Abstract
BACKGROUND:
The progressive nature of Parkinson disease (PD), together with a lack of curative treatments, contributes to its economic burden.
OBJECTIVE:
To estimate the longitudinal incremental costs attributable to PD among Medicare beneficiaries.
METHODS:
In this retrospective cohort study, we used data from the Chronic Conditions Data Warehouse to identify Medicare beneficiaries with and without PD-related claims identified from 2006 to 2014 with follow-up until 2015. We grouped PD cases and controls based on their survival profiles using a grouping algorithm that used the following baseline measures: age, race, sex, and comorbidity. We identified 3 survival groups and used them to stratify the descriptive annual cost estimates in the 9 years after the index date. We estimated the incremental 1-, 3-, and 5-year costs of PD using generalized linear models (GLM) that controlled for baseline factors.
RESULTS:
We identified 27,394 cases and controls who were grouped into 3 survival groups. The mean age of the full study sample was 73 years. No material differences were found in the incremental cost of PD across the survival groups. Based on the multivariable GLM, the 1-year incremental cost of PD was $9,625 (95% CI, $9,054-$10,197). The 3-year incremental cost of PD was $20,832 (95% CI, $19,390-$22,274). The 5-year incremental cost of PD was $27,466 (95% CI, 25,088-$29,844).
CONCLUSIONS:
Among Medicare beneficiaries, PD is associated with excess costs compared with controls. We did not identify substantial differences in the incremental cost of PD across the survival groups.
What is already known about this subject
The annual incremental cost of Parkinson disease (PD) ranges from $8,000 to $10,000 per patient based on population-based studies.
There is a lack of longitudinal information about cost accumulation per patient with PD, including costs associated with the use of emergency department services.
A longitudinal examination of cost accumulation in PD is important given the progressive nature of the illness.
What this study adds
This study covers a 9-year follow-up period and provides information about costs associated with emergency department visits among other categories of health care resource utilization.
The 1-year, 3-year, and 5-year incremental costs of PD were $9,625 (95% CI, $9,054-$10,197), $20,832 (95% CI, $19,390-$22,274), and $27,466 (95% CI, $25,088-$29,844), respectively.
We identified 3 distinct survival groups using a novel machine-learning grouping algorithm, characterized group differences at baseline, and reported the group-specific PD costs over time using cohorts defined by follow-up periods that range from 1 year to 9 years.
Parkinson disease (PD) is a progressive neurodegenerative disorder characterized by various motor and nonmotor signs and symptoms.1 Patients with PD can experience physical, cognitive, and other nonmotor symptoms that lead to increased health care needs and impaired health-related quality of life.2-4 Current treatments ameliorate many of the motor and nonmotor symptoms5 but do not cure or prevent the neurodegenerative process. The progressive debilitating nature of the disease, together with a lack of curative treatments, contributes to the clinical and economic burden of PD. In the United States, it was estimated that 680,000 individuals aged 45 and above had PD in 2010, and the prevalence was expected to increase to 930,000 by 2020.6 Given the projected increase in the prevalence of PD, it is important to quantify its economic burden using recent data. In addition, longitudinal information regarding the economic burden of PD will be valuable given the progressive nature of the illness.
The number of published cost-of-illness studies related to PD has decreased in recent years following a spike in published studies between 2011 and 2013.7 In the United States, a recent study reported on the costs of PD using 2013 data,8 whereas the other available sources used claims data that date back to 2009 and earlier.9,10 Importantly, the majority of the available studies were characterized by a relatively short follow-up and the absence of longitudinal information regarding the economic burden of PD.8,9,11-13 Results from previous studies that used population-based data indicate that the annual incremental cost of PD ranges from $8,000 to $10,000 per patient.8,11,12 Additionally, it was found that PD costs vary by the study population (eg, Medicare population vs privately insured population) and disease severity as represented by the use of assistive device, institutionalization, or PD-related medication.8-11
This study estimated the economic burden of PD in patients aged 66 years or above at diagnosis. The long follow-up time in Medicare data allowed us to expand beyond prior work to estimate the cost attributable to PD longitudinally and use more recent data. In addition, applying a novel machine-learning grouping algorithm, we estimated the cost of PD within 3 groups defined by the distinct survival patterns.
Methods
STUDY DESIGN AND DATA SOURCE
This was an observational burden-of-illness study. We used data from a random 5% sample of Medicare enrollees selected from the Chronic Conditions Data Warehouse (CCW). The CCW data consist of Medicare files that contain fee-for-service Parts A and B claims in addition to prescription drug Part D claims. We used Parts A and B files for the years 2005-2015 and Part D files for the years 2006-2015 for our retrospective analyses. The study period was from 2005 to 2015.
