Introduction to diagnosis

What is diagnosis?: The process of labeling someone with a disease or risk factor for disease

What makes a good study about diagnosis?: 1). Is the test relevant? Will it help me improve care of my patients? 2). Is the test reliable? How reproducible is the test? If two people do it, will they get the same result? 3). Is the test valid? Was the study designed to avoid intentional and unintentional bias? 4). Is the test accurate? Is it negative in those without disease (specific) and positive in those with disease (sensitive)? Choose a high quality reference standard. Mask or blind participants.

Avoid “spectrum bias”: not on very health or sick persons. This type of “case-control design” creates a bias that makes the test look more accurate than it actually is. Avoid “spectrum bias” by using a cohort design. Use a “cohort design” by studying patients with suspected disease, rather than perfectly healthy and very sick. What is the best population for your study? For pancreatic cancer from abdominal pain and jaundice. For colon cancer from annual physical.

Key

The threshold model helps us understand the diagnostic process: when we can rule out a disease, when we can rule it in, and when we need more information. A good test is relevant (provides information that improves patient outcomes), valid (accurate), and reliable (reproducible). Well designed studies of diagnostic tests use a valid reference standard, apply it to all patients, and study a relevant spectrum of persons with and without disease (ideally not known ahead of time)