Trends in 21st Century Epidemiology: From Scientific Discoveries to Population Health Impact

David F. Ransohoff Presentation: Epidemiology and Evidence-Based Cancer Prevention

Slide 1 of 28: Trends in 21st Century Epidemiology: From Scientific Discoveries to Population Health Impact

Session 4: Use of epidemiologic research to advance clinical and public health practice: bridging the evidence gap with observational studies and randomized clinical trials

Moderator: Sheri D. Schully, Ph.D., Division of Cancer Control and Population Sciences, NCI

Epidemiology and evidence-based research along the cancer care continuum
David F. Ransohoff, M.D.
University of North Carolina at Chapel Hill

Panel and Audience Discussion

  • What are new ways in which epidemiology can be used to fill evidence gaps between discoveries and population health impact in the cancer care continuum?
  • How can observational epidemiology make the greatest scientific contributions in understanding cancer-related risk factors that cannot be studied through randomized clinical trials?

Cultivate Observational Cohorts


Slide 2 of 28: Cultivate Observational Cohorts

  1. Definition, Importance
  2. Past
    • examples, lessons
  3. Future
    • opportunities, challenges, recommendations

Slide 3 of 28: Cultivate Observational Cohorts

Definition (of cohort): defined group followed over time

Importance: Can cohort be used to answer question(s)?

  • Cohort can have "strong design" for questions of diagnosis, prognosis, response to rx (molecular markers) [RCT better, but may be not appropriate or impossible.]
  • Strength of design to answer question is related to features:
    • fair 'comparison' (avoid bias) for quest.: internal validity
    • relevant question: external validity
      • details: ascertain baseline state, exposure, outcome, etc

Devils in design/detail. One 'wrong' feature can be fatal.


Slide 4 of 28: Cultivate Observational Cohorts

"Observational" does not mean:

  • "passive" (e.g., PI is passive; or 'no design')
  • "annotated specimens" + "technology/data" + "bioinformatics"

Concept: "Specimens and data=product of a study.

With cohort data, you have to fashion a "study" (regarding comparison, bias, relevance, etc.) and describe it in Methods.

It's not "data+analysis."

It's a "study," whether thought about/not.

Ransohoff. JCO 2010;28:698


Slide 5 of 28: Cultivate Observational Cohorts

In cohorts that already exist, can strong design be arranged?

  1. PI imagines ideal design: specify question, data source, comparison, anticipate/avoid bias, etc.
  2. PI asks "In existing cohort, is inherent design close to ideal?" Could added design make it, overall, satisfactory, to answer that question?"

Concepts

  • Design (inherent, added) determines study strength.
  • If don't think about design early (re kinds of data, comparison, relevance), may limit kinds and strength of questions that can be addressed later.

Slide 6 of 28: Examples of Observational Cohort: Mostly T1, Lessons for Other Ts

(From Khoury et al., Am J Epidemiol. 2010 September 1; 172(5): 517–524 with permission of Oxford University Press.)

[Image] describing epidemiology and the phases of translation and knowledge synthesis from discovery to population health impact.


Slide 7 of 28: Examples of Observational Cohort: Mostly T1, Lessons for Other Ts

(From Khoury et al., Am J Epidemiol. 2010 September 1; 172(5): 517–524 with permission of Oxford University Press.)

[Image] from previous slide describing epidemiology and the phases of translation and knowledge synthesis from discovery to population health impact. Text boxes overlaid on table to explain that phase T0 includes etiology, T1 includes studies that relate to diagnosis and prognosis, and T2 relates to RCTs and outcome research. T1 was emphasized by speaker as needing improvement.


Slide 8 of 28: Cultivate Observational Cohorts

  1. Definition, Importance
  2. Past
    • examples, lessons
  3. Future
    • opportunities, challenges, recommendations

Slide 9 of 28: In examples, consider design, lessons

Design

  • What is inherent; what is added?
  • How much effort to add?
  • Did overall design have strength to answer question?

Lessons

  • How, in future, to cultivate observational cohorts that are strong?

Slide 10 of 28: 1. Prognosis BrCa

Paik S et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. NEJM. 2004; 351: 2817.

