Background Population-based studies on physical therapy use in acute care are lacking.

Objectives The purpose of this study was to examine population-based, hospital discharge data from North Carolina to describe the demographic and diagnostic characteristics of individuals who receive physical therapy and, for common diagnostic subgroups, to identify factors associated with the receipt of and intensity of physical therapy use.

Design This was a cross-sectional, descriptive study.

Methods Hospital discharge data for 2006–2007 from the 128 acute care hospitals in the state were examined to identify the most common diagnoses that receive physical therapy and to describe the characteristics of physical therapy users. For 2 of the most common diagnoses, logistic and linear regression analyses were conducted to identify factors associated with the receipt and intensity of physical therapy.

Results Of the more than 2 million people treated in acute care hospitals, 22.5% received physical therapy (mean age=66 years; 58% female). Individuals with osteoarthritis (admitted for joint replacement) and stroke were 2 of the most common patient types to receive physical therapy. Almost all individuals admitted for a joint replacement received physical therapy, with little between-hospital variation. Between-hospital variation in physical therapy use for stroke was greater. Demographic and hospital-related factors were associated with physical therapy use and physical therapy intensity for both diagnoses, after controlling for illness severity and comorbidities.

Limitations Data from only one state were examined, and the studied variables were limited.

Conclusions The use and intensity of physical therapy for stroke and joint replacement in acute care hospitals in North Carolina vary by clinical and nonclinical factors. Reasons behind the association of hospital characteristics and physical therapy use need further investigation.

In the past 2 decades, population-based databases, patient registries, and administrative health care databases have become much more common. These databases, which often are quite large, also have become more accessible. Advances in computer technology and more robust statistical software packages have significantly decreased the cost and computational time needed to analyze these large databases.

Despite the growth in and access to large health care databases, use of these databases to conduct research relevant to physical therapy has been modest. Much of the work has focused on the delivery of physical therapy in the outpatient setting.14 Some studies also have focused on physical therapy use by Medicare beneficiaries5,6 and patients in skilled nursing facilities.7 Few studies have been conducted on physical therapy use in the acute care setting.

We found 2 studies that examined University HealthSystem Consortium (UHC) data from 1996 to determine the relationship between physical therapy use in the acute care setting and outcomes of care for stroke and joint replacement.8,9 One limitation of these 2 studies is that the data were limited to academic health center hospitals that were members of the UHC. Findings from these studies may not be generalizable to other types of acute care hospitals. Data from these studies also are outdated considering the dramatic changes in Medicare and other health insurance plan payment policies over the past 15 years.

Current data on hospital inpatient stays in the United States are available to researchers through the Healthcare Cost and Utilization Project (HCUP). The HCUP is a family of health care databases, software tools, and other products developed through a federal-state-industry partnership sponsored by the Agency for Healthcare Research and Quality (AHRQ).10 The HCUP databases can be used to research a broad range of health policy issues, including the cost and quality of health care, patterns of health care use, access to health care, and outcomes of health care. The Nationwide Inpatient Sample (NIS), one of several HCUP databases, is the largest all-payer inpatient care database in the United States. Although the NIS is a powerful database for examining national trends in access to, use of, and quality of hospital inpatient care, its data are currently not detailed enough to examine physical therapy use in particular.

The State Inpatient Databases (SIDs), another set of HCUP databases, do have the detail needed to examine physical therapy use. The SIDs consist of state-specific hospital databases from 40 states.11 Each state database contains all inpatient discharge abstracts from community hospitals in the state, essentially making the database population-based. The SIDs contain clinical and nonclinical variables, including: diagnoses, procedures, admission and discharge status, patient demographics, insurance information, length of stay (LOS), and total charges. Many of the SIDs also include detailed information on hospital charges, which allows for the examination of physical therapy use and intensity. Some SIDs contain hospital, county, and ZIP code identifiers that allow linkage to other data.11

The SIDs have a number of strengths.12 They are relatively inexpensive to obtain and use when compared with the cost of similar data collected through surveys or medical record abstraction; they are more reliable than other sources of data, such as patient self-reporting of medical expenditures or diagnosis; they are superior to data obtained from third-party payers due to the inclusion of information on patients without insurance; they cover entire populations (ie, all acute care admissions in a state); and they often can be linked to other databases to gather additional information that may be useful. Studies also support the validity of hospital discharge data for accurately identifying diagnoses, procedures (eg, joint replacement), and patient comorbidities.1316

