Minimal Pain Decrease Between 2 and 4 Weeks After Nonoperative Management of a Displaced Midshaft Clavicle Fracture Is Associated with a High Risk of Symptomatic Nonunion

Abstract

Background

The main long-term benefit of operative treatment of displaced midshaft clavicular fractures is the reduction in nonunion risk, and as this risk is generally low, the ideal approach would be to operate only patients at high risk of nonunion. However, most current surgical decision models use baseline variables to estimate the nonunion risk, and the value of these models remains unclear. Pain in the early weeks after fracture could be potentially be an indirect measurement of fracture healing, and so it is a potential proxy variable that could lead to simpler prediction models.

Questions/purposes

(1) Is pain a possible proxy variable for the development of symptomatic nonunion after nonoperative treatment of midshaft clavicular fractures? (2) How reliable is the model we created that uses pain as a proxy variable for symptomatic nonunion of nonoperatively treated clavicle fractures?

Methods

In this secondary retrospective analysis of an earlier randomized trial, we studied prospectively collected data from 64 nonoperatively treated patients aged 18 years to 60 years. In the original randomized trial, we compared operative and nonoperative treatment of displaced midshaft clavicular fractures. In all, 150 patients were included in the study, of whom 71 received nonoperative treatment. Patients were predominantly males (75%, 48 of 64) with a mean age of 38 ± SD 12 years; most fractures were comminuted and shortened more than 1 cm. All 71 patients who were nonoperatively treated were potentially eligible for this secondary analysis; of those, 11% (8 of 71) were lost to follow-up, leaving 63 patients from the nonoperative treatment arm and one patient from the operative treatment arm (who declined surgical treatment after randomization but was followed in this group according to the intention-to-treat principle) for analysis here. Nonunion was defined as lack of callus formation, persistent fracture lines and/or sclerotic edges of the bones at the fracture site on plain radiographs at 6 months follow-up. Nonunions were regarded as symptomatic if pain, tenderness, and local crepitation were present at the fracture site. Seventeen percent (11 of 64) of patients had symptomatic nonunions. After investigating differences in early pain scores between the union and nonunion groups, we defined the VAS_ratio as the VAS pain score at 4 weeks divided by the VAS pain score at 2 weeks. Week 2 VAS pain score was chosen as baseline after visual inspection of a linear mixed model that showed increased divergence in pain scores between union and nonunion group at 2 weeks after fracture. Week 4 was chosen as the cutoff because we wanted a reasonable time frame for the detection of pain reduction and did not want to delay surgical treatment more than necessary. Odds ratios for various risk factors were calculated using logistic regression analyses. We used a receiver operating characteristic curve analysis to identify cutoff values for the VAS_ratio.

Results

An increase in absolute pain score at 4 weeks after fracture (odds ratio 1.8 per 1 point increase [95% confidence interval 1.1 to 3.4]) was associated with an increased risk of nonunion 6 months after fracture. Likewise, we found that an increasing VAS_ratio (OR 1.02 per 0.01 point increase [95% CI 1.002 to 1.06]) was also associated with nonunion. Receiver operating curve analysis found that the best cutoff value of VAS_ratio was about 0.6. Patients with a VAS_ratio above 0.6 had a relative risk of developing nonunion of 18 (95% CI 2 to 130) compared with patients with a VAS_ratio below 0.6. Sparse-data bias could be present, as is evident from this wide confidence interval, though even at the low end of the confidence interval, the relative risk was 2, which may still improve surgical decision-making.

Conclusion

A pain score that exhibits no or minimal change from 2 to 4 weeks after nonoperative treatment of a displaced midshaft fracture of the clavicle is associated with a high risk that symptomatic nonunion will develop. Patients with no or minimal change in pain in the early weeks may be candidates for surgery to reduce the risk of symptomatic nonunion. As this was a retrospective study, with a risk of sparse-data bias, the predictive value of the VAS_ratio needs to be further investigated in large prospective studies before clinical use.

Level of Evidence

Level III, diagnostic study.

