Utilizing Patient-Specific Factors to Predict Costs in Home-Based Medicare Part B Outpatient Physical Therapy

Abstract

The US health care system faces rising costs related to population aging, among other factors. One aspect of the high costs related to aging is Medicare outpatient therapy expenditures, which in 2010 totaled $5.642B for ∼4.7 million beneficiaries. Given the magnitude of these costs and the need to maximize value, this study developed and tested a predictive model of outpatient therapy costs. Retrospective analysis was performed on electronic medical record data from October 31, 2014–September 30, 2016 for 15,468 Medicare cases treated by physical therapists associated with a large, national rehabilitation provider. The analysis was a multiple linear regression of cost per case by 27 predictor variables: age group, sex, recent hospitalization, community vs. facility residence, the 10 states served, time from admission to initial evaluation, initial functional limitation reporting level, functional limitation reporting category, and 9 chronic conditions. The model was designed to be predictive and includes only variables available at the start of a case. The model was statistically significant (P < .0001) but explained only 7.4% of the variance in cost. Of the predictor variables, 16 had statistically significant effects. Those most highly predictive included state in which service was provided (8 of the 16 effects), and 3 variables indicating physical functioning at initial evaluation (initial functional limitation category and level, and residence in community vs. facility). There is need for more research focusing on the effects of specific types of treatment, and also for a more proactive model for outpatient therapy reimbursement that emphasizes prevention as well as treatment.

Introduction

The US health care system is ranked eleventh in outcomes and first in cost worldwide. In addition, costs are continuing to increase.¹ A “silver tsunami” (ie, population aging) is projected to result in 25 million more Medicare beneficiaries by 2036, further straining the health care system as it assumes responsibility for long-term care needs and costly chronic conditions.² Hospital costs for older adults aged 65 and older represent 41% of total hospital costs and the percentage is rising.³ More specifically, Medicare outpatient therapy expenditures for calendar year 2010 for 4,697,349 beneficiaries totaled $5.642B, with $4077B, or 72.3%, for physical therapy, $1089B for occupational therapy, and $476M for speech/language therapy.⁴ Given the magnitude of such costs and the need to maximize value, the objective of this research is to develop and test a predictive model of outpatient therapy costs.

Methods

Sample

A retrospective analysis was completed on electronic medical record (EMR) data from October 31, 2014 – September 30, 2016 for 15,468 Medicare patient cases treated by physical therapists associated with FOX Rehabilitation, Cherry Hill, NJ. These 15,468 patient cases were drawn from a set of records (N = 38,201) using the following inclusion/exclusion criteria: (1) the case included initial and discharge functional limitation reporting level and category; (2) the patient achieved the goal of “no skill required” at discharge, which was a selection within the EMR system and is a proxy for progression to self-management; (3) the patient received 4 or more visits; and (4) the patient was seen at an average visit frequency of ≥1.5 visits per week (Table 1).

Table 1.

Sample Attrition

Initial data set	N = 38,201
Functional Limitation Reporting required at both IE and DC	N = 21,367
Must achieve self-management	N = 17,868
Must have ≥4 visits	N = 17,506
Must have ≥1.5 visits per week	N = 15,468

DC, discharge; IE, initial evaluation.

These exclusion criteria resulted in a sample that includes all required variables and included only patients who were motivated and adherent to their plan of care. Many, if not all, of the cases excluded as not having an initial and discharge functional limitation reporting level and category were unexpected discharges that did not have discharge reports. Having a total number of visits less than 4 would be unlikely in the population treated by FOX clinicians unless the plan of care was completely compensatory, which is not of interest in this study. A frequency of ≤1.5 visits per week would not meet the standard of care at FOX Rehabilitation, which is drawn from current best practice related to rehabilitation of the older adult. Including only patients who achieved self-management is an objective criterion with face validity.