Population-based databases, such as the NIS and SIDs, can be used to answer basic questions about the prevalence of diseases and conditions that are treated by physical therapists. For example, one could determine the number of hospital admissions with diagnoses commonly treated by physical therapists (eg, stroke, hip fracture) and examine changes over time to better understand current and future workforce needs. Population-based databases also can be examined to identify the proportion of and types of individuals receiving physical therapy for a given diagnosis, condition, or disability to determine if there are variations in care (eg, by hospital, by state, over time), if patients are receiving appropriate care (eg, Are current practice guidelines being followed?), and whether there are potential problems with access to care (eg, Are people who are underinsured and minority populations less likely to receive physical therapy?). Information garnered from examinations of population-based health care databases can be used to identify areas that need improvement and to inform the development of initiatives or interventions to improve the delivery and quality of health care.

This study examined 2006 and 2007 data from the North Carolina SID to describe the demographic and diagnostic characteristics of individuals who receive physical therapy in North Carolina community hospitals, to identify common diagnostic subgroups, and to identify factors associated with the receipt of and intensity of physical therapy use.


North Carolina State Inpatient Database

Our primary source of data was the North Carolina SID (2006–2007), which contains discharge abstracts from the universe of all inpatient stays at short-term, acute care hospitals (N=128) in North Carolina. Data were retrieved from the UB-92 (Uniform/Universal Billing) claim form, which is used by hospitals and health care centers when submitting bills to Medicare and other third-party payers for reimbursement of health care services.11 The database contains one record for each hospital discharge. Data include the following sociodemographic information: age, race, sex; insurance information; geographic information: patient county, patient zip code, hospital county, hospital zip code; clinical information: diagnoses, procedures, LOS, source of admission (eg, home, emergency department, nursing home), discharge status (eg, home, skilled nursing facility); and charge data: revenue codes and charges for services provided (eg, intensive care unit charges, physical therapy charges), and total charges. Hospital identification numbers (ie, Centers for Medicare and Medicaid Services IDs; American Hospital Association IDs), as well as the county and zip code information, allow linkage to other data sets.

Hospital-Level Data

Data on hospital characteristics were obtained from the Centers for Medicare and Medicaid Services (CMS) Provider of Services (POS) Files,17 CMS Hospital Cost Reports,18 and the Area Resource File (ARF).19

POS Files.

CMS POS files are created quarterly from the Online Survey and Certification Reporting System (OSCAR) database. The OSCAR database is maintained by CMS in cooperation with state surveying agencies. State surveying agencies gather information from Medicare- and Medicaid-certified institutional providers (eg, hospitals, skilled nursing facilities, home health agencies) via survey and standard CMS forms, every 1 to 3 years, depending on institution type.20 Data collected include information on staffing, number of beds, type of facility, and services provided. These data are entered electronically into the OSCAR database by the state surveying agency and are used by CMS to certify or maintain certification of institutional providers. Data elements are checked during data entry for extreme values or extreme changes in values from the previous data collection year.20 The POS file (last quarter of 2006) was used to obtain data on hospital size (ie, number of beds), for-profit status, medical school affiliation, and physical therapist full-time equivalents (FTEs).

Hospital Cost Reports.

Hospital Cost Report data are obtained from the Healthcare Cost Report Information System (HCRIS) database maintained by CMS. Information in the HCRIS database are collected and supplied by Medicare fiscal intermediaries.21 Data from 2006 were used to determine whether the hospital used contract physical therapists and had an affiliated inpatient rehabilitation facility.

Area Resource File.

The ARF is a national, county-level health resource information database supported by the Department of Health and Human Services.22 We used data from the 2006–2007 ARF19 to indicate whether the hospital's county was located in a metropolitan or nonmetropolitan area.

Final Analytic Data Set

Our final analytic data set, created by merging the North Carolina SID data and the hospital-level data, consisted of 2,285,793 discharge records from 128 hospitals. Because this data set was a different subset for each study objective and because results of our analyses for objective 1 guided the analyses for objective 2, we present the remainder of the methods by study objective.

