Chapter 4 – National Estimate Model Abstract Background Methods
Prospective Payment System: Effective January 1, 2002, the Health Care Financing Administration shifted Medicare payment for inpatient rehabilitation to a Prospective Payment System system.1 Although Medicare has paid acute-care hospitals under a PPS since 1983, rehabilitation facilities, which provide extensive occupational, physical, and speech therapy services, had been exempt from that system. With PPS, rehabilitation facilities are paid on a per-discharge basis based on the patient diagnoses, with hospitals paid more to care for patients with greater needs. For the most part, private insurers appear to have adopted the Medicare reimbursement rates for rehabilitation. The Federal payment data, thus, provides a reliable basis for estimating rehabilitation expenditures. The prospective payment rates were reported in the Federal Register, Inpatient Rehabilitation Facility Prospective Payment System report, which provided payment rates for three tiers of serious co-morbidities (e.g., stroke plus hip fracture) and one payment rate for cases with no serious co-morbidities. Although the PPS provides four payment tiers based on co-morbidities, all injuries fall into the no-co-morbidity rate unless complicated by an illness (e.g., stroke, tuberculosis). Our estimates ignore such illness complications, since the cost differential is attributable to the illness, not the injury. The Federal Prospective Payments for Case-Mix Groups (CMG) only provided allowable payments broken down by CMG (categories determined by age and motor and cognitive FIM scores) as well as tier and diagnosis (Table 2 in the regulations). Uniform Data System for Medical Rehabilitation: UDSMR collects and redistributes data from rehabilitation hospitals nationwide for use in evaluating the effectiveness and efficiency of their rehabilitation programs. It provides the most comprehensive data available on rehabilitation patients across diagnostic categories. In 2002, 783 comprehensive medical rehabilitation (CMR) facilities sent data to UDSMR. Of the CMR subscribers, 590 agreed to provide data for this study. Only cases containing cause-of-injury codes (E-codes) were selected for analysis. Five years of data (1998–2002) were combined into one dataset. The data were cleaned. For example, obvious miscoding in the E-codes was corrected and variables with missing data or codes out-of-range were excluded from the appropriate analyses. Because E–codes were indicated in multiple fields, we assigned each case to one unique etiology using a hierarchy scheme. Motorcycle riders, lacking the protection of a steel-encased vehicle, are more vulnerable to injury. Therefore motorcycle injuries were primary, followed by other motor vehicle, suicide, assault, and other unintentional injury. This resulted in 1,437 rehabilitation patients who incurred injury in motorcycle crashes. We collapsed the cases into the diagnosis groupings used in the PPS data’s CMGs (e.g., traumatic brain injury, lower-limb amputation). Table 1 lists the groupings. This was done by using UDSMR data on each patient’s diagnoses, admission FIM motor score and cognitive score, and where relevant, age. Within each diagnosis group, the PPS payment rates for no-co-morbidity were multiplied by the proportion of motorcycle injury rehabilitation cases in the CMG, then summed over the CMGs within the impairment class to get mean payment rates for motorcycle injuries by impairment class. (Appendix A table 1 provides adjusted Federal prospective payments by diagnosis group across five cause categories and appendix A table 2 provides length of stay.) Five years of data (1998–2002) were combined into one dataset and cleaned. For example, obvious miscoding in the E-codes was corrected and variables with missing data or codes out-of-range were excluded from the appropriate analyses. Because E-codes were indicated in multiple fields, we assigned each case to one unique cause using a hierarchy scheme. Motorcycle injuries were primary, followed by other motor vehicle, suicide, assault, and other unintentional injury. This resulted in 84,870 rehabilitation patients. We collapsed the cases into the diagnosis groupings used in the PPS data’s CMGs (e.g., traumatic brain injury, lower-limb amputation). This was done by using UDSMR data on each patient’s diagnoses, admission FIM motor score and cognitive score, and, where relevant, age. Within each diagnosis group, the PPS no-co-morbidity payment rates were multiplied times the proportion of injury rehabilitation cases in the CMG, then summed over the CMGs within the impairment class to get mean payment rates by impairment class. This calculation was as follows: Let, pIj = proportion of cases in an impairment class (I) in CMGIIj Table 1. Adjusted Federal Prospective Payment Rehabilitation Rates* and Mean Length of Stay for Motorcycle Injuries by Diagnosis Group (2000 Estimates in 2002 dollars)
Under the PPS, rehabilitation hospital costs would have ranged from $7,817 for neck and back pain to a high of $28,966 for spinal cord injury with complete quadriplegia, which also is reimbursed for the longest length of stay–on average 38 days. The Healthcare Cost and Utilization Project - National Inpatient Sample 2000 is a large, statistically representative sample of U.S. hospital discharges compiled by the Agency for Health Research and Quality of the U.S. Department of Health and Human Services. We used the 2000 HCUP-NIS data file to develop estimates of the number of injury episodes resulting in hospitalizations with live discharges in 2000. The HCUP-NIS provides information annually on approximately 5 million to 8 million inpatient stays that resulted in discharges in 2000 from about 1,000 hospitals. These hospitals represent a 20 percent cluster sample of non-Federal, short-term, general, and other specialty hospitals, excluding hospital units of institutions, drawn from a convenience sample of 28 States that agreed to supply AHRQ with discharge census data. All discharges from sampled hospitals are included in the HCUP-NIS database, and sampling weights are included to allow for generating