Stress fracture (SF) is an overuse injury usually occurring in bones bearing heavy weight, mainly the tibia, femur, and metatarsals. These fractures, which may be nascent or complete, result from repetitive subthreshold loading that, over time, exceeds the bone's intrinsic ability to repair itself. Repetitive vigorous stress creates a region of accelerated bone remodeling, which may progress to an SF if the stress continues (4,6,11). The SF occurs when bone is exposed to cyclical, intense, or novel exercise, particularly when an individual significantly increases his or her level of physical activity over a short period of time. High-volume exercise regimens without a proper progressive train-up period will probably result in SF development (9,10,13,17). The severity of this injury is determined according to the impact, the intensity, and the duration of the training. This injury frequently occurs in athletes and in military recruits during army basic training (BT).
Recently, a number of studies have been published that have the potential to make patient-specific modeling a reality. These include finite element based approaches (14,21) and statistical shape modeling (3,5,12,22). Unfortunately, these approaches and models are not easily adaptable to the data obtained from potential military recruits and not applicable because they require more calculations and computational resources than does a simple weighted-factor statistical regression model.
The SF occurs mainly during military training and is a significant concern during combat recruit training in the Israeli Defence Forces (IDF). However, although clinicians are able to diagnose bone stress injury after it occurs with nearly 100% sensitivity using bone scintigraphy (15,16), a method for predicting the occurrence of bone overuse injuries before they occur remains elusive. Several variables collected from individuals because they enter recruit training are also potentially associated with SF risk (i.e., activity history, anthropometrics, fitness) and could be used to develop a model that might predict individuals at risk for SF upon entry to the military.
Recently, Moran et al. (20) suggested a prediction model for SF in young female recruits. This study concluded that a young female recruited to a gender-integrated unit is at risk for SF if she is tall, lean, feels “burnout,” has iron deficiency, and is at the high end of the normal ferritin range. The study was based on data collected over 16 months from 227 female participants.
The purpose of this study was to use data collected from male military recruits at the beginning of combat BT to develop a new, simple to use prediction model for SF, thus indicating those individuals most at the risk of developing SF.
Experimental Approach to the Problem
Data on anthropometrics, activity history, psychological factors, nutrition, fitness, blood samples, and bone quality measures were collected on the induction day from soldiers entering an elite combat unit training program to look at and determine those variables able to predict SF development of soldiers from this military training program. All the measurements and collected data were analyzed and used for the construction of the predicted model for SF. The newly constructed model we have proposed would then be able to help identify individuals at the risk of developing SFs during basic and advanced training and enable the prevention of orthopedic injuries through the adjustment of the training program for these individuals.
The research design was a prospective study of male army recruits who volunteered to serve in an elite combat unit. Collectively, there were 116 soldiers, 18–19 years old, from 2 different companies of the same unit that participated in this study from their first day of recruitment to BT. Data from the first company (n = 57) were used to develop the prediction model, and data from the second company (n = 59) served to validate the model. Data from both companies were collected from November in 2 consecutive years.
All the subjects gave their verbal approval and written informed consent, and the study was approved by the institutional review board of the Committee for Research on Human Subjects, IDF, Medical Corps, the Human Use Review Committee of the Sheba Medical Center, Tel Hashomer, Israel, and by the Human Use Review Committee of the U.S. Army Research Institute of Environmental Medicine, Natick, MA. Inclusion criteria for all the subjects included written and verbal approval for participating in the study and a medical examination before the beginning of the study by the study physician to determine the physiological condition of the subjects. All the subjects successfully completed a 4-day mandatory training course that was conducted 4 months before induction and used as a criterion for acceptance for combat units. Exclusion criteria for all the subjects included medical consideration, subjects' objection for continuation, and intermission of military service or switching from combat duty to a position not requiring physical activity.
Overall, 77 different variables were collected from the subjects at baseline. Data from anthropometrics, activity history, fitness, bone quality, nutrition, blood samples, and psychological factors were collected. A medical evaluation was conducted at baseline and as necessary during the course of the study to document the occurrence of SFs and other overuse injuries.
