. The Medical Expenditure Panel Survey is designed to provide estimates that are representative of the civilian, noninstitutionalized U.S. population. Carroll and associates based their estimate on 48 hospitalized fallers who were aged 65 years or older.
There was no apparent reason to prefer one of these estimates over the other. Roudsari and colleagues based their estimate on a larger number of participants, but from a sample that was not designed to be nationally representative. Carroll and associates based their estimate on a nationally representative sample, but with a much smaller number of participants. Both estimates were based on a population of elderly (65 years or older) individuals. As a result, we used the mean of the two figures—$14,769—as our estimate of the mean reimbursement for fall-related hospitalizations.
We then estimated fall-related costs by first calculating changes in reimbursement from pre- to post-period for each group, then calculating the difference in changes between the groups. We did this for hospital costs, LTCF costs, and the sum of hospital and LTCF costs.
We analyzed data using SAS 9.1 (SAS Institute, Cary, NC). Calculations of mean (SD) numbers of falls, fractures, hospitalizations, and costs in each period were weighted by the number of days the resident spent in the facility in the period. This ensured that residents who spent longer in the facility had a proportionately larger impact on mean calculations. We used chi-square tests to test associations between categorical variables and t tests to test differences between interval-level variables. We used dollar reimbursements, rather than the logarithmic transformation of reimbursements, in comparisons of mean costs. Thompson and Barber (2000) stressed that the estimate of interest in cost studies is the arithmetic mean, which is based on the dollar estimate, rather than the geometric mean, which is assessed by logarithmic transformation. In addition, Zhou, Melfi, and Hui (1997) have indicated that it is appropriate to use a t test for differences in arithmetic means for sample sizes greater than 1,000. We report statistical significance at p <.05.
The study was approved by Virginia Commonwealth University's institutional review board.
|
Results |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
|
As shown in Table 3, the difference in pre- to post-period changes in total (sum of hospital and LTCF) costs between fallers and non-fallers was $6,259 (51,210) PRPY. The 95% confidence interval was $2,034 to $10,484. Increases in hospital reimbursement accounted for about 60% of the difference, and changes in LTCF costs accounted for the remainder.
|
|
|
Discussion |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Our study's estimate of $6,200 is considerably higher than Sorensen and colleagues' (2006) judgment-based estimate of $1,892 for the year following a typical-case fall, but it is within the range of their estimates of $700 for the best-case scenario and $12,817 for the worst-case scenario. One would expect Sorensen and associates' estimates to be higher because they included not only inpatient hospital and LTCF costs, but also costs of outpatient physician and other ambulatory services, ERs, and ambulance service. However, Sorensen and colleagues estimated costs per fall, whereas our study estimated cost per faller. The typical faller in our study experienced 2.52 falls in the 212 day post-fall period. This would be equivalent to 4.3 falls in the year following the fall (2.52 falls/212 days x 365 days = 4.3 falls per year). Our estimate of $6,259 per faller per year divided by 4.3 falls per year provides a rough estimate of $1,456 per fall. This is much closer to Sorensen and associates' estimate. (Our estimate of costs per fall is not exactly comparable to Sorensen and colleagues' because it does not reflect the cost of care in the year following each fall—only in the year following the initial fall. Sorensen and associates' estimates calculated the cost for the year following each fall.)
Health care in the United States is provided in an increasingly cost-conscious and cost-constrained environment. In such an environment, widespread adoption of new health care interventions requires that interventions be proven not only effective, but also cost effective. Recent reviews have indicated that multifactorial falls risk assessment and management programs are effective in decreasing the incidence of falls (Gillespie, Gillespie, Lamb, Cumming, & Rowe, 2003; Rubenstein, 2006). Given that these programs are individualized to each resident and that they require a multidisciplinary approach, it is likely that they are expensive. Consequently, they may not be widely adopted until they are shown to be cost effective. The results of our study provide a baseline estimate that one can use to assess the cost-effectiveness of these and other interventions designed to decrease the incidence of falls in LTCFs. For example, our results indicate that the direct costs of a fall in a LTCF are roughly $1,500 per fall. This suggests that any intervention with a cost-effectiveness ratio of less than $1,500 per prevented fall not only would be cost effective, but would be cost saving. Our results also indicate direct costs of about $6,200 per faller for the year following a fall. This indicates that any intervention that would prevent a resident from falling for a year-long period for a cost of less than $6,200 would be cost saving.
