Journal of Orthopaedic Research 13309-316 The Journal of Bone and Joint Surgery, Inc.

0 1995 Orthopaedic Research Society

Mathematical Model for Repair of Fatigue Damage and Stress Fracture in Osteonal Bone
Bruce Martin
Orthopaedic Research Laboratories, University of California at Davis, Davis, California, U X A .

Summary: This paper assembles current concepts about bone fatigue and osteonal remodeling into a mathematical theory of the repair of fatigue damage and the etiology of stress fracture. The model was used to address three questions. (a) How does the half-life of fatigue damage compare with the duration of the remodeling cycle? (b) Does the porosity associated with the remodeling response contribute to stress fracture? (c) To what extent is a periosteal callus response necessary to augment repair by remodeling? To develop the theory, existing experimental data were used to formulate mathematical relationships between loading, damage, periosteal bone formation. osteonal remodeling, porosity, and elastic modulus. The resulting nonlinear relationships were numerically solved in an iterative fashion using a computer, and the behavior of the model was studied for various loading conditions and values of system parameters. The model adapted to increased loading by increasing remodeling to repair the additional damage and by adding new bone periosteally to reduce strain. However, if too much loading was encountered, the porosity associated with increased remodeling caused the system to become unstable; i.e., damage, porosity, and strain increased at a very high rate and without limit. It is proposed that this phenomenon is the equivalent of a stress fracture and that its biological and mechanical elements are significant in the etiology of stress fractures. Additional experiments must be done to test the model and provide better values for its parameters. However, the instability characteristic is relatively insensitive to changes in model parameters.

Stress fractures are an important medical problem in the training of soldiers; in ballet dancers, race horses, and high-performance athletes; and in recreational athletes. Beyond this obvious area of significance, fatigue damage may play an important role in the fragility of bone among the elderly (18,32). Knowledge of the means by which fatigue damage is controlled by remodeling is a prerequisite for understanding stress fractures in these populations. Stress fractures are the clinical manifestation of the accumulation of fatigue damage in bone (5). Engineering analyses of fatigue in bone have established that the number of load cycles required for failure (NfaiJis inversely proportional to the applied strain range (s) (7,9-11). These studies have shown that Nf,,, = k
s-4

(1)

where k and q are coefficients. It is of particular significance that q may be as large as 15, so that small increases in strain greatly reduce Nfa,,. Fatigue damage is not a well defined entity for any

Received January 13.1994; accepted August 17,1994. Address correspondence and reprint requests to R. B. Martin at Orthopaedic Research Laboratories, Research Facility, Room 2000,4815 Second Avenue, University of California Davis Medical Center, Sacramento, CA 95817, U.S.A.

material. In bone, microcracks normally are present and can be distinguished from artifactual cracks by en bloc staining with basic fuchsin (5,15). These cracks are thought to be fatigue cracks because their numbers increase following repetitive loading (3,8,29). Many investigators have hypothesized that internal bone remodeling serves to repair fatigue microcracks (14,16,23,33). Several of these authors have suggested that fatigue damage itself stimulates increased remodeling, so that the repair process involves a negative feedback loop. Burr et al. (3) established an animal model for studying the connection between fatigue damage and remodeling. They loaded canine forelimbs in three-point bending for lo4 cycles, applied at 2 Hz on one day and producing 1,500 microstrain (pz) on the surface of the radius. When the dogs were killed 1-4 days later, microcrack density (cracks/mm2) was increased significantly. Also, osteonal resorption spaces, the first stage of osteonal remodeling, were located adjacent to microcracks more often than would be expected by chance alone. Mori and Burr (29) similarly loaded the left limbs of dogs for lo4 cycles at 2,500 p&. The right limb was loaded in the same manner 8 days later, immediately before the dogs were killed. Other dogs served as controls. There was a 5-fold increase in crack density in the loaded limbs. There also was a signifi-

