Abstract
Background/Rationale:
Alzheimer’s disease (AD) represents a serious public health issue affecting approximately 5.4 million individuals in the United States and is projected to affect up to 16 million by 2050. This study examined health care resource utilization (HCRU), costs, and comorbidity burden immediately preceding new diagnosis of AD and 2 years after diagnosis.
Methods:
This study utilized a claims-based, retrospective cohort design. Medicare Advantage members newly diagnosed with AD (n = 3374) were compared to matched non-AD controls (n = 6748). All patients with AD were required to have 12 months of continuous enrollment prior to AD diagnosis (International Classification of Diseases, Clinical Modification [ICD-9] 331.0), during which time no diagnosis of AD, a related dementia, or an AD medication was observed. Non-AD controls demonstrated no diagnosis of AD, a related dementia, or a prescription claim for an AD medication treatment during their health plan enrollment. Medical and pharmacy claims data were used to measure HCRU, costs, and comorbidity burden over a period of 36 months (12 months pre-diagnosis and 24 months post-diagnosis).
Results:
The HCRU and costs were greater for AD members during the year prior to diagnosis and during postdiagnosis years 1 and 2 compared to controls. The AD members also displayed greater comorbidity than their non-AD counterparts during postdiagnosis years 1 and 2, as measured by 2 different comorbidity indices.
Conclusions:
Members newly diagnosed with AD demonstrated greater HCRU, health care costs, and comorbidity burden compared to matched non-AD controls.
Introduction
The prevalence of Alzheimer’s disease (AD) in the United States during 2012 has been estimated to be 5.4 million affected individuals, 5.2 million of which are aged 65 and older. 1 By 2050, the prevalence is projected to reach as high as 16 million, with nearly a million new cases of AD expected to be diagnosed annually. 1 The demands on health care resource use and the associated costs are anticipated to increase in parallel to the significant growth in AD, with health care costs increasing from $200 billion in 2012 to $1.1 trillion in 2050. 1 The need to better understand the course and burden of AD from onset is imperative, given the disease has reached epidemic proportions and will be a significant strain to the US health care system.
A number of previous studies have investigated health care resource utilization (HCRU) and costs associated with AD and related dementias in a variety of populations. 2 –13 In these studies, AD has been consistently associated with substantial direct health care costs and medical service utilization. When compared to non-AD control groups, the comorbidity level, medical service utilization, and direct health care costs are generally higher among patients with AD compared to controls; however, the methodology of the previous research is heterogeneous, leading to substantial variation in findings and specifically cost estimates. 14 Case identification has been frequently based on a range of diagnosis codes for related dementias and not specific to AD. Some studies exclusively utilize diagnosis codes for case identification, 8,9,13 whereas others allow for medication proxies of diagnosis of AD. 9,10 Much of the previous research is dated and may not reflect current utilization patterns and treatment options. Finally, much of the existing research based on the secondary data sources has focused on prevalent case identification with follow-up periods of 12 months or less and annualized or imputed health care cost data. 8 –11
In order to better understand the impact of AD on health care cost and utilization proximal to the period of onset, we conducted a multiyear investigation of HCRU and costs among Medicare beneficiaries. We identified newly diagnosed Medicare Advantage Prescription Drug (MAPD) plan members with AD, in order to examine patterns of HCRU and direct health care cost over a 3-year period, one year prediagnosis and 2 years post diagnosis of AD. This study utilized recent administrative claims data from 2007 to 2011 and compared HCRU and costs for AD members to a non-AD control group in order to estimate the incremental impact associated with AD.The comorbidity burden and the comorbidity prevalence were also examined during the period after initial diagnosis of AD.
Methods
Patient Selection
This study is a retrospective, observational, and longitudinal cohort study based on the administrative claims data. The research protocol was reviewed and approved prior to study initiation by an independent institutional review board and received a waiver of Health Insurance Portability and Accountability Act authorization.
