The growing understanding of the use of biomarkers in Alzheimer's disease (AD) may enable physicians to make more accurate and timely diagnoses. Florbetaben, a beta-amyloid tracer used with positron emission tomography (PET), is one of these diagnostic biomarkers. This analysis was undertaken to explore the potential value of florbetaben PET in the diagnosis of AD among patients with suspected dementia and to identify key data that are needed to further substantiate its value. A discrete event simulation was developed to conduct exploratory analyses from both US payer and societal perspectives. The model simulates the lifetime course of disease progression for individuals, evaluating the impact of their patient management from initial diagnostic work-up to final diagnosis. Model inputs were obtained from specific analyses of a large longitudinal dataset from the New England Veterans Healthcare System and supplemented with data from public data sources and assumptions. The analyses indicate that florbetaben PET has the potential to improve patient outcomes and reduce costs under certain scenarios. Key data on the use of florbetaben PET, such as its influence on time to confirmation of final diagnosis, treatment uptake, and treatment persistency, are unavailable and would be required to confirm its value.
Alzheimer’s disease (AD) is a fatal and progressive neurodegenerative disorder that affects millions of people worldwide [
There is still no cure for AD. Cholinesterase inhibitors and memantine are the only major pharmacological treatments currently available to slow the symptoms associated with disease progression. Clinical studies have demonstrated modest benefits of these treatments in improving the symptoms related to AD [
The current diagnosis of AD is mainly based on clinical grounds according to the guidelines developed jointly by the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA). Although sensitive for AD, the NINCDS-ADRDA guidelines have poor specificity [
Florbetaben, a beta-amyloid tracer, is one of the novel diagnostic tools that can be used to detect neuropathological changes related to AD in vivo. It binds to beta-amyloid plaques and can be detected using a PET scan. The predictive values of florbetaben PET are currently under a phase III assessment, but preliminary phase II data have shown promising results in discriminating AD from other dementias and healthy controls [
The model explores the potential clinical and economic consequences of using florbetaben PET in the usual diagnostic process for the diagnosis of AD from both the US payer and societal perspectives. Usual diagnostic care refers to a period of diagnostic work-up during which specific diagnostic tests are performed over a series of medical visits to obtain the needed information for confirmation of a specific type of dementia diagnosis. Different combinations of diagnostic tests and assessments, along with other tests of pathophysiological (e.g., amyloid tracers and total tau) and/or topographical makers (e.g., FDG and MTA), received by patients during the diagnostic work-up period, could have different predictive values in terms of differentiating AD from other forms of dementia; there are numerous possible combinations of the tests and assessments that can be received by patients during the work-up period. Thus, for the purpose of simplicity and availability of the existing data to inform the predictive values of each specific combination of tests received by patients, the possible combinations of the diagnostic tests in this model were categorized into the following algorithms: clinical guidelines alone (e.g., NINCDS-ADRDA) and clinical guidelines plus one of the following tests: (1) MRI of MTA, (2) computerized tomography (CT) of MTA, (3) FDG-PET, (4) single photon emission computer tomography (SPECT), (5) CSF of beta-amyloid, total tau, phospho-tau, or beta-amyloid plus total tau, and (6) PET of florbetaben beta-amyloid tracers. Patients in the nonflorbetaben group (or called usual diagnostic care group hereafter) can be proportionally assigned to any of these diagnostic algorithms, except for the clinical guidelines plus florbetaben PET, to obtain an aggregate measure of sensitivity and specificity associated with the diagnostic algorithms assigned.
This model was implemented as an individual patient simulation using discrete event simulation (DES) [
Figure
Schematic representation of the model concept.
Each patient in the model undergoes a period of diagnostic work-up following their initial doctor visit. At the end of the diagnostic work-up period, all the predementia patients are assumed to be correctly confirmed with a predementia diagnosis, and those assigned to the florbetaben group are given a PET scan, with dementia treatment, such as cholinesterase inhibitors and memantine, initiated for those with a positive AD result. For those assigned to the usual diagnostic care group, all or some proportion of the patients may receive dementia treatment without screening after their predementia diagnosis is confirmed.
For the dementia patients, all patients are confirmed with a specific type of dementia at the end of diagnostic work-up, but the correctness of the diagnosis is dependent on the predictive values of the diagnostic algorithm assigned to the patient. Dementia treatment is initiated to all dementia patients in the florbetaben group with a positive AD result, but not to those with a negative result at the time of diagnosis confirmation. For those in the usual diagnostic care group, some dementia patients also have a chance to receive dementia treatment at the time of confirmation, depending on the result of diagnosis. Patients with a positive AD result but not treated at the time of confirmation could receive dementia treatment at a later time.
