🧪 This platform is in early beta. Features may change and you might encounter bugs. We appreciate your patience!
detection_bias
Detection bias occurs when the process of identifying or measuring outcomes differs systematically between comparison groups. If one group is monitored more closely, tested more frequently, or assessed by evaluators who know the treatment assignment, differences in detected outcomes may reflect the surveillance intensity rather than genuine treatment effects.
In a drug safety study, patients on the experimental drug receive monthly blood tests while the control group is tested annually. The drug group shows a higher rate of liver enzyme abnormalities, but this is largely because abnormalities were caught more frequently through more intensive testing.
A workplace wellness program screens participating employees for hypertension every quarter, while non-participants are only checked at their annual physical. The program appears to be associated with higher rates of diagnosed hypertension, but the difference reflects more frequent measurement rather than a true increase in disease.
A study comparing depression rates between urban and rural populations relies on clinical diagnosis records. Urban residents have greater access to mental health services and are diagnosed more often, leading researchers to conclude urban living causes more depression — when in reality rural depression is simply underdetected.
Binary (yes/no) questions an LLM must answer to identify this aspect:
Were outcomes assessed differently between the treatment and control groups?
Type: binaryDid one group receive more frequent monitoring, testing, or follow-up than the other?
Type: binaryCould knowledge of group assignment have influenced how outcomes were detected or classified?
Type: binaryWere outcome assessors blinded to participants' group assignment?
Type: binaryDetection bias occurs when the process of identifying or measuring outcomes differs systematically between comparison groups. If one group is monitored more closely, tested more frequently, or assessed by evaluators who know the treatment assignment, differences in detected outcomes may reflect the surveillance intensity rather than genuine treatment effects.
The harder you look, the more you find. When one group is scrutinized more intensely than another, more outcomes are detected in that group regardless of whether the true rate differs. This asymmetry in observation is easily mistaken for a real difference in outcomes.
Ensure identical follow-up schedules, testing protocols, and assessment criteria for all groups. Blind outcome assessors to group assignment. Use adjudication committees with predefined criteria to standardize outcome classification.
Screening programs create detection bias on a population scale. Countries that screen aggressively for prostate cancer detect more cases and report higher incidence, but much of this reflects overdiagnosis of slow-growing tumors that would never have caused harm.
Researcher expectations systematically influence how observations are recorded.
Systematic differences in care or treatment between groups beyond the intervention studied.
Measurement error that differs between comparison groups, biasing results in either direction.
How participants are identified or recruited systematically distorts the sample.
A statistical artifact where the average of every group improves when members are reclassified from one group to another, without any actual improvement in individual outcomes. Named after Will Rogers' joke: 'When the Okies left Oklahoma and moved to California, they raised the average intelligence in both states.'
Temporal trends or changes in practice during a study period distort comparisons.
Differential accuracy in remembering past events between study groups.
Systematic error arising from faulty or poorly calibrated measurement instruments.
An interviewer's expectations or behavior systematically influence participant responses.
Use these tools to detect, analyze, or train this aspect.