Title: Retrospective Studies of Clinical Outcomes A Primer for Clinicians
1Retrospective Studies of Clinical OutcomesA
Primer for Clinicians
- Marc D. Silverstein, MD FACP
2Overview
- Role of retrospective studies in a research
program - Anatomy physiology of research
- Observational research designs
- Cross-sectional studies
- Cohort studies
- Case-control studies
- Human Subjects IRB Review
3Objectives
- Describe 4 research designs
- Describe 3 threats to validity
- List advantages of retrospective studies
- List disadvantages of retrospective studies
- Understand requirements and processes for IRB
review of retrospective studies
4Research
- Definition
- Types of Research
5Definition of Clinical ResearchNIH Directors
Panel, 1997
- Patient-oriented research
- Research conducted with human subjects (or on
material of human origin such as tissues,
specimens and cognitive phenomena) for which an
investigator (or colleague) directly interacts
with human subjects - Mechanisms of human disease
- Therapeutic interventions
- Clinical trials
- Development of new technologies
- Epidemiologic and behavioral studies
- Outcomes research and health services research
6Health Services Research
- Health services research is the multidisciplinary
field of scientific investigation that studies
how social factors, financing systems,
organizational structures and processes, health
technologies, and personal behaviors affect
access to health care, the quality and cost of
health care, and ultimately our health and
well-being. - Its research domains are individuals, families,
organizations, institutions, communities, and
populations. - AcademyHealth
7Outcomes Research
- Research on measures of changes in patient
outcomes - patient health status and satisfaction
- resulting from specific medical and health
interventions - Attributing changes in outcomes to medical care
requires distinguishing the effects of care from
the effects of the many other factors that
influence patients health and satisfaction - AcademyHealth
8Patient Care and Outcomes Research
9Physiology of Research
- Design the study
- Implement the study
- Make valid (causal) inferences
10The Goal
11Design Implementation
Design
Implement
Research Question
Actual Study
Study Plan
12Causal Inferences
13Errors May Occur in Design Implementation of
Research
Error
Error
Design
Implement
Infer
Infer
14Errors May Occur in Making Inferences about
Internal or External Validity
Design
Implement
Error
Error
Infer
Infer
15Errors May Occur Anywhere
Error
Error
Implement
Design
Error
Error
Infer
Infer
16Threats to Validity
17Threats to Validity
- Chance random error due to unknown sources of
variation that distort sample and measurements in
either directions - Bias systematic error that distorts sample and
measurements in one direction - Confounding an external factor that is
associated with a predictor variable and an
outcome variable
18Reducing Random Error
- Increase sample size
- Statistical analyses
19Reducing Systematic Error (Bias)
- Population-based studies
- Inclusion and exclusion data
- Standardize measurement
- Train and certify observer
- Refine instrument
- Automate instruments
- Blinded measurements
20Reducing Confounding
- Anticipate potential confounders
- Measure potential confounders
- Matching, restriction, stratification
- Multivariate analysis
21I cannot give any scientist of any age better
advice than this the intensity of the conviction
that a hypothesis is true has no bearing on
whether it is true or not. P.B. Medawar
22Anatomy of Research
- Research Question
- Significance
- Methods
23Research Question
24Characteristics of a Good Research Question
- Feasible
- Interesting
- Novel
- Ethical
- Relevant
25It can be said with complete confidence that any
scientist of any age who wants to make important
discoveries must study important problems. P.B.
