Module 1 (+ Module 3-1) Flashcards
Epidemiology
Study of frequency/occurrence of dis-ease in populations
- differences/similarities in frequency between populations helps identify causes
Process of epidemiology
1) describe a population
2) count total population
3) count number of cases of dis-ease
Formula:
E = N/D/T
Need for age standardisation/adjustment
Can only compare ‘like with like’ (confounding)
Numerical values
- convert into categorical measures
- use mean or median level of outcome
Cohort study
Allocation into EG and CG: measured exposures
Measurement of outcome: followed over a period of time
Cross-sectional study
Allocation into EG and CG: measured exposures
Measurement of outcomes: at the same time as measurement of exposures
Incidence
- Outcome events counted forward from starting point, over a period of time
- time included in calculation
- rate
- preferred method if easy to observe events
- depends ONLY on no. Events during specified time period
Incidence type of study and data
ONLY cohort, ONLY categorical
Prevalence
- no. People with disease counted at ONE POINT in time
- time not included in calculation but when counted is mentioned
- state
- depends on incidence, deaths and cures
Prevalence types of study and data
Both cross-sectional and cohort (one point during the study)
Categorical and numerical
Change in prevalence
Difference between prevalence measured at two points in time
Retrospective info
If events come and go frequently
1) use incidence over a retrospective time period to group people into categories
2) total number of episodes for each group is outcome (no. Episodes for each individual person is not used in calculation)
Measures prevalence because deaths/cures are lost
Cohort/cross-sectional study depending on when exposures are measured in relation to outcomes
If most people died rapidly or were cured
High incidence disease has low prevalence
If few people died or were cured
Low incidence disease has high prevalence
Incidence strengths
- determined only by disease risk - clean measure
- includes N, D AND time - more info
Incidence weaknesses
- can be difficult to measure
- must be observed over time
Prevalence strengths
Relatively easy to measure
Prevalence weaknesses
- determined by incidence, cure rate and death rate - dirty measure
- doesn’t include time - less info
Ecological studies
Populations allocated to EG and CG (exposures are an average of a group of people)
Individual participant studies
Individuals are allocated to EG and CG
Reasons against RCT
Unethical, impractical
Blinding
Double-blind: neither participants nor investigators know which intervention was given to which participant
Single-blind: participants don’t know which intervention was given to which participant but investigators do
Only really works for experimental studies
Estimates of effect
Comparisons of disease occurrence in EG and CG
- gives idea of size of effect of study exposure on disease outcome
RR description
The risk of x in A is n times higher/lower than in B
- use RRI if higher, RRR if lower
Relative mean
Disease occurrence measures are calculated as averages and RR is comparison between the means
RR no-effect value
RR = 1.0
- closer RR is to 1.0 smaller the difference
RRR
= (1-RR) x 100%
EGO < CGO
RR < 1.0
RRI
= (RR - 1) x 100%
EGO > CGO
RR > 1.0
RR range and units
RR > 0
No units
RD description
Observational:
- there are y fewer/more x per n A than per n B
Experimental:
- if n people are treated rather than not treated, there will be y fewer/more events/occurrences
RD no-effect value
RD = 0 units
- closer RD is to 0 less effective drug/difference in outcome between EG and CG
ARR
EGO < CGO
RD < 0
ARI
EGO > CGO
RD > 0
RD range and units
-infinity < RD < infinity
Same units as EGO and CGO - events per n people per T
RR vs RD
- Decisions should be based on RD as this depends on original risk and so do benefits of treatments whereas RR doesn’t (only a ratio)
- RD gives more info (CGO must be known) - groups can have same RR but diff. RD
- beware of large RR but small RD
Occasional study
Don’t have comparison groups though implicitly present by subdividing EG by age, gender etc.
Error
Incorrect EGO/CGO due to:
- wrong people recruited
- right people put in wrong category (EG/CG)
Non-random error
Due to poor study design, processes or measurement
Valid study
Only has small amount of random and non-random error
RAMBOM
Recruitment, allocation (+-adjustment in analyses), maintenance, blind or objective measurement
Recruitment
- representative?
- sufficient info about process to apply?
- recruitment error aka external validity error
Response rate
No. Who took part / no. Eligible
- if < ~70%, could cause significant recruitment error
Allocation methods
By measurement/observation
Random allocation
Types of allocation error
Measurement error
Confounding
Allocation measurement error
- exposures measured incorrectly
- participants tell the truth?
Allocation measurement error solution
Use well-designed, validated questionnaires or biological tests measuring chemicals
Confounding
When exposure mixed with another factor (confounded) that is also associated with outcome
- EG and CG similar at the beginning of the study?
Confounding solution
- adjustment
- RCT
- concealment of allocation
Baseline comparison
Checking for differences between EG & CG at start of study
Selection bias
Causes confounding
- equivalent to two separate/overlapping triangles
Maintenance error
- remain in allocated groups by maintaining exposure and not being exposed to other factors?
- lost to follow-up? More lost from EG that CG, vice versa? (Not unusual to lose 30%)
Maintenance error solution
- blind studies - minimise difference in degree of error between EG and CG
- checking periodically especially in long term observational studies
BOM
1) objectivity
2) blind - helps reduce effect/error from subjective measures
Analyses
Confounding adjusted for?
- stratified analysis
- adjusted analysis (standardisation)
Stratified analysis
Diving participants into ‘strata’ and analysing data as if they were two sub-studies
- similar results: strata combined
- different result: reported separately
Confounding in ecological studies
Common but difficult to measure and adjust for
Crude death rate
No. Deaths from disease / size of pop.
Age specific death rate
No. Deaths from disease in age group / no. People in that age group
Expected deaths
Age specific death rate x no. People in age group in standard pop.
