Determinants of paraspinal muscle cross-sectional area in male monozygotic twins.

Key Words: Back pain, Magnetic resonance imaging Muscle strength, Physical activity, Twins.

Exercises for the paraspinal and abdominal muscles are common elements of treatment regimens for patients with low back symptoms. In a recent study of the intervention preferences of physical therapists, more than half of those surveyed stated that they would include strengthening exercises in the treatment of patients with acute recurrent low back pain, and 46% said that they would use the exercises for patients with chronic low back pain.[1] Training to increase strength usually is expected to result in an increase in muscle cross-sectional area (CSA) because CSA is a primary determinant of the capacity of a muscle to generate force.[2]

The CSA of the paraspinal muscles has been of interest in studies of back strengthen,[3-7] low back pain,[3,4,6] disk degeneration," and biomechanics[7,9-13] (Tab. 1). The CSAs were determined from computed tomography or magnetic resonance imaging (MRI). Larger paraspinal muscle CSAs were associated with greater trunk force in only one of the studies. Hultman et al[3] found that isokinetic trunk muscle torque measured in a sitting position was related ([Rho] = .61, P [is less than] .05) to the CSA of the erector spinae muscles among subjects who reported no history of low back pain. Patients with chronic low back pain had smaller paraspinal muscle CSAs than did control subjects in 2 studies,[3,6] but in a third study a difference was not found.[4] Smaller CSAs also have been associated with greater disk degeneration.[8]

[TABULAR DATA 1 NOT REPRODUCIBLE IN ASCII]

Researchers have investigated the association between various anthropometric factors and paraspinal muscle CSA.[5,9,11-13] Fat, lean body mass, and body mass index (BMI) were not correlated, but height, weight, and the width, depth, length, and circumference of various parts of the body were related to some of the CSAs (Tab. 2). In most studies there were very few subjects, so lack of statistical significance cannot necessarily be interpreted as a negative finding, because the power to detect an association often was weak.

[TABULAR DATA 2 NOT REPRODUCIBLE IN ASCII]

We studied 130 male monozygotic twins selected from a population-based sample. The fact that the subjects were identical twins allowed us to examine familial aggregation, that is, the combined effects of genetic influences and shared early environment. Our goal was to examine anthropometric factors, physical activity, back and neck pain, overall health, and familial aggregation as determinants of the CSAs of the erector spinae, psoas major, and quadratus lumborum muscles. The general magnitude and relative contributions of these determinants can have implications for the feasibility of increasing and sustaining changes in muscle mass through lifestyle changes in adulthood.

Method

Subjects

The subjects in this study were male monozygotic twins selected from the Finnish Twin Cohort, a population-based twin cohort.[14] The twin pairs were selected on the basis of differences between the twins in occupational materials handling, sedentary work, exercise, occupational driving, or cigarette smoking.[14] Sixty-five pairs of twins with complete data on all factors of interest were included in this investigation. The 130 men ranged in age from 35 to 67 years ([bar]X = 49), in height from 152 to 191 cm ([bar]X = 175 cm), and in weight from 48 to 109 kg ([bar]X = 79 kg). The average difference in lifetime physical work and sport activities between siblings was clear because of the selection process for the twin pairs (Tab. 3). All subjects gave informed consent prior to participating.

Table 3.
Descriptive Summary Statistics for Study Subjects

                                                Mean Difference
                                                Between Twins
                                                 (65 Pairs)
Variable                (n= 130)      SD      [bar]X       SD

Weight (kg)             78.7         11.9      7.1        6.4
Percentage of
  body fat              24.2          5.6      4.2        3.6
Job code, past
  year (scale from
  1-4)(a)                2.1          1.0      1.2        0.9
Maximum weight
  lifted at work,
  past year (kg)        25.4         28.4     28.4       25.9
Sport and exercise
  activities (years,
  at least twice
  per week)             16.5         13.0      9.3        8.5
Exercise frequency
  per week,
  post year              2.7          3.0      2.8        3.1
Frequency of
  leisure-time
  physical activity
  per week, past year    2.2          2.4      2.5       2.3

(a) Scores represent progressively greater use of the back muscles.