Introduction

Clavicular fractures account for up to 4% of all fractures, and midshaft fractures are the most common type [15]. Operative treatment of displaced midshaft clavicular fractures reduces the nonunion risk compared with nonoperative treatment, but it offers no clinically relevant long-term improvement in functional outcome scores [2]. The nonunion risk after nonoperative treatment was reported to be between 9.5% and 24% [1, 3, 20, 38, 39, 41, 42], and based on the union rates reported in a recent network meta-analysis [2], the number needed to treat (NNT) to avoid one nonunion is 10. The main long-term benefit of operative treatment of displaced midshaft clavicular fractures is the reduction in nonunion risk, and because this risk is generally low, the ideal approach would be to only operate patients at high risk of nonunion. This approach, however, would depend on the surgeon’s ability to identify patients who are at a high risk of developing a nonunion.

In a systematic review, increasing age, female sex, and tobacco smoking along with fracture comminution and shortening were described as unlikely to be risk factors for the development of nonunion, and only the presence of displacement was thought to be a risk factor [14]. Two statistical models estimated the risk of nonunion at 24 weeks based on fracture morphology and patient demographics [22, 37]. A small study based on retrospective data [5] showed that smoking tobacco and a DASH score of more than 35 points at 6 weeks after fracture increased nonunion risk. In a larger cohort, the same study group showed that at 6 weeks, a QuickDASH score of at least 40 points, no callus seen on radiographs, and fracture movement on examination were associated with nonunion [24]. The lack of a bridging callus on a sonographic evaluation at 6 weeks after fracture has also been proposed to predict nonunion [26]. A study of 222 patients with all types of clavicular fractures showed that pain at rest and pain during activity in the first 24 weeks after fracture was more common in patients with nonunion and long-term sequelae than in patients with no long-term sequelae [27]. None of the abovementioned prediction models have been externally validated. The predictive value of known baseline variables such as sex, age, tobacco smoking, fracture displacement, and fracture comminution remain controversial [14]. Hypothetically, several possible baseline predictors such as the degree of interposed tissue or genetic predisposition for slow healing may exist, perhaps along with other undiscovered variables. Instead of using baseline variables to predict nonunion risk, it might be possible to use proxy variables to estimate the likelihood of fracture healing. These possible proxy variables could be pain, DASH score at 6 weeks, or early signs of callus formation, all of which have been associated with nonunion [5, 24, 26, 27].

In general, existing prediction models were not developed using modern principles [32], and the clinical utility of these models remains unproven. As an indirect measurement of fracture healing, a proxy variable contains the effects of all known and unknown baseline variables, which could lead to simpler prediction models.

We therefore performed a secondary (retrospective) analysis of an earlier randomized trial to answer the following main questions: (1) Is pain a possible proxy variable for the development of symptomatic nonunion after nonoperative treatment of midshaft clavicular fractures? (2) How reliable is the model we created that uses pain as a proxy variable for symptomatic nonunion of nonoperatively treated clavicle fractures?

Patients and Methods

This is a secondary (retrospective) analysis of a previous multicenter randomized trial that compared operative and nonoperative treatment of displaced midshaft clavicular fractures in adults (ClinicalTrials.gov Identifier: NCT01078480) [33].

Displacement was defined as no cortical contact between the two main fragments on at least one of two radiographs. Exclusion criteria were bilateral fracture, impending skin perforation, open fracture, neurovascular injury, other fracture in the same extremity, pathologic fracture, history of shoulder disorder or previous clavicular fracture, fracture older than 14 days, cognitive impairment, inability to follow treatment regimens, or contraindications to general anesthesia or surgery.

Base Cohort

Between February 2010 and May 2014 150 patients were enrolled and randomized to either operative or nonoperative treatment in equal size groups [33]. Four patients withdrew consent after randomization. Patients were predominantly males with a mean age of about 40 years and most fractures comminuted and shortened more than 1 cm. No comorbidity index was used, but patients were generally well matched in the randomized trial.

The nonoperative care consisted of immobilization with a sling (Collar’n’Cuff, Mölnlycke Health Care, Göteborg, Sweden) for as long as the patient felt it necessary but for a maximum of 3 weeks. The patient was encouraged to use the shoulder as tolerated, and no physiotherapy was instituted.

