Patterns of Ordering Diagnostic Tests for Patients with Acute Low Back Pain

Methods

Our study was done in North Carolina, the population of which is almost equally divided between urban and rural areas. Twenty-two percent of the population is black [11]. Almost 600 chiropractors practice in North Carolina, and previous research by our group [1] showed that 39% of persons who seek care for acute back pain first see a chiropractor.

Practitioners

We examined community practitioners who commonly treat patients with acute low back pain. Patients were randomly selected from medical and chiropractic licensure files and were recruited until enough practitioners were available in each of six strata: urban primary care physicians, rural primary care physicians, urban chiropractors, rural chiropractors, orthopedic and neurologic surgeons, and physicians and a few family nurse practitioners at a group-model health maintenance organization. We defined “primary care physician” as one who was in family medicine, general internal medicine, or general practice. Few osteopathic physicians practice in North Carolina. Neurosurgeons were sampled, but none had seen a substantial number of patients with acute low back pain. Because few orthopedic surgeons practice in rural areas, these surgeons were sampled statewide.

Practitioners were eligible for the study if they 1) practiced in an ambulatory care setting more than half time and 2) saw patients with acute low back pain and no previous referral as part of their practice. Practitioners were aware of the overall purpose of the study but not of the specific outcome or utilization variables. Practitioners received $40 per recruited patient as compensation for time spent screening patients, obtaining consent, and completing baseline clinical evaluations. Of the eligible practitioners invited, an average of 74% agreed to participate (range, 65% of primary care providers to 87% of practitioners in the health maintenance organization and orthopedic strata). The number of practitioners in each stratum was as follows: 39 urban primary care physicians, 48 rural primary care physicians, 32 urban chiropractors, 32 rural chiropractors, 29 orthopedic surgeons, and 28 general and mid-level practitioners at the health maintenance organization. In 1993, at the conclusion of the study, practitioners were given a survey that examined practitioner and practice demographic characteristics, perceived treatment patterns, and treatment of hypothetical patients by clinical vignette. The vignettes had previously been used to assess interspecialty diagnostic and therapeutic aggressiveness in a nationwide survey [12]. The study design allowed us to examine practitioner response to clinical vignettes and the relation of this response to clinical behavior with similar actual patients seen in the practitioners' offices. The vignettes asked practitioners to indicate whether they would order CT or MRI for three patients: one with sciatica and a diminished ankle reflex, one with chronic back pain, and one with acute back pain and unremarkable results on a physical examination. The following are the three vignettes:

1. Acute sciatica: A 35-year-old auto mechanic presents with a 4-day history of severe acute low back pain with radiation to the posterior calf and lateral foot. He has some sensory deficits in this distribution and a diminished ankle reflex but has no motor weakness. Straight-leg raising is limited to 45 degrees in the affected leg. Plain x-ray of the lumbar spine is normal except for postural changes suggesting muscle guarding.

2. Chronic back pain: A 50-year-old homemaker presents with a 3-year history of intermittent, excruciating low back pain. She has seen other doctors and chiropractors during this period but was disappointed with the results of care. She currently has severe back pain but there is no radiation to the legs, and physical examination reveals a limitation of lumbar spine flexion. Plain lumbar spine films show osteophytes at several vertebral levels but no disc space narrowing.

3. Acute back pain: A 28-year-old woman who runs her own catering service complains of having acute severe low back pain for a week. The pain is not radiating but is so severe she has been unable to work for the past 5 days. She is anxious to return to her usual activities but feels immobilized by the pain at present. Physical examination reveals markedly limited anterior flexion and left paraspinous tenderness and a normal neurologic examination. Lumbar spine films are normal.

Two hundred eight practitioners agreed to participate in our study, 188 (90%) recruited at least one patient, and 162 returned completed surveys. Clinicians who contributed no patients reported that they saw few patients with acute back pain in their practices.

