The authors evaluated the quality of clinical trials published in four anesthesia journals during the 20-yr period from 1981-2000.
Trials published in four major anesthesia journals during the periods 1981-1985, 1991-1995, and the first 6 months of 2000 were grouped according to journal and year. Using random number tables, four trials were selected from all of the eligible clinical trials in each journal in each year for the periods 1981-1985 and 1991-1995, and five trials were selected from all of the trials in each journal in the first 6 months of 2000. Methods and results sections from the 160 trials from 1981-1985 and 1991-1995 were randomly ordered and distributed to three of the authors for blinded review of the quality of the study design according to 10 predetermined criteria (weighted equally, maximum score of 10): informed consent and ethics approval, eligibility criteria, sample size calculation, random allocation, method of randomization, blind assessment of outcome, adverse outcomes, statistical analysis, type I error, and type II error. After these trials were evaluated, 20 trials from the first 6 months of 2000 were randomly ordered, distributed, and evaluated as described.
The mean (+/- SD) analysis scores pooled for the four journals increased from 5.5 +/- 1.4 in 1981-1985 to 7.0 +/- 1.1 in 1991-1995 (P < 0.00001) and to 7.8 +/- 1.5 in 2000. For 7 of the 10 criteria, the percentage of trials from the four journals that fulfilled the criteria increased significantly between 1981-1985 and 1991-1995. During the 20-yr period, the reporting of sample size calculation and method of randomization increased threefold to fourfold, whereas the frequency of type I statistical errors remained unchanged.
Although the quality of clinical trials in four major anesthesia journals has increased steadily during the past two decades, specific areas of trial methodology require further attention.
THE randomized clinical trial is the highest level of evidence available for evaluating new therapies. 1However, serious deficiencies in study design and data analysis have been reported in reviews of clinical trials in medical and surgical journals. DerSimonian et al. 2noted that approximately half of the criteria deemed to be essential qualities of good study design were either ambiguously reported or not reported at all in their survey of 67 clinical trials from four medical journals published between July 1979 and June 1980. Emerson et al. 3reported similar results in a survey of clinical trials published in general surgical journals between 1981 and 1982. These deficiencies in study design introduce an element of bias into the trial results, which may lead to exaggerated treatment effects. 4,5To address these deficiencies in study design and preclude misleading results, criteria for excellence in randomized clinical trials have been published. 6–9Despite these publications, the quality of clinical trials in virtually all contexts that have been assessed has not improved substantively. 10–15For example, Emerson et al. 11observed that the quality of trials in drug development improved only 9% in each of the past three decades. With the ever-increasing dependency of new interventions in anesthesia on evidenced-based outcome trials, we hypothesized that the quality of trials in the anesthesia literature has improved in the past 20 yr. Accordingly, we evaluated the quality of the study design in prospective clinical trials that were published in four leading anesthesia journals during the 20-yr period from 1981–2000.
Each issue of four anesthesia journals, Anesthesia and Analgesia , Anesthesiology, the British Journal of Anaesthesia , and the Canadian Journal of Anaesthesia , was manually searched for all prospective comparative clinical trials published between 1981 and 1985, between 1991 and 1995, and during the 6-month period from January to June 2000. Trials that involved animals and human volunteers, evaluation of equipment, comparative trials using case, cohort, or historical controls, and retrospective and observational studies were excluded. Of the trials from the periods 1981–1985 and 1991–1995 that satisfied the eligibility criteria, four were randomly selected from each journal for each of the 10 yr being considered using random number tables. A total of 160 clinical trials were selected for review by one of the authors who did not participate in the review process. The methods and results sections from each trial were selectively copied, blinded to the journal and year of publication, and then randomly ordered using random number tables. The randomization codes were concealed until the review process was completed. Each trial was examined for 10 characteristics of quality related to the study design by three of the authors (J. L., M. W. C., and J. G. W.) independently. 2The rating system modified from that of DerSimonian et al. 2was used because it was most appropriate for the type of clinical trials conducted in anesthesia:
1. informed consent and ethics approval: statements reflecting institutional ethics committee approval and informed consent from the patient or guardian
2. eligibility criteria: inclusion and exclusion criteria were provided for patients in the trial
3. sample size calculation: a sample size calculation or specification for the size of detectable differences before the trial commenced was provided
4. random allocation: a statement or phrase that indicated that the patients were assigned to their treatments in a randomized manner
5. method of randomization: details about the mechanism used to generate random assignment; randomization by birth date, hospital registration numbers, and so forth were considered unacceptable
6. blind assessment of outcome: a statement that the observer who assessed the outcome was unaware of the treatment assignment
7. adverse outcomes: details of the presence or absence of side effects or complications, not related to the primary outcome variable
8. statistical analysis: appropriate statistical tests were listed
9. type I error: defined as more than 10 comparisons among the treatment groups without correction or consideration of multiple tests
10. type II error: for trials with no statistically significant differences between groups in which neither a sample size estimation nor a power analysis was performed
When the reviews for the first two periods were completed, five prospective comparative clinical trials were randomly selected from those published in the same four journals between January and June 2000 using random number tables. The methods and results sections of these trials were prepared as described and analyzed by the same three authors using the same scoring system.
