The incremental direct costs associated with behavioral symptoms in AD

Patients and informants.

We recruited patients from nine clinical practices (six neurology practices, three geriatric medicine practices) in Michigan. At each site, patients and their families were contacted about the study by mail if they met inclusion criteria. The inclusion criteria consisted of a patient diagnosis of probable AD based on National Institute of Neurological and Communicative Disorders and Stroke–Alzheimer’s Disease and Related Disorders Association criteria16 and the availability of a knowledgeable informant who could be interviewed about the patient’s health and healthcare use. Patients were excluded if they met criteria for probable dementia with Lewy bodies (DLB)17 or had Down’s syndrome even if they met probable AD criteria, so that a more precisely defined and homogeneous patient sample could be examined in this study. Both the patient and the informant were required to speak English, so that the measurement instruments used in this study would be valid. Informants were excluded if they did have direct contact with the patient at least weekly.

Study design.

This is a cross-sectional study that compares patient characteristics at baseline to healthcare utilization and costs in the year preceding the baseline interview. The dependent variable in this study is direct costs of care for patients with AD, including estimates of formal and informal costs. Predictors of costs in this study include severity of cognitive impairment, behavioral symptoms, extrapyramidal features, and comorbid medical conditions. The Institutional Review Board of Michigan State University approved our study and informed consent procedure. After obtaining informed consent, patients with AD and their informants were interviewed separately for approximately 1 hour. Patients completed a short battery of neuropsychological tests and underwent a brief neurologic examination focused on signs of parkinsonism. Informants were interviewed to determine the patient’s use of healthcare services, receipt of unpaid care, frequency and severity of behavioral symptoms, and comorbid medical conditions.

Measurement of patient characteristics.

Patient demographic information including age, sex, education, race, marital status, living arrangement, household income, and health insurance coverage were collected. Behavioral symptoms were measured with the Neuropsychiatric Inventory (NPI).1 The NPI measures 12 neuropsychiatric disturbances common in dementia, specifically delusions, hallucinations, agitation, dysphoria, anxiety, apathy, irritability, euphoria, disinhibition, aberrant motor behavior, nighttime behavior disturbances, and appetite abnormalities. An informant rates each neuropsychiatric domain for the frequency and severity of symptoms in the patient with dementia. Frequency and severity scores are multiplied for each domain and added together for the total NPI score. The maximum score in each domain is 12 and the maximum total score is 144. Cognitive function was measured with the Mini-Mental State Examination (MMSE)18 and the severity of medical comorbidities was measured using the Cumulative Illness Rating Scale (CIRS).19 The CIRS scale quantifies significant health conditions in 13 major organ groups, whether they occurred in the past or are currently present, with those requiring treatment at the present time scoring higher than those not currently requiring treatment. The presence and severity of signs of parkinsonism were quantified using a scale that measures extrapyramidal signs (EPS) developed for use in patients with dementia.20

Healthcare utilization and unpaid caregiving time measurements.

A structured interview asked informants about the patient’s use of healthcare services in the past year. We quantified the number of days of hospital care with an overnight stay and the number of days of long-term care including stays in foster care homes, assisted living facilities, and nursing homes. Emergency room visits without an overnight stay and outpatient surgical procedures were not measured or included in cost estimates. Outpatient physician visits were quantified, but the type of physician and length of the visit were not specified. Inpatient physician visits were not quantified, but average costs for inpatient physician visits were included in the per diem cost estimates for overnight hospital stays. Home care services were measured and included professional home health care (e.g., nurse, physical therapist, speech therapist) and nonprofessional paid home health aides. The number of hours of paid home health care per week and the number of weeks of care in the past year were quantified for professional and nonprofessional care. Finally, use of prescription medications was determined including type of medication, dosage, and frequency of use.

