Reading...
Play button
Play button
Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
57 Cards in this Set
- Front
- Back
|
Describe different types of bias.
|
Information bias - differences in a measurement in the groups being studied ( correct people, wrong information)
Selection bias - differences in the groups being studied due to the way they were selected ( wrong people, correct information) Recal bias: differences in the groups due to forgetting information or more inclined to give you the information they think you want to hear. Blinding might be needed. Non responder bias - better to have moderate size sample with high response rate. Healthy worker effect - those working are more likely to be healthy |
|
|
Explain causality?
|
Causality is the relationship between a first event(cause) and a second event( effect) where the 2nd event is understood as a consequence of the first.
Deterministic causality - systematic certainty, observations to predict furture events. Stochastic causality - likelyhood and probability. Association does not equal causality. |
|
|
Define a confounding factor.
|
A confounding factor is a factor that is associated with both exposure and outcome but is not itself on the causal pathway. eg alcohol is a confounding factor between smoking an lung cancer. Maybe people who smoke are more likely to drink more.
|
|
|
Define a census.
|
A census is the simultaneous recording of demographic data eg age, gender eg by the gouvernment at a particular time pertaining to all the people who live in a particular area.
|
|
|
By whom are census forms not usually returned and why?
|
Young people: more likely to live in inner city areas where high turnover rate and multi occupied housing makes it difficult to gain access.
Babies: Might be in hospital on census day so are left off form Elderly: Living alone, may have difficulty answering the door and filling out the form. Students: confusion whether resident is at parent's house or term time address. Minority ethnic groups: language barrier difficulties, suspcion. |
|
|
How is census data used?
|
census data is used to make projections on population size in order to plan services such as schools, youth centres etc.
Birth, fertility, death and migration rates are assumed to stay the same. |
|
|
Define morbitity.
|
Morbidity is any departure from a state of physiological or psychological well being.
|
|
|
Define CBR.
|
CBR- crude brith rate is the number of live births per 1000 population.
Negative factors are that age and gender aren't taken into account. |
|
|
Define GFR and TPFR.
|
GFR- general fertility rate.
The number of live births per 1000 females in popultion aged 15-44. Negative factors: women may have children before or after this age range. TPFR- total period fertility rate. Total perioid fertility rate is the average number of children that would be born to a hypothetical woman in her life (at that time) TPFR = sum off ( all current age-specific fertility rates) eg 301,600 women aged 15 had 1090 births = 0.0036 births per 15 year old + age specific FR of 16 yr old + 17 yr old and so on until 44. rate = no of births/ no. of people |
|
|
Define fertility and fecundity and explain why fertility rates are difficult to predict.
|
Fecundity is the physical ability to reproduce. it decreases with sterilisation and hysteorectomies.
Fertility is the realisation of this potential to reproduce as births. Fertility increases with sexual activity and economic climate. Fertility decreases with contraception and abortion. It is difficult to predict fertility rates as it is difficult to predict social behaviour. Conceptions = live births + micarriges + abortions but not all miscarriages are known. |
|
|
Define CDR and ASDR.
|
CDR- crude death rate is the number of deaths per 1000 population (per time)
ASDR - age specific death rate is the number of deaths per 1000 in age group. (number of deaths/ age group population ) X 1000. |
|
|
Define SMR.
|
Standardised mortalilty ratio compares observed number of deaths with number expected if age- sex distributions of populations were identical. It is an adjusted ratio for sex and age, COMPARES LIKE FOR LIKE - better comparison when considering likelyhood of death then crude mortatlity rate ratios.
exepcted deaths uses number in population X rate from compared population. eg (total observed female deaths( all ages) / total expected female deaths ( all ages) ) X 100 = SMR SMR> 100 - More deaths than expected SMS = 100 - null hypothesis not rejected. SMS < 100 - less deaths than expected |
|
|
Define incidence rate and prevalence.
|
Incidence rate is the number of new cases of a disease per year, per population. It is a measure of individual risk.
