Background and Purpose. Little information is available on factors associated with physical therapist use by people with back or neck pain. Identifying the characteristics of people who seek care from physical therapists is a useful first step in determining whether there is appropriate use of physical therapy services. The purpose of this study was to identify factors associated with physical therapist use by people with back or neck pain. Subjects. The subjects were 29,049 people who had back pain or neck pain, or both, and who were seen for an initial evaluation at 1 of 21 US spine care centers. Each subject and evaluating physician completed a written survey at the time of the initial evaluation. Methods. Multiple logistic regression analyses were conducted to identify factors associated with physical therapist use. Results. Several measures of health and illness severity were associated with physical therapist use. Age and being male were negatively associated with physical therapist use. Education level, having workers' compensation coverage, and being in litigation were positively associated with physical therapist use. Physical therapist use also varied by previous physician use and census region. Discussion and Conclusion. The results suggest that variations in physical therapist use are associated with factors other than health and illness severity. The results also suggest that people who would benefit from physical therapy may not be receiving it or that people who would not benefit from physical therapy are receiving it, or both.
Physical therapists commonly treat people with back and neck pain,1 and there is a fair amount of evidence to support the use of at least some of the interventions that they deliver.2–17 Despite the substantial use of physical therapists by people with back pain or neck pain, or both, and the potential that physical therapy has for improving outcomes, information on the characteristics of people who see physical therapists and how they compare with the general population with back or neck pain is limited. Identifying the characteristics of people who seek care from physical therapists for the management of back or neck pain is important from a health policy perspective because it will help lead to an understanding of whether there is appropriate use of physical therapy services. Variations or differences in physical therapist use, explained by factors other than health status or need (eg, race, socioeconomic status), may be indicative of underuse or overuse of services for certain groups. Numerous studies have suggested that people of a lower socioeconomic status, racial and ethnic minority groups, and certain geographic groups are not receiving necessary care or are receiving care of a lower quality.18,19 More recent studies also have raised questions about whether women, children, elderly people, and people with chronic illnesses are receiving necessary health care.18 Whether or to what extent physical therapy services are underused is largely unknown.
Differences in the use of health care services (health care use) may not always be indicative of underuse of services. For example, regional differences in surgical procedures20 and in health care use by Medicare beneficiaries21,22 are considered to be indicative of overuse of services in some parts of the United States. Variations in physical therapist use, explained by factors other than health and need, also may be indicative of overuse of services (ie, people who would not benefit from physical therapy are receiving it). Overuse of health care services is important from a health policy perspective because it leads to unnecessary health care costs.
We found 1 study that specifically focused on factors associated with physical therapist use for the management of low back pain (LBP) in the United States.23 Mielenz et al23 analyzed data collected in 1992 and 1993 for 1,580 people with acute LBP in North Carolina. In multivariate analyses, they found that the following factors were positively associated with physical therapist use: higher Roland-Morris Disability Questionnaire scores (indicating greater disability), pain below the knee in 1 or both legs, greater than a high school education, receipt of workers' compensation, and previous physical therapist use for LBP. Demographic characteristics not associated with physical therapist use were age, sex, race, income, insurance status, and marital status. Although this study provided important information on factors associated with physical therapist use, the generalizability of the results is limited because only people with acute LBP were studied. In addition, the data were from 1 state and are over 10 years old.
Other studies24–31 that have addressed factors associated with physical therapist use for the management of back pain or neck pain, or both, are summarized in Table 1. These studies were quite varied with regard to samples, specific characteristics assessed, and data analyses. Despite these differences, 1 consistent finding was that the severity of back or neck pain (measured in a number of different ways) was positively associated with physical therapist use. The relationships among demographic and socioeconomic characteristics and physical therapist use were less clear, and comparisons across studies are difficult because of differences in samples. In addition, some studies did not include demographic and socioeconomic variables in their analyses. A majority of the studies also were conducted in countries other than the United States that have different health care systems and population demographics. These studies, therefore, are not particularly useful in increasing the understanding of demographic and socioeconomic characteristics associated with physical therapist use in the United States.