nationally representative estimates. From this dataset we selected only those cases with an injury diagnosis in any of the first three diagnosis fields. When E-codes were missing from the record (approximately 20 percent of the cases), we assigned them probabilistically2. We then dropped fatalities, rehabilitation visits, and all visits that were not for acute traumatic injury (e.g., poisonings). Based on the primary injury E-code, we selected cases whose cause was a highway crash (E–codes in the range E810–E819) and whose victim was a motorcyclist (fifth digit of 2 or 3). We then classified injuries into the Rehabilitation Impairment Code (RIC) groups prescribed in the PPS according to the primary injury diagnosis using 13 categories collapsed from the Barell Injury Diagnosis Matrix, which groups ICD–9-CM codes by body region and nature of injury. In order to capture surgical amputations as well as traumatic amputations, when any procedure codes indicated that an amputation was performed, the case was re-categorized as an amputation of the appropriate body region – unless the case was a TBI, SCI, or burn. In these cases it was not re-categorized. The three-state 1997 census of hospital discharge records: PIRE previously had obtained, cleaned and pooled the injury discharges from hospital discharge census data for 1997 from acute and rehabilitation hospitals in California, Maryland, and Pennsylvania. These three States were selected because they report data on rehabilitation specialty hospitals, in addition to general acute-care hospitals. Validity checks were completed among the States, and when needed, variables were recoded to produce uniform coding categories and value labels across States for variables such as discharge status and ethnicity. The collecting agencies required the hospitals to report E-codes for the acute care discharges, with 92 percent compliance. Rehabilitation discharges, however, were voluntarily coded and according to some States’ coding rules (oriented toward getting an unduplicated count of injury incidents) should not have been cause-coded. California and Pennsylvania identified the hospital type, including rehabilitation. Although the Maryland data does not explicitly indicate rehabilitation visits, it contained Diagnosis-Related Group (DRG) codes and diagnosis codes that allowed us to identify these visits. We then looked at the distribution of patient rehabilitation status by hospital. If a hospital's patients were predominantly rehabilitation patients (80% or more), we labeled the hospital a rehabilitation hospital. (Other hospital types are acute care, psychiatric, and nursing.) Note that many acute-care hospitals have rehabilitation wards; therefore not all rehabilitation patients are treated in rehabilitation specialty hospitals. We classified as rehabilitation visits both patients coded as receiving inpatient rehabilitation treatment and all patients admitted to rehabilitation facilities. We tabulated the rehabilitation probabilities by diagnosis using the same diagnosis categories we used in the HCUP-NIS calculations. We then multiplied the HCUP-NIS injury case counts by the three-state rehabilitation probabilities for each diagnosis group to produce estimates of the number of admissions for rehabilitation in the United States for 2000. The estimates of rehabilitation admissions were then multiplied by the PPS-based rehabilitation costs per case to yield an estimate of aggregate rehabilitation costs in 2000. Final computations are based on unrounded, weighted numbers and are shown to the nearest whole numbers. Table 2 shows the rehabilitation probabilities by diagnosis group and cause. Table 2. Percentage of Hospital Admitted Injuries that Involve Inpatient Rehabilitation by Cause and Diagnosis Group, California, Maryland, and Pennsylvania, 1997
Findings Overall, HCUP-NIS suggests 243,229 patients were admitted for other motor vehicle injuries in 2000 and 24,028 patients were admitted for motorcycling injuries. Five percent of the other motor vehicle and 5 percent of the motorcycling injury patients had received inpatient rehabilitation either separately from or as a part of their hospitalized acute-care stays. For both motor vehicle and motorcycle injuries, spinal cord injury victims with or without other major injuries had the highest probability of receiving rehabilitation services, of admissions involving rehabilitation facilities. Table 3. Rehabilitation Costs of Hospital-Admitted Other Motor Vehicle Injuries (Excluding Motorcycle), 2000 (in 2002 dollars)
Table 4. Rehabilitation Costs of Hospital-Admitted Motorcycling Injuries, 2000 (in 2002 dollars)
Other Motor Vehicles Motorcycle Injuries Discussion Our PPS-based estimates for 2002 tend to be a bit lower than the AMRPA average cost data for 1999. That is predictable; PPS was designed to contain or sharply reduce inflation in inpatient rehabilitation care costs Nevertheless, the two sets of costs are similarly ranked by diagnosis, providing corroboration for our estimates. The PPS forced down prices, but some payers still may be paying the higher rates in the AMRPA data. To the extent they are, our estimates are conservative. Table 4. Comparison of APRMA Average Cost Data to UDSMR/PPS Cost Estimates
2When a record identified as an acute injury admission lacked an E-code, we assigned E-codes probabilistically based on the primary injury diagnosis. We determined the frequency distribution of E-codes from all E-coded records in the dataset with the same primary injury diagnosis. We then created a series of duplicate records, one with each E-code that was found, and weighted them by their frequency of occurrence. Example: A non-E-coded record has a primary injury diagnosis of 830.0, closed dislocation of jaw. The dataset includes three E-coded cases with this diagnosis - two E812.0 (motor vehicle driver in traffic collision) and one E960.0 (unarmed fight). So two copies of this record are created - one with an E-code of E812.0 and the case weight multiplied by two-thirds, and one with an E-code of E960.0 and the case weight multiplied by one-third. |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||