➢ Many risk factors for fracture nonunion are well supported in the orthopaedic literature, including location of the fracture site, surgical treatment, bone displacement, type of fixation, treatment delay, comminution, inadequate treatment, and wound infection. However, evidence from a systematic review suggests that patient-related or biological causes of nonunion may not be as well understood.
➢ Understanding the biological causes of nonunion is important for several reasons. Risk factors might identify patients prone to nonunion who could benefit from more aggressive intervention, and a clear idea of nonunion risk could be important when choosing between competing therapeutic options. Risk factors also can inform the design of clinical trials and clarify patient inclusion and exclusion criteria, so that small studies can yield more definitive answers. Finally, an understanding of patient risk profiles may enable clinicians to counsel patients more effectively and to set appropriate expectations for success.
➢ Patient age appears to be a strong risk factor for nonunion in some bones, a weak risk factor for nonunion in other bones, and perhaps not a significant risk factor for nonunion in certain bones. This observation suggests that there can be substantial bone-to-bone variation in nonunion risk. Age also may be a surrogate for the prevalence of risk factors that potentially increase with age, such as smoking, diabetes, obesity, or nonsteroidal anti-inflammatory drug (NSAID) use.
➢ Smoking has been replicated as a risk factor for nonunion only in retrospective studies involving the humerus and tibia. Smoking appears to have an important effect on nonunion, yet the incremental risk may be rather small, except in context with additional risk factors. Diabetes has been confirmed as a risk factor for nonunion only in retrospective studies involving the foot and ankle. Nonunion risk from diabetes alone may be small and the context of additional risk factors crucial; this could account for why it has been so difficult to predict nonunion in the individual patient.
➢ Prediction of fracture nonunion in the individual patient is a difficult problem. Fracture nonunion may be influenced by complex interactions between biological and surgical risk factors, and the nonunion rate varies from bone to bone. A randomized clinical trial may not appropriately control for all potential correlates and confounders and may require impractical sample sizes. A large claims-based study of real-world fracture-healing outcomes is needed to provide guidance for randomized clinical trials that can test risk factors more rigorously.
The rate of fracture nonunion averages between 5%1 and 10% for long bones2, and the rate may be increasing with time3. Nonunion risk is clearly related to the severity of the injury resulting in fracture4-10. Randomized clinical trials (RCTs) have also shown that variations in nonunion rate or time to union can be associated with different surgical treatments11-21. However, variation in nonunion rate cannot be fully explained on the basis of fracture severity or surgical technique. For example, the rate of nonunion following intramedullary nailing in the femur ranged from 1% in one large study22 to 10% in another large but more recent study13, with little information given to interpret this difference.
Biological risk factors (e.g., age, smoking, diabetes, obesity) may contribute to variation in the fracture-healing rate among patients4,23,24, but such risk factors often are not well understood. Risk factors predictive of fracture nonunion are important for several reasons. Risk factors can identify patients prone to nonunion who might benefit from more aggressive intervention, and a clear idea of nonunion risk may help when choosing between competing therapeutic options. Risk factors also can inform the design of RCTs and can clarify patient inclusion and exclusion criteria25 so that small studies yield more definitive answers. Finally, an understanding of patient risk profiles may enable clinicians to counsel patients effectively and to set appropriate expectations for success. To be useful, prognostic factors for nonunion must be accurate; a putative risk factor that is not reliably associated with nonunion potentially could motivate overly aggressive treatment.
We hypothesize that clinical certainty related to biological risk factors for fracture nonunion is not commensurate with the quality of the published evidence. To test this hypothesis, we critically assessed the literature on risk factors for nonunion and compared these findings with those of a recent survey of orthopaedic surgeons25.
We performed a systematic search of PubMed for all refereed publications relating to risk factors for nonunion over the last twenty years. The search was limited to PubMed as this database contains 19 million references and indexes roughly 89% of relevant literature26,27; PubMed is as sensitive as, and more specific than, Google Scholar for primary medical literature28,29. PubMed initially was searched on January 16, 2014, and all later searches used that index date.
