Comparison of logistic regression and neural networks to predict rehospitalization in patients with stroke.

CONTEXT: Rehospitalization following inpatient medical rehabilitation has important health and economic implications for patients who have experienced a stroke. OBJECTIVE: Compare logistic regression and neural networks in predicting rehospitalization at 3-6-month follow-up for patients with stroke discharged from medical rehabilitation. DESIGN: The study was retrospective using information from a national database representative of medical rehabilitation patients across the US. SETTING: Information submitted to the Uniform Data System for Medical Rehabilitation from 1997 and 1998 by 167 hospital and rehabilitation facilities from 40 states was examined. PARTICIPANTS: 9584 patient records were included in the sample. The mean age was 70.74 years (SD = 12.87). The sample included 51.6% females and was 77.6% non-Hispanic White with an average length of stay of 21.47 days (SD = 15.47). MAIN OUTCOME MEASURES: Hospital readmission from 80 to 180 days following discharge. RESULTS: Statistically significant variables (P <.05) in the logistic model included sphincter control, self-care ability, age, marital status, ethnicity and length of stay. Area under the ROC curves were 0.68 and 0.74 for logistic regression and neural network analysis, respectively. The Hosmer-Lemeshow goodness-of-fit chi-square was 11.32 (df = 8, P = 0.22) for neural network analysis and 16.33 (df = 8, P = 0.11) for logistic regression. Calibration curves indicated a slightly better fit for the neural network model. CONCLUSION: There was no statistically significant or practical advantage in predicting hospital readmission using neural network analysis in comparison to logistic regression for persons who experienced a stroke and received medical rehabilitation during the period of the study.

Characteristics of persons rehospitalized after stroke rehabilitation.

OBJECTIVE: To develop classification models for risk of hospital readmission 80 to 180 days after discharge based the demographic and functional characteristics of persons discharged from acute inpatient rehabilitation after stroke. DESIGN: Retrospective, using information from US facilities subscribing to the Uniform Data System for Medical Rehabilitation (UDS(MR)). SETTING: Information submitted to the UDS(MR) from 1994 through 1996 by 167 hospital and rehabilitation facilities from 40 states was examined. PARTICIPANTS: A total of 15,992 records of patients (mean age +/- standard deviation, 70.97 +/- 12.19yr) with a diagnosis of stroke were included in the final sample. The sample included 52.7% women and was 80% non-Hispanic white with an average length of stay (LOS) of 25.31 +/- 14.72 days. INTERVENTIONS: Not applicable. MAIN OUTCOME MEASURES: Six subscales of the FIM instrument (self-care, sphincter control, transfers, locomotion, communication, social cognition), total FIM, and other predictor variables for regression analysis (gender, age, ethnicity, marital status, prehospital living setting, LOS, primary payer source, level of function-related group). RESULTS: A logistic regression model included the following statistically significant variables (p <.05): ethnicity, sphincter control, self-care ability, gender, and LOS. The greatest variability occurred among men. Exactly 18.1% of non-Hispanic white men and 17.9% of African-American men were rehospitalized. In contrast, only 10.1% of Hispanic men and 11.4% of Asian men were rehospitalized. The odds of rehospitalization were lowest for Hispanic men. CONCLUSION: As prospective payment systems are introduced for postacute care, it is important that the relationship among functional abilities, demographic characteristics, and incidence of hospital readmission following medical rehabilitation be examined.

Satisfaction with medical rehabilitation in patients with cerebrovascular impairment.

OBJECTIVE: Overall satisfaction has important social and economic implications for patients who have received inpatient medical rehabilitation. We conducted this study to examine the overall satisfaction level at 3- to 6-mo follow-up for inpatients with cerebrovascular impairments discharged from medical rehabilitation. DESIGN: The study was retrospective using information from a national database representative of medical rehabilitation patients across the United States. Information submitted in 1997 and 1998 to the Uniform Data System for Medical Rehabilitation by 177 hospital and rehabilitation facilities from 40 states was examined. The final sample included 8,900 patient records. The main outcome measure was the level of satisfaction with medical rehabilitation at 80-180 days postdischarge follow-up. RESULTS: A logistic regression model including ten independent variables was used to predict satisfied vs. dissatisfied at follow-up. Three statistically significant variables were included in the final model and correctly classified 95.1% of the patients. Higher FIM instrument discharge scores were associated with increased satisfaction. Further analysis of the FIM instrument subscales indicated that higher ratings in transfers, social cognition, and locomotion were significantly associated with increased satisfaction. CONCLUSION: We identified several functional variables associated with increased satisfaction after medical rehabilitation in persons with stroke. The ability to objectively assess patient satisfaction is important as consumer-based outcome measures are integrated in accreditation and healthcare evaluation.