Medawar
26Hypotheses
- Simple (versus complex)
- Specific (versus vague)
- In advance (versus after-the-fact)
27Estimation
- In clinical studies the goal is often to estimate
a risk of an outcome or the magnitude of impact
on a clinical measurement
28Observational Studies
- Strengths
- Intervention is not feasible or ethical
- Rapid efficient
- Existing data
- Less time
- Expenses are lower
- Case-control studies for rare events
- Limitations
- Do not permit true assessment of time sequence of
factors and outcomes - Subject to bias and confounding
- Limited power to study rare risk factors or rare
outcomes (surveys cohort studies)
29Study Designs
- Observational studies
- Cross sectional studies
- Cohort studies
- Case Control studies
- Experiments
30Cross-Sectional Study Design
31Cross Sectional Study
Population
Sample
32Cross Sectional Study Example
- The Probability of Malignancy in Solitary
Pulmonary Nodules - Swensen, Silverstein, Ilstrup et al
- Arch Intern Med 1997 157 849
33The Probability of Malignancy in Solitary
Pulmonary Nodules
- Can clinical and radiological SPN characteristics
predict malignancy in SPNs - Retrospective cohort at multi-specialty group
practice - New diagnosis of 4mm 30 mm solitary pulmonary
nodule - Radiological indeterminate SPN with no
calcification on thin section CT - Exclude patients with primary lung cancer or
other cancer within 5 years
34The Probability of Malignancy in Solitary
Pulmonary Nodules
- Outcomes determined by radiological follow-up for
2 or more years, surgical diagnosis,
transthoracic needle biopsy, bronchoscopy biopsy
or washings - Clinical characteristics (age, smoking, history
of other cancer) and radiological characteristics
(size, location, edge characteristics
lobulation, spiculation, shagginess) - Multivariate analysis with logistic regression
- Predictors developed in 2/3 random sample and
validated in remaining 1/3 sample
35DiscriminationROC Curve Area (0.83, 0.80)
36Calibration Observed vs Predicted Probability
37Threats to ValidityMalignancy in SPN, 1
- Large number of SPNs (629)
- Referral population
- Clinically relevant SPNs (5-30 mm)
- Excludes low risk (lt 4mm)
- Excludes high risk (gt 30 mm)
- No calcification on thin section CT (benign)
- Cancer diagnosis
- Cohort study for outcomes after 2 years
- Some SPNs indeterminate classification
38Threats to ValidityMalignancy in SPN, 2
- Independent review single radiologist
- Swenson, Silverstein, Edell et al. SPN
Clinical Prediction vs Physicians, Mayo Clin
Proc, 1999 - Analysis
- Discrimination and calibration
- Independent sample for validation
- Distance to assess referral bias
- Included all SPNs
- Malignant vs (indeterminate benign)
- (Malignant indeterminate) vs benign
39Cohort Study Design
40Cohort Study Design
41Prospective Cohort Study Design
The Present
The Future
42Retrospective Cohort Study Design
The Past
The Present
43Cohort Study Example
- Long-term Survival of a Cohort of Community
Residents with Asthma - Silverstein, Reed, OConnell et al
- N Eng J Med 1994331 1537
44Long-term Survival of a Cohort of Community
Residents with Asthma
- Asthma mortality based on general US population
death certificates with asthma listed as
underlying cause of death - Residents of Rochester, MN with first asthma
diagnosis 1/1/1964-12/31/1983 - Explicit pre-defined criteria, review of all
medical records from all providers of care - Medical records and autopsy reports used to
classify deaths as due to asthma or other
conditions
45Long-term Survival of a Cohort of Community
Residents with Asthma
- 2499 patients with definite or probable asthma
- Mean duration follow-up 14 years (range 0-29
years) - 140 deaths in 32,605 person-years of follow-up
- Survival not significantly different form
expected - Survival worse in asthmatics with other lung
disease - 4 of deaths in persons with asthma were due to
asthma
46Observed vs. Expected Survival
47Survival in Asthma only vs Asthma and Other Lung
Disease
48Threats to ValidityLong-term Survival in Asthma,
1
- Large population-based cohort (2499)
- Yunginger, Reed OConnell, A Community based
study of the Epidemiology of asthma, Am Rev Resp
Dis, 1992 - Asthma diagnosis
- Beard, Yunginger, Reed et al, Interobserver