The search terms used were “nonunion fracture” and “risk factors,” and the search was limited to articles in the English language, with abstracts, that were published within the last twenty years. This initial search returned 361 articles, all of which were read, at least in abstract, to identify potential risk factors for nonunion. A risk factor was scored as positive if the article describing that risk factor indicated that there was a significant relationship (p < 0.05) with fracture nonunion. A risk factor was scored as negative if a relationship was tested between a particular risk factor and fracture nonunion but no significant relationship was reported. Case reports and small case series were excluded from detailed consideration as they are generally unable to demonstrate a significant relationship because of small sample size. Case series were included if a minimum of approximately twenty patients were studied, with this threshold being used as a soft cutoff.
All data on bone nonunion were entered into an Excel spreadsheet, which was used to tabulate the following features of each study: journal of publication, year of publication, study title, sample size (N), study type (e.g., retrospective, prospective, RCT), bone evaluated, and risk factors evaluated for nonunion. If a study included no explicit claim that it was prospective in design, it was assumed to be retrospective.
A limitation of the present study is that we relied solely on PubMed to find relevant articles and this strategy may have missed some articles. However, we believe that we found all articles in journals most relevant to practicing orthopaedists.
We found 361 articles about risk factors for nonunion that were published between 1994 and 2014. Of these, eighty articles were reviews or case reports and five articles were otherwise noncontributory to this review. Overall, 154 articles demonstrated a positive association (p < 0.05) whereas 122 demonstrated negative associations. Among the 276 studies that were evaluated, the average number of patients (and standard deviation) was 188.9 ± 337.0. A progressive increase in interest in nonunion over the past twenty years was evident (Fig. 1). The number of studies published per year has increased (R = 0.89; p < 0.0001), but the average study size (N) has not changed.
A range of bones were evaluated for fracture nonunion (Table I). With the elimination of bones that were evaluated only once, 253 articles examined risk factors for nonunion, and forty-five (18%) were prospective. A great deal of heterogeneity existed in terms of the number of patients (N) reported; the largest retrospective study included 3060 patients, and the largest RCT included 2132 patients (Table I), although the average number of patients studied was 188.9.
A wide range of potential risk factors for nonunion have been evaluated in the literature (Table II). Fracture site was the risk factor that was evaluated most often, perhaps because this information is readily available in patient charts. Positive studies—defined as those demonstrating a significant relationship between some risk factor and nonunion—outnumbered negative studies, which evaluated a potential risk factor and failed to demonstrate a significant relationship. Most reports of negative results were made in the context of positive results, perhaps because studies with positive results are more likely to be published than are studies with negative results30.
Orthopaedists are generally concerned with nonunion of the tibia. However, in an analysis of potential risk factors for tibial nonunion (Table III), it became apparent that evidence supporting any particular risk factor can be relatively weak. Although nine of twelve articles indicated that fracture type can predispose to tibial nonunion, only one study each indicated that obesity31, sex32, wound infection33, and bone translation34 affect tibial nonunion, and no study confirmed that diabetes impacts tibial nonunion. Furthermore, none of the tibial studies were prospective.
A discrepancy exists between the ambiguity of the orthopaedic literature and the near-unanimity in interpreting that literature. A recent practitioner survey identified key prognostic features used to assess the risk of fracture nonunion25. A sample of 335 orthopaedic surgeons indicated that several patient-related risk factors predict nonunion, including patient age (identified by 82% of surgeons as being related to nonunion), smoking history (98%), and history of diabetes (97%)25. The evidence supporting these and several additional risk factors is provided below.
Patient age was confirmed as a risk factor for nonunion in thirty-eight (61%) of sixty-two studies (Table II), and 82% of surgeons identified age as a risk factor for nonunion25. A total of six studies demonstrated a positive association between increasing patient age and the risk of nonunion in the humerus35-40, whereas four studies tested but demonstrated no significant association41-44. Seven studies demonstrated a positive association between age and the risk of nonunion in the humerus45-51, whereas five studies tested but demonstrated no significant association9,52-55. Four studies demonstrated a significant association between age and the risk of nonunion in the clavicle56-59.