STUDY SAMPLE
The study sample consisted of individuals with and without PD-related claims identified between 2006 and 2014. Individuals were identified as PD cases if they had 2 PD-related claims 6 to 12 months apart (Supplementary Figure 1 (161.3KB, pdf) , available in online article). The index date was defined as the date of the first qualifying claim. A PD-related claim was defined as a claim with a diagnosis code for PD; International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 332.0. We identified the “control eligible” group as Medicare beneficiaries with no PD-related claim at any time during the study period and continuous 24 months of Parts A, B, and D enrollment. The control eligible group also had to have an index date assigned at random that mirrored the distribution of index dates among the individuals diagnosed with PD.
Cases and controls with index date in 2007-2014 were required to have continuous Parts A, B, and D enrollment in the 12 months prior to the index date. Since Medicare Part D was started in 2006, cases and controls with an index date in 2006 were required to have Parts A and B enrollment in the 12 months prior to the index date as well as Part D enrollment for the portion of 2006 preceding their index date. All cases and controls were required to have continuous enrollment in Parts A, B, and D for a minimum of 12 months of follow-up post-index date. Patients were excluded if they had a health maintenance organization enrollment during the pre-index period or if they died or were censored within the first 12 months of follow-up.
COSTS
We conducted the study from a US payer (Medicare) perspective and calculated costs based on Medicare payments (reimbursements). Only direct costs were considered in the analysis. Cost was measured from the index date until the end of follow-up. We calculated the annual cost per patient longitudinally up to 9 years following the index date using monthly cost data from Medicare claims files. Cost data included claims from the following sources: hospital, emergency department, outpatient clinic, skilled nursing home, and home health hospice in addition to claims related to providers’ fees, durable medical equipment, and Part D prescription drugs costs. We estimated the total cost by summing costs from all sources. We inflated all costs to 2015 United States dollars (USD) using the medical care component of the Consumer Price Index.
COVARIATES
We evaluated a set of demographic covariates including age, sex, race, and comorbidity. We created age categories based on 10-year increments for ages between 65 and 84 years, and we created one group for ages less than 65 years and another group for ages 85 years and above. To account for comorbidities, we calculated the total count of conditions within the Hierarchical Condition Category (HCC) measure using claims from the 12-month pre-index period.14 We identified HCC comorbidities from claims for health service utilizations associated with the ICD-9 codes of the HCC conditions. We categorized the count of HCC conditions into quartiles based on the distribution in the data. We included information about the region of residence at the index date with the South region as the reference group. We also included a categorical variable to identify receipt of the low-income subsidy (LIS) at any time during the pre-index period. The Part D LIS program assists Medicare beneficiaries with Part D premiums and cost sharing for low-income beneficiaries.
STATISTICAL ANALYSES
Grouping Patients by Survival.
The process of grouping patients based on their survival profiles and using the Grouping Algorithm for Cancer Data (GACD) has been described previously.15,16 The GACD was originally developed for the analysis of survival patterns among individuals diagnosed with cancer, and it was shown to improve their cost prediction.16 As a grouping algorithm, the GACD can be applied to any clinical condition in which the examination of survival patterns is of interest. In the present study, the GACD was used to investigate and characterize cost accumulation based on survival patterns. The algorithm identified all the possible combinations of the age, race, sex, and HCC factors and estimated a survival curve for each combination. Combinations with fewer than 10 individuals were excluded, consistent with the statistical assumption of this approach.17 Then, using hierarchical clustering, the GACD grouped multiple combinations of baseline factors depending on the similarity between the survival curves. A dendrogram provided a graphical representation of the process whereby the closest survival curves were merged first until only one survival curve remained. Through cutting the dendrogram at the appropriate level, we identified the maximum number of clusters such that all survival curves were well separated. This process used the full sample and did not consider the PD group assignment (ie, “PD case” or “non-PD control”) when generating the survival curves. We identified 3 groups through this process. Then, we used these 3 groups to stratify the cost estimates. We also compared cases and controls within each survival group in terms of baseline factors. We compared between PD cases and controls within survival groups because individuals within a survival group were more similar to each other than they were to individuals outside their group.
Descriptive Statistics.
We reported the frequency and percentage distributions of the baseline categorical variables. The mean and SD were estimated for continuous variables, including age and length of follow-up. We compared these variables between cases and controls using the Pearson chi-square test for categorical variables and the t-test for continuous variables. The mean cost per person was estimated for the PD cases and controls in each of the 9 years after the index date, stratified by the survival group.
Total and Incremental Cost of PD.
In order to estimate the incremental costs of PD with the appropriate confidence intervals, we used generalized linear models (GLM) for each year of follow-up. The selection of model link and distribution family was based on the Pregibon and modified Park tests, respectively.18 To obtain incremental cost estimates in monetary units, we estimated the marginal effects of PD cases.19
We estimated the incremental 1-, 3-, and 5-year costs of PD separately by summing the total costs in each time period. Then, we modeled the costs using GLM including the baseline demographic and clinical factors discussed above as well as the indicator for PD cases.
