Evaluating Evidence Flashcards
What are the complexity of idenfitying diet-disease relationships factors?
- Multiple factors may contribute to disease risk
- Possible confounding factors in diet-disease relationships
- Long latenct period (time btw exposure & disease appearance)
- For prospective studies & RCT, need large sample size and long follow-up
- Difficult to assess nutrient intake
What
What are multiple factors contribute to disease risk?
- genetics
- alchol
- smoking
- age
- phycial exercise
- diet
- medication
What are possible confounding factos in diet-disease relationships?
- another explanation for the exposure and outcome
E.g., Fruit & Veg Intake & Lung Cancer - observational studies: lung cancer patients report lower intake of fruits & veg than controls
- However, smokers generally consume less fruit and veg than non-smokers, and smoking known to inc. risk of lung cancer
- satiety; consuming less veg & fruits
- have non-smokers in the cohort
What does long latency period (time between exposure & disease appearance)?
- take years or even decades to develop, making it difficult to pinpoint a specific dietary cause
- difficult to established direct casual relationships between diet and disease
E.g., Does vitamin D reduce risk of colon cancer? - 300 older women, which are already at higher risk b/c of age, menopause
- vitamin D and placebo
- 3-years follow-up and outcome is colon cancer; follow-up is too short, not feasible, attendance
What does prospective studies & RCT, need large sample size and long follow-up?
Larger Sample Size Required
E.g., Concerend about the possible relationships between cell phone use and head & neck cancer:
- a prospective chort study
- 10, 000 subjects, all determined to be ‘cancer-free’ at baseline
- 5,000 are ‘regular’ cell phone users
- 5,000 claim never to have used one and promise not to until after the study is completed
- 5 year follow-up
- At the end of the study: 4 head and neck cancer cases in the exposed group and 2 in the non-exposed group; not enough cases
Statisical significance
- not likely due to chance alone
- result is likely real and not just duet to chance
- measue using p-value (usually less then 0.05 means significant)
Clinical significance
- results are actually meaningful/differencev in real life
- change is real, it might be too small to matter
Why is it difficult to assess nutrient intake?
- Diet assessment tools
- Food frequency questionaire (i.e., how often do you eat potatoes?); recall bias
- 24 hr recal (over the last 24 hrs what did you eat?); reflect the last day
- Diet records - Potential for misreporting or forgetting foods
- Require accurate nutrient database of food composition
- translate food intake into nutrient composition
How do we evaluate the evidence for a diet-disease relationships?
- Consider the toatality of the evidence:
- ecologial studies, cross control, case control, chort, expierments
- diver populations - Is the evidence convincing?
- consistent evidence from good quality studies - Does the evidence show causation?
- consider Hill criteria for causation
How do we determine whether a dietary component or pattern increases/decreases risk of chronic disease?
E.g., Red Meat Intake and Colon Cancer
Correlation between Meat Consumption and Colon Cancer Rates in Differenct Countries
- lowest #Japan and Highest #USA
Cross Sectional
exposure & outcome measured at same time
Case-control
start with outcome; look ‘backwards’ for exposure
- study in which people with a disease or other condition (cases) are compared to a those without the disease or condition, who are matched for important variables such as age, sex, area of residence
Cohort
start with exposure; follow over time for outcome
- study in which subjects who presently have a certain condition and/or recieved a particular treatment are followed over time and compared with another group are not affected by the condition under investigation
- more meaningful
Experimental Study
- independent variable manipulated (or controlled) by researcher
- randomization controls for confounding effects
- can show ‘causation’
- Feasibility issues:
- “Placebo” group still consumes nutrient from foods; can not remove other nutrient
- Compliance; can not control if they are taking it
What distinguishes a TRUE experimental study from a QUASI experimental study?
A) True experimental studies use a control group
B) Participants are randomly assigned to groups in a true experimental study
C) The control group is given a placebo in true experimental studies
D) Participants are blinded to their treatment in true experimental studies
B) Participants are randomly assigned to groups in a true experimental study
Double Blind
Neither subject nor researcher aware of group assignment
- minimizes potential for bias and confounding
Placebo Controlled
Neither subject nor researcher aware of group assignment
- minimizes potential for bias and confounding
Randomized Controlled Trail (RCT)
Neither subject nor researcher aware of group assignment
- minimizes potential for bias and confounding
- assigned randomly to an ‘active treatment’ group or to a ‘control’ group
Hiaerchy of evidence for clinical descisions
What are consistent evidence from good quality studies?