Three of the studies summarized in Table 1 also incorporated physical therapist use with the use of other health care providers. For example, the dependent variable in a study by Carey et al24 was a dichotomous measure representing use or no use of 1 or more of the following: physician, chiropractor, and physical therapist. Findings from studies such as this one are limited because factors that determine whether an individual seeks care from a physical therapist may not be the same as factors that determine whether an individual seeks care from another type of provider (eg, physician). Some of the available research supports this contention.28,29 Studies in which bivariate analyses were conducted (eg, determining whether the mean age of people who saw a physical therapist differs from the mean age of people who did not see a physical therapist) also are limited because such analyses do not control for confounding by other factors (eg, pain severity) that may contribute to variations in physical therapist use. Finally, most of the studies that have been conducted on health care use for back or neck pain have been conducted for people with LBP. Whether care-seeking differs between people with neck pain and those with LBP has not been well investigated. Some data suggest that rates of care-seeking differ for people with neck pain and those with LBP.32,33
The primary objective of this study was to use a large, current national database, the National Spine Network (NSN) database, to identify factors associated with physical therapist use by people with back or neck pain. A secondary objective was to determine whether factors associated with physical therapist use varied between people with LBP and those with neck pain. On the basis of empirical data on factors associated with physical therapist use and health care use in general, our hypotheses were that physical therapist use would vary by health-related, demographic, and socioeconomic factors and that factors associated with physical therapist use would differ between people with LBP and those with neck pain.
The NSN is a consortium of US spine care centers that collaborate in collecting outcome data on their patients.34 Centers include private practice clinics, academic medical centers, and multidisciplinary spine care centers. As of August 2004, 33 spine care centers were members of the NSN.
The NSN database contains self-report survey data from patients and their physicians. Patient and physician surveys are completed during the patient's initial or baseline visit and subsequently at selected follow-up visits. The survey instrument was developed jointly by the NSN, the American Academy of Orthopedic Surgeons, the Council of Musculoskeletal Specialty Societies, and the Council of Spine Societies. Data reported by patients include demographic information, symptoms, comorbidities, health status, functional status, medications used, work status, use of care, expectations about care, and satisfaction with care. Data reported by physicians include patient signs and symptoms, surgical history, diagnosis, tests ordered, treatment plan, and assessment of patient progress. Patients, physicians, and centers are identified in the database by identification numbers. No data that could be used to specifically identify a patient, physician, or center are provided.
On a weekly basis, participating clinics mail completed survey questionnaires to the central coordinating center. Survey questionnaires are returned to participating clinics if key data are missing or the data are invalid. Data from the questionnaires then are keyed in to a preliminary database by a data-entry technician. A second, independent data-entry technician keys in the data again, and any discrepancies are resolved. Data then are loaded into the central NSN data repository.
The NSN database offers a unique source of data to explore issues related to the usual care received by people with spine problems. The sheer number of records included in the database (over 60,000 as of December 2002) and the fact that spine care centers across the United States contribute to the database also increase the generalizability of analyses conducted with the database. People with chronic spine problems make up a majority of the database, and the literature suggests that these people, in particular, may be the most likely to benefit from physical therapy.16
Although specific data on participation rates (ie, the number of people who agree to complete the survey questionnaires/the number of people who are eligible to complete the survey questionnaires) at each of the spine care centers are not available, participation rates at 1 center that contributes to the database are more than 98% (B. Hanscom, personal communication, August 2004). Information on how participants (ie, people who agree to complete the survey questionnaires) compare with nonparticipants (ie, people who choose not to complete the survey questionnaires) also is not available.
The analyses represented here are based on NSN data from 1998 to 2002. Twenty-one spine care centers (Appendix) contributed data over these 5 years, with a mean (SD) of 1,383 (2,056) records per center. Our sample consisted of subjects who were seen for an initial evaluation and for whom complete information on previous use of health care providers was available (N=29,049).
The analytic framework for this study (Figure) derives from the behavioral model of health care use of Andersen and Newman.35 This model is the most widely adopted framework for studying health care use and is amenable for framing secondary analyses. The model views health care use as a function of need, enabling, and predisposing characteristics of the individual. Need characteristics are considered the most immediate cause of health care use and can include a variety of measures (reported by both people seeking health care services and clinicians) that reflect an individual's health. People in need of health care services must have some means of obtaining them. This factor is reflected by the enabling component of the model, which includes family resources, such as income and insurance coverage, and community resources, such as the availability of health care providers. The predisposing component reflects the fact that some people have a greater propensity than others to use health care services. Predisposing characteristics include sociodemographic characteristics and attitudes and beliefs about health care. Predisposing characteristics, in and of themselves, are not directly responsible for health care use. For example, race is not considered a reason for seeking health care services. Rather, people of different races have different experiences, beliefs, and attitudes that affect their health care use.