IR= no. of new cases/ no. of person years. Prevalence is the measuring of existing cases - ratio NOT rate. prevalence = no. of cases/ no/ of population at a given time. P is associated with ( incidence X length of disease) Point prevalence is used for planning as more useful. |
|
|
Define IRR and MRR
|
Incidence rate ratio compares the incidence rate of one population with another
IRR= rate A/ rate B Mortalilty rate ratio compares the mortality rate of one population with another. MRR may compare new treatment with existing treatment eg >1, new treatment is more effective. >1 - group A is at higher risk <1 - group B is at higher risk rate - measure of absolute risk, how likely an individual in the population is likely to get it. Ratio -measures relative risk - compared with another group |
|
|
Describe the relationship between tendency and observation.
|
The observation value is the best estimate of the true or underlying tendency. However random variation influences everything we observe so it is difficult to know if the true value is increasing or decreasing. Hypothesis testing uses observed data to test hypotheses about what we have have measured. Estimation from the hypthesis testing uses observed data to provide a range within which the true value is likely to lie.
|
|
|
What is a hypothesis.
|
A hypothesis is a statement about the true value of the epidemiological measure we are interested in. The observation is then checked to see if it is compatible with the hypothesis.
eg Hypothesis is that there is no real excess risk of leukaemia near the power station ie consider true IRR for leukaemia =1 observation: observed IRR =4 Calculate probability of observing IRR =4 by chance, if true IRR =1. if the probability is quite high then chance can explain the observed difference and the hypothesis is not rejected. The more improbable the observation, under the assumption that the hypothesis is true, the stronger the evidence against the hypothesis. null hypothesis: the hypothesis that the 2 groups do not differ ie IRR= 1, SMR=100 |
|
|
Define the p-value.
|
The p value is the probability of an observation occuring by chance. The smaller the probability (the p-value) the less likely the observation is due to random variation, and the stronger the evidence to reject the hypothesis. A hypothesis is rejected if p-value is lower than 0.05 - 5%. It is statistically significant as some other factor apart from chance is likely to be responsible for the observed difference ( between hypothesis value and observed value)
|
|
|
what is a 95% confidence interval and how are they calculated.
|
The 95% confidence interval is the range of values that we can be 95% certain will contain the true value of the underlying tendency. The range is centred on the observed value as it is the best estimate for true, underlying value, therefore the observed value has to be the middle of the range. It the null hypothesis is inside the range then any observed difference may be due to chance, p>0.05. If the null hypothesis is outside the 95% CI then p< 0.5 and is statistically significant.
In order to calculate the CI limits, first the error factor must be calculated using the rate, IRR or SMR values. lower CI = observed value/ e.f higher CI = observed value X e.f |
|
|
How is calculating a 95% confidence interval different with a rate, a rate ratio and an SMR.
|
Rate:
the error factor is calculated using the number of cases/births/deaths. Rate ratio: the error factor is calculated using number of cases in both populations( a first then b) CI = IRR/ e.f to IRR x e.f null hypotheis = IRR =1, no observed difference in 2 groups. SMR: = (observed/ expected) x 100 e.f is calculated using only observed deaths. CI = SMR/ e.f to SMR x e.f the null hypothesis is SMR = 100 and there is no observable difference between the 2 groups. If CI does not contain 100 then null hypothesis can be rejected as probability that observed difference is due to chance is too low p> 0.05 - statistically significant. |
|
|
Describe how you would carry out a cohort study.
|
A cohort study compares the relative risk in one group of people to the other.
1. recruit disease free individuals 2. classify into exposed and unexposed or varrying levels of exposure 2. follow all individuals over time until the end of the study 3. compare outcomes Internal comparison - sudividing levels of exposure and comparing them. A cohort that is exposed and one that is unexposed, or varrying levels of exposure - comparison of subcohorts. - use IRR to find relative risk and use 95% CIs EXTERNAL comparison- compare cohort with reference population. Relative risk is worked out using SMR and 95% CIs. The cohort study may be retrospective - define exposure status in disease free individuals using historical records eg have they developed disease of interest between 1990-2004 Prospective studies - recruit and define exposure in disease free individuals and then follow them up. |
|
|
What are the pros and cons of a cohort studie - internal and external
|
Pros:
- suitable for rare exposure - less prone to bias (only in prospective) - study range of outcomes from single exposure - can calculate absolute risk - IR (only in prospective) - removes uncertainty that exposure procedes disease Cons: - confounding factors - subcohorts may need to be very large - survivor bias - the healthy ones - takes a long time ( especially if latent disease) - not suitable for rare disease - expensive - external comparisons - healthy worker effect form of selection bias - reference group must be in similar jobs to cohort - problems with loss to follow up eg maybe individual has died, or moved |
|
|
Compare error factors in internal comparisons and external comparisons.
|
The error factor decreases with number of events/cases.
