Hur tolka resultat i observa/onella studier? Johan Sundström johan.sundstrom@ucr.uu.se
Tolkning av resultat! Slump! Confounding! Bias! Omvänd kausalitet! Kausalitet
Slumpen! Sampling error (urvals- fel) & measurement error (mädel)! Sampling error är den största slumpfel- komponenten i kliniska studier! Bestäm vad som är en acceptabel nivå (sample size- uträkning), och! Bedöm storleken (P- värden, konfidensintervall)
Confounding En confounder är kausalt kopplad Lll både exponeringen och udallet Exponering? Utfall Confounder
Hur handskas med confounding? Definiera noga vad som är en confounder: dagi,y.net Data på potenlella confounders måste samlas in med samma precision som data om exponeringen Minska confounding i designfasen genom:! RestrikLon! Randomisering (intervenlonsstudier)! Matchning (fall- kontrollstudier) Eller i analysfasen, genom:! StraLfiering! Justering
Bias SelecLon bias:! Val av kontrollpersoner (fall- kontroll- studier)! Bias i bordallet (kohortstudier och RCT) InformaLon bias:! Icke- differenlell felklassificering! DifferenLell felklassificering (recall bias, observer bias)
Hur handskas med bias?! Blinda undersökarna! Blinda försökspersonerna för vilken behandling de får! Använd automalserade processer istället för mänskliga observatörer! Använd journaldata/registeruppgiyer istället för minnesberoende rapportering! Om bias inte kan unvikas, uppskaza storleken och riktningen.
Omvänd kausalitet Var extra försiklg i fall- kontrollstudier! Du är bjuden Lll ez bröllop på lördagen. På fredagen går du Lll frisören. När du sizer där i frisörsstolen så påverkar framlden nulden. Bröllopet har inte inträffat ännu. Det kanske aldrig kommer az inträffa. I found a twenty dollar bill on the ground because later I would need it. The only reason for Lme is so that everything doesn't happen at once (Albert Einstein)
Hill s Criteria of Causa/on The minimal condilons needed to establish a causal relalonship between two items. (AusLn Bradford Hill, 1897-1991)! Temporal relalonship! Strength! Dose- response relalonship! Consistency! Plausibility! ConsideraLon of alternate explanalons! Experiment! Specificity! Coherence
Å@a resultat a@ tolka Kom ihåg:! Slump! Confounding! Bias! Omvänd kausalitet! Kausalitet
Interpreta/on Case #1: Garlic pills and giving up smoking Four subjects who smoked were randomized into receiving garlic pills or placebo to help them quit smoking. The two who received garlic pills stopped, the other two did not. InterpretaLon: Garlic pill treatment facilitates giving up smoking London School of Hygiene and Tropical Medicine
Interpreta/on Case #2: Low infant mortality in families where the father wears a silk /e In a study of environmental risk factors for infant mortality, a strong relalon was found between wearing a silk Le and low infant mortality. InterpretaLon: DistribuLon of silk Les among the male populalon would reduce infant mortality. London School of Hygiene and Tropical Medicine
Interpreta/on Case #3: Amyl nitrite and AIDS In the early days of the AIDS epidemic, before HIV was idenlfied as its cause, a study looked at the behaviour of gay men with and without AIDS. An associalon was found between inhaling amyl nitrite (a drug used oyen in gay sex encounters) and AIDS. InterpretaLon: Use of amyl nitrite causes AIDS. London School of Hygiene and Tropical Medicine
Interpreta/on Case #4: Smoking and lung cancer In a case- control study of the associalon between smoking and lung cancer, the smoking habits of lung cancer palents were compared to those of other palents of the same lung clinic. The same proporlon of cases and controls smoked. InterpretaLon: Smoking is not a cause of lung cancer. London School of Hygiene and Tropical Medicine
Interpreta/on Case #5: Recent use of antacids and stomach cancer A case- control study of people diagnosed with stomach cancer found that cases had a much higher usage of antacids in the four months prior to diagnosis than matched controls. InterpretaLon: Use of antacids is a cause of stomach cancer. London School of Hygiene and Tropical Medicine
Interpreta/on Case #6: Gender and risk of heart a@acks A case- control study of people with a heart azack, and matched controls, INTERHEART. Female controls were oyen recruited from screening programs (cervix, mammography), male controls oyen palents with other diseases. The authors found that cardiovascular risk factors were more closely related to heart azacks in women than in men. InterpretaLon: Women would benefit more from prevenlon strategies than men. INTERHEART results have been published five times in the Lancet
Interpreta/on Case #7: Physical ac/vity and risk of heart failure Cardinal symptoms of heart failure are shortness of breath, Lredness and low exercise capacity. The authors conducted a populalon- based cohort study among healthy middle- aged and elderly men, and found that low habitual physical aclvity was related to an increased risk of subsequent heart failure. InterpretaLon: Physical aclvity prevents heart failure. Sundström et al, work in process
Interpreta/on Case #8: Sta/ns and mortality In a large randomised intervenlon study, ALLHAT, the effect of treatment with a staln (cholesterol lowering drug) vs. usual care on mortality was examined. This was during an era when staln use increased dramalcally. The authors found no difference in mortality between the groups. InterpretaLon: StaLn treatment does not save lifes. JAMA 2002;288:2998