Public health and EBM Flashcards
PICO question
P – patient/population the question applies to
I – intervention/treatment being considered
C – comparison/alternative treatment being considered
O – relevant clinical outcome
Sensitivity and specificity
Sensitivity = proportion of patients with a disease who test positive (true positives/total cases) Specificity = proportion of patients without a disease who test negative (true negatives/total non-cases)
Standard error
= standard deviation of the sample
- population SD is usually unknown, so use sample SE instead
- SE = SD / √n
- SE gives narrower range by confidence intervals
- SD gives wider range by reference ranges
Reference range
mean±1.96×SD
or mean±1.96×SE gives confidence interval, narrower
P values
- Null hypotheses = no effect, no difference between populations
- P value = strength of the evidence against the null hypothesis
- at 0.05%, 5% chance of concluding effectiveness of an intervention when it isn’t (arbitrary decision to <0.05!)
Risk difference
Risk difference = risk of outcome occurrence in exposed – risk of outcome occurrence in unexposed
Risk ratio
Risk ratio = risk of outcome occurrence in exposed / risk of outcome occurrence in unexposed
Odds ratio
Odds = patients with disease / patients without disease
Odds ratio = odds of disease in exposed (cases) / odd of disease in unexposed (controls)
approximately = risk ratio in rare cases
How to address confounding
Stratification (but gives smaller samples)
Standardisation, e.g. age-standardised mortality
Regression (multivariable)
Randomisation, e.g. RCTs, genetic epidemiology
When analysing results, consider 5 things…
- is it due to chance?
- is it due to confounding?
- is it due to bias?
- is it reverse causality?
- is it causal?
Rate ratio
- for considering treatment harms commonly
Rate ratio = rate of outcome in exposed / rate of outcome in unexposed
- RR > 1 suggests exposure predisposes to outcome (or harm)
- RR < 1 suggests exposure protects against outcome
- Indicates strength of association but not causality
Number needed to benefit (or harm)
NNTB (or NNTH)
- NNTB = 1 / R0 – R1) = 1 / risk difference
- R0 is rate of outcome in unexposed, R1 is rate in exposed
Heritability
= proportion of total phenotypic variance attributable to genetic effects
- h2 = additive genetic variance / total variance
- For phenotypes arising from a large number of genetic loci
- Only applies to population measured in
Gene mutations
= single nucleotide polymorphism if frequency >1% in population (mutation if <0.05%)
- can be silent, missense (substitute one amino acid for another), nonsense (amino acid converted to stop codon), sense (stop codon converted to amino acid), and can also affect splicing, RNA initiation and termination
- Insertion/deletion (indels)
- Structural variation (eg microsatellites, tandem repeats, translocations, inversion, segmental duplication etc)
Penetrance (impact) depends on magnitude of effect on protein (position and type of variant) and the biological importance of protein
Three aims of pubic health
Health protection
- response to incidents, outbreaks and emergencies
Health improvement
- develop primary prevention programmes via public behaviour (eg ‘Bike it’ to work, plain packaging on cigarettes)
Health services
- set up secondary prevention programmes (eg bowel cancer screening, care quality control in HCAIs)
Disease prevention
Primary prevention
- life course
- avoiding development of disease and removing risk factors
- long term effects on chronic disease risk of physical and social hazards from gestation-childhood-adulthood etc (or even previous generations)
- individual and also addressing the wider cause – poverty, legislation, healthy school lunches etc (from government level)
Secondary prevention
- early detection, treatment and preventing progression
- eg identifying those at risk of CHD (eg NHS health checks) and treating risk factors (eg cholesterol, hypertension) and referring to services (eg weight management, smoking cessation)
Tertiary prevention
- reducing complications of established disease
- eg rehabilitation after cardiac event
Preventative paradox
- many people at low risk -> more cases than few people at high risk
BUT
- a preventative measure which brings much benefit to the population offers little to each participating individual
Population vs high-risk approaches
Intervention for the whole population, or just for those already identified as high risk?