The error factors for internal comparisons are larger than for the external comparisons because the observed cases are being subdivided, thus number of events is decreasing, The uncertainty about the SMR only comes from the uncertainty about the rate in the exposed. The uncertainty about the IRR comprises of the uncertainty about the rate in exposed and the uncertainty about the rate in unexposed. The confidence interval for SMR is always narrower than that for the IRR. |
|
|
A new drug is shown to significantly improce survival in large bowel cancer. Why might you want to see the confidence interval as well as the p-value.
|
A confidence interval gives the reader more information than the p-value. With a confidence interval the null hypothesis of new drug not improving survival can be tested. By seeing the position of the null hypothesis ie 1, you can predict the effect more data would have on it. Also the CI gives a range within which we can be 95% certain that the true value for the mortalilty rate ratio lies. The range also gives an indication of the sample sized used. The larger the sample size, the smaller the error factor.
Clinicians need to have as much information as possible about the magnitude of the effect to decide whether benifits outweigh possible side effects. |
|
|
Define a case control study.
|
case control studies are retrospective
-Identify on disease outcome and compare exposure in the past - large proportion of available cases in the population - small proportion of available controls in the population |
|
|
Define Odds ratio and explain how it is calculated.
|
Odds ratios are used in case control studies: divide the odds of being exposed in cases by the odds of being exposed in controls - estimate of relative risk.
cases Controls exposed a b Unexposed c d (cases = diseased and controls = non diseased) headings in table or in alphebetical order. Odds of being exposed in cases: a/c Odds of being exposed in controls: b/d OR = a/c divided by b/d = ad/bc The range within which the true underlying OR value lies can be calculated using the ef. |
|
|
How could you increase the precision of the odds ratio within a case control study.
|
To increase the precision of the odds ratio ( ie reduce error factor and narrow 95% confidence interval) you can increase the number of controls. This can be done because there are 4 components which contribute to the error factor a,b,c,d.- the bigger these numbers are the smaller the error factor.
b and d are controls and can be increased but only to 5 or so controls as beyond that, the error factor is mainly due to the number of cases as the change in controls makes no difference. eg a=30,b=20, c=10,d=20( 1 control per case) e.f= 2.63 a=30, b=100, c=10 d= 100 ( 5 controls per case) e.f 2.19 - decreased. a=30, b=10000, c=10, d=10000. ( 500 controls per case) e.f = only 2.08!!! - no effect Bias underestimates odds ratio by increasing proportion of exposed cases and controls and decreasing proportion of unexposed cases and controls - lower odds ratio, smaller difference in odds of getting disease with exposure compared to without exposure!!! |
|
|
Why is it not usually posible to calculate absolute incidence rates in case control studies.
|
In case control studies we generally select a large proportion of the available cases in the population but a small proportion of the available controls in the population. Because we don't know exactly what these 2 proportions are, it is usually impossible to obtain a direct estimate of the incidence or prevalence of the disease in the general population.
|
|
|
Describe the possitives and negatives of case control studies
|
Possitives:
- suitable for rare diseases-no need to follow up 1000s of healthy people - Look into many different potential exposures at once - preliminary investigations to narrow down causal exposure - cheap and quick Negatives: - not suitable for rare exposures - difficult to be sure exposure preceded disease - confounding factors - prone to recall bias (information bias) - selection bias - not representative sample of general population ( more so than cohort studies) - bias underestimates/overestimates odds ratio: if cases understate how much exposure they've had - UNDERESTIMATE. If controls understate how much exposure they/ve had - OVERESTIMATE. If both controls and cases underestimate -> shrink to the null |
|
|
Describe a nested control study.
|
A nested case control study collects data from the evolving outcome and exposure database of a prospective cohort study - case control is nested within the cohort study.