309

310

B. MARTIN
For convenience, fatigue damage is defined as millimeter of observable crack length per square millimeter of cross section of bone, although the theory would work as well for any definition of damage that can be expressed per unit area of cross section. The rate of darnage formation is assumed to be a function of the loading rate (RL,in cycles per day) and the resulting strain range (s). O n the basis of Eq. 1, the rate of damage formation is assumed to be (28):

cant increase in resorption spaces 8 days after loading but not immediately afterward; this eliminates the possibility that cracks were caused by resorption cavities rather than the other way around. Again, resorption spaces were found adjacent to cracks more often than would be predicted by random associations. When a bone experiences a fatigue challenge, its mechanical properties may be changed not only by damage but also by increased remodeling space (resorption cavities and refilling osteons) produced by the induced remodeling. It has been postulated that this may accelerate progression to a stress fracture (31)-increased remodeling introduces more porosity, which decreases the elastic modulus, elevates the strains produced by the applied loads, and further increases damage formation. Thus, internal remodeling is required to repair fatigue damage, but in the process it may make the bone more susceptible to damage. Sufficient fatigue damage can stimulate the proliferation of woven bone from the periosteal surface (21,35). Presumably, this external modeling response reduces strains in the weakened structure by increasing its cross section, counteracting the negative effects of internal remodeling. It is therefore postulated to be an equally important mechanism for preventing stress fracture. With the assumption that these observations and hypotheses summarize the essential elements of bones response to fatigue damage, what implications do they have, taken together, for the control of such damage and the etiology of stress fractures? This paper is an attempt to assemble these observations into a unified theoretical model to gain a better understanding of the interactions between loading, damage, internal remodeling, and formation of periosteal callus in healthy people, in patients with stress fractures, and in various experimental situations. Specifically, three questions are addressed. (a) How does the half-life of fatigue damage (the time required for half of a bolus of damage to be removed) compare with the duration of the remodeling cycle? (b) Does the porosity associated with the remodeling response contribute to stress fracture even as damage is repaired? (c) To what extent is the periosteal response necessary to augment repair by remodeling?

DIF = k n sqRL = k,@

(2)

where kD is a damage rate coefficient that depends on loading conditions and bone structure and QU = s ~ R , . is defined as the damage potential. (Here, q may or may not be equal to that in Eq. 1.) Equation 2 can be thought of as representing a situation in which only one kind of loading exists (as in most in vitro experiments). Alternatively, s may represent an effective strain range resulting from a mixture of loads that repeat with frequency R, and produce the same QD as the actual loads. Each iteration period (At = 1 day). a number of new basic multicellular units are aclivated within the section: i.e.. a number of tunnelinl: basic multicellular units reach the plane of the section and begin to open resorption cavities in it. The activation frequency (I,) is the number of basic multicellular units that reach the section per square millimeter per day. As each basic multicelM a r unit passes through the section, its resorption cavity expands to the diameter of an osteonal cement line. removing a portion of the cortical area (which is subsequently refilled) and any damage it contains. Assuming that fatigue damage can be removed only by this process and that the rate of removal is proportional to the amount of existing damage (because the more there is. the greater the probabiility that a basic multicellular unit will hit it), the rate of damage removal (or repair) is

DK= D f,nr,F,

(3)

where D is the existing damage and ru is the radius of the osieonal cement line (fa.nr,2is the amount of cross section removed by a l l of the basic multicellular units created i n 1 day). FS is a damage repair specificity factor, which accounts for any mechanisms that serve to direct remodeling to sites of fatigue damage as opposed to strictly random remodeling. Fs is estimated to be about 5. on the basis of the ohserved frequency with which cracks are associated with new resorption cavities (3,6,29). The net rate of accretion of fatigue damage is D = D, - DK.In the equilibrium state. DR= DF and thee corresponding burden of damage waiting to be repaired is

Do = (ku@ncl)/(rrr,2f,,FS.)

(4)

THEORETICAL MODEL
The model melds Martins mathematical formulation of this problem (28) with a computer model for osteonal remodeling ( 2 5 ) . The analysis considers fatigue damage and remodeling as seen in a representative cross section of the diaphysis of a long bone. For the sake of simplicity, the present model assumes that the cross section and adjacent regions of the diaphysis are uniformly loaded in simple axial compression. A sequence of calculations simulates daily mechanical and remodeling changes in the bone. Much of the model is based on simple geometric characteristics of osteonal remodeling by basic multicellular units. These characteristics have been histomorphometrically established for various species (26).

where the 0 subscripts indicate equilibrium values. Initially. a normal pattern of daily loads is assumed io be applied to ithe section, producing an equilibrium strain range (so) and load rate (RLO), and the corresponding equilibrium damage potential (db,,,) is assumed to maintain the resulting damage at an equilibrium level (D,,).To simulate a fatigue challenge, the model allows R,, cycles per day of additional loading, producing a strain range (sE) to be superimposed on the normal activity. The rate of damage formation predicted by Eq. 2 for this additional loading is added to thmt produced by the equilibrium loading.
Db

