Customize your JAMA Network experience by selecting one or more topics from the list below.
Jarvik JG, Hollingworth W, Martin B, et al. Rapid Magnetic Resonance Imaging vs Radiographs for Patients With Low Back Pain: A Randomized Controlled Trial. JAMA. 2003;289(21):2810–2818. doi:10.1001/jama.289.21.2810
Context Faster magnetic resonance imaging (MRI) scanning has made MRI a potential
cost-effective replacement for radiographs for patients with low back pain.
However, whether rapid MRI scanning results in better patient outcomes than
radiographic evaluation or a cost-effective alternative is unknown.
Objective To determine the clinical and economic consequences of replacing spine
radiographs with rapid MRI for primary care patients.
Design, Setting, and Patients Randomized controlled trial of 380 patients aged 18 years or older whose
primary physicians had ordered that their low back pain be evaluated by radiographs.
The patients were recruited between November 1998 and June 2000 from 1 of
4 imaging centers in the Seattle, Wash, area: a university-based teaching
program, a nonuniversity-based teaching program, and 2 private clinics.
Intervention Patients were randomly assigned to receive lumbar spine evaluation by
rapid MRI or by radiograph.
Main Outcome Measures Back-related disability measured by the modified Roland questionnaire.
Secondary outcomes included Medical Outcomes Study 36-Item Short Form Health
Survey (SF-36), pain, preference scores, satisfaction, and costs.
Results At 12 months, primary outcomes of functional disability were obtained
from 337 (89%) of the 380 patients enrolled. The mean back-related disability
modified Roland score for the 170 patients assigned to the radiograph evaluation
group was 8.75 vs 9.34 for the 167 patients assigned the rapid MRI evaluation
group (mean difference, −0.59; 95% CI, −1.69 to 0.87). The mean
differences in the secondary outcomes were not statistically significant :
pain bothersomeness (0.07; 95% CI −0.88 to 1.22), pain frequency (0.12;
95% CI, −0.69 to 1.37), and SF-36 subscales of bodily pain (1.25; 95%
CI, −4.46 to 4.96), and physical functioning (2.73, 95% CI −4.09
to 6.22). Ten patients in the rapid MRI group vs 4 in the radiograph group
had lumbar spine operations (risk difference, 0.34; 95% CI, −0.06 to
0.73). The rapid MRI strategy had a mean cost of $2380 vs $2059 for the radiograph
strategy (mean difference, $321; 95% CI, −1100 to 458).
Conclusions Rapid MRIs and radiographs resulted in nearly identical outcomes for
primary care patients with low back pain. Although physicians and patients
preferred the rapid MRI, substituting rapid MRI for radiographic evaluations
in the primary care setting may offer little additional benefit to patients,
and it may increase the costs of care because of the increased number of spine
operations that patients are likely to undergo.
Advances in magnetic resonance imaging (MRI) have led to faster and
therefore less expensive examinations. Several groups, including ours, have
reported the development of a rapid MRI examination for the lumbar spine.1-6 Using
rapid MRI early in the care of patients with low back pain might benefit patients
by providing a swifter definitive diagnosis, obviating further imaging or
referral, and reassuring both patient and physician that there is no serious
disease. However, early imaging with rapid MRI risks discovering incidental
anatomic findings. In studies of subjects without low back pain, disk herniations
are seen in approximately one third,7-10 disk
bulges in half to two thirds,7,8,10,11 and
disk degeneration in up to 90% of the scars.7,10,11 Identifying
incidental abnormalities with early MRI might lead to unnecessary interventions
that otherwise would not have been performed, potentially resulting in both
worse patient outcomes and higher costs.
Because low back pain is extremely common, any change in the diagnostic
and treatment approach may have a large impact on health care resources. Spine
imaging may detect a wide range of abnormalities, for which a wide range of
treatments exists, some of which have only modest evidence of efficacy. Furthermore,
the frequent occurrence of incidental findings often makes the true source
of symptoms ambiguous. We therefore thought that a randomized trial evaluating
treatment decisions, costs, and patient outcomes was the best approach to
compare rapid MRI with radiographs as an initial diagnostic imaging test.
