Abstract
Objective:
To examine the effects of occupational therapy (OT) services on patient discharge outcomes.
Design:
This is a retrospective cohort study. We developed decision tree algorithms to investigate how 3 measures of OT service delivery—(1) the total number of completed OT visits, (2) the total minutes of all completed OT visits, and (3) the proportion of hospitalized days that included an OT visit (frequency)—affect outcomes.
Setting:
Acute care settings.
Participants:
The target population was patients admitted to and discharged from the Cleveland Clinic between 2017 and 2021, who received at least 1 OT session and stayed in the hospital for <30 days, with an orthopedic physician designated as the primary treating provider (N=36,300).
Interventions:
Not applicable.
Main Outcome Measures:
Reduced 30-day readmission rates, increased rates of home discharge, and improved daily activity abilities after hospital discharge.
Results:
Our prediction models for predicting our outcomes showed good to excellent performance metrics (accuracy, 0.69–0.91). Our model demonstrated that variations in the total number, minutes, or frequency of OT sessions are associated with the likelihood of readmission within 30 days after hospital discharge and minimal detectable improvements in daily activities for subgroups of patients with orthopedic conditions.
Conclusions:
Our study suggests that the effective use of limited OT services may require prioritizing certain subgroups of patients for providing more frequent OT sessions to optimize the impact of OT services on patient outcomes.
Keywords: Decision trees, Learning health system, Rehabilitation
Acute care hospitals are the primary employers of occupational therapy (OT) practitioners, accounting for 27% of all OT practitioners in the United States.1,2 Research has demonstrated that a higher frequency of OT services is significantly associated with lower readmission rates for patients with acute myocardial infarction, chronic obstructive pulmonary disease, coronary artery bypass graft, heart failure, pneumonia, and elective total hip and knee arthroplasty.3 The OT services are also the only category of expenditure in acute care hospitals with a statistically significant positive correlation to reduced readmission for patients with heart failure, pneumonia, and acute myocardial infarction.4 Furthermore, OT interventions have been linked to decreased delirium occurrence and duration and enhancements in physical and cognitive functions on patients’ discharge from the hospital.5
Current research highlights the importance of OT services; however, these studies were predominantly conducted with limited exploration of how different conditions demonstrate unique patterns and interact with clinical and demographic factors. Although traditional statistical models provide valuable insights, they present inherent limitations when analyzing complex relationships between OT service delivery and patient characteristics. These models often struggle to capture nonlinear interactions and hierarchical relationships in health care data.
Decision trees offer several advantages for exploring these complex relationships.6–8 First, decision trees effectively capture and model nonlinear relationships among multiple variables without requiring explicit specification. This capability facilitates discovering intricate patterns that may remain undetected or difficult to model using conventional statistical methods. Decision trees capture interactions through their hierarchical structure, allowing variables higher in the tree to modify the influence of lower variables—a more flexible approach than traditional statistical interaction terms. Second, decision trees work seamlessly with categorical and numerical variables without extensive preprocessing. This feature is particularly beneficial when analyzing diverse patient characteristics and service delivery patterns. Third, by incorporating length of stay (LOS) as a covariate, decision trees adjust for differences in therapy opportunities. When LOS substantially impacts outcomes, the model accounts for this influence, ensuring that the study predictors are interpreted adequately within the appropriate hospitalization duration. Lastly, decision trees manage high levels of missingness through surrogate splits, maintaining analytical robustness. In tree construction, records with missing values are omitted from initial split decisions but are later assigned to appropriate branches using available data patterns.
The objective of this study is to identify optimal OT service patterns for patients with orthopedic conditions using decision tree models. We examine how frequency, duration, and total sessions correlate with improved outcomes—reduced 30-day readmission rates, increased home discharge, and minimal detectable improvements in daily activities—among different patient subgroups. The Centers for Disease Control and Prevention data from 2022 indicate 25.7 million US adults experience activity limitations because of arthritis, with projections of 35 million by 2040.9 This study supports addressing the substantial health care needs of the growing orthopedic population through evidence-based resource allocation by identifying patient characteristics associated with better outcomes.
Methods
Study design
This is a retrospective cohort study.
