Abstract
Background:
Limited research is available regarding return-to-running (RTR) time after femoral neck stress fractures. While studies have shown the prognostic value of image-based grading scales for stress fractures at other sites, few have included femoral neck stress fractures.
Purpose:
To determine if the grade of femoral neck stress fractures based on magnetic resonance imaging (MRI) characteristics correlates with RTR time.
Study Design:
Cohort study; Level of evidence, 3.
Methods:
This study included 24 patients (mean age, 32.9 years; range, 18-51 years) who were diagnosed with 27 femoral neck stress fractures by MRI from 2009 to 2015 at a single sports medicine clinic. All fractures were compression sided and were treated nonoperatively. Charts were reviewed for patient demographics and RTR time. Images were graded from 1 to 4 using the Arendt stress fracture severity scale. Statistical analysis was performed using survival analysis and Cox proportional hazard model to compare the RTR time between grades. Cox regression was repeated, adjusted for age, bone mineral density (BMD), and body mass index (BMI).
Results:
The mean (±standard error of the mean) RTR time in weeks for patients with fractures graded 1 to 4 was 7.4 ± 2.7 (range, 4-11), 13.8 ± 3.8 (range, 6-21), 14.7 ± 3.5 (range, 8.5-24), and 17.5 ± 3.4 (range, 10-32), respectively. Survival analysis indicated that there was a statistically significant effect of fracture grade on RTR time (P = .0065). The Cox model indicated a statistically significant difference in RTR time between grades 1 and 2 (P = .036), 1 and 3 (P = .014), and 1 and 4 (P = .002). The unadjusted hazard ratio was significant (P = .037). There were no statistically significant differences between the remaining grades (P = .82 for grades 2 and 3, P = .37 for grades 2 and 4, and P = .31 for grades 3 and 4). Age (P = .71) and BMD (P = .81) did not have an effect on RTR time. The hazard ratio remained significant (P = .05) after adjusting for age and BMD. BMI tended to have an effect on RTR time (P = .09). After adding BMI to the adjustment, the hazard ratio decreased in significance (P = .13), although sample size also decreased.
Conclusion:
Grade 2 to 4 femoral neck stress fractures require longer RTR time than do grade 1 injuries. Patients with lower BMI tend to require a longer RTR time.
While uncommon, the femoral neck is susceptible to stress injury because of the strong forces during impact activities, with loads exceeding 4 to 8 times body weight during running. 11 Femoral neck stress fractures (FNSFs) occur when repetitive or excessive stress to the trabecular bone of the femoral neck is greater than its ability to sustain these heavy loads, repair microtrauma, and remodel.20,21 They account for less than 5% of all stress fractures, most commonly affecting young endurance athletes, primarily runners and military recruits.7,11 The reported prevalence ranges from 0.3% to 4.7% in military recruits,27,29 while prevalence outside of the military is not well reported.
FNSFs are categorized as compression sided or tension sided. 8 Compression-sided injuries occur at the inferomedial neck and have minimal risk of displacement, as the torque created by gravity results in compression. 11 This is a low-risk stress fracture and can typically be managed nonoperatively. Tension-sided injuries occur at the superolateral neck and are at increased risk for displacement, as the torque created by gravity creates tension.8,11 This is a high-risk stress fracture. Early and appropriate treatment is critical to prevent progression to a displaced fracture, which can result in delayed union, nonunion, and osteonecrosis.9,11,17,19,28 While some cases of tension-sided fractures may be managed conservatively based on the exact location on magnetic resonance imaging (MRI), most require surgical fixation. 25
FNSFs in young, healthy athletes are typically compression sided and frequently managed nonoperatively with weightbearing and activity restriction.6,14-16 The extent and duration of restriction are dictated by symptoms. A stepwise progression from painless nonweightbearing, to partial and then full weightbearing, followed by return to painless nonimpact exercise, and finally to impact exercise is recommended. Typical criteria for return to sport are asymptomatic weightbearing activities and painless passive range of motion, with or without signs of healing on imaging.6,13,16 However, the time needed to progress through these stages can be long and unpredictable. In the population at most risk, endurance athletes and military trainees, prolonged time away from training can have significant consequences, including rapid cardiovascular deconditioning, decreased quality of life, and financial loss. 22
MRI has been shown to be highly sensitive and specific for the diagnosis of stress fractures.17,18 Studies have examined the use of grading scales for injury severity on MRI to predict return-to-running (RTR) time after stress fractures, with promising results (Table 1).1,3,10,12,23 Such studies focused on more common stress fractures in long bones, primarily the tibia and metatarsals. Those that investigated other sites included few, if any, FNSFs. Given the differences in bone composition (trabecular vs cortical) and cumulative load distribution, one cannot assume these results are applicable to the femoral neck. When studies were stratified by characteristics representative of the femoral neck (ie, high-risk site or trabecular bone), they found inconsistent results regarding the utility of MRI grade as a predictor of RTR time.10,23
MRI Grading Scales for Stress Fracture Severity a
STIR, short tau inversion recovery.