We recruited patients with low back pain from 4 imaging centers in western
Washington State: a university outpatient clinic; a private, nonprofit teaching
hospital; a private, for-profit, multispecialty clinic with on-site radiology;
and a private, for-profit, free-standing imaging center. We identified potential
patients when their physicians ordered radiographs of the lumbar spine. We
targeted general internal medicine and family practice physicians, but we
also enrolled patients visiting medical and surgical subspecialty physicians
for an initial presentation of back pain. Eligible patients had low back pain
with or without radiating leg pain, no lumbar surgery for 1 year prior to
enrollment, no history of acute external trauma, no metallic implants in the
lumbar spine (eg, Harrington rods or pedicle screws), and no contraindications
for MRI. We also required that patients have a telephone, be at least age
18 years, not be pregnant, and speak English.
Of 1250 patients seen with low back pain at the recruiting clinics between
November 1998 and June 2000, 547 did not participate because either the research
coordinator was not available to enroll the patient or the primary physician
decided not to refer the patient to the study (Figure 1). Of the 703 remaining patients, 154 did not meet inclusion
criteria and 169 refused to participate, leaving 380 who enrolled.
The study was approved by the institutional review boards and radiation
safety committees of the participating institutions. All patients gave written
After enrollment, patients completed questionnaires focused on pain
and functional status and were then randomly assigned to undergo radiograph
or a rapid MRI of the lumbar spine. The random allocation scheme used a computer-generated
block design with block sizes varying between 4 and 8 and stratified by site.
Research assistants were not involved in generating the sequence and were
unaware of the block randomization scheme. Group assignments were placed in
opaque sealed envelopes that were opened by the research assistants after
completing baseline questionnaires.
The number of radiographic views was left up to the ordering clinician
although we excluded patients if the physician had ordered flexion or extension
views or special views of the sacroiliac joints. There were 161 patients who
had anteroposterior and lateral views only, 9 who had additional views (such
as oblique), and 11 for whom the number of views was not available. Although
we attempted to schedule patients randomized to rapid MRI on the day of enrollment,
when that could not occur, we scheduled them within a week. Most MRI scans
were conducted on systems with a field strength of 1.5 T (n = 136), with the
rest scanned at either 0.3 or 0.35 T (n = 46), reflecting real-world variation
in imaging equipment. The pulse sequences used for the rapid MRI were sagittal
and axial T2-weighted fast spin echo images whose total acquisition time was
approximately 2 minutes on a 1.5-T scanner (Figure 2).
Radiologists interpreted images as part of their normal workflow. Preliminary
reports were faxed or personally delivered to the referring clinician.
The primary outcome measure was the modified Roland back pain disability
scale,12,13 conducted 12 months
after randomization, testing the hypothesis that rapid MRI would result in
better 12-month scores than radiograph. The modified Roland Scale is a 23-item
functional status scale that focuses on how back pain affects common daily
activities, with higher scores indicating worse function. Patrick et al13 showed high internal consistency, validity, and responsiveness
of this scale in a sciatica cohort, with an effect size of −1.6 (the
score change for improved patients divided by the SD at baseline).
Secondary outcome measures included back pain frequency and bothersomeness;
the Medical Outcomes Study 36-Item Short Form Health Survey (SF-36), version
1; days of reduced or lost work; patient satisfaction with care; and reassurance
and preference scores. The pain frequency and bothersomeness indices are symptom
scales that measure the frequency and bothersomeness of pain on a 24-point
scale.13,14 Patients rate 4 separate
symptoms on a 1-to-6 scale: (1) leg pain; (2) numbness or tingling in the
leg, foot, or groin; (3) weakness in leg or foot; and (4) back or leg pain
while sitting. Scores are summed to create the symptom score, with higher
scores indicating worse symptoms. The SF-36 is a highly reliable and valid
generic measure of health-related quality of life.15-18 We
measured patient satisfaction using a modified version of the 11-item Deyo-Diehl
patient satisfaction questionnaire.19 We deleted
a question that asked if patients thought they should have an imaging study
and added 2 questions about reassurance that patients attributed to the imaging
Patient-elicited preference scores were measured with interactive software,
U-titer II.20 Preference scores aim to combine
all aspects of quality of life in 1 summary score anchored at 0 (death) and
1 (perfect health). As an introduction to the preference assessment, we asked
patients to use the time trade-off technique21 to
value the theoretical health states of monocular and binocular (complete)
blindness. We then asked patients to value their current health state using
the time trade-off technique and rating scale methods. We excluded from subsequent
preference score analysis patients who rated binocular blindness as a better
health state than monocular blindness. All measures except for the time trade-off
technique had been previously validated for patients with low back pain.13
We contacted patients by telephone or mail 1, 3, 6, and 9 months after
randomization. At 12 months, we asked patients to return to the clinic to
undergo a brief physical examination, repeat the preference scores assessment,
and answer the outcome questionnaires. We collected the primary outcome measure,
the modified Roland Scale, at months 3, 6, and 12. We measured patient satisfaction
at months 1, 3, and 12. We obtained rating scale and time trade-off technique
measures only at baseline and month 12. We administered all other outcome
questionnaires at baseline and months 3 and 12.
