Study design. The bottom panel provides an example of when ascertainment of low back pain (LBP) diagnosis, exclusion criteria, imaging, and subsequent clinical events were measured for a hypothetical patient A. CT indicates computed tomography.
Study population. CT/MRI indicates computed tomography or magnetic resonance imaging; CTS PCP, Community Tracking Study primary care physician; LBP, low back pain; XR, radiography.
Pham HH, Landon BE, Reschovsky JD, Wu B, Schrag D. Rapidity and Modality of Imaging for Acute Low Back Pain in Elderly Patients. Arch Intern Med. 2009;169(10):972–981. doi:10.1001/archinternmed.2009.78
Most quality metrics focus on underuse of services, leaving unclear what factors are associated with potential overuse.
We analyzed Medicare claims from 2000-2002 and 2004-2006 for 35 039 fee-for-service Medicare beneficiaries with acute low back pain (LBP) who were treated by 1 of 4567 primary care physicians responding to the 2000-2001 or 2004-2005 Community Tracking Study Physician Surveys. We modified a measure of inappropriate imaging developed by the National Committee on Quality Assurance. We characterized the rapidity (<28 days, 29-180 days, none within 180 days) and modality of imaging (computed tomography or magnetic resonance imaging [CT/MRI], only radiograph, or no imaging). We used ordered logit models to assess relationships between imaging and patient demographics and physician/practice characteristics including exposure to financial incentives based on patient satisfaction, clinical quality, cost profiling, or productivity.
Of 35 039 beneficiaries with LBP, 28.8% underwent imaging within 28 days and an additional 4.6% between 28 and 180 days. Among patients who received imaging, 88.2% received radiography, while 11.8% received CT/MRI as their initial study. White patients received higher levels of imaging than black patients or those of other races. Medicaid patients received less rapid or advanced imaging than other patients. Patients had higher levels of imaging if their primary care physician worked in large practices. Compared with no incentives, clinical quality–based incentives were associated with less advanced imaging (10.5% vs 1.4% for within 28 days; P < .001), whereas incentive combinations including satisfaction measures were associated with more rapid and advanced imaging. Results persisted in multivariate analyses and when the outcome was redefined as the number of imaging studies performed.
Rapidity and modality of imaging for LBP is associated with patient and physician characteristics but the directionality of associations with desirable care processes is opposite of associations for measures targeting underuse. Metrics that encompass overuse may suggest new areas of focus for quality improvement.
Insurers press for standardized measurement of physicians' performance to broaden programs that link performance to financial incentives. Existing metrics focus on patient experience, relative costs for comparable conditions, and clinical quality. Yet most currently available measures on clinical quality focus on the underuse of services.1 Few focus on the overuse of services that might induce harm through complications from unnecessary follow-up testing or treatment2 and/or raise health care costs with little improvement in outcomes. Moreover, if patients tend to prefer more rather than fewer services, programs that only measure underuse but omit overuse may induce even more overuse of services.
Policy makers call for the development of overuse measures and their inclusion in performance measurement and incentive programs,3 yet little is known about the patterns of care that these measures would reveal. To help anticipate the effects of applying overuse measures, we investigated associations between characteristics of patients and their primary care physicians (PCPs) and the rapidity and modality of imaging those patients receive for uncomplicated acute low back pain (LBP). We focused on LBP as a prevalent condition4 for which imaging—particularly with advanced modalities—is rarely indicated, even for elderly patients.5- 7
We analyzed claims data for Medicare fee-for-service beneficiaries treated by respondents to the 2000 to 2001 or 2004 to 2005 Community Tracking Study (CTS) Physician Surveys. Samples for these nationally representative surveys are clustered in 60 metropolitan statistical areas. Responses from 2000 to 2001 were linked to claims for years 2000 to 2002, while responses from 2004 to 2005 were linked to claims for years 2004 to 2006. Details on the survey and its linkage to Medicare claims are described elsewhere.8,9 Our analysis focuses on CTS primary care physicians (ie, specialties of general internal medicine, general practice, family practice, internal medicine/pediatrics, geriatrics). The pooled data set included information on 4567 unique PCPs and 3 years of claims for each beneficiary (Figure 1).