Age-standardised death rate
Sum of expected deaths / standard population
Age standardisation
Process of converting different age structures in each population into one (standard) population age structure then working out death rates
Random errors
- Occur due to chance
- present in every measurement in every study
Regression to(wards) the mean
Repeating measurements/studies with extreme results (often chance events) usually give less extreme results
- more measurements = less randomness = closer to middle
Random error solutions
- increase study size (allocation)
- increase no. Times a factor is measured (measurement, biological)
- objective measuring instrument (measurement)
Types of random error
Sampling, measurement/assessment, inherent in biological phenomena, allocation
Random sampling error
Unrealistic to study entire pop. Thus each sample from pop. Will be different each time
- results are ‘estimates’
Random allocation error
Differences due to chance in RCTs esp. in smaller studies
Measures of random error
Unrealistic to estimate all error thus measures generally underestimate total random error
- confidence intervals
(- P-values)
Confidence intervals
Measure amount of random error in estimates of EGO, CGO, RR, RD in whole pop. When only one study has been done
- describes range of results likely to include true result in whole pop.
95% CI acceptable def.
About 95% chance that true value in pop. Lies within 95% CI
95% CI accurate def.
In 100 identical studies using samples from same pop. 95/100 of 95% CIs will include true value for pop.
CI can be calculated for
Both categorical and numerical variables
Point estimates
Estimated value from study
- represented by square
CI interpretation
Width: wider interval, more random error in measure
- estimates degree of uncertainty
- upper/lower confidence limits
Statistical significance
No overlap between CIs of EGO and CGO
CI of RD/RR doesn’t cross no-effect line
- reasonable to assume EGO and CGO are truly different in underlying pop.
Not statistically significant
Wider CIs => large overlap between CIs of EGO and CGO
CI of RD/RR crosses no-effect line
- study unable to determine if EGO truly different from CGO (actually no difference or just too much random error)
Clinical/practical significance
If clinician would make similar clinical decision on whether the true result was near one end of CI or the other of statistically significant results
Meta-analyses
Mathematical combo of results (usually multiple studies that are too small) which generates summary of estimate of effect
- alternative to conducting one large study
Study classification by allocation
1) experimental: allocated by investigators
2) observational: allocated by measurement
Study classification by measurement of outcomes
1) longitudinal: followed over time (allocated EITHER randomly or by measurement)
2) cross-sectional: outcomes measured at same time as exposures (allocated by measurement)
Recall bias
Common in retrospective studies
Systematic reviews
1) review literature systematically (rigorously) to find all relevant studies
2) assess quality of studies and only keep good ones
3) combine results in meta-analysis if similar enough
- studies in review must be valid for valid review
Determinants for individuals
- any event, characteristic or other definable entity that brings about a change for better or worse in health
- may vary at different life stages
Determinants for groups
- concepts similar as for individuals but nature of determinants is often different
- includes characteristics of population itself + context in which it exists
- population health is greater than the sum of its parts
Downstream interventions
at micro (proximal) level
Proximal
Near to change in its health status
- any that’s readily and directly associated with change in health status
Upstream interventions
At macro (distal) level
Distal
Distant in time and/or place from change in health status
Effects of discrimination/inequities on health
Indirect: biased/limited healthcare
Direct: mental health struggles
Importance of considering social inequities
- urge to actions
- indicate possibilities to improve health conditions for groups at particular risk
- otherwise both ethically unsound and inefficient in a health development perspective
Dahlgren and whitehead model levels
1) individual level (micro)
- non-modifiable/fixed determinants
- individual lifestyle factors and attitude
2) community level (meso)
- social/community networks
- living and working conditions
3) major structural environment level (macro)
Single gene disorders
Rare among pop.
Polygenic inheritance
Influences likelihood of offspring developing a disease
Individual lifestyle factors and attitude
= environment
- certain degree of choice as an individual - impacts health
- ability to change behaviour(s) may vary by social group
Social capital
Value of social networks that facilitate bonds between similar groups of people
- inclusive environ. Diverse backgrounds
- mutual support
- strengthen defence against health hazards
- civic participation, volunteerism, supportive communities
Community level
Attitudes/behaviours of families, friends, people living/working in local community influence perception of ‘normative’ behaviours
Major structural environment level
Require political actions at national or international levels
- physical, built, cultural, biological, political environs and ecosystem
Caution about level of Dahlgren and whitehead model
Permeability between factors - no arch operates in isolation from others
- synergetic effects = more effective
- action at each can impact others - offset
Four capitals
Natural, human, social, financial/physical
Structure
Social and physical environmental conditions/patterns (social determinants) that influence choices and opportunities available
Agency
Sociological concept of the capacity of an individual to act independently and make free choices
RCT uses
investigating effects of interventions (therapies, treatments)
cohort uses
causal associations
cross-sectional uses
measure disease prevalence
ecologial uses
- prevalence in different populations
- when majority of some pop exposed but not others
- rare outcomes
RCT strengths
minimise confounding
cohort strengths
- ethical
- cheaper than RCT
- clear time sequence
- avoid recall bias
- participants more likely to be representative of general pop than RCT
cross-sectional strengths
- cheaper + quicker than cohort and RCT
- no maintenance error
ecological strengths
- use already collected data => cheaper + quicker
- large size => low random error
RCT weaknesses
- ethical limitations
- logistically difficult
- expensive
- small => too much random error
- participants often not representative of general pop (motivated volunteers)
- maintenance error
- random allocation error
cohort weaknesses
- confounding
- maintenance error
- not good for investigating interventions
cross-sectional weaknesses
- uncertain time sequence => reverse causality (not good for causal associations)
- confounding
- not good for investigating interventions
ecological weaknesses
- confounding common, difficult to measure and adjust for
- measurement error