Attempted Development of a Clinical Decision-making Model to Assist in Femoral Neck Stress Fracture Screening in Military Trainees

Background: Femoral neck stress fractures (FNSF) are serious injuries prevalent throughout the military, particularly within the initial entry training (IET) population. Delay in diagnosis and management can lead to complications secondary to fracture progression and displacement. Effective screening and identication of FNSF can reduce patient risk, mitigate costs associated with treatment, and prevent the potential medical discharge of IET personnel. There is currently no validated testing to assist in decisions for ordering advanced imaging to rule out FNSF. Military healthcare providers, therefore, must rely on advanced imaging of patients with hip region complaints to rule out suspected FNSF resulting in potential MRI over utilization. The purpose of this project was to determine if a clinical decision model can be developed from clinical ndings to assist clinicians in deciding when magnetic resonance imaging (MRI) is warranted for FNSF screening in IET personnel with hip or groin pain. Methods: This was a prospective observational cross-sectional cohort process improvement project. We screened all IET Soldiers presenting to a medical clinic with hip or groin pain and evaluated 18 clinical variables. The reference standard was a FNSF diagnosed using fast-sequence MRI. Statistical analyses included correlation, receiver operating characteristic (ROC) curve assessments, and regression. Results: Final statistical analysis included 184 IET Soldiers, of which 49 had a FNSF. Frequency analysis observed the highest occurrence of FNSF in non-Hispanic white females. Ten subjective and physical examination variables had a weak correlation to positive MRI ndings for FNSF. Regression analysis revealed six variables with positive MRI outcome predictive abilities; however, small to medium effect sizes, low ability to account for MRI outcome variance, and poor estimated sensitivity on ROC curve analysis limit this regression model’s clinical usefulness utilizing the identied variables. Conclusions: The regression models evaluating the clinical variables within this project failed to demonstrate adequate predictability and clinical usefulness when screening for the high-risk injury FNSF in an IET population. Clinicians should maintain a low threshold for ordering an MRI to screen for FNSF in this population. Clinicians also cannot rely on subject-reported pain locations or objective testing as denitive assessment signs to distinguish between IET Soldiers with and without a FNSF. Trial Registration: study deemed Brooke

as 21%. 10 Femoral neck stress fractures (FNSF) make up between 2-10% of all lower extremity stress fractures in military recruits and may have the greatest potential for long-term complications. 4,7,10,12 Sequelae of femoral neck stress fractures include fracture displacement, avascular necrosis, and in the most severe cases may require total hip arthroplasty. 5,7,13,14 In a retrospective study, Chalupa et al. reported that 66% of initial entry training (IET) Soldiers who required surgical xation and 48% who underwent nonsurgical care for FNSF could not return to full training, resulting in a medical discharge. 5 Timely, accurate identi cation of these serious injuries may reduce the risk of fracture progression and associated sequelae. 8,15 Early diagnosis of FNSF can be challenging. Because clinical presentation is variable and physical examination is of questionable utility, 16,17 the current diagnostic process for FNSF relies almost exclusively on imaging. 7,17−19 Radiographs are speci c diagnostic tests 19,20 and may identify stress fractures in the advanced stages, 16,20 but are not typically su ciently sensitive to screen for FNSF. 7,16,17,20 Fast-sequence magnetic resonance imaging (MRI) is an alternative to traditional sequence MRI studies. 6,7 Fast-sequence T1-weighted MRI has a sensitivity and speci city greater than 99% for diagnosing a fracture, avascular necrosis, or muscle injuries in the hip or pelvic regions, making this the screening modality of choice when available. 6,7,21 Fast sequence hip MRI is capable of completing examinations within 10 min, as compared to normal sequence MRI which can take up to 40 minutes without contrast. 6,7 The intent of this project was to evaluate the utility of single and combined history and examination items in determining when a FNSF should be considered in the differential diagnosis of IET Soldiers with hip or groin pain, thus warranting an MRI. The goal is to improve clinical screening decisions resulting in timely diagnosis of FNSF without unnecessary imaging procedures. We hypothesized that a cluster of commonly reported variables will predict FNSF in military recruits using fast-sequence MRI as the diagnostic gold standard.

