The article by Dr. Memtsoudis et al.  describing perioperative outcomes after unilateral knee arthroplasty (UKA) and bilateral knee arthroplasty (BKA) uses the technique of using a large database, the Nationwide Inpatient Sample (NIS), to undertake as described in the accompanying editorial a form of “comparative effective research.”1,2In their introduction, the authors state that to date, one of the major limitations of studies comparing UKA with BKA has been their small sample sizes. To avoid this problem, the authors use a large database to compare the outcomes of UKA versus  BKA. However, in designing a study using a large database, it is important to define the population groups under study accurately.

In the total knee population, the authors defined a population of 670,305 admissions between 1998 and 2006 (9-yr period) requiring knee arthroplasty. In the “Materials and Methods,” the authors state that the NIS database contains information from approximately 8 million admissions per year. By using this number, we calculated that knee arthroplasty represented 0.93% of all admissions (670,305/72 million). Can the authors define the actual percentage that knee arthroplasty represents of all NIS database hospital admissions and more importantly of all NIS database surgical admissions for the period 1998–2006? Do the authors believe that this incidence is representative of the general population, and if not, how would this alter their results?

In the authors' article, the only significant benefit to the BKA group in this NIS population was the incidence of device-related complication, which was 0.52% in the BKA group compared with 0.86% in the UKA group. How do the authors account for this difference?

Of the 670,305 patients, 626,601 (93.75%) underwent UKA as defined by the code 81.55. This code does not differentiate between a first UKA and a second opposite-side UKA. In the BKA group, both knees require arthroplasty. Thus, should the authors include only patients who have undergone two UKA knee arthroplasties (opposite side) in the control group? If the control group consists of patients who have undergone two UKAs, should the morbidity of the two UKAs be combined? If the patient had the first UKA and then decided, for whatever the reason, not to have the second opposite-side UKA, should this be classified as a complication of the first UKA?

In the article, the authors refer to four different groups in the BKA population. The first is the total BKA population of 43,703 (6.52% of 670,305). The second is a subtotal BKA population of 34,015 patients (the total BKA group minus a group of 9,688 [22.3%] that could not be defined). The third and fourth are subgroups and consist of simultaneous BKAs 25,443 (74.8% of 34,015) and staged BKAs of 8571 (25.2% of 34,015). Except for the length of stay, the authors refer to percentages and not actual “n” values in most of the results, and without stating which of the BKA groups, it makes it difficult to discern the actual values. Can the author supply the reader with actual value for mortality in the simultaneous BKA group and how this compares directly with the UKA group?

Based on the NIS database and the definitions used by the authors, the number of patients required to undergo BKA compared with UKA to cause one additional mortality is 625, that is, 625 people need to undergo BKA to cause one additional death that would not have occurred if they had only received a UKA. The evidence as demonstrated by the authors may be compelling; however, their results are based on how the authors defined the BKA and UKA groups and the NIS database studied.

*Mayo Clinic, Jacksonville, Florida.

Memtsoudis SG, Ma Y, Gonzalez Dellas Valle A, Mazumdar M, Gaber-Baylis LK, MacKenzie CR, Sculco TP: Perioperative outcomes after unilateral and bilateral total knee arthroplasty. Anesthesiology 2009; 111:1206–16
Kheterpal S: Perioperative comparative effectiveness research. Anesthesiology 2009; 111:1180–2