Dementia treatment may delay progression to the dementia phase for patients with predementia. When a patient develops dementia, treatment may slow progression to a more severe stage of the disease, as well as need for institutional care, but the effect of treatment could be negatively impacted by misdiagnosis and nonpersistence with treatment. Disease progression in this model was modelled through the interrelated changes in 3 domains over time: cognition, using MMSE scale; behaviour, using the Neuropsychiatric Inventory (NPI) scale; function, using both the activities of daily living (ADL) and instrumental activities of daily living (IADL) scales. In this model, it is assumed that only those with prodromal AD, AD, or mixed AD as the underlying cause of dementia would benefit from the treatment with cholinesterase inhibitors or memantine. Treatment initiated in patients with non-AD would only have an impact on treatment costs.
Disease progression continues during the diagnostic work-up period. Shortening the time required to correctly confirm a diagnosis would allow appropriate treatments to be initiated at an earlier stage of the disease and thus could result in greater health benefits at lower costs. Use of florbetaben PET in this model could directly influence 4 major areas: (1) time required to confirm a diagnosis, (2) accuracy of diagnosis, (3) proportion of patients receiving appropriate dementia treatments at confirmation, and (4) persistence with treatment. Each of these impacts is associated with specific clinical and economic consequences. Finally, this model allows those who are misdiagnosed to be correctly rediagnosed at a later time and receive treatment. Death can occur at any point in time, and is dependent on patient age, gender, underlying cause of dementia, and stage of the disease (i.e., predementia and dementia phases). A simplified model flow diagram showing how patients are simulated is displayed and explained in Appendix
The primary data source used to populate this model was based on the administrative databases from New England Veterans Healthcare System (VISN 1) from January 1, 2002 through December 31, 2009 (fiscal years 2002–2009). Specific analyses from the VA VISN 1 data were performed to obtain the majority of the model inputs. Detailed information on the VA VISN 1 data can be seen in Appendix
A reference-case analysis was performed based on 1,000 simulated patients per group per run for a total of 10 replications. The model time horizon for the reference-case analysis was lifetime, which is commonly used for the assessment in this therapeutic area. Costs and benefits were discounted at 3% per annum [
Data used to create the model population in the simulation were mainly obtained from the VA VISN 1 data, supplemented with data from literature. Among the patients with a confirmed diagnosis, 68% of them were confirmed with a dementia diagnosis and 32% with a predementia diagnosis. For those with a dementia diagnosis, 67% had a confirmed diagnosis of AD or mixed AD, 28% had a VaD, and the remaining 5% had other dementia diagnoses, such as LBD, FTD, and mixed non-AD. For those diagnoses with predementia, 65% were assumed to have prodromal AD, which was estimated based on a chart review of a subset of these patients. The mean age at initial diagnosis was about 78 years for patients with predementia and 82 years for patients with dementia. Data used to assign gender to the model populations (about 30% male) were obtained from literature [
Baseline MMSE scores were obtained from a chart review of a subset (
The proportions of patients in the usual diagnostic care group undergoing a specific diagnostic algorithm were obtained from the VA VISN 1 data (Table
Distribution of diagnostic algorithms and corresponding accuracy by severity (based on MMSE).
Diagnostic algorithm | % | Mild | Moderate and severe | |||
---|---|---|---|---|---|---|
Usual care | Florbetaben | Sensitivity | Specificity | Sensitivity | Specificity | |
Clinical guidelines only | 73% | 0% | 87% | 59% | 77% | 73% |
Clinical guidelines with | ||||||
MRI of MTA | 8% | 0% | 82% | 66% | 85% | 80% |
CT of MTA | 19% | 0% | 80% | 87% | 80% | 87% |
FDG-PET | 0% | 0% | 91% | 75% | 91% | 86% |
SPECT | 0% | 0% | 79% | 81% | 68% | 86% |
CSF A |
0% | 0% | 72% | 75% | 74% | 79% |
CSF A |
0% | 0% | 86% | 64% | 84% | 72% |
CSF Ttau | 0% | 0% | 77% | 73% | 82% | 71% |
CSF Ptau | 0% | 0% | 77% | 73% | 82% | 78% |
Florbetaben PET | 0% | 100% | 90% | 90% | 90% | 90% |
MRI: magnetic resonance imaging; MTA: medial temporal lobe atrophy; CT: computer tomography; PET: positron emission tomography; FDG: fluorodeoxyglucose; SPECT: single photon emission computer tomography; CSF: cerebral spinal fluid.