Interview

Information on work, exercise, and other activities involving the use of the back muscles, as well as a history of neck and back pain and other health problems, was gathered by use of a structured interview. Each man was asked to compare his health with that of others his own age by using a 5-point scale ("much worse" to "much better"). A detailed health history was taken, and any condition in the past year requiring bed rest of at least 1 week was recorded. Four men had received back surgery. Any subject with a cold, untreated high blood pressure, heart problems, or other physical conditions that could affect maximal exertions, which were required as part of an adjunct study, was excluded.

Every job held during a subject's lifetime was discussed in detail and later coded into I of 4 categories representing progressively greater use of the back muscles. Total years of sedentary and of heavy physical work were calculated. If a subject had held more than one job, all job codes were weighted by the number of months worked and then averaged. The maximum weight lifted at work at least once per month also was recorded. Retired and unemployed men were assigned to the sedentary job category, with zero weight lifted.

Participation in sport and exercise activities was also reviewed and included the frequency and duration of sport and exercise activity since 12 years of age. Running, skiing, walking, soccer, and volleyball were the most common activities, followed by bicycling, weight lifting, and ice hockey. Similarly, the frequency and duration of any other leisure-time activities involving the use of the back muscles were recorded. The most common leisure-time activities were gardening, home maintenance, hiking, hunting, fishing, and woodcutting. The total time spent on a sport or exercise activity was weighted by the number of months per year the sport or exercise activity was conducted. Seven pairs of men had very different histories of weight lifting (mean difference = 12.0 years, at least twice per week; range = 4.2-24.0). One pair of 40-year-old twins differed in their current and past bodybuilding. Since their teens, one brother had been working out 2 hours per day, 6 days per week, at a level that he described as "strenuous," and the other had worked out for 1.5 hours once per week at a "moderate" level, Sixteen other pairs had only one twin with recent activities that could, in our view, have had a training effect on the trunk muscles (mean frequency in past year = 4.6 times per week; range = 2-14). The activities were cross-country skiing (11 men), swimming (4 men), and gymnastics (1 man).

Each subject was questioned about his history of neck and back symptoms. The frequencies of pain in the neck, middle and upper back, and lower back in the preceding 12 months were classified with 7-point scales ranging from "none" to "daily" (Tab. 4). Each subject rated the pain intensity of the worst episode of low back pain in the past 12 months on a scale from 0 to 100. The effect of the worst episode of back pain on 11 common activities of daily living was used to form a disability score ranging from 0 to 4.

Table 4.
Frequencies of Back and Neck Pain in 12 Months Before Testing

                                      Middle and
                      Low Back       Upper Back         Neck
Frequency of Pain     n       %      n        %      n       %

Never                44      34     110      85      52     40
Once                 16      12       3       2      10      8
Two or 3 times       28      22       9       7      13     10
Several times        13      10       3       2      18     14
Monthly              12       9       2       2      15     12
Weekly               11       8       1       1      15     12
Daily                 6       5       2       2       7      5

Body Fat

Bioelectric impedance measurement was used to obtain percentage of body fat, which ranged from 9% to 40% ([bar]X = 24%).[16] Lean body mass was computed on the basis of percentage of body fat.

