Article Outline
To the Editor:
We read with interest with the well-written article entitled “Comparison of the Early Results of Minimally Invasive Vs Standard Approaches to Total Knee Arthroplasty: A Prospective, Randomized Study” by Karpman and Smith [1]. Their eligible patients were randomized into 3 groups: a quad sparing, mini midvastus, or standard approach to total knee arthroplasty. The authors used repeated t tests to compare continuous variables among 3 groups, and significance levels were not adjusted for multiple testing, which are not appropriate. The authors set the significance level at .05, the probability of a type I error. If the 3 comparisons can be assumed to be independent, then the chance of not committing a type I error in any one of them is (1 − .05)3 = .857. The chance of committing a type I error in at least one of the comparisons is 1 − .857 = .143, which is the overall type I error rate, although for each individual test the type I error rate is .05. We have to use analysis of variance (ANOVA) to avoid this problem [2]. When a significance has been found by ANOVA, it is necessary to do post hoc comparisons that maintains a low overall type I error, such as Bonferroni correction, Tukey test, Scheffé test, Duncan test, and many other procedures 2, 3.
After scrutinizing the data in the Table 2, we noticed the numbers did not add up between ambulation assistive (last day) and ambulatory status (last day), with different loss to follow-up among the 3 groups. We repeated 1-way ANOVA and Fisher exact test where appropriate 4, 5, and we found no significant difference among 3 groups concerning using ambulation assistive on the last day of rehabilitation (P = .119), ambulatory status on the last day of rehabilitation (P = .266), pain assessment score on the last day of rehabilitation (P = .073, assuming no loss to follow-up), and flexion degrees on the last day of rehabilitation (P = .071, assuming no loss to follow-up).
The authors omitted 3 patients from postoperative analysis because of death or infection. Both of these infections occurred in the quad sparing group, but we did not know the death occurred in which group, which should be clearly stated in the article. The authors analyzed their results by per-protocol analysis with the problems for patients lost to follow-up 6, 7, though in no instances was the surgical approach modified or abandoned. We performed sensitivity analysis [6] for quadriceps strength at week 2, and we noticed quad sparing group did not differ from midvastus mini-incision group statistically with Bonferroni correction.
We also concerned about expressions as P ≤ .05, and exact P values are preferable. Exact P values are especially desirable when no significant differences are demonstrated [3].
References
- . Comparison of the early results of minimally invasive vs standard approaches to total knee arthroplasty: a prospective, randomized study. J Arthroplasty. 2009;24:681
- . Statistics review 9: one-way analysis of variance. Crit Care. 2004;8:130
- . Statistics in the Journal of Bone and Joint Surgery: suggestions for authors. J Bone Joint Surg Am. 1992;74:319
- . Statistics in orthopaedic papers. J Bone Joint Surg Br. 2006;88:1121
- . Clinical epidemiology and biostatistics: a primer for orthopaedic surgeons. J Bone Joint Surg Am. 2004;86:607
- . The intention-to-treat principle: a primer for the orthopaedic surgeon. J Bone Joint Surg Am. 2006;88:2097
- . Intention-to-treat analysis and accounting for missing data in orthopaedic randomized clinical trials. J Bone Joint Surg Am. 2009;91:2137
PII: S0883-5403(10)00128-2
doi:10.1016/j.arth.2010.01.095
© 2010 Elsevier Inc. All rights reserved.