Enrollment, medical, and pharmacy claims data from a large MAPD plan were examined for the time period covering January 1, 2007, to June 30, 2011. Members with a diagnosis of AD were identified based on the presence of an initial diagnosis for AD (International Classification of Diseases, Clinical Modification [ICD-9] code 331.0) in their medical claims records. The date of first observed diagnosis of AD was assigned as the index date. Members were required to be aged ≥30 years as of the index date, to have continuous enrollment for ≥12 months preindex date, and to have continuous postindex enrollment ≥ 24 months. Members were excluded if they displayed a diagnosis for AD, a related dementia, or an AD-related prescription claim (acetylcholinesterase inhibitor or memantine) during the preindex period. This approach for patient selection was intended to limit the AD cohort to newly diagnosed AD patients.
The control cohort was constructed by identifying MAPD members aged ≥ 30 years without an AD diagnosis. Members were ineligible to serve as controls if they had a diagnosis of AD or a related dementia, or a prescription claim for an acetylcholinesterase inhibitor and/or memantine at any point from January 1, 2007, to June 30, 2011. The control cohort was constructed by performing a 1:2 match of AD members to eligible non-AD members on plan year enrollment, age ±1 year, gender, region, and race/ethnicity. Matched non-AD members were assigned an index date that corresponded to the index date of their AD cohort match.
Measures
Health Care Resource Utilization
The HCRU was based on the medical claims, and utilization was assigned to outpatient, inpatient, emergency department, home health service, or care facility categories based on the place of service. The percentage of each cohort utilizing services and the number of encounters for each service were calculated for each study year (preindex, postindex year 1, and postindex year 2). Medication use for the treatment of AD was determined based on the National Drug Code numbers associated with pharmacy claims.
Health Care Costs
All cost calculations included both plan paid and member paid cost (including copayments, coinsurance, and deductibles). Total medical expenses were defined as the cumulative expenses associated with medical claims for outpatient, inpatient, emergency department, home health service, and care facility encounters during the period of observation. Total pharmacy expenses were defined as the cumulative cost associated with pharmacy claims during the period of observation. Pharmacy cost for members included in the AD cohort was stratified into AD and non-AD pharmacy cost. Total non-AD pharmacy expenses for members included in the AD cohort were defined as the total pharmacy cost minus AD pharmacy costs. All cost data were adjusted to 2011 costs, using the medical care component of the Consumer Price Index.
Comorbidity Measures
The performance of different approaches to measuring comorbidity using administrative claims data has not been well characterized in the AD population; therefore, comorbidity burden was measured via 2 complementary approaches, one using medical claims data and second measure based on the pharmacy claims data. Deyo-Charlson Comorbidity Index (DCI) scores were determined based on the diagnostic codes associated with medical claims. 15 The Deyo implementation of the Charlson Comorbidity Index consists of identifying 17 disease state categories based on the ICD-9 codes, and the index score is determined based on weighting the individual categories based on the disease state severity such that a higher score indicates a greater comorbidity burden. 16 –18 The RxRisk-V score is determined based on the identification of 45 distinct comorbid conditions via their associated medication treatments. We calculated an unweighted RxRisk-V score by summing the number of unique comorbid conditions, such that a higher score indicates a greater comorbidity burden. Three RxRisk-V categories (neurogenic bladder, ostomy, and urinary incontinence) were not included because claims for products associated with these conditions (urinary catheters, colostomy supplies, and diapers/pads) are not captured in pharmacy claims. We did not map these 3 comorbidity categories to new medication exposures and used the original RxRisk-V specifications provided by Sloan et al. 16
The DCI and RxRisk-V scores were calculated for the 12-month preindex period, and independently for each 12-month period of postindex follow-up. The dementia category was omitted from the calculation of both the DCI and the RxRisk-V scores. In addition to these summary scores, the prevalence of specific comorbid conditions of interest was determined in order to provide resolution of key comorbid conditions that may not be reflected in the DCI. Specific comorbidities of interest were identified based on the ICD-9 codes observed during the entire 24-month postindex observation period. Raw prevalence ratios were calculated by dividing the prevalence of the specified comorbidity among members with AD by the prevalence of the comorbidity among non-AD control group.