The impact of age on different bones is noteworthy. The largest single study of humeral nonunion evaluated 1027 patients, eleven of whom had nonunions; the prevalence of proximal humeral nonunion was only 1.1%, and there was no identified effect of age on the healing rate42. The largest study of clavicular nonunion evaluated 868 patients but demonstrated a nonunion rate of 6.2%57. On multivariate analysis, both surgical treatment and patient age were significant predictors of nonunion57. The difference in nonunion rate between the humerus and the clavicle was substantial and may have had an effect on the ability of the investigators to detect age effects with a reasonable sample size.
These results suggest that age may be a strong risk factor for nonunion in some bones (e.g., the clavicle) but a weak risk factor for nonunion in other bones (e.g., the humerus). Furthermore, age could be a surrogate measure of the prevalence of other comorbidities that increase with age and contribute to nonunion, including diabetes, obesity, osteoporosis, smoking, heart disease, and nonsteroidal anti-inflammatory drug (NSAID) use. Without controlling for potential confounders, it is unclear whether age per se is a risk factor for nonunion. The average number of patients (N) across all studies was only 188.9, which may not provide an adequate number of patients to control for confounders of age. The paucity of patients in most studies may make it impractical to identify subtle age effects, such as a threshold below which nonunion is unlikely to occur.
Smoking was confirmed as a risk factor for nonunion in sixteen (64%) of twenty-five studies (Table II). A positive relationship was reported between smoking and nonunion in the mandible60, clavicle61, scaphoid62, scapula63, spine64, long bones65-67, long-bone diaphysis68, proximal part of the humerus69, humeral diaphysis70, middle of the shaft of the humerus36, tibia31,34,71, and femur72. Negative studies demonstrated no link between smoking and nonunion in the spine73-75, femoral diaphysis76, distal part of the humerus41, lower extremity77, long bones78, and tibia33,79. These discrepancies confirm that there can be bone-to-bone variation in nonunion risk.
It is noteworthy that several studies of tibial nonunion disagreed as to whether smoking was a significant risk factor; three studies enrolling a total of 314 patients demonstrated a positive association between smoking and nonunion31,34,71, whereas two studies enrolling a total of seventy-five patients did not demonstrate such an association33,79. This may mean that the effects of smoking can only be seen with a large sample size80 or that the risk of nonunion from smoking is small in comparison with the risk from infection, which was the focus of one study33.
A recent systematic review demonstrated that the multivariate-adjusted odds ratio (OR) for bone nonunion was 2.32 when current smokers were compared with nonsmokers80. That review is exemplary in that 6356 patients from nineteen studies were pooled. Cigarette smoking was associated with a trend toward prolonged fracture-healing time and a significant effect on fracture nonunion, especially in patients with open fracture80. However, smoking was not found to increase nonunion risk dramatically when multivariate logistic regression was used to control for risk factors that correlate with smoking but might affect nonunion risk independently (e.g., heart or lung disease)80.
Although smoking was confirmed as a risk factor for nonunion in 64% of studies, 98% of surgeons identified smoking as a risk factor for nonunion25. One issue might be acceptance by orthopaedists of established wisdom from other medical specialties81. For example, in a review article on wound-healing that was published in 1992, Silverstein noted that, “The association between cigarette smoking and delayed wound healing is well recognized in clinical practice, although extensive controlled studies have yet to be performed”81. That cigarette smoking can impair digital blood flow and superficial wound-healing in the hand82 does not necessarily mean that smoking can impair fracture-healing.