Our choice of variables and our hypotheses were based on the model of Andersen and Newman35; the data available in the NSN database; the results of previous studies that examined factors associated with health care use for neck pain or back pain, or both23–31; our previous work on determinants of health care use36–40; and our clinical experience.
Descriptive statistics on the study variables are presented in Table 2. The dependent variable for the analyses was whether subjects had seen a physical therapist for their spine-related conditions. This variable was created on the basis of the responses to the following question: “What types of health care providers have you used for your spine-related condition?” Seventeen possible response categories were provided; 1 of these was “a physical therapist.” Forty-seven percent of the subjects (n=13,710) indicated that they had previously seen a physical therapist for their spine-related conditions.
Need characteristics included primary diagnosis, level of involvement, duration of the problem, number of comorbidities, general health, history of depression, history of injection, and history of surgery. Although the NSN survey instrument includes a number of specific questions on functioning and symptoms, we did not include these data in our analyses because these questions focus on the subjects' symptoms over the preceding 1 to 4 weeks. The dependent variable in our analyses represented any physical therapist use since the subjects' spine-related problems began. Over 88% of the subjects in the database reported having spine-related problems for more than 6 months.
The physician portion of the NSN survey instrument has a diagnosis section that lists 37 different diagnoses. Physicians are instructed to mark 1 diagnosis as the primary diagnosis. They also are instructed to indicate levels of involvement. On the basis of the distribution of the data, the format of the diagnosis portion of the physician evaluation, and diagnostic categories developed by Hart et al,41 we grouped diagnoses into the following categories: herniated disk, spinal stenosis, spondylosis, pain syndrome, sprain or strain, deformity, and “other.” The diagnostic categories and their associated diagnoses are presented in Table 3. We created 3 dichotomous variables to represent levels of involvement: cervical (occiput–T2), thoracic (T3–T10), and lumbosacral (T11–ilium). These 3 variables were not mutually exclusive. We hypothesized that physical therapist use would vary by diagnosis and by level of involvement.
The remainder of the need characteristics were based on subject report. With regard to the duration of the spine-related problems, subjects were asked, “Overall, how long have you had spine-related problems?” Subjects could choose from 9 response categories ranging from “2 weeks or less” to “more than 3 years.” On the basis of the distribution of the responses and because chronic back pain usually is defined as pain that lasts for 3 months or more,42 we categorized the duration of the problem as less than 3 months, 3 months to 1 year, and more than 1 year.
Subjects also were asked whether they had any of 21 comorbidities. The comorbidity list was adapted from lists used by the North American Spine Society and the American Academy of Orthopaedic Surgeons. The mean and median numbers of comorbidities were 1.7 and 1, respectively. Therefore, we created a dichotomous variable to indicate whether subjects had 2 or more comorbidities. The general health variable was based on the subjects' responses to the following question: “In general, would you say your health is: (1) excellent, (2) very good, (3) good, (4) fair, or (5) poor?” We dichotomized this variable as “general health very good or excellent” and “general health good or less.” We also created a dichotomous variable to indicate the history of depression on the basis of the responses to 2 questions that have been found to be sensitive and specific screening questions for depression.43 The 2 questions were: “In the past year, have you had 2 weeks or more during which you felt sad, blue, depressed or when you lost all interest in things that you usually cared about or enjoyed?” and “Have you felt depressed or sad much of the time in the past year?” Response categories for both questions were “yes” and “no.” If subjects responded “yes” to either or both questions, then we classified them as having a history of depression.
Finally, we included 2 dichotomous variables to indicate whether the subjects had previously had surgery or an injection for their spine-related problems. We considered these 2 variables proxies for overall illness severity. We hypothesized that duration of the problem, number of comorbidities, poorer general health, history of depression, history of surgery, and history of injection all would be positively associated with physical therapist use. We also hypothesized that the need characteristics, in general, would explain the largest amounts of variations in physical therapist use.
Enabling characteristics included education level and health care payment characteristics. The NSN database does not include specific information on income; therefore, we used education level as a proxy measure for income and socioeconomic status. The only health care payment questions included in the NSN survey ask about Social Security disability coverage, disability insurance, and workers' compensation. Specific questions about other types of insurance (eg, private, Medicare, Medicaid) are not included. For each of the health care payment sources (ie, Social Security disability coverage, disability insurance, and workers' compensation), the response categories were: (1) am receiving, (2) applied for it, (3) planning to apply for it, (4) used to receive, and (5) not applicable. We created 2 dichotomous variables from the responses for these 3 payment sources. One variable indicated whether the subject was receiving or used to receive any type of disability insurance. The second variable indicated whether the subject was receiving or used to receive workers' compensation. We chose to code the disability insurance and workers' compensation variables in this manner because the dependent variable for this study was whether an individual had seen a physical therapist for a spine-related condition. Therefore, we were interested in identifying people who had received or were receiving disability insurance or workers' compensation at the time of data collection. We hypothesized that people who were receiving or had received workers' compensation or disability insurance would be more likely to have seen a physical therapist.