= k~(so~R +~ Si I, , ~ R L= Eku(@no ) + @rw)

(5)

Values for k,, and q were obtained from the two fatigue doseresponse experiments by Burr and co-workers (3.29) (Appendlx). Then, allowing F, = 5 and f,!,, = 0.0064immiday. Eq. 4 was rear,,, = soqRLo = 9.83 x lo-$ cycles per day. This is ranged to obtain @ equivalent, for example, to RLo= 1,448 cycles per day and s,)= 300 p&.These values, or any equivalent set. can be used as equilibrium loading conditions in the model, and it will approximate thr damage in the t:xperiments of Burr and co-workers (3,29). (An exper-

J Orrhop

Key, Vol. 13, No.

3, I995

M O D E L I N G S T R E S S FRACTURE D Y N A M I C S

311

<
0

0.8 -

1 studied the behavior of the model as this parameter was varied, starting with O.OOS/mm-a value slightly larger than the 0.003/mm measured in the experiment of Mori and Burr. which produced no significant periosteal response. The daily change in periosteal radius is calculated as Arp = MpAt,and cortical area is calculated with the assumption that the endosteal radius is constant at 3 mm. To determine strain, the applied load is divided by the model's crosssectional area and elastic modulus. All of the aforementioned calculations were incorporated in a compiled BASIC program (Fig. 2). The behavior of the system was tested extensively. but. for brevity, only the results relevant to the three questions posed at the end of the introductory section are presented here.

0.000

0.005
DAMAGE, /rnm

0.01 0

FIG. 1 . Graph of Eq. 7, the hypothetical relationship between the activation frequency of basic multicellular units and fatigue damage. The two data points represent values for the normal controls (left) and the experimental dogs (right) in the study of Mori and Burr (29).

iment with many different values for sF must be done to determine more accurately such variables as q and k".) A sigmoidal dose-response relationship is assumed between the activation frequency (f,,) and the current damage (D) (Fig. I ) . Mathematically. this is expressed as
fa =

(faofmax)/(fao + [fattm

- fa01

exp[k,f,,,,(D

DoYDol) ( 6 )

where fa,,, is the maximum allowable activation frequency, taken to be l.0/mm2/day, a value slightly greater than the largest observed value of 0.38/mm2/day (30). kK is a coefficient. which was adjusted so that the numbers of newly activated basic multicellular units per square millimeter approximated those in the study by Mori and Burr (29) (Fig. 1). The model uses the history of f;, to calculate porosity as previously described (25) (Appendix). The elastic modulus (E) is calculated from porosity (P) using the relationship (24.27): E = 15(1 - P)' GPa. Because the model is normalized to an equilibrium condition. it is insensitive to the modulus at zero porosity (15 GPa here). It is recognized that E also may be diminished by fatigue damage. but this effect is not included in the present model. Damage also is assumed to stimulate formation of periosteal bone, which increases the area of the cross section. No doseresponse data have been found relating the apposition rate or periosteal bone (Mp) to fatigue damage, but Turner el al. (34) found that when rat tibias were bent 36 times a day for 12 days. surface bone formation was lamellar at lower loads and converted to woven hone at higher loads. Apposition rates of lamellar bone were linearly proportional to the magnitude of load or strain, but f woven bone was an "all or nothing" response (the production o rate of bone formation was not correlated with load). Mathematically, this may be represcnted by
Mp = k p

RESULTS When the model's parameters were adjusted to simulate the experiments of Burr and co-workers (3,29), as already described, the model reproduced data on resorption spaces and damage (Fig. 3). This is not presented as validation of the model but simply to show how, when porosity, strain, activation frequency, and damage are plotted as functions of time, the model provides insight regarding the overall behavior of the system. The model could be validated by additional experiments of a similar nature. Using this version of the model, I addressed the first question regarding the comparison of the half-life of fatigue damage with the duration of the remodeling cycle. When 10' cycles were applied on a single day, at sE = 2,500 or 3,500 p,Eq. 3 caused the model to remove damage faster in proportion to (a) the amount of damage introduced and (b) the increase in FS(Fig. 4). The duration of the remodeling cycle was 71 days. For sE= 2.500 p,the damage half-life was considerLOADING

t
=

FiFA

I P I

u I
'

I I

H] I ' .