It remains unclear if MRI grade can serve as a predictor for RTR time after FNSF. Given the consequences of prolonged time out of training for endurance athletes and military recruits, the populations at most risk, it is important to identify predictors for recovery time and safe return to sport and combat duties. The aim of this study is to determine if the MRI grade of injury, based on a previously established MRI grading scale, correlates with RTR time after FNSF. The null hypothesis is that there will be no difference in the RTR time when comparing FNSFs of different MRI grades.
Methods
Subjects
After institutional review board approval was obtained, a retrospective review was performed on patients diagnosed with FNSF at a single sports medicine clinic from January 2009 to December 2015. Records were reviewed for patients with an International Classification of Diseases, 9th Revision, code for FNSF (733.96) or stress fracture of other bone (733.95). Patients included were those with a documented FNSF by MRI. Patients were excluded if they were diagnosed with a stress fracture at any location other than the femoral neck (n = 100), they had no MRI available for review (n = 1), they refused activity restriction (n = 1), or if there was no clear documentation of RTR time (n = 1) (Figure 1).
Study inclusion and exclusion criteria. FNSF, femoral neck stress fracture; ICD-9, International Classification of Diseases, 9th Revision; MRI, magnetic resonance imaging; RTR, return to running.
Treatment of FNSFs
All injuries were compression sided. All patients were evaluated and treated by the same sports fellowship–trained physiatrist. A nonoperative approach, with a standard, graduated return-to-play protocol guided by pain, was used for all FNSFs. All patients were required to remain nonweightbearing for a minimum of 4 to 6 weeks. Once pain free, they were upgraded to partial weightbearing and then full weightbearing. They gradually began nonimpact exercise and, if free of pain, progressed to low-impact exercise and, finally, running. If they developed pain at any stage, they returned to the prior stage. All patients were enrolled in a standard physical therapy regimen to address deficits in strength throughout the kinetic chain (particularly hip abductors) and range of motion. All were instructed to take acetaminophen as needed for pain.
Data Collection
The medical record of each injury case was reviewed for RTR time. All patients were initially cleared to start running with a program of alternating running and walking (interval training), followed by gradual progression of running time and distance while remaining symptom free. RTR time was, therefore, defined as the time (to the nearest week) from initiation of treatment to initiation of any running as documented in physician or physical therapy notes. All patients were either runners or had running in their training program. Basic patient demographics were reviewed in the medical record and recorded.
The MRIs in all cases were reviewed by a fellowship-trained musculoskeletal radiologist with more than 15 years of experience. The radiologist was blinded to all clinical data except for patient age. The Arendt grading scale was used to grade each stress fracture (Table 1 and Figure 2). Short tau inversion recovery (STIR) and T1- and T2-weighted images were reviewed and graded based on the presence or absence of periosteal or bone marrow edema and/or a fracture line. Grade 1 injuries were defined as having signal change representing periosteal or marrow edema on STIR imaging only. Grade 2 injuries involved signal change in STIR and T2 images. Grade 3 injuries involved signal change on all imaging sequences (STIR, T2, and T1) without fracture line. Grade 4 injuries were defined as having signal change on all imaging sequences with fracture line. This quantitative system was chosen because it is consistent with the biologic classification of bone stress phenomena, 1 has been previously validated,1,3,10,23 and enabled statistical analysis.
