Evidence Based Medicine

EVIDENCE BASED MEDICINE

Extensive scientific investigation of drugs in man and introduction of numerous new drugs over the past few decades is gradually transforming the practice of medicine from ‘experience based’ wherein clinical decisions are made based on the experience (or rather impression) of the physician to ‘evidence based’ wherein the same are guided by scientifically credible evidence from well designed clinical studies. Evidence based medicine is the process of systematically finding, evaluating and using contemporary research findings as the basis of clinical decisions. Results of well designed multicentric interventional trials are forming the basis of constantly evolving guidelines for disease management. Today’s physician has to be skilled in searching and evaluating the literature on efficacy, safety and appropriateness of a particular therapeutic measure (drug). Therapeutic evaluation of a drug includes:

§  Quantitation of benefit afforded by it.

§  The best way (dosage, duration, patient selection, etc.) to use it.

§  How it compares with other available drugs.

§  Surveillance of adverse effects produced by it.

Clinical studies are basically of the following three types:

a. Clinical trials

b. Cohort studies

c.  Case control studies

Clinical Trial

It is a prospective ethically designed investigation in human subjects to objectively discover/verify/ compare the results of two or more therapeutic measures (drugs). Depending on the objective of the study, clinical trial may be conducted in healthy volunteers or in volunteer patients. Healthy volunteers may be used to determine pharmacokinetic characteristics, tolerability, safety and for certain type of drugs (e.g. hypoglycaemic, hypnotic, diuretic) even efficacy. For majority of drugs (e.g. antiepileptic, antipsychotic, antiinflammatory, antitubercular, etc.) therapeutic efficacy can only be assessed in patients.

The inclusion of a proper comparator (control) group in clinical trials is crucial. The control group, which should be as similar to the test group as possible, receives either a placebo (if ethically permissible) or the existing standard treatment. Separate test and control groups may run simultaneously (parallel group design), or all the subjects may be treated by the two options one after the other (cross over design) so that the same subjects serve as their own controls. In the cross over design, some patients are treated first by drug ‘A’ followed by drug ‘B’, while in others the order is reversed. This nullifies the effect (if any) of order of treatment. This design is applicable only to certain chronic diseases which remain stable over long periods.

It is well known that both the participants and the investigators of the trial are susceptible to conscious as well as unconscious bias in favour of or against the test drug. The greatest challenge in the conduct of clinical trial is the elimination of bias. The credibility of the trial depends on the measures that are taken to minimize bias. The two basic strategies for minimizing bias are ‘randomization’ and concealment or ‘blinding’.

Randomization

The subjects are allocated to either group using a preselected random number table or computer programme so that any subject has equal chance of being assigned to the test or the control group. Discretion (and likely bias) of the investigator/subject in treatment allocation is thus avoided. If considered necessary, stratified randomization according to age/sex/disease severity/other patient variable may be adopted.

Blinding (masking)

This refers to concealment of the nature of treatment (test or control) from the subject (single blind) or both the subject as well as the investigator (double blind). For this purpose the two medications have to appear similar in looks, number, weight, taste, etc. and are to be supplied in unlabelled packets marked for each patient. In double blind, the key/code to treatment allocation is kept by a third ‘data management’ party who is not involved in treating or recording observations. The code is broken at the completion of the trial and the results are analysed according to prespecified statistical method. However, all clinical trials need not be blinded. Those in which the nature of treatment is not concealed are called ‘open’ trials.

Randomized controlled double blind trial is the most credible method of obtaining evidence of efficacy, safety or comparative value of treatments.

Inclusion/Exclusion Criteria

The characteristics of the subject/patient (age, sex, disease/symptom, severity and/or duration of illness, coexistant and past diseases, concurrent/preceeding drug therapy, etc.) who are to be recruited in the trial or excluded from it must be decided in advance. The trial results are applicable only to the population specified by these criteria.

End point

The primary and secondary (if any) end points (cure, degree of improvement, symptom relief, surrogate marker, avoidance of complication, curtailment of hospitalization, survival, quality of life, etc.) of the trial must be specified in advance. The results are analysed in relation to the specified end points.

Higher efficacy may not always be the aim of a clinical trial. A trial may be designed to prove ‘non inferiority’ (of the new drug) to the existing treatment, and possibly afford advantages in terms of tolerability, safety, convenience, cost or applicability to special patient subgroup(s).

Sample Size:

The number of subjects in the trial for obtaining a decisive conclusion (test better than control/control better than test/no difference between the two) must be calculated statistically beforehand. Because the trial is conducted on a sample of the whole patient population, there is always a chance that the sample was not representative of the population.

Two types of errors are possible:

Type I (α) error: a difference is found between the two groups while none exists. Its possibility is called ‘significance’ of the result, e.g. if test drug is found to be better than control at a significance level of 0.05, it means that there is 5% chance that this is not real.