Population strategy
- good - radical (removes reason the disease is common), creates cultural shift (the change becomes the social norm so potentially powerful and long-term impact at the population level), individuals with unidentified risks/problems benefit, and avoids stigma
- eg home visiting to new mothers and babies by health visitors includes Postnatal Depression questionnaire, traffic light system for labelling foods in supermarkets
High-risk strategy
- good - prevention is appropriate to the individual, so high acceptability, easily implemented, cost-effective use of medical resources (although may involve screening costs), selectivity improves the benefit to risk ratio, and opportunity to address inequalities (often the people who engage with a population intervention are those that need it least, so might only serve to widen inequalities)
- eg C-card condom distribution scheme for under 25s in areas of deprivation, follow-up of recently discharged psychiatric patients to reduce risk of suicide
-> So should combine by ‘proportionate universalism’ – universal service for all, but with additional targeted services for those more at risk
Passive vs active immunisation
Passive
- Antibodies given to individual (immunoglobulins produced by B or T cells, Y shaped, two-headed - neutralise toxins – eg diptheria, tetanus vaccines, neutralise viruses, kill cells directly or with complement, block microbial adhesion/cell entry, promote opsonisation and phagocytosis)
- Instant but temporary, and risks involved
Active
- Needs antigen exposure (anything that can bind to antibody, B or T cell (usually polysaccharide proteins))
- Takes time, but long(er) lasting and low risk
Innate vs adaptive immune response
Innate/natural
- physical barriers, physiological factors, protein secretions, phagocytic cells
- instant response, no memory
Adaptive/specific
- B and T cells
- 2nd level of defence, specific, response is better but slower, has memory
Innate and adaptive work together (eg opsonisation of phagocytes to make more effective)
Vaccine process
- Vaccine/antigen administered
- Recognised by naïve B cells
- Recruit T cells
- T cells make memory cells and plasma cells
- Plasma cells produce antibodies
- Memory cells activate the immune system for next time
- IgM first
- Then IgG second (makes most of immunity)
Live attenuated, or dead (inactivated, toxoid, subunit, conjugate)
Childhood immunisations
Types - universal rolling, universal catch up, or target programme
2 months - rotavirus, PCV (pneumococcal), menB 3 months - rotavirus 4 months - PCV, menB 1 year - PCV, MMR, menB Pre-school - annual flu nasal spray, MMR Girls 12yo - HPV 14yo - tetanus, diptheria, polio 19-25yo - MenACWY MSM - HPV 65yo+ - PPV, annual flu, shingles
Reproduction number (infectious disease control)
Basic reproductive number
= the average number of secondary cases produced by one primary case in a wholly susceptible population (measure of intrinsic potential for infectious agent to spread)
- R0 = probability of effective contact (eg condoms) x number of contacts (education) x duration of infectiousness (screening)
Effective reproductive number
= R = R0 x proportion susceptible
Herd immunity = proportion of population that need to be immune for a disease to become stable (when R = 1)
- if R<1, disease dies out
Positive and negative predictive values
Positive predictive value
- the probability of having disease if test positive
- PPV = True positives / (TP + false positives)
Negative predictive value
- the probability of being disease free if test negative
- NPV = True negatives / (TN + FN)
Potential biases in diagnostic study
- spectrum bias – type of patients recruited (need to use grey area cases where there is clinical uncertainty as well as classical cases, test should be performed on the group of patients who it would be applied in the real world)
- work up/verification bias (all patients should receive both diagnostic and gold standard tests, even though gold standard tests often more expensive and invasive, so clinically unwilling to perform on ‘normal’ patients)
- loss to follow-up bias
- reporting/review bias (how were tests done, and were they blinded)
Interpreting diagnostic test results
Sensitivity = TP/TP+FN Specificity = TN/TN+FP PPV = TP/TP+FP NPV = TN/TN+FN
+ve Likelihood ratio = sensitivity / (1 – specificity)
-ve LR = (1 – sensitivity) / specificity
Likelihood ratios
+ve likelihood ratio = probability of +ve test result in people with the disease / the probability of +ve result in people without the disease
- LR+ = sensitivity / (1 – specificity)
(How much more does a +ve test occur in people with disease compared to those without disease?)
