Presentation is loading. Please wait.

Presentation is loading. Please wait.

Lecture 8: Prospective cohort studies: planning and execution (part 2)

Similar presentations


Presentation on theme: "Lecture 8: Prospective cohort studies: planning and execution (part 2)"— Presentation transcript:

1 Lecture 8: Prospective cohort studies: planning and execution (part 2)
Jeffrey E. Korte, PhD BMTRY 747: Foundations of Epidemiology II Department of Public Health Sciences Medical University of South Carolina Spring 2015

2 Follow-up of cohorts So you have assembled a cohort of people. How do you follow them over time to see who develops the disease?

3 Disease measurement Basic principle:
Disease ascertainment should be performed in the same way for exposed and unexposed participants If possible, disease ascertainment should be performed by people unaware of participants’ exposure status

4 Disease measurement Ascertainment without participant contact
Death certificates, medical records, hospital surveillance, reportable disease registry, etc. Ascertainment using patient contact Questionnaires, physical exams, lab tests, etc.

5 Disease measurement Mortality study
Death certificate/National Death Index Cause(s) of death Death certificate should only be a starting point Medical records Physician consultation Autopsy reports Next of kin

6 Disease measurement Diseases requiring hospitalization: local hospitals can be monitored Inexpensive Good for short follow-up periods (longer follow-up periods: cohort members move out of area, leading to BIAS) Population-based disease registries: similar information available

7 Disease measurement Physician records
Good if patients normally see primary health providers before referrals e.g. prepaid health plans in US Good for diseases generally seen by physician

8 Disease measurement Disadvantages of using existing records (hospital, physician, etc.): Not standardized What information is recorded? Differences in diagnostic criteria Some records incomplete Self-selection leads to BIAS Some patients (e.g. exposed) are less likely to seek care

9 Disease measurement Advantages of periodic study visits for cohort members Standardized follow-up, time dependent exposure measurement, detection of preclinical disease (intermediate endpoint) Non-standard tests for state-of-the art assessment Active ascertainment of all variables Alternative: periodic mailed questionnaires

10 Diagnostic criteria Determination of disease status must be the same for exposed and unexposed Diagnostic criteria should therefore be established before study begins

11 Diagnostic criteria Mortality study Diagnosis on death certificate
Some misclassification is inevitable Some exposures may increase likelihood of certain diagnoses, leading to BIAS e.g. workers known to have been exposed to asbestos: physician more likely to put down mesothelioma as cause of death

12 Diagnostic criteria Mortality study
Supplement death certificate with other info (medical records, autopsy report, next of kin) Clear criteria must be established a priori to determine which data sources have precedence Cause of death should always be assigned without knowledge of the exposure

13 Diagnostic criteria Hospital records
Quickest: accept discharge diagnosis from face page of medical record More accurate and complete: search all parts of medical record Some parts of medical record are definitive for certain diseases (e.g. pathology report for cancer; x-ray for fracture)

14 Diagnostic criteria Physical exams and testing
Must apply existing diagnostic criteria, or develop new criteria Medical committee can be assembled to develop and revise outcome definitions Example: next slide

15

16 Diagnostic criteria Subdivision by certainty of diagnosis
Useful if definitive criteria cannot be applied to all cases e.g. “definite”, “probable”, “suspect”, or “no” myocardial infarction

17 Diagnostic criteria Infectious diseases:
Clinical diagnosis must usually be confirmed by the isolation of the microorganism and/or the detection of antibodies (or a rise in antibodies) False positives and negatives depend on variations in lab technique, presence of a co-infection, inhibitors of the test being used, cross-reactivity in serologic tests, etc.

18 Diagnostic criteria Questionnaire Ask about condition of interest
Ask about symptoms that define condition of interest Ask about physician visits for symptoms or conditions, or physician diagnoses Confirm with physician

19 Diagnostic criteria Combine sources of information
e.g. questionnaire data leading to physician confirmation of condition e.g. signs, symptoms, and serologic evidence of infectious disease

20 Spectrum of disease Many diseases are not “present” or “absent”
Arbitrary thresholds can be used to categorize variables like blood pressure, osteoporosis, etc.