To account for the costs of right-censored subjects, we used the inverse probability weighting (IPW) method suggested by Bang and Tsiatis.20 The method estimates the mean cost for the full sample based on the costs of uncensored observations. It gives higher weight for subjects with longer follow-up time. For the descriptive longitudinal cost estimation, we partitioned the follow-up time into 9 years and used the partitioned IPW approach.20 For the multivariable 3-year and 5-year cost estimates, we used the simple IPW approach without partitioning.
All statistical analyses were conducted using SAS Studio 3.71 (SAS Institute) and STATA 13.1 (StataCorp. 2013. Stata Statistical Software: Release 13). Statistical significance was based on two-sided α of P < 0.05. This study was approved by the University of Maryland, Baltimore Institutional Review Board (HP-00072257).
Results
We identified 15,254 PD cases and an equal number of Medicare beneficiaries without PD, who were selected at random (Supplementary Figure 2 (161.3KB, pdf) : study flowchart). After running the survival grouping algorithm, 14,118 cases and 13,276 controls (N = 27,394 individuals) were grouped into 3 main survival groups (groups 0, 1, and 2) and were included in the final analytic sample. The Kaplan-Meier survival curves of the 3 groups that were obtained after running the grouping algorithm are illustrated in the supplementary material (Supplementary Figure 3 (161.3KB, pdf) ). Cases and controls in group 1 had the highest survival rates, whereas cases and controls in group 2 had the lowest survival rates. Cases and controls in group 0 were in the middle between groups 1 and 2. The median survival for each group is shown in Table 1. The mean (SD) age of the full study sample was 73 (12.2) years, and the mean (SD) follow-up time was 47 (27) months. The majority of the sample was White (93%) and female (62%).
TABLE 1.
Median Survival of Parkinson Disease Cases and Controls Stratified by the Survival Group Generated From the GACD (N = 27,394)
| Group | N | Median survival (years) | 95% CI | |
|---|---|---|---|---|
| Lower | Upper | |||
| Group 0 | ||||
| Cases | 8,167 | 4.92 | 4.75 | 5.00 |
| Controls | 6,887 | Not reached | 9.50 | Not reached |
| Group 1 | ||||
| Cases | 3,693 | 6.00 | 5.67 | 6.25 |
| Controls | 4,517 | Not reached | Not reached | Not reached |
| Group 2 | ||||
| Cases | 2,258 | 4.25 | 4.08 | 4.50 |
| Controls | 1,872 | 8.83 | 7.92 | Not reached |
GACD = Grouping Algorithm for Cancer Data.
Table 2 illustrates the characteristics of the cases and controls in each survival group. The characteristics of cases and controls differed by survival group. For example, PD cases were more likely to be males (74%) in group 1 (the group with the best survival), whereas they were more likely to be females (93%) in group 2 (the group with the worst survival). Comparing cases with controls, PD cases were older, had a higher proportion of males, and had more comorbidities in all 3 survival groups. Compared with controls, cases had higher proportions of White patients in groups 0 and 1, and lower percentage of White patients in group 2. Compared with controls, PD cases had a higher proportion of LIS recipients in group 1.
TABLE 2.
Baseline Demographic and Clinical Characteristics of the Study Population (2006-2014)
| Full sample N = 27,394 n (%) | Group 0 | Group 1 | Group 2 | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Cases N = 8,167 n (%) | Controls N = 6,887 n (%) | P value | Cases N = 3,693 n (%) | Controls N = 4,517 n (%) | P value | Cases N = 2,258 n (%) | Controls N = 1,872 n (%) | P value | ||