Study sample
Patients included in this study were as follows: (1) admitted to and discharged from 1 of the 11 Cleveland Clinic hospitals between January 1, 2017 and December 1, 2021: this timeframe was established to allow for the inclusion of all readmissions occurring within 30 days after discharge—up to December 31, 2021; (2) received at least 1 OT session during their hospital stay; (3) stayed in the hospital for <30 days; and (4) orthopedic physician designated as the primary treating provider during hospitalization. For individuals with multiple admissions during the study period, the most recent admission was selected for analysis, and the number of prior admissions from 2017 to 2021 was considered a covariate in our analysis. This study is approved by the Cleveland Clinic Institutional Review Board [IRB#22–362].
Outcome variables
We had 3 primary outcomes as follows: (1) 30-day hospital readmission; (2) home discharge; and (3) minimal detectable improvement in daily activities at hospital discharge. To assess improvements in patients’ abilities to perform daily activities, we used the Activity Measure for Postacute Care (AM-PAC) 6-Clicks daily activity short form,10 which was scored by occupational therapists at each visit. We used the minimal detectable change of 5.49 between the first- and last-recorded scores.10 For home discharge, we classified outcomes into 2 groups: home (home self-care and home health) versus nonhome settings (skilled nursing facilities, inpatient rehabilitation facilities, and all other nonhome locations). Furthermore, we examined hospital readmissions by categorizing patients based on whether they were readmitted to any of the Cleveland Clinic hospitals within 30 days or less after discharge.
Predictors: OT sessions
We used 3 measures to capture the delivery of OT services: (1) the total number of completed OT visits; (2) the total minutes of all completed OT visits; and (3) the proportion of hospitalized days that included an OT visit. The proportion of hospitalized days with an OT visit was calculated by dividing the number of OT visits by LOS (in d). The 3 measures of OT service delivery are retained to provide a comprehensive view of care. The standardized proportion measure accounts for varying lengths of stay, whereas the unstandardized total minutes and visits capture important absolute values of service intensity and volume.
Covariates: Demographic, environmental, and clinical features
We included a comprehensive set of 41 variables for each patient. Demographic and environmental variables included age, race, sex, payor type, home setup, availability of assistance at home, living environment, and the presence of stairs at the home entrance. Clinical variables included whether the hospital admission was elective or nonelective, whether surgery was performed during the hospital stay, Charlson comorbidity index (CCI) score, prehospital daily activity scores (AM-PAC 6-Clicks daily activity short form), prehospital daily mobility scores (AM-PAC 6-Clicks mobility short form), duration from admission to the first physical therapy (PT)-recorded AM-PAC 6-Clicks basic mobility score, duration from admission to the first PT evaluation, and duration from the last PT evaluation to discharge. Equivalent PT service measures (total number, total minutes, and proportions) were also included as covariates. Detailed information on the covariates of the study population can be found in table 1 and appendices 1 and 2.
Table 1.
Sample demographics.
| Variable | Description | Category | N (%) |
|---|---|---|---|
| Sex | Patient’s sex | Male | 15,465 (43%) |
| Female | 20,835 (57%) | ||
| Race | Patient’s race | White | 30,474 (84%) |
| Black | 4283 (12%) | ||
| Other | 1543 (4%) | ||
| Primary payor | Primary payor for hospitalization | Traditional Medicare | 11,751 (32%) |
| Medicare HMO | 9681 (27%) | ||
| Medicaid | 2641 (7%) | ||
| Commercial | 11,212 (31%) | ||
| Self-pay | 347 (1%) | ||
| Other | 668 (2%) | ||
| Elective | Elective (vs nonelective) hospital admission | Nonelective | 9284 (26%) |
| Elective | 27,016 (74%) | ||
| Assist available | Level of assistance available for patients at home at the time of discharge | 24 h | 18,661 (53%) |
| Occasional | 15,666 (44%) | ||
| None | 1127 (3%) | ||
| Missing response | 846 (2%) |
Statistical analysis
Describing OT services
This study summarized the 3 metrics of OT services across 3 patient outcome criteria. We calculated means and SDs, stratifying the results by LOS duration (<3d vs 3d or longer) and 3 outcome criteria.
Decision tree model
We developed and evaluated a decision tree algorithm to identify optimal OT service patterns and their associated patient characteristics. The impact of various patient subgroups was evaluated separately across 3 key outcomes. Appendix 3 provides detailed information regarding the selection of the decision tree model for this study, the creation of a balanced dataset, the partitioning of the dataset into training and testing sets, and the selection of the optimal complexity parameter.
