Please enable JavaScript.

Coggle requires JavaScript to display documents.

Epidemeology - Coggle Diagram

- - - - Exposed N
        
        Disease +ve
        
        Disease -ve
      - Not Exposed N
        
        Disease +ve
        
        Disease -ve
  - - - Ratio of risks/rates: how much more likely is one group to get a disease than another?
        
        estimate of the relative effect of exposure
        
        The associative strength of risk factors
        
        Relative risk/rate of a disease and risk factor
        
        Ratios - Relative rate, relative risk and odds ratio
        
        Measure the strength of the relationship between exposure and disease
        
        Ratio <1 = exposure associated with decerased risk of disease
        
        Ratio >1 = exposure associated with increased risk of disease
      - Differences between rates/risks: how much greater is the frequency of disease in one group vs another? (see later in the module
        
        estimate of the absolute effect of exposure
        
        Excess - attribute risk
  - - - Leads to incorrecrestimates of measures of association/effect
      - Can be above or below the true value depending on the estimates
      - Many types of bias exist, most falling into two groups
        
        Selection bias- Systematic difference in characteristic between participants and the population from which they are selected or between groups compared in the study
        
        Explained by: non-participation or attrition (loss of follow up)
        
        Can be introduced by the researcher/observer - responeder (social desirabilty or error in recall) - instrument (e.g. measuring device used)
        
        Differential vs non-differential bias:
        
        Random (non-differential) - misclassification not associated with disease or exposure status
        
        Systematic (differential) - misclassification associated with disease or exposure status
        
        Information (measurement) bias - systematic differences in the classificationof exposure or outcome of study participants
        
        Explained by: measurement error or misclassification
- - - - Attack rate
        
        How many at risk people get infected - cumulative incidence
      - Secondary attak rate
        
        proportion of contacts with primary case that become infected
      - force of infection
        
        rate at which succeptable get infected - can be identified by case notification or serological data
      - reproductive number
        
        Numberf of new infections caused by a single infected person in a population OR average number of successful offspring that a parasite is cabable of producing
  - - - can be high low or seasonal incidence
      - pathogen and hostat equilibrium. Relatively constant number/proportion of infected hosts in population
      - Terms: holoendemic, hyperendemic, mesoendemic, hypoendemic
  - - - The avg number of secondary infections produced when one infected individuals is introduced into a real population with a mix of succeptables and immunes
      - Varies over time as immunity changes in the population
      - R/Re Describe the dynamics
        
        Re > 1 = each infected N infects more than 1 succeptable - epidemic infection - proloferation of of the pathogen (is it successful)
        
        Re = 1 = Each infected N infects roughly 1 individual - endemic infection (equilibrium) - independent of the prevelence
        
        Re < 1 = each infected N infects less than 1 individual - infection fades from the population
    - - Pathogenicity - measure of the exposed/infected. Similar to exposure specific incidence rate
      - Virulence - proportion of infected hosts who develop severe disease. If severity = death rate then this is equivilent to the case fatality rate
      - Case Fatality Rate - mortality rate per number of infected hosts - the proportion of infected hosts that die
    - - Who has been identified?
        
        N who develop clinical illness
        
        Sought medical care
        
        survived?
        
        were detected by screening programme
        
        how sensitive or specific is the diagnostic test?
      - Are N likely to be asmptomatic?
        
        UK swine flu 2009 - 30-70% of infected N actually sought care
        
        How do these change over time?
  - - - In a closed population N = X + Y + Z
      - X - succeptables
      - Y - infected
      - X - Resident
      - Beta - probablilty of transmission per contact
      - I greaca (latin Y) - Recovery
- - - - wide range of data
      - resource efficient but less accurate
    - - specific purpose, e.g. covid case finding and quarantine
      - resouce intensive but more accurate
    - - The Numerators
        
        The numbers:
        
        health behaviour
        
        risk factors/behaviours
        
        health outcomes
        
        disease onset
        
        All events must be clearly defined
        
        Health conditions by diagnosis
        
        various other health information and methods also exist
        
        How much do events tell us?
        