The amount of time taken to confirm a specific type of dementia diagnosis from the initial office visit under usual diagnostic care was predicted using parametric equations derived from the VA VISN 1 data. Table
Equations for prediction of time to events and disease progression.
Equation | Coefficient and predictor | SD/shape | Distribution |
---|---|---|---|
Time to confirmation of a diagnosis | |||
AD or mixed AD | 4.571 + 0.327 Male + 0.252 MixedAD − 0.353 CKD | 0.965 | Lognormal |
VaD | 4.529 + 0.158 Diabetes + 0.203 Hypertension + 0.385 Stroke | 1.005 | Lognormal |
Other non-AD dementia | 6.558 − 0.029 Age + 1.554 Stroke + 0.654 LBD + 0.400 FTD | 0.896 | Lognormal |
Predementia | 3.981 + 0.009 Age − 0.243 CKD − 0.179 CVD | 0.994 | Lognormal |
Time to treatment initiation if not started at diagnosis | |||
Dementia | 7.149 − 0.022 Age + 1.056 VaD + 2.091 Other non-AD + 0.004 (time, in days, to confirmed diagnosis) | 1.517 | Lognormal |
Predementia | 18.781 − 0.150 Age | 3.996 | Lognormal |
Time to treatment discontinuation | |||
Dementia | 7.487 − 0.0008 (time to confirmed diagnosis) | 0.922 | Weibull |
Predementia | 7.122 + 0.443 (conversion to dementia) | 1.131 | Weibull |
Time to conversion to dementia | Scale = 0.0212 | 0.952 | Weibull |
Rate of change in MMSEa | 5.4663 − 0.4200 PM1 − 0.0042 PM2 + 0.1415 PM3 − 0.079 PrevRate + 0.07474 Age + |
N/A | N/A |
Rate of change in NPIb | (5.74 − 0.64 Treatment + 0.03 Weeks − 0.59 NPIbase − 0.59 NPI Weeks + 0.24 NPIrecent − 1.74 White − 3.82 Black + 2.34 PsyMed + 0.12 MMSEbase − 0.22 MMSErecent + |
N/A | N/A |
Rate of change in ADL | 1.35 − 0.81 Treatment + 0.06 Weeks − 0.79 ADLbase + 0.71 IADLprevious + 0.12 MMSEbase + 0.09 Age + 0.81 PsyMed − 3.05 Black − 0.49 MMSErecent + |
N/A | N/A |
Rate of change in IADL | 1.27 + 0.63 Treatment + 0.17 Weeks − 0.06 Treatment * Weeks − 0.84 IADLbase − 0.002 IADLbase * Weeks + 0.84 IADLprevious − 0.67 Male + 0.20 MMSEbase − 0.28 MMSErecent − 0.16 ADLbase + 0.18 ADLrecent + |
N/A | N/A |
Time to institutional care | |||
Dementia | 9.883 − 0.02 Age + 0.295 VaD + 1.154 Other non-AD − 0.001 Time to confirmed diagnosis + 1.079 Dementia treatment | 0.933 | Weibull |
Predementia | 11.469 − 0.028 Age | 1.373 | Weibull |
Time to death | |||
Male | Scale = −9.697 | 0.087 | Gompertz |
Female | Scale = −10.787 | 0.097 | Gompertz |
Patient utility | 0.408 + 0.010 MMSE − 0.004 NPI − 0.159 Institutionalized + 0.051 Living with Caregiver | N/A | N/A |
Caregiver utility | 0.90 − 0.003 AgeCG + 0.03 MaleCG + 0.001 Male − 0.001 NPI − 0.001 ADL − 0.0004 IADL − 0.01 PsyMed | N/A | N/A |
SD: standard deviation; AD: Alzheimer’s disease; VaD: vascular dementia; LBD: Lewy body dementia; LTD: frontotemporal dementia; CKD: chronic kidney disease; CVD: cerebrovascular disease; MMSE: mini-mental state examination; NPI: neuropsychiatric inventory; ADL: activities of daily living; IADL: instrumental activities of daily living, CG: caregiver.
aPM represents patients’ previous MMSE measurement, partitioned over the scale of MMSE. PrevRate is the patients’ last known rate of decline. Age represents patients’ age at baseline.
bTreatment is dementia medication, Weeks represents weeks of followup in the simulation, NPIbase is the patient’s baseline NPI, and NPIrecent is the patient’s last NPI. White and Black are dummy variables for race, PsyMed is a dummy variable for patients on psychiatric medications at baseline, MMSEbase represents the patient’s MMSE at baseline, and MMSErecent represents the patient’s current MMSE.