Magnetic Resonance Imaging

Back muscle morphology was analyzed with MRI. Transverse sections were obtained with the inclination positioned parallel to the L3-4 inter-vertebral space by use of 1.5-T Siemens Magnetom MRI equipment(*) with a surface coil and a spin-echo sequence with a repetition time of 2,450 milliseconds and an echo time of 22 to 90 milliseconds. The slice thickness was 3 mm, and the gaps between the slices measured 0.3 mm. The matrix was 192 X 256, and the field of view was 260 mm. Subjects were positioned supine, and their paraspinal muscles were in a relaxed state. The images were analyzed by use of a computer program written for this purpose. The perimeters of the left and right erector muscle mass (including the erector spinae and multifidus muscles) and the psoas major and quadratus lumborum muscles were traced with a mouse by the primary author (LEG), and the program calculated the areas. The muscle tracing was repeated for 18 subjects chosen at random. The intraclass correlation coefficients[17] (ICCs) for the reliability of the cross-sectional areas of the 6 muscle masses were [is greater than] .94. No correction was made to the digitized area because the differences between the digitally obtained areas and the true physiological areas previously were found to be small ([is less than] 3%).[10]

Data Analysis

Most statistical methods assume that subjects are independent observations, with independent error terms. Because our subjects were selected as twin pairs, however, we believed that this assumption did not hold. The error terms within the twin pairs were correlated, and we believed that it was necessary to use statistical methods to account for this correlation. In the regression analyses on suspected determinants, general estimating equations (GEEs) were used.[18] The appropriate scale of each variable was determined, and the effects of outliers were evaluated. Variables measured in years were age adjusted in the univariate analyses. Because of the number of univariate associations examined, the issue of multiple comparisons arose. The use of a significance level of .05 meant that even if there were no true associations, 5% of the findings could appear significant because of chance alone. Therefore, we believe that the findings must be evaluated in light of their consistency with other factors and the multivariable results. Multivariable modeling with GEEs was used to assess which factors were independent predictors of muscle CSA. Score tests were used to determine variable selection. The probability values for Wald tests are given for the significance of each variable in the multivariable model. All variables with a univariate significance of [is less than] .20 were candidates for the model, but familial aggregation (ie, pair-wise status) was not entered into the models as a variable. Plausible interactions were assessed, but none were significant.

Two methods were used to assess the relative contribution of familial aggregation. The first method was the ICC,[17] traditionally used in twin studies, which estimated the proportion of the variability in CSA attributable to familial aggregation. The second method was ordinary least squares (OLS) regression analysis with the adjusted [R.sup.2] to indicate what percentage of the variability in a given outcome was explained by the variables of interest. It was adjusted for the number of terms in the model, because each additional term will automatically improve the fit to some degree. The contribution of familial aggregation was examined by forming, for each twin pair, a variable that indicated membership in that pair. With 65 pairs of twins, 64 indicator variables were needed (the 65th variable was covered by the overall mean), so the adjustment of the [R.sup.2] value was particularly important.

Results

Erector Muscle Mass

The average CSA for the erector muscles was 48.6 CM2 (SD = 7.9). The average difference between the twin brothers was 4.5 [cm.sup.2] (SD = 3.8). For the 7 pairs of twins with very different histories of weight lifting, the twins who had been involved in weight lifting had, on average, CSAs that were 4.6 cm 2 (SD = 7.4) greater than those of their brothers. The CSAs for twins with recent training activities were slightly smaller than were those of their brothers, with a mean difference of 0.8 (SD = 5.1). The intensive bodybuilder had a CSA 27% greater than that of his brother.

Taller height and greater weight, BMI, lean body mass, percentage of body fat, and maximum weight lifted at work in the preceding year were univariate predictors of greater CSA of the erector muscle mass, as determined with GEEs (Tab. 5). Weight was the strongest predictor in multivariable analyses, followed by the disability score for the worst back pain in the past 12 months and the maximum weight lifted at least once per month at work (Tab. 6). The ICC for the erector muscle mass CSA was .72. In the OLS regression, familial aggregation explained 72% of the variance (3% attributable to age), weight explained 11%, and the remaining factors explained less than 1%, for a total of 83% (Figure).

