On 30 October 1948, the British Medical Journal published a Medical Research Council report on streptomycin for pulmonary tuberculosis, and the study enrolled 107 young patients and compared streptomycin plus bed rest with bed rest alone. Treatment assignment came from concealed schedules based on random sampling numbers. Chest radiographs were read without knowledge of treatment or sequence.
The report is widely described as the first modern randomized controlled clinical trial. That description needs two qualifications. It was not the first controlled comparison in medicine, and it was not the first time a random device had assigned patients; its landmark status comes from how clearly several protections worked as one design.
The history matters because “randomized trial” is more than a label. Random sequence generation, allocation concealment, a concurrent comparator, consistent outcome measurement, and analysis of the assigned groups solve different problems. The streptomycin trial made that architecture visible.
Before randomization, medicine still made comparisons#
Clinical experiments long predate modern statistics. Which study counts as the “first trial” depends on which feature you insist on: a prospective comparison, a concurrent control, alternate allocation, a random mechanism, concealment, blinding, or a published protocol. Change the requirement and you change the answer, which is why several studies belong in this history without any one of them making the others irrelevant.
James Lind's scurvy comparison#
On 20 May 1747, Royal Navy surgeon James Lind selected 12 sailors with scurvy aboard HMS Salisbury; he placed them in six pairs, kept their basic conditions and diet similar, and gave each pair a different proposed remedy. The pair receiving oranges and lemons improved rapidly; one returned to duty after six days, when the fruit supply ended.
Lind's study was a powerful concurrent comparison. It reduced some background differences and tested competing treatments at the same time. His 1753 account did not describe a random assignment process, however, and with two people per group, it also could not use randomization to create stable balance across many prognostic factors. Calling Lind's work a controlled trial is accurate. Calling it the first randomized trial blurs a later methodological distinction.
Alternation and rotation#
Researchers later assigned patients by alternation, admission date, or rotation. These methods can produce numerically balanced groups, but the next assignment is predictable: if a clinician knows that the next patient will receive the new treatment, conscious or unconscious enrollment choices can alter who enters at that moment. Predictability is the weakness: a sequence can look perfectly orderly while selection bias walks in ahead of assignment.
Random devices before 1948#
In a 1931 US trial of sanocrysin for pulmonary tuberculosis, investigators used a coin toss to divide matched groups. Other experiments used random numbers or comparable devices. These studies show that random allocation did not suddenly appear in 1948. The historical question is therefore not “Who first thought of chance?” It is “When did a rigorously described system make chance assignment credible, concealed, clinically workable, and interpretable?”
The 1944 patulin trial was an important bridge#
During the Second World War, early reports suggested that patulin might treat the common cold. The MRC organized a large multicenter, placebo-controlled study reported in 1944.
Patients received coded solutions in a strict four-group rotation: two patulin groups and two placebo groups. The extra coding was intended to make the pattern harder for staff to predict. The trial found no convincing benefit.
The patulin study was methodologically sophisticated. It used concurrent placebo controls, masking, standardized procedures, and multiple centers. Yet rotation remained deterministic. Someone who discovered the sequence could foresee a future assignment.
The trial's negative result may help explain why it became less famous. Methodological history often remembers successful treatments more readily than rigorous null findings. Its design directly informed the team that later planned the streptomycin study.
Why streptomycin created an urgent test#
Tuberculosis was a major cause of death, and effective treatment options were scarce. Streptomycin, discovered in the 1940s, showed activity against Mycobacterium tuberculosis. Britain had only a limited supply.
The MRC faced a scientific and allocation problem. Pulmonary tuberculosis had a variable course. A series of treated patients could improve because of case selection, supportive care, or the disease's natural variation, and at the same time, there was not enough drug for every eligible patient.
A concurrent randomized comparison could estimate benefit while distributing a scarce, unproven treatment by a rule that recruiters could not manipulate. Scarcity did not itself make the study ethical; uncertainty about benefit, a meaningful control condition, and fair allocation were central.