Dementia medications can be initiated at either the time of diagnosis confirmation or a later time for those patients in the usual diagnostic care group (Table
Model parameters for treatments.
Parameter | Dementia | Predementia | Data source | ||
---|---|---|---|---|---|
Usual care | Florbetaben | Usual care | Florbetaben | ||
% of patients receiving dementia medication at diagnosis | N/A | N/A | 28% | N/A | VA VISN 1 and user specification |
If Dx = AD+ | 77% | 100% | N/A | 100% | |
If Dx = non-AD | 67% | 0% | N/A | 0% | |
Distribution of dementia medication |
|
|
VA VISN 1 | ||
Donepezil | 63% | 66% | 76% | ||
Galantamine | 25% | 9% | 6% | ||
Rivastigmine | 5% | 4% | 1% | ||
Memantine | 7% | 21% | 17% | ||
Maximum dementia treatment duration allowed, years | Life time | 5 | User specification | ||
Stopping dementia medication if MMSE score is below 10 | Yes | N/A | User specification |
AD: Alzheimer’s disease; Dx: dementia diagnosis; MMSE: mini-mental state examination; N/A: not applicable.
Patients on any dementia treatment may discontinue over time. Table
In this model, dementia treatment can also be forced to stop under some conditions specified by the users. In the reference-case analysis, patients with dementia were allowed to receive dementia treatment for lifetime as long as their MMSE scores were greater than 10; patients with predementia were assumed to receive treatment for no longer than five years if the patient did not covert to the dementia phase.
Patients with predementia may convert to the dementia phase at a later time. Data on time to conversion, also from the VA VISN 1 data, were fitted to a Weibull function (Table
Disease progression for patients with AD or mixed AD was modelled based on the interrelated changes in MMSE, NPI, ADL, and IADL over time. Data used to simulate the disease progression were based on the predictive equations (Table
Two Weibull equations derived from the VA VISN 1 data were used to predict the time to institutional care for dementia and predementia patients (Table
Time to death was predicted using two gender-specific Gompertz functions derived from the US Life Table based on patients age 55 years and above (Table
Cost inputs and their corresponding sources are shown in Table
Cost inputs.
Cost item | Value | Unit | Data source |
---|---|---|---|
Diagnostic work-up | |||
AD+ | $5,120 | Per year | |
VaD | $5,885 | Per year | VA VISN 1 and [ |
Other non-AD | $6,638 | Per year | |
Predementia | $6,187 | Per year | |
Imaging and biomarker tests | |||
MRI + MTA | $437 | Per test | |
CT + MTA | $300 | Per test | |
FDG-PET | $1,042 | Per test | [ |
SPECT | $596 | Per test | |
CSF | $304 | Per test | |
Florbetaben PET | $2,300 | Per test | |
Dementia medication | |||
Donepezil | $7.79 | Per day | |
Galantamine | $6.36 | Per day | [ |
Rivastigmine | $6.11 | Per day | |
Memantine | $7.89 | Per day | |
Medical care for predementia | $5,548 | Per year | [ |
Medical care for AD+ | |||
Mild | $8,315 | Per year | |
Mildly moderate | $12,806 | Per year | |
Moderate | $12,806 | Per year | [ |
Moderately severe | $18,526 | Per year | |
Severe | $23,227 | Per year | |
Nonmedical care for AD+ | |||
Mild | $154 | Per year | |
Mildly moderate | $3,692 | Per year | |
Moderate | $12,166 | Per year | [ |
Moderately severe | $14,209 | Per year | |
Severe | $23,355 | Per year | |
% of additional cost of care for non-AD relative to AD | |||
VaD | 84% | [ | |
Other non-AD | 37% | ||
Institutional care | $373 | Per day | [ |
Caregiver time | $7.25 | Per hour | [ |
Caregiver burden for predementia | 2.10 | Hours per day | [ |
Caregiver burden for dementia | |||
Mild | 2.10 | Hours per day | |
Mildly moderate | 3.58 | Hours per day | |
Moderate | 3.58 | Hours per day | [ |
Moderately severe | 3.76 | Hours per day | |
Severe | 5.10 | Hours per day |
AD: Alzheimer’s disease; VaD: vascular dementia; MRI: magnetic resonance imaging; MTA: medial temporal lobe atrophy; CT: computer tomography; PET: positron emission tomography; FDG: fluorodeoxyglucose; SPECT: single photon emission computer tomography; CSF: cerebral spinal fluid.
