2nd CUTEHeart Workshop Constantine Gatsosis Presentation

The link of diagnostic tests and patient outcomes: methodologic challenges

April 2016 CUTEheart Workshop

Constantine GatsonisDepartment of Biostatistics and Center for Statistical SciencesBrown University School of Public Health

Outline 1. The problem: Broadly shared concern about the

effectiveness of modalities for diagnosis and prediction

2. The solutions, classical and new:

a. Randomized studies

b. Observational studies:

– Large databases: EMR, Health Insurance Claims

– Large registries: The National Oncology PET Registry studies

– Combining registries with administrative and insurance databases

3. Evaluation of effectiveness of diagnostics revisited



Diagnostic testing is receiving much attention and skepticism

1 ‐ Specificity

0.0 0.2 0.4 0.6 0.8 1.0

0.0

0.2

0.4

0.6

0.8

1.0

Sensitivity


CMS to Trim Spending on Diagnostic Lab TestsThe Centers for Medicare and Medicaid Services appear to be executing a two‐step strategy to shrink the $8 billion annual price tag of clinical diagnostic laboratory tests.On Sept. 25, CMS released a final payment determination for the Clinical Laboratory Fee Schedule that cuts payment by over 90 percent for some tests

October 2, 2015

The Health Care Value Transparency Movement and Its Implications for Radiology Daniel J. Durand, MD et al, JACR 2015 ….radiology, along with other “ancillary services,” has been a major focus of early efforts to enhance consumer price transparency. By contrast, radiology as a field remains in the “middle of the pack” with regard to quality transparency.

Addressing Overutilization in Medical Imaging Hendee et al, Radiology 2010Key forces influencing overutilization .. include payment mechanisms and financial incentives; the practice behavior of referring physicians; self‐referral, including referral for additional radiologic examinations; defensive medicine; missed educational opportunities when inappropriate procedures are requested; patient expectations; and duplicate imaging studies.

Tests and therapeutic interventions

Fundamentally tests provide information for use in selecting course of care. Both long‐ and short‐term effects of tests materialize in context of available health care options, including therapeutic interventions.Not possible to define and measure test effects outside the particular health care context in which the test will be used. However, oftentimes diagnosis may be ahead of therapy: e.g. DCIS in breast cancer, Amyloid plaque testing in Alzheimer’s


Main questions in test evaluation


Accurate? • Accuracy in detection• Accuracy in prediction

Affects Care ?

Affects Outcome?

Process of care:• Dx thinking/decision making• Tx thinking/decision making

Patient outcomes:• Quality of life, satisfaction, cost• Mortality, morbidity

The challenge for diagnosis and prediction

Dx informationOutcomes

How can we illuminate the complex path

The main pathway goes through therapyApril 2016 CUTEheart Workshop

THERAPY

CUTEheart WorkshopApril 2016

A heuristic argumenthighlighting key constraints

R

Test A

Test B

Tx1

Tx1

Tx2

Tx2

Outcome

Outcome

Outcome

Outcome

+

+

‐

‐

Simple randomized design, comparing two tests


Difference in success rates between two arms:D=(r1-r2)xpx(SensA – SensB)

r1 and r2 = success rates for therapeutic interventions Tx1 and Tx2, when performed on cases that have the clinical condition (irrespectively of which test detected them).

p = prevalence of the clinical conditionSens= test sensitivity (Specificities are assumed equal above).