Height (centimeters) was measured using a stadiometer (accuracy ± 1 cm), and weight (kilograms) was determined with a metric scale (accuracy ± 100 g). Percent body fat was estimated by skinfold thickness measured using a 4-site method (biceps, triceps, subscapula, and suprailiac) (7), using Lange skin fold calipers (Beta Technology, Santa Cruz, CA, USA). The same investigator performed all skinfold measurements on all the subjects.
(a) Maximum Volume of Oxygen Consumption (V[Combining Dot Above]O2max) was measured to assess aerobic capacity using a continuous, uphill, stepwise, treadmill protocol using a metabolic measurement system (SensorMedics Co, CA, USA). The subjects reached maximum oxygen uptake within 10–15 minutes of starting the test. (b) Anaerobic Capacity was measured by the Wingate Anaerobic 30-second cycle Test (1). This test is used to determine peak anaerobic power and anaerobic capacity. The testing device consists of a mechanically braked bicycle ergometer (Ergomedic 894 Ea, Monark, Sweden). (c) Lower Extremity Power and Force (Vertical Jump) were measured and assessed by the Leonardo Ground Reaction Force Platform (Orthometrix, Inc., White Plains, NY, USA). The subject was positioned on the force platform (wearing sneakers) and jumped as high as possible 3 times. The subject was then asked to perform three 1-legged vertical jumps on each leg. Recorded parameters included power (watts), velocity (meters per second), peak velocity (meters per second), displacement (meters), and work (joules). (d) Bar-Or Basic Fitness Test. All the volunteers completed this fitness test, which assesses the subject's basic level of fitness (1). This 3-event test assessed the time to complete a 2-km run and the maximum number of continual push-ups and sit-ups able to be performed until exhaustion (stopping for 2 seconds).
Peripheral Quantitative Computed Tomography (Stratec/Medizintechnik XCT 2000) was used to measure bone and muscle characteristics of the tibia, according to previously established methods (24) at 4, 38, and 66% of the approximated segment length proximal to the distal endplate of the tibia (20).
Different questionnaires were used for the following: (a) Activity Assessment 1 year before recruitment was evaluated by a detailed physical activity questionnaire that included questions on the type (aerobic/nonaerobic), frequency, duration, and the age span during which the subject was involved in the activity. The average number of hours each activity was performed per week was also included. These data were analyzed to place the soldiers into 1 of 3 categories: inactive, moderately active, and very active. (b) Psychological Assessment was evaluated by questionnaires, based on 0–5 or 1–7 scales, which were administered to collect data relating to stress, exhaustion, cohesion, burnout, motivation, self-confidence, psychological parameters, which mediate between stress and its outcomes. (c) Nutrition Profile. The subjects were interviewed on site regarding their dietary intake using a Food-Frequency-Questionnaire developed specifically for the Israeli population. This is a common method used to assess individual long-term dietary intake of foods and nutrients. The questionnaire elicits a subjectively reported “usual frequency” of consuming an item from a list of foods (8,23). (d) Health/Injury History. A health history assessment was conducted during the prerecruitment period. This questionnaire was based on a new recruit survey conducted in 2003 by the IDF's Military Combat Fitness Center.
Blood drawing procedures were performed as follows: Recruits fasted for 10 hours before a blood sample was collected while seated (between 0700 and 0800 hours) according to the after procedure: Approximately 30 ml of blood was drawn from an antecubital vein using sterile venipuncture techniques. Assays were performed in duplicate with an average of both assays being used as the final measure. Samples from each subject were analyzed in the same assay to minimize the effects of assay variability. Assay results for albumin were used to adjust for plasma volume shifts, as appropriate.
Injury surveillance and SF diagnosis took place over the course of the entire 1-year training period. Data were inserted in a personal surveillance table and collected by the commanders of the clinics on the military base where the participants served. Orthopedic monitoring was conducted every 2–3 weeks by an orthopedic surgeon. The physician followed up any orthopedic-related injuries and problems and decided on the appropriate treatment based on clinical need. The SFs were clinically diagnosed and confirmed by 2 orthopedic surgeons after bone scintigraphy or magnetic resonance imaging, and treated based on the IDF Medical Corps Protocol (19).