        Rates of morbidity in a given population (typically a population of morbid N)
      - The Denominators - At-risk populations:
        
        Population Definition: all individuals who share a common characteristic
  - - - health and social care planning
      - Long term changes in population health
      - greater relevance for chronic illnesses
  - - - This requires follow up at specified time to confirm and alter expected morbidity rates
      - Issues with incidence proportion:
        
        May have some missing data - if N do not respond, drop out or die then that data cannot be used in the follow up, you will underestimate the numerator
        
        May have assumed all N participated - this will mean you overestimate the denominator
        
        It is important to note and record when certain individuals stopped submitting feedback and using that data relevant to the time frame they were involved in the research - This calculates the incidence rate
  - - - person-days, person-months, person-years. 10 events in 18 months = .56 events per person-month
      - important to consider which is more effective for your form of research: 1 person for 10 years or 10 people for 1 year
  - - - Find target population: all individuals with the condition of interest
      - Find study population: the population a sample can be drawn from
      - Find study sample: those who provide data
    - - data not collected for the research question - can it be directly applied to the research intention
      - Bias: coverage, opt in and out, reporting
      - Coding framework: is it consistent?
      - Data Quality: missing data, misclassification bias
    - - Large data pool - can be applied generally
      - transparent and repeatable methodology if reported
      - enables research otherwise unavailable - e.g. too expensive for a priori
      - You must know how the data used was collected originally
- - - - lower respiritory tract infections - 3 million
      - Diarrhoeal diseases - 1.4 million
      - TB - 1.3 million
      - HIV/AIDS - 1 million
      - Malaria - .45 million
  - - - Host: incubation period - has no signs or symptoms of disease until after the infectious period starts
    - - Host: symptomatic period - sick and seeking care. Typically the first time the host realises they have been infected
      - When both stages end the host is recovered and no longer sick or infectious. Failure to overcome the disease may otherwise lead to death
      - thus many measures of disease occurance only observe the occurance of individuals in this stage
      - Key Question: To what extent can asymptomatic people spread infection?
    - - genetics - both host and pathogen
      - dosage
      - often not independent and other factors are likely involved (including alcohol and drug consumption
      - however, people tend to fit within general ranges and period lengths tend to correlated in some way
        
        short latency periods will often cause short incubation periods
  - - - DIfferent life cycles between pathogens impose structure/constrains on transmission
        
        Microparasites - viruses, bacteria, protozoa, (malaria and schistosomiasis are notable exceptions
        
        short generation time at high replication rates
        
        Simpler life cycle - asexual reproduction with fewer obligate hosts
        
        Macroparasites - helminths and arthropods
        
        longer generation time and lower replication rate
        
        complex life cycles - sexual reproduction often required, many obligate hosts often required
      - Host Characteristics
        
        individuals - humans, aniamls, plants
        
        cells within an individual - target cells, immune cells
        
        cluster of individuals - lakes, pens, herds, households, cities
        
        Host attributes affect transmitability and can vary within a population
      - Types of Transmission:
        
        Direct: requires contect type
        
        sexual
        
        dropplets (sneezing)
        
        transplacental
        
        Contact may not be physical but proximal
        
        Dynamics depend on population contect structure and biology of agent - type of contact (see above) and mixing among hosts (spacial and/or social organisation
        
        indirect:
        
        Environment - waterborne, airborne, foodborne
        
        shared exposure to single external source (see above) and intemittent contact with contaminants
        
        Vector - biological, mechanical
        
        carried by a parasite or mechanical carrier. Dynamics may be complex - intermittent contact with a vector, variation in vector biology or distribution, disease burden in vector
        