Objective 1: Demographic and Diagnostic Characteristics of Patients Who Received Physical Therapy

We first identified all records that had one or more physical therapy charges associated with them (n=512,018) (Fig. 1). Because there were more than 500 primary ICD-9-CM diagnostic codes associated with these records, we used AHRQ's Single-Level Clinical Classification Software (CCS)23 to group these codes into broader diagnostic categories. These diagnostic categories may be more useful for conducting descriptive and other types of analyses, particularly when examining a diverse clinical group. The specific ICD-9-CM codes associated with each CCS diagnostic category are presented in the CCS user's manual.23

Figure 1.

Sample development. PT=physical therapy.

Once the diagnostic categories were assigned, we tabulated these categories overall and by age categories. We also conducted descriptive analyses of the demographic and hospital-related characteristics of the sample. All analyses were conducted using Stata version 10.1 (StataCorp, College Station, Texas).

Objective 2: Factors Associated With Use and Intensity of Physical Therapy for Common Diagnostic Subgroups

Based on our analyses for objective 1, the 3 most common diagnostic categories for individuals who received physical therapy were osteoarthritis (OA), rehabilitation care, and acute cerebrovascular disease (Tab. 1). Because almost all individuals with a primary diagnosis of OA were in the hospital for a lower-extremity joint replacement, our first common diagnostic subgroup comprised patients with lower-extremity joint replacement. Any individual record with a diagnostic category of OA and a primary procedure of a hip or knee joint replacement or revision (ICD-9-CM procedure codes: 81.51–81.55) was placed in this subgroup (n=49,653).

Table 1.

Top 15 Agency for Healthcare Research and Quality Single-Level Clinical Classification Software (CCS) Diagnostic Categories Associated With Receipt of Physical Therapy, Overall and by Age Groupa

Rehabilitation care was the second most common diagnostic category (Tab. 1). Individuals in this diagnostic category had one of the following primary ICD-9-CM codes: V52.0, V52.1, V52.4, V52.8, V52.9, V53.8, V57.0, V57.1, V57.2, V57.21, V57.22, V57.3, V57.4, V57.81, V57.89, V57.9, and V58.82, which are associated with the fitting of prostheses and rehabilitation care use. The V-codes are supplementary and intended to indicate a reason for care in patients who have already been treated for a disease or injury not currently present, who have residual impairments, or who may be receiving care to prevent recurrence.24 Because individuals in this category represent a heterogeneous group in regard to underlying diagnosis, we chose cardiovascular accident (CVA) (the next most common diagnostic subgroup) as our second diagnostic subgroup.


To conduct analyses examining factors associated with physical therapy use, we needed to include participants in the same diagnostic categories who did not receive physical therapy. Our final samples for the joint replacement and stroke subgroups were n=49,653 and n=26,422, respectively (Fig. 1).

Study variables.

Based on clinical experience and the body of literature on health care utilization, we hypothesized that individual and hospital-level factors would affect whether and the extent to which individuals received physical therapy in the acute care setting. For each diagnostic subgroup, we generated 2 dependent variables: (1) a dichotomous variable to indicate whether the individual received any physical therapy and (2) a measure of “physical therapy intensity” for those who did receive physical therapy. Because information on number of visits was not available, we used data on physical therapy charges to create our measure. In preliminary analyses we found that charging practices varied across hospitals and that hospitals with high total charges also had high physical therapy charges. We, therefore, chose to standardize our measure of physical therapy intensity by dividing physical therapy charges by total charges. This measure also accounts for the fact that charges often are not reflective of “true costs” (eg, what is paid by the insurer or what it costs the hospital to deliver the care). Because our measure of physical therapy intensity (total physical therapy charges/total charges) was derived, we conducted sensitivity analyses to determine how our results changed using total physical therapy charges and physical therapy charges per day as measures of physical therapy intensity.

Independent variables included patient-level demographic characteristics: age, sex, race (white, non-white, missing), and insurance (Medicare/private/other, Medicaid, or uninsured); and measures of illness severity: admission through the emergency department, LOS (categorized by tertiles because of the skewed nature of the data), and comorbidities. We created our comorbidity measures based on the work of Elixhauser et al,25 who developed a comprehensive set of 30 comorbidity measures for use with large administrative inpatient databases. These measures were created using secondary ICD-9-CM diagnostic codes and include comorbidities such as diabetes, hypertension, congestive heart failure, and so on. A table of the codes associated with each of the comorbidity measures is available on AHRQ's Web site.26 Because Elixhauser and colleagues found the independent effects of each of the comorbidity measures varied by patient population, they suggested using the comorbidity measures as separate indicators in analyses, rather than creating a count.