21 Spectrum of disease Disease subtypes May be etiologically distinct
Hemorrhagic stroke, ischemic stroke Low-grade cervical lesions, high-grade/cancerous cervical lesions Lumping everything together may obscure risk factors (disease misclassification)

22 Spectrum of disease Subclinical disease
Infectious agents (non-symptomatic) Pre-cancerous or cancerous processes Slow-growing tumors Subclinical atherosclerosis, coronary heart disease, etc. Risk factors may differ for clinical vs. sub-clinical disease

23 Non-participation Not everyone will agree to join study
Non-participators will differ from participators Usually, individuals who join study are healthier and have fewer risk factors This affects generalizability of results This may also bias relative risk estimates

24 Non-participation Bias of risk estimates, relative risk estimates
More likely when proportion of nonparticipants increases More likely when nonparticipants are more different from participants Bias also depends on which measure of relative risk is used (examples on next few slides)

25 Non-participation Participation varies by exposure status, but not by disease status (likely for cohort studies) Risk ratio, rate ratio, and odds ratio are all unbiased Example: study of smoking and CHD: higher proportion of nonsmokers participate, but otherwise participants and nonparticipants are equally likely to develop CHD

26 Non-participation Participation varies by exposure status and by some other risk factor for disease This can be controlled in analysis (the other risk factor may function as a confounder)

27 Non-participation Participation varies by disease status, but not by exposure status (likely for case-control studies?) Risk ratio and rate ratio are biased, but odds ratio is not For rare diseases, magnitude of bias is small e.g. CHD risk is 10% in smokers, 5% in non-smokers; 80% of people destined to develop CHD participate vs. 40% of others: risk ratio changes from 2.00 to 1.91; odds ratio does not change

28 Non-participation Participation varies by some combination of disease status and exposure status Risk ratio, rate ratio, and odds ratio can be strongly biased e.g. participation rates are 90% for nonsmokers; 50% for smokers destined to get CHD; 90% for other smokers: risk ratio changes from 2.00 to 1.16; odds ratio changes from 2.11 to 1.17

29 Non-participation Participation varies by some combination of disease status and exposure status This can be bad! (see previous slide) This can occur if participants are aware of the study hypothesis This can sometimes be ameliorated by control for confounding

30 Follow-up Current US populations are highly mobile
Ask each participant for at least one contact who should always know where they are Name, address, phone number Can track people through DMV, credit bureaus Electronic records, social security numbers Post office can provide forwarding address

31 Follow-up Mortality study: National Death Index
Computer index of all deaths since 1979 Identification: name, birthdate, social security number Information provided: date of death, state where death occurred, death certificate number of matching individuals Investigator can then acquire death certificate from state where death occurred

32 Follow-up Mortality study: National Death Index
Quality has been shown to be very high Successful matching demonstrated for 98.4% of known deaths in a cohort of men (Wentworth et al 1983) and 96.5% of known deaths in a cohort of women (Stampfer et al 1984) If no social security number: matching better for males vs. females, and whites vs. nonwhites

33 Follow-up Mortality study: National Death Index
Middle initial helps matching significantly Nicknames are problematic False-positive matches: computer algorithms or manual review of results can identify which (if any) is correct match

34 Follow-up Outcomes other than mortality
Loss to follow-up must be held to minimum Use study visits, periodic phone contact, etc. to maintain cohort Loss to follow-up may produce bias similar to nonparticipation Loss to follow-up may be more likely (than nonparticipation) to be related to disease status Investigator can assess loss to follow-up relative to exposure status (differential loss to follow-up may produce bias)

35 Last word: avoid bias! Assess exposure and outcome the same for everyone Definition of exposures Definition of disease(s) Minimize loss to follow-up Maintain scheduled study visits


Download ppt "Lecture 8: Prospective cohort studies: planning and execution (part 2)"

Similar presentations


Ads by Google