| Follow-up time, mean (SD), m | 47 (27.0) | 44 (24.5) | 51 (28.5) | < 0.001a | 44 (25.7) | 52 (28.7) | < 0.001a | 43 (24.7) | 51 (29.3) | < 0.001a |
| Age, mean (SD), y | 73 (12.2) | 76 (9.2) | 71 (15.2) | < 0.001a | 72 (9.4) | 68 (9.1) | < 0.001a | 82 (7.3) | 72 (16.7) | < 0.001a |
| Age category, y | ||||||||||
| < 65 | 4,299 (16) | 807 (10) | 1,784 (26) | < 0.001 | 412 (11) | 655 (15) | < 0.001 | 78 (3) | 563 (30) | < 0.001 |
| 65-74 | 9,269 (34) | 1,616 (20) | 1,392 (20) | 2,622 (71) | 3,639 (81) | 0 (0) | 0 (0) | |||
| 75-84 | 9,929 (36) | 4,700 (58) | 2,833 (41) | 0 (0) | 0 (0) | 1,443 (64) | 953 (51) | |||
| > 84 | 3,897 (14) | 1,044 (13) | 878 (13) | 659 (18) | 223 (5) | 737 (33) | 356 (19) | |||
| Sex | ||||||||||
| Male | 10,540 (38) | 2,965 (36) | 2,350 (34) | 0.005 | 2,723 (74) | 2,275 (50) | < 0.001 | 162 (7) | 65 (3) | < 0.001 |
| Female | 16,854 (62) | 5,202 (64) | 4,537 (66) | 970 (26) | 2,242 (50) | 2,096 (93) | 1,807 (97) | |||
| Race | ||||||||||
| White | 25,363 (93) | 7,830 (96) | 6,470 (94) | < 0.001 | 3,481 (94) | 4,066 (90) | < 0.001 | 1,865 (83) | 1,651 (88) | < 0.001 |
| African American | 1,401 (5) | 152 (2) | 240 (3) | 212 (6) | 451 (10) | 197 (9) | 149 (8) | |||
| Other | 630 (2) | 185 (2) | 177 (3) | 0 (0) | 0 (0) | 196 (9) | 72 (4) | |||
| Count of HCC conditionsb | ||||||||||
| 0 | 4,751 (17) | 916 (11) | 1,257 (18) | < 0.001 | 599 (16) | 1,599 (35) | < 0.001 | 94 (4) | 286 (15) | < 0.001 |
| 1 | 5,343 (20) | 1,384 (17) | 1,569 (23) | 838 (23) | 1,187 (26) | 53 (2) | 312 (17) | |||
| 2-3 | 8,399 (31) | 2,372 (29) | 2,447 (36) | 839 (23) | 815 (18) | 1,118 (50) | 808 (43) | |||
| > 3 | 8,901 (32) | 3,495 (43) | 1,614 (23) | 1,417 (38) | 916 (20) | 993 (44) | 466 (25) | |||
| Region | ||||||||||
| Midwest | 7,235 (26) | 2,149 (26) | 1,794 (26) | 0.296 | 1,043 (28) | 1,226 (27) | 0.171 | 529 (23) | 494 (26) | 0.131 |
| Northeast | 5,391 (20) | 1,626 (20) | 1,428 (21) | 672 (18) | 763 (17) | 500 (22) | 402 (21) | |||
| South | 10,612 (39) | 3,192 (39) | 2,610 (38) | 1,406 (38) | 1,795 (40) | 906 (40) | 703 (38) | |||
| West and other/unknownc | 4,156 (15) | 1,200 (15) | 1,055 (15) | 572 (15) | 733 (16) | 323 (14) | 273 (15) | |||
| LIS | ||||||||||
| Yes | 9,892 (36) | 3,095 (38) | 2,713 (39) | 0.159 | 1,152 (31) | 1,138 (25) | < 0.001 | 970 (43) | 824 (44) | 0.642 |
| No | 14,877 (54) | 4,256 (52) | 3,490 (51) | 2,275 (62) | 3,003 (66) | 1,028 (46) | 825 (44) | |||
| Missing | 2,625 (10) | 816 (10) | 684 (10) | 266 (7) | 376 (8) | 260 (12) | 223 (12) | |||
The table presents column percentages for categorical variables.
a Based on two-sample t-test.
b HCC is defined based on the health care utilization in the 12 months before index date.
c The West and other/unknown regions were combined to avoid cell sizes less than 11 per data use agreement.
HCC = hierarchical condition categories; LIS = low-income subsidy; m = month; y = year.
Table 3 presents the descriptive annual mean cost per person among cases and controls, conditional on surviving the year of interest. For groups 1 and 2, the mean annual costs of patients with PD were higher than those of controls in years 1 through 9 (P < 0.05 for all time points). For group 0, the mean annual costs of patients with PD were higher than those of controls in years 1 through 8 (P < 0.05), and the costs were not statistically different in year 9. The incremental cost of PD was the highest in year 1 for the 3 survival groups. In year 1, the incremental cost of PD was $12,822 for group 0, $12,796 for group 1 and $12,731 for group 2. The annual incremental costs of PD were lower in the subsequent years, and they varied by survival group.
TABLE 3.
