A systematic review of nonspecific low back pain trials published between 1980 and 2012.Objective.
To explore what proportion of trials have been powered to detect different bands of effect size; whether there is evidence that sample size in low back pain trials has been increasing; what proportion of trial reports include a sample size calculation; and whether likelihood of reporting sample size calculations has increased.Summary of Background Data.
Clinical trials should have a sample size sufficient to detect a minimally important difference for a given power and type I error rate. An underpowered trial is one within which probability of type II error is too high. Meta-analyses do not mitigate underpowered trials.Methods.
Reviewers independently abstracted data on sample size at point of analysis, whether a sample size calculation was reported, and year of publication. Descriptive analyses were used to explore ability to detect effect sizes, and regression analyses to explore the relationship between sample size, or reporting sample size calculations, and time.Results.
We included 383 trials. One-third were powered to detect a standardized mean difference of less than 0.5, and 5% were powered to detect less than 0.3. The average sample size was 153 people, which increased only slightly (∼4 people/yr) from 1980 to 2000, and declined slightly (∼4.5 people/yr) from 2005 to 2011 (P < 0.00005). Sample size calculations were reported in 41% of trials. The odds of reporting a sample size calculation (compared to not reporting one) increased until 2005 and then declinedResults.
Sample sizes in back pain trials and the reporting of sample size calculations may need to be increased. It may be justifiable to power a trial to detect only large effects in the case of novel interventions.Conclusion.
Level of Evidence: 3