Variability in Medical Record Review An
Epidemiological Study of Asthma, J Clin Epid,
1992
49Threats to ValidityLong-term Survival in Asthma,
2
- Asthma deaths
- 14 years of follow-up
- Small number of deaths (140)
- Classification of deaths
- Hunt, Silverstein, Reed et al. Accuracy of Death
Certificate in a Population-Based Study of
asthmatic Patients, JAMA, 1993 - Review of all death certificates autopsy
reports - 13 out of state
50Case-Control Study Design
51Case-Control Study Design
The Past
The Present
52Case Control Study Example
- Risk Factors for Deep Vein Thrombosis and
Pulmonary Embolism - Heit, Silverstein, Mohr et al
- Arch Intern Med 2000 160 809-15
53Risk Factors for Deep Vein Thrombosis and
Pulmonary Embolism
- Reported risk factors vary in magnitude and
independence of each are unknown - Population based nested case-control study
- Cases 625 Olmsted County residents with first
lifetime VTE 1/1/1976-12/31/-1990 - Controls 625 Olmsted County residents without
VTE - Matched on age, sex, calendar year, and medical
record number (duration of observation)
54Risk Factors for Deep Vein Thrombosis and
Pulmonary Embolism
- Mean duration of medical records documentation
reviewed 34.7 years - 30 potential independent risk factors
- Conditional logistic regression to identify
independent risk factors
55Risk Factors for VTE
56Threats to ValidityRisk Factors for VTE, 1
- Population-based inception cohort
- Definite VTE confirmed by imaging
- Antemortum and postmortum ascertainment of VTE
- Silverstein, Heit, Mohr et al Trends in the
Incidence of DVT and PE, Arch Intern Med 1998 - Risk factors
- Large number of pre-specified risk factors
analyzed - Matching limits ability to analyze age and sex as
risk factors - Unable to analyze immobilization due to potential
diagnostic suspicion bias - Protein C had not been reported at time study was
initiated
57Threats to ValidityRisk Factors for VTE, 2
- Multivariate analysis
- Large number of potential confounders anticipated
to control for confounding - Bootstrap validation of multivariate analysis
- Age effect for risk due to varicose veins
- Treatment effect for risk due to cancer
58Study Components
- Abstract
- Specific Aims
- Background significance
- Preliminary Data
- Methods
- Human Subjects
- References
59Methods
- Design
- Setting
- Subjects
- Measurements
- Interventions
- Analysis Sample Size
- Interpretation Limitations
60Human Subjects
- Risks
- Benefits
- Risks in Relation to Benefits
- Privacy Confidentiality
- Informed Consent
- Women, children, minorities, priority populations
61IRB Review - Observational Studies
- Record review study (retrospective study)
- Observational study (prospective study)
62IRB ReviewRecord Review Studies
- Risks
- Criminal or civil liability
- Damaging employability, financial status or
reputation - Ethical Issues
- Privacy
- Confidentiality
- Issues fro investigators
- Data collection
- Data security
63Informed Consent
- Must be obtained, unless IRB approves waiver of
informed consent - Minimal risk
- Right and welfare of subjects not adversely
affected by waiver of informed consent - Research could not be practically carried out
without the waiver
64Category of IRB Review
- Exempt Studies
- Record review studies may be exempt form IRB
review - Determined by Chair of IRB
- Minimal risk
- Expedited Review
- Some IRBs allow expedited review for record
review studies - Determined by chair of IRB
- Publicly available data
- No personal identifiers and records can not be
linked to subject - Full IRB Review
65Informed Consent
- Protect human subjects/volunteers
- Ensure that study subjects understand benefits
and risks - Provide potential subjects with all the
information to reach a decision on whether or not
to participate in a research study
66Informed Consent Requirements
- Benefits
- Alternatives
- Confidentiality
- Compensation
- Medical treatment for injury
- Where to obtain further information
- Contact person
- Voluntary participation
67Review
68Objectives
- Describe 4 research designs
- Describe 3 threats to validity
- List advantages of retrospective studies
- List disadvantages to retrospective studies
- Understand requirements and processes for IRB
review of retrospective studies