Patients

Practitioners invited consecutive patients with acute low back pain to participate in the cohort study. Patients were eligible if their current episode of back pain had lasted for less than 10 weeks and if they had not previously received care for the current episode of pain, had never had back surgery, had no history of metastatic cancer, were not pregnant, owned a telephone, and spoke English. Patients were the unit of analysis when the outcome was patient-oriented; the practitioner was the unit of analysis when the analysis was based on the practitioner survey.

Practitioners obtained patient consent and gathered minimal information on the history and results of physical examination at the initial office visit. Twice weekly, we obtained the names and telephone numbers of the patients with back pain. Personnel from the University of North Carolina Survey Research Unit conducted all interviews. No attempt was made to influence the practitioners' diagnostic tests or treatments; our study was observational. Staff of the practices kept lists of the recruitment of patients into the cohort, thereby allowing assessment of approximate recruitment rates. Fifty percent of patients with back pain seen in the offices were eligible for the study; only 8% of those who were eligible refused enrollment. The main reasons for ineligibility were chronic pain and previous treatment of the current episode of pain. Patients received $20 for their time spent responding to the many surveys. They were told that the study was to examine “how long back pain usually lasts and the types of treatments used.” The Survey Research Unit contacted patients by telephone shortly after the index visit. The median time from the index office visit to the baseline telephone interview was 6 days (range, 0 to 62 days).

Data Analysis

Information on demographic characteristics, health care utilization, and functional status was collected at the time of the baseline interview and 2, 4, 8, 12, and 24 weeks after the baseline interview or until the patients declared themselves “completely better.” Functional status was examined by assessing time off from work and by using the Roland-Morris 23-item back disability scale, a validated subset of the Sickness Impact Profile [13, 14]. All patients were interviewed at 24 weeks. In telephone surveys, patients were asked about all care from all providers. Chart abstraction was used to gather details on care provided by the index practitioner. Charts were abstracted by a single research assistant. For key chart elements, reliability exceeded 0.9. Chart abstractors could not be blinded to practitioner type because specialty was often designated on the chart. Telephone interviewers asked patients about care received; the type of practitioner initially seen was identified in the baseline interview but not in the follow-up interview. “Functional recovery” was determined to be the date on which the patient could perform daily activities as well as he or she could before the episode of back pain.

We examined bivariate associations between the outcome of interest (use of imaging studies) and each patient or practitioner characteristic. We used t-tests or the Mann-Whitney test for continuous variables and the Pearson chi-square test for categorical data. These bivariate analyses were followed by multivariable modeling techniques that used logistic regression; this allowed us to simultaneously control for multiple independent variables. The covariate-adjusted probabilities with which patients would receive an imaging test and 95% CIs by category of practitioner and patient characteristics were calculated from the model-estimated β-coefficients and SEs. For appropriate analyses, we used design-based survey estimators [15] to correct the SE for any intrapractitioner correlation resulting from the cluster sampling scheme. We used SAS (SAS Institute, Cary, North Carolina) and Stata (Stata Corp., College Station, Texas) for the statistical analyses.

Results

A total of 1633 patients were enrolled in the cohort. Follow-up during the 6-month study period was excellent; 97% (n = 1580) of the initial cohort was assessed during this time. The main study results, reported elsewhere [10], indicated that function improved rapidly; 95% of patients could perform their usual daily activities at 6 months as well as they could before the back pain episode. Baseline demographic and clinical characteristics of the cohort are shown in Table 1. Previous analyses [10] have shown only minor differences in demographic and baseline clinical factors among the six practitioner strata.

According to the sampling strategy, patients were included only if 1) they had acute back pain for which they had not previously visited a practitioner and 2) they had not previously had surgery. This strategy allowed us to compare like patients across the practitioner strata. The patients were young, and two thirds had had symptoms for 2 weeks or less before seeing the practitioner. One quarter of the patients had sciatica (pain at the level of the knee or below) at presentation. Eleven percent had abnormal results on neurologic examination (asymmetric foot strength or ankle jerk reflex) that was done at the baseline visit.