Caregiving time was quantified using the Caregiver Activities Time Survey.21 This survey subdivides caregiving into paid and unpaid activities. For unpaid caregiving, the informant was asked to estimate the number of hours and minutes spent in a typical day for nine categories of care, including patient supervision. Hours during which both the patient and caregiver were sleeping were not included in caregiving time estimates. The informant was asked to include only caregiving activities that were new since the onset of dementia. The time spent in each of the subcategories for a typical day was summed and then annualized. This survey has shown good test-retest reliability, is significantly correlated with the severity of the patient’s cognitive and noncognitive symptoms, and was responsive to change in a clinical trial of a cholinesterase inhibitor.21

Direct cost estimates.

In this study we used cost terminology suggested by Gold et al.22 The term direct cost encompasses the value of all goods, services, and other resources associated with an illness or its treatment. We further divided direct costs into formal and informal costs. Direct formal costs included those costs in which a payment was made for services (e.g., physician visits, medications, hospital care, paid home care, and long-term care services). Direct informal costs were the economic value of unpaid caregiving. Total direct costs were the sum of formal and informal direct costs.

To estimate direct costs we used a gross costing method in which we totaled utilization of important types of care and then multiplied this utilization by a unit cost for each type of care.22 Table 1 shows the unit costs used in this study and their source.23-28⇓⇓⇓⇓⇓ These unit costs represent national or regional averages of charges or expenditures for health care services. Cost estimates from different time periods were inflated to 2001 dollars using the Medical Care component of the Consumer Price Index (CPI).29 The unit cost of foster home care for patients with dementia was the average of charges reported in a survey of a representative sample of Michigan adult foster homes that provided care for patients with dementia. This survey collected information on 465 adult foster care homes and 1,100 residents with dementia (unpublished data provided by Clare Collins, RN, PhD).

A replacement wage method was used to estimate the economic value of unpaid caregiving. In this approach, the unit cost of unpaid caregiving time is the hourly wage of a worker who would need to be hired to provide the same care that an unpaid caregiver is providing. For this study we used a low, midrange, and high estimate of the economic values of unpaid caregiving. Our low estimate is the national minimum wage ($5.11/hour), which is the lowest wage that could be used to hire a worker to provide care in the home. Our midrange estimate is $7.92/hour, which is the hourly cost of adult daycare services, which provide supervision and activities for patients with dementia, but minimal or no physical care for basic activities of daily living (ADL).28 Our high estimate of $9.38/hour is the national average wage for a nonprofessional home care worker.27

Statistical analysis.

We used two approaches to analyze incremental costs associated with greater noncognitive behavioral symptoms. The first approach compared direct costs in two NPI groups, after adjusting for covariates. One NPI group had active and problematic behavioral symptoms and the other did not. These results are valuable for health administrators and policy makers, because they estimate the incremental costs associated with a group of patients with AD who are identified because of the severity of their behavioral symptoms. These incremental cost estimates provide an idea of the resources that could be used to target this group of patients with AD for healthcare programs designed to optimally manage behavioral symptoms in patients with AD. The second approach used the entire sample to estimate the incremental total direct costs associated with a one-point increase in the NPI score, after adjusting for covariates. These results can be combined with the findings of efficacy trials for pharmacoeconomic evaluations.

NPI groups.

To facilitate the examination of direct costs and their subcomponents in groups of patients with AD distinguished by the severity of their behavioral symptoms, we divided patients into two groups based on their NPI score. For this analysis the high NPI group consisted of patients at or above the median value (i.e., 50th percentile) of 13, whereas the low NPI group consisted of patients below the median value. We chose this cutpoint because it distinguishes those patients with active, problematic noncognitive behavioral symptoms. For example, if a patient was rated as having symptoms occasionally (i.e., less than once per week) and the severity of these symptoms was rated as mild in all 12 categories of behavior measured by the NPI, then they would achieve a score of 12. Any other combination of more frequent or severe behavior could achieve a score of 13 or higher. The maximum score in any one of the 12 domains is 12 (occurring once or more per day and of marked severity). Thus, a patient would have to have symptoms in at least two domains to be classified in the high NPI group. Statistical comparisons of group differences in baseline characteristics were performed using t-tests for continuous variables and χ² analysis for categorical or frequency data.