We also included a variable to indicate whether a subject had taken any legal action that was either pending or resolved for a spine-related condition. This variable was created on the basis of the responses to the following question: “What legal action, if any, are you considering for your spine-related symptoms?” Response categories were: (1) none; (2) I am considering an attorney; (3) my legal action is pending; (4) my legal action has been resolved, but not in my favor; and (5) my legal action has been resolved in my favor. We hypothesized that subjects who had taken legal action for their spine-related conditions would be more likely to have seen a physical therapist.
We also included 6 dichotomous variables to represent the subjects' past use of the following types of providers: (1) general practitioner or internist, (2) orthopedic surgeon, (3) neurosurgeon, (4) physiatrist, (5) rheumatologist, and (6) chiropractor. Because most insurance plans will reimburse only for physical therapy prescribed by a physician, physicians play an important role in access to and appropriate use of physical therapy. There are also data to suggest that the likelihood of physical therapy referral varies by physician specialty.40,44 We included chiropractors in this category because they are often the primary provider for people with spine problems.45 We hypothesized that physical therapist use would vary by physician specialty. We also hypothesized that subjects who saw a chiropractor would be less likely to see a physical therapist. Although we were unable to find any research on the similarities or differences between people who visit chiropractors and people who visit physical therapists, we reasoned that subjects who had allopathic physicians as their primary care providers would be more likely to be referred to physical therapists and, conversely, that subjects who had chiropractors as their primary care providers would be less likely to be referred to physical therapists.
Finally, we included a variable to represent the census region in which the spine care center was located. Since the seminal work of Wennberg and Gittelsohn in 1982,46 some studies46–48 have documented variations in health care use based on geographic locations and have attributed these variations to differences in the availability of health care resources or physician practice style. Therefore, census region served as a crude proxy for physical therapist availability or physician practice style, each of which can have an impact on physical therapist use. Because of data privacy issues, we were unable to obtain more specific information on the geographic locations of the centers. However, we did control for center effects in our analyses by clustering on center. Clustering on center accounts for the nonindependence of observations within center (ie, unobserved characteristics of subjects visiting a particular center are likely to be correlated).
Predisposing characteristics were represented by sex, age, race, and Hispanic ethnicity. Race was categorized as white, African American, or “other.” The category “other” included subjects who indicated that they were more than 1 race. We hypothesized that physical therapist use would be positively associated with female gender and age and would not be associated with race or ethnicity.
We also included a dichotomous variable to indicate whether an individual had previously used 1 or more of the following complementary care providers: acupuncturist, homeopath, or massage therapist. We included this variable as a predisposing characteristic because data suggest that the use of complementary care providers reflects particular values, beliefs, and attitudes toward life and health.49 Because some data suggest that people who have LBP and who use complementary care are more likely to use all types of medical care,24 we hypothesized that the use of complementary care providers would be positively associated with physical therapist use.
All analyses were conducted with Stata, version 8.0.* We first conducted a multiple logistic regression analysis to identify need, enabling, and predisposing characteristics associated with physical therapist use for the entire sample (N=29,049). Because we chose what we considered to be a parsimonious set of independent variables, we did not conduct any a priori analyses to assess collinearity.
Using the odds ratios (ORs) from this analysis, we calculated risk ratios (RRs) with the following formula50: where Po is the probability of the occurrence of the outcome (ie, physical therapist use) in people without the characteristic of interest. For example, for subjects with a diagnosis of herniated disk, the proportion of subjects who did not have this diagnosis and who saw a physical therapist was .47. Therefore, the RR was 1.33/[(1−.47)+(.47×1.33)]=1.15. This RR can be interpreted as follows: subjects with a diagnosis of herniated disk were 15% more likely to have seen a physical therapist than subjects with a diagnosis of sprain or strain. We chose to calculate RRs to assist with the interpretation of our results. Because it is not statistically appropriate to calculate confidence intervals for RRs computed in this manner, we refer the reader to the ORs and their 95% confidence intervals to assess the precision of the estimates.