        Continuous time model (Ross, 1917) of malaria dynamics including mosquito vectors show cycles
      - Zoonotic Infections: occur in animal populations but spill over into human populations
        
        use contact, vector or environmental tranmission
        
        patterns and freq rely on contact with host species and infectious dynamics in host species
        
        e.g. rabies, plague, lyme disease
  - - - Minimum SI - The time between N1 entering the symtomatic period and N2 entering their symptomatic after disease spreads from N1 to N2 during N1 latency period at the earliest possible moment
      - Maximum SI - The time between N1 ending their symptomatic period and N3 entering theirs following infection
      - SI presentation - the actual seial interval value is show as a range between min and max SI values
      - If the time between two cases fits within this range then those cases may be linked in a transmission chain
        
        Example used relies on the incubation period for all N being the same however IRL this is not the case
        
        The SI ranges would be further appart if incubation variation was accounted for
        
        -ve SI - The min SI can be -ve. THis means that N2 can get sick before N1 due to variation in incubation periods
    - - Involves both frequency and type of contact. There is a massive list of determinants for this
      - types of Contact Structures
        
        Homogenous: The cahnce of any two individuals making appropriate contect for infection is possible and equally likely
        
        Structured Population: contact within a strata is more likely than between strata
        
        Sufficient connections in a clustered population can act like a homogenous population
- - - - As the average age of a country changes then the distribution of morbidity will also change in a given time frame and associated with the change in age in a given nation
      - Thus, age will confound your comparrison of death between countries - How to you measure a direct correlation without confounding variable in the experimental structure
  - - - Direct Standardisation
        
        Question: What would the rate of death be if two populations had the same structure?
        General Approach: study population stratum specific rates are compared to standard population data
        
        age stratified rates of disease applied to an age stratified population
        
        The outcome is the the rates of disease in the populations being compared with the underlying difference is population structure being removed
        
        Commonly used for comparing populations from different countries
        
        The age specific rates are compared between individuals with the specific age strata of the "standard population". This shows the rates of mortality in the standard population - you have controlled for age
        
        Applying the same hypothetical standard of age between two countries you can observe the association between nationality and morbidity without the confounding influence of age
        
        in order to compare different standardised results to other you must use the same standardised structures in each study
      - Indirect standardisation
        
        Question: How do the observed deaths compare to the number of deaths which wouldhave been expected if the age specific rates of death were the same in all populations?
        General Approach: standard population strata-specific rates are applied to corresponding study population strata
        
        This shows the expected number of cases in a population comparred to the obsered number of cases in the population - calculating a ratio
        
        often used for comparing subgroups within a national population - standard rates are usually the total or national rates
        
        Age specific rates in two countries are selected (relevant to areas of comparison) apply these rates to age stratified populations in each to get the total expected cases in each country.
        
        then divide the number of observed cases by the number of expected cases in each area to get a standardised incince ratio (SIR)
        
        If observing mortality rates this is known a standardised mortality rates (SMR)
        
        Interpretting SIR:
        
        The output is a number equal to or different than 1 (may be expressed as 100 to equal the percentage difference)
        
        1 means the SIR is equal to the expected standardised rates in a larger (usually national) population
        
        1< means the rates are greater than expected. 1> means the rates are less than expected
        
        Observing rates of morbidity in an erea =/= the being in that erea causes that morbidity. Those rates are a reflection of the population of that area other than the confound controlled for
        
        the SIR is a measure of the strength of the association between a given area and the morbidity rates
        
        You can convert this to an actual rate (although rarely needed) by multiplying the national rate and the area SIR
  - - - clinical sign or symptom
      - Biological Indicator - medical tests
      - survey responses - mental health
      - Gold standard is a test
    - - Discrete outcomes - often:
        
        isolation/culture
        
        visualisation/imagery
        
        physical exam
        
        detection
        
        questions
      - continuous or ordinal outcomes - often:
        
        Biochemistry
        
        serology
        
        Pysical parameter
        
        Questions
      - When to do a test
        
        Surveillance (active or passive)
        
        estimate occurance
        
        detect outbreaks and initial response
        
        detect disease early to interviene
        
        Diagnosis - initiate treatment
        
        Research
        
        estimate occurance
        
        define individual status of disease, exposure or risk factors
        
        Choice of test indicate the strength of our estimates of occurance/exposure are - important to avoid misclassification bias
        
        Ideally using the gold standard test would be considerably more effective at diagnosing everyone. However this standard may be too epensive ro otherwise inaccessible to justify usage on asymptomatic people
        
        Tests better for widespread use are employed but less effective at accurate diagnosis
        
        The cutt-off is a trade off:
      - Defining performance
        
        Sensitivity: how effective is this test at identifying a disease?
        
        the proportiong of affected individuals accurately identified by the test
        
        SN = true positive / all affected
        
        Specificity: How good is this test at excluding diseases?
        
