EBH Summary

  • Published on
    16-Oct-2015

  • View
    8

  • Download
    0

Embed Size (px)

DESCRIPTION

summary for exam

Transcript

<p>EBH</p> <p>EBH SummaryR1 Intro EBH is Traditional health care based on theories, clinical experience, expert and mentor opinions and patient preferences (Dr. Archie Cochrane) ( this also applies to public health policy not just EBM, but EBH (i.e. service provision and funding, and assessment of need/effectiveness of services)(i.e.disease prevention and health promotion in knowledge of risk factors and environment)</p> <p> John Sinclair coined the term statistics (pieces of info useful to the state). He was the 1st person to calculate cost of living. Health stats are pieces of info useful in providing health services (based on research data). i.e. how can we detect, who likely to get it, how can we treat, etc. Prevalence the proportion of people with the problem ( from cross-sectional studies</p> <p> Incidence the proportion of people who will develop the problem in a given period of time ( follow up/cohort studies</p> <p> Risk Factor = factor modifies your risk of disease up or down, may or may not be causal (i.e. education and diseases) Cohort studies can identify groups with higher or lower incidences of disease</p> <p> Case control studies compare people with and without disease to identify factors that distinguish them</p> <p> How to ID by screening (health prob to intervene early), diagnose ( validate these</p> <p> Screening detecting people with a health problem so we can intervene early</p> <p> Diagnosis idenfiying the disease in a particular person (trying to use test results to outrule or diagnose)</p> <p> How diseases develop over time (info comes from studies of natural history and prognosis) Natural history: the course the disease follows over time (in a specific person, i.e. even in their teens these men were developing fatty precursors in their arteries)</p> <p> Clinical course: the natural history starting with the first observable manifestations (first clinical manifestation in many ppl is an acute coronary episode). </p> <p> Prognosis: the expected course the disease will take</p> <p> How to treat Clinical trials are the primary source of the evidence. Very significant proportion of all medical research. Ideal for simple treatments like drugs, harder to design good evaluations of complex interventions like </p> <p> Practitioner-delivered therapies</p> <p> Community interventions</p> <p> Interventions designed to produce hard-to-measure benefits such as improved quality of life</p> <p> Prevention can operate at each phase of the natural history</p> <p> Primordial: you eradicate the disease</p> <p> Primary: people don't get the disease</p> <p> Secondary: the disease is stopped from getting worse</p> <p> Tertiary: the consequences of the disease are reduced</p> <p> Prevent spread</p> <p> Optimise quality of life </p> <p> Each type of question needs a specific study to answer it (and each study has own features/strengths/limitations)</p> <p>R2 How to read a scientific paper Paper like a supermarket know layout to get everything fast</p> <p> Introduction authors justify the research, IDs context/purpose of it usually has 3 sections: what is currently known problem area, what is not, and general statement/conceptual of research question (Conceptual question in plain language determines relevance of study)</p> <p> Methods translate the conceptual question into research question by defining who will be studied and how it will be measured) ( why is it in: to allow replication, to see if methods turn conceptual q into research q, and to let reader judge if study is good investigation of research q ( should describe: ppl studied (how recruited, selection bias; eligibility criteria, clinically important?; exclusion criteria, lmt applicability of findings?) how things were measured (appropriate to research q, vlid and reproducible, what precautions against bias, all appropriate factors assessed, how study was conducted [data quality and completeness checks] and analyzed) Results 3 sections: characteristics of ppl studied, outcome measures, factors assoc w outcome measures ( Table 1 (describe who was studied, subgroup differences in diff columns [i.e. men/women] versus characteristics in rows) Table 2 (what you found, descriptive stats, CI) Table 3 (relationship b/n who you studied and what you found, measures of effect size [relative risks, odd ratios, differences b/n groups, CI]). Discussion 3 sections: summary of knowledge to date, integration of findings of study into existing knowledge, implications for practice/further research</p> <p> Abstract structured summary of paper (map of supermarket) read first, write last</p> <p>R3 Populations and Samples Our knowledge is based on/only as good as our samples (i.e. you may know little about Japanese food if hardly ever eaten any) </p> <p> Key issues in choosing a sample: Representative and sample size big samples more trustworthy (but also need to be representative and therefore unbiased (no