We next conducted 2 separate analyses of a sample of subjects with neck pain only and no LBP (n=4,584) and of a sample of subjects with LBP only and no neck pain (n=18,202). The latter 2 analyses eliminated subjects for whom data on the locations of their problems were missing (n=3,930), who had a midback problem only (n=1,760), who had problems in the neck and low back (n=285), or who had problems in the neck, midback, and low back (n=288).
We used the cluster option in all models to control for the nonindependence of measures obtained from the same center.51 The cluster option specifies that observations are independent across groups or clusters (ie, centers) but not necessarily within groups. Specifically, this option corrects the SEs, which would tend to be smaller if not corrected, leading to more accurate parameter estimates. We also controlled for unreported data in all analyses by creating dummy variables to identify missing observations. For example, we created a variable (level not reported) that was coded “0” if data on the location of the problem were reported and “1” if data on the location of the problem were not reported. This variable was included as an independent variable in our analyses. With this approach, records with missing data on 1 or more variables are not eliminated from the analyses. After the logistic regression analyses were run, the fit of the models was assessed by running Hosmer-Lemeshow goodness-of-fit tests.52
The results of the logistic regression analysis of the entire sample are presented in Table 4. Seven of the need characteristics had RRs of 1.10 or more. Subjects with a diagnosis of herniated disk, spinal stenosis, spondylosis, or pain syndrome were 10% to 16% more likely to have seen a physical therapist than subjects with a diagnosis of sprain or strain. The duration of the problem also was positively associated with physical therapist use. Subjects with a problem lasting 3 months or more were 34% to 39% more likely to have seen a physical therapist than subjects with a problem lasting less than 3 months. Subjects who had previously received an injection also were more likely to have seen a physical therapist (RR=1.40) than subjects who had not previously received an injection.
Several of the enabling characteristics had RRs of ≥1.10 or ≤0.90. Subjects with more than 4 years of college education were 10% more likely to have seen a physical therapist than subjects with a high school education or less. Subjects who were receiving or had received workers' compensation or who had taken legal action also were more likely to have seen a physical therapist (RR=1.43 and RR=1.31, respectively) than subjects who had not received workers' compensation or who had not taken legal action. The variables representing previous use of allopathic physicians were all positively associated with physical therapist use, with RRs ranging from 1.14 to 1.44. Previous use of a chiropractor was not associated with physical therapist use. Geographic variations in physical therapist use also were present, with physical therapist use being 16% lower in the Midwest and 27% lower in the South than in the Northeast.
With regard to predisposing characteristics, male subjects were 13% less likely to have seen a physical therapist than female subjects, and subjects 50 years of age and older were 12% to 28% less likely to have seen a physical therapist than subjects 35 to 49 years of age. There was no association between race or ethnicity and physical therapist use. Subjects who had previously used 1 or more complementary care providers were 36% more likely to have seen a physical therapist than subjects who had not previously used complementary care providers.
Analyses of Subjects With Neck Pain or LBP
For several of the variables, the relationships with physical therapist use were similar for the subgroups of subjects with neck pain or LBP and followed the trends seen in the analysis of the entire sample. However, there were some differences. Diagnosis explained less of the variation in physical therapist use for subjects with neck pain than for subjects with LBP. Herniated disk was the only diagnosis with an RR of greater than 1.10 for subjects with neck pain. Diagnoses of herniated disk, spondylosis, spinal stenosis, and pain syndrome all had RRs of greater than 1.10 (RRs=1.14–1.20) for subjects with LBP. The duration of the problem also had a greater effect on physical therapist use for subjects with neck pain than for subjects with LBP. Compared with subjects with a spine problem lasting less than 3 months, those with neck pain lasting more than 1 year were 48% more likely to have seen a physical therapist, and those with LBP lasting more than 1 year were 33% more likely to have seen a physical therapist. With regard to previous physician use, the use of a rheumatologist was associated with physical therapist use for subjects with LBP only (RR=1.17), and previous use of a physiatrist had a stronger association with physical therapist use for subjects with neck pain than for subjects with LBP (RR=1.59 and RR=1.44, respectively). Variations in physical therapist use also were seen in the Midwest for the 2 subgroups. Subjects who had neck pain and who lived in the Midwest were 27% less likely to have seen a physical therapist than subjects who lived in the Northeast. Subjects who had LBP and who lived in the Midwest were 11% less likely to have seen a physical therapist than subjects who lived in the Northeast.