Typically, overall effect D between arms will be much smaller than difference in efficacies of therapies r1-r2

If r1-r2 = 0 (therapy not effective) or SensA = SensBthen overall effect of test D=0. April 2016 CUTEheart Workshop

Key observations:

R

CT, 3 annual screens

WorkupTherapy

No further intervention

Lung Ca. Mortality+

+

‐

‐

Example: Lung Cancer Screening Trial


X‐ray , 3 annual screens No further

intervention

Lung Ca. Mortality

Lung Ca. Mortality

Lung Ca. Mortality

N=53546

A randomized study of strategies formanaging stable angina patients

CUTEheart Workshop

http://www.acrin.org/PROTOCOLSUMMARYTABLE/ACRIN4701RESCUE/tabid/747/Default.aspx

April 2016


Randomized designs in practice

Key design question: Randomization to test only or to strategy that includes the test?Example 1: CT Angiography for patients with stable angina. – Randomize to anatomic (CTA) vs functional test (e.g. SPECT) without specified followup (as done in the PROMISE trial) or

– Randomize to strategies linking test results to patient management ( is patient candidate for optimal medical therapy (OMT)?) (RESCUE trial)

Example 2: Preoperative MRI in breast cancer– Randomize to MRI or no MRI before surgery, without algorithms for

use of MRI results (COMICE trial, )– Randomize to MRI or no MRI, with algorithms to guide use of test

results (Alliance A011104 / ACRIN 6694, active trial)

A more elaborate design-Paired tests-Randomized discordantpairs


Outcome

Tx2

Outcome

A B

+ + Tx1

Tx1

Tx2

Outcome

Outcome

Outcome

Tx1

- -

Tx2 Outcome

+ -

- +

R

R


Randomized designs in practice, cont.

Paired test designs– Efficient but rarely adopted. MINDACT trial is example.

– Ethical issues‐ can positive or negative test results be ignored?

– Implementation difficulties. Strong potential for non‐compliance.

Beyond clinical trials

• Observational studies– Registry studies – Electronic Medical Records– Claims datasets– Cohort studies

• Research synthesis– Systematic reviews and meta‐analysis– Modeling


Use of large databases (e.g. EMR, Claims)

General approach:– Construct longitudinal record including test and outcome information.

– Use methods for observational data to make comparisons.

Recent examples:– Coronary calcium study using Medicare claims.– Testing of ER patients with chest pain study using EMR data.



Outcomes After Coronary Artery Calcium and Other Cardiovascular Biomarker Testing Among Asymptomatic

Medicare BeneficiariesJacqueline Baras Shreibati, MD, MS; Laurence C. Baker, PhD; Michael V. McConnell, MD, MSEE; Mark A.

Hlatky, MD

Effectiveness of CAC and hs-CRP testing

Circ Cardiovasc Imaging.2014;7:655‐662.)

20% random sample of Medicare beneficiaries

without CVD claims in previous 6 mos.

Propensity score matching of patients receiving CAC and high sensitivity C-reactive protein testingCompared utilization and outcomes

Testing of ER patients with chest pain

>420K ER patients. Records from 2011 MarketscanPatients classified into 5 testing strategiesOutcomes compared using regression analysis


JAMA Intern Med. 2015;175(3):428‐436.

Limitations of Claims and EMR data

EMR and claims data typically deficient in key information about

Context of the test (indication, filtering)Test interpretation and findings Clinical characteristics of patientSubsequent diagnostic and therapeutic decisions


DiagnosticProcedure

OutcomesTHERAPY

Clinical Context





DiagnosticProcedure

OutcomesTHERAPY

Clinical Context

Defensivemedicine





DiagnosticProcedure

OutcomesTHERAPY

Clinical Context

Defensivemedicine

Creative Coding

National Oncologic PET Registry: A Nationwide Collaborative Program

Sponsored by

Managed by

Advisor

Goal: Assess the effect of PET on referring physicians’ plans of intended patient management.