        The proportion of unaffected individuals correctly identified in the test
        
        SP = true negatives / all unaffected
        
        Klinkman et al (1998) - Evaluating MDD tests:
        
        SN = 31/81 - 38% will be identified correctly for depression (62% of cases will be missed)
        
        SP = 257/291 - 88% non-depressed people will be correctly identified (12% will be incorrectly identified as beign depressed)
        
        Cut off values
        
        To maximise SN you lower the cut off score - more N will be classed as depressed
        
        Consequences:
        
        Positive More correctly diagnosed patients would pursue therapy
        
        Negative More incorrectly diagnosed patients would pursue unneccesary treatment
        
        To maximise SP you increase the cut off score - Fewer N will be classed as depressed
        
        Consequences
        
        Positive Fewer people without depression would be flagged for therapy
        
        Negative Fewer people with depression would be flagged for therapy
        
        Evaluating diagnostic tools
        
        Total prevalence = TP + FN / n
        
        Apparent Prevalence (the prevalence recorded by the tool) = TP + FP / n
        
        The sensitivity of a tool can always be altered to better fit the scientific context
        
        Evaluating Bias
        
        Bias is dependent on the true prevalence of the disease
        
        Low prevalence = over estimation and high prevalence = under estimation
        
        Predictive Values - What Propertion of postive results are correct
        
        Positive Predictive Value (PPV) = true positives / test positives
        
        Negative Predictive Value (NPV) = true negatives / test negatives
        
        PVs are presented as %
- - - - Distributions: person, place, time, etc
      - Determinants: biological, physical, social, behavioural
      - Health-Related States: diseases, behaviours (e.g. smoking), effectsof preventative programs and use of health services
      - Specified Populations: characterised by geography, demographics or common characteristics
    - - What: causes ill health?
      - Who: does and does not suffer under this?
        
        Does the frequency differ from place to place?
        
        Does the frequency change over time
        
        What can be done to improve the health and health servies between a population or within a defined population?
  - - - Born in York, eldest of 9
      - Advocate of germ theory, denounces miasma theory
      - Cholera:
        
        Found that epidemics followed common trade routes
        
        Carefully examined the vital registration and parish records from the outbreaks in 1831-32 and 1848-49
        
        Hypothesised the agent was waterbourne, comes from faeces and ingested orally
        
        He tested this in the outbreak in 1853-57
        
        This lead to the understanding of geography in disease spreading. Areas in lower altitudes (where water will flow to) showed greater rates of disease.
        
        New legislation about reservoires moving up stream caused a decrease in cholera. Following this he went door to door to record rates of death and correlates them with the water supplier for that area
        
        over 17 weeks he found that one company correlated with increased rates of death and incidence rates
        
        Also looked at the water companies which supplied water to the north vs the south which showed him the waterbourne correlation of cholera
        
