Systematic Errors/Accuracy

ROC

(receiver operation characteristics): line graph that plots the probability of a true positive result against the probability of a false positive result for a range of cutoff points (help define cut-off value of test and compare performance of two tests)

publicity bias

(awareness bias): media-driven attention so people report differential experiences; people living close to nuclear plant are exposed to poor health are more likely to report symptoms more frequently and perhaps over-report

loss to follow-up

(if the likelihood of being lost of follow up is related to outcome status and exposure status); patients with multiple sclerosis is a severe and progressive disease so study participation is a burden to continue

performance of test in a population

PPV, NPV, impact of disease prevalence, sensitivity and specificity on predictive values

specificity

ability of test to identify correctly non-affected individuals = proportion of persons testing negative among non-affected individuals TN /(TN+FP) = true negative/ true negative + false positive affected by: characteristics of non-infected people (diversity may increase FPs)

sensitivity

ability of test to identify correctly the affected individuals = proportion of persons testing positive among affected individuals TP/(TP + FN) = true positive / true positive + false negative affected by: characteristics of infected people (antigenic characteristics of pathogen in the area, not picking up ppl in the area)

high standard quality study characteristics

appropriate study design valid and reliable measurements control for possible bias good cooperation between research group and study population

ways to reduce subject bias

blind the study participants as to the hypothesis under investigation, use questionnaires that are carefully constructed; ask specific as well as multiple questions; for socially sensitive questions (alcohol/drug use) use a self-administered questionnaire instead of an interview; if possible, asses past exposures from biomarkers or from pre-existing records

ways to reduce observer bias

blinding, use standardized questionnaires consisting of closed-end, easy to understand questions with appropriate response options, train all interviewers to adhere to the question and answer format strictly, with the same degree of questioning for both cases and controls

random errors

chance; lack of precision always present and influence the precision of a study; random error doesn't affect the average, only the variability around the average

sensitivity and specificity are:

characteristics of a test and population does not affect results

observer bias

consistent distortion of measurement by observer; more intensive measurement in certain subjects (blood pressure in treatment and control groups; better response from educated relative to illiterates); ask questions about specific exposures several times of cases but only once of controls (smoking and lung cancer)

subject bias

consistent distortion of measurement by study subject recall bias: selective recall of an event reporting bias: related to issues of social desirability or sensitivity

methods to control confounding

design: randomization, restriction, matching analysis: stratification, statistical modeling (multivariate)

selection bias definition

distortion in the estimate of effect resulting from how study subjects are selected and from factors influencing their participation

confounding

effect of extraneous variable that entirely or partially explains the apparent association between the study exposure and the disease - Independent risk factor - Unevenly distributed among exposed and non-exposed - Not on the causal pathway between exposure and the disease

instrument bias

faulty function of an instrument (biased scale giving a higher/lower reading that actual); inappropriate use of technique or tool to objective of measurement (endoscope not reaching to the site of interest and giving false info from a distance; leading questions)

choice of sampling frame

how we select, such as population registry, telephone book etc

Berson's bias

if hospital cases and controls are systematically different from one another because of the combo of exposure and disease under study that increases the probability of admission to hospital (alcohol on liver cancer, select gastrological and liver department patients and control group is from another department (trauma), maybe trauma patients are more likely to use alcohol instead)

precision

in epi measurements corresponds to the reduction of random error (obtaining similar results with repeated measurements; reliability, repeatability, reproducibility, consistency, agreement)

how to deal with random error

increase sample size, repeated measurements

precision and accuracy limited by

instrumentation and data gathering techniques

self selection bias

likelihood to participate is related to the exposure and disease (volunteers can be different from the general population, perhaps more health-conscious

factors influencing PPV

sensitivity, specificity (more specific, less false-positive results), and prevalence of disease (low prevalence, low pre-test probability for positives = more false positives; high prevalence, high pre-test probability for positives = more true positives)

performance characteristics of a test:

sensitivity, specificity, choice of a threshold

sources of bias

methodological aspect of study design or analysis selection of subject quality of info obtained misclassification confounding

differential referral or diagnosis

more intensive interview to desired subjects (pulmonary disease and smoking effect. Physician might be more prone to select the subjects who smoke, or might look at his xrays more intensively, and less likely to diagnose those who don't smoke)

systematic errors

nonrandom error in the collection, analysis, interpretation or publication of data that can lead to conclusions that are systematically difference from the truth (inaccurate results) does affect the average (affect accuracy, influence validity)

types of information bias

observer, subject, instrument

sources of random error

sampling errors, measure error, individual biological variation

major types of bias

selection bias information bias/misclassification (observer, subject, instrument) confounding

types of selection bias

self selection, choice of sampling frame, publicity bias, Berkson's bias, loss to follow-up, health worker effect, differential referral or diagnosis

factors influencing NPV

sensitivity (more sensitivity, the less false-negatives, specificity, and prevalence of disease (low prevalence, high pre-test probability for negatives = more true negatives; high prevalence, low pre-test probability for negatives = more false negatives)

increasing threshold in test:

sensitivity decreases, specificity increases

decreasing threshold in test:

sensitivity increases, specificity increases

information bias/misclassification

subjects are incorrectly categorized with respect to their exposure status or outcome

accuracy

the degree to which a measurement/estimate based on measurements, represents the true value of the attribute that is being measure (obtaining results close to the truth; validity, conformity, lack of bias)

validity

the degree to which a variable actually represents what it is supposed to represent (comparison with a reference standard) (accuracy, conformity) > systematic error bias internal vs external

reliability

the degree to which a variable has nearly the same value when measured several times (comparison among repeated measure) (precision, reproducibility) > random error

NPV

the probability that an individual testing negative is truly non-affected = proportion of non-affected persons among those testing negative TN/(FN+TN) = true negative / (false negative + true negative)

PPV

the probability that an individual testing positive is truly affected = proportion of affected persons among those testing positive TP/(TP+FP) = total positive/ (total positive + false positive)

healthy worker effect

using the general population as a comparison group for an occupational study; mortality rate of people in a certain occupation to the general population; may be lower in occupation because they may be healthier (perhaps those who aren't working are ill)