Patients reported current pain at the fracture site on a VAS once per week in a paper journal the first 6 weeks after fracture. VAS score was marked on a 10 cm scale with no digits. The scale was marked in one end as: “No pain at all” and in the other end as: “Worst imaginable pain”. A study nurse converted the VAS score to a numerical value from 0 to 10 (0 cm to 10 cm). Patients were examined clinically and radiographically after 6 weeks and 3, 6, and 12 months. The DASH score was obtained at 6 weeks and after 3, 6, and 12 months [13]. The DASH ranges from 0 to 100 points, with a score of 0 representing normal function and increasing score representing increasing dysfunction. Nonunion was defined as lack of callus formation, persistent fracture lines and/or sclerotic edges of the bones at the fracture site on two AP radiographs with a 15° to 30° angle between them in the sagittal plane at 6 months’ follow-up. Three reviewers (SLJ, CMJ, MTV) analyzed radiographs. The nonunion was regarded as symptomatic if any pain, tenderness, and local crepitation were present at the fracture site.

Study Cohort

Seventy-five patients were randomized to nonoperative treatment in the original study [33]. For this study cohort, the exclusion criteria were similar to that of the clinical trial. Four patients withdrew consent, and at 6 months, eight patients were lost to follow-up. The study cohort consisted of the remaining 63 patients and one patient, who opted for nonoperative treatment after initial randomization to surgery (Fig. 1). Patient demographics were comparable to the base cohort (Table 1).

Fig. 1.

The STROBE flowchart demonstrates the selection of the cohort for this study.

Table 1.

Characteristics of patients in the union and nonunion groups

Parameter	Union (n = 53)	Nonunion (n = 11)	p value
Mean age, years (SD)	37 ± 13	43 ± 11	0.2
% male (n)	77 (41)	64 (7)	0.5
% who smoked tobaccoa (n)	17 (9)	45 (5)	0.05
% with fracture comminution (n)	68 (36)	54 (6)	0.5
% with fracture shortening more than 2 cm (n)	25 (13)	18 (2)	1.00
Mean VAS ratio (95% CI)	0.5 (0.3 to 0.6)	0.9 (0.7 to 1.1)	< 0.001
Mean pain score at 4 weeks, points (95% CI)	1.9 (1.5 to 2.6)	4.2 (3.3 to 5.1)	< 0.001
Mean DASH score at 6 weeks, points (95% CI)	22 (20 to 29)	37 (31 to 44)	< 0.001

^aDefined as daily smoking of any tobacco products.

At 6 months, nonunion had developed in 19% (12 of 64) of patients in the study cohort (95% confidence interval 10% to 31%), and 11 of 12 patients were symptomatic. Ten patients with symptomatic nonunion underwent surgery with open reduction and plate fixation. All surgically treated patients with nonunion had healed with no complications at 12 months after the initial fracture. One patient with symptomatic nonunion did not undergo surgical treatment because of a high risk of anesthetic complications after a newly discovered aortic aneurysm.

Statistical Analysis

All statistical analyses were conducted using R (Foundation for Statistical Computing, Vienna, Austria) [34] and the nlme [28] and pROC [36] packages. One investigator (AHQ) completed all primary analyses. The linear mixed effects model was developed with the help of the local department of mathematical sciences. Continuous variables were analyzed using a t-test for normally distributed data. Categorical variables were tested using a chi-square test or Fisher’s exact test if one value was 5 or less. We performed a linear mixed-effects analysis of the relationship between pain score and nonunion. Parameters were estimated via the restricted maximum likelihood method. As a fixed effect, we entered nonunion into the model. Participant identification was included as a random factor and development over time was included as a random slope. Model diagnostics were validated by a visual inspection of residual plots and fitted data on an individual participant scale. We were only interested in the analysis of possible proxy variables, so we made no adjustment for possible confounding baseline variables.

After visually inspecting the linear mixed-effects analysis plot, we defined the VAS_ratio as the pain score at 4 weeks divided by the pain score at 2 weeks. We chose week 2 VAS pain score as the baseline after visual inspection of the linear mixed model, which showed increased divergence in pain scores between union and nonunion groups at 2 weeks after fracture. We chose week 4 as the cutoff because we wanted a reasonable time frame for the detection of pain reduction and did not want to delay surgical treatment more than necessary. A VAS_ratio of 0 represents a pain-free patient at 4 weeks, and a VAS_ratio of 1 represents a patient with an unchanged pain score from 2 weeks to 4 weeks. No patients reported no pain at 2 weeks, and no data were missing at 2 and 4 weeks. A logistic regression analysis was used to identify and quantify the effects of different possible proxy variables that may reflect an increase in the nonunion risk. We used different VAS_ratio cutoff points to divide the cohort into two groups, and we calculated absolute and relative nonunion risk in the groups. We used a receiver operating characteristic curve to analyze the discriminatory value of the VAS_ratio. Youden’s index [43] was calculated for each possible VAS_ratio cutoff point and was used as a criterion for selecting the optimum cutoff point at a VAS_ratio value where the sensitivity and specificity of the test was maximal. The level of significance was set at p < 0.05.