For each trial, the 10 criteria were evaluated and reported as present, absent, or not applicable. Inadequate or ambiguous information was reported as absent. The 10 criteria were weighted equally. The first eight items were scored 1 if present and 0 if absent, whereas the last two items (type I and II errors) were scored 1 if absent and 0 if present. The scores for the 10 characteristics were summed to give a maximum score of 10 for each trial. Scoring discrepancies among the reviewers were resolved by discussion and a consensus decision before the randomization code was revealed.
For the trials published in 1981–1985 and 1991–1995, the frequency of reporting each characteristic (nominal data) was compared among journals and between years using the Fisher exact test. The total trial scores were compared among the journals and between the two review periods using two-factor analysis of variance with the Newman–Keuls test for multiple comparisons. Mean analysis scores for the four journals in the 1981–1985 and 1991–1995 periods were tested for normality using the Kolmogorov–Smirnov test. To account for the number of between-group comparisons (15 in total), the threshold for statistical significance was designated to be P < 0.01. The trials published in the first 6 months of 2000 were included for descriptive statistics; however, the numbers were too small to include them in the statistical analysis.
All 180 trials were evaluated by the same three reviewers, and a consensus was reached on each criterion reviewed. For each of two periods, 1981–1985 and 1991–1995, twenty trials were reviewed from each of the four journals, giving a total of 160 trials. The frequencies of responses for each of the 10 criteria are summarized in table 1. Of the 10 criteria that were evaluated in each of the 160 trials (for a total of 1,600 evaluations), less than 1% were nonapplicable. For these few evaluations, the trial was omitted from the analysis of that criterion, and the number of journals in that period decreased accordingly. For the first 6 months of 2000, 20 trials, representing a mean sampling rate of 13% (range, 6–18%) of the published trials, were reviewed from the four journals. None of the criteria was deemed not applicable.
The mean analysis scores of the four journals in both the 1981–1985 and 1991–1995 periods were normally distributed. Within each period, the analysis scores for the four journals were similar and were therefore pooled (fig. 1). The pooled mean analysis scores increased significantly from 5.5 ± 1.4 in 1981–1985 to 7.0 ± 1.1 in 1991–1995 (P < 0.00001 compared with 1981–1985) and then to 7.8 ± 1.5 in 2000. The percentage of trials with scores of 9 or more out of 10 increased from 0% in 1981–1985 to 10% (3.5–16.5%, 95% confidence interval) in 1991–1995 (P < 0.014 compared with 1981–1985) and to 30% (11–49%, 95% confidence interval) in 2000.
Within the periods 1981–1985 and 1991–1995, the percentage of trials that fulfilled each criterion was similar among the four journals. Accordingly, we pooled their percentages into one value for the respective period. For 7 of the 10 criteria, consent, eligibility, randomization, methods of randomization, blinded assessor, and adverse events, the pooled values increased significantly between 1981–1985 and 1991–1995 (table 1). The reporting of sample size calculations was poor in the periods 1981–1985 and 1991–1995, appearing in fewer than 20% of trials in 1991–1995, but the reporting increased to 55% by 2000 (fig. 2and table 1). Although the reporting of the method of randomization increased between 1981–1985 and 1991–1995 (fig. 3), the difference reached statistical significance for only one journal (P < 0.0083). By 2000, 75% of the trials in three of the four journals reported the method of randomization. In the fourth journal, none of the trials reported this criterion. Statistical analysis was reported in most trials and did not change significantly between 1981–1985 and 1991–1995. The risk of a type I statistical error occurred in approximately 50% of the trials during all three periods (fig. 4). The risk of a type II statistical error was present in approximately 10% of trials (table 1) and did not change significantly between 1981–1985 and 2000.
