Cummings SR, Palermo L, Browner W, Marcus R, Wallace R, Pearson J, Blackwell T, Eckert S, Black D, for the Fracture Intervention Trial Research Group . Monitoring Osteoporosis Therapy With Bone DensitometryMisleading Changes and Regression to the Mean. JAMA. 2000;283(10):1318–1321. doi:10.1001/jama.283.10.1318
Author Affiliations: Departments of Epidemiology and Biostatistics (Drs Cummings and Black, Ms Palermo, and Mr Pearson), and Medicine (Drs Cummings and Browner and Ms Blackwell), University of California, San Francisco; Veterans Affairs Medical Center, Palo Alto, Calif (Dr Marcus); Department of Preventive Medicine, University of Iowa, Iowa City (Dr Wallace); and Eli Lilly & Co, Indianapolis, Ind (Dr Eckert).
Context The principle of "regression to the mean" predicts that patients with
unusual responses to treatment might represent outliers who are likely to
have more typical responses if treatment is continued without change.
Objective To test whether women who lose bone mineral density (BMD) during the
first year of treatment for osteoporosis continue to lose BMD if the same
treatment is continued beyond 1 year.
Design and Setting Two randomized, double-blind, placebo-controlled trials in 11 US clinical
research centers for the Fracture Intervention Trial and 180 centers in the
United States and other countries for the Multiple Outcomes of Raloxifene
Participants and Interventions Postmenopausal women with low BMD assigned to treatment with 5 mg/d
of alendronate sodium in the Fracture Intervention Trial who completed 2 years
of BMD monitoring and adhered to study medication (n = 2634), and postmenopausal
women with osteoporosis assigned to treatment with 60 or 120 mg/d of raloxifene
hydrochloride in the Multiple Outcomes of Raloxifene Evaluation trial who
similarly completed 2 years of monitoring while adhering to study medication
(n = 3954).
Main Outcome Measures Baseline, 12-, and 24-month hip and spine BMD.
Results Women with the greatest loss of BMD during the first year of treatment
were the most likely to gain BMD during continued treatment. Specifically,
among women taking alendronate whose hip BMD decreased by more than 4% during
the first year, 83% (95% confidence interval [CI], 82%-84%)had increases in
hip BMD during the second year, with an overall mean increase of 4.7%. In
contrast, those who seemed to gain at least 8% during the first year lost
an average of 1% (95% CI, 0.1%-1.9%) during the next year. Similar results
were observed among women taking raloxifene for 2 years.
Conclusions Our data suggest that most women who lose BMD during the first year
of treatment with alendronate or raloxifene will gain BMD if the same treatment
is continued for a second year. These results illustrate the principle of
regression to the mean and suggest that effective treatments for osteoporosis
should not be changed because of loss of BMD during the first year of use.
It is common practice to monitor the effectiveness of medical treatments,
such as treatments for osteoporosis, with surrogate measurements of clinical
outcomes, such as bone densitometry. When the measurement worsens or fails
to improve as expected, clinicians may change treatments or add new ones.
These changes are often followed by satisfying improvements in the measurements.
However, the measurements that are used to monitor the effects of treatment
are imperfect: they may vary from time to time because of inherent variation
in the measurement or natural variations in the patient. This variability
may produce changes in measurements that clinicians could mistake for therapeutic
benefit or failure.
Extreme results of measurements are sometimes due, in part, to random
error. Thus, when extreme measurements are repeated, the result is often closer
to the mean for a population. This is called "regression to the mean."1 This principle predicts that patients with unusual
responses to treatment (such as loss of bone mineral density [BMD] while being
treated with an effective agent) are likely to have more typical responses
(gain in BMD) if treatment is continued and the measurement is repeated.
Most women gain BMD during the first year of treatment with alendronate
sodium; on average, their BMD increases by about 2% to 4%.2,3
We considered the small proportion of women who lose BMD during treatment
with alendronate. If bone loss during the first year of treatment is mainly
due to an inherent biological failure to respond to the drug, then those who
lose BMD during the first year should continue to lose BMD during the next
year of the same treatment. However, if the loss of BMD during the first year
is mainly due to variations in the measurement, then those who lose BMD during
the first year should regain it the next year of treatment. We tested this
hypothesis by analyzing data from the first 2 years of the Fracture Intervention
Trial (FIT), a large randomized trial of alendronate. We then extended these
analyses to women receiving raloxifene hydrochloride in the Multiple Outcomes
of Raloxifene Evaluation (MORE) Trial.4
The FIT trial included postmenopausal women who had femoral neck BMD
of 0.68 g/cm2 or less, randomly assigned to receive 5 mg/d of alendronate
sodium or an identical looking-placebo for the first 2 years.2,3
Participants with a daily calcium intake less than 1000 mg were given a daily
supplement containing 500 mg of calcium (as the carbonate salt) and 250 IU
of cholecalciferol. Women who had used fluoride or bisphosphonates at any
time in the past or had used estrogen or calcitonin during the previous 6
months were excluded. Because investigators and participants were blinded
to study treatment and to the results of follow-up bone densitometry, there
were no changes in treatment, and no changes in treatment were made based
on the results of bone densitometry.