Advantages over normal case control studies - proportions are known so incidence rates ( absolute risk) can be calculated - population for sampling of controls is already defined. Advantages over normal cohort studies - can collect more detailed information for a minority of participants |
|
|
Explain the difference between bias and confounding.
|
Bias is a characteristic of a flawed study - either the population studied is unrepresentitive ( selection bias - wrong people, right information) or the information is systematically wrong ( information bias ( wrong information, right people)
Confounding is a characteristic of the population eg drinkers are more likely to smoke than non smokers and this would affect a perfect study. |
|
|
What would the effect on the odds ratio be:
a. the cases understated more than the controls how much they smoked. b. the controls understated more than the cases how much they smoked. c. both cases and controls understated randomly how much they smoked. |
a. If the cases understated more than the controls then this would bias the odds ratio towards underestimating the true value.
b. If the controls understated more than the cases then this would bias the odds towards overestimating the true value. c.Both cases and controls understated how much they smoked causes shrinkage to the null where the odds ratio moves closer to 1. |
|
|
Define a cause
|
A cause is a factor which increases the probability of disease
Necessary cause - must always precede a disease eg tubercle bacillus and tuberculosis. Sufficent cause - when it can cause the disease on its own. important causes of diseas - alcohol consumption and liver cirrhosis - ionising radiation and leukaemia - smoking tobacco and heart disease |
|
|
Describe epidemiological reasoning for cause- effect relationship,
|
1. Hypothesis - generated from observations and/or theories
2. Analytical study - systematic observations of comparisons. 3. Observed Associations - possible explanatons of non-causal associations ( chance, confounding, bias, reverse causality) 4. cause-effect relationship- based on judgement of how the observed association fits in with evidence from other sources. |
|
|
Explain the bradford hill's criteria for inferring causality.
|
Association feature:
- strength of association - strong association more likely to be causal - specificity of association - specific to the exposure outcome ( most diseases are multifactorial and many factors can cause several diseases) - consistency of association: association demonstrated by different studies on different groups of people in different places at different times Exposure/outcome: - Temporal sequence - exposure procedes outcome of interest - Dose response - different levels of exposure = different levels of outcome - reversibility - removal of exposure, decrease in risk of disease Over evidence - Coherence of theory - association conforms with current knowledge and theory. - biological plausibility - biologically plausible mechanism - Analogy - an analogy exists. |
|
|
Define epidemiology.
|
Epidemiology is the study of health event patterns in a society and is used in evidence based medicine to identify risk factors for disease and determine optimal treatment approaches.
|
|
|
Describe Henle koch's prostulates.
|
1. the agent must be present in every case of the disease NECESSARY
2. the agent must not be found in cases of any other diease SPECIFIC 3. the agent must be capable of reproducing the disease in experimental animals and must be recovered from the experimental disease produced SUFFICIENT. however nowadays we know that many factors can stimulate several diseases. |
|
|
Define a clinical trial
|
A clinical trial is any form of planned experiments using patients and designed to find the best method of treatment for future patients with a given medical condition.
A clinical trial is a fair, controlled (comparative), reproducible experimental study as opose to observational trials eg cohort studies. The purpose of a clinical trial is to provide reliable evidence of treatment efficacy and safely |
|
|
Outline the steps inbolved in a randomised control trial.
|
DEFINE FACTORS - disease, treatment, outcomes, bias confounding - predefining outcomes prevents data dredging and repeat analysis.
CONDUCT TRIAL: identify, recruit, consent and maintain (follow up patients) 2 comparable groups of participants - treatment allocation must be by chance to minimise allocation bias and confounding (known and unlknown) - new vs standard treatment COMPARISON OF OUTCOMES: statistically significant and clinicaly significant. eg patho-physiological eg tumour size, clinicaly defined eg death/disability, patient focused eg quality of life Placebo used if no standard treatment! - an inert substance that looks, tasts, texture, dosage requiremnet, warnings are all the same as the new treatment, Illness may be improved by a feeling that something is being done about it. |
|
|
Why do randomised controlled trials give strong evidence about causality.
|
Strong evidence is achieved by minimising bias and confounding.