Length of stay and hospital readmission for persons with disabilities.

OBJECTIVES: Length of stay (LOS) and hospital readmission for persons receiving medical rehabilitation were examined. METHODS: A total of 96,473 patient records (1994-1998) were analyzed. Mean age of patients was 68.97 years; 61% were female and 83% were non-Hispanic White. RESULTS: A decrease in LOS of 6.07 days (SD = 3.23) and increase in hospital readmission were found across all impairment groups (P < .001). Readmission increases ranged from 6.7% for amputations to 1.4% for orthopedic conditions. LOS was longer (2.1 days) for readmitted patients (P < .01). Age was not a significant predictor of rehospitalization. CONCLUSIONS: Understanding variables associated with rehospitalization is important as prospective payment systems are introduced for postacute care.

Relation of disability costs to function: spinal cord injury.

OBJECTIVE: This study evaluated the validity of the Functional Independence Measure (FIM instrument) in predicting (1) the number of minutes of daily assistance provided, (2) the cost of durable goods currently used, and (3) the number of paid helper hours provided daily to persons with spinal cord injury living in the community. DESIGN: A cross-sectional study. SUBJECTS: One hundred nine persons with spinal cord injury who were a median 6 years postdischarge from initial medical rehabilitation. RESULTS: A significant inverse linear relationship was observed between FIM scores and the square root values of the three cost-related measures. The FIM-18 and the FIM motor scores were the best single predictors of the square root of minutes of assistance (paid and/or unpaid) per day, explaining 85% of variance. The FIM motor measure was the best single predictor of square root of cost of durable goods, explaining 29% of variance. The Self-Care, FIM motor, and FIM-18 scores equally predicted square root of hours of paid help per day, explaining 58% of variance. CONCLUSION: The findings indicate FIM-related scores predict the amount of assistance needed and certain costs for persons with spinal cord injury disability.

Predicting follow-up functional outcomes in outpatient rehabilitation.

Functional outcomes at an average of six months after outpatient rehabilitation were investigated in a pilot study with a sample of 42 patients receiving physical therapy for low back, neck, and other musculoskeletal problems. Logistic regression analyses were used to study variables related to the achievement of a predetermined level of physical functioning or to whether improvement occurred from initial to follow-up assessment. Scales used in the analyses were obtained from two measures of functioning, the Medical Outcomes Trust Short Form 36 (SF-36) and the Medical Rehabilitation Follow Along (MRFA instrument). Demographic and program characteristic variables obtained from clinic records were also included in the analyses. The independent variables entered into two groups of regression equations included age, gender, presenting problem, workers' compensation coverage, functioning at initial assessment, number of visits, length of program, and intensity of program. For the SF-36, the General Health scale and presenting problem predicted above or below a level greater than one standard deviation less than the mean on the Physical Functioning scale at follow-up. The combination of the Role Physical and Role Emotional scales predicted follow-up level on the Physical Component Summary (PCS) scale. The interaction of age and intensity of treatment was predictive for both the Physical Functioning and PCS scales. For predicting improvement or no improvement using the SF-36 scales, General Health and the interaction between age and intensity were predictive of change on the Physical Functioning scale, whereas age and the interaction between age and number of visits were predictive of change on the PCS scale. For the MRFA instrument, the interaction between age and the intensity of treatment was predictive for both a predetermined level of physical functioning and amount of improvement in physical functioning at follow-up. An intensity rating of pain and a measure of affective distress at initial assessment both improved the identification of patients at risk for not reaching a given level of functioning, whereas treatment intensity improved the identification of those patients at risk for not showing improvement at follow-up. Length of time between the end of outpatient therapy and follow-up assessment was generally not related to follow-up physical functioning. Interpretation and implications of these findings, as well as applications of this analytic approach to outcomes assessment, are discussed.