In this study, we used the exploratory data analysis (EDA) approach (25), which is used for analyzing data sets and to summarize their main characteristics to help us to build the most appropriate simple model. No prior information about the variables that would be included in the analysis outcome (i.e., final prediction model) was available.
The basic concept of constructing the SF prediction model was to calculate the probability of young recruited men to develop SF during combat BT and advanced elite combat training programs. The variables were divided into 4 categories according to the ability and simplicity to measure them. The first category was based only on answers from questionnaires, and the second category was based on anthropometry measurements. The third category involved blood samples, which is an invasive procedure and requires laboratory analysis. The fourth category included more sophisticated and cumbersome measurements taken by special and expensive equipment and that required technicians for operation, logistics, coordination, and availability. Two phases were used in the model construction. Accordingly, for the first phase, all the collected and measured variables from the various categories were pooled to construct the new prediction models for SFs. In the second phase, we used only the noninvasive and easy-to-use variables from the first 2 categories.
As a consequence, logistic regression was used to predict the probability of the discrete outcome by using a set of independent measured variables. In principle, the goal of logistic regression was first to correctly predict the individuals with and without SF using the most parsimonious model constructed from all the variables and second to construct a prediction model that would be user friendly and more applicable. To accomplish this goal, univariate statistical analysis for correlation and significance was performed between each of the 77 collected variables from the 4 different categories and the response variable, which was the existence and nonexistence of SFs.
Unpaired t-tests were used to compare between the diagnosed subjects with SF and those with no SF (NSF). The bivariate correlation between SF existence and the continuous variables was computed by Pearson product-moment correlation with the corresponding p value, which is computed after Fisher's z transformation. Logistic regression was used for multivariate analysis. The dependent variable in our analysis was SF presence, which was a dichotomous variable. The predictor variables (independent) were selected from the set of all the measured variables collected before the start of BT. The final model included only “simple to measure” variables (from questionnaires and anthropometry). Selection methods were used to reveal the best simple subset of variables for SF prediction.
The SCORE algorithm was used to find a specified number of models with the highest likelihood score (chi-square) statistic for all possible model sizes, from 1, 2, 3 effect models, and so on, up to the single model containing all of the explanatory effects, and the FORWARD algorithm was PROC LOGISTIC first estimates parameters for effects forced into the model. These effects are the intercepts and the first explanatory effects in the MODEL statement. Next, the procedure computes the chi-square statistic for each effect not in the model and examines the largest of these statistics. If it is significant at the chosen level (0.1 in our case), the corresponding effect is added to the model. Once an effect is entered in the model, it is never removed from the model. The process is repeated until none of the remaining effects meet the specified level for entry. The statistical computations were performed by SAS9.2 procedures MEANS, TTEST, CORR, and LOGISTIC.
For a description of the accuracy of the SF prediction model and the true positive rate (sensitivity) and the false positive rate (1 − specificity), we used the receiver operating characteristic (ROC) curve method (18). In this method, accuracy of a prediction model is measured by the area under the ROC curve, whereby an area of 1.0 represents a perfect prediction model (100%). The curve is constructed by computing the sensitivity and specificity of the prediction model.
The backward stepwise elimination method was used to achieve a parsimonious model, which adequately describes the data and does not include noncontributory factors at the p > 0.10 level.
Twenty-three of the 57 recruits from the first company (40.3%) and 22 of 59 recruits from the second company (37.3%) were diagnosed with SFs of different grades during the 4-month BT and after 6 months of advanced training, Overall, there were 60 SFs in 45 recruits as follows: 35 in the tibia, 5 in the femur, and 20 in the metatarsal. Of note, during the first 6 months of training, SF occurred mainly in the tibia (71%), and during the after 4 months, mainly in the metatarsal (80%).