The model estimates utilities for both patients and their caregivers. Health utilities for patients with AD were estimated based on a published regression equation shown in Table
Of the 1,000 simulated patients, 32% had predementia and 68% dementia, replicating the underlying input data. Of those patients with predementia, 65% had prodromal AD as the underlying cause. Patients with predementia had a mean age of 78 years, and mean scores of 27.5 on the MMSE, 2.5 on the NPI, and 10.1 on both ADL and IADL scales. On the other hand, for those with dementia, 67% had AD or mixed AD as the underlying cause, 28% had VaD, and 5% had LBD, FTD, or other mixed no-AD dementia with a mean age of 82 years and mean scores of 21.9 on the MMSE, 16.3 on the NPI, 29.7 on ADL, and 29.1 on IADL scales.
The reference-case analysis shows that the average time to confirmation of predementia diagnosis was 4.64 months under usual diagnostic care, which was slightly lower than the time indicated by the VA VISN 1 data due to death and early conversion to dementia during the diagnostic work-up period, and 2.49 months with use of florbetaben PET.
Due to death and early conversion, only about 92% (
Reference-case results.
Outcome (per patient) | Predementia cohort ( |
Dementia cohort ( |
||||
---|---|---|---|---|---|---|
Usual care | Florbetaben | Net | Usual care | Florbetaben | Net | |
Survival, years | 6.84 | 6.94 | 0.10 | 4.57 | 4.57 | 0.00 |
Time to confirmed diagnosis, months | 4.64 | 2.49 | −2.15 | 5.08 | 2.66 | −2.42 |
Time in predementia, years | 3.22 | 3.56 | 0.34 | N/A | N/A | N/A |
Time to institutional care, years | 5.48 | 5.72 | 0.24 | 3.17 | 3.29 | 0.12 |
Time in severity, years | ||||||
Mild | 3.53 | 3.82 | 0.29 | 0.56 | 0.60 | 0.04 |
Mildly moderate | 0.46 | 0.44 | −0.02 | 0.77 | 0.78 | 0.01 |
Moderate | 0.48 | 0.45 | −0.03 | 0.71 | 0.71 | 0.00 |
Moderately severe | 0.42 | 0.40 | −0.01 | 0.57 | 0.56 | −0.01 |
Severe | 1.96 | 1.83 | −0.13 | 1.96 | 1.92 | −0.05 |
Caregiver time, years | 0.92 | 0.91 | −0.01 | 0.77 | 0.76 | −0.01 |
Costs (discounted) | ||||||
Total direct medical care | $301,599 | $289,225 | −$12,374 | $314,156 | $303,070 | −$11,086 |
Caregiver time | $47,914 | $47,271 | −$643 | $42,311 | $42,008 | −$303 |
Total | $349,514 | $336,496 | −$13,018 | $356,466 | $345,077 | −$11,389 |
QALYs (discounted) | ||||||
Patients | 3.53 | 3.68 | 0.15 | 1.75 | 1.78 | 0.03 |
Caregivers | 4.29 | 4.41 | 0.12 | 2.59 | 2.60 | 0.01 |
Total | 7.82 | 8.09 | 0.27 | 4.34 | 4.37 | 0.03 |
ICERs (discounted) | ||||||
Patients | Dominant | Dominant | ||||
Caregivers | Dominant | Dominant | ||||
Total | Dominant | Dominant |
QALYs: quality-adjusted life years; ICERs: incremental cost-effectiveness ratios.
Note: inconsistency may occur due to rounding.
The reference-case results for the dementia cohort are also shown in Table
Figure
Results of univariate sensitivity analyses based on predementia cohort.
Impact of percent reduction in time to diagnosis on net cost and quality-adjusted life-years.
The top 15 parameters influencing the net cost based on the dementia patients were similar to those observed in the predementia cohort, but the levels of significance for some model parameters were somewhat different (Figure
Results of univariate sensitivity analyses based on dementia cohort.
Detailed information on how the probabilistic sensitivity analyses were performed can be viewed in Appendix
Incremental cost effectiveness plane for predementia cohort.
Incremental cost effectiveness plane for dementia cohort.