We do not mean to imply that smoking has been accepted as a risk factor for nonunion in the absence of evidence; in fact, twenty-five articles have evaluated smoking as a risk factor since 1994. Furthermore, some articles that were published too early to be included in the present review concurred that smoking is a risk factor for nonunion. A retrospective study that was published in 1986 evaluated 100 patients undergoing laminectomy and spinal arthrodesis (including fifty smokers and fifty nonsmokers) and used the rate of pseudarthrosis as an end point83. Smoking was found to be a risk factor for nonunion independent of age, sex, or race (p < 0.001). A 1993 study of patients with delayed healing of tibial shaft fractures indicated that smokers had a 4.1-fold higher risk of fracture following low-energy injury; the more cigarettes smoked, the longer the time to bone union84. However, despite this long record of careful evaluation of smoking as a risk factor for nonunion, 36% of recent studies failed to confirm a significant relationship (Table II). This may be because many studies were too small to provide a definitive answer. We propose that most individual risk factors are incremental, acquiring substantial predictive power only in the context of additional risk factors.
Diabetes was confirmed as a risk factor for nonunion in five (71%) of seven studies (Table II), yet 97% of surgeons perceived diabetes as a risk factor for nonunion25. Evidence that diabetes is a risk factor for nonunion was strong in studies of foot and ankle fracture85-88. Analysis of 165 diabetic patients showed that, after adjusting for covariates, peripheral neuropathy, duration of surgery, and hemoglobin A1c levels of >7% were significantly (p < 0.05) associated with healing complications88. Another large study indicated that both Type-I and Type-II diabetes roughly doubled the risk of nonunion at the sites of both long and short-bone fractures24. In contrast, negative findings were reported in studies involving the lower extremities77 and tibia79.
As long ago as 1978, diabetes was recognized as a risk factor for nontraumatic osteonecrosis89, and a case series of thirteen diabetic patients, two of whom developed nonunion, indicated that ankle arthrodesis should be undertaken with caution in diabetic patients90. A nonunion rate of 15% (two of thirteen) was reported90, but the sample size was small enough that misclassification of a single nonunion could change this particular series into a series more typical of nondiabetic patients. In studies with few subjects, small changes in the distribution of events can change the results from significant to nonsignificant91. Nevertheless, the results of a recent large study confirmed that diabetic peripheral neuropathy and hemoglobin A1c levels were risk factors for complications following ankle surgery88.
Obesity was confirmed as a risk factor for nonunion in six (75%) of eight studies (Table II). The studies that demonstrated a significant relationship between obesity and the risk of nonunion evaluated the humerus36,70,92, femur45,93, and tibia31, whereas those that demonstrated no relationship evaluated the femur55 and the upper and lower extremities78. Differences between two studies that did45 and did not55 demonstrate a relationship between obesity and femoral nonunion are hard to interpret; although the studies addressed different methods of treatment, both had large sample sizes (ninety-nine and 156 patients, respectively). It may be noteworthy that the study that did not demonstrate obesity to be a risk factor for nonunion evaluated reamed nailing; this method of fracture fixation may be so effective that it obviates obesity as a risk94. Furthermore, obesity may only be a risk factor for nonunion in weight-bearing bones, which could explain why the study that evaluated the upper as well as the lower extremity failed to confirm obesity as a risk factor78.
Comorbid Disease and Alcoholism
Osteoporosis was confirmed as a risk factor for nonunion in two (40%) of five studies (Table II). The studies that demonstrated a significant relationship between osteoporosis and the risk of nonunion evaluated the humerus95 and the tibial plateau96, whereas the studies that failed to find such a relationship evaluated the spine97, femoral shaft98, and all bones99. Two particularly strong studies led to directly conflicting conclusions: the positive study evaluated eighty-one humeral nonunions and compared them with a tenfold-larger group of matched controls95; in contrast, a nested case-control study of 1498 patients with fragility fractures indicated that, although bone quality is diminished in the elderly, bone fragility has no impact on nonunion risk99. In the face of such discordance, more research is clearly needed.