We used the first year of observation in each 3-year period to identify exclusion criteria and adjust for comorbidities. Using the second year of observation, we identified beneficiaries as having an episode of acute LBP if they had relevant diagnostic codes from an ambulatory or emergency department visit but lacked similar diagnostic codes within the 6 previous months. To assess imaging, we examined Medicare claims from inpatient, outpatient, and professional services files to detect radiologic studies of the lower back within 6 months of the incident LBP diagnosis. We derived diagnostic codes used to identify LBP and procedure codes used to identify imaging procedures from the measure of inappropriate imaging for LBP developed by the National Committee on Quality Assurance (NCQA).10
Starting with the earlier round of the CTS survey, we used the Unique Physician Identification Number (UPIN) to identify the physician who billed for the greatest number of evaluation and management visits during the entire observation period (January 1, 2000, through June 30, 2002) for a given beneficiary. We similarly identified eligible PCPs from the later round of the survey using claims for years 2004 to 2006. We focused on beneficiaries whose “plurality” physician was a CTS PCP. To assess the robustness of our results to methods of attribution, we conducted 2 separate analyses attributing beneficiaries who (1) had any evaluation and management visits with the CTS PCP during the observation period and (2) had their initial LBP diagnosis coded by the CTS PCP. To identify whether the imaging study was conducted in the same organization in which the CTS PCP worked, we used the tax identification number associated with each physician's UPIN on claims.
Eligible patients were enrolled in Medicare for at least 6 months after the incident diagnosis of LBP. Thus, identified episodes of LBP occurred between January 1, 2001, and June 30, 2002, for the first cohort. For years 2000 to 2002, we included beneficiaries 65 years and older as of January 1, 2000, who were continuously enrolled in fee-for-service Medicare through the subsequent 18 months (until June 30, 2002) and did not enter hospice or nursing homes. We excluded those with end-stage renal disease or who received care in more than 1 state during 2000. This procedure was repeated for the next CTS round linked to 2004 to 2006 Medicare claims. We excluded beneficiaries whose initial LBP diagnosis was coded by a radiologist because this may reflect patient self-referrals.
To focus on cases of uncomplicated LBP, we used the NCQA definition10 to identify these episodes in claims. We applied additional, more stringent criteria to exclude patients who might have had a legitimate indication for imaging. Thus, patients whose claims indicated neurologic deficit, trauma, low impact injuries, cancer, infection, nonspecific anemias, or constitutional symptoms suggestive of increased cancer risk were excluded. Detailed criteria for identifying patients with uncomplicated LBP are given in eTable 1.
In primary analyses, we examined the rapidity and modality of imaging. We defined rapidity of imaging as an ordinal variable based on timing of the first imaging study—no imaging (none within 180 days of diagnosis), delayed (within 29-180 days), or rapid (within 28 days). Modality of imaging was defined as a separate ordinal variable—no imaging, radiography only, or computed tomography or magnetic resonance imaging (CT/MRI). The CT/MRI category included patients who received radiography and CT/MRI concurrently or serially. In secondary analyses, we considered the number of imaging studies performed within 180 days (none, 1, or >1).
We included patient sex, race (white, black, or other [because data on other racial categories in Medicare data are not as reliable]), and Medicaid eligibility, while adjusting for comorbidities.11,12 Drawing on data from the Area Resources File, we also adjusted for median household income in the beneficiary's zip code, and the percentage of adults 25 years and older in the county with 12 or more years of education, as aggregate measures of sociodemographic status.
We assessed the rapidity and modality of imaging relative to PCPs' self-reported financial incentives, which included the following: (1) practice ownership (employed, part owner, or full owner); (2) the overall effect of incentives, derived from responses to the question “How would you describe your overall personal financial incentives in your practice? On balance, do these incentives favor reducing services to individual patients, favor expanding services to individual patients, or favor neither?”; and (3) a composite 9-category variable constructed from a set of questions identifying factors on which physicians' compensation was based (only his or her own productivity; only clinical quality measures; only patient satisfaction surveys; productivity and clinical quality measures; productivity and patient satisfaction surveys, clinical quality measures and patient satisfaction surveys, all 3, or none; and solo practice physicians who were not asked questions on incentives). We did not include in this composite responses about physicians' exposure to incentives based on cost profiling because this variable was not associated with rapidity or modality of imaging in bivariate or multivariate analyses.
We considered the individual physician's specialty (family/general practice vs general internal medicine); number of years in practice; board certification; and medical school site (United States/Canada vs elsewhere). Practice characteristics included type (solo/2-person practice, small group of 3-10, medium group of 11-50, large groups of >50, medical school, group/staff health maintenance organization, and all other types); and the percentage of practice revenues derived from Medicare, Medicaid, and capitation (each categorized as tertiles). We adjusted for radiologist supply (number of patient care radiologists per capita in the metropolitan statistical area), and urban vs rural location.13 Finally, to control for unmeasured market factors, we included a dummy for each of the 60 metropolitan areas in the CTS sample.