Unlike the results based on the predementia patients, almost all the ICERs based on the dementia patients spread in the fourth quadrant of the incremental cost effectiveness plane (Figure
To our knowledge, this is the first model to assess the cost effectiveness of a biomarker in the early diagnosis of AD with DES to simulate the course of disease progression from predementia to dementia phase and its clinical management from initial diagnostic work-up to treatment initiation. The greater detail underlying the DES framework allows exploration of the potential value of biomarker use in the early diagnosis of AD, identification of major data gaps, and assessment of the uncertainty in outcomes associated with those gaps. As the model closely resembles the course of disease and its management at individual patient level, it inevitably requires richness of data to support the simulation. To deal with the data issue, we undertook a comprehensive analysis of longitudinal data from VA VISN 1 to characterize usual care pertaining to the diagnosis and treatment of AD and other forms of dementia in the US and to provide direct empirical estimates of various aspects of usual care of the diseases. With the use of advanced modelling technique, along with the support of comprehensive data from the VA VISN 1, this model provides a better understanding of how patients would be affected over time if a diagnostic biomarker like florbetaben PET tracer is used and which model parameters would have major influence on the model outcomes for specific patient groups. However, it should be noted that the results from the reference-case analysis are based on many important assumptions. Solid evidence to support or refute these assumptions is necessary before more conclusive estimates can be produced.
The reference-case scenario indicates that use of florbetaben PET in the diagnosis of AD results in both health benefits and cost savings. The probabilistic sensitivity analyses suggest that such model outcomes are positive in the great majority of cases when florbetaben PET is used in patients with dementia but are subject to a greater uncertainty when used in patients identified with predementia. The greater uncertainty in the latter case is mainly due to lack of data on several critical model parameters. The deterministic sensitivity analyses indicate that improved accuracy of diagnosis alone would not be adequate to yield sufficient clinical benefits and cost offsets to justify the use of florbetaben PET in the diagnosis of AD. Other clinical benefits, especially if it would shorten the time taken to confirm a dementia or predementia diagnosis, are needed to further support its cost effectiveness.
The reasons the reduction in time to confirmed diagnosis is so important to the cost effectiveness of florbetaben PET are not only that early diagnosis could allow appropriate dementia treatment to be initiated at an earlier stage of the disease, but also that reduction in time to diagnosis confirmation has a direct beneficial impact on time to institutional care. The risk of needing institutional care would be reduced by about 12% for every 100-day reduction in time to diagnosis confirmation, as indicated by the analyses of the VA VISN 1 data. The causal relationship between them is still unclear. It is possible that early confirmation of diagnosis would allow patients and their family members to plan ahead and make needed adjustments to keep patients living independently as long as possible before they are sent to long-term-care facilities. Given that institutional care is costly, any minor delay to institutional care would have a meaningful impact on offsetting the cost of the scan. Data from a survey study, conducted alongside the florbetaben PET phase IIA trial [
Three additional important findings from the present analyses are worth mentioning as they may have important implications for future economic assessment of the florbetaben PET tracer in the early diagnosis of AD. First, a large discrepancy on time to confirmed diagnosis was found from the VA VISN 1 data when different approaches, that is, analysis of administrative data versus chart review of a subset of the study cohort, were used to quantify this duration. The estimated duration based on the analyses of the VA VISN 1 data may better capture the time taken to confirm a dementia diagnosis from the initial office visit because a diagnosis code would normally be recorded to represent the main complaint for a particular office visit. Yet, the estimated duration of 1.5 years based on the review of medical records should be a good proxy for the time to confirmation of diagnosis from the early signs and symptoms of AD as these could be recorded in the medical charts during the office visits for other medical problems. The implication of this discrepancy seems to pose a great opportunity for florbetaben PET tracers to identify patients with prodromal AD even at a much earlier stage if clinicians know when to use them. This would have a substantial, favourable impact on the cost effectiveness of florbetaben PET tracers.