MRIs of femoral neck stress fracture grades 1 to 4 according to Arendt grading scale. (A) Coronal short tau inversion recovery (STIR) MRI demonstrating hyperintense edema consistent with grade 1 injury. (B) Coronal T2 MRI demonstrating hyperintense edema consistent with grade 2 injury. (C) Coronal T1 MRI demonstrating significant hypointense edema without fracture line consistent with grade 3 injury. (D) Coronal T1 MRI demonstrating fracture line at the femoral neck consistent with grade 4 injury.
Statistics
Frequencies, means (±SD), and ranges of patient characteristics were calculated to identify trends in all patients and within each grade of FNSF. Patient characteristics included were age, sex, body mass index (BMI), symptom duration before evaluation, weekly running mileage, history of prior stress fracture, evidence of low bone mineral density (BMD) by dual-energy x-ray absorptiometry, and evidence of low vitamin D level. Patient sex was treated as a dichotomous variable, as was history of prior stress fracture (yes or no, without notation of prior number), BMD level (normal or low), and vitamin D level (normal or low). BMD and vitamin D levels are routinely tested in this clinic in patients with trabecular stress injuries (including FNSFs), high-risk stress injuries, or a history of 2 or more stress injuries. Low BMD was defined as a Z score of less than −1.0 for premenopausal females and men younger than 50 years or a T score of less than −1.0 for postmenopausal females and men aged 50 years and older. Given the mix of age and sex among our patients, the use of both Z and T scores was required. For this reason, numeric BMD values could not be directly compared among patients, and BMD was treated as a dichotomous variable. Similarly, low vitamin D level was defined as below the reference range for a specific testing facility. As the reference range for normal vitamin D varies among testing facilities, and the laboratory work was conducted at different facilities, numeric values for vitamin D could not be directly compared among patients; therefore, vitamin D was treated as a dichotomous variable. Age, BMI, symptom duration, and running mileage were treated as continuous variables. Symptom duration and running mileage were obtained from clinic notes and are routinely assessed in this clinic during the evaluation of confirmed or suspected stress injury. BMI was obtained using the most recent height and weight data recorded in the medical record within the prior 6 months. A 1-way analysis of variance or Fisher exact test was performed for continuous and dichotomous variables, respectively, to identify any statistically significant differences among grades.
Injuries were grouped by grade, and the mean RTR was calculated for each group. Survival analysis and Cox proportional hazard model were used to determine if a statistically significant difference in RTR time could be explained by grade of injury. A separate Cox model was repeated to assess if age, BMD, or BMI accounted for any apparent relation between grade and RTR time, as these were felt to be potential confounding variables. A pairwise log-rank test was done for pairwise comparisons among grades. P ≤ .05 was considered to be statistically significant, and >4 weeks was considered clinically significant as substantial deconditioning has been shown to occur during 3 to 4 weeks of inactivity in athletes. 22 Finally, a Kaplan-Meier curve was used for visual inspection of RTR time for each grade.
Three fractures included in the data analysis were repeat fractures in a single patient. Therefore, as a secondary check, a linear mixed-effect model was used to rule out any confound due to potential within subject correlation. All statistics were performed using R version 3.2.1 (R Foundation).
Results
A total of 27 FNSFs occurring in 24 patients were included. Three cases were excluded from data analysis due to (1) unavailability of MRI, (2) patient refusal of activity restriction, and (3) no clear documentation of RTR time (Figure 1).
Of the 27 FNSFs, 23 occurred in females (85.2%). The mean age was 32.9 ± 9.2 years (range, 18-51 years). All patients but one (96.3%) were self-identified runners, with an average weekly mileage of 29.6 ± 20.2 miles (range, 10-75 miles). The average symptom duration before medical evaluation by a specialist was 6.3 ± 6.4 weeks (range, 1.5-33 weeks). The rate of low BMD for all patients was 41.0%; by grade, the rate of low BMD was 0% for grade 1, 60% for grade 2, 50% for grade 3, and 44.4% for grade 4. The rate of low vitamin D for all patients was 18.5%; by grade, the rate of low vitamin D was 0% for grade 1, 20% for grade 2, 25% for grade 3, and 22.2% for grade 4. While grade 2 to 4 injuries had a higher frequency of low BMD (44.4%-60%) and low vitamin D (20%-25%) compared with grade 1 injuries (0.0% for low BMD and low vitamin D), no statistically significant difference was found. One-third (33.3%) of the patients diagnosed with FNSFs had a prior stress fracture. Eight (29.6%) occurred in patients with a prior FNSF (4 contralateral, 4 ipsilateral). There was no significant difference in any of the above variables among grades. The mean BMI was 23.0 ± 4.6 (range, 15.06-32.92). Of note, only 3 patients were classified as underweight by BMI, and all suffered from grade 4 injuries. Using a 1-way analysis of variance, a nonsignificant trend (P = .06) was found for BMI among grades. Further evaluation using a post hoc test showed that the BMI of grade 4 injuries was different than that of grade 3 injuries (P = .05). See Table 2 for details on patient characteristics. The injury occurred at a similar rate on either side (52% left, 41% right) and rarely occurred bilaterally (7%).