Data

The outcome variable was total cost per case. The candidate predictor variables included: age group (<60, 60–64, 65–69, 70–74, 75–79, 80–84, 85–89, and ≥90 years); sex; recent hospitalization (yes/no); place of service (home/community vs. facility); the 10 states served by FOX at the time (Connecticut, Delaware, Massachusetts, Maryland, New Hampshire, New Jersey, New York, Pennsylvania, South Carolina, and Virginia); time from admission to initial evaluation; initial functional limitation reporting level; the 4 functional limitation reporting categories present in the physical therapy specific data set (Mobility [moving by changing body position or location or by transferring from one place to another, by carrying, moving, or manipulating objects, by walking, running or climbing, and by using various forms of transportation]; Body Position [getting into and out of a body position and moving from one location to another, such as getting up out of a chair to lie down on a bed, and getting into and out of positions of kneeling or squatting. Inclusions: changing body position from lying down, from squatting or kneeling, from sitting or standing, bending and shifting the body's center of gravity]; Carrying [raising up an object or taking something from one place to another, such as when lifting a cup or carrying a child from one room to another. Inclusions: lifting, carrying in the hands or arms, or on shoulders, hip, back, or head; putting down]; and Other); and 9 chronic conditions (Parkinson's disease [PD], congestive heart failure, chronic obstructive pulmonary disease, cerebrovascular accident (CVA), diabetes, total hip arthroplasty, total knee arthroplasty, dementia, and depression). Table 2 displays frequency distributions and summary statistics of these variables.

Table 2.

Frequency Distributions and Summary Statistics for All Variables

Continuous variables
Cost per case	Mean = $3308.12, sd = $1757	Median = $3044.55
Time from Admission to Initial Evaluation (Days)	Mean = 10.34, SD = 12.41	Median = 6.00

Variable/Value	Frequency	Percent
Age group, years
<60	388	2.51
60–64	230	1.49
65–69	688	4.45
70–74	1120	7.24
75–79	1797	11.62
80–84	3109	20.10
85–89	4100	26.51
≥90	4036	26.09
Sex
Female	10,832	70.03
Male	4636	29.97
Recent hospitalization
Yes	2987	19.31
No	12,481	80.69
State
Connecticut	726	4.69
Delaware	794	5.13
Massachusetts	80	0.52
Maryland	305	1.97
New Hampshire	10	0.06
New Jersey	5752	37.19
New York	2793	18.06
Pennsylvania	3717	24.03
South Carolina	559	3.61
Virginia	732	4.73
Place of Service
Community	9646	62.36
Facility	5822	37.64
Functional Limitation Reporting at Initial Evaluation
0% impaired, limited, or restricted	6	0.04
1%–19%	98	0.63
20%–39%	1287	8.32
40%–59%	4330	27.99
60%–79%	5916	38.25
80%–99%	3522	22.77
100% impaired/limited/restricted	309	2.00
Functional Limitation Reporting Category
Mobility	14,019	90.63
Body Position	1276	8.25
Carrying	136	0.88
Primary Other	37	0.24
Chronic condition
Parkinson's disease	1410	9.12
Congestive heart failure	961	6.21
Chronic obstructive pulmonary disease	742	4.80
Cerebrovascular accident	1564	10.11
Diabetes	2653	17.15
Total hip arthroplasty	783	5.06
Total knee arthroplasty	959	6.20
Dementia	4518	29.21
Depression	2704	17.48

SD, standard deviation.

Analysis

The main analysis was a multiple linear regression regressing cost per case on the 27 candidate predictor variables described. The study team also conducted 1-way analyses of variance (ANOVAs) to examine the nature of the effects on total cost per case of age group and functional limitation reporting level.

The authors received approval for this study from the Thomas Jefferson University Institutional Review Board.

Results

The FOX Rehabilitation patient population is predominantly female (70.0%) and older, with only 4.0% younger than age 65, 46.6% in their 80s, and 26.1% in their 90s (Table 3). Roughly one fifth (19.3%) experienced a recent hospitalization before their rehabilitation case. More than one third, 37.6%, were in a facility, with facility being defined as senior living community, assisted living, or adult day care, rather than the community. An average latency of 10.34 days occurred between admission and evaluation. Of 10 states of residence, those with the largest patient populations were New Jersey (5752), Pennsylvania (3717), and New York (2793). The most common initial functional limitation reporting level was 60%–79% impaired (38.2%), followed by 40%–59% impaired (28.0%), and 80%–99% impaired (22.8%). Of the 4 functional limitation reporting categories reported in the data, the most common by far was Mobility (90.6%), followed distantly by Body position (8.25%). There were very few cases with Carrying (0.88%) or Other (0.24%). Finally, of the 9 chronic conditions, the most common were dementia (29.2%), depression (17.5%), and diabetes (17.2%).