Second, our deterministic sensitivity analyses show that younger patients would have a greater gain in net cost savings and QALYs from the use of florbetaben PET. This is due to a longer life expectancy in this population. In order to treat patients with prodromal AD or AD dementia at a younger age, screening general populations at younger ages seems to be a reasonable strategy. Although screening the general population for AD is not the focus of this assessment, the significant gain in economic and clinical benefits in the younger group from our analyses does suggest a promising possibility to support such an application for florbetaben PET tracers. This could also have valuable benefits to some of the patients who choose to know their disease propensity as early as possible despite the absence of effective treatments during the preclinical or predementia phase as it allows them to plan ahead with their life for personal and financial reasons [
Third, as there is still no convincing evidence to support that treatment with cholinesterase inhibitors would yield any survival benefit, this model assumes that survival is independent of treatment effect, consistent with the assumption made in other published models [
The present model has several major limitations. First, because the current analyses are mainly based on the data from the New England VA Healthcare System, the findings from this analysis may not be generalizable to patients in other regions of the VA Healthcare System, as well as in other healthcare systems outside of the VA system. Second, an external validation of this model to examine how well the model can predict the results observed in other studies has not yet been conducted due to lack of an appropriate external data source. However, the results of key model components, including disease progression during the dementia phase and time to clinical events shown in Table
This economic model provides a comprehensive framework to explore the potential clinical and economic value of florbetaben PET in the early diagnosis of AD among patients who present to their physicians’ office for the first time due to cognitive complaints, to identify key value drivers as well as potential data gaps. Our exploratory analyses suggest that florbetaben PET has the potential to be a valuable tool in the diagnosis of AD as it would improve the health benefits of patients (with dementia as well as predementia) and their caregivers at a lower cost under certain scenarios. While the findings from the analyses to a large extent are supported by the data from a large longitudinal database and published literature, they rest also on many key assumptions and are subject to great uncertainty. Data on how the technology would impact clinical decision making and outcomes, such as time to confirmation of diagnosis, treatment uptake, and treatment persistency, will be needed to further substantiate its value.
A simplified flow diagram showing how patients are simulated in the model is displayed above. At the beginning of the simulation, the model creates 1,000 patients with different types of dementia, based on the prevalence of dementia for each type specified by the user, and assigns patient and disease characteristics conditional on their underlying cause. These characteristics include age, gender, race, baseline scores for Mini-Mental State Examination (MMSE), Neuropsychiatric Inventory (NPI), activities of daily living (ADL), instrumental activities of daily living (IADL), location of care (either home or institutional care), use of antipsychotics, comorbidities (i.e., chronic kidney disease, cardiovascular disease, diabetes, and hypertension), and caregiver’s gender and age. These characteristics are used to predict the rate of disease progression and other model outcomes, such as time to death, time to institutional care, time to confirmation of diagnosis, costs of care, and health utilities. After the assignments of baseline patient characteristics and event times, each patient is cloned; one clone is assigned to the usual diagnostic care group and the other to the florbetaben group. The cloning step used in the simulation resembles a perfect randomization where both groups are comprised of exactly the same patients. For those assigned to the florbetaben group, their time to confirmation of diagnosis is updated based on a percent reduction in time to diagnosis specified by the user. Then, all patients are sent to the “search next event” module, where the next event for each patient is identified based on the event with the shortest time to occur. There are a total of 10 events conceptualized in this model, as shown in diagram above. After identifying the next event for a patient, the model then fast-forwards its clock to the next event time. Before the event is processed for its related consequences, such as updates of patient and treatment statuses and time to next event, all time-dependent outcomes—including survival, quality-adjusted life-years (QALYs), costs of care, caregiver time, time alive at each stage of the disease, and time spent in institutional care—are tallied and accumulated. After processing the consequences of the event, the patient proceeds to the “search next event” module again, and the same process is repeated until the patient dies or the model time ends see Figure
Model flow.
The primary data source used to populate this model was based on the administrative databases (clinical, laboratory, and pharmacy databases) from New England Veterans Healthcare System (VISN 1) from January 1, 2002 through December 31, 2009 (fiscal years 2002–2009). The New England region consists of eight Veterans Administration (VA) Medical Centers and affiliated clinics providing inpatient and outpatient medical care. The VISN 1 pharmacy files were obtained from Information Resource Management, Boston, MA. ICD-9-CM diagnoses and laboratory data were captured by accessing the VA National Patient Care and Decision Support Systems administrative databases (Patient Treatment File and Outpatient Care File) located at the Austin Automation Center, Austin, TX. All database analyses were conducted at the Massachusetts Veterans Epidemiology Research and Information Center (MAVERIC), VA Boston Healthcare System, Boston, MA. The study was approved by the Institutional Review Board of the VA Boston Healthcare System.