[TABULAR DATA 5 NOT REPRODUCIBLE IN ASCII]

Table 6. Best General Estimating Equation Multiple Regression Model for Explaining Cross-sectional Areas (CSA) of Erector Muscle Mass(a)

Factor(b)                     Predicted Effect        P (Wald Test)
                              on CSA ([mm.sup.2])

Weight (10-kg increase)            +393                 <.001
Disability from worst
  back pain in past
  12 months (1 point/4)             -86                  .002
Maximum weight lifted
  at work, past year
  (10-kg increase)                  +30                  .049

(a) Erector muscle mass includes erector spinae and multifidus muscles. The mean CSA was 4,856 [mm.sup.2].

(b) Values in parentheses indicate what quantity of the factor was associated with the regression coefficient.

[Figure ILLUSTRATION OMITTED]

Psoas Major Muscles

The average CSA for the psoas major muscles was 28.5 [cm.sup.2] (SD=6.3). The average difference between the twin brothers was 2.8 [cm.sup.2] (SD = 2.4). Among the twins with very different histories of weight lifting, the brothers with a history of extensive weight lifting had s slightly smaller psoas major muscle CSAs, on average 0.6 [cm.sup.2] (SD = 2.5). Similarly, the CSAs of the twins with recent training activities were, on average, 0.7 cm 2 (SD = 4.7) smaller than those of their brothers. Even the CSA of the intensive bodybuilder was only 3% greater than that of his brother.

Taller height; greater weight, BMI, and lean body mass; and more years of sport and exercise activities (age adjusted) were associated with a larger CSA of the psoas major muscles (Tab. 5). Ten years of additional age was associated with a 2.6-[cm.sup.2] (9%) decrease in the CSA. Weight, years of sport and exercise activities, and age remained significant in multivariable modeling (Tab. 7). The ICC for the psoas major muscle CSA was .83. In the OLS regression, familial aggregation explained 83% of the variance (10% attributable to age), weight explained 3%, and sport and exercise activities explained 2%, for a total of 88% (Figure).

Table 7. Best General Estimating Equation Multiple Regression Model for Explaining Cross-sectional Areas (CSA) of Psoas Major Muscles(a)

                            Predicted Effect
Factor(b)                   on CSA ([mm.sup.2])     P (Wald Test)
Weight (10-kg increase)          +199                   <.001
Sport and exercise
  activities (10 years)          +107                   <.001
Age (10-year increase)           -236                    .001

(a) The mean CSA was 2,852 [mm.sup.2].

(b) Values in parentheses indicate what quantity of the factor was associated with the regression coefficient.

Quadratus Lumborum Muscles

The average CSA for the quadratus lumborum muscles was 12.9 [cm.sup.2] (SD = 3.1). The average difference between the twin brothers was 1.8 [cm.sup.2] (SD = 1.6). Again, those subjects with a history of weight lifting had slightly smaller CSAs, with a mean difference of 1.0 [cm.sup.2] (SD = 1.8), as did those subjects with recent training activities, with a mean difference of 0.6 cm 2 (SD = 2.5). The intensive bodybuilder, however, had a CSA 11% greater than that of his brother.

Taller height and greater weight, BMI, and lean body mass were associated with a larger CSA of the quadratus lumborum muscles (Tab. 5). Ten years of additional age was associated with a 1.0-[cm.sup.2] (7%) decrease in the CSA. Increased frequency of low back pain during the year preceding our study was associated with having a smaller quadratus lumborum muscle CSA. In multivariable analyses, lean body mass was the strongest factor, followed by frequency of low back pain in the preceding 12 months and years of leisure-time activities with use of the back muscles (Tab. 8). The ICC for the quadratus lumborum muscle CSA was .71. In the OLS regression, familial aggregation explained 71% of the variance (5% attributable to age), lean body mass explained 2%, back pain explained 1%, and leisure-time use of the back muscles explained 3%, for a total of 77% (Figure).

