HSCI 1130 Exam2 - Bulk Test

Descriptive statistics

Do: Describes the data / Don't: Present any conclusions

Most common way to describe data

Sample Size (n) / Ratios, Binary variables (M:F) / Continuous variables: Mean, Median, Mode, Range / Distribution (unimodal, bimodal, standard-bell curve)

Sample size

# of people in study sample / Larger sample, more powerful the study

Reasons why large sample size is better

Better represents all variations of population / Less affected by outliers / Randomization works better / Higher power / Better at finding real and smaller effect as opposed to by chance (power)

How to select proper sample

Select representative sample via random sampling

Bias when sample can't be generalized

Selection / Sampling bias

Other types of bias

Measurement bias / Inconsistent / inaccurate measurements

Mean

Affected by outliers / The average / Doesn't accurately represent data

Median

Not affected by outliers / Accurate representation of data / Middle value

Mode

Most represented number

Range

Min - Max value (DO NOT SUBTRACT)

Normal distribution

68-95-99.7 rule

Correlation

Relationship between two variables changing in relation with one another

Relationship

Perfect / High / Low +/- correlation and No correlation

Causal relationship conditions

Must be correlation between cause/effect / Cause must be before effect / No alternative explanation or confounders

Determining causation

Find if correlation exists then search for evidence of causation

Importance of distinguishing correlation and causation

Can predict / modify / understand health outcomes

Confounding

3rd factor distorting relationship between exposure and outcome

Bradford Hill Criteria (SPECTACSB) - Strength

Strong association (large RR/OR) is more likely causal

Bradford Hill Criteria (SPECTACSB) - Plausibility

There is a biologically plausible mechanism for how the exposure causes the outcome

Bradford Hill Criteria (SPECTACSB) - Experiment

Experimental evidence supports it; removing or reducing the exposure reduces the disease

Bradford Hill Criteria (SPECTACSB) - Consistency

The same association is seen repeatedly in different studies, places, and populations

Bradford Hill Criteria (SPECTACSB) - Temporality

The cause must come before the effect; exposure before outcome

Bradford Hill Criteria (SPECTACSB) - Analogy

A similar exposure causes a similar disease

Bradford Hill Criteria (SPECTACSB) - Coherence

The association fits with existing knowledge about the natural history and biology of the disease

Bradford Hill Criteria (SPECTACSB) - Specificity

One specific exposure leads to one specific disease / outcome

Bradford Hill Criteria (SPECTACSB) - Biological Gradient

Dose-response — more exposure, higher risk / more disease

Diseases are

Not random in people / place / time

Risk factors

Characteristics in people that increase / decrease disease

Characteristics

Place (Geographic location, humidity and environment) / Time (season, winter) / People (Inuit lowest diabetes risk 3.2%, First Nation on reserve highest 15.9%)

Exposure

Determinant of health

Outcome

Change in health status

Study of relationship between exposures and outcomes

Prevents / modifies outcome of disease / Improve public health

Descriptive study

Describe frequency and pattern of data in people, place and time / No analysis / Can't establish correlation / Useful for making hypothesis

Analytical study

Analyze correlation between exposure / outcome / Observational: Observe participants and record info about exposure / outcome, no control over exposure / outcome / confounders / Experimental: Strongest study, Randomized trials, Can control exposure

Target population

Population sharing specific characteristics we are interested in and want to draw conclusions about

Study sample

Sample from target population

Selection bias

Sample has different characteristics than target population, can't generalize

Measurement bias

Inaccurate / Inconsistent measurements / classification of exposure / outcome

Bias

Factor or trend in collection / analysis of data leading to incorrect and misleading results

Sources of data

Surveillance / Medical Files / Questionnaire / Medical Tests

Aggregate Data

Population level data / Low quality

Individual level data

High quality / Can measure outcome and exposure at same time

Data analysis

After measuring data, correlation of exposure / outcome measured / Statistical inference made

Ecological studies

Study existing aggregate data / Cannot measure correlation between exposure / outcome / Vulnerable to ecological fallacy

Ecological fallacy

Assuming every individual had same (average) exposure

Ecological study strengths and weaknesses

Fast and inexpensive (existing data) / Ecological fallacy

Cross sectional studies (prevalence studies)

Exposure / outcome measured and collected at same time / Quick / inexpensive / Can't establish temporal sequence of exposure / outcome / Can't establish causation / Can't assess risk of disease / Calculated by PR (Prevalence ratio) / Prev of Exposed / unexposed / 1 to infinity = + association / 0 to 1 = - association / Value near 1 = weak association

Case control studies

Identification of cases (experience outcome) / Appropriate controls identified / People with similar characteristics without the outcome / Exposure status determined in both case + control and compared

Case

Must have outcome (disease of interest) / Diagnosed with same disease definition and diagnostics criteria

Controls

Must be at risk but free of disease

Matching

Case-control matching can reduce known confounders / Individual (age, sex) / Group (Similar proportions of confounders)

Max # of controls

4 before it gets more expensive with no benefit

Potential exposure

Usual (Genetic, age, sex) / Likely (implicated by informed reasoning / played a part in similar outcomes)

Potential exposure + Critical thinking

Critical thinking helps select exposure / Explain relationship between exposure / outcome / Plan potential ideas for intervention

Measure of CC study

OR Odds ratio / Association similar to PR (0-1-infinity)

Null Hypothesis

Base hypothesis, X will do nothing

Alternate Hypothesis

A difference or relationship does exist between variables / Only after evidence rejects null hypothesis

Measure of statistical significance (ss)

P-value: Measure of probability by chance 0-1 / p<0.05 chance of being random is <5% so statistically significant

CC studies strengths and weaknesses

Ideal for rare diseases / multiple potential exposures could be associated with one outcome / Risk of recall bias / Can't calculate incidence, prevalence rate

Cohort studies

Measure exposure first and measure outcome over time / Strongest of all observational study

Study design of cohort study

Exposed / Unexposed free of outcome / Both followed over time and recorded / Compared at end

Prospective Cohort study

Participants from present study to future / Can take long time

Retrospective cohort study

Use historical data / Quality might be worse than the data from that time

Risk ratio / Relative Risk (RR)

Compares risk (cumulative incidence) of outcome in 2 groups over a period of time / Risk of outcome in exposed / Risk of outcome of unexposed / RR <1 negative association / RR >1 positive association / RR =1 no association / result is X times as likely for Y factor to be outcome

Rate ratio / Incidence Rate Ratio (IRR)

Compares incidence rate that incorporate person-time in denominator

Person-time at risk

Length of time each person in study is at risk:

# of new events / sum of time each person was at risk for event *1000

Background risk

Risk in unexposed group

Excess risk

Remaining after removing background risk / Risk exposed - Risk non exposed

Selection bias in cohort study

Poor choice of cohort / Lost follow up or drop outs

Strengths of cohort study

Can be sure exposure before outcome / Stronger causality evidence / Person-time incidence rate (accurate risk estimates)

Weaknesses of cohort study

Requires funding and time / Susceptible to loss-follow-up / Impossible to count all confounders

CI value

Range of values, if null value is in range then it is not statistically significant

If CI includes 1 then it is not statistically significant

Bigger ranges are usually less significant / powerful