Specific analyses from VISN 1 were performed to obtain the majority of the model inputs, including the following: (1) prevalence of dementia for each type and baseline patients and disease characteristics by dementia diagnosis, (2) time to confirmation of diagnosis from the initial visit, (3) resource use during the diagnostic work-up period, (4) proportion of the patients treated with cholinesterase inhibitors or memantine at time of diagnosis confirmation by dementia diagnosis, (5) time to treatment initiation if not treated at time of confirmation, (6) time to treatment discontinuation if treated, (7) time to conversion to the dementia phase for patients with predementia, and (8) time to institutional care. The study cohort for these analyses was based on 2,783 patients who were confirmed with a dementia (
Among the patients with a confirmed diagnosis (
The costs of diagnostic work-up, including a series of office visits to primary care physicians and specialists, lab tests, and outpatient clinic visits, were estimated based on the time taken to confirm a diagnosis from the initial office visit. Data for healthcare resource use during the diagnostic work-up period were obtained for each dementia diagnosis from the VA VISN 1 data and were translated into costs by applying the unit costs from the Centers for Medicare & Medicaid Services (CMS) [
The costs of brain imaging and biomarker analyses, a one-time cost based on the distribution of diagnostic algorithms assigned, were estimated using the cost data from CMS hospital outpatient fee schedule [
Unit costs for dementia medications were based on the average wholesale price reported in the Red Book [
Costs of care were separated into costs of medical and nonmedical care. For patients with AD, these costs were obtained from a longitudinal study which followed 172 patients with probable AD for 4 years to examine the effects of patient dependence, measured by the Dependence Scale, on the following: (1) medical care costs including hospitalizations, outpatient treatments and procedures, and assistive devices; (2) nonmedical care costs, including overnight respite care, adult day care, and home healthcare; (3) informal caregiving time, including time used for ADL and for supervision [
The cost of managing predementia in the US setting was not identified from the literature search. Although some studies were found for other countries [
Model outcomes for the base-case analysis included percent of patients misdiagnosed, time to diagnosis confirmation, number of predementia patients progressing to dementia, life-years, time alive at each severity stage, percent of patients needing institutional care, time to institutional care, caregiver time, costs, and QALYs for patients and caregivers. Deterministic sensitivity analyses, including one-way, subgroup, and scenario analyses, were performed to assess how the model outcomes vary in relation to changes in model parameters. Finally, in order to account for uncertainties from multiple key parameters, probabilistic sensitivity analyses were performed by simultaneously varying multiple parameters under the following assumptions. First, for the inputs which have no or little prior data to support, uniform distributions were used as it is a more conservative assumption. These parameters included the following: (1) sensitivity (80% for the lower bound–96% for the upper bound) and specificity (80–96%) of florbetaben PET, (2) percent reduction (0–100%) in time to diagnosis confirmation by florbetaben PET, (3) percent reduction (0–100%) in risk of conversion to dementia phase by treatment, and (4) percent reduction (0–100%) in treatment discontinuation by florbetaben PET. Second, for model inputs with prior data to support, beta distributions were used for categorical variables and normal distributions within 2 standard errors of the mean as the upper and lower bounds were used for continuous variables. For some of the parameters included in the probabilistic sensitivity analyses, standard errors were available from the parameter source data and thus used to measure parameter uncertainties. Where a standard error was not available for a selected parameter, we used 25% of the mean as an assumed standard error. Parameters included in the probabilistic sensitivity analyses and assumed to be beta distributed were as follows: (1) percents of the predementia and dementia patients treated at time of confirmation under usual diagnostic care, (2) ratios of cost of care for AD versus VaD and non-AD, and (3) patient and caregiver utilities for predementia patients. Parameters assumed to be normally distributed included the following: (1) all predictor coefficients for the parametric equations, including time to diagnosis, time to treatment initiation, time to treatment discontinuation, and time to institutional care; (2) coefficients for the treatment effects on rates of change in MMSE, NPI, ADL, and IADL over time; (3) hazard ratios of death (with log transformation) compared to general population for predementia, AD, and no-AD; (4) predictor coefficients for patient and caregiver utility equations for dementia patients; (5) cost of dementia medications; (6) costs of medical and nonmedical care for predementia and dementia; (7) cost of institutional care; (8) caregiver burden by severity level.
All authors participated in the conception and implementation of the model, data analyses for obtaining model inputs and analyses of the simulation, and writing of the paper. S. Guo, D. Getsios, L. Hernandez, and S. Lanes are employees of United BioSource Corporation (UBC), a consultancy that has also received grants for other unrelated research from various pharmaceutical companies. E. Lawler and K. Cho are employees of the Massachusetts Veteran’s Research and Information Center, which is supported by the Veterans Affairs Cooperative Studies Program. A. Altincatal is an employee of Boston VA Research Institute, Inc. M. Blankenburg is an employee of Bayer Healthcare Pharmaceuticals, located in Berlin, Germany. This work was sponsored by Bayer Healthcare Pharmaceuticals, the manufacturer of florbetaben tracers. The authors thank Jens Kuhlmann, Martin Pessel, and Andreas Kramell for critically reading the paper.