• across a wide spectrum of cancer indications for PET, not covered currently by the Medicare program, •in relation to cancer‐type, indication, performance status, physician’s role in management, and scan type


http://www.cancerpetregistry.org

Biostatistics Center atBrown University

NOPR Accrual (Cases Completed/Business Day)

More than 200,000 patients registered

PET Changed Intended Management in 36.5% of Cases

Non-Treat Treat 23.2 31.6 28.6 29.2 28.3Treat Non-Treat 7.9 7.9 7.5 9.7 8.2Patients with change post-PET 31.1 39.5 36.1 39.0 36.5

Hillner et al J Clin Onc 2008

Clinical Indication for PET Study (Percent)

Pre-Pet Plan

Post-PET Plan

Dxn=5,616

Staging n=6,464

Restaging n=5,607

Recurrence n=5,388

Alln=22,975

Treat Same 16.0 46.5 15.8 20.4 25.5Non-Treat Same 52.9 14.0 48.0 40.7 37.9


Summary of NOPR Results (Before 2009 NCD)Overall Impact on Patient Management

– Diagnosis, Staging, Restaging, Recurrence– Data on 22,975 scans from May 8, 2006 – May 7, 2007– J Clin Oncol 2008; 26:2155‐61

Impact on Patient Management by Cancer Type– Confirmed Cancers– Staging, Restaging, Recurrence– Data on 40,863 scans from May 8, 2006 – May 7, 2008– J Nucl Med 2008; 49:1928‐35

Treatment Monitoring– Data on 10,447 scans from May 8, 2006 – Dec 31, 2007– Cancer 2009:115:410‐18



Do NOPR findings agree with those from • other studies or • other sources of information on the same participants

How does registry information compare to other sources?

Combining registries with administrative and insurance databases

– Linking NOPR and CMS claims data


Registry record

CMS claims, longitudinal

record

NOPR strengths

Provided extensive, timely, “real world” data, that are very difficult and expensive to collect in clinical trials.Prospectively organized and controlled data collectionRegistry data can be effectively linked to data from other information sourcesResults consonant with more tightly controlled studies (later in presentation)Supports the Coverage with Evidence Decisions approach of CMS


NOPR weaknesses (long list!)


Evidence documents change in intended management, not actualmanagementNo evaluation of whether management changes were appropriate.No information on whether the use of PET improved long‐term outcomesHow should PET should be used in the flow of patient care.Plus the concerns about generalizability and bias in data obtained from registries

Imaging Amyloid Plaques (PIB-PET)A harder case: Diagnosis ahead of therapy


18F-florbetaben (Piramal) March 2014 (Neuraceq)

18F-florbetapir (Avid/Eli Lilly) April 2012 (Amyvid)

18F-flutemetamol (GE) October 2013 (Vizamyl)

CMS National Coverage Decision

Insufficient evidence of clinical utility to justify coverage of Aβ PETReimbursement would be considered under coverage with evidence development (CED) in clinical studies designed to:• Develop better treatments or prevention strategies for AD

• Identify subpopulations at risk for developing AD• Resolve clinically difficult differential diagnoses (e.g., frontotemporal dementia versus AD)

• Must demonstrate Aβ PET improves health outcomes (short‐term outcomes related to changes in management as well as longer‐term dementia outcomes)CUTEheart WorkshopApril 2016

Combining registries with administrative and insurance databases

– Amyloid plaque imaging: Comparison of

• Registry of participants undergoing Dx, and

• Matched Controls identified from CMS claims data


?? ?

?

IDEAS Steering CommitteeCore Science TeamGil Rabinovici, UCSFMaria Carrillo, Alzheimer’s AssociationConstantine Gatsonis, Brown UBruce Hillner, VCUBarry Siegel, Wash URachel Whitmer, Kaiser Permanente

Additional committee membersWilliam Abbot, Piramal ImagingRosemarie Hakim, CMSMeredith Johnson, GE HealthcareMark Mintun, Avid RadiopharmaceuticalsDon Rosen, ACR

ACR OperationsCharlie ApgarCynthia Olson

MITA representative :Terri WilsonConsultant: Brian Carey, FoleyHoag LLP

Biostatistics Center (Brown U)Ilana GareenRoee GutmanErin GrecoLucy HannaBenjamin HermanRajesh Makineni

Scientific and logistical supportJim Hendrix, Alzheimer’s Association

[email protected]


IDEAS (Imaging Dementia—Evidence for Amyloid Scanning)