The initial prediction SF model was constructed from the 10 potential predictor variables (Table 1) that had the most significant correlations with SF, indicating meaningfulness in predicting SF. The subject status data were fit to this initial prediction model, constructed from the 10 variables, and found to correctly predict presence or absence of SFs in 99.4% of soldiers. This value is the percentage (99.4%) of randomly drawn pairs of subjects with and without SFs and predicts correctly their status: with or without SF.
Next, a logistic regression was used only from the noninvasive and easy-to-measure variables from the first 2 categories (preinduction physical activity questionnaires, anthropometrics, etc.) as specified in Table 1. A prediction model for SF was constructed from 7 variables from these 2 categories and found to correctly predict the presence or absence of SFs in 88.4% of the soldiers. However, we noticed that some variables in this model were intercorrelated (e.g., lean body mass and fat percentage) thus causing some specific odds ratios to be unreliable and, therefore, did not improve the reliability of the model.
To improve the reliability of the suggested model and to follow our prediction model strategy, we reduced the number of variables and chose only variables with a significant contribution to the model. For that purpose, we also used the backward stepwise elimination analysis to find the most prevailing variables for SF. This method involved repeated iterations of model construction and elimination of the least contributory variable in each iteration. Thus, the least significant variable to each model was eliminated before construction of the subsequent model. The final model was completed when all remaining variables from the 2 noninvasive categories were found to contribute to the model significantly, when p < 0.07 (Table 2). This last model, which was constructed from only 3 simple explanatory variables, predicted the presence or absence of SFs in 85.3% of the male soldiers from the first group that participated in this study.
The final model to predict SF (PSF) was as follows:
where PSF is the stress fracture prediction according to the Log Odds(SF), Odds(SF) is the ratio between probability of SF existence and nonexistence, ATn is the aerobic training (times per week), ATt is the aerobic training (minutes per week), and Waist is the circumference (in centimeters).
We analyzed using the ROC method the true positive rate (sensitivity) against the false positive rate (1 − specificity) of SFs for the different possible cut points of the prediction model. Notably, sensitivity is defined as the proportion of subjects with SFs who tested positive, and specificity is defined as the proportion of subjects without SF who tested negative. The sensitivity and the specificity described how well the model discriminated between soldiers with and without SFs. The area under the curve was found to be 0.853, which is considered an excellent prediction model.
To explain the specific contribution of each variable to the prediction model, we used the relative odds ratios for a risk factor with its specific 95% confidence interval. In this case, the odds ratio is defined as the odds of a subject with SF being exposed to the risk factor divided by the odds of a subject without SF being exposed to the same risk factor. To quantify the relative influence of each of the explanatory variables to the prediction SF model, we computed the adjusted odds ratio (Table 3). For each variable, a change in 1 unit will be reflected in a corresponding change in the odds ratio for SF prediction. For example, a subject whose questionnaire responses registered an “aerobic training duration” of 40 minutes is 16.6 times more likely to suffer SF than a subject with an “aerobic training duration” of 30 minutes. Thus, an increase of 10 minutes, from 30 to 40 minutes, in a running period will cause an increase of 16.6 times in the odds of developing SF. To note, this example is also dependent on the other 2 variables used to construct the model (waist circumference and ATt). Accordingly, a newly recruited male to a combat unit with narrower waist circumference (<78 cm) that ran only 1–2 times per week (each training time >60 minutes), will be more likely to develop SF than a matched individual with a wider waist circumference who ran more frequently but not >30 minutes each training period.
In an EDA study, no formal power analysis is possible because no prior model is known (25). In this work, we studied a unique population of highly trained young subjects (Special Forces) with a very high rate (40%) of SFs during their BT and advanced training. Therefore, higher samples are not practically possible in such a population. The usage of an independent validation sample from a similarly trained population is a must in an EDA study. Therefore, a separate database from another company was applied to test the validity of the present model. This database was compiled from measurements taken 1 year later from another 59 recruits and where 22 SFs were diagnosed. Thus, the model predicted the presence or absence of SF in 76.5% of the soldiers who participated from this group.