Descriptive Mean Annual Cost for Parkinson Disease Cases and Controls Surviving at the End of Each Follow-Up Year, Stratified by Survival Group (N = 27,394)
| Year | Group 0 | Group 1 | Group 2 | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Cases N = 8,167 | N | Controls N = 6,887 | N | Incremental cost | Cases N = 3,693 | N | Controls N = 4,517 | N | Incremental cost | Cases N = 2,258 | N | Controls N = 1,872 | N | Incremental cost | |
| 1 | 25,379 | 8,167 | 12,557 | 6,887 | 12,822a | 23,909 | 3,693 | 11,113 | 4,517 | 12,796a | 24,901 | 2,258 | 12,170 | 1,872 | 12,731a |
| 2 | 21,446 | 7,048 | 12,813 | 6,467 | 8,633a | 19,855 | 3,248 | 10,903 | 4,355 | 8,952a | 20,567 | 1,861 | 11,831 | 1,692 | 8,736a |
| 3 | 20,845 | 6,207 | 12,721 | 6,125 | 8,124a | 19,226 | 2,920 | 10,684 | 4,219 | 8,542a | 20,554 | 1,581 | 11,836 | 1,577 | 8,718a |
| 4 | 20,871 | 5,504 | 12,351 | 5,857 | 8,520a | 20,195 | 2,680 | 11,475 | 4,109 | 8,720a | 20,563 | 1,356 | 11,677 | 1,497 | 8,886a |
| 5 | 21,782 | 5,001 | 12,453 | 5,650 | 9,329a | 18,849 | 2,530 | 11,118 | 4,024 | 7,731a | 20,662 | 1,224 | 11,546 | 1,431 | 9,116a |
| 6 | 20,512 | 4,685 | 12,231 | 5,483 | 8,281a | 18,393 | 2,428 | 11,062 | 3,984 | 7,331a | 18,989 | 1,111 | 10,298 | 1,368 | 8,691a |
| 7 | 20,535 | 4,476 | 12,540 | 5,373 | 7,995a | 18,407 | 2,356 | 10,490 | 3,932 | 7,917a | 19,805 | 1,046 | 11,747 | 1,335 | 8,058a |
| 8 | 20,500 | 4,358 | 13,535 | 5,291 | 6,965a | 18,812 | 2,319 | 11,883 | 3,902 | 6,929a | 21,347 | 1,006 | 9,410 | 1,313 | 11,937a |
| 9 | 19,163 | 4,309 | 14,405 | 5,261 | 4,758 | 22,137 | 2,299 | 12,790 | 3,880 | 9,347a | 18,619 | 989 | 10,054 | 1,304 | 8,565a |
Cost estimates are for individuals who were alive or censored during the year of interest. Mean costs (USD) were adjusted for censoring using the inverse probability weighting method.
aStatistically significant at a two-sided α of < 0.05.
USD = United States dollars.
The cumulative mean annual costs for cases and controls are shown graphically in Figures 1, 2, and 3 for group 0, group 1, and group 2, respectively. The mean 9-year cumulative cost amounted to $118,799 for patients with PD in group 0 compared with $94,111 for controls. In group 1, the 9-year cumulative costs were $118,927 for cases and $90,834 for controls. In group 2, the 9-year cumulative costs were $109,148 for cases and $81,491 for controls.
FIGURE 1.
Cost Accumulation Over Time for Parkinson Disease Cases and Controls in Group 0 (N = 15,054)
FIGURE 2.
Cost Accumulation Over Time for Parkinson Disease Cases and Controls in Group 1 (N = 8,210)
FIGURE 3.
Cost Accumulation Over Time for Parkinson Disease Cases and Controls in Group 2 (N = 4,130)
Based on the multivariable GLM, the 1-year incremental cost of PD was $9,625 (95% CI, $9,054-$10,197). The mean 1-year cost of PD was 73% higher than the cost of controls. The 3-year incremental cost of PD was $20,832 (95% CI, $19,390–$22,274). The mean 3-year cost of PD cases was 53% higher than the cost of controls. In addition, the 5-year incremental cost of PD was $27,466 (95% CI, $25,088-$29,844). The mean cost of PD cases was 45% higher than the cost of controls based on the 5-year time horizon (Table 4). In terms of other cost determinants, age younger than 65 years was associated with higher 1-, 3-, and 5-year costs compared with age group 65-74 years, whereas age group 85+ was associated with lower 3-year costs. Age groups older than 74 were associated with lower 5-year costs. Females had higher 5-year costs than males. Compared with White race, African American race was associated with higher 1-, 3-, and 5-year costs, whereas other race groups were associated with higher 5-year costs. In addition, having one or more HCC conditions was associated with higher 1-, 3-, and 5-year costs compared with not having any HCC condition (Table 4).
TABLE 4.