Rather than including all 30 comorbidities, we limited our list to those that had support for inclusion based on the literature, those that occurred with a frequency of at least 2%, and those that were related to our outcomes in preliminary analyses. For the joint replacement data, we included the following indicators (0=no, 1=yes): diabetes (with or without complications), obesity, chronic pulmonary disease, rheumatoid arthritis, neurological disorder or paralysis, and congestive heart failure2729; and for the stroke data: congestive heart failure, diabetes with complications, depression, psychoses, renal failure, and chronic obstructive pulmonary disease.3033

We also included condition-specific variables for both data sets. For the joint replacement data set, we created dichotomous variables to indicate whether the procedure was a revision or a hip replacement (versus knee replacement) and whether the individual had a hip fracture.34 For the stroke data set, we created dichotomous variables to indicate whether the stroke was hemorrhagic and whether the individual had atrial fibrillation.31,35 The condition-specific variables were created by examining the ICD-9-CM procedure and diagnostic codes.

Hospital variables included hospital bed size (categorized by tertiles), which served as an indirect measure of procedure or diagnosis volume (both stroke and joint replacement volume have been shown to be positively associated with higher quality of care3639); whether the hospital had a major medical school affiliation, an additional indicator of higher-quality care40,41; urban or rural location of hospital; whether the hospital had an affiliated rehabilitation facility; the use of contract physical therapists; and physical therapist FTEs per 1,000 admissions (categorized by tertiles). The latter 2 variables represented measures of physical therapist supply. We also included a variable to indicate the for-profit status of the hospital, as literature suggests that patient outcomes and incentives at for-profit hospitals differ from those at other hospitals.36,42

Data Analysis.

For both the stroke and joint replacement samples, we conducted multivariable logistic regression analyses to identify determinants of physical therapy use and multivariable linear regression analyses to identify factors associated with physical therapy intensity (physical therapy charges/total charges), conditional on receiving physical therapy. Due to the skewed distribution of physical therapy charges/total charges, we transformed this variable by taking its natural log. Our multivariable analyses were limited to individuals who survived their inpatient stay and to hospitals that had at least 10 admissions for stroke or joint replacement over the 2-year period (n=32,139 for the stroke sample; n=49,684 for the joint replacement sample). When conducting our multivariable analyses, we used the cluster option in Stata to account for the fact that measures from the same hospital were not independent and were likely more highly correlated than measures from different hospitals.43 The cluster option uses the Huber-White sandwich estimator of variance to correct the standard errors of the parameter estimates (eg, the odds ratio in the logistic regression analysis or the beta coefficient in the linear regression analysis), which may be smaller due to the correlation of measures within hospitals. Not correcting the standard errors can lead to the conclusion that findings are statistically significant when they are not.

Less than 0.1% of the records (n=907 for joint replacement sample; n=511 for stroke sample) had missing insurance data. These records were dropped from the multivariable analyses. Because a large percentage of race data were missing, we created a dummy variable to identify those records with missing data. This variable allowed us to retain those records, which otherwise would have been dropped in the multivariable analyses. No other variables had missing values.

Role of the Funding Source

This study was funded, in part, by the Division of Physical Therapy, University of North Carolina at Chapel Hill.


Of the more than 2 million patients discharged from short-term, acute care hospitals in North Carolina in 2006–2007, 22.4% received physical therapy during their inpatient stay.

Common Diagnostic Categories

Table 1 presents the 15 most common diagnostic categories associated with physical therapy use for the entire sample and by age category. Osteoarthritis was the most common category overall and for individuals older than 45 years of age. Leg fractures were the most common diagnostic category for individuals 1 to 45 years of age. Rehabilitation was the second or third most common category overall and for all age categories.

The 5 most common secondary diagnoses associated with individuals who had rehabilitation as the primary diagnostic category were: V54.81 aftercare after joint replacement, V599.0 urinary tract infection, V427.31 atrial fibrillation, V438.20 late effects of cerebrovascular disease, and V428.0 congestive heart failure. These 5 diagnoses accounted for 20% of the records.