Patient Demographics, Overall and by Grade a
BMD, bone mineral density; BMI, body mass index; FNSF, femoral neck stress fracture.
One-way analysis of variance (ANOVA).
An initial 1-way ANOVA revealed a near-significant difference in BMI among grades (P = .06). A post hoc test showed that the mean BMI of grade 4 injuries was different from that of grade 3 injuries (P = .05).
Fisher exact test for testing the null of independence of rows and columns in contingency table.
The mean (± standard error of the mean) RTR time in weeks was 7.4 ± 2.7 (range, 4-11) for grade 1, 13.8 ± 3.8 (range, 6-21) for grade 2, 14.7 ± 3.5 (range, 8.5-24) for grade 3, and 17.5 ± 3.4 (range, 10-32) for grade 4 injuries (Table 3 and Figure 3). The 6- to 10-week increase in RTR time in grades 2 to 4 compared with grade 1 injuries was felt to be clinically significant, while the 1- to 4-week difference between grades 2, 3, and 4 injuries was of less clear clinical significance.
MRI Grade and RTR Time for Each Patient, by Grade a
P value based on Cox model comparing each specified grade to grade 1 injuries; bolded values indicate significant difference (P ≤ .05). MRI, magnetic resonance imaging; RTR, return to running; SEM, standard error of the mean.
Contralateral fracture in the same patient.
Contralateral fracture in the same patient.
Ipsilateral fracture in the same patient.
Average return-to-running time by MRI grade for patients with femoral neck stress fracture (FNSF).
Using survival analysis, grade was found to have a significant effect on RTR time (P = .0065). Using the Cox proportion hazard model, a statistically significant difference in RTR time was found between grade 1 injuries and all other grades (P = .036 for grade 2, P = .014 for grade 3, and P = .002 for grade 4) (Table 3 and Figure 3). The hazard ratio was significant (P = .037). Age (P = .71) and BMD (P = .81) did not have an effect on RTR time. After adjusting for age and BMD, the Cox analysis was not significantly changed, and the adjusted hazard ratio remained significant (P = .05). BMI tended to have an effect on RTR time (P = .09). Inclusion of BMI as a third covariate in the Cox model resulted in a loss of statistical significance of the hazard ratio (P = .13), although the difference between grades 2 to 4 and grade 1 injuries remained significant (P = .062 for grade 2, P = .065 for grade 3, and P = .045 for grade 4). There was no significant difference found between grades 2 and 3 (P = .82), grades 2 and 4 (P = .37), and grades 3 and 4 (P = .31). A Kaplan-Meier curve demonstrates the number of patients, stratified by grade, who were not able to RTR at given time intervals in order to emphasize the difference between grades, particularly grade 1 injuries compared with higher grade injuries (Figure 4). Last, the linear mixed-effect model confirmed that inclusion of the 3 recurrent fractures in the single patient did not alter the statistical significance of these results (P = .05).
Kaplan-Meier curve analysis: return-to-running time by MRI grade for patients with femoral neck stress fractures.
Discussion
The principal findings of this study demonstrated that grade 2 to 4 FNSFs required a longer time to RTR than did grade 1 injuries, independent of age or BMD. BMI was identified as an additional factor that may affect RTR time. Common characteristics of patients who suffer FNSF, stratified by MRI grade of severity, were also identified.