What the 1948 trial actually did#
The study focused on patients aged 15 to 25 with acute, progressive, bilateral pulmonary tuberculosis that was bacteriologically proved and unsuitable for collapse therapy, and the restrictions created a comparatively defined population at substantial risk.
Fifty-five patients were assigned to streptomycin plus bed rest, and 52 to bed rest alone. Streptomycin was given for four months. Both groups otherwise received the prevailing hospital management.
A random sequence generated away from recruitment#
Austin Bradford Hill prepared allocation schedules from random sampling numbers, separately by sex and center; this structure aimed to preserve balance within the small center and sex groupings while retaining unpredictability.
Concealment until a patient was accepted#
Investigators and the central coordinator did not know the sequence. Each assignment was kept in a numbered, hospital-specific sealed envelope. A patient had to be judged eligible and accepted into the trial before the envelope identified the group.
The order is what does the work. If the assignment is revealed before enrollment is irrevocable, a recruiter can delay, exclude, or redirect a patient. Random numbers cannot repair that selection.
A concurrent control group#
The control patients received bed rest, then standard management. Because both groups were recruited in the same period and centers, secular changes in diagnosis, nursing, nutrition, and background risk were less likely to create the entire difference. Historical controls would have been far less credible, because tuberculosis prognosis varied and supportive care was changing underneath the comparison.
Outcome assessment without treatment knowledge#
The study could not mask patients or ward clinicians easily: streptomycin injections and toxicity were apparent. It did protect a central radiographic outcome. Serial chest films were copied, stripped of dates and identifiers, and assessed in a concealed order by readers who did not know treatment group. Keep that distinction: an open-label trial can still blind its outcome adjudicators, especially for images or records that can be reviewed centrally.
Prespecified clinical and bacteriological follow-up#
The report described mortality, general condition, temperature, weight, sedimentation rate, sputum results, radiographic changes, and drug toxicity. Follow-up at six months captured the early contrast between groups and emerging resistance.
What the trial found#
During the first six months, four of 55 streptomycin patients died, compared with 15 of 52 control patients. Radiographic and clinical improvement also favored streptomycin.
The benefit was dramatic but incomplete. Some patients deteriorated after an initial response, and streptomycin-resistant organisms emerged. The study therefore delivered two lessons at once: the drug had real activity, and monotherapy could select resistance.
Later combination regimens transformed tuberculosis care. The first trial did not answer the final treatment question. It established a causal effect in its population and exposed the next problem to solve.
Randomization and concealment are not synonyms#
These two safeguards are often merged in casual summaries.
Random sequence generation addresses how assignments are ordered, and a valid random process gives each eligible participant a known chance of each treatment and balances measured and unmeasured prognostic factors on average.
Allocation concealment addresses who can know the next assignment before enrollment. It prevents selection into groups based on prognosis.
A trial can have a genuinely random computer sequence and still be biased if an open list sits on the recruiter's desk; it can also have excellent sealed envelopes and a predictable alternating sequence inside them. Look for both, or you have neither. Blinding begins after allocation and addresses a third family of problems: differences in co-interventions, participant behavior, outcome assessment, or analysis caused by knowing the assigned treatment.
What randomization does not guarantee#
Randomization does not automatically produce identical groups, especially in a small trial. Chance imbalances remain possible, but its value is that any baseline difference arises from a known chance mechanism rather than a systematic enrollment rule, allowing valid probability-based inference when the design and analysis align.
It also does not protect against:
- loss to follow-up that differs by group;
- selective outcome reporting;
- treatment crossover or poor adherence;
- measurement error;
- a biased or clinically weak endpoint;
- a sample that does not represent the target population;
- too few events to estimate effects precisely.
The MRC report is instructive because it paired randomization with explicit eligibility, concealment, concurrent follow-up, protected assessment, and transparent reporting.
Why the phrase “gold standard” needs care#
Randomized trials are especially strong for estimating causal effects of interventions because assignment breaks the usual link between prognosis and treatment choice. They are not best for every question.