Adjusted Incremental 1-, 3-, and 5-year Costs of Parkinson Disease Cases Compared With Controls Based on Multivariable GLM (N = 27,394)
| 1-year cost | 3-year cost | 5-year cost | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Variable | Cost ratio | 95% CI | Cost ratio | 95% CI | Cost ratio | 95 % CI | |||
| PD | 1.73 | 1.67 | 1.78 | 1.53 | 1.48 | 1.57 | 1.45 | 1.40 | 1.50 |
| Marginal effects (USD) | 9,625 | 9,054 | 10,197 | 20,832 | 19,390 | 22,274 | 27,466 | 25,088 | 29,844 |
| Age group, y | |||||||||
| < 66 | 1.14 | 1.09 | 1.20 | 1.18 | 1.13 | 1.24 | 1.23 | 1.17 | 1.30 |
| 66-74 | Reference | Reference | Reference | ||||||
| 75-84 | 1.01 | 0.98 | 1.05 | 0.98 | 0.95 | 1.01 | 0.92 | 0.89 | 0.95 |
| ≥ 85 | 1.01 | 0.96 | 1.05 | 0.91 | 0.87 | 0.95 | 0.79 | 0.76 | 0.82 |
| Sex | |||||||||
| Male | Reference | Reference | Reference | ||||||
| Female | 1.00 | 0.97 | 1.03 | 1.03 | 1.00 | 1.06 | 1.04 | 1.01 | 1.08 |
| Race | |||||||||
| White | Reference | Reference | Reference | ||||||
| African American | 1.13 | 1.05 | 1.21 | 1.15 | 1.07 | 1.22 | 1.14 | 1.06 | 1.23 |
| Other | 0.96 | 0.88 | 1.05 | 1.05 | 0.96 | 1.14 | 1.11 | 1.02 | 1.22 |
| HCC | |||||||||
| 0 | Reference | Reference | Reference | ||||||
| 1 | 1.43 | 1.35 | 1.53 | 1.37 | 1.30 | 1.44 | 1.37 | 1.30 | 1.45 |
| 2-3 | 1.83 | 1.73 | 1.94 | 1.73 | 1.64 | 1.81 | 1.67 | 1.59 | 1.76 |
| 4+ | 3.36 | 3.17 | 3.55 | 2.77 | 2.64 | 2.91 | 2.49 | 2.36 | 2.62 |
| Region | |||||||||
| South | Reference | Reference | Reference | ||||||
| Midwest | 1.01 | 0.97 | 1.04 | 0.99 | 0.95 | 1.02 | 0.99 | 0.96 | 1.03 |
| Northeast | 1.00 | 0.96 | 1.04 | 1.01 | 0.97 | 1.05 | 1.01 | 0.98 | 1.06 |
| West | 1.05 | 1.01 | 1.10 | 1.05 | 1.01 | 1.10 | 1.05 | 1.01 | 1.11 |
| Unknown/other | 0.50 | 0.35 | 0.71 | 0.59 | 0.41 | 0.85 | 0.58 | 0.41 | 0.82 |
| LIS | |||||||||
| No | Reference | Reference | Reference | ||||||
| Yes | 1.28 | 1.24 | 1.33 | 1.24 | 1.20 | 1.27 | 1.18 | 1.14 | 1.22 |
| Missing | 1.21 | 1.15 | 1.27 | 1.17 | 1.13 | 1.22 | 1.15 | 1.11 | 1.20 |
GLM = generalized linear models; HCC = hierarchical condition categories; LIS = low-income subsidy; PD = Parkinson disease; USD = United States dollars; y = years.
Discussion
We undertook this study to characterize cost accumulation over time among Medicare beneficiaries diagnosed with PD. With the projected increase in the prevalence of PD and given its progressive nature, it was important to quantify the economic burden of PD over time while comparing costs between PD cases and controls. Our study sample, which included PD cases and controls with up to 9 years of follow-up, provided a unique opportunity to consider the economic burden of PD over a relatively long period of time (ie, 9 years). Based on the results from multivariable regression analyses, the 1-year, 3-year, and 5-year incremental costs of PD were $9,625, $20,832, and $27,466, respectively. We did not extend the GLM beyond 5 years because using data with follow-up beyond 5 years would have resulted in the use of data with high censoring proportions. The censoring proportions for year 6 ranged from 35% to 62% and were higher than would be desirable for the application of the IPW method. Our results from the 1-year and 5-year models are consistent with the most recent cost-of-illness studies conducted from a payer perspective, which documented positive excess costs of PD.8-10
However, our estimated costs are not directly comparable with these studies because of differences in study design whereby prior studies used claims-based proxy indicators for disease progression (eg, the use of assistive devices, PD-related medication use, or evidence of admission to a skilled nursing facility), whereas we report costs for the full sample. The ranges from prior studies across the less severe to more severe groups could be useful for providing boundary estimates. Risk-adjusted 1-year costs (in 2013 USD) for mild to moderate and advanced PD groups were $8,751 and $14,839, respectively, among Medicare beneficiaries.8 Ranges for the 1-year excess costs (in 2009 USD) across groups categorized as newly diagnosed, early disease, requiring ambulatory assistive device use, and admitted to a nursing home were $2,197, $3,886, $21,565, and $50,277, respectively, for a cohort of Medicare beneficiaries.10 Ranges for the 5-year excess costs (in 2009 USD) across groups categorized as newly diagnosed, early disease, requiring ambulatory assistive device use, and admitted to a nursing home were $28,422, $14,806, $50,923, and $102,750, respectively.10 By comparison, 1-year excess costs (in 2010 USD) for groups defined by newly diagnosed, requiring ambulatory assistive device use, and admitted to a nursing home were $4,072, $26,467, and $37,410, respectively, for a cohort of privately insured individuals.9 Our results for the 1-year and 5-year excess costs (ie, $9,625 and $27,466, respectively) were within the ranges estimated by these prior studies.