We evaluated the quality of study design in clinical trials published in four leading anesthesia journals between 1981 and 2000. The overall quality of the trials was strikingly similar among the four journals within each period, a finding that is consistent with preliminary data previously reported (Alex Mathieu, M.D., Department of Clinical Epidemiology and Biostatistics, University of Cincinnati, Cincinnati, OH, written communication). The mean analysis scores for the four journals increased approximately 20% between successive periods, an increase that exceeded that reported in other medical subspecialties. 10–14The mean analysis score of clinical trials from the four anesthesia journals in 2000, 7.8 out of a maximum value of 10, is encouraging, but with only 30% of the trials scoring 9 or greater of a maximum score of 10, deficiencies remain in the majority of published clinical trials. Although the percentage of trials scoring 9 or greater out of 10 in 2000 is threefold greater than the 10% reported in 1991–1995, the overall quality seems lacking. Greater emphasis on clinical study design is required if we are to remain a competitive medical research subspecialty.
Sample size calculation is frequently omitted from clinical trials. 14,16–19Reviews of trials published in journals from medicine, surgery, and family practice have reported low rates, between 5% and 52%, for sample size calculation 3,10–14,17–21that are consistent with the 12% reporting rate for sample size calculation during the 1991–1995 period in this study. Although other specialties have not reported comparable data for sample size calculations in the past 5 yr, our finding of a fourfold increase in the reporting of this criterion suggests that sample size calculation is more widely included in the study design of clinical trials in the anesthesia literature than it has been in the past two decades. Nonetheless, that 45% of trials in 2000 did not report a sample size calculation reflects poorly on the scientific, ethical, and fiscal aspects of study design in clinical anesthesia. Scientifically, a sample size calculation is essential because it provides a reasonable estimate of the number of patients needed to reject the null hypothesis if it were truly false, thereby minimizing the risk of type II statistical error. 16Ethical concerns mandate that a trial should enroll the minimum number of patients required to reject the null hypothesis. The basis for the ethical concern is that a minimum number of patients should be exposed to the potential harms associated with randomizing their care. To address this concern, most ethics committees now require investigators to justify their sample size before ethics committee approval is granted. Finally, the sample size calculation dictates fiscal implications, including the number of patients that should be enrolled, the duration of the study, and the overall study cost. Failure to include a sample size calculation may result in excessive fiscal costs due to overenrollment of patients.
Randomization, a strategy that is used to minimize bias in clinical trials, 6,17,22,23consists of two distinct operations: random allocation and an appropriate method of randomization. Random allocation refers to the assignment of patients to treatments such that the chance of receiving any one treatment is the same for all comparative treatments. Randomization sequences that provide an appropriate method of randomization are based on random number tables, computer programs, or any other technique for which the chance that any single treatment is assigned to a patient is the same for all comparable treatments. Arbitrarily assigning patients to treatments, termed nonrandomized allocation, may introduce bias by interfering with or manipulating the treatment assignment. 22–24Flipping a coin or use of birth dates or hospital record numbers are acceptable methods of randomization; however, they are unacceptable methods for allocating patients to treatments because the investigator could be unblinded to the treatment assignment. 6,17,22–24These techniques are termed nonconcealed randomization. Trials in which the allocation assignment is not concealed are more likely to lead to exaggerated treatment effects, resulting in more trials with positive outcomes than trials in which the treatment is concealed. 4–6,20,21,25Some authors have suggested that concealed allocation, an issue that was not evaluated in the current study, may be more important in minimizing bias than the actual randomization. 21The results of this study indicate that almost all of the trials reviewed reported random allocation of the treatments.