Two main goals:Aim 1: To assess the impact of amyloid PET on the management of patients meeting Appropriateness Use Criteria – This aim will be addressed using a registry of

participants undergoing amyloid PET.Aim 2: To compare hospital admissions and emergency room visits within 12 months in patients enrolled in the registry to matched controls (no amyloid PET), to be identified from CMS claims data. – This aim will be addressed using linked registry and CMS claims data.

https://clinicaltrials.gov/ct2/show/NCT02420756?term=IDEAS&rank=1


Sample size considerations

Aim 1 Hypothesis: Determine whether amyloid PET imaging will lead to a ≥ 30% change between intended and actual patient management within 90 days in a composite measure consisting of the following:a) AD drug therapy; b) Other drug therapy; orc) Counseling about safety and future planningN=11,050 for 80% power (separate analyses for dementia and MCI subgroups)

Aim 2 Hypothesis: Determine whether amyloid PET is associated with a ≥ 10% relative reduction in registry patients in comparison to matched controls in the following endpoints: a) Inpatient hospital admissions over 12 months. b) Emergency room visits over 12 monthsN= 18,500 for 90% power (separate analyses for each endpoint)



In closing

(Can we assess) how effective is diagnosis in health care?

Current approaches can be enhanced: Linking tests to downstream outcomes is a challenging undertaking.Large secondary databases can be used, but may lack necessary information.Registries offer flexibility but typically have narrow focus.Combination of information from several sources holds promise.



Solutions to the Problem: Coverage with Requirement for Data Collection through a National Registry or Registries1) Require Standardization of High Quality Testing2) Compare Back to Companion Diagnostics Where Available3) Collect Outcomes: High‐level clinical outcomes need to be collected and compared back to the new testing standard. 4) Attach Testing to Existing and Future Clinical Trials5) Aggregate All Data, Nationally and Internationally

Editorial by ASCO


New realities in health care:Integrated Systems, ACOs, etc


Comment in “Modern Health Care” The CMS stance on PET is an artifact of fee‐for‐service medicine. In a few years, as insurers in both the public and private sectors gravitate toward episode‐based and bundled care, the risk of using expensive and unproven technologies that don't improve outcomes won't fall on payers. It will fall on the purchasers of those expensive technologies which, for PET scanners, are mostly hospitals.

Decision Memo for Positron Emission Tomography (NaF-18) to Identify Bone Metastasis of Cancer (CAG-00065R)CMS has concluded that the evidence is not sufficient to determine that the results of NaF-18 PE imaging to identify bone metastases improve health outcomes of beneficiaries with cancer. Therefore we have determined that this use is not reasonable and necessary under §1862(a)(1)(A) of the Social Security Act.

New realities in health care:Integrated Systems, ACOs, Bundled Care etc


The CMS stance on PET is an artifact of fee‐for‐service medicine. In a few years, as insurers in both the public and private sectors gravitate toward episode‐based and bundled care, the risk of using expensive and unproven technologies that don't improve outcomes won't fall on payers. It will fall on the purchasers of those expensive technologies which, for PET scanners, are mostly hospitals.

• New incentives for health care systems to developmandexpand their own research efforts.

• Realistic hope for better data from more effective and efficient information systems.

• Continuous, integrated, enterprise‐wide data collection • Effectiveness questions posed from a “systemic” perspective in contrast to the current la carte evaluation of components of health care.

• Metrics and endpoints would be adapted to new realities.• Validity and Generalizability of findings may be a challenge.


Thank you!

Key challenges

Best target population (age range?; MCI vs. atypical dementia?)No established clinical management algorithms based on scan results (unlike situation for FDG‐PET in cancer)No good historical control dataSome patient‐centered outcomes will take quite long to be detectable (slowing of functional decline vs. avoiding futile Rx)


NOPR linked to CMS claims data


Approach

Construct health care utilization history using CMS claims dataCohort of 8640 consenting registry participants, age 65+, enrolled in 2007‐2008 undergoing PET for cancer restaging.Extensive database work involved to identify patterns of care and classify into post‐PET treatment categories.