With the relatively longer study follow-up and the availability of mortality data, we were also interested in distinguishing survival patterns in the event that there were implications for the cost estimation. The GACD allowed us to consider cost accumulation amongst groups of patients that were similar based on their survival patterns. Within these groups, we estimated the incremental cost of PD. We did not identify substantial differences in the descriptive incremental cost of PD across the GACD groups and thus pooled the data for the regression analyses reported in Table 4. The results of the GACD yielded intriguing results for future studies that examine survival differences between PD cases and controls. Given that the survival curves resulted from data driven clustering, it was instructive to consider the combinations of baseline characteristics of the resulting groups (and cases and controls within each group) as any resulting differences in these combinations may identify potential predictors of survival. Comparing group 1 (best survival) and group 2 (worst survival), group 2 had above average proportions for the oldest age group, female sex, and the highest counts of HCC conditions.
Comparing group 1 (best survival) and group 2 (worst survival), we found that group 2, the smallest of the 3 groups, was nearly entirely composed of females (93% among cases and 97% among control), whereas females were not the majority in the group 1 sample (26% among cases and 50% among controls). These results appear to be inconsistent with published findings showing that female sex is associated with lower risk of death.21 Ad hoc survival analyses using our dataset (ie, main effect Cox proportional hazards models among PD cases) were consistent with the published literature in that females had lower risk of death. The reason behind these seemingly contradictory results using our dataset is that the GACD creates combinations (ie, interactions) of baseline factors and then clusters the combinations with similar survival patterns into main survival groups. Overall, among PD cases, female sex was not a risk factor for worse survival, but combinations of female sex with other specific baseline factors, such as age 75-84 years, White race, and HCC count of 2-3 placed the individual in the worst survival group. In comparison, male PD cases with the same baseline characteristics were placed in the intermediate survival group (group 0). Our sample includes a higher proportion of females than prior studies. However, it is not the only study to report a majority female proportion in the PD setting. The sex distribution in our original sample was 57% female among the PD cases and 63% female among unmatched controls. Prior claims-based studies on the use of health services, costs, and outcomes among individuals with PD have reported proportions of females ranging from 43% to 60%.10,21-24
LIMITATIONS
This study had a few limitations that should be taken into consideration. First, we did not investigate costs in subgroups defined by disease progression owing to limitations associated with examining progression-related health care resource utilization. Ongoing work has highlighted the challenges associated with identifying all individuals with progression using claims data and it remains an active area of research. The investigation of the cost implications associated with disease progression can be better conducted using datasets that contain rich measures of comorbidity, progression, and health care resource utilization. Second, the study is based on data for Medicare beneficiaries with fee-for-service coverage, and the results may not be generalizable to individuals with other types of coverage. Third, the identification of cases was based on a claims-based algorithm, and there could be a possibility of overlooking true PD cases who did not meet the identification algorithm, eg, individuals who have been diagnosed with PD but who did not have the qualifying number and pattern of PD-related claims during the study period. Lastly, the comparison of incremental cost estimates across sample subsets defined by the follow-up time is subject to survival bias. Survival bias arises because individuals with PD who contribute to the ninth year cost estimates are those with longer survival, who are expected to be healthier and generate lower costs than patients with shorter survival.
Conclusions
This study estimated incremental costs associated with PD using a large sample of PD cases and controls identified between 2006 and 2015. Leveraging the relatively long follow-up, we compared costs between cases and controls over varying lengths of follow-up and estimated covariate-adjusted incremental costs over 1-year, 3-year, and 5-year periods. We found that PD is associated with excess costs among Medicare beneficiaries. The 1-year, 3-year, and 5-year incremental costs of PD were $9,625, $20,832, and $27,466, respectively. Future studies using datasets with richer clinical measures can provide additional insight into clinical factors (eg, disease progression) associated with costs.
REFERENCES
- 1.van Rooden SM, Heiser WJ, Kok JN, Verbaan D, van Hilten JJ, Marinus J. The identification of Parkinson disease subtypes using cluster analysis: a systematic review. Μου Disord. 2010;25(8):969-78. doi: 10.1002/mds.23116 [DOI] [PubMed] [Google Scholar]