Table 3.

Multiple Linear Regression Model: Cost Per Case by Candidate Predictor Variables


Overall Model	df = 28	F = 45.36	<.0001
R²	7.4%
Adjusted R²	7.2%

Variable	DF	Estimate	SE	t	P
Parameter Estimates
Intercept	1	2446.34	147.51	16.58	<.0001	^***
Age group	1	−1.98	1.62	−1.22	0.2221
Sex: male vs. female	1	56.62	29.74	1.9	0.057
Recent hospitalization	1	112.66	34.17	3.3	0.001	^***
Facility vs. Community	1	287.81	31.81	9.05	<.0001	^***
Connecticut	1	1006.99	66.28	15.19	<.0001	^***
Delaware	1	−54.15	63.25	−0.86	0.392
Massachusetts	1	439.17	187.74	2.34	0.0193	^*
Maryland	1	947.82	98.03	9.67	<.0001	^***
New Hampshire	1	1240.82	525.97	−2.36	0.0183	^*
New York	1	588.34	38.96	15.1	<.0001	^***
Pennsylvania	1	−104.98	36.04	−2.91	0.0036	^**
South Carolina	1	949.32	74.09	12.81	<.0001	^***
Virginia	1	341.87	65.29	5.24	<.0001	^***
Time from admission to IE	1	−6.32	1.12	−5.67	<.0001	^***
Functional Limitation Reporting at IE	1	253.22	14.08	17.98	<.0001	^***
Body Position	1	−92.19	49.20	−1.87	0.061
Carrying	1	−701.50	143.48	−4.89	<.0001	^***
Other	1	−471.44	274.02	−1.72	0.0854
Parkinson's disease	1	165.65	47.43	3.49	0.0005	^***
Congestive heart failure	1	−63.72	56.33	−1.13	0.258
Chronic obstructive pulmonary disease	1	33.68	62.97	0.53	0.5927
Cerebrovascular accident	1	241.17	45.03	5.36	<.0001	^***
Diabetes	1	37.97	35.74	1.06	0.2881
Total hip arthroplasty	1	15.32	62.21	0.25	0.8054
Total knee arthroplasty	1	−30.84	56.58	−0.55	0.5857
Dementia	1	−192.69	31.71	−6.08	<.0001	^***
Depression	1	8.57	36.75	0.23	0.8156
^*** P < .001
^** P < .01
^* P < .05

IE, initial evaluation; SE, standard error.

The regression model was designed to be predictive and thus only includes variables available at the start of a case. The overall model (Table 3) was statistically significant (P < .0001), but it explained only a modest proportion of the variance in total cost (7.4%). Of the 27 candidate predictor variables included in the model, 16 had statistically significant effects. Demographic variables, age group or sex, did not have a significant effect on total cost in this sample. Higher cost was predicted by both recent hospitalization ($113 higher) and living in a facility ($288 higher). Comparing each of the other 9 states FOX serviced with New Jersey (the state with the highest volume of cases), 8 out of 9 (all but Delaware) were significantly different in one direction or the other: 6 states had higher total cost than New Jersey (Connecticut, Massachusetts, Maryland, New York, South Carolina, and Virginia). Two states had lower total cost than New Jersey: New Hampshire and Pennsylvania (Table 3). Time from admission to initial evaluation predicted lower total cost. A higher level of disability predicted higher total cost, by $253 per 1-level rise in functional limitation reporting. Of the 3 functional limitation reporting categories compared to Mobility, only Carrying significantly predicted total cost. Finally, of the 9 chronic conditions, the 3 that significantly predicted total cost were PD, CVA, and dementia (Table 3).

Two sensitivity analyses were run to assess whether high outliers in total cost might bias the regression model results: first excluding the top 1% of the cases by total cost (≥$9312) and then excluding the top 5% of the cases by total cost (≥$6531). There was very little change: while the original model had an R² of 7.4%, the other models accounted for only slightly less of the variance in total cost – 7.1% for the model excluding the top 1% of cases, and 6.8% for the model excluding the top 5% of the cases.