Results

Pain as a Proxy Variable for Nonunion

We found an association between pain in the early weeks after fracture and the development of nonunion. An increase in absolute pain score at 4 weeks after fracture (odds ratio 1.8 per 1 point increase [95% CI 1.1 to 3.4]) was associated with an increased nonunion risk 6 months after fracture. Likewise, we found that an increasing VAS_ratio (OR 1.02 per 0.01 point increase; 95% CI 1.00 to 1.06) was also associated with nonunion (Table 2). In the linear mixed-effects analysis, we saw a deviation in the VAS score over time for the two groups (Fig. 2). The difference in slope was = 0.4 points per week (p = 0.003). We found that VAS_ratio, pain score at 4 weeks and DASH score at 6 weeks were associated with the development of nonunion (Table 1). Fractures that united had a lower mean VAS_ratio of 0.5 (95% CI 0.3 to 0.6) compared with fractures that did not unite (0.9 (95% CI 0.7 to 1.1]; p < 0.001). At 4 weeks, the mean pain score for fractures that united was 1.9 (95% CI 1.5 to 2.6) and 4.2 (95% CI 3.3 to 5.1; p < 0.001) for fractures that did not. The DASH score at 6 weeks (37 [95% CI 31 to 44]) was worse in the nonunion group than in the union group (22 [95% CI 20 to 29]; p < 0.001).

Table 2.

Logistic regression analysis of possible proxy variables

Variable	Odds ratio	95% confidence interval	p value
Pain at 4 weeks (per 1 point increase)	1.84	1.1 to 3.4	0.03
VAS ratio (per 0.01 point increase)	1.02	1.00 to 1.06	0.04
DASH at 6 weeks (per 1 point increase)	1.35	0.96 to 1.09	0.49

Fig. 2.

This graph shows the results of the linear mixed model. The bold lines indicate prediction intervals on the population level with 95% confidence interval. Individual patient data are plotted as single lines. The red lines indicate the nonunion group and the blue lines indicate the union group. A color image accompanies the online version of this article.

Reliability of Pain as a Proxy Variable for Nonunion

We saw an increase in the relative risk of developing nonunion with a decrease in the VAS_ratio and an increase in the absolute risk with an increasing VAS_ratio (Table 3). Receiver operating curve analysis found that the best cutoff value of VAS_ratio was about 0.6; at a VAS_ratio of 0.6, the relative risk of developing nonunion in the exposed group compared with that in the control group was 18 (95% CI 2 to 130) (Table 3). For the VAS_ratio the area under the receiver operating characteristic curve was 0.84 (95% CI 0.72 to 0.96), indicating a diagnostic test with acceptable discrimination [12] (Fig. 3). At a VAS_ratio of 0.6 the absolute risk for developing nonunion was 44 % (95% CI 23 to 64). This means that patients who did not experience a reduction in pain by 40% or more from 2 to 4 weeks had an absolute increase of 41% (95% CI 20 to 62) in the risk of developing nonunion compared with patients with a larger pain reduction. At a higher VAS_ratio cutoff, the absolute risk of nonunion increases but using a higher VAS_ratio cut-off would also result in a higher number of false negatives (Table 3).

Table 3.

Analysis of the cohort at various VAS_ratio cutoff points

VASratio	Union (n)	Nonunion (n)	Absolute riska	Relative riska	p value
Cutoff 0.20
Above cutoff	41	11	21% (95% CI 11 to 35)	∞b	0.11
Below cutoff	12	0
Cutoff 0.60
Above cutoff	13	10	44% (95% CI 23 to 64)	18 (95% CI 2 to 130)	< 0.001
Below cutoff	40	1
Cutoff 1.00
Above cutoff	2	3	60% (95% CI 15 to 95)	4 (95% CI 2 to 12)	0.03
Below cutoff	51	8

^aAbsolute and relative risk of symptomatic nonunion in the “Above cutoff” group.