The results of this study may also be useful to policy makers. The costs of falls in LTCFs can be compared to the costs of other diseases and conditions to determine their relative economic importance. Policy makers can use these comparisons to inform their decisions about allocation of resources. One can estimate the total direct costs of falls for the U.S. population residing in LTCFs as follows: There were approximately 1.5 million nursing home residents 65 years of age or older in 2003 (National Center for Health Statistics, 2006). Between 45% and 61% of nursing home residents experience a fall each year (Gryfe et al., 1977; Lord et al., 2003; Perry, 1982; Thapa et al., 1996; Tinetti, 1987; Tinetti et al., 1992; van Doorn et al., 2003). Given our estimate of fall-related costs of $6,259 for the year following the fall, these statistics indicate that the total direct costs of falls in LTCFs for 2003 were between $4.2 billion and $5.7 billion. (0.45 x 1.5 million x $6,259 = $4.2 billion; 0.61 x 1.5 million x $6,259 = $5.7 billion.)
One can also compare our estimates to the annual direct costs of other common diseases to determine how the economic burden of falls in LTCFs compares with that of other common conditions. (Cost estimates for other conditions are taken from National Institutes of Health estimates and adjusted to 2003 dollars using the Medical Consumer Price Index [Kirschstein, 2000]. The National Institutes of Health estimates apply to the total U.S. resident population—both institutionalized and community dwelling.) The cost of falls in LTCFs is substantially lower than the direct costs of asthma ($15.6 billion) and slightly lower than the costs associated with atherosclerosis ($6.5 billion), but greater than the direct costs of treating allergic rhinitis ($2.5 billion), Parkinson's disease ($3.0 billion), or multiple sclerosis ($4.0 billion). The cost of falls in LTCFs is roughly comparable to the direct cost of treating epilepsy ($4.5 billion). Decision makers in government and industry can use these comparisons to determine which diseases and conditions have the greatest economic impact and to inform decisions about which should receive a greater share of available funding.
Our results indicated that slightly more fallers than non-fallers died in the year following the fall (15.1% vs 12.4%, respectively). This suggests that falling may increase the risk of mortality. A number of studies have found higher mortality rates for fallers than for non-fallers (Donald & Bulpitt, 1999; Jantti et al., 1995; Nurmi, Luthje, & Kataja, 2004; Wild, Nayak, & Isaacs, 1981; Wolinsky, Johnson, & Fitzgerald, 1992). For example, a study of institutionalized patients in Finland found 5-year mortality rates of 75% for fallers compared with 58% for non-fallers (Nurmi et al., 2004). Similarly, an earlier study found 1-year mortality rates of 13.8% for non-fallers and 35.7% for fallers among elderly nursing home residents in Finland (Jantti et al., 1995). However, none of the studies that found significant mortality differences controlled for baseline differences in health or functional status between fallers and non-fallers. Dunn, Rudberg, Furner, and Cassel (1992) found significant differences in the unadjusted mortality rates of fallers and non-fallers among respondents in the Longitudinal Study on Aging. However, after the researchers controlled for differences in demographic factors, chronic diseases, and disabilities, the differences between fallers and non-fallers were no longer statistically significant. This suggests that falls have, at best, a small effect on mortality and that differences noted in earlier studies were probably a result of fallers having poorer baseline health than non-fallers.