One-way ANOVAs also were run (Table 4) for variables with 3 or more categories (ie, age group and functional limitation reporting level). Although the effect of age group is not statistically significant in the multiple regression model, the trend is direct and linear (ie, increasing age is associated with higher cost). The effect of functional limitation reporting on cost is highly statistically significant, and linear except for the reversal from functional limitation reporting level 5 to 6.

Table 4.

One-Way Analysis of Variance of Cost by Age Group and Functional Limitation Reporting Level

Age group, years	N	Mean	SD	P
<60	516	$3077	$2043	0.019
60–64	302	$3198	$2004
65–69	853	$3180	$1861
70–74	1352	$3303	$1849
75–79	2063	$3324	$1758
80–84	3567	$3341	$1821
85–89	4659	$3324	$1750
≥90	4541	$3326	$1616

Functional limitation reporting level	N	Mean	SD	P
0	115	$400	$913	<.0001
1	196	$1555	$1508
2	1559	$2693	$1516
3	4960	$3170	$1629
4	6674	$3486	$1712
5	3992	$3607	$1863
6	357	$3245	$2248

SD, standard deviation.

Discussion

Of the 27 predictor variables in the multiple linear regression model, 16 were statistically significant. When looking at which variables are most predictive, it is clear that physical functioning at the time of evaluation (initial functional limitation reporting category and level, and residence in community vs. facility) has the greatest influence on the overall cost of a rehabilitation case for this population. Both functional limitation reporting level and living in a facility are indicators of a patient's amount of disability. The Carrying functional limitation reporting category predicted lower cost (by $702), as those classified as such are more likely to be functionally well than those assigned the Mobility functional limitation reporting on evaluation. Moreover, one of the strongest relationships identified is that between initial functional limitation reporting level and cost (ie, an increase of $253 per functional limitation reporting level increase). Costs increase steadily as the level of disability increases until a patient reaches 100% disability, at which point costs may decrease as the plan of care becomes primarily rooted in patient-centered compensatory techniques and education. Thus, it appears that additional focus on, and sound measurement of, variables related to functional ability may allow for better prediction of rehabilitation costs.

In addition, the state in which the patient was treated has a significant effect on costs. Although 8 out of 9 states, when compared with New Jersey, were significantly different in one direction or the other, this high impact on cost, relative to other variables, is likely the result of state-specific reimbursement rates and relative differences in the health needs of specific populations.

By contrast to the strong effects of level of disability and state noted, the demographic variables of age and sex had no statistically significant effects. The most likely reason for this lack of significant effects is the relative homogeneity of the sample: about 70% were female, and almost 50% were in their 80s.

On the other hand, roughly one fifth were recently hospitalized, and recent hospitalization predicted higher total cost by $113. In turn, recent hospitalization is associated with lower functional status. In the FOX sample, those with recent hospitalizations have lower mean functional status: mean functional limitation reporting level of 3.88 vs. 3.78. Similarly, Greysen et al⁵ identified a 40% increase in hospital admission risk if one is dependent in at least 3 activities of daily living, which would characterize the vast majority of the FOX sample based on their functional limitation reporting.

Recent analyses of the overall FOX patient population indicate that 78% of patients have at least 2 chronic conditions and that 30% have at least 5. Thus, it is rare for a FOX patient to have just 1 chronic condition, the effect of which can be examined in isolation. This may explain why, of the 9 chronic conditions, only 3 (dementia, PD, and stroke) significantly influenced total cost of care in the sample. However, it does demonstrate some ability to modulate costs in outpatient therapy based on key chronic conditions that might become more apparent if Medicare beneficiaries were referred when they have fewer chronic conditions.

