|
|
| Estimations
of prognosis may be severely biased when determined retrospectively.
For example, determining prognosis by selecting a sample of patients who
present with a disease or disease complication, and then looking back to
see if they have experienced the same previously, will usually overestimate
the risk of a second event. Those who have experienced the disease
or complication multiple times have greater chance of being entered into
the prognosis investigation. Thus patients with more severe disease
are selected (sackett).
Adequate description of the study population is critical for determining the presence or extent of referral bias, and the appropriateness of the inception cohort. . For example, studies of prognosis conducted in tertiary care or referral centers tend to include a higher proportion of "difficult" cases, and prognosis may appear worse. It is often convenient to use a natural sample, like patients who attend a clinic, and follow them forward in time. This preferentially excludes patients with very severe disease who die early or are too sick to attend, or patients with mild disease who do not feel compelled to attend. For example, a study of MI prognosis that enrolls prevalence cases from an HMO clinic appointment roster might yield very different results from a study that enrolls incidence cases from emergency department admission logs. Prospective population based
investigations often yield reliable estimates of disease prognosis.
Nevertheless, if you are practicing in a tertiary referral center you may
be more interested in studies conducted in this setting. Population based
investigations, often conducted on large databases such as insurance
enrollees (medicare), not on patients presenting with a disease or visiting
a specific type of clinic clinic, help eliminate the biases
associated with selective patient enrollment. In any case, all
measurements of prognosis needs to be evaluated in context of the inception
cohort from which they are derived.
|