Randomisation achieves pefect balance in the long run eg large trials - minmises allocation bias and confounding ( known and unknown) Avoids allocation bias as random allocation will tend to result in equal numbers of similar age, sex, social status and other possible sources of counfounding distributions in the 2 groups, thus avoiding problems of confounding. Random number tables or computer generated random numbers. |
|
|
What are some problems with randomised controled trials and how may they be biased.
|
- There can be ethical problems eg consent from children, people with cognitive impaiement
- There can be problems with loss to follow up eg people die, fail to fill in diary, get bored. Bias - selection bias eg selecting an unrepresentive group of patients - allocation bias eg by passing random allocation process - treatment bias eg differential treatment of participants depending on which treatment group they are in the trial - assessment bias eg treatment of group's outcomes are measured more favourably many of these biases can be minimised by blinding. In general large well conducted RCTs are preferred to well conducted systematic reviews that depend on small trials |
|
|
Explain blinding and why it is done.
|
Blinding minimises non-treatment effects and measurment bias- ensures the allocation of treatment to patients is not known by:
- Patient- who may change their behavoir depending on the treatment - The clinician - May alter their treatment, care and interest in a patient ( non treatment effect) - The assessor - may alter approach when making measurements and assessing outcomes ( measurment bias) Single blind - one of patient, clinician or assessor does not know treatmnet Double blind - 2 of patient, clinician or assessor but usually assessor=clinician. Blinding: aim to make treatments appear identical in every way eg appearance, taste, texture, dosage regime, warnings, label identical containers for treatments with code numbers instead of names. Blinding is difficult with surgical problems, psychotherapy vs anti-depressant, alternative medicine vs western medicine, lifestyle interventions, prevention programmes. |
|
|
How could you avoid loss to follow up in randomised controlled trial.
|
Appropriate losses - their clinical condition may necessiate their removal from the trial
Unfortunate losses - they may choose to withdraw from the trial Minimise losses to follow up -minimise hospital visits - honnest about commitments - avoid inducements - maintain contact |
|
|
Describe the difference between intention to treat analysis and as treated analysis and decide which one is better.
|
Intension to treat analysis is based on a pragmatic trial which inlcudes everyone, including non-compliers. It therefore gives a more realistic idea of what would happen in the real world, taking account of side effects, non compliance and so on and would give a better idea of what the treatment would actually achieve for a population of patients. Also the intension to treat analysis maintains the original randomisation and therefore deals with all cofounders, knownn and unknown.
As treated analysis from an explanatory trial doesn't include non compliers. It therefore does not preserve randomisation and although you can adjust for known confounders, you can't for unlknown confounders. Also it ignores the impact of side effects and palatability of the treatment on the treatment effect. As treated compares physiological effects of the treatments. Intention to treat compares likely effects of using the treatments in routin clinical practice, more realistic sizes of effect. |
|
|
Compare collective vs individual effects in a randomised controlled trial.
|
Collective ethic RCTs aim to properly test treatments for efficacy and safety.
Individual ethic: - RCTs do not guarantee benefit -RCTs may cause harm - RCTs allocate treatment by chance - RCTs place burdens and confer benefits RCTs are for the benefit of future patients |
|
|
Describe issues to consider for a clinical trial to be ethical.
|
Clinical equipoise - is when there is genuine uncertainty about whether the drug will be beneficial
Scientifically robust - relevant issue, appropriate design can justify treatment and available risk Ethical recruitment - inappropriate inclusion eg pregnant women ( high risk of harm), inappropriate exclusion eg elderly or children, mentally ill ( people difficult to get valid consent from) Valid consent - informed, competent decision Voluntariness- avoid coercion. |
|
|
Define the role and function of a research ethics committee.
|
NHS trust/PCT R&D office
- research governance - financial management - resource implications Research effects committee - ensures that research respects th dignity, rights, safet and well being of individual research participants. |
|
|
Describe evidence based healthcare.