Incremental, total costs in NPI groups.

Total direct costs and total formal costs in the high and low NPI groups were compared using analysis of covariance models (ANCOVA). Total direct costs were not transformed because the distribution did not violate normality assumptions. Total formal costs were log transformed because the untransformed distribution was highly skewed and violated normality assumptions for regression. For log transformed cost variables, one was added to zero values before log transformation and the expected cost on the log scale was retransformed to the non-log scale (i.e., US dollars) using the retransformation method of Duan.30 Costs in the high and low NPI groups were adjusted for group differences in severity of cognitive impairment (i.e., MMSE score), comorbid medical conditions (i.e., CIRS score), and extrapyramidal features (i.e., EPS score) using ANCOVA.

Subcomponents of costs in NPI groups.

For subcomponents of costs, we compared the low and high NPI groups by examining first the percentage of patients who utilized each type of care or service and then by examining the unadjusted mean utilization data and associated costs for those who used the service and for all members of the group. To test for significant differences between the NPI groups, we used the χ² test for percentages and the t-test for continuous variables that were normally distributed. For cost subcomponents with greater than 50% zero responses (e.g., long-term care costs, hospital costs, home health care costs, daycare costs for entire group), we used the nonparametric Wilcoxon test to look for significant differences between the NPI groups. In addition, for those utilization and cost subcomponents that were significantly different in the unadjusted comparisons, we compared the NPI groups using ANCOVA models that adjusted for severity of cognitive impairment and comorbid medical conditions. The low, midrange, and high estimates of the value of unpaid caregiving time were used for unpaid caregiving cost estimates. A logistic regression model was used to test the significance of differences in the probability of having a long-term care cost in the high and low NPI groups, after adjusting for severity of cognitive impairment and comorbid medical conditions.

Estimates of incremental, total direct costs per one-point NPI increase.

We developed multiple regression models using the entire sample that included NPI score as a continuous variable to predict total direct costs. This approach allowed the estimation of the incremental costs associated with a one-point increase in the NPI score, after adjusting for important covariates. We estimated these incremental costs using three different hourly values of unpaid caregiving time (i.e., low, midrange, and high estimates), because the economic value of an unpaid caregiver’s time is uncertain. The dependent variable, total direct costs, did not require transformation because the distribution did not violate normality assumptions. Covariates examined in these predictive models included patient age, education, sex, marital status, income, severity of cognitive impairment (i.e., MMSE score), comorbid medical conditions (i.e., CIRS score), and extrapyramidal features (i.e., EPS score). We chose a final model that included NPI score and all other significant covariates.

NPI groups.

Table 2 lists the characteristics of the entire sample and patients in the high and low NPI groups. The groups did not differ significantly in age, education, sex, percent living at home, income, percent with prescription drug coverage, or severity of extrapyramidal features. The high NPI group consisted of 71 patients and had a mean NPI score of 29.9, whereas the low NPI group consisted of 57 patients and had a mean NPI score of 4.7. The high NPI group had significantly greater cognitive impairment on the MMSE and significantly more comorbidities on the CIRS scale. These significant group differences are adjusted for in the cost comparisons of NPI groups.

Incremental total costs in NPI groups.