Statistical Analysis
Between-group differences in HCRU were examined using chi-square tests for comparison of percentages and t tests for comparison of means. Differences in health care costs were evaluated using t tests for comparison of means. A repeated measures analysis was performed to assess the relationship between AD status and cost outcomes over time, after adjusting for a specified list of covariates. The outcomes data used in the analysis consist of information collected on each member over 3 discrete 12-month time periods: preindex year, year 1 postindex, and year 2 postindex. To assess the relationships of interest, analysis of covariance models was employed using an autoregressive covariance specification to account for the within-member correlation between repeated measures. The cost outcomes modeled were total health care cost, total medical cost, and total prescription drug cost. Selection of variables to be included in the models was performed a priori. All models included age, gender, race/ethnicity (white vs other), geographic region (south vs other), dual eligibility/low-income subsidy (LIS) status, study year, preindex RxRisk-V, and AD status. In separate analyses using those same models, an interaction term for AD status by study year was added to assess the relationship between AD status and outcome at each time point.
All analyses were conducted using SAS Enterprise Guide version 4.2 (Cary, North Carolina) and all models were run with the GLIMMIX procedure. The least-squares means (LSMEANS) statement with PDIFF option was used in the interaction models to compare adjusted mean estimates for each time period. These comparisons were performed to assess the relationship between AD status and outcome during each period. To control for data skewness and possible overdispersion, all models used a log normal distribution and the identity link function. Zero values on cost outcomes were set to 1 to allow for the inclusion of all data in the model. 19 Adjusted costs were back transformed from the log scale to allow for interpretation in the original units of the outcome measure. Back transformation was conducted via the smearing method using the LSMEANS estimate from the interaction model. 20 Incremental cost associated with AD was calculated as the difference in adjusted cost between the AD and the control cohorts. Statistical significance was determined using α <.05.
Results
Patients
A total of 42 426 members were initially identified with diagnosis of AD on a medical claim between January 1, 2008, and June 30, 2009. After applying the remaining study patient selection and continuous enrollment criteria (Figure 1), a total of 3374 MAPD members with newly diagnosed AD were identified and matched to 6748 non-AD controls. As expected based on the matching approach employed, there were no between-group differences in the demographic composition of the cohorts (Table 1); however, a greater proportion of patients in the AD cohort were receiving a Medicare part D LIS (17.4% vs 15.4%, P = .01) or were dual eligible (13.4% vs 11.5%, P = .006).
Attrition flow chart for Alzheimer’s disease (AD) cohort patient selection.
Demographic and Clinical Characteristics of Members Newly Diagnosed With AD and Matched Non-AD Controls.
Abbreviations: AD, Alzheimer’s disease; n, number of members; SD, standard deviation.
a P value from t test of means.
b P value from chi-square test.
Utilization and Costs
The between-group difference in percentage utilizing and mean encounters was statistically significant for each utilization category at all time points (Table 2). As shown in Table 3, the between-group difference in mean unadjusted health care costs was statistically significant for each cost component at all time points, with the exception of preindex total pharmacy costs ($1718 ± $2784 vs $1651 ± $2726, P = .25). Total pharmacy and non-AD pharmacy costs were significantly greater during postindex years 1 and 2, compared to the control group (P < .001 for both).
Health Care Resource Utilization for Members Diagnosed With AD and Matched Non-AD Controls.a
Abbreviations: AD, Alzheimer's disease; n, number; Enc, encounters; SD, standard deviation; N/A, not applicable.
aAll between-group comparisons at individual time points are statistically significant (t test of means, P < .001).
bPercentage of cohort with 1 or more claims for a utilization category during the given observation period.
cMean number of encounters associated with utilization category during the given observation period.
dStandard deviation associated with mean number of encounters.
Unadjusted Mean Annual Health Care Costs per Member Diagnosed With AD and Matched Non-AD Controls.a
Abbreviations: AD, Alzheimer’s disease; n, number; SD, standard deviation.
aAll between-group comparisons at individual time points are statistically significant (t test of means, P < .001) with the exception of preindex pharmacy cost (P = .25). All costs are reported in 2011 dollars based on the medical care component of the Consumer Price Index.