        Demonstrated that correlation =/= causation and disprove miasma theory
        
        This was solidified by the north street well being contaminated and causing an outbreak
- - - - Comparison of exposure histories: people with vs people without disease to determine whether exposures occur more or less in these cases compared to controls - shows increased or decreased risk of disease
    - - As a result it is impractical to study rare outcomes in the GP using a cohort study as N cases will be significantly fewer than controls
    - - Presents an increase in the number of N with the disease that are identified
      - Takes advantage of: notifications, registrations and referals
      - Wider population
      - Relative to a cohort study - less expensive and less time costly
      - Case control studies can still be expensive and time consuming depending on their specific design
    - - Provide estimates of the frequency of exposures among individuals with disease from the target population
      - N must adhere to a clear case definition: is this a clinical definition, more objectively defined?
      - Potential sources: routine testing, notifications, registration, case-finding
    - - provide an estimate of the expected frequency of exposure in the target population
      - must represent where the cases came from and be at risk of being a case N
      - Population: health registers/electoral rolls, proximity to case
      - Hospital: admission to the same/different ward for non-case condition
      - Rest of the cohort: nested designs
    - - none available from a case-control study
      - The proportion of the diseased N in the study is set by the design
      - Although: disease incidence might be determined in a nested case-control study based on information of the cohort it is nested within
      - Measuring exposure frequency
        
        Proportion (typically) - did/did not happen (the simplest case involves the following for this proportion calculation)
        
        How many times did exposure occur?
        
        How long ago did exposure occur?
        
        Historical:
        
        Who provides the data - cases are ill (are they dead?)
        
        How is it measured? - questionnaire, blood test (before or after diagnosis?), contemporary records (medical census, employment - how accessible is this info?)
    - - exposure histories of diseased/non-diseased N within the cohort are compared
      - Why do a nested case-control study? - cost effective approach for larger cohorts
        
        use of baseline biological material - finite and expensive resource to analyse
        
        further data required for those who develop the disease of interest - additional exposure data be be needed, may be costly in time/money
    - - Pre-select a subset of the cohort (more detailed follow up from the outset)
      - All cases in the wider cohort are compared to this subset
      - alows for multiple case sets and more complex analytics
      - Matching: used to control confounding data
        
        Assures comparable distribution of known confounders among cases and controls
        
        Matching as a part of the study design:
        
        efficient statistically
        
        multiple matched controls per case increases power
        
        must be accounted for in analysis
        
        Cons:
        
        Effects of matched factors cannot be estimated
        
        can restrict control selection
        
        overmatching: many different factors, factor of interest closely linked to a matching factor
        
        makes cases and controls too similar - no observable differences
  - - - Using odds ratio in a cohort runs the risk of increased risk - estimation of risk by odds improves as risk decreases
      - Odds (exposure in cases) = DP + target exposure / DP / DP - exposure / DP
      - Odds in controls = control + exposure / controls / controls - exposure / controls
    - - Valid when disease prevalence is low
      - measures the strength of the relationship between exposure and disease
      - ratio >1 = exposureis more likely in diseased individuals than in non-diseased individuals
      - ratio <1 = exposure less likely in diseased population compared to non-diseased population
    - - Odds ratio = (exp + disease / all diseased) / (non-exp + disease / all diseased) / (exp + non-desease / all non-diseased) / (non-exp + non-diseased / all non-diseased)
    - - We calculate the odds of exposure given disease status
      - The odds ratio - "the odds of [control] having [exposure] are [calculation output value] more/less likely to have [disease] than that of [control] without [exposure]
  - - - Do controls accurately relflect the karger population they belong to?
      - Sampling methods should ensure that controls comprise a representitive sample of the population from which the cases originated
        
        Important to acknowledge the demographic determinants of who would particpate in the study also (poverty, employment, etc)
        
        This demographic confound may not be equally reflected in the case condition - this will result in a systematic selection bias affecting the controls
    - - Interview based case-control studies are prone to this
      - recall of previous exposures may be related to individuals case-control status
      - Solution: obtain exposure data from sources independent from N's own reporting - records or tests
      - Low individual-level agreement between recall and medical records
        
        Genreally poor results - the Sensitivity (L2) will be genreally lower than objective records
        
        However specificity (L2) is generally relatively accurate
        
        This leads to potential misclassification that depends on the outcome of this recall (case vs control status)
- - - - Unit of analysis = population of groups
        
        typically measures of disease occurance are correlated with other characteristics of the population
        
        units of observation are geographically defined
        
        countries/regions within a country
        
        same population at different points in time
        
        Data used can be obtained from many sources
        
        natoinal statistics
        
        cross-sectional data
      - Correlational Analysis - pearsons test of correlation is sued measure the strength of linear associations between two variables
        