Bone mineral density of the lumbar spine and hip was measured at baseline,
and annually thereafter. This article refers to total hip BMD as "hip BMD."
All densitometry operators received training and were certified by the University
of California, San Francisco Coordinating Center and a random sample of all
BMD scans was reviewed centrally. To calibrate, phantoms with known BMD were
scanned daily, and some phantoms were circulated between centers. The mean
coefficient of variation for the longitudinal phantom measurements for all
centers for more than 5 years was 0.49% (range, 0.4%-0.7%).
Of the 3236 women randomly assigned to the alendronate group, 2970 (92%)
completed measurements of hip and posteroanterior spine BMD at baseline, 12,
and 24 months. Defining adherence as taking at least 70% of pills by pill
count and 75% by diary, 2634 (89%) of the 2970 women were adherent.
The MORE Trial included 7705 postmenopausal women aged 80 years or younger
whose BMD at either the femoral neck or lumbar spine was at least 2.5 SDs
below the mean for normal young women or who had at least 1 moderate or 2
mild vertebral fractures detected by spine radiographs.4
The women were randomly assigned to receive 60 or 120 mg/d of raloxifene hydrochloride
or an identical-looking placebo for 3 years. All participants were also given
a daily supplement of 500 mg of calcium and 400 to 600 IU of cholecalciferol.
Women who had secondary causes of osteoporosis or who had taken a bisphosphonate
within the previous 6 months or oral estrogen within the previous 2 months
were excluded. Investigators and participants were blinded to study treatment.
Femoral neck BMD was measured annually. A central reading facility provided
correction factors to adjust for intersite differences and changes in the
performance of the densitometers over time. The mean coefficient of variation
for BMD measurements at all centers on a spine phantom, for more than 4 years,
was 0.5% (range, 0.3%-1.6%).
Bone densitometry of the hip was performed on a subset of women participating
in the MORE trial. Of the 5129 women randomly assigned to the raloxifene group,
4099 had measurements of femoral neck BMD at baseline, 12, and 24 months after
entry. Defining adherence as taking at least 70% of pills by pill count, 3954
(97%) of the 4099 women were adherent and were included in this analysis.
During the first year of the study, women who adhered to treatment gained
an average of 2.2% in total hip BMD and 4.5% in spine BMD. They gained an
additional 0.9% in the hip and 1.7% in the spine during the next year of treatment.
During the first year, 82% of the women taking alendronate had an apparent
increase in hip BMD; 1.4% appeared to have more than a 4% decline in total
Those who seemed to lose BMD during the first year of alendronate usually
gained BMD during the second year (Table
1). On average, those who lost the most during the first year were
most likely to gain hip BMD and gained more BMD than other groups during a
second year of treatment (Figure 1).
For example, those who appeared to lose more than 4% during the first year
had a 92% chance of gaining BMD and, on average, gained 4.8% during the next
year of continued treatment. In contrast, those who seemed to gain more than
8% in the first year had only a 36% chance of gaining BMD in the second year
and, on average, lost 1% during the second year of treatment with alendronate.
Similar findings were observed for changes in BMD at the spine: those who
lost more during the first year were more likely to gain spine BMD during
the second year.
A similar phenomenon was seen in women in the placebo group. Among the
2603 women who had hip BMD and adhered to placebo, the 37 who lost more than
4% of their hip BMD in the first year had an average increase in the second
year of 4.8% (95% confidence interval [CI], 2.7%-6.9%), while among the 61
who gained more than 8%, experienced an average loss of 1.0% (95% CI, −0.1
to −1.9) during the second year. Among the 484 women who lost hip BMD
during the first year, those who also lost spine BMD (n = 80) had a 75% chance
of gaining hip BMD with an average increase of 1.7% (95% CI, 1.0%-2.3%); those
who also had increased spine BMD (n = 404) during the first year had an 80%
chance of gaining BMD, with an average increase of 2.0% (95% CI, 1.7%-2.3%).
Women who lost BMD in the femoral neck during the first year of taking
raloxifene usually gained BMD during the second year of treatment (Table 2). For example, the women who appeared
to lose more than 4% during the first year had a 79% chance of gaining BMD
and, on average, gained 4.0% during the next year of continued treatment.
In contrast, those who seemed to gain more than 8% in the first year had a
22% chance of gaining BMD in the second year and, on average, lost 2.8% during
the second year of treatment with raloxifene.