matter how big the sample), a key issue ( i.e. may have lived in NY whole life and suspect other USers more courteous. i.e. eating Japanese in Irish restaurants A population is any group that shares a common characteristic. Can be things/events/ppl (i.e. ppl in Dublin, ppl w angina, ppl admitted to CCU in 2002, all cervical smears sent for reporting in 2002, fish caught in Irish sea, all calls to computer help desk in July, all cyclist accidents) Defining a population: 1st step) inclusion and exclusion criteria population definition determines relevance of study ( 2nd step) once defined population need a representative sample (since cant include whole population) Probability (when data analyzed statistically) Aimed at generalizing the population every member of population has defined probability of being included, where best ones are where every pop member has equal chance. </p> <p> Best way of doing this is A) random (from sample frame = list of pop members, usually from computer generated random #s, i.e. any list which uniquely IDs members of population like dialing random phone extension #s to select employees in a building OR hospital charts ( problems: how get a sample of ppl being treated for high bp (super scattered) so do. Or if groups being compared are unbalanced in population, or B) multistage/clustering sampling to ID naturally occurring clusters in population and then select some of these randomly process has several stages: 2 stage sample (primary sample unit: select school at random, then kids) -3 stage sample (primary sample unit select random health boards, then medical practices w/n each area, then eligible patients with HT at random in the med practice)</p> <p> Needs special statistics members multistage sample more alike than in random (i.e. kids w/n same school!), so less variation! Statistics can be misleading (CI and hypothesis tests especially) so need good modern stat software. </p> <p> C) stratified sampling (if study involves making comparisons b/n groups w diff prevalences, i.e. male/female nurses in random sample there are more women than male nurses so its an unbalanced comparison ( stratify your sample by taking 2 random sample groups, 1 from each group makes comparison b/n groups easier</p> <p> D) systematic sampling (not necessarily random i.e. putting questionnaire into every 10th employee pay packet) kinda random and every employee has an equal chance of being selected unlikely to be biased, but not strictly spreaking random (since selection of 1st participant effects selection of others) but saves a lot of time Non-probability (qualitative for key themes) grab participants most likely to be informative (i.e. key informants, ppl that represent the population like members of parliament represent public). Ppl that can articulate the needs/concerns of ppl they represent and give information if cant ID population members cant do random sample but can still draw a representative one or one with minimum of bias. ( i.e. wanna sample callers to a helpline dont really have option of sampling ALL calls every and picking randomly (or even to specific line) so cant do any random sample ( do you study all calls for a month or every nth call? no answer!</p> <p> grab (convenience sample) taking whatever is available (asking colleages) probable serious bias dont know if representative sometimes used for exploratory research/piloting methodology but NOT to draw conclusions quota (grab samples w specific quotes of ppl) i.e. 10 male/10 female makes comparison easier for same problems w grab use for general conclusions in comparison and same stuff as above snowball use w hard-to access groups (drug abusers) researchers asks recruited person to pass the survey along to others (increases coverage and compliance due to social networks) produces biased results (less biased than simply surveying known pop members) self-selected groups useful in shaping your ideas (ppl who answer to ads, patient advocacy/self-help groups, key informants IDd from asking around) not trustworthy to draw conclusions 3rd step) choose this after relevance and representative issues resovled (due to their bias factors) ( in calculating sample size: precision = amt of uncertainty surrounding findinds of a piece of research (usually expressed using CI) ( need a large sample to increase precision of conclusions (margin of error and confidence intervals will be smaller) ( precision depends on context (i.e. age, km, adults weighed in kg, babies in grammes) so need to define precision of our research question (how much uncertainty can we tolerate in the findings) variability must be taken into account b/c larger samples needed when measuring something that is more variable problems: how much knowledge is enough, how big is big enough, when is knowledge precise enough? larger sample sizes are more expensive!!R4 Treatment trials/Controlled Trials Scientific medicine differs from other medicine sorts by being based on evaluation of evidence for treatment effectiveness (in way that is replicable, public, peer-reviewed) one