        Ecological Fallacy - as association observed between variables on aggrigate level doesnt necessarily reflect the association on an individual level
    - - Collect info fomr N at a particular point in time - repeat surveys provide time trends
      - information of different types - questinairres/interviews, biological samples
      - Target population - entire or subset
      - Commonly used to:
        
        Commonly used to: determine the distribution of factorswithin the population (demographics, smiking, drinking, etc)
        
        estimate levels of common conditions with reasonable long durations (LT illness, serological status, etc)
        
        The data in cross-sectional studies are often used for ecological studies
      - Prevalence - measure of disease frequency/burden of disease in a population
        
        made up of all ongoing cases of disease
        
        Point Prevalence: the proportion of the population affected by the condition of interest at a specific point in time
        
        Can be calculated from a cross-xectional study and is important for health and social care planning
  - - - Cohort - rare exposure, multiple outcomes, temporal relations, incidence (poor for rare disease, multiple exposure)
      - Case Control - rare disease, multiple exposures (poor for rare exposure, temporal relations and incidence - cannot measure multiple outcomes)
      - Cross-Sectional - weak for all - cannot measure incidence
      - Ecological - moderate strength for rare disease - weak for rare exposure, multiple outcomes and multiple exposures - cannot measure temporal relations of incidence
    - - Str: cheap and ethically safe
      - Lmt: only estimates prevalence not incidence, only assocaition not causality, succeptible to recall bias, confounders may be unequally distributed, group sizes may be unequal
    - - Str: shows temporal relation between exposure and disease, can make some attempt to asses cause and effect (RCT are preferable), sometimes more representative of clinical population
      - Lmt: large cohort and long follow uo for rare disease, loss of follow-up risk, cannot rule out unmeasured confounding
    - - Str: realtively cheap, quick and easy, no attrition, suitable for rare events with long lag between exposure and outcome
      - Lmt: potential recall bias, timing cannot always be reliably established, cannot assess incidence (proportion of disease is fixed), relies on recall or recorsd to determine exposure status, confounds, selection of groups can be difficult
    - - Str: quick and cheap
      - Lmt: association not causation, confounds, ecological fallacy
- - - - This observe the target exposure and outcome seperates for N who fit or do not fit the confound to see if there is a significant difference in the outcome - if their is, you can identify which factor does not have a causal relationship and is thus a confound
      - if the strata-specific odds ratio is even but only the crude odds ratio (taken from the original outcome of factor and outcome) is onyl similar for one factor then that is your causal factor, the other is a counfound
    - - This shows the effect of exposure on the outcome controlling for the confounding factor
      - Reporting final summary statistics: the odds of [outcome] are x times higher in individuals with high [causal exposure] compared to those with low [causal factor], adjusting for the confounding effects of [confound]
    - - confounds are a theat to internal validity
        
        Mainly in observational research (cohort, cross-sectional and case-control studies)
        
        Confounds can lead to over/underestimation of effect sizes
        
        Confounds can cause reverse direction of association (Simpsons Paradox)
      - Controlling Confounds
        
        Study Design
        
        Restrict sampling
        
        Randomisation (trials)
        
        Matches cases to controls
        
        Data Analysis
        
        Adjusting for confounds (see above)
        
        Stratification
        
        Regression models
  - - - If the relative risk of disease is not equal between individuals who have and have not experiences this other factor then it is likely this is an effect modifier
      - The EM does not have to associate with the exposure (although it can) but must be associated with the outcome
    - - The two ORs shouldhave predicitive power for the outcome however the strength of assocaition will be higher in the group with a positive EM
      - For an EM to be positive: the crude OR should not equal either SOR and the SORs should not be equal to eachother - Whichever SOR is stronger is the group with the EM exposure
    - - Stratification: stratify the data by the levels of this third factor
      - compare the stratum specific effect estimates to the druce effect estimates
      - Null Results: If the SSEE are equal to eachother AND the crude EE this third facotr is neither a confound or an EM
      - Confounding Results: if the SSORs are equal but different than the COR this third factor is a confound
        
