3.2
Experimental Units and Treatments
- Definitions:
- Individuals on which experiments are conducted:
- Inanimate objects: Experimental units
- Animals: Subjects
- Humans: Participants
- A specific experimental condition applied to the units is referred to as a treatment.
Example of a Clinical Trial
- Case Study:
- A randomized trial evaluating chemoradiotherapy and chemotherapy following the surgical resection of pancreatic cancer, conducted by Neoptolemos et al., published in the New England Journal of Medicine in 2004.
- Objective: To assess the benefits of radiation therapy and chemotherapy.
Patient Sample and Treatment Groups
- Study Population:
- Total of 289 Patients involved in the trial.
- Treatment Groups:
- 69 patients received Chemoradiotherapy
- 73 patients received Chemotherapy
- 75 patients received both Chemoradiotherapy & Chemotherapy
- 72 patients were placed in the Observation (Control) group.
Five-Year Survival Rates
- Survival Rate Statistics for Different Treatments:
- Observation only: 11%
- Chemoradiotherapy: 7%
- Chemotherapy only: 21%
- Chemoradiotherapy & Chemotherapy: 13%
- Significance of Randomized Controlled Experiments:
- Considered the gold standard in health science due to their ability to provide robust evidence for causation.
Nature of the Clinical Trial
- The described clinical trial exemplifies a comparative experiment.
Control Group
- Definition:
- The control group refers to the cohort of patients who did not receive any treatment post-surgery.
Placebo Effect
- Consideration in Clinical Trials:
- Acknowledgment that patients might show improvement due to the placebo effect irrespective of the treatment administered.
- Note: In this specific context, it is deemed unethical to subject cancer patients to a sham treatment placebo.
Experiment Design Elements
- An experiment is designed based on several key components:
- Response Variable: The outcome measured in the experiment.
- Factors: The different treatment conditions being tested.
- Layout of Treatments: The organization of these factors in the experimental structure.
- For the clinical trial example:
- Response Variable: 5-year survival
- Factors:
- Radiation therapy
- Chemotherapy
- Layout of Treatments: A 2 by 2 factorial experiment is utilized.
Sources of Bias and Randomization
- Bias in Treatment Assignment:
- The allocation of experimental units to treatments can introduce bias.
- Mitigating Bias:
- Can attempt to match patient characteristics (age, gender, physical condition, etc.), although this is often inadequate.
- The preferable method is to randomly assign patients to treatment groups.
Randomization
- Definition:
- The use of chance to allocate experimental units to treatment groups is termed randomization.
- Pioneering of Randomization:
- Introduced by R. A. Fisher in agricultural field trials to enhance crop yields.
Principles of Experimental Design
- Key Principles:
- Control the influence of lurking variables on the response by comparing two or more treatments.
- Randomize the assignment of experimental units to treatment groups.
- Repeat treatments across numerous units to minimize variation due to chance.
Statistical Significance
- Definition:
- An observed effect large enough that it is unlikely to occur by random chance is defined as statistically significant.
Random Number Generation
- Methods of Randomization:
- Use a random number generator available on calculators or statistical software.
- Alternatively, employ a traditional table of random digits.
Types of Experimental Design
- Completely Randomized Design (CRD):
- All experimental units are randomly distributed among all treatments.
- Double-Blind Trials:
- In such trials, neither the participants nor the medical staff are aware of which treatments have been assigned.
- Triple-Blind Trials:
- In addition to the double-blind aspects, the statistician analyzing the data is also unaware of treatment allocation.
- Requires an additional statistician and at least one medical staff member to track safety, who will know the treatment assignments.
Matched Pairs and Block Designs
- Matched Pairs Design:
- Simplest method of utilizing matching, typically involving just two treatments.
- Block Design:
- A block is defined as a collection of experimental units known to share common characteristics.
- In a block design, random assignment of units to treatments is performed separately within each block.