( D - Do)/D,,

M P = M,

for D < DC lor D t Dc.

CORT A R E A . A

(7)

where k, is a coefficient and M, is the apposition rate for woven bone formation. If M,* is the maximum apposition rate for lamellar bone formation, achieved at D = Dc, then k P = M,*/([D, /Do] - 1). The data of Turner et al. (34) suggested using M w = 0.030 mm/day and Mp* = 0.004 mm/day. Finding no information on Dc.

FIG. 2. Flow chart for the model. The calculations are keyed t o equation numbers in the text. Integrations are over the iteration period (At = 1 day) unless otherwise indicated. The calculation of Q (the net bone added daily by all active basic multicellular units) from f, (activation frequency) history is shown in abbreviated form. (Sec the Appendix for details.)

312
0.10

B. M A R T I N

0.05

a
0 2 0. .0 00

b-

0.01 0.00

0.04

r-----l

250

0.000

2 0 40 6 0 80 100
TIME, d a y s

20 40 6 0 8 0 100
TIME, d a y s

FIG. 3. Simulation of the experiments by Burr and co-workers (3,29), showing plots of porosity, strain, activation frequency (ACT F), and damage (DAM) as functions of time, with sE = 2,500 p& (broken line) and sE = 1.500 F&(solid line). Note that the strain graph shows the strains being produced by the normal daily loading (so and Rm),not those produced by the experimental loading, which lasted only 1 day (arbitrarily set at day 5). Data points show values for damage, which were measured in the experiments, and activation frequency, which was deduced from measurement of resorption spaces. There are transient increases in porosity and strain due to increased remodeling, and the rate at which damage is removed is relatively slow. The half-life of the bolus of damage is several months.

ably longer than this unless the repair specificity factor (F,) was quite large (greater than 10). When sE= 3,500 ~ L the E , half-life was about half the remodeling period for Fs = 5. The cost of removing more damage or removing it less efficiently (with a smaller F , ) was a greater transient increase in porosity due to the greater numbers of basic multicellular units activated (left panels in Fig. 4). To address the other two questions regarding the
0.30 I
3
I , , ,

g
g K
a

0.20 0.15 0.25

0 0.10

::::
0.05
0.00 0.20

1 1 ii-.
0.008 I
i i
I

negative effects of the remodeling space and the importance of the periosteal response, I first examined the behavior of the model when no periosteal response was present. Suppose, in contrast to Burr's studies, a constant amount of additional daily loading is superimposed on CDD0using load control. That is, RLE cycles of a load are applied each day such that the resulting strain range is sk = sE,on the first day. Later, the strain produced by the load may change due to
r

- 4
J
0.004

0.006

F=2500pE

-_-_ _ $ 0.002 - - - -_ --___

0.000

------ _ _ _ _ _ _ ___ _

---_--_

g cn
K

.
5
0
0.002
0.000

0 0.15

--_

0.05
0.00

2 0 40 6 0 8 0 1 0 0
TIME, d a y s

20 40 60 8 0 100
TIME, days

-_-

F , =

F , =

.-___-

F , =

15

FIG. 4. The effects of the magnitude of damage and repair specificity factor (Fs) on the transient increase in porosity (left) and the rate of damage removal (right). The top and bottom panels simulate experiments for which R L = ~lo4 cycles on a single day and sE = 2,500 and 3,500 p, respectively. (Fs = 5 was used in all other parts of the study.)

J Orthop Rrs, Voi. 13, No. 3, 1995

M 0D E L I N G S T R E S S FRACTURE DYNAMICS
0.3
7J

313

0.3

g Ln
0
LT

0.2
0.1

; ; . E 0.2
\
k -0.1 t-

o
Q

0.0
7000

0.0 0.01 0

6000

5000 4000
I

'0.006
2 0.004 2 0.002
5
W '

E 0.008

3000
0

5 10 1 5 20 25 30
TIME, rno.

0.000
0 5 10 1 5 20 25 3 0

TIME, mo.

LE

90-

1 0 9 .......