The present model successfully predicted SF development in new elite combat recruits. The final model was based only on 3 variables: 2 from a questionnaire and another from an anthropometry measurement. These variables adequately identified 85.3 and 76.5% of new recruits from 2 different groups, respectively, for the future development of SF.
The purpose of this new model was to develop a method of screening for SF risk in new recruits. The usage in this simple prediction model by the commanders can lower the occurrence of SF, mainly by revisions in the physical training regimens for individuals at risk. However, this suggested model is not directly relevant to other risk factors, for example, inferior bone quality and geometry, nutrition deficiency, hematology profile, and physical fitness. The inclusion of all the potential risk factors into an SF prediction model would create a better prediction and accuracy of SF occurrence. However, the more variables we have in a model and the more complicated data collection involved, the higher the likelihood that no one will use this model, in addition to the increased potential of errors for the prediction.
Physiological research can be used as a scientific process for guidance, improving performance, and preventing injuries. However, it is commonly known that the translation of the scientific research to daily practice is usually poor. Furthermore, military researchers have often been criticized for studying irrelevant topics, and the outcome of their studies is difficult to implement within a practical setting (2). In this study, we could achieve a better prediction model for SF; however, that model probably would not be used because implementation would most likely not be practical.
The strategy for the construction of this SF prediction model was based on 2 principles. The first was to try to choose the most effortless and comfortable variables to measure from all the 77 collected variables. The second principle was to consider the ability to implement and use the model to predict SFs and thus potentially prevent some of these injuries. The current developed model proposes that the relevance of SF occurrence during training should be considered regardless of the military setting. It is suggested that a model, based on history of physical fitness habits of the recruits 6–12 months before induction and an additional anthropometric measurement, can predict the probability for SF occurrence.
The prediction SF model constructed in this study includes 3 variables from 2 different categories. Aerobic training (times per week) and aerobic training duration (minutes per week) 6–12 months before induction are the most dominant variables in predicting SFs in a young Israeli male soldier recruited for elite combat units. These 2 variables give us information about each individual's personal training program. The high relevancy of the 2 aerobic training variables in the model is most probably influenced by the background of the surveyed soldiers. To be accepted for a combat unit, one must pass a preinduction course, and for that, one has to be very aerobically fit. As a consequence, most of the candidates train alone before recruitment, which, unfortunately, is not always according to a correctly guided training program, thus increasing the risk of orthopedic injuries. We found in this study that aerobic training (times per week) contributes positively and aerobic training duration (minutes per week) negatively for SF occurrence. Therefore, it is suggested that those subjects who a year before recruitment aerobically trained 4–6 times a week for 20–30 minutes are at a lower risk of developing SF.
The third and final variable included in the prediction SF model was “waist circumference.” The inclusion of this variable for the study population represents a stability in body type. Thus, the physical activity and exercise training executed in combat BT and elite combat training have a more pronounced impact on those young recruits that had a narrower waist circumference and were more prone to SF.
The suggested prediction model constructed in this study was found to be accurate with a high sensitivity and specificity with an area under the curve of 0.853 in the ROC curve analysis. However, a tradeoff was found between sensitivity and specificity, with any increase in sensitivity being accompanied by a decrease in specificity. Thus, for most cases with high sensitivity (0.85–1.00), the corresponding specificity was within a lower range of values (0.45–0.80). The practical interpretation of this distribution is that the model better predicts soldiers who will develop SFs during training than soldiers who will not.
Today, almost every recruit that is healthy and relatively fit can be accepted for BT for combat units. The candidates are usually not asked their history of physical activity and previous fitness habits. Thus, in each BT course, a percentage of the recruits will be diagnosed with SF, resulting in the loss of many training days. We suggest here a new approach for consideration, which is the outcome of scientific research and presents evidence based data. However, there is still a need to validate this approach on a larger number of recruits from different backgrounds, countries, cultures, and mentalities. In any case, the cost of implementing the suggested model is negligible in comparison to the benefits that will most likely be achieved.