- study designs
- appropriate choice of measurements tools
- sample size
- statistics
- control for confounding or other factors
What are consider Hill criteria for causation?
- consistency
- strength of relationships; how big is the difference? - 2 fold to consider it’s strong, not practical b/c we do estimates
- dose response - be aware of sigmoid relationships between nutrition * disease
- biological plausibilty
- temporality - dependence on time
Sigmoid Relationships & Threshold effects
- linear relation
- desribes a relatioships where small changes at first have little effect, then suddenly make a big impact and final level off again
- All consuming sufficient amount of nutrient, further increase in intake shows no effect
- All consuming insufficient amounts, no apparent relationship with outcome
S-curve (Sigmoid)
- Slow start - at first, changes happen slowly
- Rapid growth - after certain point small changes cause a big effect
- Leveling off - eventually, the effect slows down and stabilizes
Which observational study design(s) are not able to show temporality?
A) Cross sectional
B) Case control
C) Cohort
D) Cross sectional & case control
E) All of the above (none of the observational study designs can show temporality)
D) Cross sectional & case control
What is the criteria to judging the evidence?
WCRF/AICR Third Expert Report:
Convincing: evidence from ≥ 2 study types, ≥ 2 cohort studies, consistency of findings, biological plausibility, good quality studies, experimental evidence (human or animal)
Probable: evidence from ≥ 2 cohort or ≥ 5 case control studies, consistency, biological plausibility, good quality studies
Limited – suggestive: evidence from ≥ 2 cohort or ≥ 5 case control studies, biological plausibility
Limited- no conclusion: insufficient evidence
E.g., Judging the evidence
Processed Meat & Colorectal Cancer: The evidence
Positive correlation between processed meat & colorectal cancer in
- 12 out of 18 cohort studies
- 6 out of 9 case-control studies
Meta-analysis of 10 cohort studies showed
dose response and
- relative risk: probability of exposed group vs. not exposed/low exposure
- Relative risk (RR) for 50 g/day increase = 1.18, 95% CI = 1.10−1.28
“Moderate” mechanistic evidence from experimental animal studies
Forest Plots
**Each “box” **shows the RR (or mean diff) for an individual study
- size of the black box reflecs the weight of the study in the meta-analysis (and, usually size)
Middle line: “no effect” line - study’s result crosses this line, means the effect is unclear
Horizontal line: the line extending from the box shows the 95% CI
- shorter lines mean more precise results, longer lines mean more uncertainty
Diamond: weighted mean of studies
- if it does not overlap with RR (or OR) if 1.0 (or a mean diff of ), it means the effect is statically significant
- fully on one side of the middle line, the effect is likely real
- right side = positive correlation
- left side = negative correlation
Relative Risk
Likelihood that outcome occurs in exposed
group compared to unexposed group
RR = Risk of outcome among exposed/Risk of outcome in unexposed
Null hypothesis: RR = 1
RR>1 positive association
Statistial Significance of RR
statisically significant when the confidence interval does NOT include 1.0
How certain do we need to be in making recommendations?
Decision plot
- visualize the evidence; enough certainity to take action
- cut-plane is a threshold line that separates when we take action from when we do not
Implemented for drugs
- cut-plane is high b/c need very strong evidence for approval
- Below line no approval
- Above line approval
**Implemented for nutrients **
- cut-plane is lower b/c nutrients usuallt have lower risks
Explanations for associations
- cause-effect
- chance or random error
- bias or systematic error
- effect-cause
- confounding
What if the evidence is lacking, Fat intake and colon cancer?
Ecological study: Incidence of colon cancer strongly correlated with per capita disappearance of animal fat and meat
Case control studies: weak positive association between colon cancer and fat intake (and total energy intake)
- limited by memory, recall bias
Prospective studies: most do not show fat intake (independent of energy intake) associated with colon cancer
No randomized trials
Confounding - other diet/lifestyle factors contribute to risk of colon cancer