Question

  • In node-neg BrCa, is prognosis (i.e., low recurrence rate) discriminated by RNA signature?

Inherent design

  • In banked RCT, control group followed: dx to outcome.

Added design

  • measure RNA in FFPE specimen at diagnosis

Results

  • RNA signature prognostic: low recurrence rate

Slide 11 of 28: 1. Prognosis BrCa

Paik S et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. NEJM. 2004; 351: 2817.

Lessons

  • Inherent design has RCT strength: ascertain l.t. outcome, blinded, etc; clear relevant question
  • Piggybacking (adding) to strong inherent design:
    useful, if possible
  • This example:
    • NIH-funded, already banked
    • "Old" study can assess new molecules (validation or discovery)

Future: add 'specimens' to selected studies?


Slide 12 of 28: 2. Diagnosis OvCa (blood)

Zhu CS et al. A framework for evaluating biomarkers for early detection: validation of biomarker panels for ovarian cancer. Can Prev Res. 2011; 4: 375.

Question

  • Can blood proteomics screen for OvCa?

Background

  • Strong claims (2002), disappointment (2002-8) b/o weak design (bias in comparison etc.)

Inherent design

  • RCT (PLCO) ~1990; biorepository added mid-1990s, included serial bloods.

Added design ~2008

  • elect a blood just <dx for proteomics assay
  • blinded hypothesis testing'

Result

  • 5 groups' assay panels: no better than CA125.

Slide 13 of 28: 2. Diagnosis OvCa (blood)

Zhu CS et al. A framework for evaluating biomarkers for early detection: validation of biomarker panels for ovarian cancer. Can Prev Res. 2011; 4: 375.

Lessons

  • Diagnosis question addressed by serial specimens (blood), by selecting blood near time of diagnosis.
  • Expensive, difficult (big N subjects, specimens; small N cancer and of "relevant specimens")
  • NIH-funded; NIH arranges strong comparisons
  • "Old" study can assess new molecules
  • "If only bigger"... (what lessons from 'mega-cohort')

Slide 14 of 28: 3. Diagnosis CRC (stool DNA)

Imperiale TF et al. Fecal DNA versus occult blood for colorectal-cancer screening in an average-risk population. NEJM. 2004; 351: 2704.

Question

  • Can stool DNA screen for early CRC?

Inherent design

  • prospective cohort; industry (EXACT) DNA assay
  • expensive: specimen<colonoscopy; >5000 persons, 31Ca

Added design: (none)

Result

  • bad news: better than gFOBT, but expensive; biologically promising, clinically disappointing
  • good news: answer strong (reliable) because of design

Slide 15 of 28: 3. Diagnosis CRC (stool DNA)

Imperiale TF et al. Fecal DNA versus occult blood for colorectal-cancer screening in an average-risk population. NEJM. 2004; 351: 2704.

Lessons

  • If was greater amount of stool or blood, others could study new molecules (validation or discovery).
  • Industry resource is not 'shared.'

Slide 16 of 28: 4. Outcome CRC screening

Selby JV et al. A case-control study of screening sigmoidoscopy and mortality from colorectal cancer. NEJM. 1992; 326 (10): 653.

Question

Can sigmoidoscopy reduce CRC mortality in L colon?

Inherent design (1970s+)

HMO cohort, some sig screening was done

Added design (years later)

  • nested case-control study
  • learn cause of death
  • learn whether exposure occurred (sig for screening)
  • create internal control group

Slide 17 of 28: 4. Outcome CRC screening

Selby JV et al. A case-control study of screening sigmoidoscopy and mortality from colorectal cancer. NEJM. 1992; 326 (10): 653.

Result

  • L-sided CRC mortality reduced ~60%.

Lesson

  • Assess RCT question in case-control (observ.) study.
  • Strength: nested c-c; exposure reason known.
  • Could one add bloods, other specimens, and answer other questions.

Slide 18 of 28: 5. PrCa Prognosis

[Image] of Canary Foundation webpage with information about prostate cancer clinical studies.

Question

  • Can markers identify lethal vs non-lethal PrCa?