^bNo nonunions in “Below cutoff” group.

Different VAS_ratio cutoff points were used to divide the cohort into groups “Above cutoff” and “Below cutoff”.

Fig. 3.

This graph shows the receiver operating characteristic curve, using the VAS_ratio as a predictor of symptomatic nonunion. The area under the curve is equal to 0.84 (95% CI 0.72 to 0.96).

Discussion

The main long-term benefit of operative treatment of displaced midshaft clavicular fractures is the reduction in nonunion risk [2]. The nonunion risk is generally low [1, 3, 20, 38, 39, 41, 42], and the ideal management of midshaft clavicular fractures would be to only operate on patients at high risk of nonunion. This approach, however, depends on the surgeon’s ability to identify patients at a high risk of having nonunion. Several models for the prediction of nonunion risk have been proposed [22, 24, 37], but these models are often complex and based on baseline variables of questionable predictive value [14], and none of these models have been externally validated. We proposed that a measurement of pain in the early weeks after fracture could be used as an indirect measurement of fracture healing. This proxy variable would contain the effects of all known and unknown baseline variables, which could lead to simpler prediction models. In our secondary analysis of a randomized trial [33] of patients with nonoperatively treated, displaced midshaft clavicular fractures, we found that no or minimal changes in pain scores from 2 to 4 weeks after fracture is associated with a high risk of symptomatic nonunion by 6 months. A high VAS_ratio was independently associated with an increased risk of nonunion, and that risk increased with an increasing VAS_ratio.

Limitations

This study had several limitations. In our original study, five patients were lost to follow-up at 6 weeks (early dropouts), and a total of eight patients were lost to follow-up at 6 months (late dropouts). We do not know why these patients dropped out of the study. The risk estimates and predictive value of the VAS_ratio might be affected if patients drop out early because of high levels of pain or late because of the symptoms of possible nonunion. This possible worst-case scenario further illustrates the point that the predictive value of the VAS_ratio needs to be further investigated before clinical use. We found that the risk of developing a nonunion is 18 times higher (95% CI 2 to 130) for patients with a VAS_ratio above 0.60 compared with patients with a lower ratio (Table 2). This wide CI is most likely a function of small sample size along with a low number of events; sparse-data bias could be present [17]. However, at even the lowest estimate of relative risk, the VAS_ratio still has the potential to improve surgical decision-making, as a doubling of the nonunion risk would mean a nonunion risk of about 25% for patients with a VAS_ratio above 0.60. Surgical treatment in this extreme case could still reduce the NNT to 4 compared with the existing NNT of 10 as was proposed recently in a network meta-analysis [2].

Also, we defined the VAS_ratio as changes in pain from 2 to 4 weeks. This definition was based on an inspection of the linear mixed model, where pain scores started to deviate at 2 weeks after fracture (Fig. 2). We did not investigate other possible time period combinations. The association with nonunion could become stronger if the definition of the VAS_ratio included a longer period. The analysis of different time ratios is beyond the scope of this study, in which we simply illustrated an association between no or minimal improvement in early pain scores and the risk of nonunion, and different time periods should be investigated in future studies. Likewise, absolute pain at 4 weeks was not further investigated. We believe that the use of a single absolute pain score could be unreliable in a clinical setting because pain perception is highly individual and influenced by age, race, gender, and socioeconomic status [9, 11, 16, 21].

The clinical reliability of the VAS_ratio depends on the existence of ratio properties for reporting pain scores on a VAS. For the VAS_ratio model to work, a patient with a pain decrease from 8 points to 4 points must have the same pain experience as a patient with a decrease from 4 points to 2 points. Supporting our model, other studies have shown these linear and ratio properties [8, 23, 30, 31]; furthermore, the percentage of pain reduction based on raw pain scores has a good correlation with the patient-reported percentage of reduction in pain [4]. In a study of patients with chronic pain, the authors showed that subjective perception of changes in pain and the measurable percentage of change in pain scores was consistent, regardless of baseline pain scores [7]. It may be possible to use the patient-reported percentage of pain reduction in future studies to determine the predictive value of the VAS_ratio.

In general, a dataset consisting of a minimum of 100 events and 100 nonevents is recommended for external validation studies of predictive logistic regression models [40]. For clavicular nonunions, this would require a dataset substantially larger than ours, and we were therefore not able to test our data against other models [22, 37]. In our logistic regression analysis, we did not find DASH to be an independent variable, as did a retrospective analysis of another randomized study [5]. No description of the subdomain scores of DASH was done in that study, and the role of the pain subdomains of the DASH score at 6 weeks remains unclear.