Results
Demographic
A total of 36,300 patients with orthopedic conditions were included in this study. The average age of the patients was 66.7 years; 57% were women, and 84% were White. Three primary payors predominantly provided insurance coverage: traditional Medicare (32.0%), commercial insurance (31.0%), and Medicare Advantage Health Maintenance Organization (HMO, 26.7%), with Medicaid accounting for the remaining 7.3% of patients. Furthermore, 74.4% of patients received elective hospital admissions, and 95.7% underwent surgery during their stay. At discharge, 53% of patients had access to 24-hour assistance, 44% had occasional assistance available, and 3% of patients had no assistance. A total of 8597 patients had multiple admissions, averaging 1.46 admissions per patient. On average, patients stayed in the hospital for 3 days. Table 1 and appendices 1 and 2 include additional demographic information.
Descriptive OT sessions
Patients readmitted to the hospital within 30 days after discharge received 0.18 additional OT sessions and 3.8 minutes longer total OT session duration. Nonreadmitted patients had OT encounters that were 16.9% more frequent than their readmitted counterparts. A slightly different trend was observed between patients discharged to their homes and those not. Patients discharged to non-home locations received 0.49 additional OT sessions, 11.6 minutes longer in total OT session duration, and 30.4% fewer frequent OT encounters. In addition, on average, patients with minimal detectable improvement in daily activity received OT sessions that were 6.1 minutes longer, 14.6% more frequent, and negligibly more OT sessions (0.05). Patients with a LOS of 3 days or more received more and longer OT sessions, though at a lower frequency, compared to those with a LOS of <3 days across all 3 outcomes. Refer to table 2 for additional information.
Table 2.
Average provided OT sessions by length of stay (<3 vs ≥3d).
| Variables | Category | LOS | N (%) | LOS (SD) | Total OT Number (SD) | Total OT Minutes (SD) | OT Proportion (SD)* |
|---|---|---|---|---|---|---|---|
| Readmission | No | All | 35,477 (98%) | 2.92 (3) | 1.43 (0.87) | 48.2 (36.5) | 60.3 (29.4) |
| No | <3 d | 22,022 (61%) | 1.35 (0.48) | 1.19 (0.44) | 41.4 (23.1) | 74.5 (23.8) | |
| No | ≥3 d | 13,455 (37%) | 5.49 (3.56) | 1.83 (1.19) | 59.3 (49.4) | 37 (22) | |
| Yes | All | 823 (2%) | 5.17 (4.43) | 1.61 (1.17) | 52 (45.5) | 43.4 (28.6) | |
| Yes | <3 d | 273 (1%) | 1.44 (0.5) | 1.2 (0.48) | 43.4 (23.8) | 71.3 (25.1) | |
| Yes | ≥3 d | 550 (2%) | 7.02 (4.34) | 1.81 (1.35) | 56.3 (52.6) | 29.6 (18.4) | |
| Discharge home | No | All | 9151 (25%) | 5.58 (3.87) | 1.8 (1.2) | 56.9 (47.3) | 37.3 (22.7) |
| No | <3 d | 1142 (3%) | 1.8 (0.4) | 1.26 (0.51) | 43.1 (24.6) | 60.5 (26.6) | |
| No | ≥3 d | 8009 (22%) | 6.11 (3.84) | 1.87 (1.25) | 58.9 (49.4) | 34 (20.1) | |
| Yes | All | 27,005 (74%) | 2.05 (1.97) | 1.31 (0.69) | 45.3 (31.7) | 67.7 (27.4) | |
| Yes | <3 d | 21,127 (58%) | 1.32 (0.47) | 1.18 (0.44) | 41.3 (23) | 75.2 (23.3) | |
| Yes | ≥3 d | 5878 (16%) | 4.68 (2.87) | 1.76 (1.11) | 59.7 (49.5) | 40.7 (23.6) | |
| Missing | All | 144 (0.4%) | 9.06 (7.18) | 1.68 (1.49) | 51 (59.5) | 32.5 (33.7) | |
| Missing | <3 d | 26 (<0.1%) | 1.62 (0.5) | 1.23 (0.43) | 35.4 (17.1) | 81.7 (48.5) | |
| Missing | ≥3 d | 118 (0.3%) | 10.7 (6.93) | 1.78 (1.62) | 54.4 (64.8) | 21.7 (15.1) | |
| Activity improved | No | All | 8348 (23%) | 4.83 (4.01) | 2.47 (1.02) | 81.6 (47) | 66.8 (36.3) |
| No | <3 d | 2631 (7%) | 1.75 (0.43) | 2.1 (0.35) | 72.2 (28.3) | 105 (31.5) | |
| No | ≥3 d | 5717 (16%) | 6.25 (4.12) | 2.65 (1.17) | 85.9 (52.8) | 49.3 (22.4) | |
| Yes | All | 2128 (6%) | 3.7 (3.37) | 2.52 (1.12) | 87.7 (55) | 81.4 (35.6) | |