Characteristics of Physical Therapy Users

Demographic and hospital-related characteristics for physical therapy users and for individuals who had a stroke or joint replacement are presented in Table 2. Physical therapy users had a mean age of 66 years, 58% were female, and 42% were white, although race data were missing for almost half of the sample. A majority of the sample was 65 years or older and had Medicare insurance. Median total physical therapy charges were $1,121 for all physical therapy users, which represented approximately 5% of the total charges. Almost 50% of the sample was admitted to the hospital through the emergency department, and more than half of the sample was discharged to post-acute care (ie, home health, rehabilitation facility, skilled nursing facility). Differences in demographic and hospital-related characteristics by diagnosis were present.

Table 2.

Sample Characteristicsa

Variation of Physical Therapy Use by Hospital

Figure 2 presents data on variation in physical therapy use by hospital for the stroke and joint replacement samples. There was considerable between-hospital variation in use of physical therapy for stroke. The mean (SD) and median (interquartile range [IQR]) percentages of patients with stroke who received physical therapy at each hospital (n=122) were 71% (12%) and 73% (66%–79%), respectively. In contrast, there was very little between-hospital variation in the use of physical therapy for joint replacement. The mean (SD) and median (IQR) percentages of patients who received physical therapy at each hospital (n=93) were 98% (2%) and 99% (99%–100%), respectively. Hospitals with percentages lower than 90 (n=7) had fewer admissions overall. Three of these hospitals had 10 or fewer joint replacement admissions over the 2-year period.

Figure 2.

Percentage of patients receiving physical therapy (PT) by hospital. Each circle represents a hospital.

Factors Associated With the Receipt and Intensity of Physical Therapy

Table 3 presents results of our multivariable regression analyses on physical therapy use for stroke. Factors positively associated with physical therapy use (odds ratios significantly greater than 1.00) included older age, being on Medicaid insurance relative to Medicare/private/other, being uninsured relative to Medicare/private/other, having an emergency department admission, having an LOS greater than 4 days, or being seen at a hospital with a major medical school affiliation. Having congestive heart failure, atrial fibrillation, or a hemorrhagic stroke was negatively associated with physical therapy use (odds ratios significantly less than 1.00).

Table 3.

Factors Associated With Use of Physical Therapy for Strokea

Of those who received physical therapy, intensity of physical therapy use was higher (beta coefficients positive and significantly greater than 0) for older individuals, individuals on Medicaid relative to Medicare/private/other, individuals with an LOS from 4 to 6 days, relative to those with an LOS less than 4 days, and individuals with depression. Intensity of physical therapy use was lower (ie, beta coefficients significantly less than 0) for females and individuals with hemorrhagic stroke, renal failure, or chronic obstructive pulmonary disease.

Factors positively associated with physical therapy use for joint replacement (Tab. 4) included longer LOS, being seen at larger hospitals, and being seen at hospitals with major medical school affiliations. Factors negatively associated with physical therapy use were having a revision procedure, hip fracture, or hip replacement (relative to knee); and being seen at a hospital that used contract physical therapists. Of particular note is the imprecision of some parameter estimates (ie, LOS variables, medical school affiliation variable). This imprecision was because most people received physical therapy, which led to small cell sizes for some of the cross tabulations (eg, physical therapy use by medical school affiliation).

Table 4.

Factors Associated With Use of Physical Therapy for Joint Replacementa

Of those who received physical therapy, older individuals, females, and those with a longer LOS received a higher intensity of physical therapy. The presence of comorbidities, having a hip fracture, having a revision procedure, or having a hip replacement was associated with a lower intensity of physical therapy. Individuals seen in urban hospitals received higher-intensity physical therapy, whereas those in for-profit hospitals or hospitals that used contract physical therapists received lower-intensity physical therapy.

Sensitivity Analyses

Findings of our multivariable linear regression analyses using physical therapy charges or physical therapy charges per day (rather than physical therapy charges/total charges) were similar, with the exception of one variable: urban location of the hospital. In the stroke models, urban location became significant. In the joint replacement models, urban location became nonsignificant.


To our knowledge, this is one of the first population-based studies examining physical therapy use in acute care. We have reported on a number of findings related to the characteristics of patients who receive physical therapy in acute care settings, as well as determinants of physical therapy use and intensity for individuals with stroke or lower-extremity joint replacement. Approximately one quarter of patients admitted to short-term, acute care hospitals in North Carolina receive physical therapy. These users are older, with a majority insured by Medicare. Osteoarthritis and stroke were 2 of the most common diagnostic categories to receive physical therapy, which is consistent with the literature on the growth in joint replacement procedures44,45 and the incidence of stroke.46 Both are common reasons for acute care hospitalizations.