Research assistants collecting outcome data were blinded to patient
allocation but were asked at the end of each interview to which group they
thought the patient had been assigned on a 5-point scale ("definitely x-ray"
to "definitely MRI"). We considered them unblinded if they reported that they
were "definitely" or "probably" aware of the patient's true randomized assignment.
We tracked societal resource use with patient diaries and medical and
billing record reviews. Record abstraction documented the name, dose, route,
frequency, and duration of each drug prescription. For office visits, tests
and procedures, we recorded current procedural terminology codes.22 From inpatient records, we identified the relevant
diagnosis related group and International Classification
of Diseases, Ninth Revision, Clinical Modification procedure codes
for surgery, radiology, anesthesiology, and pathology. To track surgeries,
we queried the Washington State Comprehensive Hospital Abstract Reporting
We supplemented record abstraction with a resource utilization questionnaire
that patients completed every 3 months. This included (1) office visits to
conventional or alternative health care practitioners (2) drug use, including
over-the-counter medications; (3) hospitalizations; (4) patient time spent
obtaining health care for back pain; and (5) transportation, home care, and
other expenses incurred while seeking health care. To avoid double counting,
we examined each questionnaire entry and excluded those already in the medical
record abstraction. We imputed missing values for drug dose and duration by
the relevant modal and mean values, respectively. We excluded resources unrelated
to back pain.
We used the federal upper limit and average wholesale price to estimate
the unit cost of generic and branded drugs, respectively.23 The
cost of over-the-counter drugs was based on prices from a nationwide pharmacy.24 We used Medicare local fee schedules25-27 to
estimate the cost of ambulatory care. Services not covered by Medicare were
valued using reimbursements from a Washington state health insurance plan.28 We used the Medicare prospective payment system29 to value inpatient facility costs, supplemented with
reimbursements for surgeon, radiologist, anesthesiologist, and pathologist.
We conducted a microcosting exercise based on a time-motion study to
estimate the cost of rapid MRI and radiographs.30 In
brief, the time-motion study tracked radiologist, technologist, and patient
imaging room time, for a series of patients receiving rapid MRI or radiograph
for low back pain. We multiplied staff time by relevant compensation rates.
Equipment time was valued using the amortized capital cost of equipment, room
construction, and maintenance plus consumable and overhead costs. All costs
were in year 2001 values.
We classified patients as either (1) in the labor force (eg, full-time
or part-time worker), (2) in the nonlabor force (eg, homemaker, retired, or
receiving disability compensation), or (3) in the nonproductive labor force
(eg, unemployed). We multiplied the number of hours spent seeking treatment
for back pain by the age, sex, and labor-type specific mean earnings per hour
in the United States in 2001.31,32
We compared baseline demographic characteristics, clinical findings,
symptoms, and functional status of patients allocated to the 2 study groups
by calculating means and SDs or frequencies and proportions. For the primary
outcome variable, the 12-month modified Roland score, we used analysis of
covariance to compare the diagnostic groups controlling for baseline Roland
score and recruitment site. Using an intention-to-treat strategy, we used
the Mann-Whitney U test for ordinal variables and
the χ2 or Fisher exact test for dichotomous variables. We did
not adjust for multiple comparisons over time or between different outcome
measures. Analyses were performed with SPSS33 or
SAS34 software. Statistical tests were 2-sided
with P<.05 being considered statistically significant.