Multiple regression analysis found that members with AD had higher total health care, medical, and prescription drug costs than non-AD members during the overall study observation period (all P < .001, from models without the AD status by study year interaction term). The models that included the interaction term for AD status by study year found that members with AD had higher total health care and total medical costs than non-AD members at each of the time points examined (all P < .001).
The net differences in adjusted total health care, medical, and pharmacy costs between the AD and the control cohorts are presented in Table 4. The incremental annual total health care cost associated with AD was $2331 during the prediagnosis year, increased to $9333 during the year immediately after diagnosis, and was $5761 during the second year after diagnosis (P < .001 for all). Pharmacy costs were lower among the AD cohort during the preindex period (incremental cost preindex: −$135, P < .001); however, pharmacy costs were greater among the AD cohort during both postindex periods (incremental cost postindex: year 1, $1744, P < .001; year 2, $2210, P < .001).
Incremental Adjusted Annual Health Care Costs Associated With AD.
Abbreviation: AD, Alzheimer’s disease.
aIncremental adjusted cost is equal to the adjusted cost estimate for the AD cohort minus the adjusted cost estimate for the control cohort. P < .001 from the model with the AD status study period interaction.
Comorbid Conditions
The DCI scores were greater among the AD group compared to the control group at each observation period (all P < .001, Table 5). RxRisk-V scores were similar between the groups during the preindex period (P = .06) but greater among AD members during both postindex years (both P < .001). The greatest prevalence ratios were observed for Parkinson’s disease (8.87), epilepsy (5.26), and mood disorders (3.12); however, the prevalence of a broad range of included comorbid conditions was higher among the AD cohort compared to the control (Table 6).
Mean (SD) Comorbidity Index Scores During Preindex Year, Postindex Year 1, and Postindex Year 2 for Members Diagnosed With AD and Matched Non-AD Controls.a
Abbreviations: AD, Alzheimer’s disease; n, number of members; SD standard deviation.
aAll between-group comparisons at individual time points are statistically significant (t test of means, P < .001) with the exception of between-group comparison of preindex RxRisk-V (P = .06).
Prevalence of Comorbid Conditions Among Members Diagnosed With AD and Matched Non-AD Controls During 24-Month Postindex Follow-Up Period (Sorted by Prevalence Ratio).
Abbreviations: AD, Alzheimer’s disease; n, number of members.
a P value from chi-square test.
Discussion
This study reports HCRU and direct health care cost over a 3-year period proximal to an initial diagnosis of AD for Medicare beneficiaries enrolled in a Medicare Advantage plan. Similar to previous research, members with AD consistently demonstrated higher levels of HCRU and greater health care costs when compared to non-AD controls. 2 –13,21 Medical cost drove the observed difference in total health care cost across all time frames, including the period of time prior to diagnosis. Within the AD cohort, medical costs increased substantially during the first year after diagnosis but decreased during the second postindex year. With the incidence of AD expected to nearly double over the next few decades, it will be important to identify factors that contribute to the observed trends, to develop effective tactics to coordinate care, and to provide cost-effective, evidence-based treatment for patients with AD. The findings of the current study are useful for health care providers, health plan administrators, and others to understand the impact of AD on HCRU and direct health care costs proximal to the time of new diagnosis of AD.
After controlling for covariates including RxRisk-V score, pharmacy cost during the preindex period was greater among non-AD controls, albeit this difference was small relative to incremental pharmacy cost observed in the postindex periods. The incremental pharmacy cost estimates reported during the postdiagnosis period in the current analysis are similar in magnitude to findings from Zhao et al 10 ($1711) but higher than that reported in other analyses that examined medication costs. 7,9,13
Our findings indicating HCRU and costs are elevated immediately preceding initial diagnosis of AD align with prior research. 22 –24 For example, Albert et al 22 observed an increase in use of ambulatory and outpatient care among Medicare beneficiaries in the 2 years preceding diagnosis of AD and estimated excess cost of 26% for female and 85% for male patients with AD compared to non-AD counterparts. The percentage difference in unadjusted preindex costs in the current study is approximately 44%, falling within these bounds. Additional research by Ramakers et al 23 found general practitioner contact frequency was elevated among individuals with AD. Furthermore, a case–control analysis, based on the data from the German health insurance system, demonstrated increased utilization of ambulatory care services in the year prior to diagnosis of AD, with the increase in utilization compared to non-AD controls persisting the year after diagnosis. 24 Collectively, these data and ours suggest the impact of AD is substantial prior to, and early after, diagnosis. Increased utilization and costs in the period prior to diagnosis of AD may be related to a variety of factors including early manifestations of cognitive impairment, increased utilization related to assessment for a diagnosis of AD, and medical complications present prior to the diagnosis of AD.