Our analyses of determinants of physical therapy use for joint replacement and stroke provide an opportunity to compare and contrast findings on 2 very different problems. Joint replacement often is an elective procedure, with patients having a choice of physicians and hospitals. Stroke, on the other hand, is an event that occurs unexpectedly. Where the patient is (ie, geographically) at the time of the stroke will greatly affect where he or she is admitted.

On average, 98% of individuals who were admitted for a joint replacement procedure in North Carolina hospitals received physical therapy. There also was little variation in physical therapy use across hospitals. Although we do not understand fully the effectiveness of physical therapy in the acute care of patients after joint replacement, particularly in regard to longer-term outcomes, some studies suggest that starting rehabilitation as soon as possible after surgery leads to better outcomes.4749 Most current guidelines for postoperative, acute care after joint replacement also recommend physical therapy.50 In addition, patients and health care providers believe physical therapy is useful in the acute care setting.51 The fact that most individuals who undergo joint replacement in North Carolina hospitals receive physical therapy is consistent with current best evidence and guidelines.

On average, 70% of patients admitted to North Carolina hospitals for treatment of stroke received physical therapy. Between-hospital variation in the percentage of patients who received physical therapy was greater than for joint replacement. Although a shorter LOS was associated with a decreased likelihood of physical therapy use, even when we limited our sample to individuals who survived their inpatient stay and were in the hospital for at least 3 days, only 85% of the sample received any physical therapy. When looking at the percentage of individuals who received physical therapy by hospital, 5 of the hospitals had proportions below 70%.

As with joint replacement, our understanding of the effectiveness of physical therapy for the acute care of patients with stroke is limited. Several current stroke guidelines recommend mobilization and assessment by rehabilitation professionals as soon as possible after an acute care admission.5255 Recommendations of the Northeast Cerebrovascular Consortium are that all patients admitted to the hospital for stroke be evaluated by a physical therapist and an occupational therapist.56 Incorporating rehabilitation into stroke unit care also has been shown to be one of the key components in disability reduction after stroke.57 Considering current evidence and guidelines, variation in the use of physical therapy in the acute care setting needs further exploration and may be indicative of suboptimal care.

Although demographic characteristics were not associated with physical therapy use for joint replacement, we found older age, being on Medicaid, and being uninsured associated with physical therapy use for stroke. One explanation for these findings is these variables are capturing unmeasured illness severity. Individuals who are older have more comorbidities, complications, and problems in response to a CVA and have greater functional limitations.58,59 People with Medicaid coverage or no insurance, proxies for lower socioeconomic status, have poorer health relative to individuals of higher socioeconomic status.6062

In both the stroke and joint replacement models, several of the illness severity and comorbidity measures were associated with physical therapy use in the expected directions. For example, individuals with a hemorrhagic stroke and those who had a joint replacement after hip fracture were less likely to receive physical therapy. These 2 variables are proxies for greater illness severity.34,62

With respect to hospital characteristics, we found that individuals who were seen at hospitals with major medical school affiliations (ie, academic health centers) were more likely to receive physical therapy for stroke and joint replacement. Assuming the receipt of physical therapy for these conditions is appropriate and necessary (considering current evidence and guidelines), this finding supports other literature that suggests, at least for some aspects of care, academic health centers deliver higher-quality care.40,41

We also found that hospital size was associated with use of physical therapy for joint replacement. The reasons for this finding are unclear. Although imprecise, hospital size may be a proxy measure for joint replacement procedure volume. Larger hospitals are likely to perform more procedures than smaller hospitals. In some of our work with other hospital databases, we found joint procedure volume and hospital size were moderately correlated (r=.60). Studies have shown a positive relationship between joint replacement procedure volume and quality of care.37,38 Whether use of physical therapy is more likely at hospitals with higher procedure volumes needs further exploration, along with the relationship between physical therapy use and quality of care for joint replacement.

In both the stroke and joint replacement models, individuals seen at hospitals who used contract physical therapists were less likely to receive physical therapy. This may suggest potential physical therapy supply issues. Physical therapy FTEs per admissions, however, were not significantly associated with physical therapy use in any of the models. The reasons for this finding are unclear. Further exploration of the accuracy of the measure may provide some insight. Although there is a growing body of literature on the relationship between higher nurse staffing and better quality of care,63,64 little has been done in the area of rehabilitation staffing. Such information would be useful for determining optimal staffing levels for physical therapy.