-ve likelihood ratio = probability of person who has disease testing -ve / probability of person who does not have the disease testing -ve
- LR- = (1 – sensitivity) / specificity
(How less likely is a negative test result in people with the disease compared to those without the disease?)
Likelihood ratios giving values close to 1 indicate no better than random
- but eg if LR+ = 18 for strep rapid antigen test, then a positive test is 18x more likely in a person with strep than a person without strep
Before applying a test to patients
- Satisfactory reproducibility?
- Generalisable results to my patient load?
- Results likely to change my management?
- Will patients benefit from the test results?
Can use if only low pre-test probability of disease:
- Screening for disease in clinical population
- Test is cheap and has reasonable LR
- Test does not cause distress or harm
- Disease can have atypical / variable clinical features
- Effective therapy which will alter prognosis
- Waiting until presentation more obvious -> bad effect on prognosis
Screening outcomes and consequences
True positives
- sometimes helpful – outcome better because of early intervention
- sometimes not helpful… - outcome good but early detection made no difference, outcome poor and early detection made no difference, condition would never have had any impact so intervention was unnecessary (and anxiety and overtreatment is detrimental to the individual)
False positives
- unnecessary anxiety, followed by relief or anger
False negatives
- anger or loss (though less now, as people understand that in intervals between screenings disease can develop)
- due to miscommunication/mistakes, unavoidable eg tumour doesn’t show on mammogram, below threshold at first point (detectable only on hindsight), genuine miss despite training, or rapid growing cancer develops in interval
True negatives
- no improvement in health, but maybe reassurance
- but poor use of resources! (and maybe increased initial anxiety)
… screening needs to balance maximum benefit and minimum harm
Three key biases in evaluating screening programme via RCT
Healthy screenee effect (the type of people that elect to attend screening are biased towards being healthy/educated)
Length time effect (screening is best at picking up long-term slow pathological conditions which are biased towards good prognosis, not others) = overdiagnosis bias + inability to pick up fast developing conditions
Lead time bias (screening looks like they extend their ‘survival time’ when actually they will still die at the same time, they’ll just have been diagnosed sooner)
Technical vs allocative efficiency
Technical efficiency
- producing output in the best way possible, without wasting scarce resources
- eg is it better to vaccinate against meningitis, or give abx for meningitis
Allocative efficiency
- producing pattern of output that best satisfies the pattern of consumer wants/needs
- eg should we vaccinate against meningitis, or perform more cardiac surgeries, or provide more social housing
Quality adjusted life years
QALY
= quantity x quality of life
Q (quality):
1 = best imaginable health, 0= dead (or as bad as), 0.5 = intermediate health state
- valued by time trade off (20 years (Y) in this health state, or how many years (t) in best imaginable health until death, QoL = t/Y)
ICER
To decide cost-effectiveness (worthiness) of intervention
ICER = Incremental cost-effectiveness ratio (costs/QALY)
- ICER < £20k/QALY – accept, unless evidence base is poor
- ICER £20-30k/QALY – judgement, referencing certainty of evidence and other factors including end of life
- ICER >£30k/QALY – reject, unless strong arguments against
Decision to either reject (not free on NHS), fund but for limited period with mandated RCTS (rare), or fund
- achieving allocative efficiency, technical efficiency, and equality in priority setting
Ecological studies
- look at average exposure vs rate of outcome
- good for generating hypotheses for other studies, cheap and quick, no ethical issues
- but ecological fallacy (assuming that average characteristics apply to the individual eg smokers get lung cancer), hard to control confounding, and dependent on previously collected data
(bottom of hierarchy of evidence)
Cross-sectional studies
- measures prevalence of disease in a population at a snap shot moment, with any measurement of risk factors in the same point in time
- very good for measuring true disease burden in population