185 ............. c p d

FIG. 5. Graphs of porosity, strain, activation frequency (ACT F), and damage (DAM). The strain plotted is due to the continuing experimental loads and shows the behavior of the model without a periosteal response. As described in the text, ongoing experimental loading initially producing sEI = 2,500 FLE is superimposed on normal loading. If RI.Fis 100 cycles per day (cpd) or less, the system is stable over a lifetime; when RLEslightly exceeds this critical value, instability is precipitated.

remodeling-induced changes in porosity, but the load remains the same, as it would, for example, if an army recruit began to march 5 km with the same pack every day. When this situation is pertained and the experimental loading had a damage potential QDE = sErqRLE, which was less than a critical value cDDc, the additional damage increased the remodeling rate and the porosity and the bone became less stiff, but a new equilibrium was reached (Fig. 5). However, when the damage

potential exceeded QDc, remodeling never "caught up" with the increased rate of damage formation. Porosity, strain, and damage increased at an ever increasing rate, without limit. It is reasonable to consider this instability to be fatigue failure of the model (i-e., a complete stress fracture). If aDD was only slightly super-critical, this instability could be quite insidious. Damage leveled off at what seemed to be a new equilibrium level and then, suddenly, after what could be

\
k -

0.2

0.1
0.oL

'

'

'

'

'
w

0.0
0.01 0
0.008

6000

2
tv,
0

4000

'0.006
2
0.004

9
I

LT

2000

2 0.002
0.000

p-j
- .. :..... -_ ....
0 3 6 9
1 2 1 5
TIME, mo.

12

15

TIME, rno.

321 - - - - 322R LE = 185 ............ 225-------

CPd

FIG. 6. Behavior of the system when a periosteal response is present. This is similar to the behavior depicted in Fig. 5, except that greater values of RLEcan be tolerated before failure. When instability occurred. strain values quickly became greater than the values shown here, exceeding failure strains for bone (approximately 2-3%).cpd = cycles per day.

J Orthvp Res, Vol. 13, No. 3, 1995

314

B. MARTIN

a long time, the damage exploded, rising rapidly at a very high rate. When the model parameters were set to simulate those in Burrs experiments, QDc. = 3.05 x lo- cycles per day (eg., RLE= 101 cycles per day and sE,= 2,500 PE). When 102 cycles per day were applied, it took 20 years for failure to occur, and with 103 cycles per day, it took about 4 years. Slightly higher loading rates resulted in failure after a few months or weeks (not shown). Clearly, without a periosteal response, the model can withstand only relatively small fatigue challenges because of the inherent instability caused by the remodeling space. Now, consider the third question: the models behavior with a periosteal response in place and Dc set at 0.005/mm. Again, sEIwas set at 2,500 p& and RLE was incremented to increase the experimental damage potential. When RLE= 185 cycles per day (which previously produced early failure, Fig. 5), the addition of external bone acted to reduce strain, and the system returned toward a new equilibrium level after a transient response lasting about a year (Fig. 6 ) . Increasing RLEcaused larger transients that ended more quickly. until another critical damage potential was reached, and instability implying failure ensued. Again, the inherent instability of the system was remarkable. As R,E increased, remodeling responded ever more vigorously, and the challenge was overcome more and more quickly. Then, when one more cycle per day was added, instead of damage not decreasing, the system precipitously failed. The corresponding changes in cortical area were SO-lSO%, depending on how long bone was formed. Woven bone formation only began after 3 months when R, = 321 or 322 cycles per day and began after 6 months when RL.E = 225 cycles per day. Once the periosteum rcturncd to lamellar bone formation, the challenge had been met successfully; no oscillations between these two modes were observed. When the parameter Dc was increased, less loading was required to produce failure because the periosteal response was retarded. A very small change in Dc converted a stable system to an unstable one. Instability also was observed when the periosteal doseresponse function was modeled as a sigmoidal curve. DISCUSSION The goal of this analysis was to continue the development of a mathematical model for the repair of fatigue damage by remodeling, incorporating temporal aspects of internal remodeling and the effects of a periosteal response. Predictions were obtained regarding three specific questions. First, the model predicted that the half-life of a bolus of fatigue damage is several months and is substantially reduced if the amount of damage is increased or the remodeling is well directed at foci of damage. Second, the model