Rare or delayed harms may require very large observational data sets. Prognosis is usually studied in cohorts. Diagnostic accuracy needs representative patients and an appropriate reference standard. Mechanistic experiments explain how a treatment might work. Qualitative studies reveal experiences that a numerical endpoint can miss.
Even for treatment benefit, a poorly designed randomized trial can be less trustworthy than a strong alternative design, and the correct lesson from 1948 is not that one label outranks appraisal. It is that deliberate protections make comparisons fairer.
How the 1948 architecture appears in a modern trial report#
Reading a modern trial report, you can trace:
- who generated the sequence and by what method;
- whether blocking or stratification was used;
- how assignment was concealed from recruiters;
- when eligibility became final relative to disclosure;
- who was blinded after assignment;
- whether all randomized participants appear in the analysis;
- whether the registered outcomes match the reported outcomes.
Central web-based randomization now often replaces envelopes. The technology changed, but the logic did not. The person deciding whether a patient enters should not be able to forecast the group.
Sealed envelopes can still work if they are opaque, sequentially numbered, tamper-evident, and opened only after enrollment. Poorly prepared envelopes can be held to light, opened early, or substituted. A method's name is less important than its implementation.
The broader legacy#
The MRC trial showed that a multicenter clinical question could be coordinated with shared eligibility rules, centralized allocation, standardized records, and protected outcome assessment, and its report explained enough of the process for readers to judge why the groups should be comparable.
It also illustrated a physician-scientist habit: uncertainty should be designed around, not hidden. The investigators did not treat an exciting antibiotic or a biologically persuasive theory as proof. They built a comparison capable of being wrong.
The finding led to use of streptomycin, but resistance redirected the field toward multidrug therapy. Strong experiments do not end inquiry. They sharpen the next question.
The fair-comparison conclusion#
There is no single birthday for the clinical trial. Lind demonstrated the force of concurrent comparison. Later investigators used alternation, masking, placebo controls, and random devices. The MRC patulin study assembled many of those pieces.
The 1948 streptomycin trial became the landmark because it made a full chain of protection explicit. Eligibility came before concealed assignment. Random numbers replaced a predictable sequence. A concurrent control represented what would otherwise happen. Protected readers judged radiographs. Follow-up documented both benefit and resistance.
Its lasting contribution is a design principle: a persuasive medical comparison depends on preventing knowledge of the answer from changing who enters, what they receive, how outcomes are judged, and which results are shown.
References#
- Medical Research Council. Streptomycin treatment of pulmonary tuberculosis. BMJ. 1948;2:769-782.
- Chalmers I. Why the 1948 MRC trial used treatment allocation based on random numbers. James Lind Library.
- Yoshioka A. Use of randomisation in the MRC streptomycin trial. BMJ. 1998.
- Chalmers I, Clarke M. The 1944 patulin trial of the British Medical Research Council. International Journal of Epidemiology. 2004.
- Bartholomew M. Lind and scurvy: 1747 to 1795. Journal of the Royal Society of Medicine. 2002.
- Crofton J. The MRC randomized trial of streptomycin and its legacy. BMJ. 2006.
Questions and answers
Was James Lind's scurvy study the first randomized controlled trial?
It was an early and famous controlled clinical comparison, but Lind's report did not describe random assignment; the citrus result was striking, yet the design does not meet the modern definition of a randomized trial.
Why is the 1948 MRC trial called the first modern randomized trial?
It clearly documented random-number allocation, concealment before enrollment, concurrent controls, defined eligibility, standardized follow-up, and blinded radiograph assessment as one coherent system.
Did any medical study randomize patients before 1948?
Yes. Earlier studies used coin tosses or other random devices. The 1948 study's special status reflects its methodological clarity and influence, not an absolute claim that chance had never been used.
What is the difference between randomization and allocation concealment?
Randomization creates an unpredictable assignment sequence. Concealment keeps that sequence hidden from those enrolling participants until entry is final. Each prevents a different source of bias.
Was the streptomycin trial blinded?
Patients and treating clinicians knew whether injections were given, but radiographic assessment was protected by removing identifying and chronological information, showing that outcome-assessor blinding can remain possible in an open-label trial.