A 2-point difference on the modified Roland Scale is likely the smallest
clinically important difference.13 In order
to detect a clinically important difference with 80% power, and a 2-tailed α
level of .05, we calculated a final target sample size of 314, based on pilot
data (pooled variance, 40) for the modified Roland scale.5 Anticipating
a 15% drop-out rate, our recruitment goal was 372.
Because of concern that 11 patients referred by orthopedic surgeons
were systematically different from the rest of the cohort, we performed the
main analyses with and without these patients. There was no important difference
between the results, so we report only results for the entire cohort.
We performed an incremental cost-effectiveness analysis from the societal
perspective. We used bootstrap analyses35 with
5000 iterations to derive a 95% confidence interval (CI) for total costs and
We enrolled 380 patients. The characteristics of the radiograph and
rapid MRI groups were similar at baseline (Table 1), although the rapid MRI group was slightly more likely
to have comorbidities and to be receiving or applying for disability compensation.
The most common comorbid conditions were osteoarthritis (n = 112), hypertension
(n = 106), and depression (n = 90).
There were 187 patients referred by general internal medicine or family
practice physicians, 131 by rheumatologists, 38 by physical medicine and rehabilitation
specialists, 11 by orthopedic specialists, and 13 by other provider types.
There were 109 primary care physicians, with a mean of 3.5 patients referred
Six patients randomized to undergo rapid MRI instead received radiographs
(4 due to claustrophobia and 2 because of scheduling difficulties). Two additional
patients randomized to the rapid MRI group did not undergo MRI scan (1 due
to a coordinating error, and 1 who failed to return for the diagnostic scan).
All patients randomized to the radiograph group underwent radiographs (Figure 1).
Imaging results for the 2 groups are summarized in Table 2. In the rapid MRI group, disk findings and facet degeneration
were common: 35 patients (20%) had moderate or severe central spinal stenosis
and 31 (17%) had lateral recess stenosis. Only 13 (7%) of patients had nerve
root impingement and 9 (5%) had compression fractures. In the radiograph group,
disk-space height loss and facet degeneration were also common, and 24 (13%)
had compression fractures. The scans or x-ray films showed no evidence of
infection, tumor, or inflammatory spondylitis among any of the patients.
The coordinators collecting follow-up data were successfully blinded
to 82% of the patients at 12 months. At 12 months, 337 (89%) of 380 patients
had completed the modified Roland scale, but the rate at the sites ranged
from 73% to 96%. Among those assigned to the rapid MRI group, 167 (88%) of
190 (95% CI, 0.84-0.93) and 170 (89%) of 190 assigned to the radiograph group
(95% CI, 0.85-0.94) completed the study (P = 0.73).
Those lost to follow-up had baseline Roland Scale scores of 16.0 vs 12.9 for
those who completed the study (12.9; 95% CI of difference, 1.13-4.96; P = .002). But baseline scores were quite similar between
groups among those who dropped out (16.4 rapid MRI vs 15.6 radiograph; 95%
CI of the difference, −4.27 to 2.70; P = .65).
Those who dropped out had more comorbid conditions than those who did not
(2.4 vs 1.6; 95% CI of the difference, 0.34 to 1.31; P =
.002). There were no significant differences between those lost to follow-up
and those contacted with respect to age; sex; and smoking, depression, body
mass index, or disability compensation status. There were no important adverse
events in either group. Because of an administrative error, only two thirds
of patients completed the Roland Scale that was administered at 3 months.
After adjusting for baseline modified Roland score and study site, the
12-month Roland Scale score in the radiograph group was 8.75 vs 9.34 in rapid
MRI group, which did not represent a clinically or statistically significant
difference (mean difference −0.59; 95% CI −1.69 to 0.87; P = .53). Although modified Roland scores were more than
2 points lower at 2 sites compared with the other 2 sites, this difference
was not statistically significant when adjusted for baseline scores. The range
of observed Roland scores, SF-36 subscale scores, pain indices scores, and
disability days were also similar between groups.
At 3 months, after adjusting for baseline scores and site, the mean
modified Roland score for the rapid MRI group was 10.4 vs 8.6 in the radiograph
group (95% CI of the difference, −3.47 to −0.19; P = .03), but this difference narrowed and was no longer significant
at 12 months (Table 3).