The P values for Hosmer-Lemeshow goodness-of-fit tests52 (computed from the chi-square distribution with df=8) were .15, .68, and .35 for analyses of the entire sample, subjects with neck pain, and subjects with LBP, respectively. P values greater than .05 indicated a good fit; that is, the model's estimates fit the data at an acceptable level.52
In an ideal health care system, and assuming that people's preferences are the same, need characteristics alone should explain variations in health care use. Theoretically, people with greater need (ie, poorer health) would use more care services than people with less need (ie, better health). When factors other than need explain variations in health care use, it can be questioned whether health care services are being used appropriately by all people (ie, services may be underused or overused by certain subgroups).
As we hypothesized and as has been reported in the literature,23–31 several need characteristics were associated with physical therapist use. For example, physical therapist use varied by diagnosis and was positively associated with measures of severity (eg, duration of the problem, previous history of surgery). Contrary to what we hypothesized, need characteristics were not the strongest predictors of physical therapist use in our models. Enabling characteristics as a group explained the greatest amount of variation in physical therapist use. One explanation for this finding is that we used very general measures of need. Need characteristics may have appeared to be the strongest predictors of physical therapist use had we included other, more specific measures of need, such as functional level and pain severity at the time when a subject saw a physical therapist. Unfortunately, this type of information was not available in the database.
With regard to enabling characteristics, previous use of a physiatrist was 1 of the strongest predictors of physical therapist use, with RR=1.44. Relative to previous use of other allopathic physicians, previous use of a physiatrist was the strongest predictor of physical therapist use. This finding may be related to the fact that physiatrists, of all physician specialists, probably have the most interaction with physical therapists and the greatest understanding of the care that they provide. In our subgroup analyses of subjects with neck pain or LBP, we found that previous use of a physiatrist was a stronger predictor of physical therapist use in subjects with neck pain than in subjects with LBP. Again, this finding may be related to a physiatrist's knowledge of and experience with physical therapy. Physical therapy for neck pain may not be considered as often by other physicians because neck pain is less prevalent53 and because we believe there is generally less evidence to support the use of physical therapy for the management of neck problems.
Because of the large sample analyzed in this study, some variables were statistically significant (P<.05) but had relatively small RRs. For example, for the general health variable, P=.001 and RR=0.96, indicating that subjects with a very good or excellent general health rating were 4% less likely to have seen a physical therapist than subjects with a general health rating of good, fair, or poor. Therefore, our interpretation of the results focused on RRs and not P values. What is considered a meaningful RR will vary depending on the outcome and the context of the study. For example, one may consider RR=1.05 (indicating a 5% greater risk) significant if the outcome is death. Because previous research on health care use offered us little guidance with regard to determining meaningful RRs and because data on factors associated with physical therapist use are very limited, we considered RRs that deviated 10% or more from 1.00 (ie, ≥1.10 or ≤0.90) to be worthy of discussion.
Research on factors associated with physician referral to physical therapists is limited. One study that examined factors associated with physician referral to physical therapists showed that orthopedic surgeons were more likely than general practitioners to refer subjects with musculoskeletal conditions to physical therapists.40 In our analyses, we found that the likelihoods of physical therapist use were similar for subjects who had previously seen an orthopedic surgeon and subjects who had previously seen a general practitioner. Contrary to what we hypothesized, we found no association between chiropractor use and physical therapist use. Although our hypothesis that chiropractor use would be negatively associated with physical therapist use was not supported, the fact that the variables indicating previous use of allopathic physicians were all positively associated with physical therapist use is notable.
Having received workers' compensation coverage and having taken legal action also were relatively strong predictors of physical therapist use, with RR=1.43 and RR=1.31, respectively. However, receiving or having received disability insurance did not increase the likelihood of physical therapist use. Our finding regarding a higher level of physical therapist use among subjects receiving workers' compensation has been reported in the literature.23,40,54
Education level, which we considered a proxy for socioeconomic status, was positively associated with physical therapist use. This finding is consistent with reports in the health services research literature on the use of other types of health care providers.18,19 People of a lower socioeconomic status often face barriers to receiving health care services. Mielenz et al,23 in an analysis of North Carolinians with acute back pain, also reported a positive association between education level and physical therapist use. We did not find any association between physical therapist use and race or ethnicity. These findings also agree with those of Mielenz et al.23 Although our results are encouraging because they suggest no racial or ethnic differences in physical therapist use for back or neck pain, the sample used in our analysis had a low representation of minority groups relative to 2000 census data.55 Census data for the year 2000 indicate that the US population was 12% African American and 12% Hispanic or Latino.