Demographic Characteristics of Patients

Forty-seven percent of patients had lumbar spine radiography at some time during their pain episode, and 9% had either CT or MRI. Table 2 shows the patient characteristics associated with receipt of an imaging study. White race was associated with receipt of both lumbar spine radiography and CT or MRI. An unexpected finding is that patients with better insurance or more education or those claiming workers' compensation were less likely to receive lumbar spine radiography.

Clinical Characteristics of Patients

Presence of a neurologic deficit, practitioner diagnosis of disc disease, practitioner assessment of severe pain, long duration of pain before the index visit, and presence of sciatica were all associated with use of radiography and CT or MRI. The number of previous episodes of back pain was minimally associated with use of radiography and had no effect on use of CT or MRI. Time (in days) to functional recovery was strongly associated with use of radiography and use of CT or MRI. For radiography, time to functional recovery indicated only baseline functional impairment because 85% of all radiographs were obtained at the time of the index visit to the practitioner. Previous receipt of lumbar spine radiography did not prevent receipt of radiography for the current episode of acute back pain; previous CT or MRI increased the likelihood that the patient would receive CT or MRI for the current episode of pain.

Characteristics of Practitioners

Chiropractors and orthopedic surgeons were much more likely to obtain radiographs than were primary care physicians (both those in private practice and those in a health maintenance organization). Table 3 also shows a more modest but statistically significant difference among practitioner strata in the proportion of practitioners ordering CT or MRI during the 6-month follow-up period. Patients who sought care from orthopedic surgeons were most likely to receive CT or MRI. Ownership of a radiograph machine was associated with use of radiography, but this finding did not persist when we controlled for practitioner specialty in a logistic regression model. Almost all chiropractors and orthopedic surgeons (97%), the specialists most likely to obtain radiographs, owned radiograph machines. Among primary care providers who did not belong to a health maintenance organization, those who owned radiograph machines were no more likely to use radiography than were those who did not own them. Solo practice was associated with use of radiography, and this finding persisted after adjustment for specialty. Solo practice was not associated with greater ownership of radiograph machines. Practitioners who had attended a continuing medical education meeting on back pain were more likely to use radiography. These findings were not seen for CT or MRI.

Multivariable Analysis

We used logistic regression to examine the independent effects of the patient and practitioner factors on use of lumbar spine radiography or CT or MRI. Overall, the logistic regression analysis confirmed the findings of the bivariate analysis. Ownership of or access to a radiograph machine was not independently associated with use of radiography. Long duration of pain before the index visit, number of previous episodes of back pain, lack of workers' compensation, practitioner assessment of pain severity, poor baseline functional status, practitioner specialty, and small practice size all remained significantly associated with use of lumbar spine radiography. The bivariate findings on race and insurance were not seen in the multivariable analysis. On logistic regression, patient receipt of CT or MRI was independently associated with white race, neurologic deficit at baseline, practitioner assessment of disc disease at baseline, severe baseline functional status, sciatica, and small group-practice size. Use of imaging studies was also associated with time to functional recovery—the longer the patient had been ill, the greater the likelihood that he or she would receive an imaging procedure. According to the logistic regression analysis, practitioner specialty was not significantly associated with use of CT or MRI.

After 6 months of follow-up, 89 patients (5%) had not had functional recovery. Thirty-four percent of these patients did not receive lumbar spine radiography during the study period, and 60% had not received CT or MRI. Despite these small numbers, previous receipt of an imaging study did not change the proportion of patients with chronic back pain who had not received these studies.

Correlation of Responses to Clinical Vignettes with Test-Ordering Behavior

Each practitioner completed a survey at the end of the cohort study. No study results were shared with practitioners before the survey was completed. Cherkin and colleagues [12] found high rates of use of CT or MRI on responses to vignettes. When responding to the vignettes, more than 70% of primary care respondents in Cherkin and colleagues' study indicated that they would use CT or MRI for patients who had acute pain with sciatica and neurologic signs (vignette 1). Vignette 3 describes a patient with a 5-day history of back pain without sciatica; Cherkin found that 18% of family physicians and 25% of internists would order CT or MRI for this patient (Cherkin DC. Personal communication).