        You now produce a pooled OR adjusting for this confound on the relationship between exposure and outcome
      - EM Results: if the SSEM are different from eachotehr AND the COR then the third factor is an EM
        
        We can report that there is heterogeneity in the SSEM when the SSEMs are not equal
        
        Do not produce a pooled OR - do not adjust for the EM
      - Diagram For Identification: (bold = stats, Roman = result)**
        
        Crude OR
        
        SSOR
        
        Similar
        
        Pooled OR
        
        Pooled = Crude
        
        1 more item...
        
        Pooled =/= Crude
        
        1 more item...
        
        Disimilar
        
        EM - use SSOR
- - - - Necessary Causes: "You wont develop the disease without this exposure"
      - Based on these, smoking is neither sufficient or necessary for lung cancer
      - Component Causes: Even with a necessary exposure, compenent factors will contribute to the likelihood of developping the disease in addition - a combination of factors = a sufficient cause
        
        When a necessary exposure occurs, multiple single components have causal effects in addition, these can be different characteristics which interact with the necessary cause
        
        The contribute toward one sufficient causal mechanism
        
        Removing even just one component cause can decrease the chances of developping the disease by compromising the causal mechanism
      - Sufficient Causes: "if you have the exposure you will get the disease"
    - - Prevention can be effective in reducing disease rates by removing the removable component causes to stop sufficient causal mechanisms
      - This compnent does not have to be the necessary cause as that is often impossible to prevent (e.g. HPV)
    - - Criteria for Causality in an Observed Association: Nine Viewpoints (Bradford Hill, 1897-1991) - Not all need to be present as they do not reflect every disease
        
        Strength of association
        
        consistency of association
        
        Specificity
        
        Temporality
        
        Dose response or biological gradient
        
        Plausability
        
        Coherence
        
        Experimental evidence
        
        analogy
      - Reporting strength/consistency: Compare odds ratio findings over time between the target cause and rates of disease following exposure
      - Temporality: the necessity that the cause preceeds the outcome
        
        descriptive studies - can we see this occur naturally?
        
        Follow-up Studies: find disease occurance rates following exposure
        
        retrospective cohorts/nested case-control studies: documented presence of exposure before disease occurance
      - Dose-Response Relation: Is a unidirectional cause and effect observed? (very difficult to record)
        
        Nested Case-Control Studies
      - Specificity/Plausability/Coherence:
        
        Specificity - a cause leads to a single effect, not multiple - this is often refuted and invalidated due to life beign more complicated
        
        Plausability: Does this relationship have a valid biological basis?
        
        Coherence: a cause-effect interpretation does not conflict with existing knowledge of the antural history of biology and disease
    - - Mediators: An intermiediate variable between an exposure and an outcome which is influenced by the exposure on the causal pathway to the outcome (Ananth & Brandt, 2023)
      - E.g. obesity causes CVD - a mediator of this relationship would be blood lipid levels rising - observing the association between blood lipids and CVD can explain how the exposure affects the outcome on a biological level
- - - - Calculate disease occurance in N without exposure as Y.
      - Y = expected background occurance.
      - X - Y = excess of background occurance.
      - Calculate disease occurance in N following exposure as X.
      - Example: smoking and lung cancer
        
        Smokers: 140 cases for ever 100,000 person years. Non-smokers: 10 cases every 100,000 person years
        
        140 - 10 = 130
        
        Reporting: An additional 130 cases of lung cancer were observed for each 100,000 person years in smokers compared to non-smokers
      - AR: how many occurances occur in one group compared to another and what is the burden of disease associated with the exposure group
        
        AR > 0 = exposure associated with increased cases
        
        AR < 0 = exposure associated with fewer cases
        
        The larger the AR the greater estimate of disease outcome the exposure is
    - - What proportion of cases observed in exposure group are caused by that exposure?
      - Attributable Factor = AR / Risk in exposed group
        
