In clinical trials or other comparative studies, randomization refers to the technique of assigning subjects (usually patients) to an experimental intervention (often a new treatment) group and a control groups based only on chance distribution. The purpose of randomization is to reduce the opportunity for selection bias when assigning patients to one group or the other. Proper randomization of patients is an indifferent yet objective technique that tends to neutralize the impact that any risk factors or other prognostic factors, known or unknown, may have on outcomes by spreading them evenly among the experimental and control groups. That is, randomization reduces the chances for any prognostic factor to be allocated unevenly between the experimental and control groups, which could thereby confound discerning the causal relationship between the experimental intervention and the outcomes of interest. For randomization to be successful, the number of patients (or other subjects) to be assigned must be large enough to achieve a high probability of evenly distributing any prognostic factors.
Given the need to minimize the influence of known as well as unknown sources of bias in comparative studies involving telemedicine, it is desirable to use random assignment whenever possible. Depending upon the investigation, it may be one or more of patients, physicians, or delivery sites that are randomized. For example, the randomization of patients to telemedicine intervention or standard of care minimizes the chances for differences in such potential prognostic factors as age, disease severity, or socioeconomic status to confound results.
For telemedicine, randomization often is not a straightforward matter. As noted above, telemedicine interventions are not always discrete or self-contained technologies, and thereby present challenges to randomization. In the instances of a trial of a new drug, for example, the molecular entity itself is reasonably assumed to be the same from pill to pill. As long as physicians or patients follow the trial protocol for dispensing or taking the pills, there is presumed to be no interactive effect between the delivery of the medication and the molecular entity itself. However, in a trial of teleconsultations, the causal effects of the teleconsultation may be confounded by differences among the participating physicians, among the participating institutions (for a multicenter trial), or among other factors. For example, physicians include stronger and lesser proponents of telemedicine, have varying levels of confidence or satisfaction in teleconsultations, and have varying levels of skill in conducting them. Many telemedicine evaluations to date have involved physicians that have been self-selected as proponents, rather than also including physicians who may have tried telemedicine but rejected it, physicians that have been reluctant to participate in it, or physicians that have been interested but without access to telemedicine facilities. Clinics or other delivery settings vary in many ways that may affect the provision of teleconsultations. It may be difficult to standardize or control for these potentially confounding factors.
In principle, then, in addition to randomizing patients to either teleconsultation or standard care, it may be desirable to randomize participating referring physicians to teleconsultations or standard care, and to randomize participating clinics to providing teleconsultations or standard care. In practice, however, the numbers of participating physicians and delivery sites may be too small for randomization to distribute prognostic factors to teleconsultations or standard care evenly enough to neutralize their effects on the outcomes of interest. Although it is more preferable to randomize than not to randomize, even where small numbers prevail, the practical constraints may outweigh the benefits of doing so. Therefore, investigators need to make explicit assumptions about the similarities among physicians and among clinics, i.e., that they are not sufficiently different to have independent effects on the outcomes of interest. Clearly, these are important design considerations that can affect the validity of any findings of comparative studies of telemedicine evaluations.
Several of our expert interviewees called for large, multicenter RCTs of teleconsultations in which patients would be randomly assigned to teleconsultations or standard care, and health outcomes would be followed over time. To the extent that the centers involved and the nature of the teleconsultations can be assumed to be comparable, such RCTs could provide convincing evidence about the value of teleconsultations.