Alcoholism was confirmed as a risk factor for nonunion in four (80%) of five studies (Table II), and 94% of surveyed surgeons identified alcoholism as a risk factor for nonunion25. The studies that demonstrated a significant relationship between alcohol intake and the risk of nonunion evaluated the mandible60,100, femoral neck47, and humeral diaphysis92, whereas the study that failed to demonstrate such a relationship evaluated the lower extremity77. The largest study of the mandible involved 906 patients, twenty-five of whom had nonunion, so it is likely to be robust; among patients who failed to heal, alcohol and drug abuse were both common60,100. Another study that compared 563 patients with 2252 matched controls indicated that alcohol intake had no significant impact on nonunion, although the mandible was not one of the bones evaluated24. These results confirm that there can be bone-to-bone variation in nonunion risk.
NSAIDs were confirmed as a risk factor for nonunion in one (50%) of two studies (Table II), yet 86% of surgeons identified NSAIDs as a risk factor25. A study that evaluated 563 patients indicated that the use of NSAIDs twelve months prior to fracture was associated with a 2.6-fold higher odds ratio of nonunion or delayed union24. Another study demonstrated a significant (p = 0.000001) relationship between NSAID use and the risk of nonunion in the femoral diaphysis76, whereas the study that failed to demonstrate that NSAIDs were a risk factor evaluated spinal fusion101.
Steroids were not confirmed as a risk factor for nonunion in any of the four studies in which they were assessed (Table II)77,97,99,102, although 96% of surveyed surgeons identified steroid use as a risk factor for nonunion25. Neither glucocorticoids (used for the treatment of rheumatological diseases)103 nor inhaled steroids (used for the treatment of asthma)104 were associated with an increased risk of nonunion, although steroids may be associated with an increased risk of fracture, similar to that conferred by osteoporosis103,104.
Studies Are Biased to Report Positive Outcomes
Roughly 78% (216) of 276 published orthopaedic studies were retrospective (Table IV), and retrospective studies demonstrated positive findings in 75% of all cases. These 216 retrospective studies evaluated 325 different potential risk factors, but many additional risk factors may have been evaluated but not reported; nonreporting could have a major impact on the apparent “hit rate” (proportion of positive findings reported). The hit rate for RCTs was 63% (Table IV); because RCTs can be published whether or not they demonstrate a positive finding, the hit rate for RCTs may reveal the hit rate expected without bias. However, RCTs represented only 2% of the orthopaedic literature overall.
Small Studies Are Problematic
Studies with relatively few enrolled patients are open to error resulting from statistical outliers. Among 198 orthopaedic clinical trials, the average sample size was ninety-five patients allocated to two treatment arms91. Studies that demonstrate a significant difference between treatment arms often have few patients enrolled, so treatment differences could be spurious. If an average of only four outcomes had changed from positive to negative among 769 outcomes in 198 clinical trials, the reported significance levels would have been lost91. On average, a significant finding was based on fewer than 8% of patients in a treatment arm, suggesting that small trials might be too unstable to trust91.
Lack of Study Replication Is a Problem
Relatively few biological risk factors for nonunion have been replicated in more than one fracture site (Table V), and only age and comorbid illness were studied in at least five different bones. In contrast, the risk of nonunion has been assessed in five or more bones for many surgical risk factors, including fracture site, displacement, comminution, surgical technique, fixation type, and treatment delay. Risk factors for fracture nonunion that are confirmed across multiple bones are most likely to be important, but this threshold excludes some factors commonly believed to be important nonunion risk factors. Smoking, obesity, sex, and time to weight-bearing may have an impact on fracture nonunion, but the evidence supporting these potential risk factors has been replicated in only two bones each. Diabetes and alcoholism have been confirmed as nonunion risk factors in one bone each.
Prospective Studies Are Relatively Rare
An even smaller group of risk factors has been replicated in prospective studies (Table VI). Fracture type was confirmed as a significant risk factor for nonunion in the spine105-107, humerus40,108, tibia109,110, and long bones111,112. Some widely accepted risk factors such as infection, smoking, obesity, time to weight-bearing, diabetes, and alcoholism have never been confirmed in a prospective study, to our knowledge.
