Discussing hidden bias in observational studies

Article Abstract:

The randomized controlled experiment is regarded by many medical scientists as the gold standard for research. However, data based on observations must not be automatically dismissed as lacking scientific rigor. One reason is that certain experiments are neither ethical nor practical. For example, a randomized group of people cannot be asked to smoke cigarettes for 20 years to confirm the link with lung cancer. A second reason that observational studies should not be dismissed is that they may indeed be quite valid although lacking random control groups. However, observational studies may have hidden bias which must be accounted for. A observational study might, for instance, find that people who take vitamin C have a lower rate of cancer. To conclude that vitamin C prevents cancer is to overlook that the people who take vitamin C regularly may also eat different foods and have different habits which alters their risk of cancer independently of vitamin C. Can the effects of hidden bias be eliminated? The answer is, no. But the author presents methods by which the possible effects of bias can be estimated. By estimating the characteristics of the hidden bias which would be necessary to explain the observed results, it is then possible to draw reasonable inferences from the observed data. For example, when the first studies found that lung cancer was more likely among people who smoked, some researchers objected that perhaps smokers were different in other ways. Perhaps the same genetic makeup which makes some people prefer smoking also makes them more liable to some cancers. If such a hidden bias is presumed to exist, then it is possible to calculate that the bias factor, whatever it is, must lead to cancer in virtually every instance. Furthermore, the bias factor must be 10 times as common among smokers than nonsmokers. No one has been able to offer a suggestion of any sort of factor which comes close to these requirements. While this example is perhaps extreme, it is clear that the estimation of possible bias can reveal the amount of confidence which may be placed in the results of an observational study. (Consumer Summary produced by Reliance Medical Information, Inc.)

Medical statistics

---

Survivor treatment selection bias in observational studies: examples from the AIDS literature

Article Abstract:

Statistical biases associated with differences in patient health may confuse the conclusions of observational research studies. Observational studies follow patient health as the patient and doctor freely choose the kind and timing of treatment. Examples from the literature on AIDS illustrate how these patient health biases can alter the survival benefits of drug treatment. There are also ways to better calculate and present survival results. Patients who are sicker at the start of a study will more likely choose treatment earlier or may die before starting treatment than healthier patients. Without weighing these variables in the statistical analyses, researchers can show that ineffective treatment improves survival. Statistically weighting the patients' changing health status, the time treatment begins, and the time that disease develops can provide a clearer and more balanced picture of a drug's effectiveness.

Care and treatment, AIDS patients

---

When to base clinical policies on observational versus randomized trial data

Article Abstract:

A model is presented for helping researchers decide whether to use readily available observational data in electronic databases or mount an expensive randomized research study when studying treatment practices. This model is based on predictions of cost-effectiveness and relative health benefits. Ways to estimate the number of study participants needed to arrive at meaningful conclusions is also included. Uncontrolled variables inherent in observational data have the potential to lead researchers to misinterpret the relationship between the proposed treatment change and its effect.

Planning, Usage, Clinical trials, Centralized databases

Similar abstracts:

• Abstracts: Vulvovaginal human papillomavirus infections: clinical implications and management. An evaluation of human papillomavirus testing for intermediate- and high-risk types as triage before colposcopy
• Abstracts: Understanding human genetics. Charles Darwin and the foundations of clinical genetics in dentistry. Nutrients and micronutrients: progress in science-based understanding
• Abstracts: Setting up a foot clinic. Leg ulceration in rheumatoid arthritis. Diabetic foot ulceration
• Abstracts: Limitations in characterization of heterosexual HIV transmission risk: commentary on the models of Downs and De Vincenzi
• Abstracts: Changing focus on unexplained esophageal chest pain. Diagnostic value of esophageal studies in patients with angina-like chest pain and normal coronary angiograms

This website is not affiliated with document authors or copyright owners. This page is provided for informational purposes only. Unintentional errors are possible.