CHLH 274Quiz 3 Evaluating the role of random errorThree ways to determine if studies are valid- Random error- Bias- Confounding- All must be eliminated for it to be valid You can calculate the probability that the measure of association you observed was due to chance by hypothesis testingHypothesis testing- You are performing a statistical test in order to get a P value- A statistical test quantifies the degree to which sampling variability or chancemay explain the observed association - The null hypothesis is assessed by a statistical test that gives you a P value- Definition of P value- given that the null hypothesis is true, the p-value is the probability of seeing the observed result by chance aloneP Value- P value ranges from 0 to 1- Tells you the extent to which the null hypothesis is compatiable with the data- Small P value indicates a low degree of compatibility between the null hypothesis and the observed data- A small P value implies the alternate hypothesis is a better explanation for the data- Small P values indicate that chance is an unlikely explanation for the result- P value test hypothesis for statistical significance- P<0.05 significant, not due to chance- P>0.05 not significant, due to chance - P<0.01 highly significant, more confident not due to chance- Example study RR= 1.4 and P= .1- since the P value is not less than 0.05 these results are not considered “statistically significant”o These results indicate that the best estimate of the increased breast cancer risk associated with pesticide exposure is 1.4 Point estimate- Actual measure of association given by the data- Given sampling variability it is important to indicate the precision of the point estimate i.e. give some indication of sampling variability- This is indicated by the confidence intervalConfidence Intervals (read all slides)- Quantifies sampling variability - Strict statistical definition- if you did the study 100 times and got 100 pt- Estimates and 100 CI’s, in 95 of 100 results, the true point estimate would liewithin the given interval. In 5 instances the true point estimate would not lie between the given interval - Note that the point estimate is RR or OR- Confidence intervals estimate the value of the parameter within a margin- Example pesticides and breast cancer- RR= 1.4 95% CI= .07-2.6o Results indicate that the best estimate of the increased breast cancer risk with pesticide exposure is 1.4 however we are 95% confident thatthe true RR lies between .07 and 2.6o That is the data are also consistent with hypothesis of .07 to 2.6 P Value & CI- Both tell you nothing about the other possible explanations for an observed result: bias and confounding- Both tell you nothing about biological, clinical or public health significanceBiasInternal validity - The ability of a study to measure what it sets out to measure- Inference from participants in a study should be accurate- Study must be free of biasExternal Validity- The ability to extrapolate from a study to the general population- Important b/c a total population census approach is usually impossible and expensive- Usual tactic: choose a sample, study it, and hopefully extrapolate the result to the general populationError- Any epidemiological study presents many opportunities for error in relation to- Selection of participant, classification and measurement, and comparison/interpretationBias- Is a systematic error in inference that result in an incorrect(invalid) measure of association- Bias is primarily introduced by the investigator or study participants- example maybe didn’t think through selection protocol or bias by the way participants answer question- A poorly designed and conducted study will have bias- Bias can arise in all types of studies- Two main types selection and observation biasDirection of bias- Toward the null means effects are underestimated- Away from the null means effects are overestimated- Positive association is biased toward the null value- Preventative association is biased away from the null value- See diagramsSelection Bias- Selection of study participants in a way the favors a certain outcome- Stems from a absence of comparability between groups being studied- Most likely to occur in case-control studies because exposure and outcome have occurred at the time of study selection - Can also occur in prospective cohort and experimental studies from differential loss to follow upCase Control Selection bias- Two types berkson and neyman- Berkson- admission rate bias, results from differential rates of hospital admission for case and controls- Neyman- incidence- prevalence bias or selective survival biaso Arises when a gap occurs between exposure and selection of study participantso Crops up in studies of diseases that are quickly fatal or transiento Creates a case-group not representative of cases in the communityCohort Selection Bias- Occurs when exposed and unexposed subjects is not independent of outcome- See example slidesSolutions selection Bias- Little or nothing can be done to fix this bias once its occurred- You just need to plan very well in design phase- Can avoid it during design phase by using the same criteria for selection casesand controls, obtaining all relevant subject, obtaining high participation rates,and taking into account diagnostic and referral patterns of disease Observation Bias- An error that arises from systematic difference in the way of information of exposure or disease is9 obtained from study group - Results in participants who are incorrectly classified as either exposed or unexposed or as diseased or not diseased- Occurs after the subjects have entered the study- Recall bias-people with disease remember or report exposures differently than those withouto Can result in under-estimate of measure of associationo Solutions: use controls whoa re themselves sick; use standardized questionnaires that obtain complete information, mask subject to study hypothesis - Interviewer bias- systematic differences in soliciting, recording, and interpreting information o Can occur in case control or cohort studieso Solutions: mask interviewers to study hypothesis and disease or exposure status of subjects, use standardized questionnaires or standardize methods of outcome - Information bias (interviewer?)- a loaded question is a question with a false, disputed or question begging presuppositionObservation Bias (random or one direction)- Is it random or in one direction?- Misclassification-