Common postdiagnosis comorbidities that were associated with AD in this study included Parkinson’s disease, epilepsy, mood disorders, and cerebrovascular disease. The prevalence ratio of Parkinson’s disease was highest, which is consistent with the epidemiology of concurrent dementia and Parkinson’s disease. 25 In fact, Aggarwal et al 26 showed that individuals with parkinsonian-like symptoms, such as gait or bradykinesia, were 2 to 3 times more likely to develop AD subsequently compared to individuals without such symptoms. Following Parkinson’s disease, epilepsy and mood disorders had the next highest prevalence ratios among the observed comorbidities. Seizures are common occurrences in patients with AD, and temporal lobe epilepsy has been suggested to be comparable with AD in terms of neuropathology. 27 Neuropsychiatric symptoms, including depression, may be risk factors for incident AD or even represent preclinical signs of AD onset. 28 While the findings of the current analysis are aligned with previous research examining comorbidity burden among individuals with AD, greater comorbidity burden in the current study may also be related to increased medical utilization among individuals with AD and greater resulting opportunities for surveillance, diagnosis, and/or coding of medical conditions by health care providers.
We observed a greater level of postdiagnosis comorbidity burden among members newly diagnosed with AD as compared to matched controls. However, findings related to preindex comorbidity burden differed depending on whether the score was based on the ICD-9 codes (DCI) or on the prescription claims (RxRisk-V). The basis for this is not entirely clear, and the performance of medical versus prescription claim-based comorbidity measures among patients with AD has not been extensively researched. A recent analysis has found that a prescription claims-based comorbidity measure (the Chronic Disease Score, a progenitor of the RxRisk-V) was not as robust as a comorbidity measure that integrates medical records, interviews, and clinical assessment for patients with AD (the Cumulative Illness Rating Scale-Geriatric). 29 Although comorbidity assessments based on the multiple sources of clinical information are ideal, they may not be feasible within the context of a retrospective claim-based analysis. In non-AD populations, prescription claims-based comorbidity measures such as the RxRisk-V have been shown to perform as well, if not better, than ICD-9-based comorbidity in predicting health care costs. 18 The findings of the current study highlight the need for detailed assessment and comparison of the performance of various claim-based comorbidity measures in AD populations.
As opposed to a number of previous studies that have used broad inclusion criteria, including a range of dementia codes or AD medication utilization to identify cases, 3,5,7,9,10 the current study focused on members identified via a diagnosis code specific to AD (331.0). The study did not include other codes for dementia or case identification based on the medication use to populate the AD cohort. While our approach was more focused, it is possible that our method resulted in misclassification of some cases due to false diagnosis, misdiagnosis, or lack of diagnosis. The AD may be uncoded or undercoded for a variety of reasons including stigma, maximization of reimbursement, and other factors. Furthermore, the method of case identification may influence utilization and cost estimates, because identification based on the diagnosis code associated with a medical claim may bias medical cost estimates. 30 On the other hand, case identification based on the medication exposure may bias estimates of drug expenditure.
We attempted to identify individuals newly diagnosed with AD via first observed diagnosis of AD in the member medical claims records and excluded members who displayed a diagnosis of a related dementia or a medication indicated for the treatment of AD during the preindex enrollment. The identification window for the current analysis covered an 18-month period of time, and all patients were required to have a minimum of 12 months preindex enrollment. Although research indicates that claim data, self-report medication use, and restricted ascertainment period may result in underrecognition of AD, 30 –32 we used a broad range of AD and related dementia codes as well as claim-based medication exposure to identify and exclude prevalent cases from the AD cohort. With regard to the control cohort, members were not eligible for inclusion in the control cohort if they demonstrated a diagnosis of AD, a related dementia, or an AD medication at any time in their available medical and pharmacy claims records (minimum of 36 months of enrollment).