The minority of patients who seem to lose BMD during the first year
of treatment with alendronate or raloxifene are those most likely to gain
BMD if treatment is continued. This counterintuitive result demonstrates the
principle of regression to the mean: individuals who have measurements that
differ from the mean for a population tend to have repeat measurements that
are closer to the mean.1,5 This
tendency is greatest for measurements that are farther from the mean. This
explains why those who lost the most BMD the first year also gained the most
the next year of treatment.
Bone densitometry, with coefficients of variation in the range of 1%
to 2%, is one of the most precise measurements in clinical medicine.6,7 Other measurements used to assess response
to medical therapy, such as cholesterol, blood pressure, and spirometry, have
larger coefficients of variation.7 Thus, it
is likely that the phenomenon with bone densitometry also holds true for measurements
used to monitor lipid-lowering, antihypertensive, bronchodilator, and others.
Our results have several practical implications for the use of bone
densitometry to monitor therapy. Treatment should be continued in patients
who appear to lose BMD initially, because most patients will gain it with
continued treatment. These gains will usually exceed the losses of the first
year, so, on the whole, the patient will gain BMD. Conversely, when patients
make unusually large gains the first year, they will likely lose, or have
modest gains in, BMD during the second year. Physicians and patients should
not view this as development of resistance to treatment, because this may
simply be a natural correction of random error in the earlier estimation of
change in bone density.
Physicians sometimes monitor patients at 2 sites at the hip and spine.
Our results also indicate that adding a measurement of change at a second
skeletal site added little to the ability to predict change in hip BMD during
the second year of treatment.
Some patients may lose BMD during treatment for other reasons, including
nonadherence or, in the case of alendronate, improperly taking it with meals
or medications that interfere with its absorption. Others may have responded
to treatment by losing less BMD than they would have without treatment. Some
patients may be resistant to the effects of alendronate or raloxifene, although
no such cases have been documented and the chance of such biological nonresponse
is not known. If clinicians monitor antiresorptive drug treatment with BMD,
then loss of BMD during treatment should prompt physicians to ask about adherence
to treatment and, for women taking alendronate, to make sure that they are
taking it properly. It may not be possible to discern the reason for unexpected
losses in BMD during treatment, but the patient can be reassured that continued
treatment is likely to produce better results during the next year.
On average, women who gained the most during the first year of treatment
had the greatest cumulative increases during the first 2 years of treatment.
This might mean that changes in the first year are predictive of greater long-term
changes in BMD. However, if the first measurement was in error, the cumulative
change during 2 years or longer will carry that error forward into long-term
estimates of change.
Sometimes patients who are not receiving drug treatment for osteoporosis
are also monitored to determine if they are losing BMD at an unusually rapid
rate. We found that women who lost the most BMD while taking placebo were
also the most likely to gain during the next year of monitoring. Thus, physicians
should be aware that large changes in BMD during observation of untreated
patients might also be due to measurement error. If a new treatment is begun
after observing a large loss in BMD, it is likely to increase in the next
year, but much of this increase may be due to regression to the mean.
Our findings raise questions about the value of monitoring BMD during
treatment. It might be valuable to prompt inquiries about adherence and proper
use of the medication among those who lose BMD, but such inquiries could be
included in routine follow-up visits regardless of BMD testing. Monitoring
may reinforce adherence to therapy; however, this may have relatively limited
value because most problems with adherence occur within the first 3 months
of starting treatment, long before an annual measurement of BMD. It has been
suggested that measuring markers of bone turnover might be a better and earlier
way to assess response to therapy,8 but available
markers have relatively poor reproducibility, with coefficients of variation
in the range of 10% to 30%.8,9
The value of using markets to monitor therapy in individual patients is uncertain.9
This study has limitations. This analysis was not designed to determine
whether change in BMD is related to degree of reduction in risk of fractures
during treatment. We only studied women taking 5 mg/d of alendronate sodium;
since 10 mg of alendronate produces a somewhat greater increase in bone density,
these results might differ a little for women taking a higher dose. This study
also has several strengths. In both trials, patients and staff were blinded
to treatment assignment and in the FIT trial they were also blinded to all
measurements of bone density, so the changes we observed cannot be attributed
to changes in treatment. Our analysis demonstrated that this phenomenon occurs
with 2 different drugs, treatment with placebo and 2 measurement at 2 skeletal
sites. Thus, it is highly likely that regression to the mean influences monitoring
of other antiresorptive drugs with all types of densitometry. Measurements
of BMD had excellent reproducibility; regression to the mean will probably
be magnified if measurements are done with poorer reproducibility.
We conclude that women who lose BMD during treatment with alendronate,
raloxifene, and probably other antiresorptive drugs, will usually have increases
in BMD if treatment is continued. This illustrates the principle of regression
to the mean and this phenomenon may influence other measurements that are
used to monitor many types of treatments used in clinical practice.