Both groups had significant and clinically important improvement over
time on many measures, with most of this improvement occurring within the
first 3 months. However, there were no significant differences between groups
in SF-36 scores, pain scores, or days lost from work.
Although there was no significant difference in overall patient satisfaction
scores between groups, satisfaction was associated with the degree of reassurance
patients received from the MRI scans. At months 1, 3, and 12, the overall
satisfaction score had a positive correlation with the degree of self-reported
reassurance that patients had received from the MRI scan results (Pearson
correlation coefficients ranging from 0.55-0.59, P<.001
for all 3 times). Patients consistently rated reassurance from the MRI scan
results higher than for the radiographic results, but only by a half point
on a 5-point scale. When asked simply if they were reassured by the results
of the diagnostic test, 58% of patients in the radiograph group responded
affirmatively at 12 months compared with 74% of those in the rapid MRI group
(P = .002). Reassurance due to the diagnostic test
did not decrease over time.
The time trade-off technique preference scores were significantly higher
in the rapid MRI group at 12 months after adjusting for baseline scores (mean
difference −0.03; 95% CI of the difference, −0.12 to −1.02; P = .005). However this improvement correlated only weakly
with other quality-of-life measures and patient reassurance.
Patients initially randomized to undergo radiograph were more likely
to be referred for a conventional MRI scan in the year following randomization
(Table 4). Conversely, patients
randomized to undergo rapid MRI had more subsequent radiographs of the lumbar
spine or pelvis although the economic impact was small due to their low cost.
There were 131 specialist consultations after randomization to rapid MRI compared
with 90 radiographs. In contrast, patients in the radiograph group had approximately
twice the total number of physical therapy, acupuncture, massage, or osteopathic
and chiropractic manipulation appointments as patients randomized to undergo
rapid MRI. This finding was consistent for treatments likely to be initiated
by the physician (eg, physical therapy) and those more likely to be patient
initiated (eg, acupuncture).
Ten patients (6%) randomized to undergo rapid MRI had lumbar spine surgery
within 1 year compared with 4 patients (2%) in the radiograph group (risk
difference, 0.34; P = .09). The mean time from enrollment
to surgery was greater, but not significantly, in the rapid MRI group (138
vs 97 days, P = .31). Patients who underwent surgery
had similar clinical and demographic characteristics compared with nonsurgically
treated patients at baseline. The respective mean scores for the 14 patients
who had surgery vs the remaining patients who did not were not significantly
different: 14.4 vs 13.2, Roland score (P = .46);
13.2 vs 11.9, pain frequency (P = .33); and 11.9
vs 12.3, pain bothersomeness (P = .74). Pain radiating
into 1 leg was more common among surgical patients (69%) than the others (46%, P = .08).
The MRI scan results most predictive of future surgery were those that
detected either a disk herniation or central stenosis, observed in 8 (80%)
of 10 patients who had surgery vs 69 (41%) of 168 patients who did not have
surgery Figure 2. However, 69 (90%)
of 77 patients with herniations or central stenosis did not have surgery and
12 (92%) of 13 patients with nerve root impingement did not have surgery.
Because of the higher surgical rate, the mean cost of surgery, averaged across
all patients, was higher for patients randomized to undergo rapid MRI ($417)
than patients randomized to undergo radiograph ($156, P = .09).
The mean cost of health care services was higher among patients randomized
to undergo rapid MRI than radiograph ($2121 vs $1651, respectively), primarily
due to more inpatient admissions (Table
4). This difference was not statistically significant (mean difference
−$470; 95% CI, −$1044 to $105; P = .11).
Patients in the radiograph arm reported more hours spent attending health
care appointments (mean difference 5.3 hours), reflecting their greater use
of physical and other manual therapies. Similarly, patients randomized to
radiograph reported higher out-of-pocket expenses than those randomized to
MRI. In total, the societal cost in the year following rapid MRI was 16% higher
($2380) than lumbar radiograph ($2059) but not statistically significantly
(−$321; 95% CI, −$1100 to $458; P = .42).