(Med Care 2013;51: 361–367)

Australian Prospective Studies of Impact of PETAgreement in Post-PET Plan and Actual Care (2003-2006)


Cancer Pts Centers Indication Change in Plan

F/U (mo.

)

Agree-ment

Ovarian 90 3 SR 58.9 6 67.8 Esophagus 129 5 IS 38.0 12 53.2 Lymphoma 74 6 IS 34.0 6 74.3 Colorectal 93 4 SR 65.6 6 62.0 Colorectal 98 4 Resect

Hepatic 49.0

6

70.1

Head/Neck 71 3 IS 33.8 3 74.7 SR: Suspected Recurrence. IS: Initial Staging

Referring MD requests amyloid

PETPatient consent.

Visit 1 Pre‐PET

assessment and form completed

Visit 2PET done, interpreted

& reported.MD communicates results to participant

Visit 390‐day Post‐PET Assessment andform completed

Ongoingpatient

management

IDEAS Registry Workflow (main steps)



Additional material

Impact of PET Used for Treatment Monitoring

• Chemotherapy 82%, chemoRT 12%, RT 6%• Ovarian, pancreas, NSCLC, SCLC most frequent• Metastatic disease in 54%• PET findings led to:

– Switch to another therapy in 26%

– Adjust dose or duration of therapy in 17%

– Switch from therapy to observation/supportive care in 6%

• Management change more often if post-PET prognosis worse rather than improved/unchanged (70% vs. 40%)

Hillner et al., Cancer 2009; 115:410


NOPR Analysis by Cancer Types

Two years of data (40,863 scans)Confirmed Cancers

Initial staging, restaging, suspected recurrencesAll Cases Change in

Management95% C.I. Imaging adjusted

impact

Initial staging 39.8 39.0-40.6 13.4Restaging 35.9 35.1-36.7 12.6Suspected Recurrences 38.5 37.6-39.3 14.7


All Cases Initial Staging Restaging Suspected Recurrence

Prostate 32.0 34.0 39.4Ovary 43.1 37.7 44.5Bladder 39.9 36.4 36.7Pancreas 39.2 38.3 39.3

Changes in Intended Management (%) Stratified by Pre-PET Plan

Image n=9,518

Biopsy n=3,552

Watch n=2,199

Treatm. n=7,706

Post-PET PlanImage 5.8 6.0 4.6 3.0Biopsy 9.5 24.0 9.0 6.8Watch 37.2 33.6 62.3 15.6Same Rx NA NA NA 42.4New or Major Change in Rx

47.6 36.3 24.1 8.7

Minor change Rx

NA NA NA 23.5

Pre-PET Plan

Hillner et al., J Clin Oncol 2008; 26:2155


Patient presentation

Treatment Decisions

Diagnostic workup

Patient outcomes


The study of modalities for diagnosis and Prediction: Conceptual Framework

Fundamentally tests provide information for use in selecting course of care. Both long‐ and short‐term effects of tests materialize in context of available health care options, including therapeutic interventions.Not possible to define and measure test effects outside the particular health care context in which the test will be used. However, oftentimes diagnosis may be ahead of therapy: Diagnosis of DCIS; Amyloid plaque imaging for Alzheimer’s.


Pre‐PET Form: Intended Patient Management

Observation (with close follow‐up)Additional imaging (CT, MRI) or other non‐invasive diagnostic testsTissue biopsy (surgical, percutaneous, or endoscopic). Treatment (if treatment is selected, then also complete the following)Treatment Goal: (check one) Curative PalliativeType(s): (check all that apply)

– Surgical Chemotherapy (including biologic modifiers)– Radiation Other Supportive care

If PET were not available, your current management strategy would be (select one)?