We used these same vignettes to compare the responses of our sample of providers with those of the national sample and to evaluate the extent to which such responses reflected actual ordering of tests (Table 4). Because we examined the effect of practitioner characteristics on practitioner behavior, the unit of analysis in Table 4 is practitioners (n = 162), for whom we have complete survey data. All patients described in the vignettes had received radiography; practitioners were asked to recommend or not recommend further imaging studies. We found that the responses to the vignettes overestimated actual practitioner ordering of CT or MRI. Most of our patients were similar to those described in vignette 3 (mechanical back pain); some were similar to those described in vignette 1 (sciatica with reflex loss). Fourteen percent of respondents indicated that they would order CT or MRI for patients with mechanical pain, and 34% said that they would obtain these scans for a patient with sciatica. Six percent of study patients with mechanical pain and 18% of those with sciatica received CT or MRI. According to the responses to the vignettes, orthopedic surgeons were more likely to recommend CT for the patient with acute pain (vignette 3) than for the patient with sciatica (vignette 1); this counterintuitive finding may be related to the small sample size (n = 19) in this stratum.

Approximately half of all practitioners ordered at least one CT or MRI for the patients studied. Practitioners who indicated on their responses to the vignettes that they would order CT or MRI were as likely to actually order CT or MRI for their patients as were practitioners who indicated a negative response on the vignette. For example, 48% of practitioners who answered “yes” on vignette 1 and 48% of those who answered “no” ordered CT or MRI for at least one patient. We used logistic regression to estimate the probability of ordering CT or MRI, controlling for between-practitioner differences in patient characteristics of duration of pain, presence of sciatica, income, baseline Roland-Morris disability score, income, and workers' compensation. We estimated nine logistic regression models, one for each practitioner type (primary care, chiropractic, orthopedics) by each of the three vignettes. Of the resultant regression equations, a positive response on a vignette predicted that CT or MRI would be ordered (P < 0.05) only by primary care providers and only for the sciatica vignette (vignette 1) and the chronic pain vignette (vignette 2). Overall, little evidence indicated that response to the vignettes predicted whether individual physicians would order CT or MRI.

Discussion

Extensive published literature and recent Agency for Health Care Policy and Research guidelines have emphasized the advisability of using imaging conservatively in patients with acute low back pain. The natural history of the disorder is characterized by moderately severe functional disability at the onset of the illness with relatively rapid clinical improvement. The median time to functional improvement in our cohort was 9 days; half of the patients declared themselves “all better” by 4 weeks. Relatively few of the patients in our low-risk cohort would have received radiography had the clinicians in our study followed current guidelines. Patients who had a relatively long duration of illness before presentation to the practitioner plus sciatica might be considered more reasonable candidates for radiography. We found that among primary care providers, 41% of patients with both of these clinical characteristics received radiography but 20% of patients with neither characteristic received radiography. Chiropractors and orthopedic surgeons frequently used radiography; unlike primary care physicians, their use of radiography was little influenced by clinical presentation. Even among patients who had neither clinical characteristic, 62% of the patients seen by chiropractors and 70% of those seen by orthopedic surgeons received radiography.

We found that almost all orthopedic surgeons and chiropractors owned radiograph machines; we could not separate the effect of specialty from the effect of ownership of this equipment. We found no relation among primary care physicians between use of radiography and ownership of a radiograph machine. This finding contrasts with those of previous studies. The study by Swedlow and colleagues [16] took place in California and did not examine the effect of practitioner specialty. Hillman and colleagues [17] examined the use of various imaging tests in a large national secondary dataset. They found that ownership of a radiograph machine was associated with a fourfold increase in the likelihood of obtaining radiographs for back pain. This relation persisted regardless of specialty, but the investigators did not consider practice size. Similar results were found when the United Mine Workers' Insurance database was examined [18]. Our most consistent finding was the major effect of specialty on use of radiography. We cannot completely explain the differences between our findings and those of these previous studies. North Carolina may differ from the previous sampling frames, but the prospective nature of our study is a strength.