We should note several limitations of our study. First, our results underestimate total fall-related costs. We did not have estimates of the costs of fall-related physician visits, ER visits, or outpatient hospital visits. These would add to the total costs of falls. However, studies of the cost of falls in community-dwelling elders suggest that these services compose a small part of total fall-related costs (Carroll et al., 2005; Rizzo et al., 1998; Roudsari et al., 2005). For example, Carroll and colleagues estimated that inpatient hospitalizations accounted for 60% to 65% of fall-related costs in a population of community-dwelling elders and that home health costs accounted for an additional 10% to 15%. LTCFs provide their residents with most of the services that home health agencies provide to community-dwelling patients. Both provide skilled nursing care; aides to help with personal care such as bathing, toileting, and dressing; social services; and physical, speech, and occupational therapy. This suggests that the 10% to 15% of costs attributable to home health agencies in the outpatient setting are included in the LTCF costs of nursing home residents. This suggests that our estimates, which include inpatient hospitalization and LTCF costs, probably cover at least 75% of the total direct costs of falls in LTCFs.
Furthermore, our estimates did not include indirect and intangible costs. Indirect costs measure the productivity that is lost as a result of disease or injury. Intangible costs include the pain and suffering that result from injury or disease. Although indirect costs are probably minimal for individuals in LTCFs, it seems likely that these persons would suffer substantial fall-related pain and suffering. In addition, relatives of LTCF residents may incur both indirect and intangible costs when their institutionalized relatives fall. Indirect costs would result from missing work or other productive activity, such as child care or volunteer work, to deal with activities such as hospital admissions, readmission to the LTCF after hospitalization, and changes in personal care resulting from the relative's fall. Intangible costs would result from the stress and anxiety associated with an injury to an institutionalized relative.
We restricted our sample to residents who had received care in a facility for a minimum of 6 months. As a result, the sample excluded residents who were admitted to a facility for long-term care but who died or were transferred within 6 months of admission. This would have biased our results only if residents who fall within 6 months of admission suffer different types or severities of injury than those who fall 6 or more months after admission.
We based our estimate of the cost of a fall-related hospitalization on studies conducted in (primarily) community-dwelling samples. This could have biased our results if the costs of fall-related hospitalizations for LTCF residents are different than those for community-dwelling fallers. We have no data on whether there is a difference. LTCF residents, as a group, are older and less healthy than community-dwelling elders, but this does not necessarily imply that fallers in LTCFs are older or sicker than elderly fallers in the community. Finkelstein et al. (2005), in a sample composed primarily of community-dwelling elders, found that fallers were substantially older and had substantially more comorbidities than non-fallers. The differences between fallers and non-fallers were much less pronounced in our sample of LTCF residents. This suggests, at the least, that the age and health differences between fallers in LTCFs and in the community are less substantial than the age and health differences between the overall populations of LTCFs and community-dwelling elders.
Finally, although we controlled for observed differences between fallers and non-fallers using propensity scores and matching, there may have been cost-related differences between fallers and non-fallers that were not measured in our study. We minimized this limitation to some extent by employing a before-and-after design to control for within-resident differences that were constant over the time period of the study.
In summary, our study indicates that LTCF residents who fall have significantly higher LTCF and hospital costs than do non-fallers. We estimate that fallers' costs average $6,200 PRPY higher than those of non-fallers. Fallers are also more likely to die or be transferred to a hospital. These estimates provide baseline values that one can use to estimate the cost-effectiveness of interventions designed to reduce the rates of falls in LTCFs and to inform policy makers about how best to allocate resources.
|
Footnotes |
|---|
A preliminary version of this research was presented at the 11th Annual Meeting of the International Society of Pharmacoeconomics and Outcomes Research, May 23, 2006, Philadelphia, PA.
1 School of Pharmacy, Virginia Commonwealth University, Richmond.
3 TNS Healthcare, Stamford, CT.
Decision Editor: William J. McAuley, PhD
Received for publication May 2, 2007. Accepted for publication August 13, 2007.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
65 years—United States, July 1991 – June 1992. Morbidity and Mortality Weekly Report, 45, 877-883.
| ||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