Table 8. 7Best General Estimating Equation Multiple Regression Model for Explaining Cross-sectional Areas (CSA) of Quadratus Lumborum Muscles(a)

                            Predicted EfFect
Factor(b)                   on CSA ([mm.sup.2])     P (Wald Test)
Lean body mass
  (10-kg increase)              +174                    <.001
Frequency of low back
  pain in post 12
  months (1 point/7)            +29                     .002
Leisure-time activity with
  physical loading (10
  years)                        -23                     .008

(a) The mean CSA was 1,294 [mm.sup.2].

(b) Values in parentheses indicate what quantity of the factor was associated with the regression coefficient.

Discussion and Conclusion

Genetic and early environmental influences appear to be the primary determinants of paraspinal muscle CSA in adulthood. Recent back pain had little effect in our nonpatient population. Occupational, sport, and leisure-time physical activities also had a negligible impact, once familial aggregation was accounted for.

Contrary to the apparent negative findings of several previous studies (Tab. 2), we found weight, height, and BMI to be associated with the CSAs of all 3 muscle groups, in the directions expected. The CSAs in our study were quite similar to those found in other studies (Tab. 9). Because of our large sample size, our study had much greater statistical power to detect such effects. We found body fat to be associated with the erector muscle mass. It was not associated with the erector muscle mass or the psoas major muscles in a study by Tracy et al,[12] but most of the subjects in the latter study were under investigation for disk problems.

Table 9.
Cross-sectional Areas (CSA) of Paraspinal Muscles in Previous
Studies
                                       CSA of Muscles(a)

Reference                  Level(b)        Sex(c)

Parkkola et al[5]          L4-5            M and F
McGill et al[9]            L4-5 (right)    M
McGill et al[10]           L3 (right)      M
Reid and Costigan[7]       Maximal         M
Tracy et al[12]            L3-4 (right)    M
Wood et al[13]             L4-5            M
Parkkola and Kormano[8]    L4-5 (right)    M
                                           F
Parkkola et al[6e]         L4-5            M
                                           F
Hultman et al[3e]          L3              M

                            CSA of Muscles(a)

Reference                  Erector Muscle
                            Mass(d)

Parkkola et al[5]           48 [+ or -] 8 (37-69)
McGill et al[9]             22 [+ or -] 4 (18-28)
McGill et al[10]            29 [+ or -] 4
Reid and Costigan[7]        54 [+ or -] 8
Tracy et al[12]             26 [+ or -] 3 (21-32)
Wood et al[13]              31 [+ or -] 8 (9-44)
Parkkola and Kormano[8]     26 [+ or -] 3
                            22 [+ or -] 4
Parkkola et al[6e]          54 [+ or -] 6
                            45 [+ or -] 6
Hultman et al[3e]           44 [+ or -] 8 (40-44)

                            CSA of Muscles(a)

Reference                  Psoas Major Muscles

Parkkola et al[5]          22 [+ or -] 4 (15-32)
McGill et al[9]            17 [+ or -] 4 (14-28)
McGill et al[10]           16 [+ or -] 4
Reid and Costigan[7]
Tracy et al[12]            15 [+ or -] 3 (8-25)
Wood et al[13]             18 [+ or -] 3 (12-25)
Parkkola and Kormano[8]    19 [+ or -] 3
                           12 [+ or -] 2
Parkkola et al[6e]         35 [+ or -] 6
                           24 [+ or -] 4
Hultman et al[3e]

                              CSA of Muscles(a)

                            Quadratus Lumborum
Reference                       Muscles

Parkkola et al[5]
McGill et al[9]
McGill et al[10]                 7 [+ or -] 2
Reid and Costigan[7]
Tracy et al[12]             7 [+ or -] 2 (4-11)
Wood et al[13]
Parkkola and Kormano[8]

Parkkola et al[6e]

Hultman et al[3e]

(a) [bar]X [+ or -] SD (range).

(b) When more than one level was imaged, L3-4 is reported here.

(c) M = male, F = female.