|
-Primary research reviews eg RCT
- Literature reviews ( expert reviews + systematic reviews) - Decision analyses - harms and benefits and cost effectiveness |
|
|
Define systematic reviews and how they are prone to bias
|
" an overview of primary studies that used explicit(transparent) and reproducible methods" - usually small trials are used
- overview - should be comprehenive - primary studies - quality assured ( internal validity of study and external validity to review) - Explicit: statements about objective, materials and methods - Reproducible: follows on from explicitness May not include meta-analysis if clinical heterogeneity is too great. Systemic reviews are prone to bias - faling to include all relevant trials eg puplication bias - failing to exclude poor quality trials eg allocation, treatment biased trials - only including trials for particular situations eg not applicable locally - inappropriate combining og results eg pooling heterogeneous results |
|
|
Define a meta-analysis.
|
" a quantitive synthesis of the results of 2 or more primary studies that addresed the same hypothesis in the same way"
Quantitive studies: pooled estimate of effect eg odds ratiom risk ratio Primary studies: used primary data Same hypothesis: similar outcomes, population, intervention Same way: similar study designs. quantitive pooling of results from individual studies. Odds ratios and their 95% CIs are calculated for all studies and then combined to give a pooled estimate odds ratio using a statistical computer program. Studies are weighter according to their size and the uncertainty of their odds ratio( smaller e.f greater weight) |
|
|
Describe the formal protocol for meta- analysis.
|
- compilation of complete set of studies
- identification of common variable or category definition - standardised data extraction - analysis alowing for sources of variation |
|
|
Describe the used of forest plots.
|
Forest plots are used to express pooled estimate odds ratios.
- individual odds ratios (squares) with the 95% CI (lines) are displayed for each study - size of square is proportional to weight given to the study based on size of study and uncertainty of odds ratio ie ef ( small ef = greater weight) - the diamond is the pooled estimate with the centre indicating the pooled odds ratio, and the width representing the pooled 95% CI. Null hypothesis line = 1 runs through all studies.(might not cross study) |
|
|
What are the problems associated with meta analysis?
|
- Heterogeneity between studies : no 2 studies are identical so modelling for variation: fixed effect model vs random effects model
- Analysing the variation - sub group analysis Variable quality of studies - eg poor study design - some studies are more prone to bias and confounding Publication bias in selection of studies modelling for variation: - fixed effect model - assumes that the studies are estimating exactly the same effect size - random effects model - assumes that the studies are estimating similar, not same, effect size.- often wider CI - can only account for variation but not explain it. |
|
|
What can sub-group analysis do.
|
Sub group analysis can help to explain heterogeneity which may provide further insight into the effect of a treatment or exposure.
2 types: - Stratification by study characteristics: subsets of whole studies are defined eg study design, participant profile - Stratification by participant profile: data is analysed by types of participants eg age group, sex |
|
|
Descirbe how the quality of the studies in meta analysis is assesed.
|
1. Define a basic quality of study required
2. score each study on quality and weight accordingly ( along with odds ratio) Assessing quality of studie: - Allocation methods - Blinding and outcome -Intension to treat - statistical analysis. |
|
|
Describe publication bias.
|
Favourable results/ statisticaly significant, are more likely to be published leading biased selection of studies towards demonstration of effect.
Bias can be detected by - funnel plots( no bias = balanced funnel), - statistical tests( weak tests if bias) - identification method for meta-analysis ( should include searching and identification of unpublished studies) |
|
|
Explain how epidemiological evidence contributes to clinical practice.
|
clinical presentation - symptoms and signs
Disease prevalence - common or uncommon Mortality rates - what is lethal? cohort/case control shudes - is it preventable? Clinical trials - is it treatable. |
|
|
Describe the differences between fixed effect model and random effects model.
|
Heterogenity between studies - Modelling for variation within meta-analysis. 2 approaches for calculating the pooled estimate odds ratio and 95% CI
- fixed effects assumes that the studies are estimating exactly the same effect size - random effects assumes that the studies are estimating similar, but not exactly the same, effect size The point estimate odds ratio is usually similar in both fixed effect and random effects. The 95% CI is often wider in random effects model than in fixed effects. Weighting of studies is more equal inbetween studies in random effects model than in fixed effects where greater weight is given towards small studies. Much debate over which model is better. Random effects can only account for variation, not explain it. |