Type II (β) error: no difference is found while it really exists. The probability of failing to detect an actual difference is expressed by the ‘power’ of the trial. A power of 0.9 means that there is 10% chance of missing a real difference.

The sample size of the trial depends on the desired level of significance and power. The other input needed for calculation of sample size is the magnitude of difference between the two groups that is expected or is considered clinically significant, e.g. a 10% reduction in pain intensity may not be considered clinically significant, while a 10% reduction in mortality may be worthwhile. Larger sample size is required to detect smaller difference. Also, higher the significance and power level desired, greater is the number of subjects.

Many large scale trials are subjected to interim analysis from time to time as the trial progresses by an independent committee which can order an early termination if a decisive result (positive or negative) is obtained; because it would be unethical to subject some of the remaining patients to a treatment (test or control) which has been found inferior.

Multicentric trial

Many large trials are conducted at more than one centre by as many teams of investigators, sometimes spread over several countries. The advantages are:

§  Larger number of patients can be recruited in a shorter period of time.

§  Results are applicable to a wider population base which may cover several countries/ ethnic groups.

§  Regulatory requirements of several countries may be satisfied.

§  Credibility of the trial is enhanced.

Sequential Trial

This design attempts to detect a significant result as soon as it is achieved, minimizing the number of subjects. The trial is conducted on matched pairs of subjects and is scored as ‘A’ treatment better than ‘B’ or ‘B’ better than ‘A’ or no difference. This is plotted continuously as the trial proceeds till the

boundries of predetermined level of significant superiority/inferiority/no difference are touched. The trial is then terminated. This design is applicable only to certain types of drugs and diseases for which clinical end points are achieved quickly and paired comparisons are possible. Moreover, it may not always be practicable to recruit matching pairs of trial subjects.

Meta-Analysis

This is an exercise in which data from several similarly conducted randomized controlled clinical trials with the same drug (or class of drugs) examining the same clinical end point(s) is pooled to bring out the overall balance of evidence by enlarging the number of test and control subjects and increasing the significance and power of the conclusions. Because individual trials are often conducted on relatively smaller number of patients, some may fail to detect a significant difference, while others may find it. Discordant results are published which confuse the medical practitioner. Though there are many criticisms of metaanalysis, such as:

§  bias in the selection of trials for analysis;

§  unintentional exclusion of negative results which are less likely to be published (publication bias);

§  nonuniformity of the trials in details of methodology and conduct;

it is a useful tool to arrive at conclusions that may influence medical practice. For example, metaanalysis of trials has strongly supported the use of βadrenergic blockers in heart failure and use of statins to reduce risk of coronary artery disease.

To be reliable, the metaanalysis should observe the following:

§  Comprehensive search of the literature to identify all eligible trials.

§  Use objective criteria for selecting the trials for inclusion.

§  Include only randomized trials of assured quality.

§  Employ proper statistical methods in pooling and treating the data from individual trials.

Metaanalysis are now frequently published on contemporary therapeutic issues.

Cohort Study

This is a type of observational study in which no intervention for the sake of the study is done. ‘Cohort’ is a group of individuals having some common feature. In the context of drug research, the common feature is that all study subjects have taken a particular drug. Occurrence of events (beneficial or adverse) in users and nonusers of the drug is compared. It can be a prospective or a retrospective study. In the prospective design, all patients who receive the study drug are followed up for therapeutic outcomes or adverse effects. A matching group of patients who have not received the drug is identified and followed up to serve as control. Cohort studies are primarily used to discover uncommon adverse effects that may be missed during formal therapeutic trials which involve fewer patients and often exclude certain type of patients who may be susceptible to that adverse effect. Its value for defining therapeutic outcomes is less credible. The limitations of cohort studies are that controls included may not be appropriate, and relatively long period of follow up is needed.

In the retrospective cohort study, health records of a population are scrutinized for exposure to the study drug and the subsequent beneficial/adverse events. Its value is questionable because many events may have been missed in the records and several unknown factors may have contributed to the findings. However, it may serve as pointer, or to arouse suspicion.

Case Control Study

This type of observational study is used mainly to reveal association of a suspected rare adverse event with the use of a particular drug. Cases of the suspected adverse event (e.g. agranulocytosis) are collected from hospital records or disease registries, etc. A matched control group similar in other respects but not having the adverse event is selected. Drug histories of both groups are traced backwards to compare exposure to the indicted drug (e.g. phenylbutazone) among patients with the adverse event to those without it. The suspicion is strengthened if high association is found. Though case control studies can be performed rather quickly because the number of patients analysed is small compared to the cohort design, they do not prove causality. Also, the causative drug and the adverse event have to be suspected first to plan the study, whereas cohort study can reveal unsuspected adverse events. Variable accuracy of retrospective records, non randomly selected control group, chances of bias and a variety of unknown factors make the case control study a weak instrument for affording convincing evidence.