This study was conducted on Israeli recruits designated to elite combat units. The participants were very highly motivated, and therefore, most of them exercised on their own before recruitment. Therefore, the developed model should also be validated with other populations joining the military for different units and characterized with less motivation. Another point of debate is whether it is appropriate to examine SF throughout the 10-month training period even though the measures were collected on the recruiting day. Because we are discussing an SF prediction model in response to the training environment, our assumption is that the variables used in constructing the model continue to be significant throughout the training period. Further research is necessary to validate this assumption. There are many risk factors for SFs, among them being the geometry and quality of the bones and hematology deficiencies, which were not accounted for in the suggested model. The fact that this model is simple and requires answers for only 2 questions and 1 anthropometric measurement is probably the explanation why it only predicted 76.5% of the SF occurrence in the validation group. Thus, on the one hand, we have a simple and an easy-to-use model, but on the other hand, we are limiting ourselves by not achieving SF prediction of >76.5%.
In conclusion, the model suggested in this study was found to be able to correctly predict the presence or absence of SFs in 85.3 and 76.5% of 2 sample populations. The applied prediction model for SF provides a framework that is easy to implement. A young male recruit for elite combat unit is at a greater risk of developing SF if, before entering BT, he ran <2 times a week and each training was >40 minutes, and his waist circumference was <75 cm. However, because this is an exploratory analytically derived model, further evaluation is required for different populations and under different conditions and protocols (e.g., non-Israeli male soldiers, different training regimes, and different climate conditions).
Currently, there is no practical method or program to assess new recruits for potential SF that might occur during BT. There is no doubt that risk factor profiling of new recruits might enable us to prevent SF by better sorting. Other benefits from this process would include injury prevention that may well preserve bone health, an adjusted training program for those with risk factors, decreased loss of training days, and less pressure on the military medical system. Overall, we should expect better training efficiency, decreased costs related to health care, and better military readiness.
In this study, we present a model that, if adopted, can reduce SF by identifying those individuals at a higher risk. Accordingly, we suggest a simple model consisting of only 3 variables that are easy to monitor. By this method, it is possible within a very short period of time to screen a large population and to spot those that are at risk for SF development. These individuals should receive more medical attention, greater awareness from commanding officers of the problem, and be under a less intensive physical program. The intention is not to prevent any recruit from participating in combat BT. On the contrary, for some new recruits, a careful training program will assure their successful BT completion and will decrease attrition overall.
The SFs in any scenario, but particularly in elite combat units in the military cause harm not only to the person but also to the organizations involved. Predetermining those persons at a greater risk of developing SFs would therefore help improve and decrease the damages for the individual and the organization.
The opinions or assertions contained herein are the private views of the authors and should not be construed as official or reflect the views of the U.S. Department of Defence or the Israeli Defense Forces. The results of this study do not constitute endorsement of the product by the authors or the National Strength and Conditioning Association. The authors have no conflict of interest to disclose. This study was supported in part by a contract from the Medical Research and Material Command (MRMC No. W911QY-08-p-0286).
1. Bar-Or O. The Wingate anaerobic test. An update on methodology, reliability and validity. Sports Med 4: 381–394, 1987.
2. Bishop D. An applied research model for the sport sciences. Sports Med 38: 253–263, 2008.
3. Boyd SK, Muller R. Smooth surface meshing for automated finite element model generation from 3D image data. J Biomech 39: 1287–1295, 2006.
4. Coady CM, Micheli LJ. Stress fractures in the pediatric athlete. Clin Sports Med 16: 225–238, 1997.
5. Cootes T, Taylor C, Cooper D, Graham J. Active shape models-their training and application. Comp Vis Image Understanding 61: 38–59, 1995.
6. Daffner RH. Anterior tibial striations. Am J Roentgenol 143: 651–653, 1984.
7. Durnin JV, Rahaman MM. The assessment of the amount of fat in the human body from measurements of skinfold thickness. Br J Nutr 21: 681–689, 1967.
8. Etzion-Daniel Y, Constantini N, Finestone AS, Shahar DR, Israeli E, Yanovich R, Moran DS. Nutrition consumption of female combat recruits in army basic training. Med Sci Sports Exerc 40 (suppl.): S677–S684, 2008.