Methods
The institutional review board at Brooke Army Medical Center (BAMC) approved this design as a process improvement project. All participating physical therapists completed a 1-hour training session on the project intent and data collection procedures. Physical therapists collected examination information on IET Soldiers seeking care for hip or groin pain at two different clinics at Joint Base San Antonio -Fort Sam Houston, TX. One clinic was a direct access physical therapy clinic and the other a troop medical clinic. This allowed IET personnel the exibility to see a physical therapist directly or go to a medical clinic where they would be screened by one or more medical personnel for their complaints. Personnel seen at the medical clinic were managed by a primary care provider, but a physical therapist involved in the study performed all data collection. IET personnel presenting to either clinic with hip or groin pain were determined to be suitable to participate if they did not have contradictions for an MRI examination or a history of orthopaedic hip or pelvic surgery.
Following arrival at one of the two clinics, IET Soldiers meeting inclusion criteria completed a health screening form (Appendix 1) that assessed individual subjective information described in the literature as potentially related to the presence of FNSF. Each evaluating therapist, regardless of clinic, performed an examination based on the Soldier's presenting symptoms and history. While providers tailored the examination to the Soldier's complaint, signs, and symptoms, there was a standardized portion of the examination speci c for data collection purposes (Appendix 2). For IET Soldiers under the care of other providers in the troop medical clinic, the physical therapist only collected project data and recorded the clinical processes for screening IET Soldiers with suspected FNSF. If the evaluating provider in either clinic thought there was su cient clinical suspicion for FNFS, the project leader recommended the patient be referred for plain radiographs and fast-sequence MRI. As an observational project, there was no requirement for MRI utilization, thus, some IET Soldiers did not receive MRI screening and were subsequently not included in our data analysis. For IET Soldiers seen in the direct access physical therapy clinic, a physical therapist provided all care and collected project data.

Outcomes
We used a positive nding for FNSF on either the prerequisite plain radiographs or a subsequent fastsequence MRI as determined by the radiologist as the reference standard for this project. We included only data from IET Soldiers who underwent fast-sequence MRI based on provider suspicion of the presence of a FNSF. Table 1 presents the categories of the 18 variables collected from each subject evaluated. We selected variables based on published association or positive predictive potential for lower extremity stress fractures. Age, gender, and race/ethnicity are potential factors in the development of stress fractures in military recruits and athletes. 11,12 Alcohol consumption and cigarette smoking negatively in uence bone quality, potentially increasing the overall fracture risk of those involved in lower extremity impact activities. 17,22 Prior exercise history, overall aerobic tness, and nutritional status also in uence the occurrence of stress fractures in the military and athletic populations. 10,23,24 We used the SCOFF and the Eating disorder Screen for Primary care (ESP) to screen for eating disorders. [25][26][27] The SCOFF corresponds to speci c words in each of the ve questions within this questionnaire (Sick, Control, One, Fat, and Food). The SCOFF and ESP are validated eating disorder screening questionnaires. 27 The SCOFF has a sensitivity of 78% and a speci city of 88% while the ESP has a sensitivity of 100% and speci city of 71%, each with a cutoff score of two or more abnormal responses. 27 We also collected data on anthropometric measurements, pain locations, gluteus medius muscle strength, and hip rotation range of motion. 10,16,17,28 Finally, we included common orthopedic special tests used for assessing lower extremity stress injuries. 16,17,29 Sample Size Given the novel purpose and exploratory nature of this project, we set a goal of 10 personnel per variable with a target of 180 personnel prior to the start of the project. 30, 31 We anticipated no more than six of the 18 independent variables of interest being clinically feasible to develop a positive predictive rule for FNSF in this population.

Statistical Consideration
We analyzed all data using IBM SPSS Version 26 (Chicago, Illinois) with a p-value of < 0.05 considered statistically signi cant. Descriptive statistics were calculated and reported. Pearson and Spearman's correlation coe cients were calculated for all variables to evaluate the strength of any potential relationships with MRI outcomes. We conducted linear regression analysis on variables found to have signi cant correlations, making note of variables retaining signi cance within the regression coe cient analysis. We removed variables that did not demonstrate signi cance in the prediction model prior to subsequent regression analyses. We continued this step-wise process until all non-signi cant variables were removed from the model. We then generated receiver operating characteristic (ROC) curves on correlated variables to assess individual performance when screening for FNSF with MRI imaging. Due to the high potential for long-term morbidity following a missed fracture 15,32 , we strived to maximize variable sensitivity at 100%, similar to previous literature developing radiography decision rules. 30 We used adjusted R 2 values to estimate regression model effect size and calculated 95% con dence for all relevant data.

Results
We collected data on 265 IET Soldiers seeking care at the two clinics. Eighty-one Soldiers did not undergo MRI screening based on provider decisions and were excluded from the nal data analysis. The nal data analysis included 184 IET Soldiers (140 females, 44 males) evaluated with MRI. Table 2 displays the frequency analysis separated by gender and MRI outcomes. Figure 1 illustrates self-reported pain locations in IET Soldiers with con rmed FNSF and those without a FNSF. Darker colors within each map signify repeated pain locations.