Reporting rates for the method of randomization of 7.5% in 1981–1985 and 26% in 1991–1995 in this study and the results of another study 26suggest that the anesthetic literature has not kept pace with the 16–50% incidence reported in medical and surgical specialties during the same intervals. 3,7,10,17,18,21,23In 2000, the method of randomization increased to 50% of the clinical trials reviewed, with marked variability among the journals. Comparable data for this criterion from other medical specialties in 2000 are not available. Although omission of the method of randomization does not necessarily indicate that an inappropriate method of randomization was used, it does raise concern regarding possible bias in the data. This criterion must be addressed in future trials.
Blind assessment of outcome is another strategy that minimizes bias in clinical trials. In this review, we scored a trial as blinded if the methodology was described as double-blinded. Implicit in this description is the notion that the patient as well as the observer were blinded to the treatment assignment. We did not require that the anesthesiologist be blinded to the treatment, provided that he or she did not determine the treatment, that he or she could not influence the outcome variables, and that he or she did not assess the outcome variables. Ideally, all participants who could potentially influence the results, including those who observe, record, or interpret any outcomes, should be blinded to the treatment assignment. In 2000, 70% of the trials indicated that the outcome variables were assessed by an individual who was blinded to the treatment assignment. Although this is almost double the rate in 1981–1985, it falls short of the 100% expectation for quality clinical trials. For circumstances in which blinding of the treatment is not feasible, at the very least, the investigators should ensure that the observers are blinded to the study hypotheses.
A key area of study design that showed little improvement and was addressed infrequently was the risk of a type I statistical error. Type I statistical errors occur when multiple comparisons are performed between treatments without statistical compensation. 27This may lead to exaggerated treatment effects (i.e. , falsely rejecting the null hypothesis). Using our threshold of 10 between-group comparisons to define the level beyond which the risk of a type I statistical error may become substantive, approximately 65% of the trials in 2000 were at risk for this error. Type I statistical errors are best controlled by prevention: identify the primary outcome variable, limit the number of variables to be evaluated, and minimize the between-group comparisons. If multiple between-group comparisons must be performed, then techniques such as multivariate analysis of variance or the Bonferroni t test must be used. 27
Our definition of a type I statistical error may be critiqued on two accounts. First, we arbitrarily defined the threshold for a type I statistical error as more than 10 between-group comparisons. No published consensus exists for the maximum number of between-group comparisons after which compensation for a type I error is appropriate. However, many would consider 10 between-group comparisons to be excessive and would recommend compensating for a type I error after even fewer comparisons. In this case, the risk of a type I error in our study would have increased. Second, we included all between-group comparisons, whether primary or secondary outcome variables or demographic variables. Some investigators consider only primary outcome variables and the comparisons that relate directly to that variable to contribute to type I errors. Had we accepted this latter threshold for type I errors, then the frequency of reporting this error would have been reduced by as much as 40%. Irrespective of the accepted threshold, to curb exaggerated treatment outcomes, we recommend that the editorial boards adopt criteria for diagnosing and implementing compensation for type I statistical errors in the outcome variable.
One additional possible limitation of this study was our modification of the criteria for evaluating the quality of trials by DerSimonian et al. 2We modified the criteria to improve the applicability and relevance of the criteria to clinical trials in anesthesiology. 2Specifically, criteria from DerSimonian et al. , 2such as “admission before allocation,”“patients’ blindness,” and “lost to follow-up,” have limited relevance to clinical trials in anesthesia. Furthermore, we modified the “statistical methods” by addressing type I and II statistical errors individually. Although the analysis scores based on the 10 criteria have not been validated, it is our contention that they provide a reasonable measure of the overall quality of clinical trials in anesthesia.
In summary, we found that trials published in four major anesthesia journals between 1981 and 2000 were of similar quality. The overall quality of the study design (mean analysis scores) increased approximately 25% between 1981–1985 and 1991–1995 but only 10% between 1991–1995 and the first 6 months of 2000. Nonetheless, serious deficiencies in study design remain, including sample size calculation, method of randomization, blinded assessors, and consideration of type I statistical errors. To remain a viable and competitive clinical–research subspecialty, it is incumbent on clinical investigators in our specialty as well as the editorial boards to acknowledge the deficiencies and support minimal standards for study design to ensure high-quality clinical trials in anesthesia.
The authors thank Ms. Nancy Sikich (Department of Anaesthesia, Hospital for Sick Children, Toronto, Ontario, Canada) for her assistance in completing this study.