predicted that the porosity associated with remodeling to remove fatigue damage can produce a highly unstable state in which strain and damage increase rapidly and without limit under continued loading. This condition may be interpreted as representing a stress fracture. Finally, it was shown that a periosteal response substantially increases the amount ot loading that can be tolerated but does not remove the tendency of the system to be unstable. The most unique feature of this theory is its inclusion of the details of osteonal remodeling. Several investigators have modeled the effects of internal remodeling on bone density and mechanical properties (12,13,19). In each of these studies, bone density was assumed to be a function of a strain-related variable (strain energy density), and that strain was in turn a function of density through the elastic modulus, creating a feedback loop. Carter et al. (12) suggested formulating the effects of fatigue damage on remodeling as in Eqs. 1 and 2, but no previous models have accounted for the details of basic multicellular unitbased remodeling that could significantly affect the mechanical behavior of the system, such as the remodeling space and the time required for the remodeling to occur. BeauprC et al. (2) accounted for the variations in the internal surface area of bone that accompany changes in density; this helped to account for the time-dependency of mechanically adaptive remodeling, but the transient increase in porosity characteristic of remodeling was not modeled. This is the primary strength of the present model, and the feature that acclounts for its unstable behavior. Turning to the limitations of the model, the first, and most easily remedied, is the simple stress state that was assumed (uniaxial compression). This simplification allowed the other aspects of the model to be isolated and explored. The next version of the model should use finite element analysis to calculate more realistic stress states for a diaphyseal region. Many stress fractures scem to occur in characteristic, localized regions, consistent with inhomogeneous stress states (22). However, there is no reason to assume that such inhomogeneities would eliminate the instability effects of the present model: indeed, they might exacerbate them. A second limitation is the assumption that woven bone added by the periosteal response has the same mechanical properties as lamellar bone when strain is calculated. Although this clearly is not the case (and should be corrected in subsequent models), the effect of the current approximation is to reduce rather than increase the instability problem. Woven bone is less stiff than lamellar bone and would thus protect against remodeling-induced increases in strain less effectively than mlodeled here. Another limitation is that the present model does not account for the mechanical

M O D E L I N G S T R E S S FRACTURE D Y N A M I C S

31.5

effects of damage on the modulus of the bone. It is anticipated that adding this effect also would enhance rather than diminish the instability behavior. A more difficult limitation is the lack of knowledge about the relationships between loading and damage. This is a very difficult problem, which can only be solved experimentally. It is, however. entirely analogous to the general problem of defining and measuring a "remodeling stimulus" in other theories of mechanically adaptive remodeling. Similarly, the estimate of F, and the assumed relationships between damage and activation frequency or periosteal apposition in the present model are analogous to the assignment of arbitrary coefficients relating changes in density to strain energy in other models. Parametric studies of the model suggest that its general prediction (that remodeling is capable of repairing damage during a fatigue challenge but produces instability when loading is too severe) is insensitive to such details of the loading-damage-remodeling interactions. Furthermore, the present model can be used to design in vivo experiments to obtain more information about such details and to test predictions concerning damage, activation frequency. porosity, and strain. The question arises as to whether the instability seen in the model is truly indicative of fatigue failure. I propose that it is. because calculated strains quickly exceeded the failure strain of the bone. It is possible that an unknown biological process halts the increase in remodeling space before failure occurs, but that also would halt the repair of damage. The fundamental problem is that damage can be repaired only by remodeling and remodeling requires an increase in porosity. These increases in remodeling space are readily observable when sites of stress fracture are examined histologically (21,31). Therefore, it is reasonable to postulate that the instability observed in the model is indicative of stress fracture. It is far too early for the model to provide conclusive insight concerning the vagaries of stress fractures, but it does support the clinical impression that remodeling intended to repair damage can instead cause damage to increase dramatically when the capabilities of the system are exceeded (31). The unpredictable nature of stress fractures may stem from the fact that the system that produces and controls fatigue damage is nonlinear and small changes in the conditions of such systems can result in vastly different outcomes. This may explain why stress fractures can occur without previous symptoms or in ways that are difficult to predict. Recently, there has been growing interest in the possibility that instability may be present in various aspects of skeletal control involving bone modeling and remodeling (17). I believe that the transient increases in porosity and delays inherent in the re-

modeling process are potentially important sources of instability not only in repair of fatigue damage but also in other mechanically adaptive remodeling. Analyses that only assume relationships between strain and density, for example, and ignore the processes that actually connect these variables. may miss clinically significant elements of the problem.