(d) Erector muscle mass includes erector spinae and multifidus muscles.

(e) Control subjects.

Many of the same factors were associated with each of the 3 muscle groups. Biomechanically, the erector muscle mass and quadratus lumborum muscles are similar in function, and most of die factors associated with the quadratus lumborum muscles were also associated with the erector muscle mass. One notable exception was that percentage of body fat was associated with the erector muscle mass CSAs and not with the quadratus lumborum or psoas major muscle CSAs. Consistent with this result, visual inspection of the MRI scans revealed various amounts of fat in the erector muscle mass, but fat was rarely observed to such notable degrees in the psoas major or quadratus lumborum muscles. Parkkola and Kormano[8] also observed no gross fat deposits in the psoas major muscles. Thus, it may be that people with a higher fat content are also likely to have more intramuscular fat in the erector muscle mass than in the other muscles studied. If there is a "training effect" of carrying around extra fat, it was not observed in the quadratus lumborum or psoas major muscle CSAs.

There were similarities in the 3 methods used. The traditional ICC and the adjusted [R.sup.2] from the OLS regression yielded identical estimates for the percentage of variability explained by familial aggregation. The use of GEEs to control for the correlation between the twins affected a few of the univariate results, but the multivariable models were quite similar to those generated from an OLS regression stepwise approach (data not shown).

In keeping with the major role of genetic and early environmental influences in determining muscle CSA, it is very unusual to find identical twins who differ substantially in their overall degree of activity or exercise involvement. We were able, however, to identify a small number of identical twins with different levels of recent and lifelong participation in strengthening exercises (Tab. 3). In addition to the use of twins with very different histories of participation in physical activities, a strength of this study was the level of detail sought in the interview, which enabled us to estimate many aspects of recent and lifetime physical activities. Explaining 77% to 88% of the variance in the CSAs is remarkable, particularly because inevitable inaccuracies in measuring activity and pain history will reduce their explanatory power, and there will be measurement error in the CSAs as well. Unfortunately, we were restricted to the study of men, because we could not find enough female twin pairs with notably different histories in the factors under consideration.

We had expected to see a clear difference in the CSAs within the pairs with notably different histories of weight lifting or recent training activities. This was not a clear finding, however, except for one pair with notably different histories involving extreme training circumstances. One explanation could be that the levels of activity engaged in by most of the 35- to 69-year-old men were not sufficient to influence muscle CSA beyond that achieved from the activities of daily living. It may be that only very intensive and specific muscle training substantially influences paraspinal muscle CSA and that such training was beyond the levels selected and sustained by most of the adult men participating in this study. It should also be noted that these results may not apply to situations in which immobilization or other marked inactivity results in muscle atrophy.

In this population-based sample of men with various histories of back pain, common variations in the levels of physical activity had little effect on paraspinal muscle CSA. Genetic and early environmental factors had more influence than constitutional and lifestyle choices in adulthood did.

Acknowledgments

We thank Markku Koskenvuo, Hannu Manninen, JDG Troup, and Eila Voipio.

(*) Siemens AG, Erlangen, Germany.

References

[1] Battie MC, Cherkin DC, Dunn R, et al. Managing low back pain: attitudes and treatment preferences of physical therapists. Phys Ther. 1994;74:219-226.

[2] Bruce SA, Phillips SK. Woledge RC. Interpreting the relation between force and cross-sectional area in human muscle. Med Sci Sports Exerc. 1997;29:677-683.

[3] Hultman G, Nordin M, Saraste H, Ohlsen H. Body composition, endurance, strength, cross-sectional area, and density of MM erector spinae in men with and without low back pain. J Spinal Disord. 1993;6:114-123.

[4] Mayer TG, Vanharanta H, Gatchel RJ, et al. Comparison of CT scan muscle measurements and isokinetic trunk strength in postoperative patients. Spine. 1989;14:33-36.

