Statistical Issues for Developing Alzheimer Screening Tests J. Wesson Ashford, M.D., Ph.D. Stanford / VA Alzheimer’s Center VAMC, Palo Alto, California & Helena C. Kraemer, Ph.D. Department of Psychiatry and Behavioral Sciences Stanford University, Palo Alto, California June 19, 2005 Washington, D.C. Slides at: www.medafile.com (Dr. Ashford’s lectures)
30
Embed
Statistical Issues for Developing Alzheimer Screening Tests J. Wesson Ashford, M.D., Ph.D. Stanford / VA Alzheimer’s Center VAMC, Palo Alto, California.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Statistical Issues for DevelopingAlzheimer Screening Tests
J. Wesson Ashford, M.D., Ph.D.Stanford / VA Alzheimer’s Center
VAMC, Palo Alto, California&
Helena C. Kraemer, Ph.D.Department of Psychiatry and Behavioral Sciences
• Annual assessments (or bi- or semi-)• Positive screen recommends more
tests• Contrast with current lack of system
– 50% not diagnosed until moderate AD
• Screening will progressively improve• Change over time can be detected
AD Can Be Readily Diagnosed
• A diagnosis of Alzheimer’s disease can be made with a high degree of certainty
• Using NINCDS-ADRDA criteria, accuracy in autopsy-verified cases is approximately 90%
• Diagnosis is a 2-step process:– Detection through screening (test vs. family
concern)– Confirmation through patient history and
physical, caregiver interview, brain imaging, and appropriate laboratory studies
McKhann G et al. Neurology. 1984;34:939-944. Kazee AM et al. Alzheimer Dis Assoc Disord. 1993;7:152-164.Ashford JW et al, Psychiaric Annals, 1996;26:262-268.
AD is Under-diagnosed• Early Alzheimer’s disease is subtle, the diagnosis
continues to be missed – it is easy for family members to avoid the problem and
compensate for the patient – physicians tend to miss the initial signs and symptoms
• Less than half of AD patients are diagnosed– Estimates are that 25% to 50% of cases remain
undiagnosed– Diagnoses are missed at all levels of severity: mild,
moderate, severe
• No definitive laboratory test for diagnosing AD exists– Efforts to develop biomarkers, early recognition by
brain scan have not provided a screening methodology
Evans DA. Milbank Quarterly. 1990; 68:267-289
Reasons to Diagnose Alzheimer’s Disease Early
Social• Undiagnosed AD patients face avoidable
problems – social, financial
• Early education of caregivers– how to handle patient (choices, getting started)
• Advance planning while patient is competent– will, proxy, power of attorney, advance directives
• Reduce family stress and misunderstanding– caregiver burden, blame, denial
• Promote safety– driving, compliance, cooking, etc.
• Patient’s and Family’s right to know– especially about genetic risks
• Promote advocacy– for research and treatment development
Reasons to Diagnose Alzheimer’s Disease Early
Medical• Early diagnosis and appropriate intervention may lessen
disease burden and early treatment may improve overall course substantially– Neurophysiological pathways in patients with AD are
still viable and are a target for treatment
• Specific treatments now available (anti-cholinesterases, memantine)– Improve cognition– Improve function (ADLs)– Delay conversion to AD from Mild Cognitive
Impairment– Slow underlying disease process, the sooner the better– Decreased development of behavior problems– Delay nursing home placement, possibly over 20
months– Delay nursing home placement longer if started earlier
UNDERLYING CONTINUUM OF ALZHEIMER SEVERITY- progressive disruption of neuroplasticity -
(Ashford & Jarvik, 1985; Ashford et al., 1995; Ashford, 2004)
Iatragenic Damage? Clinical Wash Clinical Wash Clinical Gain
Major(?) Loss Minor (?) Loss Minor(?) Loss Major(?) Gain Some gain
PP’
SeSe’
SpSp’
Helena Kraemer, 2003
$W = Cost–Worthiness Calculation
• I = incidence (new occurrences each year, by age)• $T = cost of test, time to take (Subject, Tester)• Se = sensitivity of test = True positive / I• Sp = specificity of test = True negative / (1-I)• Cost:
– $B = benefit of a true positive diagnosis• Estimate: (100 years – age ) x $1000• Save $50,000 NH cost / 1year (after treatment cost deduction)
– $C = cost of a false positive diagnosis• $500 for further evaluation (time, stress of suspecting dementia)
– True negative (real peace of mind) (no price)– False negative = false peace of mind (no price)
BRIEF ALZHEIMER SCREEN (Normal vs Mild AD, MMS>19)
9
20
1413
1211
10
9
6
7
8
2627
25
0
10
20
30
40
50
60
70
80
90
100
0 10 20 30 40 50 60 70 80 90 100
False Positive Rate (%) (1-Specificity)
Tru
e P
osi
tiv
e R
ate
(%
) (
Se
nsi
tiv
ity)
animals 1 m AUC = 0.868
animals 30 s AUC = 0.828
MMSE AUC = 0.965
Date+3 Rec AUC = 0.875
BAS AUC = 0.983
JW Ashford, MD PhD, 2003
Mendiondo et al., 2003
Issues in Screening• ROC analysis provides independent values of test performance
– how the screening test values affect the normal and patient populations– plots of their relationship with respect to each other (specificity vs sensitivity)– data must be derived from the represented population!!!
• The value of the test must be calculated with respect to the risk of the disease– In the specific population to which it is being applied– Risk is affected by age, genotype, many other factors– Accounting for a priori probability is Bayesian analysis
• The cost-benefit must be assessed: – Apply the test cost and the costs of false positive and false negative results– Apply the benefits of correct positive and negative results
• Alzheimer’s disease is not a dichotomous diagnosis but a continuum– Diagnosis would be better described with a probabilistic statement– Item Response Theory would better calculate probability (Modern Test Theory)
Item Response Theory and Factor Analysis allow combination of test components
See: Hambleton et al., 1990; Ashford & Schmitt, 2001)
AD all
0.00
0.10
0.20
0.30
0.40
0.50
0.60
-4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10
DISABILITY SCALE
TES
T IN
FOR
MA
TIO
N
MMSE Item-Response Analysis
Summary Requirements for Screening Test Evaluation
• Sensitivity, specificity for the population to which test is to be applied– values change with level of disease being screened (must be
prospective)• Portion of population at risk
– Risks according to age, gender, genotype, family hx, education, etc.• Cost of test -$1, $10, $100, $1000
– Need brief screen to replace MMSE (ASAP) - $10 Medicare allowance• Costs of false positive, false negative tests• Benefits of true positive, true negative
– Longevity is increasing over time, needs to be factored in.• Cascaded tests – preliminary screen, confirmatory exam• Longitudinal tests may provide much more reliable information• Different tests for clinic (cascaded), research (outcomes)
– Clinic: brief screen, brain scan; Research: CSF - is disease stopped?• Benefit to society relative to other societal needs.
Dementia Screening Test Requirements for the Future
• For Alzheimer’s disease (other tests for non-AD)– Validated against more stringent tests – brain scans, CSF measures– Specificities and sensitivities valid for comparison with other tests– Item Response Theory analysis of discriminatory power on disability
continuum• Multiple platforms:
– Doctor’s offices– Best if computerized for rapid, objective assessment– World-Wide Web – based testing, – CD-distribution– KIOSK administration – drug stores, shopping malls
• Very brief (about 1-minute)• Multiple test forms
– so it can be repeated often, e.g., every 3 months• Cost-effective yearly after age 50
– repeatable every 3 months over 65 years of age or with concerns• Sensitive to change over time• Nominal cost