Even though we used a very gross measure for geographic location (ie, census region), we still found geographic variations in physical therapist use in all 3 models. Relative to the Northeast census region, the South and the Midwest showed lower levels of physical therapist use. Data from the 1997 Area Resource File indicate that physical therapist availability per 100,000 people is greater in the Northeast and West than in the Midwest and South56 and may be an explanation for this finding. What we are unable to determine from this finding is the clinically appropriate rate of physical therapist use. For example, physical therapists could be overused in the Northeast or underused in the South and Midwest. Our finding on geographic variations in physical therapist use is consistent with a body of literature that has reported geographic variations in the delivery of health care services even after accounting for illness severity. Some of the literature on geographic variations in health care use also suggests that variations are particularly great for more discretionary treatments (eg, elective surgery), for which strong scientific evidence on efficacy is lacking.20,21,44 For such treatments, physicians' preferences, attitudes, and past experiences may influence whether they offer it to their patients. With regard to spine problems, we believe that physical therapy can be considered a discretionary treatment for which strong scientific evidence on efficacy is lacking (ie, evidence obtained from randomized clinical trials). Therefore, some of the geographic variations in physical therapist use may be attributable to regional differences in physician practice styles.
With regard to predisposing characteristics, we found that being male and age were negatively associated with physical therapist use. These findings agree with those reported by Ehrmann-Feldman et al27 in a study of physical therapist use among Canadian workers. Data from the 2001 National Ambulatory Medical Care Survey57 also support our findings with regard to sex. Women generally make more ambulatory care visits than men, possibly because their attitudes and beliefs about medical care are different. One explanation for the negative relationship between physical therapist use and age is that age captures some of the unmeasured illness severity not captured by our need variables. This situation may be particularly true considering the limitations in our need variables.
We also found that previous use of other complementary care providers was positively associated with physical therapist use. Carey et al,24 in an analysis of care-seeking behaviors in North Carolinians with chronic LBP, reported similar findings. They found that people who used complementary care providers were more likely to be users of all types of conventional care providers.
Most studies that have examined health care use for the management of spine pain have focused on LBP. We chose to examine subjects with pain in any location (ie, neck, midback, low back). Although a majority of subjects had LBP, approximately 20% had neck pain. A secondary objective of our study was to determine whether there were differences in physical therapist use between subjects with LBP and subjects with neck pain. Our subgroup analyses did reveal some differences between the association of need characteristics (ie, diagnosis and duration of problem) and the association of enabling characteristics (ie, physician use and geographic location) with physical therapist use.
This study has several limitations. First, the generalizability of our results is limited to people who have back or neck pain and who visit spine care centers. These people may differ from the general population of people with back pain or neck pain, or both. Participation in the NSN database also is voluntary at the level of both the individual and the spine care center. Therefore, the data are not nationally representative and may be subject to nonresponse bias. Spine care centers and therefore subjects in the West are underrepresented. Most of the data also are from spine care centers affiliated with academic institutions. Private practice spine care centers may be underrepresented.
A second limitation is that the analyses were limited by the data available in the NSN database. More specific data on the severity of an individual's pain and level of function at the time when a physical therapist was seen likely would have improved the fit of our models. More specific information on enabling characteristics, such as income, insurance status (eg, Medicare, Medicaid, private, health maintenance organization), physical therapist availability, geographic location of the center, and other center characteristics, also likely would have improved the fit of our models. Finally, more specific information on people's preferences and beliefs likely would have improved the fit of our models. However, the goodness-of-fit tests indicated that our models fit the data adequately.
A third limitation is that the reliability and validity of some of the self-report data included in the NSN database have not been established. Some of the information supplied by patients, in particular, may be subject to recall bias. Missing data also can be problematic. We chose to retain as many data as possible and created dummy variables to indicate data that were missing. For all but 1 variable (ie, level of involvement), missing information represented less than 5% of the data. Although 13% of the data on the level of involvement were missing, we found that ORs and RRs for physical therapist use for subjects for whom level of involvement was not reported were similar to ORs and RRs for subjects for whom level of involvement was reported. This finding suggests that, with regard to physical therapist use, subjects who did not report level of involvement were similar to subjects who did.
Relevance of Findings
To our knowledge, this is the only US study that has attempted to identify need, enabling, and predisposing characteristics associated with physical therapist use for back or neck pain (both acute and chronic). We believe that the most important findings of this study relate to the enabling and predisposing characteristics. Although some of the findings related to these variables are similar to those of other studies,23,24,27 our findings stand on their own as new because of the dissimilarities between those studies and this study.