FNSFs occur when repetitive or excessive stress to normal trabecular bone of the femoral neck is greater than its ability to sustain these loads and repair microtrauma.20,21 They have been reported commonly in military recruits and runners. Nearly all research to date has occurred within the military population, and none has assessed if imaging features may aid in predicting recovery time. This study included the largest civilian population suffering from FNSFs reported in the literature, to date, to the best of the authors’ knowledge, and assesses a novel question regarding prognosis.
The majority of patients diagnosed with FNSF were female (85.2%). This is consistent with previous reports of increased incidence of all stress fracture sites in female compared with male athletes. Prior reviews have attributed this, in part, to sex-related factors that affect bone health, including BMD, menstrual history, and diet.1,2,4,5 One-third of the patients suffered a prior stress fracture, with half of those occurring at the femoral neck. Nearly all patients presenting to this sports medicine clinic diagnosed with an FNSF were runners (96.3%), with an average weekly mileage of 29.6 ± 20.2 miles. While runners are at increased risk for stress fractures at other sites and many case reports document FNSFs in runners, there is limited research on FNSF in civilian runners. In a 2003 study, Arendt et al 1 found that femoral stress fractures were more common in college distance runners versus other athletes, but this included only 7 femoral stress fractures, and the site of injury within the femur was unspecified. Prior biomechanical studies have shown that the load at the femoral neck is many-fold greater during running compared with walking. 24 While this may explain the increased incidence, it does not explain why some runners incur this injury and others, with similar biomechanics and weekly mileage, do not. The latter question is beyond the scope of this study but remains of critical importance in prevention strategies. Further research is needed to explore these potential risk factors for FNSF. While no conclusions can be made from this study regarding risk factors for FNSFs, it should be noted that patients diagnosed with FNSFs were commonly female runners with an average weekly mileage ≥25 miles and a history of prior stress fracture, particular FNSF. These characteristics may aid physicians in making the decision to pursue advanced imaging given the low sensitivity of radiographs for FNSF.
While treatment guidelines exist, the RTR time after FNSF has not been well studied, and evidence-based recommendations are lacking. Severity of findings on MRI and bone scan has been shown to correlate with the recovery time needed after stress fractures at other sites. In 1995, Fredericson et al 12 published the first MRI grading scale, based on a progressive pattern of MRI findings associated with stress fractures advancing from periosteal edema to marrow edema and, finally, fracture line (Table 1). The scale was found to correlate with clinical severity of symptoms and return to full activity time in a retrospective review of 33 patients with tibial stress injuries. This scale was later used to retrospectively review 24 patients with bone stress injuries at various sites, including 10 FNSFs, and it was found that only fracture line correlated with longer recovery time. 30 In 1997, Arendt and Griffiths 3 published a modified version of this scale (Table 1), and Arendt et al 1 retrospectively reviewed 74 track and field athletes with stress injuries. These authors showed that high-grade injuries (grades 3 and 4) took longer to heal than did low-grade injuries (grades 1 and 2). They included 7 femoral stress fractures but did not publish the site of these fractures. 1 In 2012, this scale was used to compare RTR time based on fracture severity (high vs low grade) and site (high vs low risk). High-risk and high-grade fractures were found to have a longer RTR time than low-risk and low-grade fractures. However, when looking at the 2 factors together, MRI grade only had a significant effect on RTR time in low-risk fractures. As FNSFs were categorized as high risk, this study would argue against a correlation between MRI grade and RTR time for FNSFs. However, only 1 FNSF was included. 10 Recently, Nattiv et al 23 performed a prospective study of 61 stress fractures using a modified version of this scale (Table 1) and found that MRI grade was an independent predictor for RTR time. In addition, trabecular sites had a significantly longer RTR time than did cortical sites. However, only 2 FNSFs were included. Based on the paucity of FNSFs in these studies, no conclusions regarding the applicability of these scales for FNSFs could be made.
This study differs from all prior studies in that it is the first study, to date, to focus solely on the RTR time of FNSF by image grading. While prior studies have used the grade of severity on MRI as a predictor of RTR time for stress fractures, they have included stress fractures at multiple sites and have included patients diagnosed by bone scintigraphy (for which a similar grading scale exists). This study focused only on FNSF and included only patients diagnosed by MRI because MRI is the most sensitive and specific diagnostic test for stress fractures.18,26 This eliminated any biases that may arise by using different grading systems for MRI versus bone scintigraphy.