Intended management, given PET findings, asked on post-PET form


Cohort Profile• First year of NOPR

(5/8/06 to 5/7/07)• 22,975 “consented”

cases from 1,519 facilities

• Technology profile – 84% PET/CT– 71% non-hospital– 76% fixed sites

Hillner et al., J Clin Oncol 2008April 2016 CUTEheart Workshop

Intended vs Inferred Care after PET for initial staging


J Nucl Med 201354:2024–2031

Cohort of 4661 consenting registry participants, age 65+, enrolled in 2007‐2008. PET for initial cancer restaging. N PPV Agreement KappaAny systemic txAll 3,030 79.3 74.1 0.40Bladder 676 79.0 73.0 0.42Ovarian 223 86.1 82.5 0.52Stomach 648 67.6 67.6 0.34

Any radiationAll 1,293 64.7 80.8 0.53Bladder 337 68.8 83.3 0.60Ovarian 23 30.4 91.6 0.32Stomach 250 48.8 79.0 0.39Any surgeryAll 1,031 63.6 77.9 0.43Bladder 248 56.9 75.8 0.36Ovarian 57 52.6 75.9 0.31Stomach 458 71.4 75.7 0.50

ConclusionBetter concordancethan in restaging PET

Clark et al. JAMA 2011, Lancet Neurol 2012

Pathology Validation: Florbetapir PET

Median 5 raters vs. autopsySensitivity 92%(69%-95%)Specificity 95%(90%-100%)

FDA approval:• Florbetapir

(AmyvidTM) 2012

• Flutemetamol(VizamylTM) 2013

• Florbetaben(NeuraceqTM) 2014


Amyloid vs. FDG-PET in Differential Diagnosis of AD vs. FTD

Rabinovici et al. Neurology 2011Rabinovici et al. AAN 2014

AD (N=62, age 65, MMSE 22)FTD (N=45, age 65, MMSE 22)Amyloid (PIB) PET visual reads

90% sensitivity, 83% specificityInter-rater agreement κ=0.96

FDG-PET visual reads78% sensitivity*, 84% specificityInter-rater agreement κ=0.72*

47 autopsy-proven casesPIB: Sensitivity 100%, Specificity 90%FDG: Sensitivity 87%, Specificity 78%

* ‐ p<0.05 vs. PIB


IDEAS study (under development)

Organization:

Working collaboration of researchers from NOPR Working Group, Alzheimer’s Association, SNMMI, American College of Radiology Imaging Network, and Brown Biostatistics.

Stakeholders from industry and professional organizations.

Support from industry and from CMS (coverage of scans)


Control group construction: Selection of superset


Criterion Factors/Examples Time1 New Diagnosis of MCI or

Dementia*CMS claims related to dementia and MCI in the 24 mo. prior to IDEAS Study initiation

≥24 mo.

1A MCI CODE 331.83 Mild cognitive impairment-other codes to be selected in sensitivity analysis

1B DEMENTIA CODE: Prior to structural imagingInconsistent claims patterns*

≥2 different categories of dementia codes or>2 claims with exclusively limited to non-specific dementia codes) occurring within a two year (24 month) time interval

2 Minimal Structural Brain Tests Minimal Blood Laboratory w/u

Head MRI or CT (+/-) CBC, standard blood chemistry profile, TSH, vitamin B12

<12 mo.

3 Exclude cases with unanticipated brain pathology on intake MRI/CT

Such as primary or metastatic cancers at time of match and at baseline for contemporaneous cohort

4 Exclude cases with:‐ All Non-Skin Cancers,

Lymphomas, Hematologic Malignancies

‐ Hip/Pelvic Fracture

<12 mo.

PET Amyloid Imaging

Imaging biomarkers for amyloid deposits have been approved by FDA:– 18F‐florbetapir in April 2012,– 18F‐flutemetamol in October 2013– 18F‐florbetaben in March 2014 CMS National Coverage Decision in September 2013Coverage for one study per patient, but only under CED to:– Develop better treatments or prevention strategies for AD, or,

as a strategy to identify subpopulations at risk for developing AD

– Resolve clinically difficult differential diagnoses (e.g., FTD vs. AD) with goal of improving health outcomes (including short term outcomes related to changes in management as well as longer term dementia outcomes).