Inherent design (PASS)

  • Prospective cohort, N>1000, active surveillance.

Added design: (none)

Results: (none)

Comment

  • If 'lethal' PrCa is rare, are results limited?

Lesson

  • Cohorts may have limitations.

Slide 19 of 28: Observational cohorts cultivate: other examples

  1. Research studies designed as RCT, cohort
    • Framingham
    • Nurses Health Study; Physicians Health
    • WHS
  2. (used to study diagnosis, prognosis, etc)

  3. Practice settings
    • HMOs (Kaiser-Permanente, Group Health, etc)
    • Eli Lilly etc
    • other

Slide 20 of 28: Cultivate Observational Cohorts

  1. Definition, Importance
  2. Past
    • examples, lessons

Examples and concepts are not new to this group.

Our focus: Lessons about how to cultivate observational cohorts.


Slide 21 of 28: Cultivate Observational Cohorts

  1. Definition, Importance
  2. Past
    • examples, lessons
  3. Future
    • opportunities, challenges, recommendations

Slide 22 of 28: Future: Opportunity

An illustrative example: Molecular markers (blood) for CRC screening

Background

  • In design to discover/validate molecular test, specimen (e.g. blood) must be obtained <procedure; req. big N.
  • What cohorts could be cultivated?
  • In existing cohort infrastructures, add spec. collection (RCTs of EU, VA; HMOs; practices)
  • Specimens could be used for validation and/or discovery.

Slide 23 of 28: Future Opportunity

An illustrative example: Molecular markers (blood) for CRC screening

Background

  • In design to discover/validate molecular test, specimen (e.g. blood) must be obtained <procedure; req. big N.
  • What cohorts could be cultivated?
    • In existing cohort infrastructures, add spec. collection (RCTs of EU, VA; HMOs; practices)
  • Specimens could be used for validation and/or discovery.
    • Imagine big N, big volume of blood, stool; then banked specimens useful in discovery/validation.

Approach is generalizable to many problems.

Challenges: logistics, motivation.


Slide 24 of 28: Future Challenges

What available cohort sources, infrastructures

  • ongoing research studies
  • practice settings
  • e.g., CRN, HMORN, HMOs; Cohort Consortium; etc etc

What are logistics of 'cultivating'

  • How to anticipate questions and technologies; impact on "design"
  • Add what?
  • Who 'drives' research if different from who 'owns' data?
  • non-trivial: consider CRN, co-op groups

Slide 25 of 28: Future Challenges

Other challenges:

  • how to cultivate efficiently; avoid wasted effort (past examples)

Slide 26 of 28: Recommendation: Cultivate observational cohorts

But how?

  1. Make sure we understand lessons of past; ideas not new.
  2. Approaches
    • big effort; big N of smaller studies (let 1000 flowers bloom)
    • piggyback onto current infrastructure
    • role of nested case-control design
    • considering 'megacohort'? beware limitations
  3. Don't just collect data/specimens/annotate; do consider role of questions, methods/design to answer, etc.
  4. Try different approaches, get preliminary data, scale up.

How to organize, supervise this effort...


Slide 27 of 28

Session 4: Use of epidemiologic research to advance clinical and public health practice: bridging the evidence gap with observational studies and randomized clinical trials

Moderator: Sheri D. Schully, Ph.D., Division of Cancer Control and Population Sciences, NCI

Epidemiology and evidence-based research along the cancer care continuum
David F. Ransohoff, M.D.
University of North Carolina at Chapel Hill

Panel and Audience Discussion

  • What are new ways in which epidemiology can be used to fill evidence gaps between discoveries and population health impact in the cancer care continuum?
  • How can observational epidemiology make the greatest scientific contributions in understanding cancer-related risk factors that cannot be studied through randomized clinical trials?

* Cultivate Observational Cohorts


Slide 28 of 28: Acknowledgements

National Cancer Institute
Division of Cancer Prevention

  • BRG- Biometry Research Group
  • EDRN- Early Detection Research Network
  • EDRG- Early Detection Research Group (PLCO)
  • CPTAC- Clinical Proteomic Technology Assessment for Cancer

Return to Top