The findings of this study should be interpreted taking into account limitations related to the use of administrative claims data in health care research. Cohort assignment in our study predicated on the validity of diagnostic codes submitted with medical claims. Reporting of AD and other dementias in medical claims data may be impacted by a variety of factors including underrecognition, undercoding, or limitations in the medical coding and reimbursement systems. 33,34 Patients with cognitive impairment related to a variety of medical conditions, medication exposure, and other etiologies may be “channeled” into receiving a diagnostic code for AD as a result of limited coding options for providers and reimbursement disincentives associated with the use of other dementia codes. While limitations in the coding schema may result in channeling of individuals toward a diagnosis of AD, previous research indicates that, overall, AD is substantially underreported in medical claims records. 33 Another limitation of administrative claims data is that the ICD-9 coding schema does not indicate severity of AD. Previous research based on the administrative claims data has used complications commonly associated with advanced AD as a proxy for late-stage AD 13 ; however, such proxies have not been validated and we did not include a proxy for AD severity in the current analysis. Individuals with AD may not present for medical care until their dementia has progressed such that functional and/or cognitive impairment are significant; furthermore, research indicates that the pathological process of AD begins many years prior to clinical diagnosis, and the preclinical phase of AD is associated with neuropsychological findings and preclinical cognitive impairment. 35 –38 Although we tailored our methodology to identify newly diagnosed patients with AD, there is a possibility these patients may represent individuals across a spectrum of dementia severity and do not necessarily reflect new-onset AD.
A strength of the current study is the requirement that members were continuously enrolled for 36 months, allowing for resolution of longitudinal outcomes among parallel cohorts over a period of concurrent enrollment. As opposed to previous research, 8 –10,13 we were able to directly observe and calculate actual medical and pharmacy costs over a period of extended follow-up and did not impute or annualize cost data. Annualized cost estimates are sensitive to the duration of follow-up and may result in biased cost estimates. 33 Cost estimates from the current study do not reflect the health care costs associated with AD in later years, which may be higher or lower. Further research is needed to determine the long-term trends in costs for individuals after diagnosis of AD.
Based on the findings of this study, future research should identify factors driving the observed trend in year-over-year medical costs for individuals newly diagnosed with AD compared to control group. Increased medical costs during the prediagnosis period may be related to a variety of factors including diagnostic work-up, prediagnosis morbidity, prediagnosis treatment for non-AD conditions, or other factors. Similarly, postdiagnostic trends in medical costs highlight the need to identify targeted strategies that aim to ensure efficient use of health care resources for patients with AD.
Footnotes
Acknowledgment
We would like to thank Mary Costantino, PhD, a full-time employee of Complete Health Analytics, Inc, who was paid for the development of this manuscript, for reviewing and editing this manuscript.
Authors’ Note
The research concept was approved by the Joint Research Governance Committee of the Humana-Pfizer Research Collaboration, comprised of Humana Inc and Pfizer Inc employees, and plans to publish results were made known prior to commencing the study.
Declaration of Conflicting Interests
The authors declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: Suehs, Davis, and Patel are employees of Competitive Health Analytics, a wholly owned subsidiary of Humana Inc, and Suehs and Patel are stockholders of Humana, Inc. Van Amerongen is an employee of Humana, Inc. Shah, Alvir, and Faison are employees and stockholders of Pfizer Inc. At the time that this research was conducted, Joshi was an employee of Pfizer Inc.
Funding
The authors disclosed receipt of the following financial support for the research, authorship and/or publication of this article: This research was conceived, funded, and carried out collaboratively by Humana Inc, Pfizer Inc, and Competitive Health Analytics, Inc. Competitive Health Analytics, Inc received funding support from Pfizer in connection with conducting this study and for the development of this manuscript.