The fact that enabling and predisposing characteristics were associated with physical therapist use suggests that there may be inappropriate physical therapist use for the management of back or neck pain. The findings that we considered to be most relevant in this regard were the positive association between education level (a proxy for socioeconomic status) and physical therapist use, variations in physical therapist use by geographic location, variations in physical therapist use by physician specialty, and the lack of an association between race or ethnicity and physical therapist use. Because people of a lower socioeconomic status tend to be in poorer health,18,19 data indicating that they use fewer health care services than people of a higher socioeconomic status typically are considered to be suggestive of problems with access or underuse. Variations in physical therapist use by geographic location or physician specialty may be indicative of underuse or overuse. For example, we found that subjects who saw physiatrists were more likely to have seen physical therapists. This finding may indicate that physiatrists are referring appropriate people (ie, referring people who would benefit from physical therapy) and that other physicians are underreferring people (ie, not referring people who would benefit from physical therapy). Alternatively, this finding may indicate that physiatrists are overreferring people (ie, referring some people who would not benefit from physical therapy). Although the lack of racial or ethnic differences in physical therapist use implies that there may not be problems with access to physical therapists for racial or ethnic minority groups, this issue should be explored in future studies with nationally representative samples.
Our findings can serve as a point of departure for future studies examining issues related to the appropriate use of physical therapists for the management of neck or back pain. The ultimate goal of such studies is to provide information that can be used to ensure that all people in need of physical therapy are receiving it. One of the greatest challenges in improving the delivery of health care services and reducing variations in use is determining who is truly in need of care. Only then can it be determined whether variations in use are attributable to problems with access or underuse of services. Therefore, we believe that future studies should begin to determine the clinical characteristics of people who have back or neck pain and who would benefit from physical therapy. We believe that future studies also should continue to explore the roles of enabling and predisposing characteristics in physical therapist use (controlling for need characteristics) and should attempt to identify the underlying reasons for differences found.
The primary objective of this study was to identify factors associated with physical therapist use for the management of people with neck or back pain. In addition to need characteristics, several enabling and predisposing characteristics were associated with physical therapist use. Variations in physical therapist use explained by factors other than need suggest that there may be underuse of physical therapists (ie, people who would benefit from physical therapy are not receiving it) or overuse of physical therapists (ie, people who would not benefit from physical therapy are receiving it), or both.
All authors provided concept/idea/research design, writing, fund procurement, and consultation (including review of manuscript before submission). Dr Freburger and Dr Holmes provided data analysis. Dr Carey provided facilities/equipment and institutional liaisons. The authors thank Brett Hanscom, MS, and James Weinstein, DO, MS, for providing access to the National Spine Network data and for assisting with technical questions about the data set. The authors also acknowledge the National Spine Network and each of its members (Emory Spine Center; Kenton D. Leatherman Spine Institute; Rothman Institute; University of Iowa Hospitals and Clinics; Washington University Medical School; Hospital for Joint Diseases; University of California; Hospital for Special Surgery; Vanderbilt University Spine Center; University of Utah School of Medicine; Georgetown University Medical Center; Medical College of Wisconsin; Dartmouth–Hitchcock Spine Clinic; University Hospitals of Cleveland; Rush-Presbyterian–St Luke's Medical Center; UCSF Neuro-Spinal Service; SUNY Health Science Center; Spine and Scoliosis Surgery; Orthopedics and Scoliosis, Ltd; University Orthopaedics, Inc; Department of Orthopaedic Surgery, University of Pittsburgh Medical School; Washington University Medical School—Neurosurgery; Nebraska Spine Surgeons, PC; Tulane University Medical Center; University of Missouri—Neurosurgery; University of Wisconsin; University of Miami; William Beaumont Hospital; North Carolina Spine Center; Lakewood Orthopaedic Clinic; New England Baptist Bone and Joint Institute; and Providence Seattle Medical Center) for their support of this research.
This study was supported by a research grant from the Foundation for Physical Therapy and by the Agency for Healthcare Research and Quality (National Research Service Award Postdoctoral Traineeship sponsored by the Cecil G. Sheps Center for Health Services Research, grant T32-HS00032).
The results of this study were presented at the Combined Sections Meeting of the American Physical Therapy Association; February 4–8, 2004; Nashville, Tenn.
The analysis and any conclusions drawn from the data provided by the National Spine Network are the sole responsibility of the authors.
↵* Stata Corp, 4005 Lakeway Dr, College Station, TX 77845.
- Received September 15, 2004.
- Accepted March 9, 2005.
- Physical Therapy