Using claims from the 6-month period subsequent to LBP diagnosis, we examined the rate of potential “cascade effects” of imaging and clinical conditions that might post hoc justify imaging that beneficiaries received as well as its rapidity. We considered (1) hospitalization or surgery for LBP,14 (2) complications or progression of LBP, and (3) new diagnoses of low back fractures or cancers with potential for bone metastases (eTable 2).
We performed ordered logit regressions with either the rapidity or modality of imaging as the dependent variable (the referent was no imaging). To assess whether associations between the CTS PCPs’ characteristics and imaging were influenced by care delivered by other physicians, we repeated analyses excluding beneficiaries who had visits with physicians other than the CTS PCP between the dates of LBP diagnosis and imaging (or within 6 months for beneficiaries who did not receive imaging). To account for secular trends, we also evaluated the interaction between each performance incentive variable and the period of the survey (2000-2002 or 2004-2006).
Analyses were conducted using SUDAAN analytic software (release 7.0; Research Triangle Institute International, Research Triangle Park, North Carolina), which accounts for clustering of patients within physicians and multiple observations of physicians responding to both survey rounds. We applied weights to reflect the probability of sampling and known differences between respondents and nonrespondents. Weighted estimates are representative of nonfederal PCPs providing patient care at least 20 hours per week in the continental United States and the Medicare beneficiaries whom they treat.
Of 496 529 beneficiaries who met inclusion criteria and had a CTS PCP, 35 039 had a diagnosis of uncomplicated acute LBP during the 6-month assessment period (Figure 2). Table 1 gives the characteristics of their CTS PCPs. Among these beneficiaries, 11 294 (32.2%) had at least 1 imaging study within 180 days of diagnosis; 9637 (28.8%) underwent imaging within 28 days of diagnosis, and 1657 (4.6%) between 28 and 180 days. Among these patients, 88.2% had a plain radiograph, while 11.8% had CT or MRI as their initial study. The median number of days between diagnosis and imaging was zero (interquartile range [IQR], 0-7) for any modality, 11 (IQR, 2-42) for CT, and 10 (IQR, 3-30) for MRI.
For the 11 294 beneficiaries undergoing imaging within 180 days, 2250 (19.9%) were treated by an orthopedic surgeon between the dates of diagnosis and imaging. The proportions who visited chiropractors, neurosurgeons, or rheumatologists were 1619 (16.4%), 170 (1.5%), and 248 (2.0%), respectively. Among the beneficiaries who did not receive imaging, 15.9% saw a chiropractor, 0.9% a neurosurgeon, and 2.7% a rheumatologist.
In a third of cases (29%), beneficiaries received imaging within their CTS PCP's practice organization. In bivariate analyses, the site of imaging was not associated with the CTS PCP's practice ownership status (data not shown). This held true irrespective of imaging modality. However, time between diagnosis and imaging was significantly shorter for beneficiaries who received imaging within their CTS PCP's practice vs those undergoing imaging elsewhere (mean, 9.5 vs 15.7 days).
Minority beneficiaries received less rapid and less advanced imaging than white beneficiaries (24.8% [black], 18.9% [other races], and 29.7% [white] [P < .001] for imaging within 28 days and 9.1% [black], 7.2% [other races], and 10.8% [white] [P < .05] for CT/MRI). Beneficiaries also covered by Medicaid received less rapid and less advanced imaging vs other patients (22.7% vs 29.7% [P < .001] for imaging within 28 days and 7.3% vs 11.0% [P < .001] for CT/MRI) (Table 2). These results persisted in multivariate analyses (Table 3).
Beneficiaries treated in practices more reliant on Medicaid revenues received less rapid and less advanced imaging than other beneficiaries (highest vs lowest tertile, 23.8% vs 30.7% [P < .001] for imaging within 28 days and 8.9% vs 11.2% [P < .01] for CT/MRI). Similarly, patients whose CTS PCPs with greater than 25% of revenues derived from capitation or who were solo practitioners received both less rapid and advanced imaging than those treated in practices receiving no practice revenues from capitation or who were employed (Table 2).