Figure 3 shows the likely
range for the cost-effectiveness of rapid MRI. The mean estimate suggests
that rapid MRI is more costly and associated with clinically equivalent outcomes.
However, there is wide variation around this mean estimate as represented
by the 5000 bootstrap replicates in the scatter plot. The CI around the incremental
cost-effectiveness ratio is undefined due to the non-negligible probability
that cost-effectiveness for the entire population lies in any 1 of the quadrants
in Figure 3. Thus, the economic
effect of replacing radiographs with rapid MRI in this setting remains ambiguous.
Replacing lumbar spine radiographs with a rapid MRI scan in primary
care patients resulted in no long-term difference in disability, pain, or
general health status. There was a preference among both patients and physicians
for the rapid MRI scan, but there may be a higher surgical rate among patients
undergoing MRI scan, and overall societal costs may be higher as well. A major
impetus for this work was the concern that substituting radiographs with rapid
MRI scans would result in worse patient outcomes because incidental abnormalities
would foster increased interventions and unnecessary morbidity. Our study
suggests that substituting rapid MRI scan for radiographs is likely safe but
may in fact result in more specialist consultations and operations. Despite
the higher rate of surgery, average outcomes were not better among those in
the rapid MRI group.
The use of MRI scan instead of radiographs as the initial imaging for
low back pain has become more common. McNally et al6 substituted
a limited MRI scan for radiographs for 1042 patients with at least 6 weeks
of low back pain.6 They concluded that MRI
scans detect a greater number of abnormalities, having discovered neoplasms
in 17 (1.6%), more than double the percentage observed in a primary care population.36 In fact, their cohort was mostly not a primary care
population, with only 40% having been referred by general practitioners. Nevertheless,
they have incorporated limited use of MRI into their routine practice, using
it instead of radiographs for patients without radiculopathy and more than
6 weeks of low back pain.
The equivalency of patient outcomes in our study provides some reassurance
that the policy advocated by McNally et al is not harmful to patients. However,
it is worrisome that in our study more patients in the rapid MRI group had
back operations than patients in the radiograph group. In general, the rapid
MRI group had more specialist physician consultations while the radiograph
group had more manual and physical therapy consultations. The difference in
health care service costs was not statistically significant between the rapid
MRI and radiograph groups. However, the observed difference in mean health
services cost ($470) suggests that routine use of rapid MRI has the potential
to increase costs for patients with low back pain without a measurable benefit
in pain or functional status. This observed increase in surgical costs may
be off-set, to some extent, by lower costs for services such as physical therapy
and chiropractor visits, as well as lower future diagnostic imaging costs.
Our results are concordant with trials examining similar issues. Kendrick
et al37,38 reported the results
of a randomized controlled trial of radiograph vs no-imaging for primary care
patients with low back pain.37,38 There
were no differences in outcomes after 9 months except for higher patient satisfaction
in the radiograph group. However costs were also higher in that group. These
results parallel our study despite the use of different technologies. The
reassurance value provided by a diagnostic test is a well-recognized phenomenon
that may have an effect on patient outcomes.39 However,
in low back pain, our data suggest that improvement in measures of reassurance
and satisfaction do not result in measurable improvements in functional status
or health-related quality of life.
Recruiting patients from both academic and private practice settings
improves the generalizability of our results. A potential spectrum bias does
exist because patients with more bothersome back pain might have been motivated
to enroll in order to obtain a free MRI scan. However the pain and disability
scores of our cohort are comparable with other back pain studies.13
Given the current evidence, it is difficult to make strong recommendations
regarding the use of rapid MRI for patients with low back pain. On the one
hand, it does not appear that rapid MRI causes harm or greatly increases costs,
and it provides more reassuring information for both patients and physicians.
On the other hand, patient symptoms and functional outcomes are not, on average,
improved by using MRI scan as the first imaging test. Rapid MRI has the potential
to increase the number of back operations without an apparent benefit to patients
and perhaps to increase costs. If the use of rapid MRI scans disseminates
widely and surgical complications are more common than we observed, the consequences
could be detrimental. In this setting, a cautious approach is probably most
prudent, and we recommend that rapid MRI not become the first imaging test
for primary care patients with back pain until its consequences for surgical
rates and costs are better defined.
Create a personal account or sign in to: