L9 Causality Flashcards
LO2: recog that assocs may be present in absence of true cause eff rel. epidemiol study design, sys and rand var, alt assocs.
Poss expos for assoc-
sys and rand var-
-Chance- meas by p val. Pt estim by 95% CI. Never prove purely due to chance. P val=likelihood of res by chance if null hyp true.
-Bias- incorr.
Selec bias wrong peop- unrepresentative of pop being studied. Grp comparison not like with like. Sys diffs in chars due to way selec.
Info bias- wrong info- diff recall, obs, meas, classic. Sys diffs in meas or classic of subj in grps.
Bias= deviant res/ intervenes from truth, or proc that causes such deviant. Arises from any trend in collec, anal, interp, public, or Rev of data that can= conclus and systematic diff from truth. Bias invas conclusions and gens that might be drawn.
-confounding (erroneous)- kn facs eg age, sex. Poss facs eg depriv, unkn facs eg genet. Meas of eff of expos on risk of outc is distorted due to assoc of expos with other fac that infls outc of int. Probab lot unkn and unsusp confounder aff epidemiol research.
-alt assocs- obs assoc may be:
Due to unkn confounding, us only partial.
Due to comm cause eg chronic bronchitis not cause lung CA- both caused by smoking.
Rev causality- rel in other direc. Interven not work.
A true assoc.
-assoc= stat dep btw 2+ events, chars or other variables. NOT=causality.
LO1: expl what’s meant by a cause eff rel in epidemiol.
-reductionist appr- est cause by find out all steps that lead to event.
-epidemiol- if someth not occ then event stops, steps not matter. Are many spurious corrals.
-historic perspective- and Greece can’t know true val. 16C bacon sys obs/expos. 18C Hume induction not sec basis for est sci laws.
20C Popper- appr of refrig at, laws tested, test what trying to disprove, null hyp. Kuhn knowl paradigms, laws as temp soc contr its, no cert.
-apprs to dis causat-
Holistic- miasma so, soc facs, ecology.
Reductionist- germs, biol facs, ?genetics.
-kochs postulates-
NECESS- Ag present in ev case of dis and precedes dis.
SPECIF- not in other dis. WRONG.
SUFFIC- Ag can repro dis in expos animus and must be recov from exp dis produced. All that is req. one cause can= dis once predisp and precip facs in place.
Measles necess and suffic. MTB necess NOT diffic as alone not cause- can be latent. Radon suffic for lung CA not necess. Poor liv conds not necess or suffic to cause TB.
-curr- expos neither necess (other causes) nor always suffic (need other facs). Eg alc and liv cirrhosis. Ion is rad and leuk. Smok and HD. Traffic speed and pedest inj.
-epidemiol context- dis from interplay host, env and agent. Cause= expos/fac incr probab of dis. Expos NOT have to be either necess or suffic to be imp. Aim to use knowl gained to rem, av, protec Ag harmf facs. Rev fac to decr probab. Most dis multi fac. One expos can =sev dis eg smok. Dis us res of interac btw host and env eg imm, virulence, maln etc. If cause= incr risk, cause eff rel not have to be absol/direct but must be real not artefactual.
LO3: eval strength of ev in fav of cause eff rel.
-Bradford hill criteria for causality 1965. Start hyp. Study design to test. Validate any assoc by excl alt expos eg chance, bias, confounding, Rev causality. Consid if stat assoc repres cause eff rel.
assoc feats:
-Strength of assoc- strng eg large ratio less likely expl by undetec confound/bias to extent that it’s wrong. BUT not always true, and weak assoc can be causal.
-Specificity of assoc- if outc assoc with one specif fac more likely causal. BUT lack specif not necess weaken case eg smoking. Most dis now seen as multi fac-lot factors may /not be rel. eg asbestos and mesothelioma.
-Consis of assoc- if assoc seen in diff studies and diff sub grps, unlikely due to same confounding/bias. BUT inconsis poss due to diffs in other causal facs eg imm defic. Or due to study design.
expos/outc:
-Temp seq- imp in est when not possib to be causal. Good study design RCT or prospect coh. Weak X sec prevalence, CC.
-Dose resp- If diff. Levs expos= diff risk of outc. Unlikely to to unkn confound/bias on all levs. BUT eg smoking when drink. Lack of biol grad not rule out causal eg thresh eff, J/U shaped rel.
-Reversibility- rem/prev factor=decr or prev risk of outc. Strongest ev but diffic to demonstr. Lot dis have time lags. Ethic iss of RCT of preven prog-make start again. Public health prog to prev/rem expos often reqs action by soc.
other ev:
-Coherence of theory- if obs assoc conforms with curr knowl, paradigms, constructs, theories. BUT inapp rejec of unfav assoc in public bias tow studies that supp theories or interventions is bad.lack coherence not rule out causal eg H pylori.
-Biol plausib- mech or demonstr. BUT lim by curr knowl eg survey. Also presence of demonstr biol mech not necess=causal eg Mg for MI.
-Analogy- with other sp, dis, settings. Easier to infer than biol plausib mech. BUT can be inapp, weak argument.
LO4: desc how to disting causal from non causal assocs. epidemiol reasoning, heir of ev, how to read paper BMJ.
-epidemiol= study of distrib and determinants of health rel states or events in specif pops, and applic to contr health probs.
-assumps in epidemiol- dis not occ rand. Dis has causal and preventable facs that can be id’d by sys investig.
-imp to diff btw causal and non causal assoc because- atten to decr RFs eff pop health. Atten to facs assoc no causally with dis waste time, eff, money.
Necess, specif, suffic agent=dis. Host, env and agent=dis. Multi fac web of facts=dis.
Dev for causality criteria=probabilistic appr to consid causality based on assessm of liklih/risk of dis occ.
-epidemiol reasoning-
Hyp- gen from obs and/or theories. Clinic prac/studies.
Analytical study- sys obs of comps. Test hyp.
Obs assoc- poss expos of non causal assocs (CBS +/- Rev causality). Excl alt expos.
Cause eff rel- based on judgement of how the obs soc fits ev from other sources. In light of curr knowl.
-heir of ev- sys Rev and meta anal. RCTs. Coh. CC. X sec surveys. Case reps not usef eg MMR.
LO5: scheme for ass causality.
-desc of ev: what’s the expos/interven. Outc. Study design. Study pop. Main finding.
-int val- non causal assocs. has method to min bias. Are there confounder so. How likely res due to chance.
Feats of causality- is rel strong, dose resp, specif of assoc, corr time rel or is Rev causality poss.
-ext val- gen of res- can apply to source pop. Or other pops.
-comp res to other studies- particularly from more pow design. Does tot ev sugg specificity. Is there biol mech.