[5] Parkkola R, Kujala U, Rytokoski U. Response of the trunk muscles to training assessed by magnetic resonance imaging and muscle strength. Eur J Appl Physiol. 1992;65:383-387.

[6] Parkkola R, Rytokoski U, Kormano M. Magnetic resonance imaging of the discs and trunk muscles in patients with chronic low back pain and healthy control subjects. Spine. 1993;18:830-836.

[7] Reid JG, Costigan PA. Trunk muscle balance and muscular force. Spine. 1987;12:783-786.

[8] Parkkola R, Kormano A Lumbar disk and back muscle degeneration on MRI: correlation to age and body mass. J Spinal Disord. 1992;5: 86-92.

[9] McGill SM, Patt N, Norman RW. Measurement of the trunk musculature of active males using CT scan radiography: implications for force and moment generation capacity about the L4/L5 joint. J Biomech. 1988;21:329-341,

[10] McGill SM, Santaguida L, Stevens J. Measurement of the trunk musculature from T5 to L5 using MRI scans of 15 young males corrected for muscle fibre orientation. Clinical Biomechanics. 1993;8: 171-178.

[11] Reid JG, Costigan PA, Connie W. Prediction of trunk muscle areas and moment arms by use of anthropometric measures. Spine. 1987; 12: 273-275.

[12] Tracy MF, Gibson MJ, Szypryt EP, et al. The geometry of the muscles of the lumbar spine determined by magnetic resonance imaging. Spine. 1989;14:186-193.

[13] Wood S, Pearsall DJ, Ross R, Reid JG. Trunk muscle parameters determined from MRI for lean to obese males. Clinical Biomechanics. 1996;11:139-144.

[14] Kaprio J, Koskenvuo M, Artimo M, et al. The Finnish Twin Registry--Baseline Characteristics, Section I: Materials, Methods, Representativeness, and Results for Variables Special to Twin Studies. Helsinki, Finland: Kansanterveystieteen Julkaisuja; 1979:M47.

[15] Battie MC, Videman T, Gibbons LE, et al. Determinants of lumbar disc degeneration: a study relating lifetime exposures and MRI findings in identical twins. Spine. 1995;21:2601-2612.

[16] Van Loan MD. Bioelectrical impedance analysis to determine fat-free mass, total body water, and body fat. Spoils Med. 1990;10: 205-217.

[17] Bartko JJ. The intraclass correlation coefficient as a measure of reliability. Psychol Rep. 1966; 19:3-11.

[18] Liang KY, Zeger S. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73:13-22.

LE Gibbons is Research Scientist, Department of Environmental Health, University of Washington, Box 354790, Seattle, WA 98195 (USA) ([e-mail]). She was associated with the Department of Orthopedics, University of Washington, at the time of this study. Address all correspondence to Ms Gibbons.

T Videman, MD, DMSc, is Heritage Senior Scholar, Faculty of Rehabilitation Medicine, University of Alberta, Edmonton, Alberta, Canada, and Professor, Department of Health Sciences, University of Jyveskyle, Jyveskyle, Finland.

MC Battie, PhD, PT, is Professor and Chair, Department of Physical Therapy, University of Alberta.

J Kaprio, MD, PhD, is Co-Director, The Finnish Twin Cohort Study, Department of Public Health, University of Helsinki, Helsinki, Finland, and

Chief Physician for Behavior Genetics, Department of Mental Health and Alcohol Research, National Public Health Institute, Helsinki, Finland.

The study conformed to the ethical standards of the Declaration of Helsinki and US law 45-CFR-46, and received approval from the Human Subjects Committee of the University of Washington and the Ethical Committee of the Department of Public Health at the University of Helsinki.

This study was supported by National Institutes of Health Grant 1 ROI AR 40857-03, the Finnish Ministry of Education, the Alberta Heritage Foundation for Medical Research, and the Academy of Finland.

This article was submitted March 3, 1997, and was accepted December 17, 1997.