Functional task benchmarks for stroke rehabilitation.

OBJECTIVE: To determine typical outcome "benchmarks" for 18 functional tasks in patients undergoing stroke rehabilitation. The benchmarks are intended to serve as points of reference to which the outcomes of patients with similar impairments and degrees of disability can be compared. SUBJECTS: Records from 26,339 stroke patients discharged from 252 inpatient facilities across the United States that submitted 1992 data to the Uniform Data System for Medical Rehabilitation. METHODS: Stroke impairment was detailed as the presence or absence of hemiparesis resulting from stroke and the side(s) of involvement. Within each of five stroke impairment categories, patients were further classified by the Functional Independence Measure-Function-Related Groups (FIM-FRGs) into nine syndromes by degree of disability (admission motor and cognitive FIM scores) and by age. Outcomes were determined for each stroke syndrome at patients' discharge from medical rehabilitation. MAIN OUTCOME MEASURES: Patients' median performance levels on each of the 18 items making up the FIM, length of stay, and community discharge rates. RESULTS: The majority of patients whose admission motor FIM scores were above 37 were able to eat, groom, dress the upper body, and manage bladder and bowel functions independently by discharge. In addition to these tasks, most of those whose motor FIM scores were above 55 were able to dress the lower body, bathe, and transfer onto a chair/bed or toilet. The majority of patients whose initial motor FIM scores were above 62 points and whose cognitive FIM scores were above 30 gained independence in most tasks, including stair climbing and tub transfers. Community discharge rates ranged from 51.6% for the group of patients with the most severe disabilities to 99.2% for the group with the least severe disabilities. CONCLUSION: The clinician can apply these benchmarks to guideline development and quality improvement, and in establishing patient goals.

The Interest Checklist: a factor analysis.

OBJECTIVE: The purpose of this study was to determine whether the 80 items on the Interest Checklist empirically cluster into the five categories of interests described by Matsutsuyu, the developer of the tool. METHOD: The Interest Checklist was administered to 367 subjects classified in three subgroups: students, working adults, and retired elderly persons. An 80-item correlation matrix was formed from the responses to the Interest Checklist for each subgroup and then used in a factor analysis model to identify the underlying structure or domains of interest. RESULTS: Results indicated that the Social Recreation theoretical category was empirically independent for all three subgroups; the Physical Sports and Cultural/Educational theoretical categories were empirically independent for only the college students and working adults; and the Manual Skills theoretical category was empirically independent for only the working adults. CONCLUSION: Although therapists should continue to be cautious in their interpretation of patients' Interest Checklist scores, the tool is useful for identifying patients' interests in order to choose meaningful activities for therapy.

Development of function-related groups version 2.0: a classification system for medical rehabilitation.

OBJECTIVE: To present a new version (2.0) of the Functional Independence Measure-Function Related Group (FIM-FRG) case-mix measure. DATA SOURCE/STUDY SETTING: 85,447 patient discharges from 252 freestanding facilities and hospital units contained in the 1992 Uniform Data System for Medical Rehabilitation. STUDY DESIGN: Patient impairment category, functional status at admission to rehabilitation, and patient age were used to develop groups that were homogeneous with respect to length of stay. Within each impairment category patients were randomly assigned to one data set to create the system (through recursive partitioning) or a second set for validation. Clinical and statistical criteria were used to increase the percentage of patients classified, expand the impairment categories of FIM-FRGs Version 1.1, and evaluate the incremental predictive ability of coexisting medical diagnoses. Predictive stability over time was evaluated using 1990 discharges. PRINCIPAL FINDINGS: In Version 2.0, the percentage of patients classified was increased to 92 percent. Version 2.0 includes two new impairment categories and separate groups for patients admitted to rehabilitation for evaluation only. Coexisting medical diagnoses did not improve LOS prediction. The system explains 31.7 percent of the variance in the logarithm of LOS in the 1992 validation sample, and 31.0 percent in 1990 discharges. CONCLUSIONS: FIM-FRGs Version 2.0 includes more specific impairment categories, classifies a higher percentage of patient discharges, and appears sufficiently stable over time to form the basis of a payment system for inpatient medical rehabilitation.