The relative homogeneity of the FOX sample compared to typical outpatients is likely the result of a reactive approach to health care. Patients generally are not referred for therapy services under Medicare Part B, in the home, until they reach frailty and are approaching or already experiencing failure. As a result, the variance in costs is limited, given that ∼90% are at least 40% disabled on arrival, about one third already live in a facility, and only <1% were appropriate for the Carrying functional limitation reporting category, signifying lower impairment. In addition, the inclusion/exclusion criteria were designed to identify patients who were able to adhere to their plan of care and achieve self-management, which purposefully focused the analysis on a subset of the population who utilize more outpatient services. Most of those excluded from this sample were unexpected discharges. Of those who were not unexpectedly discharged (ie, had initial evaluation and discharge functional limitation reporting category and level), 83.6% met the goal of self-management. If patients seen for outpatient Medicare Part B therapy were less disabled and more well, it stands to reason that fewer would be unexpectedly discharged for medical reasons. It appears that if these patients were referred more proactively, clinical value would be enhanced via clinical stratification systems that identify unnecessary costs and improve outcomes.

Despite the prevalence of frailty in older age and its association with adverse outcomes, it is not routinely identified and treated the same way as other long-term conditions.⁶ This reactive approach makes therapeutic interventions less effective and more costly. The World Report on Aging and Health states that people are living longer because of modern medicine with little to no change in functional ability.⁷ Similarly, the Center for Healthcare Quality and Payment Reform recommended spending proactively to reduce avoidable costs later⁸ and proactively focusing on individuals' functional wellness.

A lifestyle that includes adequate physical activity and proactive physical therapy as needed may be a cost-effective way to manage the health of the older US population. Physical therapists are trained to screen and identify underlying impairments before they manifest themselves as functional limitations, which reduces downstream costs (ie, reducing risk of falls and hospitalizations).^9
–11 Intervention by a physical therapist can increase efficiency in chronic disease management, and contribute to improved quality and cost savings. Such a proactive approach would be analogous to the many medically driven approaches such as the Medicare Diabetes Prevention Program.

Some of this study's results – patients who need more therapy cost more, and patients treated in states with higher reimbursement rates cost more – are not surprising. However, these findings represent a first step in a process of improving the quality of research in this area. Several studies have shown the positive impact of early physical therapy intervention on health care utilization and downstream costs related to low back pain^12,13 and acute musculoskeletal disorders.¹⁴ Although these studies are not directly applicable to the population studied herein, they help to contextualize these findings. A similar trend is visible when the population studied is compared to Medicare Part B beneficiaries in general, whose physical therapy cases cost $850 on average.⁴ The DOT.PA study beneficiaries were younger, with fewer falls and prior hospitalizations, and less likely to live in a facility or have dementia or PD compared to the population studied herein. It may be that the cost discrepancy is at least partially related to earlier intervention.

Although FOX Rehabilitation and other providers work to reduce unnecessary practice variability and to guarantee maximal adherence to evidence-based practice, rehabilitation must be patient-centered. Therefore, there is a need for additional research focusing on the impact of specific treatment paradigms as we move toward a value-based model rooted in population health.

Limitations

This study is an analysis of EMR data for a single outpatient rehabilitation provider. Results cannot be extrapolated to all outpatient rehabilitation providers in the United States. FOX is associated with physical therapists in 10 states, not all 50. Because the data analyzed are EMR data, they are subject to limitations common with EMR data (eg, lack of standardization, occasional free text).

Additionally, the research design used demographic and clinical data to predict costs, and did not account for specific differences in treatment approaches among patients, which are likely to be influential in predicting costs. Such treatment factors need to be investigated further both to predict costs and to predict outcomes. In addition, costs were calculated at the patient case level and therefore do not reflect total cost per patient over time.

Although the FOX data do not represent outpatient therapy costs perfectly, the data do allow the examination of associations between predictive variables and costs for a wider range of costs with less regulatory influence.

Although the results have limitations, they can provide benchmarks for costs for outpatient physical therapy providers in general, and describe the degree of association between patient disability and complexity and costs. This rehabilitation-specific predictive model is consistent with a growing body of evidence relating functional limitations^5,11,15 and certain chronic conditions¹⁶ to higher health care costs. This model can have significant impact from the perspectives of intervention, payment, and value. It can help providers identify potentially high-cost patients early in their plan of care so that factors such as clinical dosage of interventions may be analyzed. FOX Rehabilitation internal data indicate that a higher frequency of initial treatment may shorten the plan of care and help reduce overall costs. Perhaps other providers can use their own internal data for the same purposes. The model also can be used to identify cost “standards” for different patient presentations that may guide third-party payments and help quantify clinical value.