- but don’t measure incidence, are susceptible to reverse causality, measurement bias and selection bias
Case-control study
- recruit available cases and comparable control group, so sample determined by outcome, retrospectively assessing exposure to risk factors and presenting results as odds ratios
- good for rare conditions, can test for multiple exposures and get quick results
- but retrospective so reverse causality, selection bias (hard to select controls fairly) and measurement bias (recall and interviewer)
Cohort studies
- select population based on initial absence of disease, whilst measuring exposure to defined factors at baseline then followed up. Results of risk ratio recorded
- good as reduces reverse causality and selection bias, allows testing of multiple outcomes and better control over confounders
- but retrospective gives recall/interviewer bias and reverse causality, whilst prospective can take years, is expensive and vulnerable to loss to follow up bias. Selection bias as recruitment and attrition is associated to exposure and disease, and is inefficient for rare diseases
Qualitative studies
- eg observations, interviews, focus groups, documents, oral history, then data analysis and coding
- good for ‘why’ things happen, for hypothesis generation, for complex and subjective outcomes
Randomised controlled trials
- interventional, two-armed studies, need clinical equipoise and informed consent
- good for causality (only study which can), best confounder control, allocation concealment and blinding to reduce selection and measurement biases
- but may have poor generalisability if inclusion/exclusion criteria too restrictive, sample size needs to be large (power>80%) so expensive to conduct, and can still have selection/performance/detection/attrition bias
Systematic reviews
- identify all relevant evidence on a given clinical question, appraising validity and synthesising results, minimising biases and random errors, in a way that should be entirely reproducible
- highest quality of evidence, to guide clinical/public health practice, and cheaper and quicker than more RCTs
- but only as good as the research available, research agenda bias and reporting biases (publication, time lag, language, and multiple publications)
Meta-analyses
- statistical analysis combining results of similar studies (similar PICO)
- fixed effects for homogeneity, assuming that any variation is only due to sampling error
- random effects for heterogeneity, assuming that large and small studies provide different information (treatment effect is context dependent)
- present forest or funnel plots
(funnel plot symmetrical - no bias, asymmetrical - reporting bias)
Prevalence and incidence
Prevalence = number of individuals with a disease / total population at risk
Incidence = number of new cases of a disease in a given period / population at risk that were initially disease free
Incidence rate = number of new cases of disease / (population at risk x time interval)
Standardised mortality ratio
= (number of observed deaths / number of expected deaths) x 100
Risk vs odds
Risk = disease/total
Odds = disease/non-diseased
Confidence intervals
Range of values within which we are 95% confident that the true population value lies
SpPin and SnNout
SpPin = when a test has a high specificity, a positive result rules in the target disorder
SnNout = when a test has a high sensitivity, a negative result rules out the target disorder
Live vaccine
- eg MMR, BCG, oral rotavirus, oral polio, nasal influenza
- gives strong immune response, longer lasting
- but may get mild infection 7-10days after (or severe if weak immunity), can’t be used in pregnancy, must be kept cold
Conjugate vaccine
- eg HiB, Men C, pneumococcal 13
- enable development of immunity to non-protein material,
- but smaller response, less long lasting, can’t be given mucosally
Toxoid vaccine
- eg diptheria, tetanus
- no risk of infection
- but smaller response, less long lasting, can’t be given mucosally
Subunit vaccine
- eg pertussis, MenB
- no risk of infection
- but smaller response, less long lasting, can’t be given mucosally
Monogenic vs multifactorial genetic conditions
Monogenic = high heritability, low prevalence, studied via linkage studies
Multifactorial = low heritability, many genes involved, environmental influences also
Inverse equity hypothesis
The availability of good medical care tends to vary inversely with the need for it in the population serve
(so public health interventions don’t reach those who really need it)