        AR = how much extra we see caused by exposure
        
        Risk in exposed group = what is observed in the exposed group
        
        Divide each element by risk in unexposed group - Relative risk - 1 / relative risk
        
        Result: Percentage of cases explained by exposure following calculation
    - - incidence rates are not observed so AR cannot be calculated but you CAN calculate AF
      - Case-control studies can calulcate relative risk. Use this to divide each element by risk in unexposed group
      - AF is only relevent for the exposed N for the study and possibly the larger population it reflects
    - - Bias: selection/inclusion and measurement/information
      - The ratio observed within the study will reflect the GP when compared to the frequency of exposure in the population - this is often derived from routine data and calculated outside of your study
      - Calculation: can be estimated using differences in incidence between the population and unexposed group OR using an estimate of effect and estimate of exposure frequency
      - Population AR: incidence in GP - incidence in unexposed group
      - Reporting: "in the general population [exposure] may account for [% chance of outcome] ([GP number of cases per PY - unexposed cases per PY numbers]) per [figure] deaths per year from lung cancer
    - - Reporting: "so (with many assumptions) if nobody smoked 83% ([PAR-GPR]) of lung cancer related deaths would not occur."
      - This % may exceed 100 when assesing many causes (especially with many weak effects)
      - PAF in Case-Control Studies: estimated by Levin's Formula (Levin 1953)
      - Calc: PAF = P(exposure) x (OR - 1) / P(exposure) x (OR - 1) +1
        
        P(exposure) = frequency of exposure in control population
        
        OR - 1 = Relative Risk in exposed vs non-exposed group
      - As OR and RR can be used we can calculate PAF in case-control studies without calculating incidence
  - - - 1 Preclinical stage: primary prevention - exposure - onset of disease - secondary prevention
      - 2 Clinical Stage: onset of symptoms/signs OR arrested progression (no further steps) - tertiary prevention - result: cure/control/disability/death
    - - increase exposure to preventative factors OR decrease exposure to causal exposures
        
        modify personal behaviour (stop smoking)
        
        Change or improve services in GP (clean water)
      - Strategies:
        
        mass or population (whole population approach)
        
        reduce burden of disease across whole population - modulation of pop behaviour or parameters
        
        High-risk (subgroups in GP)
        
        Target those at high risk - mose high risk N to lower risk levels
        
        Which to Choose
        
        Mass/Pop:
        
        potential for large impact
        
        requires widespread participation
        
        HIgh-Risk:
        
        may be a small impact if high-risk group contributes litle ot the overall burden of disease
        
        those treated more liekly to those that benefit
        
        Further Consider:
        
        cost of prevention, who bears the cost, how intrusive is it?, how does it influence compliance, how will success be calculated
    - - Screenin Programs:
        
        predicted on being able to do something
        
        not all diseases are suitable
        
        need to fit certain criteria
      - Introduction inevitably results in:
        
        increases in disease incidence and/or prevelance
        
        apparent (biased) improvements in survival
        
        Requires tests (sometimes many)
    - - evidence used to design interventions but need to know if our interventions are effective
      - Measuring impact of interventions: compare these findings to original expectations, consider timescale of impact - is there lag expected?
        
        Process measures
        
        how many N: screened, vaccinated, impacted??
        
        Was target population engaged
        
        cost
        
        Outcome measures
        
        Cases/deaths prevented
        
        life years added
        
        Behaviour changed
      - Data Sources: Be aware of previously discussed biases in data (both in testing and intervention data)
        
        Routine Data sources
        
        Bespoke mechanisms (RCT or before/after)
        
        Quantitive/qualitative
        
        Economics
      - Further Biases:
        
        Lead Time Bias
        
        Earlier diagnosis = transcient increase in incidence
        
        Artificial increase in survival - this may appear to increase time between diagnosis and death but if all screening does is diagnose earlier than the actual lead time toward death is not changed, only our understanding of how long N can have a disease before death
        
        Length Time Bias:
        
        screening is more likely to diagnose slower developping cases over accutely developping cases
        
        Overdiagnosis