The proportion of nonunion in the present cohort was 19% (12 of 63 patients), with 11 of 12 patients being symptomatic. Only one recent randomized trial clearly defined symptomatic nonunion [33]. Other randomized trials described a subpopulation of patients with nonunions who “require” surgery, with reported proportions ranging from 29% to 80% [1, 38, 39, 42]. This seemingly high proportion of symptomatic nonunions in our study is incomparable with the reported proportions because there is not a clear definition of symptomatic union, and the definition of nonunion itself differs in terms of time period and radiographic features.

Anticipating Nonunion After Clavicle Fracture

We found that a VAS_ratio of 0.6 or greater by 4 weeks after clavicle fracture was independently associated with a steep increase in the risk that the patient would develop a symptomatic nonunion by 6 months after injury. Several prior attempts to estimate the nonunion risk after clavicular fractures have been made. Some studies base the risk estimate on baseline variables such as age, sex, degree of comminution and tobacco smoking [22, 37]. Other studies were more focused on the association between nonunion and poor DASH [5] or QuickDASH [24] scores at 6 weeks after fracture. The predictive value of most baselines variables remains controversial [14], and we believe that poor DASH and QuickDASH scores at 6 weeks after fracture could be associated with pain. If associated with pain, the use of an absolute DASH or QuickDASH score would have the same clinical unreliability as the use of a single absolute pain score [9, 11, 16, 21]. None of the proposed models have been externally validated, and even the most modern model [24] was developed without sufficient internal and external validation [32]. In this study we showed that pain could be a proxy variable for fracture healing. As an indirect measurement of fracture healing, the VAS_ratio would contain all effects of all possible baseline variables and has the potential to be used as a predictor of nonunion risk without the need for baseline variables. Unfortunately, we did not have a sufficient sample size to properly develop a predictor model. According to sample size guidelines [35], the proportion of nonunions is a limiting factor because a sample size of 219 patients would be needed to develop a predictor model in a dataset with an event rate of 17%. External validation requiring a minimum of 100 events [40] is more challenging. The original data were collected at three centers during a 4-year period. At the same event rate, it would take approximately 18 years to reach 100 events at the same three centers. This calls for a national effort to collect a sufficient sample size for external validation. Given the sample size challenges, a more pragmatic approach in which surgery is offered to patients with a VAS_ratio above 0.6 may be a viable solution, but this model would need to be investigated further before widespread clinical use.

Using the VAS_ratio in Practice

Two common approaches to treating patients with clavicle fractures could be summarized as the “treat-all” approach and the “treat-none” approach. Hypothetical surgical treatment of all patients with a VAS_ratio of 0.6 or greater at 4 weeks would expose 20% of all patients in our cohort to unnecessary surgery because these patients later had union, despite having a high VAS_ratio. This is, however, a better scenario than a treat-all regimen, in which 83% of the cohort would be exposed to unnecessary surgery, or a treat-none regimen, in which 17% will have prolonged pain because of the development of nonunion over a period of many months. Since nonunion surgery has been shown to have less predictable functional results, a higher complication rate, and prolonged time to complete return to work compared with acute operative treatment [18, 19], we believe that exposing 20% of all patients to unnecessary acute surgery is a better option than allowing 17% of patients to develop a nonunion before intervention. A simple measurement of pain scores at 2 and 4 weeks along with the subsequent use of the VAS_ratio could become a valuable surgical decision tool in the management of displaced midshaft clavicular fractures. Delaying surgery to 4 weeks after fracture would have no negative impact on long-term functional outcome or union rates [6, 10, 25, 29] and would allow the surgeon to choose the right patients for surgery.

Conclusion

A pain score that exhibits no or minimal change from 2 to 4 weeks after nonoperative treatment of a displaced midshaft fracture of the clavicle is associated with a high risk that symptomatic nonunion will develop. Patients with no or minimal change in pain in the early weeks may be candidates for surgery to reduce the risk of symptomatic nonunion. As this was a retrospective study, with a risk of sparse-data bias, the predictive value of the VAS_ratio needs to be further investigated in large prospective studies before clinical use.