Beneficiaries treated by CTS PCPs with variable compensation based only on patient satisfaction received more rapid and advanced imaging than beneficiaries whose PCPs were exposed to other combinations of incentives (Table 2 and Table 3). In contrast, beneficiaries whose CTS PCP was exposed to incentives based on both clinical quality and satisfaction, but not productivity, received less rapid imaging. These associations with financial incentives persisted in multivariate analyses. Interactions between the round of the survey and each combination of incentives were nonsignificant.
Beneficiaries treated by family and general practitioners consistently received less rapid and less advanced imaging than those cared for by general internists (Table 2). Beneficiaries with CTS PCPs in practices of 10 or more physicians were more likely to receive imaging within 28 days than those seeing physicians in solo or 2-person practices. In adjusted analyses, beneficiaries treated in larger group practices were also substantially more likely to receive advanced imaging (Table 3).
Our core findings of associations between the level of imaging and patient and physician characteristics persisted across a range of sensitivity analyses. Results were similar when we attributed beneficiaries who had any visits with the CTS PCP to that PCP and when we attributed to each CTS PCP only beneficiaries for whom they had coded the initial LBP diagnosis. Our results were also robust when we redefined the level of imaging as the total number of low back imaging studies that a beneficiary received.
Rates of new cancer diagnoses were similar regardless of the timing or modality of imaging (Table 4). Hospitalization for and complications or progression of back pain were more common among beneficiaries who received imaging than among those who did not.
This study focused on the use of imaging for uncomplicated acute LBP because it is prevalent, and well-established guidelines indicate that rapid or advanced imaging is not beneficial in the absence of specific complicating features or comorbid conditions.5- 7
We found that rapidity and modality of imaging for LBP was associated with nonclinical characteristics of patients and the physicians and practices that treated them. Low-income and minority patients and those treated in smaller practices or practices more reliant on Medicaid revenues received less rapid and advanced imaging than higher-income or white patients and those in larger practices or settings less reliant on Medicaid. These results are consistent with previously reported patterns of care in terms of groups who tend to receive fewer services, although in our study, the patterns represent better quality care than when measuring quality in terms of underuse.15,16 Our findings were consistent across the timing or number of imaging studies and for all modalities. Moreover, patients cared for by physicians exposed to incentives based on patient satisfaction received more rapid and advanced imaging. Conversely, those whose physicians were exposed to both clinical quality and productivity incentives received less rapid and advanced imaging.
The association between exposure to satisfaction incentives and the level of imaging is not unexpected. Patients may consider imaging reassuring, and those with higher socioeconomic status may be more successful in obtaining testing in this context.17 However, in contrast to generally underused services such as diabetic monitoring, more rapid or advanced imaging for LBP may not benefit patients and may result in harm.18,19
Contrary to our hypothesis that incentives focused on underuse might result in greater potential overuse, we observed an inverse association between exposure to clinical quality incentives and the level of imaging. These results should be interpreted with caution because we had only a small number of observations for some combinations of incentives. Nevertheless, they suggest that quality-based incentives may improve appropriateness of care in some unmeasured arenas. This could reflect a broad ecologic effect if physicians exposed to existing incentives become more generally aware of and adherent to clinical practice guidelines.20 Alternatively, physicians exposed to quality measurement may be less vulnerable to the effects of incentives that encourage imaging.
We also found that exposure to both clinical quality and satisfaction incentives among a small fraction of PCPs resulted in less rapid or advanced imaging than with either incentive alone or no incentives at all. Although some prior studies suggest that patient satisfaction tends to correlate with clinical quality,21,22 Landon et al23 compared the experiences of Medicare beneficiaries in the fee-for-service vs managed care programs and found instead that the 2 systems had different strengths in quality vs satisfaction performance. Our findings are also consistent with those of Weyer et al,24 who reported that improvements in preventive services were associated with declines in patient satisfaction. Although our results require confirmation in larger populations of physicians facing this relatively rare combination of incentives, it is possible that physicians less pressured to maximize visit volume can better align their efforts to perform well on both clinical quality and patient satisfaction (eg, by spending more time providing reassurance about deferred imaging). However, these benefits would accrue to relatively few patients because far more physicians face productivity incentives than other types of incentives.25
We found that one-third of imaging studies were performed within the PCP's practice organization and that patients treated in large group practices were modestly more likely to receive rapid and advanced imaging than those treated in smaller practices. Since large groups are more likely to have the resources to invest in imaging equipment,26 our results are consistent with other studies suggesting that practice-owned equipment results in supplier-induced demand and physician self-referral.27 These associations were independent of the higher likelihood that physicians in large practices were exposed to performance incentives. They contrast with studies showing that larger practices tend to outperform smaller ones in quality improvement efforts and on standardized measures emphasizing underuse.15,28,29
Finally, we found that many patients received imaging on the day of diagnosis. These care patterns are inconsistent with guidelines that recommend a trial of conservative therapy first and suggest opportunities for quality improvement.