In the models that examine factors associated with the intensity of physical therapy use, females with stroke received lower-intensity physical therapy, whereas females with joint replacement received higher-intensity physical therapy. These findings need further exploration to determine whether these differences are indicative of sex disparities or potential confounding. For example, data suggest that females delay joint replacement and, therefore, are more disabled than males preoperatively.65

In both models, for-profit status of the hospitals was associated with lower-intensity physical therapy use. This finding also warrants further exploration to determine whether it is indicative of differences in care delivery or differences in patient populations (eg, patients at for-profit hospitals tend to be healthier).

Although urban location was not associated with physical therapy intensity for the stroke sample, it was for the joint replacement sample. In our sensitivity analyses using other measures of physical therapy intensity, we found the effect of urban location changed, suggesting that charging practices at urban and nonurban hospitals may vary. These findings also lend support to our choice of physical therapy charges/total charges as our measure of physical therapy intensity.


This study has several limitations. First, it is based on data from one state, which limits its external validity. The analyses also were limited by the type and availability of data. We did not have measures of function. Instead, we used proxy measures of illness severity and comorbidities. Other measures of illness severity, such as intensive care unit use66 or APR-DRG (all patient refined diagnosis related groups) measures of severity and mortality, may have been useful. The APR-DRGs expand the traditional diagnosis related group classification by adding a 4-level measure of illness severity and a 4-level measure of mortality risk.67 In addition, race data were missing for almost 50% of the sample, which may have been the reason for the generally nonsignificant findings on race. National efforts are under way to improve the reporting of race/ethnicity data in health care databases such as the SIDs.68 Another limitation was the manner in which we assessed physical therapy intensity. A measure of physical therapy costs, adjusted for factors such as hospital location, may have been better. Although not used in our study, cost-to-charge ratios, which take into account geography, have been developed from CMS data and can be used to convert hospital charge data to costs.69

Clinical Implications

Hospital discharge data can be used to identify the extent to which physical therapy is used in acute care and factors associated with use. Such information is helpful for determining whether patients are receiving appropriate care and whether there are indications of disparities in care. Variation in care also can be examined and may identify potential areas of overuse of care (ie, instances in which health care resources and procedures are used even when there is no reason or evidence to suggest they are the best way to help a patient)70 or underuse of care (ie, instances in which evidence-based, effective health care practices are not used).70 Our findings suggest that hospital characteristics influence whether individuals receive physical therapy and the intensity of their physical therapy, particularly for patients with joint replacement. Future investigations should focus on a better understanding of the reasons for this variation and the influence of hospital characteristics.

Although hospital discharge data have been used in numerous studies examining health care use and outcomes of care,12 little has been done with these data in the area of rehabilitation care. State Inpatient Databases from several other states are available for minimal to moderate fees through HCUP. Although not all SIDs have detailed information on physical therapy use, efforts are under way to improve this situation. In 2006, the AHRQ developed a new data element to indicate physical therapy use based on information from the UB-92 revenue.71 Looking at the SIDs available for purchase in 2008 (N=26), 19 of the states captured the physical therapy utilization data element. As the availability of this data element increases, it also may be represented in HCUP's NIS database.72 Having information on physical therapy use in the NIS would facilitate understanding of physical therapy use in acute care at the national level.


We used a population-based data set to describe the physical therapy use in North Carolina acute care hospitals. Osteoarthritis associated with lower-extremity joint replacement and CVA were 2 of the most common diagnostic categories that received physical therapy. Almost all patients admitted for joint replacement received physical therapy, with little between-hospital variation. Use of physical therapy for patients with stroke was more varied. Both clinical and nonclinical factors were associated with the use and intensity of physical therapy for joint replacement and stroke.


  • Dr Freburger and Dr Montmeny provided concept/idea/research design and writing. Dr Freburger provided data collection, project management, fund procurement, and institutional liaisons. All authors provided data analysis. Ms Knauer provided clerical support. Ms Heatwole Shank and Ms Knauer provided consultation (including review of manuscript before submission).

  • This research, in part, was presented orally at the Combined Sections Meeting of the American Physical Therapy Association; February 17–20, 2010; San Diego, California.

  • This study was funded, in part, by the Division of Physical Therapy, University of North Carolina at Chapel Hill.

  • Received October 15, 2010.
  • Accepted August 15, 2011.


View Abstract