This study supports the use of an MRI grading system to aid in the prediction of RTR time after FNSF. The average RTR time progressively increased as the grade of injury increased (Table 3 and Figure 3). Grade 2 to 4 injuries were found to have a significantly longer RTR time than grade 1 injuries (Figures 3 and 4). No statistically significance difference was found between the remaining grades. However, clinicians, athletic trainers, and athletes may argue that the average 1- to 4-week difference between grade 2, 3, and 4 injuries is clinically significant. Further studies with a larger sample size are needed.
These results also raise questions regarding risk factors for higher grade injuries and their effect on RTR time. All patients with low BMD suffered grade 2 to 4 fractures, suggesting that low BMD may risk a higher grade of injury. However, no significant difference was found in BMD between grades in this study. Further, this study found that BMD had no significant effect on RTR time after FNSF. Adjusting for both age and BMD did not change the relationship between grade and RTR time. This supports that grade has an effect on RTR time, independent of age or BMD. This analysis is limited by the small sample size (N = 27) and the treatment of BMD as a dichotomous variable, as discussed previously. Further studies with a larger sample size and treating BMD as a continuous variable are needed to better assess this relationship.
All patients classified as underweight by BMI suffered grade 4 injuries, and the grade 4 group had a significantly lower BMI than did other grades. This suggests that underweight individuals, by BMI, are more likely to suffer grade 4 injuries. BMI tended to have an effect on RTR time after FNSF, with lower BMI requiring a longer time out of running. After adjusting for age, BMD, and BMI, there remained a significant difference in the RTR time between grade 1 injuries and all other grades. It should be noted that, in this final analysis, the adjusted hazard ratio lost significance. This is likely due, in part, to the decreased sample size (n = 21 due to missing BMI data in 6 cases). Further investigation with a larger sample size is needed.
The major limitation of this study is the small sample size; however, this is the largest study of a runner-type population reported in the literature to date. As this injury is uncommon, the number of diagnoses made over the time frame of this retrospective review was limited. While the small sample size should not affect those comparisons in which a statistically significant difference was identified, all analyses that failed to reach significance may be attributed, at least in part, to the small sample size, and a true significance cannot be ruled out. Given the small sample size, recurrent injuries in the same patient were included in data analysis, which may be viewed as a weakness. However, only 3 cases were recurrent fractures in a single patient. All recurrent injuries occurred more than 1 year after the initial injury and had MRI images confirming complete healing of the prior injury. To account for any intrasubject correlations, a linear mixed-effect model was performed that showed no significant effect of these cases on the results. This study is a retrospective review and is, therefore, held to the limitations of all retrospective studies. While results were adjusted for age, BMD, and BMI, additional variables were not adjusted for given the limited sample size of this study. For this reason, additional confounding variables cannot be ruled out. As this study was limited to 1 sports medicine clinic, the generalizability of the results may not be uniform; however, this group represents a large, cosmopolitan, active city environment. A larger scale, multicenter, prospective study is needed to further validate the correlation between MRI grade and RTR time after FNSF. In addition, with a larger cohort, it would be possible to look for additional risk factors for higher grade injuries and prolonged recovery time to aid in diagnoses and treatment protocols, as well as identify potentially modifiable risk factors.
Conclusion
This study demonstrates that grade 2 to 4 FNSFs require a significantly longer time to RTR than do grade 1 injuries, averaging a 6- to 10-week increase, respectively. MRI can provide accurate diagnostic information for FNSFs, as well as important prognostic information regarding time to RTR. This study also suggests that patients with low BMI are more likely to have grade 4 injuries and require a longer RTR time. Additional work is needed to fully define RTR guidelines.
Footnotes
Acknowledgements
The authors thank Ross Zafonte, DO, Spaulding Rehabilitation Hospital; Matthew Provencher, MD, Massachusetts General Hospital; J. Andrew Taylor, PhD, Spaulding Rehabilitation Hospital; Can O. Tan, PhD, Spaulding Rehabilitation Hospital; and Duc Tran, MD, PhD, Spaulding Rehabilitation Hospital. Without their support, this project would not have been possible.
The authors declared that they have no conflicts of interest in the authorship and publication of this contribution.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