- 2.Kwok JYY, Auyeung M, Chan HYL. Examining factors related to health-related quality of life in people with Parkinson’s disease. Rehabil Nurs. 2020;45(3):122-30. doi: 10.1097/rnj.0000000000000179 [DOI] [PubMed] [Google Scholar]
- 3.Corallo F, De Cola MC, Lo Buono V, Di Lorenzo G, Bramanti P, Marino S. Observational study of quality of life of Parkinson patients and their caregivers. Psychogeriatrics. 2017;17(2):97-102. doi: 10.1111/psyg.12196 [DOI] [PubMed] [Google Scholar]
- 4.Dowding CH, Shenton CL, Salek SS. A review of the health-related quality of life and economic impact of Parkinson’s disease. Drugs Aging. 2006;23(9):693-721. doi: 10.2165/00002512-200623090-00001 [DOI] [PubMed] [Google Scholar]
- 5.Connolly BS, Lang AE. Pharmacological treatment of Parkinson disease: a review. JAMA. 2014;311(16):1670-83. doi: 10.1001/jama.2014.3654 [DOI] [PubMed] [Google Scholar]
- 6.Marras C, Beck JC, Bower JH, et al. Prevalence of Parkinson disease across North America. NPJ Parkinsons Dis. 2018;4(1):21. doi: 10.1038/s41531-018-0058-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Bovolenta TM, de Azevedo Silva SMC, Arb Saba R, Borges V, Ferraz HB, Felicio AC. Systematic review and critical analysis of cost studies associated with Parkinson disease. Parkinsons Dis. 2017;2017:3410946. doi: 10.115/2017/3410946 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Dahodwala N, Li P, Jahnke J, et al. Burden of Parkinson’s disease by severity: health care costs in the US Medicare population. Mov Disord. 2021;36(1):113-42. doi: 10.1002/mds.28265 [DOI] [PubMed] [Google Scholar]
- 9.Johnson SJ, Kaltenboeck A, Diener M, et al. Costs of Parkinson Disease in a privately insured population. Pharmacoeconomics. 2013;31(9):799-806. doi: 10.1007/s40273-013-0075-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Kaltenboeck A, Johnson SJ, Davis MR, et al. Direct costs and survival of Medicare beneficiaries with early and advanced Parkinson’s disease. Parkinsonism Relat Disord. 2012;18(4): 321-26. doi: 10.1016/j.parkreldis.2011.11.015 [DOI] [PubMed] [Google Scholar]
- 11.Noyes K, Liu H, Li Y, Holloway R, Dick AW. Economic burden associated with Parkinson disease on elderly Medicare beneficiaries. Mov Disord. 2006;21(3):362-72. doi: 10.1002/mds.20727 [DOI] [PubMed] [Google Scholar]
- 12.Huse DM, Schulman K, Orsini L, Castelli-Haley J, Kennedy S, Lenhart G. Burden of illness in Parkinson disease. Mov Disord. 2005;20(11):1449-54. doi: 10.1002/mds.20609 [DOI] [PubMed] [Google Scholar]
- 13.Bhattacharjee S, Sambamoorthi U. Co-occurring chronic conditions and healthcare expenditures associated with Parkinson disease: a propensity score matched analysis. Parkinsonism Relat Disord. 2013;19(8):746-50. doi: 10.1016/j.parkreldis.2013.02.019 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Pope GC, Ellis RP, Ash AS, et al. Diagnostic cost group hierarchical condition category models for Medicare risk adjustment. Health Economics Research, Inc; 2000. [PMC free article] [PubMed] [Google Scholar]
- 15.Qi R, Zhou S. A comparative study of algorithms for grouping cancer data. IAENG International Conference on Data Mining and Applications; 2014; Hong Kong. [Google Scholar]
- 16.Onukwugha E, Qi R, Jayasekera J, Zhou S. Cost prediction using a survival grouping algorithm: an application to incident prostate cancer cases. Pharmacoeconomics. 2016;34(2):207-16. doi: 10.1007/s40273-015-0368-6 [DOI] [PubMed] [Google Scholar]
- 17.Chen D, Xing K, Henson D, Sheng L, Schwartz AM, Cheng X. Developing prognostic systems of cancer patients by ensemble clustering. J Biomed Biotechnol. 2009;2009:632786. doi: 10.1155/2009/632786 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Glick HA, Doshi JA, Sonnad SS, Polsky D. Economic evaluation in clinical trials. OUP Oxford; 2014. [Google Scholar]
- 19.Onukwugha E, Bergtold J, Jain R. A primer on marginal effects—part II: health services research applications. Pharmacoeconomics. 2015;33(2):97-103. doi: 10.1007/s40273-014-0224-0 [DOI] [PubMed] [Google Scholar]
- 20.Wijeysundera HC, Wang X, Tomlinson G, Ko DT, Krahn MD. Techniques for estimating health care costs with censored data: an overview for the health services researcher. Clinicoecon Outcomes Res. 2012;4:145-55. doi: 10.2147/CEOR.S31552 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Willis AW, Schootman M, Kung N, Evanoff BA, Perlmutter JS, Racette BA. Predictors of survival in patients with Parkinson disease. Arch Neurol. 2012;69(5):601-07. doi: 10.1001/archneurol.2011.2370 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Pressley J, Louis E, Tang M-X, et al. The impact of comorbid disease and injuries on resource use and expenditures in parkinsonism. Neurology. 2003;60(1): 87-93. doi: 10.1212/wnl.60.1.87 [DOI] [PubMed] [Google Scholar]
- 23.Gross A, Racette BA, Camacho-Soto A, Dube U, Searles Nielsen S. Use of medical care biases associations between Parkinson disease and other medical conditions. Neurology. 2018;90(24):e2155-65. doi: 10.1212/WNL.0000000000005678 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Wei Y-J, Palumbo FB, Simoni-Wastila L, et al. Antiparkinson drug adherence and its association with health care utilization and economic outcomes in a Medicare Part D population. Value Health. 2014;17(2):196-204. doi: 10.1016/j.jval.2013.12.003 [DOI] [PubMed] [Google Scholar]