important fnct of diagnosis to allocate patients to treatment will benefit them most ( until recently, med was based on authority of respected teachers (changing from authoritarianism to one of scientific openness) First randomized trial medical treatment in 1948 Lind 1753 did Scurvy trial (12 sailors as alike as he could make them and formed groups of 2 got diff things) ( problem: practice lags behind evidence even now! (50 yrs before put Linds findings into practice) Phase 1 clinical pharmacology and toxicity (drug is non-toxic to humans, what dosages can be given w/o serious SE, drug metabolism/bioavailability) 20-80 normal volunteers and patients</p> <p> Phase 2 clinical investigation for treatment effect and safety 100-200 closely-monitored patients Phase 3 (clinical trials) controlled evaluation of treatment (compared to current best practice) lots of patients</p> <p> Phase 4 postmarketing surveillance (adverse/rare reactions, long-term consequences)</p> <p> Why not open trials? To monitor new treatment, establish safety, compare value of new treatment to accepted standard (clinically and statistically significant), ensure response is not to placebo/hawthorn effect or selection/observer bias</p> <p> Definition: a treatment trial has 3 key elements: defined treatment, defined patient group, against defined endpoint</p> <p> 4 key methodological features </p> <p> controlled (against defined alternative) compensates for Hawthorne where ppl act diff under observation (b/c baseline/control group receives [as close as possible] same treatment as intervention group except for intervention being assessed for clear interpretation of results)</p> <p> randomized (in allocating patients to ensure comparability [independent of their characteristics]) doesnt allow for patient preference which may be important in determining treatment success blinded (reduce bias) single, double, triple blind (subject or assessor, both subject/assessor, subject/assessor/data analyst) data analyst blind is where data is done w/o any knowledge of what drugs are intention to treat analysis i.e. all ppl in trial (not just those who complete/comply) are counted in, so evaluating the treatment policy i.e. if treatment long the analysis of those who complete the trial is per protocol analysis evaluating a paper treatment (well-described, practical), patient group (well-described, sources of selection bias [tertiary referral center/age cutoff/unusual diagnostic criteria/comorbidity excluded], practical [are they like patients youll see?], patient representatives [exclusion criteria should only be for ppl where treatment wouldnt be an option normally, or where enpoints cant be measured], all patients accounted for?), endpoints (measured using known, valid, reproducible methods SE/adverse and all outcomes documented)</p> <p> then can act on clinical trial if it is shown that the new treatment is clinically significant benefit (acting on basis of clinical trial 1) from treatment being investigated (i.e. drug that lowers cholesterol by 2% is no use) ( also analyze if its statistically significant benefits (clinical trial 2) where treatment benefit must be so large that its unlikely to have occurred by chance in patient group as big as one studied ( interpreting negative trials if didnt show significant treatment effect is that b/c wasnt big enough to have reasonable chance of finding one (i.e. small trial or small treatment effect!) ( is the new treatment realistic in setting (affordable/acceptable/safe) (clinical trial 3)</p> <p> trials versus systematic reviews (integrates evidence from similar trials: each w one dosage, endpoint, patient population eligibility, choice of comparison, etc.) ( latter is basis for decision making since no single trial can evaluate a treatment fully, and systemic gives broader pic/composite estimate of effect of treatmentR5 Cohort studies cohort study follows defined group of participants over period of time and records incidence of events of interest (observational, not experimental) basically a follow up in time (recording all relevant events) comparing exposed and non-exposed cohort to see risks for diseases Incidence: the rate of occurrence of new cases per unit population per unit time Population Cohort Study: follow population (i.e. can see dose-effect relationship of risk factor and outcome general population studies are rare, often specific pop groups and often more risks factors to more diseases, i.e. birth cohort studies for all kids born in defined period can be occupational cohorts, i.e. workers in specific industries looking for hazards) so planning one: involves 3 components: 1) defined healthy population, 2) classify into groups depending on exposure status, 3) follow up to assess defined endpoints</p> <p> Ex. Netherlands Cohort Study on Diet and Cancer is population cohort study that aims to measure incidence of various types of cancers and their risk factors (120,852 subjects multi-stage sampling) physical overactivity inconsistently linked to the silent kill...</p>