Our results should be interpreted within the context of our analytic approach. We could not determine appropriateness of imaging for a given patient. Ours is a comparative analysis of the level of imaging relative to patient, physician, and practice characteristics and not an attempt to benchmark the behavior of individual physicians. The inability to identify particular cases of overuse with certainty does not invalidate our findings because all physicians are subject to such errors in coding and ascertainment. As an example, for our results to be biased, physicians would have to be more likely to code diagnoses of “red flag” conditions for minority patients than for white patients at the same time that they are less likely to order imaging studies for minority patients. Although rates of back pain complications and hospitalizations was higher for patients receiving more rapid or advanced imaging, these differences may reflect, in part, events triggered by the initial imaging study itself and/or the relative aggressiveness of care delivered by physicians who are more likely to order imaging. Rates of cancer diagnoses were no higher for patients who had more rapid or advanced imaging.
We cannot be certain that each CTS PCP was truly responsible for imaging decisions for the patients studied because claims-based attribution may not reflect actual care relationships.29 However, our conclusions were robust when using both more liberal and more restrictive attribution approaches and when we focused on patients whose initial LBP diagnosis was coded by the PCP or those who saw only that physician between the dates of diagnosis and imaging. Thus, regardless of the involvement of other providers, the level of imaging appears to be related to the characteristics of a patient's PCP.
Our findings require confirmation with a broader set of overuse measures but nevertheless have important implications. Development of more overuse metrics would balance out current measures that are heavily weighted toward underuse, perhaps by leveraging research on the comparative effectiveness of different treatment options.30 It is not surprising that physicians and practices with attributes associated with underuse of care are less likely to overuse imaging. However, our results may provide a rationale for tailoring packages of performance metrics to practice attributes. For example, a practice with high Medicaid revenue, many minority patients, and minimal physician incentives for maximizing patient satisfaction may be more likely to benefit from measures emphasizing underuse. In contrast, practices with less Medicaid revenue, fewer minority patients, and more satisfaction-based incentives may be more likely to benefit from performance measures that include overuse.
More balanced measurement of overuse and underuse might also allow insurers to design incentives to counter both. For example, insurers could offer higher copayments for services that tend to be overused but lower or no copayments for generally underused services. Similarly, bonuses could encourage physicians to avoid overused services or increase underused services. Mandatory delays in ordering overused services in the absence of clinical “red flags” is another alternative tool. Such interventions would need to avoid limiting necessary care, particularly for low-income patients, but in the context of fee-for-service reimbursement, targeted and balanced incentives could better align the dual goals of quality improvement and cost containment.
Patterns of care revealed by the application of a single measure of potential overuse point to different foci for performance improvement than when measurement focuses on underuse. Given possible reversal in direction from the typical “quality advantage” for important subpopulations of patients and physicians, measuring overuse alongside underuse may be critical for improving the overall appropriateness of care.
Correspondence: Hoangmai H. Pham, MD, MPH, Center for Studying Health System Change, 600 Maryland Ave SW, Ste 550, Washington, DC 20024 (email@example.com).
Accepted for Publication: January 30, 2009.
Author Contributions: Dr Pham had full access to the data and takes responsibility for the integrity of the data and accuracy of the analysis. Study concept and design: Pham, Landon, Reschovsky, and Schrag. Acquisition of data: Pham. Analysis and interpretation of data: Pham, Landon, Reschovsky, Wu, and Schrag. Drafting of the manuscript: Pham and Schrag. Critical revision of the manuscript for important intellectual content: Pham, Landon, Reschovsky, Wu, and Schrag. Statistical analysis: Pham, Reschovsky, Wu, and Schrag. Obtained funding: Pham. Administrative, technical, and material support: Pham. Study supervision: Pham and Schrag.
Financial Disclosure: None reported.
Funding/Support: This study and the Community Tracking Study Physician Survey were supported by the Robert Wood Johnson Foundation. This analysis was also supported by grant R01 AG027312-02S1 from the National Institute on Aging.
Role of the Sponsors: The funding agencies had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation, review, or approval of the manuscript.