There is a major difference between the number of risk factors supported by retrospective studies (Table V) and the number supported by prospective studies (Table VI). This situation may be due to the relative ease of identifying associations with use of patient charts, which are accessible and very detailed. It is harder and more expensive to confirm such relationships with use of prospectively collected data. Retrospective studies therefore may be more prone to confirmation bias than prospective studies are; we are likely to find what we look for in the data.
Confirmation Bias May Be a Problem
Confirmation bias can be defined as reporting data to confirm a hypothesis, without noting nonconfirmatory evidence113. There is disagreement in the orthopaedic literature (Table II, Table III, and Table IV), whereas conformity would be expected if there was actual confirmation bias. Therefore, we see no evidence of confirmation bias in the orthopaedic literature. It is not known to what extent confirmation bias plays a role in determining which results are written for publication.
The prediction of fracture nonunion in the individual patient is a very difficult problem. We have searched the orthopaedic literature from 1994 to 2013 to summarize what is known about risk factors for nonunion, collating results from 361 clinical studies. We found that there is more ambiguity in the literature than in the minds of surgeons.
Smoking is an excellent example. Smoking is likely a risk factor for nonunion; the negative effects of smoking on general health are well documented, and the specific effects on fracture repair are intuitive. In a survey of board-certified orthopaedic surgeons, 98% agreed that smoking increased nonunion risk. However, only 64% of the studies that examined the link between smoking and nonunion demonstrated a significant association.
Overall, we found a large and largely unexpected lack of consensus in the literature on nonunion risk. Some factors, such as age, have been studied extensively (sixty-two studies, with 61% showing a positive association). In comparison, the literature on diabetes is thin, with only seven recent papers, five of which showed a significant correlation with nonunion.
Heterogeneity in the literature suggests that risk factors combine in complex ways. Individual risk factors are not definitive prognostic tools; smoking likely cannot predict nonunion without other knowledge, such as fracture severity and location, surgical technique, condition of the soft tissue, and a thorough patient history. Risk is relative, rather than absolute, and depends on context rather than fixed rules.
What are the implications of the lack of consensus in the literature? First, global risk factors for nonunion are unlikely to emerge from RCTs. With a low incidence of nonunion and many variables in play, it is economically infeasible to conduct RCTs for all fractures. Second, the results of an RCT on tibial nonunions that have been treated with nailing may not inform the assessment of risk of nonunion of clavicular fractures.
We believe there are two promising paths forward. The first is by harnessing the power of so-called Big Data. With the advent of electronic medical records and large databases, it may be possible to use statistical tools to tease apart the contributions of risk factors singly and in combination. The second promising path is through the compilation of large, prospective, real-world data sets that can be used to test the predictive power of nonunion diagnoses. If there are constellations of risk factors, these can be tested empirically against real patients in real time. The prognostic power of such metrics potentially can be enhanced to more reliably predict nonunion in the presence of biological risk factors.
Source of Funding: The source of funding for this work was Bioventus LLC. Three authors (R.Z., S.M., and G.J.D.R.) are consultants for Bioventus, and one author (R.G.S.) is an employee of Bioventus.
Investigation performed at Bioventus LLC, Durham, North Carolina
Disclosure: One or more of the authors received payments or services, either directly or indirectly (i.e., via his or her institution), from a third party in support of an aspect of this work. In addition, one or more of the authors, or his or her institution, has had a financial relationship, in the thirty-six months prior to submission of this work, with an entity in the biomedical arena that could be perceived to influence or have the potential to influence what is written in this work. No author has had any other relationships, or has engaged in any other activities, that could be perceived to influence or have the potential to influence what is written in this work. The complete Disclosures of Potential Conflicts of Interest submitted by authors are always provided with the online version of the article.
- Copyright © 2016 by The Journal of Bone and Joint Surgery, Incorporated