For most RCTs, proving therapeutic efficacy necessitates certain constraints in patient selection. It is not uncommon that many patients in a physician’s practice would not have fulfilled the restrictive entrance criteria of most moderate-sized RCTs. For example, some RCTs have an age cut-off that actually excludes more than half of all patients with the disorder. This by no means implies that the reputed benefit of the test drug is not applicable (effective) to the patients excluded, but it does beg the question. Entry criteria alone should never be the sole basis for denying a patient the benefit of proven therapy. Interpatient variability is inevitable in all RCTs and contributes much to the “random errors” seen in small and moderatesized trials. However, the larger the trial, the smaller the random error, and the more likely that benefit can be reliably extrapolated to some patients who do not necessarily qualify for entry.^1,2 For example, one may observe that the benefits are consistent across different subgroups, suggesting that the results may be applicable beyond the boundaries of patient selection. On the other hand there may emerge reliable evidence for a lack of benefit in certain subgroups.

Evidence-based medicine that depends solely on external evidence is disease oriented rather than patient oriented. In other words, the verifiability of RCT data is often dependent on having a given diagnosis, as opposed to a clinical spectrum of risk associated with the diagnosis. This is the so- called “ labeling ” dilemma. For example, patients labeled as having ” acute coronary syndrome ” simply because they present with chest pain associated with non-ST segment elevation are often treated alike in an RCT, whereas the clinical expression of this entity may encompass a wide range from very low- to very high-risk patients. Translating external evidence based solely on a unified diagnosis into practice guidelines or clinical pathways has the unfortunate consequence of making management decisions dependent on a label rather than the presenting clinical circumstances and risk of the underlying disorder.

Another nagging problem with the “bottom line” of clinical trials is the emphasis on primary endpoints that are measurable. Statistical dependency on hard data such as mortality rates, prespecified clinical outcome events, rehospitalizations, etc., fails to acknowledge the significance of clinically relevant “soft” data, such as impact on symptoms, quality of life, psychosocial well-being, attitudes, economic realities and patient preferences.

Finally, clinical trials all have finite time limits and, not uncommonly, the duration may be inadequate to assess long-term benefits and risks, especially for any new drug. In such cases the information from RCTs may have to be supplemented by other sources of non- randomized evidence.