Frequency Analysis
We identi ed 49 FNSFs among the 184 IET Soldiers. Table 2 shows the frequency analysis separated by gender and MRI outcomes. In the current sample, there were more females diagnosed with FNSF (n = 37) compared to males (n = 12). Within our sample, the odds of incurring a FNSF during IET were 0.35 in females and 0.375 in males. Non-Hispanic white females had the highest occurrence of FNSF; 30 IET personnel accounting for 61.2% of all positive MRI outcomes. Of all personnel diagnosed with FNSF, 79.6% were of non-Hispanic ethnicity. Only 8.2% (15) of IET Soldiers reported using chewing tobacco or smoking prior to the start of training, while 17.4% (32) reported alcohol consumption prior to training.
Outcomes of the eating disorder screening revealed six IET Soldiers (four males, 2 females) with an elevated risk when using the ESP screening criteria. Of these six Soldiers, one male and two females tested positive for FNSF. Three male Soldiers showed an elevated risk for eating disorder when using the SCOFF criteria, but none developed a FNSF. The SCOFF and ESP screening questionnaires identi ed two similar male Soldiers as having an elevated risk for eating disorder, neither of which developed a FNSF. A chi-square test for independence found no signi cant association between gender and MRI outcomes, X 2 (1) = 0.012, p = 0.912. The odds for developing a FNSF in females versus males within our project group was 1.044 (95% CI 0.487-2.238).

Correlation Analysis
Initial investigation revealed 10 signi cant correlations as noted in Table 3 displaying pooled correlations  and Table 4 displaying gender-split correlations. To capture all potential predictive variables, we reported both Pearson and Spearman's correlation coe cients as they varied slightly in identifying correlated variables across the pooled and gender-split analyses. All predictor variables were deemed to have a low to moderate strength of correlation to positive MRI outcomes. Only the male gender-split analyses noted a moderate correlation strength with two variables (weight and hop test). All the correlated variables displayed poor ability to discriminate IET Soldiers with and without FNSF based on pooled and gendersplit ROC curve analyses, as noted in Figs. 2, 3, and 4 respectively. The maximum estimated sensitivity for any of the 10 variables was 0.63 (95% CI 0.537-0.722) for pooled, 0.62 (95% CI 0.514-0.729) for female gender-split, and 0.69 (95% CI 0.510-0.860) for male gender-split.

Regression Analysis
Initial linear regression analysis included all signi cant variables the correlation analyses. We conducted both pooled and gender-split regression analyses with outcome data reported in Table 5 and Table 6 respectively. The pooled data regression model displayed statistically signi cant ability to predict MRI outcomes, F (10, 1172) = 4.617, p < 0.005; however, the predictor variables only accounted for 16.6% (adjusted R 2 value) of the variability of MRI outcomes. Adjusted R 2 also signi es a small effect size for this pooled prediction model. Five variables showed signi cance within this model as noted in Table 5. Regression analysis of the female gender-split model also showed a statistically signi cant ability to predict MRI outcomes, F (10, 128) = 5.248, p < 0.005, with the predictor variables accounting for 23.5% of MRI outcome variability. This model also displayed a small effect size, based on the adjusted R 2 value.
The male gender-split regression model was non-signi cant. Subsequent regression models only utilizing signi cant variables from previous models did not improve prediction abilities or effect sizes as noted from adjusted R 2 values.