APPENDIX
Values of q and kD were obtained from the data in Burr's experiments (3.29). as follows. Damage was calculated as the product of crack density and length for each experiment. (Crack lengths were not measured in the first cxperiment [ 3 ] but were found to be about 0.052 mm in both loaded and control regions in the second experiment [29]. Therefore, this number was used for crack length in all cases.) D,, was taken as the mean damage in the control bones of both experiments (0.000754/mm). It was assumed that the difference between the damage in the experimental bones and that in the controls (Do) was the damage produced by the loJ cycles of experimental loading: 1.e.. ADE= DF:- Do.With Eq. 2 written just for the experimental loading, D'F = ADE/At = knsEqRLE. This equation was written for each of the two experiments. and the two equations were solved for q and k,, with At = 1 day and RLE = lo4 cycles per day. For the first experiment ( 3 ) .sE= 0.0015 and ADE= 0.000494/mm and for the second experiment (29), sF = 0.0025 and ADE = 0.002158imm. The result was q = 2.89 and kn = 7.2/mm. The calculations for porosity are based on the idealized geometry of a basic multicellular unit as seen in longitudinal and cross sections (Fig. 7). as well as its longitudinal advancement rate (v) and mean refilling rate (MF).The times required for the resorptive (TK). reversal (T,), and refilling (T,) phases of the remodeling cycle are estimated as T, = LR/v,T, = L,/v, and
2 = IT (rc - r,$

A, = nr;

FIG. 7. Diagram showing the geometry of a typical basic multicellular unit in longitudinal and cross sections. The data for osteonal remodeling that are used in the model are from previous studies (1.20). The values of the variables are radius of the cement line (r<.). 0.095 mm; radius of the haversian canal (rH). 0.020 mm; tunneling rate (v). 0.039 mmlday: refilling rate (MF).0.00121 mmlday: length of reversal region (L,), 0.143 mm; and length of resorption , mm. L~ = "(1, - r,)/MF is the length of the cavity ( L ~ )0.200 refillingregion.
\ I

J Orthop R e s l'ol.

1.3, No. 3. 1995

316

B. MARTIN
7. Caler WE, Carter DR: Bone creep-fatigue damage accumulation. J Biomech 22525-635,1989 8. Carter DR, Hayes WC: Compact bone fatigue damage: a microscopic examination. Clin Orthop 127:265-274,1977 9. Carter D, Caler WE, Spengler DM, Frankel VH: Fatigue behavior of adult cortical bone: the influence of mean strain and strain range. Acta Orthop Scand 52:481-490,1981 10. Carter DR, Caler WE: Cycle-dependent and time-dependent bone fracture with repeated loading. J Biomech Eng 105:166170, 1983 11. Carter DR, Caler WE: A cumulative damage model for bone fracture. J Orthop Res 3:84-YO, 1985 12. Carter DR, Fyhrie DP, Whalen R T Trabecular bone density and loading history: regulation of connective tissue biology by mechanical energy. J Biomech 20:785-794, 1987 13. Cowin SC, Hegedus DH: Bone remodeling. I: Theory of adaptive e1:asticity.J Biomech 6:313-326. 1976 14. Enlow DH: The remodeling of bone. Yearbook Phys Anfhropol 20:19-34,1977 15. Frost HM: Presence of microscopic cracks in vivo in bone. Bull Henry Ford Hosp &25-35,1960 16. Frost H M The Bone Dynamics in Osteoporosis and Osteomalacia. Springfield, Illinois, Charles C Thomas, 1966 17. Harrigan TP, Hamilton JJ: An analytical and numerical study of the stability of bone remodelling theories: dependence on microstructural stimulus. J Biomech 25:477488,1992 [erratum appears in J Biomech 26:365-366. 19931 18. Heaney R P T h e natural history of vertebral osteoporosis. Is low bone mass an epiphenomenon? Bone 13(Suppl2):S23-S26,1992 19. Huiskes R, Weinans H, Grootenboer HJ, Dalstra M, Fudala B, Slooff TJ: Adaptive bone-remodeling theory applied to prosthetic-design analysis. J Biomech 20:1135-1150,1987 20. Jaworski ZF, Lok E: The rate of osteoclastic bone erosion in haversian remodeling sites of adult dogs rib. Calcif Tissue Res 10:103-112, 1972 21. Johnson LC, Stradford HT, Geis RW, Dineen JR, Kerley E: Histogenesis of stress fractures [abstract]. J Bone Joint Surg [Am] 45:1542,1963 22. Jones I3H,Harris JM,Vinh TN, Rubin C: Exercise-induced stress fractures and stress reactions in bone: epidemiology, etiology, and classification.Exerc Sport Sci Rev 17:379-421,1989 23. Martin RB, Burr DB: A hypothetical mechanism for the stimulation of osteonal remodeling by fatigue damage. J Biomech 15~137-139. 1982 24. Martin RB: Porosity and specific surface of bone. Crit Rev Biomed Eng 10:179-222, 1984 25. Martin RB: The usefulness of mathematical models for bone remocleling. Yearbook Phys Anthropol28:227-236,1985 26. Martin RB, Burr DB: Structure, Function, and Adaptation of Compact Bone. New York, Raven Press, 1989 27. Martin RB: Determinants of the mechanical properties of bones. J Biomech 24(Suppl 1):79-88,1991 28. Martin B: A theory of fatigue damage accumulation and repair in cortical bone. J Orthop Res 10818-825, 1992 29. Mori !<, Burr DB: Increased intracortical remodeling following fatigue damage. Bone 14:103-109, 1993 30. Scheiri SS: The effects of repetitive stress on living canine bone [Masters thesis]. Morgantown, West Virginia, West Virginia University, 1978 31. Scully TJ, Besterman G: Stress fracture-a preventable training injury. Mil Med 147:285-287, 1982 32. Sherman S, Heaney RP, Parfitt AM, Hadley EC, Dutta C [eds]: NIA workshop o n aging and bone quality. Bethesda, September 3-4,1992: Proceedings. Calcif Tissue Inr 53(Suppl l):Sl-S180,1993 33. Tschantz P, Rutishauser E: La surcharge mechanique de 10s vivant. Ann Anat Pathol 12:223-248.1967 34. Turner CH, Rho J-Y, Forwood MR: Mechanical thresholds for lamellar and woven bone formation. Trans Orrhop Res SOC113528, 1993 35. Uhthoff HK, Jaworski ZFG: Periosteal stress-induced reactions resembling stress fractures: a radiologic and histologic study in dogs. Clin Orthop 199:284-291, 1985