Although the use of imaging for diagnosis appears to occur too early and too frequently, some patients in our study with prolonged duration of symptoms received no diagnostic tests. For many practitioners, the decision to order imaging studies for patients with back pain is related to specialty; the presenting clinical picture exerts some influence. We could not measure patient anxiety or radiographs obtained at the request of patients; these factors may have had some effect. Although patients with sciatica or a high level of initial functional impairment may have a more prolonged clinical course, little is lost by delaying diagnostic testing for several weeks to determine whether spontaneous recovery will occur. Fifty-five percent of patients with sciatica and a Roland-Morris score greater than 11 (indicating moderately severe functional disability) recover to the level of baseline functional status within 3 weeks after the initial practitioner visit.

Clinical vignettes have frequently been used as an inexpensive way to predict actual practitioner practice patterns in response to standardized clinical situations. Their value as predictors of clinical behavior has been controversial [19]. The limited literature comparing practitioner responses to simulated patients and behavior in clinical situations supports our findings. In a study of British general practitioners, Morrell and Roland [20] found no correlation between responses to simulated care histories about specialty referral and practitioner-specific referral rates. Norman and Feightner [21] compared the response of medical students to patient management problems with recommendations when dealing with simulated patients. Their results were similar to ours: They found that respondents chose twice as many management recommendations on the “paper” problem than when confronted with a simulated patient having the same characteristics. They found that “differences between formats (paper problem vs. simulated patient) accounted for a greater proportion of variance in total number of options than differences between students or cases. …” We found that although vignettes may reflect what practitioners think about back pain, they do not predict how practitioners will treat their own patients. When considering whether to order diagnostic tests or make referrals, clinicians may respond to many cues: visual cues from patients, the patient's expressed desire for the test or referral, or clinical variables not included in the vignette. Vignettes do, however, predict aggregate behavior: Patients with sciatica in vignettes and in real life are more likely to receive CT.

Such clinical factors as patient age, functional status, and neurologic findings explain some of the variability in practitioners' use of diagnostic tests. Practitioners have a low threshold for using radiography and a somewhat higher threshold for using CT or MRI. Previous use of radiography does not appear to diminish the likelihood of future radiography. We do not have information on whether the previous radiograph was obtained from the same or a different provider. Both lumbar spine radiography and CT or MRI appear to be substantially overused.

Our study findings can be generalized to patients in North Carolina, but a single state may not be representative of other areas of the United States. We did not measure some factors that could influence practitioner test-ordering, such as patient anxiety, patient pain behavior in the office, or a patient's voiced fear of cancer.

Attempts to implement practice guidelines for low back pain should consider these issues. Barriers to reducing the use of radiography vary substantially depending on provider specialty. The behavior of primary care physicians may well be more malleable than that of chiropractors and orthopedic surgeons. The specialists use radiography much more frequently and do not appear to be influenced by the clinical presentation of the patient. We recommend that guideline implementation efforts specifically target these clinical and practitioner factors as factors about which health care insurers, providers, and patients will have questions. Patients with a high level of functional incapacity and its attendant anxiety at presentation may be more likely to request diagnostic studies so as to obtain reassurance that they are not seriously ill.

We have identified the clinical, demographic, and practice characteristics associated with the use of two types of imaging technology in patients with acute low back pain. Given current information on the clinical course of low back pain, a substantial reduction in the number of studies done appears possible. Collaboration with specialty societies will be very important in attempts to change practitioner behavior.

Appendix

Investigators participating in the North Carolina Back Project were P. Curtis, J. Darter, G. DeFriese, J. Fryer, N. Hadler, G. Hunter, J. Joines, A. Jackman, W. Kalsbeek, C. McLaughlin, T. Konrad, T. Ricketts, and D. Taylor (University of North Carolina, Chapel Hill, North Carolina); R. McNutt (University of Wisconsin, Madison, Wisconsin); and D. Smucker (University of Cincinnati, Cincinnati, Ohio).