The high NPI group had significantly greater formal, informal, and total direct costs, after adjusting for severity of cognitive impairment and comorbid medical conditions. Figure 1 shows average annual adjusted costs (including formal, informal, and total direct costs) of the high and low NPI groups at three levels of cognitive impairment. The average annual formal cost increases in the high NPI group ranged from $3,162 in patients with mild cognitive impairment (i.e., at MMSE = 24) to $5,919 in patients with severe cognitive impairment (i.e., at MMSE = 5). Estimates of incremental informal and total costs depend on the value of unpaid caregiving used in the calculation of informal costs. The ranges described below represent calculations using our low and high estimates of the value of unpaid caregiving. For total informal costs, the high NPI group’s average annual costs were between $6,661 and $12,133 higher than the low NPI group. Total direct costs were between $10,670 and $16,141 more per year in the high NPI group than the low NPI group.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Average annual adjusted direct costs in patients with AD with and without prominent behavioral symptoms at three severity levels of dementia. Patients are grouped into high and low Neuropsychiatric Inventory (NPI) groups as described in the text. Black bars represent total formal costs, gray hatched bars represent total informal costs, and the entire bar represents total direct costs. Error bars show SEM. These cost estimates were made with analysis of covariance models that used NPI as a grouped variable (i.e., low NPI, high NPI) and Mini-Mental State Examination (MMSE) score and Cumulative Illness Rating Scale (CIRS) score as continuous variables. The estimates of costs for mild, moderate, and severe dementia used the following MMSE values: mild = 24, moderate = 15, severe = 5. For these cost estimates, CIRS score was held constant at the mean value of 20.8.

The ANCOVA model used to predict total formal costs described above was significant, with NPI group, MMSE score, and CIRS all significant independent predictors of formal costs (adjusted R² = 0.31). The ANCOVA model used to predict informal costs was significant, with NPI group and MMSE significant independent predictors of costs but not CIRS score (adjusted R² = 0.09). Similarly, for total direct costs the ANCOVA model was significant and NPI group and MMSE score were significant independent predictors of costs but not CIRS score (adjusted R² = 0.34). For the total direct cost model using the midrange value for unpaid caregiving, the parameters estimates and SEM were NPI group = $14,253 ± $3,619, MMSE score = −$1,159 ± $206, and CIRS score = $625 ± $439.

Subcomponents of costs in low and high NPI groups.

Table 3 summarizes healthcare utilization and subcomponent costs in the low and high NPI groups. These unadjusted comparisons show that the cost of unpaid care, prescription medications, and doctor visits were significantly greater in the high NPI group than the low NPI group. Group comparisons that adjusted for severity of cognitive impairment and comorbid medical conditions were significant also for these three subcomponents of costs. The cost of unpaid caregiving was the largest cost subcomponent for each NPI group and the cost subcomponent with the greatest cost increase in the high NPI group. The high NPI group was significantly more likely to be receiving formal long-term care and had significantly greater average long-term care costs. However, these group differences were not significant after adjusting for group differences in the severity of cognitive impairment and comorbid medical conditions.

Estimates of incremental, total direct costs per one-point NPI increase.

Patient age, sex, marital status, education, income, and extrapyramidal score were not significant independent predictors of total direct costs in multivariate models. A model using NPI, MMSE, and CIRS scores to predict total costs was chosen as our final model, because it was the simplest significant model that explained the most variance and the model in which all the variables included in the model were significant, independent predictors of costs. This model predicts an average annual increase in total direct costs of between $247 and $409 for each one-point increase in the NPI score, depending on whether a low, midrange, or high estimate of the economic value of unpaid caregiving is used. Figure 2 displays these incremental costs for 1 to 20 point changes in the NPI. The parameter estimates and SEM for the ANCOVA model that used the midrange estimate for the value of unpaid caregiving costs were NPI score = $353 ± $109, MMSE score = −$1,120 ± $214, and CIRS score $956 ± $429. The adjusted R² of the model was 0.31 (model p < 0.0001, NPI score p = 0.002).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. Average annual incremental total direct costs for changes in Neuropsychiatric Inventory (NPI) score between 1 and 20 points using a low, midrange, and high estimate of the economic value of unpaid caregiving time. The y-axis shows the cost difference of changing the NPI the number of points shown on the x-axis. The low estimate is designated with -▪-, the midrange estimate with -x-, and the high estimate with -▴-.