Control group construction: Matching criteria


Criterion Factors/Examples Time1 Age Minimal age of 66 to 67 years +/-24mo.2 Gender N.A.3 Marital status N.A.4 Ethnicity N.A.5 Location/Service Area Hospital Referral Region (n=306) N.A.6 Match Chronic

Condition Warehouse per table below: Focus is on conditions related to hospitalizations

Match on number or specifics of associated CMS Codes. Non-CHF heart disease is an aggregate of Acute MI, A fib,, HTN, and CAD/IHD), CKD, COPD, diabetes

<24 mo.

7 Match for Chronic Neurological Conditions, not included in Warehouse

Prior stroke, TIA, PD, MS, epilepsy, TBI

<24 mo.

8 Hospitalization/Emergency Room Visits

Dementia or delirium <12 mo.

9 Medicare/Medicaid Eligibility

Single or dual eligibility N.A.


rA1= rate of favorable outcomes for pts randomized to test A

rB2= rate of favorable outcomes for discordant cases randomized to be treated according to result of A

Difference between arms in simple randomized design

Difference between arms in randomized of discordant cases design

D1= rA1 ‐ rB1 =(r21-r11) x p x (SeB – SeA)+(r22-r12) x (1-p) x (SpA – SpB)

D2= rA2 – rB2 =[(r21-r11) x p x (SeB – SeA)+(r22-r12) x (1-p) x (SpA – SpB)]

= D1/ fwhere f = fraction of discordant cases


Collect Registry participant data

Select superset of potential matches from national CMS claims data using inclusion/exclusion criteria

Select matching cases from superset

Streamlined process flow of selection of controls


The Recent Reversal of the Growth Trend in MRI: A Harbinger of the Future?Richard Sharpe et al. JACR 2013

JACR 2012


October 2, 2015

Referring MD requests PET

Pre-PET Form

PETdone

PET interpreted& reported

Post-PETForm sent,

including question for referring MD consent

Post-PET Form completed.

Claim submitted

Ongoingpatient

management

NOPR Workflow

Ask patient for consent


Agreement of Post‐PET Plan and Claims‐inferred Actions at 30 Days


Bladder Ovary StomachPatients (total), n 1127 2075 632Systemic tx only planned, n 261 811 134

PPV (%) 66.7 65.0 51.5Raw agreement (%) 71.2 71.1 65.7k 0.33 0.40 0.17

Radiation only planned, n 71 46 34PPV (%) 74.6 67.4 35.3Raw agreement (%) 89.4 96.9 95.3k 0.42 0.48 0.23

Biopsy planned, n 150 176 85PPV (%) 64.7 55.1 48.2Raw agreement (%) 74.4 80.5 75.2K 0.27 0.23 0.21

Watching, n 473 804 320PPV (%) 70.8 74.1 66.3Raw agreement (%) 71.0 69.1 60.8k 0.41 0.38 0.21

Med Care 2013;51: 361–367

Conclusion: Moderate concordance

National coverage of NOPR registry (2006-2009)



Randomized designs in practice, cont.

Randomization to test only or to strategy with test?– Tradeoff between generalizability and efficiency

• Test only: “noisy” (hence larger sample sizes) but may be more generalizable

• Test as part of strategy : Less “noise” but results apply to specific strategies.

– Compliance is important factor to study.Paired test designs– Efficient but rarely adopted (see recent report in JClinEpi). MINDACT trial is example.

– Ethical issues‐ can positive or negative test results be ignored?

– Implementation difficulties.

2nd CUTEHeart Workshop Constantine Gatsosis Presentation

Education

Transcript of 2nd CUTEHeart Workshop Constantine Gatsosis Presentation