9. Evans RK, Antczak AJ, Lester M, Yanovich R, Israeli E, Moran DS. Effects of a four-month recruit training program on markers of bone
metabolism. Med Sci Sports Exerc 40(Suppl.): S660–S670, 2008.
10. Evans RK, Negus C, Antczak AJ, Yanovich R, Israeli E, Moran DS. Sex differences in parameters of bone
strength in new recruits: Beyond bone
density. Med Sci Sports Exerc 40(Suppl.): S645–S653, 2008.
11. Finestone A, Milgrom C, Wolf O, Petrov K, Evans R, Moran D. Epidemiology of metatarsal stress fractures versus tibial and femoral stress fractures during elite training. Foot Ankle Int 32: 16–20, 2011.
12. Fleute M, Lavallee S. Building a complete surface model from sparse data using statistical shape models. In: Proceedings of the Medical Image Computing and Computer-Assisted Intervention—MICCAI'98, First International Conference. New York: Springer ISBN: 3-540-65136-5, 1998. pp. 879–887.
13. Gaeta M, Minutoli F, Scribano E, Ascenti G, Vinci S, Bruschetta D, Magaudda L, Blandino A. CT and MR imaging
findings in athletes with early tibial stress injuries: Comparison with bone
scintigraphy findings and emphasis on cortical abnormalities. Radiology 235: 553–561, 2005.
14. Garcia-Aznar JM, Rueberg T, Doblare M. A bone
remodeling model coupling microdamage growth and repair by 3D BMU-activity. Biomech Model Mechanobiol 4: 147–167, 2005.
15. Hod N, Ashkenazi I, Levi Y, Fire G, Drori M, Cohen I, Bernstine H, Horne T. Characteristics of skeletal stress fractures in female military
recruits of the Israel defense forces on bone
scintigraphy. Clin Nucl Med 31: 742–749, 2006.
16. Lassus J, Tulikoura I, Konttinen YT, Salo J, Santavira S. Bone
stress injuries of the lower extremity: A review. Acta Ortho Scand 73: 359–368, 2002.
17. Matheson GO, Clement DB, McKenzie DC, Taunton JE, Lloyd-Smith DR, MacIntyre JG. Stress fractures in athletes. A study of 320 cases. Am J Sports Med 15: 46–58, 1987.
18. Metz CE. Basic principles of ROC analysis. Sem Nuc Med 8: 283–298, 1978.
19. Milgrom C, Finestone A, Shlamkovitch N, Giladi M, Lev B, Weiner M, Schaffler M. Stress fracture treatment. Orthop Int Ed 3: 363–367, 1995.
20. Moran DS, Israeli E, Evans RK, Yanovich R, Constantini N, Shabshin N, Merkel D, Luria O, Erlich T, Laor A, Finestone A. Prediction model for stress fracture in young female recruits during basic training. Med Sci Sports Exerc 40(Suppl.): S636–S644, 2008.
21. Negus CH, Impelluso TJ. Continuum remodeling revisited: Deformation rate driven functional adaptation using a hypoelastic constitutive law. Biomech Model Mechanobiol 6: 211–226, 2007.
22. Rajamani K, Joshi SC, Styner M. Bone
model morphing for enhanced visualization. In: IEEE Symposium on Biomedical Imaging
. ISBN: 0-7803-8388-5, New York: IEEE 2004. pp. 1255–1258.
23. Shahar D, Shai I, Vardi H, Brener-Azard A, Fraser D. Development of a semi-quantitative Food Frequency Questionnaire (FFQ) to assess dietary intake of multiethnic populations. Eur J Epidemiol 18: 855–861, 2003.
24. Sievanen H, Koskue V, Rauhio A, Kannus P, Heinonen A, Vuori I. Peripheral quantitative computed tomography in human long bones: Evaluation of in vitro and in vivo precision. J Bone
Miner Res 13: 871–882, 1998.
25. Tukey JW. Exploratory Data Analysis. Boston, MA: Addison-Wesley. ISBN 0-201-07616-0, 1977.