A prediction model for functional recovery in stroke.

BACKGROUND AND PURPOSE: Stroke-related physical disability can diminish quality of daily living, place care burden on families, and increase need for long-term institutionalization. We developed a prognostic index for use in research and with potential for adaptation to clinical practice that establishes the likelihood of an individual achieving a specific stage of functional recovery after stroke rehabilitation. METHODS: We constructed the index using logistic regression based on 3760 patient records from 96 rehabilitation facilities in 31 states. The stage, as measured by the Functional Independence Measure, includes achievement of the following: independence in eating, grooming, and dressing the upper body; continence in bowel and bladder; and transfer between a bed and chair with supervision only. RESULTS: This stage was achieved by 26.1% of patients functioning below it at rehabilitation admission. Disability onset of less than 60 days was associated with more than a 3-fold increase in the likelihood of achieving the stage (adjusted odds ratio, 3.5; 95% confidence interval, 2.0 to 6.0). Each eight-point increase in an eight-item activities of daily living score, measured at admission to rehabilitation, increased the odds 2.5-fold (95% confidence interval, 2.3 to 2.8). For those living alone or employed before the stroke, the odds of achieving the stage increased by factors of 1.3 and 2.2, respectively. The index showed minimal shrinkage on cross validation. The achievement of this profile of function is important because 95.3% of stroke patients who achieved or exceeded it were discharged home, as opposed to only 66.8% of those who did not achieve it. CONCLUSIONS: The index can be used to establish prognoses for individual stroke patients at admission to rehabilitation with regard to achieving this stage. Achievement of the stage is associated with a high likelihood of discharge to home.

A brief outpatient functional assessment measure: validity using Rasch measures.

The Medical Rehabilitation Follow Along (MRFA(TM)) is a brief outpatient functional assessment measure that was developed using Rasch analysis. The MRFA currently has musculoskeletal, neurologic, multiple sclerosis, cardiac, and pulmonary forms. Using Rasch scoring and selected scales, the 31-item musculoskeletal form of the MRFA was compared with and contrasted to a measure of general health status, the Medical Outcomes Trust SF-36. Content, construct, and criterion validity were addressed using scale scores before and after outpatient rehabilitation, as well as therapist ratings of improvement. The results supported the validity of inferences made from the MRFA scales using Rasch measures for persons with musculoskeletal problems. Rasch and raw scoring provided similar results with respect to the validity of the MRFA scales. Implications for the use of Rasch and raw scoring approaches with the MRFA are discussed.

The reliability of the functional independence measure: a quantitative review.

OBJECTIVE: The reliability of the Functional Independence Measure (FIMSM) for adults was examined using procedures of meta-analysis. DATA SOURCES: Eleven published studies reporting estimates of reliability for the FIM were located using computer searches of Index Medicus, Psychological Abstracts, the Functional Assessment Information Service, and citation tracking. STUDY SELECTION: Studies were identified and coded based on type of reliability (interrater, test-retest, or equivalence), method of data analysis, size of sample, and training or experience of raters. DATA EXTRACTION: Information from the articles was coded by two independent raters. Interrater reliability for coding all elements included in the analysis ranged from .89 to 1.00. DATA SYNTHESIS: The 11 investigations included a total of 1,568 patients and produced 221 reliability coefficients. The majority of the reliability values (81%) were from interrater reliability studies, and the intraclass correlation coefficient (ICC) was the most commonly used statistical procedure to compute reliability. The reported reliability values were converted to a common correlation metric and aggregated across the 11 studies. The results revealed a median interrater reliability for the total FIM of .95 and median test-retest and equivalence reliability values of .95 and .92, respectively. The median reliability values for the six FIM subscales ranged from .95 for Self-Care to .78 for Social Cognition. For the individual FIM items, median reliability values varied from .90 for Toilet Transfer to .61 for Comprehension. Median and mean reliability coefficients for FIM motor items were generally higher than for items in the cognitive or communication subscales. CONCLUSIONS: Based on the 11 studies examined in this review the FIM demonstrated acceptable reliability across a wide variety of settings, raters, and patients.