A distal radius fracture displacement of >2 mm is considered significant [8]. Screw sizes vary in steps of 1 or 2 mm. We thus determined the non-inferiority margin (δ), which was 1. The standard deviation of the anterior-to-posterior (AP) size was the same as in a previous cadaver study [9]. Our hypotheses were as follows:

H0 size at fracture site (3D simulation) minus size at fracture site (real surgery) ≤−|δ|

H1 size at fracture site (3D simulation) minus size at fracture site (real surgery) >−|δ|.

Assuming that the groups were matched and using the sample size estimation method of the Centre for Clinical Research and Biostatics [10], the sample required a power of 80% and a significance level of 0.05 was seven. G*Power (version 3.1.9.7) also calculated a sample size of seven [11].

Patients were consecutively enrolled beginning in October 2021. We invited adult patients with distal radius fractures admitted for surgery to participate. Fractures with severe comminutions (AO/OTA classes 23-C2 and C3) were excluded [12]. Seven patients gave written informed consent by February 2022.

An IQon Spectral CT scanner (Philips, Amsterdam, Netherlands) was used. Preoperative wrist CT yields images in DICOM format that can be converted into STL images using 3D Slicer (ver. 4.11; Surgical Planning Laboratory, Harvard Medical School, Boston, MA, USA) [13]. Segmentation converts stacked two-dimensional images (in DICOM format) into 3D surface models (STL format) [14]. The process is as follows: the DICOM files to be converted are selected. Usually, the thinnest axial images are ideal. We obtained 299 images per patient, each with a thickness of 0.8 mm. Then, a segmentation editor is used to select an appropriate threshold for the region of interest; only the osseous portion should undergo volumetric reconstruction. After clicking on the 3D icon, the model appears. If only the distal radius is of interest, the “scissors” and “island” functions can be used to crop away other carpal bones and/or the ulna. Finally, the final 3D model is exported as an STL file (Fig. 1). The file must then be converted into geometric (G)-code to command the printer. We used Cura (ver. 4.11.0; Ultimaker, Utrecht, the Netherlands) to this end. This open-source slicing application converts STL to G-code that instructs the 3D printer. The default settings should be used, i.e. 0.2-mm thickness and 10% occupied volume. Clicking the slice icon yields G-code that is then sent to an Ender-5-Plus 3D printer (Shenzhen Creality 3D Technology, Shenzhen, China) using polylactic acid filament as ink. The product is a printed 3D model of the fractured distal radius. Fig. 2 summarizes these steps.

The AP and radial-to-ulnar (RU) widths (in mm) of the fracture sites were measured on CT images, printed models, and real bones (during surgery) (Fig. 3).

(1) To obtain measurements on CT images, the “ruler” of the picture archiving and communication system was used. The fracture site was selected in an axial cut by browsing the coronal and sagittal sequences. At the appropriate axial image, AP and RU widths between the farthest cortices were measured (Fig. 3A).

(2) After 3D models were produced, the models were separated at the fracture sites. We used a digital caliper to measure the AP and RU widths at the distal end of proximal fragments (fracture site) in the printed models (Fig. 3B).

(3) We used the same caliper (now sterilized) to measure real bones during surgery. Similarly, these measurements were also made by holding the distal end of proximal fragments with the caliper before the reduction (Fig. 3C). A gentle periosteal dissection at the fracture site facilitated the measurement using a caliper, which in turn made the reduction also simple.

We recorded the times required for modeling, printing, and simulation. Modeling refers to the conversion of the DICOM file to G-code. Reduction and fixation of printed distal radii (using plates and screws) were simulated by the junior orthopedic surgeon (author 1). During simulation and surgery, narrow or standard 2.4-mm variable angle locking compression plates and volar distal radius plates (DePuy Synthes, Warsaw, IN, USA) with two shaft holes were used. Each plate was secured with 2.4-mm variable angle locking screws. Plate locations were recorded by measuring three distances (location, sLOC [and rLOC] 1, 2, and 3) (Table 1, Fig. 4A) from the radial margins of the fracture site to the wire holes. If all three distances were similar during simulation and real surgery, we assumed that the plate locations would be similar. The screw lengths (ssc [and rsc] 1 to ssc [and rsc] 8) (Table 2, Fig. 4B) were recorded. The senior orthopedic surgeon (author 2), who was blinded to the simulation data, performed all surgeries. The corresponding plate profiles and screw sizes during the surgery were recorded.

Measurements were compared using the signed-rank test. When no significant difference was apparent, measurement consistency was further evaluated by calculating interclass correlation coefficients (ICCs). All statistical tests were two-tailed and p-values of <0.05 were considered significant.

The CT, 3D model, and real bone AP and RU widths did not differ significantly (Table 3). The ICC revealed excellent consistency among the measurements.

For cases 1 to 5, narrow plates were used for both simulation and real surgery; while standard plates were employed for cases 6 and 7. Distance 2 was significantly shorter during real surgery than simulation (mean sLOC2, 16.04 mm vs. mean rLOC2, 14.69 mm; p=0.043) (Table 1). In terms of screw sizes, the most distal and radial screw, i.e. screw 1, was significantly longer in the simulation compared with real surgery (mean ssc1, 17.42 mm vs. mean rsc1, 14.29 mm; p=0.016) (Table 2). The sizes of other screws did not differ significantly, but only the two screws for the shaft holes had ICCs indicating excellent consistency (0.927 and 0.842, respectively).