Conclusion

This study employed multiple linear regression to identify factors that predict outpatient rehabilitation costs for a single large provider. Of some 27 candidate predictors, those most predictive included the state in which service was provided and the patient's physical functioning at the time of evaluation (initial functional limitation reporting category and level, and residence in community vs. facility). There is a need for more research focusing on the effects of specific types of treatment, and also a need to consider a new, more proactive model for outpatient therapy reimbursement that emphasizes prevention as well as treatment.

Footnotes

Author Disclosure Statement

The authors declared that there are no conflicts of interest. The authors received the following financial support: Jefferson College of Population Health received funding from FOX Rehab, Inc. for performing the data analyses in this manuscript.

References

Schneider

, Sarnak

, Squires

, Shah

, Doty

. Mirror, Mirror 2017: International Comparison Reflects Flaws and Opportunities for Better U.S. Health Care. 2017. www.commonwealthfund.org/interactives/2017/july/mirror-mirror/assets/Schneider_mirror_mirror_2017.pdf Accessed January 22, 2018 .

Lewis

. Our future selves: unprecedented opportunities. Phys Ther, 2016; 96:1493–1502.

Pfuntner

, Wier

, Steiner

. Costs for Hospital Stays in the United States, 2011. Healthcare Cost and Utilization Project. 2013. https://www.hcup-us.ahrq.gov/reports/statbriefs/sb168-Hospital-Costs-United-States-2011.jsp Accessed January 22, 2018 .

Silver

, Lyda-McDonald

, Bachofer

, Gage

. Developing Outpatient Therapy Payment Alternatives (DOTPA): 2010 Utilization Report. 2013. https://www.cms.gov/Medicare/Billing/TherapyServices/Downloads/2010-DOTPA-Utilization-Report.pdf Accessed January 22, 2018 .

Greysen

, Cenzer

, Auerbach

, Covinsky

. Functional impairment and hospital readmission in Medicare seniors. JAMA Intern Med, 2015; 175:559–565.

Harrison

, Clegg

, Conroy

, Young

. Managing frailty as a long-term condition. Age Aging, 2015; 44:732–735.

World Health Organization. World Report on Ageing and Health. 2015. http://apps.who.int/iris/bitstream/10665/186463/1/9789240694811_eng.pdf Accessed January 22, 2018 .

Miller

. Creating a Provider-Led Healthcare Future. 2016. www.chqpr.org/downloads/HaroldMiller_APTA_09-17-16.pdf Accessed January 22, 2018 .

Sherrington

, Tiedemann

, Fairhill

, Close

, Lord

. Exercise to prevent falls in older adults: an updated meta-analysis and best practice recommendations. N S W Public Health Bull, 2011; 22:78–83.

10.

Avin

, Hanke

, Kirk-Sanchez

, et al. Management of falls in community-dwelling older adults: clinical guidance statement from the Academy of Geriatric Physical Therapy of the American Physical Therapy Association. Phys Ther, 2015; 95:815–834.

11.

Falvey

, Burke

, Malone

, Ridgeway

, McManus

, Stevens-Lapsley

. Role of physical therapists in reducing hospital readmissions: optimizing outcomes for older adults during care transitions from hospital to community. Phys Ther, 2016; 96:1125–1134.

12.

Fritz

, Childs

, Wainner

, Flynn

. Primary care referral of patients with low back pain to physical therapy: impact on future health care utilization and costs. Spine (Phila Pa 1976), 2012; 37:2114–2121.

13.

Gellhorn

, Chan

, Martin

, Friedly

. Management patterns in acute low back pain: the role of physical therapy. Spine (Phila Pa 1976), 2012; 37:775–782.

14.

Mitchell

, de Lissovoy

. A comparison of resource use and cost in direct access versus physician referral episodes of physical therapy. Phys Ther, 1997; 77:10–18.

15.

Simmonds

, Syddall

, Westbury

, Dodds

, Cooper

, Sayer

. Grip strength among community-dwelling older people predicts hospital admission during the following decade. Age Ageing, 2015; 44:954–959.

16.

Dieleman

, Baral

, Birger

, et al. US spending on personal health care and public health, 1996–2013. JAMA, 2016; 316:2627–2646.