Discussion
This was the rst attempt to identify clinical ndings predictive of femoral neck stress fractures in IET personnel. We identi ed 10 variables with weak correlation to positive MRI outcomes for FNSF, with no apparent combination of variables demonstrating clinical utility. Although we identi ed signi cant interactions within a regression model, it had limited predictor variable ability to account for MRI outcome variance. Based on these ndings, we are unable to recommend making advanced imaging decisions solely on the post-test results of our tested variables due to the lack of estimated variable sensitivity. With a false negative rate of up to 38%, we lack con dence in these results and therefore cannot recommend any changes to the current FNSF screening methods. We recommend that clinicians retain an extremely low threshold for ordering advanced imaging in this population with the gold standard remaining MRI.
Our results are consistent with previous studies that identify the female gender as the most consistent predictor of FNSF, although FNSF were identi ed in both genders. However, the identi cation of more stress fractures in females may be secondary to a greater proportion of females tested than males. Future efforts to develop a sensitive predictive model may need to focus on additional gender speci c factors to develop a clinically feasible prediction model. In the military training environment, factors related to the female athlete are of interest. Similar to previous reports, 12 we observed a higher incidence of FNSF in races other than black at 93.9%. Only eight participants (six females, two males) used tobacco products prior to the start of basic training and later developed a FNSF. This occurrence may be underrepresented as we did not question or specify about e-cigarette or vaping use. Prior studies have also linked eating disorders or hypocaloric states to the occurrence of stress fractures; however, these studies used subjects previously diagnosed with an eating order. 10,23,33 To our understanding, no studies have explored possible relationships between individual eating disorder screening questions and positive diagnosis of FNSF. With only 6% of personnel displaying ndings suggesting the presence of an eating disorder while also developing a FNSF, we are unable to determine a link between the two.
We corroborated past summative observations that FNSF is a multifactorial disorder posing a challenge to diagnosis with patient-reported history or physical examination methods alone. This diagnostic complexity is depicted through the pain maps of IET Soldiers with and without FNSF, Fig. 1. Patientreported pain maps can be a vital part of the clinical reasoning and diagnostic process as noted from previous studies. [34][35][36][37] While there is high variability in individual clinical presentations, better understanding of pain location as it relates to FNSF may improve initial screening. While they do not hold certainty in a diagnosis, they can provide insight into pain origin, potential referral sources, changes in symptoms over time, and overall dispersion of symptoms. The pain referral maps collected in this project illustrate similarity between IET Soldiers with and without FNSF. Those with and without a fracture reported both anterior and posterior pain commonly and a few reported referral patterns distally into the thigh and knee regions. One difference noted in these maps that may help future investigations is the lack of central low back and sacroiliac region pain in IET personnel with con rmed FNSF. Previous studies have described the positive impact pain referral maps can have on the differential diagnosis process for multiple regions of the body. 34,38,39 This project may indicate FNSF pain does not produce low back pain referral patterns as noted in other literature. 38 To our knowledge, this is the rst project documenting self-reported pain maps in IET personnel with FNSF.

Limitations:
A limitation of our project was the small sample size. Although we reached our targeted sample size of 180 personnel undergoing MRI evaluation, the small noted effect sizes could infer that our sample size was not adequate. Our sample also was predominately female, questioning the outcome of females being at a great risk. Additionally, 81 individuals did not complete MRI evaluation limiting the total number of IET personnel analyzed. This was due to the observational nature of this project and no mandate for providers to screen with MRI. Another limitation may be that our project consisted of individuals undergoing advanced individual training (AIT), which is the second part of their initial military training following basic combat training (BCT). During BCT, there is a high attrition rate secondary to musculoskeletal injuries that typically occurs early in the training process. 1 Military IET personnel completing AIT may be a population that is less susceptible to FNSF than those participating in BCT.
Regarding tobacco product utilization, we only assessed standard tobacco use and not alternative tobacco products. Vaping may pose an underlying stress fracture risk, but we are unaware of any project assessing this risk to date. Overall, our sample reporting tobacco and alcohol use is smaller than previous studies. 40 The prevalence of eating disorders within the military population is understudied to make a comparison. 41 None of the screening within this project assessed consumption of soda, energy drinks, or other consumable stimulants.

Conclusions
While we identi ed multiple variables with the potential to guide clinical decision-making processes regarding the use of MRI for FNSF screening, the clinical importance of these ndings appears limited. Failing to identify a FNSF in IET personnel can result in substantial health complications for the patient, added costs to the military health care system, and a decrease in military unit readiness. We recommend clinicians have a low threshold when deciding whether to screen IET personnel with MRI that are suspected of having a FNSF. Speci cally, when screening personnel in AIT with hip and/or groin pain with a clinical suspicion of FNSF, it may be best to immediately screen with fast-sequence MRI due to the multifactorial nature of the disorder and the likelihood the subject will undergo multiple evaluations with either a primary care physician or physical therapist prior to diagnostic con rmation. This delay can result in additional healthcare costs, possible fracture progression and displacement, and lost training time.

Declarations Ethics Approval and Consent to Participate
This study was deemed as a process improvement project by the Brooke Army Medical Center Institutional Review Board on 7 August 2017.

Consent for Publication
Not applicable.

Availability of Data and Materials
The datasets generated and/or analyzed during the this project are not publicly available due the project being deemed a process improvement project, but are available from the corresponding author on reasonable request.

Competing Interests
The authors declare that they have no competing interests.

Funding
No funding was received for this project.
Author's Contributions JZ -Author was responsible for project design, data collection, data analysis, data interpretation, and manuscript development.
GD -Author assisted in the project design, data interpretation, and manuscript development.
ML -Author assisted in the project design, data analysis, data interpretation, and manuscript development.
JM -Author assisted in the project design, data interpretation, and manuscript development.
CA -Author was responsible for project design, data collection, data interpretation, and manuscript development.