TF=(rC- rH)/MF. Also estimated from these geometric data are the mean daily amounts of bone removed by each resorbing basic multicellular unit (Qc = Ac/TR mm2 resorbed per basic multicellular unit per day) and added to the section by each refilling basic multimm2 formed per basic mulcellular unit ( a , = AB/TF ticellular unit per day). The history of fa is used to determine how many basic multicellular units are resorbing (NR) and refilling (NF)on the current day. NRis found by numerically R days ago to the present time, integrating fa from T and NF,by integrating from (TR + TI + TF)to (TR+ TI) days ago. Porosity has two components: PHC,due to completed haversian canals, and PRs,due to the remodeling space, defined as the combined areas of all resorbing, reversing, and refilling basic multicellular units. As remodeling continues. PHC increases at a decreasing rate because some new osteons overlap the approaches an canals of old ones. Consequently, PHC equilibrium value, which is the ratio of the area of the haversian canal to the area inside the cement line (24). To avoid the distraction of this secondary effect, when the porosity due to active basic multicellular units is the primary concern, PHC is assumed to be constant at its equilibrium value of 4.43%. The model keeps track of total porosity by calculating Q, the net amount of bone added daily per square millimeter of section by all active basic multicellular units:

The term (1 - PHC)serves to correct for new basic multicellular units that overlap an existing haversian canal (26). The daily change in total porosity is calculated as AP = QAt. These calculations are explained further in Martin (25).
Acknowledgment: This work was supported by National Institutes of Health Grant AR41644.

REFERENCES
1. Anderson C, Danylchuk KD: Bone-remodeling rates of the beagle: a comparison between different sites on the same rib. A m J Vet Res 391763-1765,1978 2. Beauprt GS, Orr TE, Carter DR: An approach for timedependent bone modeling and remodeling-theoretical development. J Orthop Res 8:651-661, 1990 3. Burr DB, Martin RB, Schaffler MB, Radin E L Bone remodeling in response to in vivo fatigue microdamage. J Biornech 181189-200, 1985 4. Burr DB, Milgrom C, Boyd RD, Higgins WL, Robin G , Radin EL: Experimental stress fractures of the tibia: biological and mechanical aetiology in rabbits. J Bone Joint Surg [Br] 72:370-375,1990 5. Burr DB, Stafford T Validity of the bulk-staining technique to separate artifactual from in vivo bone microdamage. Clin Orthop 260:305-308, 1990 6. Burr DB, Martin RB: Calculating the probability that microcracks initiate resorption spaces. J Biomech 26:613-616,1993

J Orthop Res. Vol. 13, No. 3, 1995