Improved healing response in delayed unions of the tibia with low-intensity pulsed ultrasound: results of a randomized sham-controlled trial

A multi-center, randomized, controlled trial was undertaken at six hospitals in Germany to determine the effectiveness of LIPUS in accelerating the healing process in delayed unions of the tibial shaft. All adult patients who had sustained a tibial shaft fracture that subsequently showed inadequate progress toward healing (i.e., delayed union) and provided informed consent were candidates for inclusion in the study. Delayed union was defined as lack of clinical and radiologic evidence of union, bony continuity or bone reaction at the fracture site for no less than 16 weeks from the index injury or the most recent intervention [6, 9, 13]. Standard anteroposterior (AP) and lateral radiographs were performed at baseline to establish lack of union. Patients were not admitted to the study if any of the following criteria were present: pregnancy, revision or reoperations at the fracture site within 16 weeks of enrollment, deep wound infection, excessive malalignment. The trial was approved by the institutional review boards at each clinical site and was conducted in adherence to the Declaration of Helsinki.

One hundred one subjects (age range: 14 to 70 years) with delayed unions of the tibia, enrolled between January 2002 and December 2005, were randomly allocated to treatment with either an active LIPUS device (n = 51) or an inactive sham device (n = 50).

Patients with established delayed unions of the tibia were randomized to receive either LIPUS or an identical nonoperative sham device. The study device was the Exogen 2000/2000+ (Smith & Nephew GmbH, Schenefeld, Germany). The sham device was inactivated in such a manner as to be indistinguishable from the active LIPUS device, except that it did not emit acoustic pressure waves.

All subjects were instructed to use the device for 20 minutes per day, for 16 weeks. Those subjects randomly assigned to active treatment had the ultrasound pressure wave signal set at the following parameters: 1.5 MHz frequency, 1 kHz repetition rate, 200 μs pulse duration, 30 mW/cm² spatial intensity. All devices recorded the amount of time per day of use as a means of assessing subject compliance with the treatment protocol.

Treatment was assigned randomly to each subject on a 1:1 basis in blocks of six and randomization was stratified within each clinical site. The randomization code was developed using a computer random number generator. The investigators, subjects and sponsor were blinded to the random allocation sequence prior to initiation of treatment and throughout the entire duration of this study. Once the study was complete and the last subject reached 16 weeks of follow-up, the randomization code was broken and treatment assignments revealed to the study statistician. Quantitative radiographic assessments of BMD and gap area also were undertaken without knowledge of treatment group assignment.

The primary outcomes of this study were BMD and gap area at the fracture site, assessed by computed tomography (CT) scanning. All radiographic imaging studies were evaluated at a central Radiology laboratory using a dedicated workstation for image processing and digitization. BMD was determined for three regions of interest: fracture site, 2-3 mm proximal and distal. BMD also was determined in a healthy reference area. For each of the three regions, three replicate observations were obtained. These three replicates were averaged to obtain a mean BMD for each region separately for pre-treatment and post-treatment evaluations. Region specific changes from baseline were then computed. Finally, the three region specific changes from baseline obtained for each subject were averaged to obtain the primary effectiveness endpoint defined as the mean change from baseline in BMD.

BMD was indirectly estimated using the mean CT attenuation coefficients, or Hounsfield units (HUs), for each region of interest. Calibration to varying concentrations of K₂HPO₄ to directly quantitate BMD in equivalent mineral density (mg/cm³) was not undertaken. Thus, BMD results are reported in HUs [14, 15]. Gap area in mm² was estimated directly from CT scans using digitized images.

The primary endpoint with respect to efficacy was change in BMD between pre-treatment and 16 weeks. Change in gap area at the fracture site was a secondary endpoint.

AP and lateral radiographs also were taken at 1, 2 and 3 month follow-up intervals. Blinded to treatment assignment, participating physicians were asked to judge the healing status (healed/not healed) of each study subject at 16 weeks.

All analyses employed Statistical Analysis Software (SAS). Subject baseline characteristics were summarized using frequency and percentage distributions or descriptive statistics, as appropriate. Proportions were compared using the Chi-square test with Yates' continuity correction or Fisher's exact test. Continuous variables were compared using the two sample t-test.

The primary analysis was intention-to-treat (ITT) and involved all subjects who received random treatment assignments and initiated device usage. Seventeen subjects had missing post-treatment outcomes, consequently 84 subjects were included in descriptive analyses of 'completers'. There was notable differential drop-out between groups with 24% (12 of 50) of sham-treated subjects and 9.8% (5 of 51) of active-treated subjects missing post-treatment BMD values. The ITT cohort was preserved by imputing missing clinical endpoints using a multiple imputation procedure that minimizes bias from differential drop-outs and properly accounts for uncertainty in imputed values when performing statistical inference. Multiple imputation uses multiple predictions of missing clinical endpoints based on patient characteristics and the baseline value of the clinical outcome variables [16]. For each of five stochastically completed data sets, analysis of covariance (ANCOVA) was used to estimate a treatment group contrast that controlled for the baseline value of the clinical endpoint as well as clinical site. This ANCOVA approach is consistent with ICH E9 guidelines for statistical analysis of a multi-site clinical trial [17]. The estimated group difference obtained for each of the five stochastically completed data sets were appropriately combined to obtain a single estimate of the intervention effect. Attention was paid to the validity of statistical model assumptions using graphical and other approaches. We found that BMD could be validly analyzed with or without log transformation, but that valid inference required log transformation for gap area. Analyses on log transformed data expresses group differences in terms of relative improvement which has a convenient clinical interpretation. Estimates of relative and absolute group differences were presented along with confidence intervals that accounted for the multiple imputation.

All primary effectiveness results using statistical modeling were based on a one-sided alpha = 0.05 with 90% confidence intervals (CIs). Preliminary descriptive analyses without imputations and descriptive assessments of effect size (ES) were two-sided with 95% CIs.

As specified a priori, study hypotheses were ordered such that the hypothesis concerning change in BMD was tested first. Only if the active device was shown to be superior to sham based on a one-sided test at α = 0.05 would the second test regarding changes in bone gap area be performed. This a priori ordering eliminated the need to adjust significance levels for multiple testing.

The success rate for physician judgment of fracture healing was compared between groups using Fisher's exact test. The statistical content of the manuscript was approved by an expert in medical statistics and epidemiology.