| Yes | <3 d | 1050 (3%) | 1.78 (0.41) | 2.12 (0.41) | 73.9 (27.6) | 105 (29.4) | |
| Yes | ≥3 d | 1078 (3%) | 5.57 (3.89) | 2.91 (1.42) | 101 (69.7) | 58.4 (24.2) | |
| Missing | All | 25,824 (71%) | 2.31 (2.33) | 1.01 (0.12) | 34.3 (15.9) | 55.9 (24.9) | |
| Missing | <3 d | 18,614 (51%) | 1.27 (0.44) | 1.01 (0.08) | 35.2 (15.9) | 68.5 (16.4) | |
| Missing | ≥3 d | 7210 (20%) | 5 (2.98) | 1.01 (0.17) | 31.8 (15.8) | 23.5 (8.46) |
LOS, length of stay; OT, occupational therapy.
The proportion unit is calculated as percentages.
Decision tree model
Our prediction models for predicting readmission within 30 days after discharge, home discharge, and minimal detectable improvement in daily activities showed good to excellent performance metrics (accuracy, 0.69–0.91; sensitivity, 0.67–0.92; specificity, 0.66–0.91; and area under the curve, 0.71–0.94). Refer to appendix 3 for additional information.
Readmission within 30 days
Our pruned decision tree model for predicting 30-day readmission determined several patterns related to OT services as follows: (1) patients with LOS <3 days, CCI score ≥4, and a proportion of days with OT visits <67% were at an increased risk of readmission; (2) for patients with LOS <3 days and CCI <4 who underwent surgical intervention, a proportion of days with OT visits <39% was associated with higher readmission risk; and (3) a complex pattern emerged for patients with LOS ≥3 days and CCI <2. Readmission risk increased when the following factors were present: proportion of days with PT visits ≥36%, <9.7 score improvement in AM-PAC 6-Clicks mobility, PT-designated complexity at evaluation as low or high, ≥1.8 hours between the last PT-recorded AM-PAC 6-Clicks mobility score evaluation and discharge, and the total minutes of completed OT visits ≥26 minutes.
To illustrate, the first pattern indicates that patients with short hospital stays (<3d) and high comorbidity burden (CCI≥4) are at an increased risk of 30-day readmission if they receive OT services on <67% of their hospitalized days. This finding suggests that more frequent OT interventions may be beneficial for high risk patients with short hospital stays. See figure 1 for additional information.
Fig 1.
Decision tree for patient readmission within 30 days after hospital discharge. OT, occupational therapy; PT, physical therapy.
Home discharge
The pruned model excluded OT-related variables. Instead, the model selected the following variables: the most recent physical therapist−recorded AM-PAC 6-Clicks mobility scores, LOS, available assistance at home, living conditions, prior functional level of independence, age, and total minutes of all completed PT visits. See figure 2 for additional information.
Fig 2.
Decision tree for predicting patient home discharge.
Minimal detectable improvement in daily activities
Our pruned decision tree model for predicting minimal detectable improvement in daily activities identified several patterns involving OT services (see fig. 3): (1) patients were more likely to achieve minimal detectable improvement in daily activities when the following conditions were met: the proportion of hospitalized days with a PT visit was <86%, the first AM-PAC daily activity score was <30, and the total number of completed OT visits was ≥3; (2) for patients with a proportion of hospitalized days including a PT visit ≥86%, the likelihood of experiencing minimal detectable improvement in daily activities increased when the time to record the first PT-recorded 6-Click mobility score from admission was <9.6 hours, and the proportion of hospitalized days with an OT visit was ≥82%; and (3) the likelihood of achieving minimal detectable improvement in daily activities increased when all of the following factors were present: the proportion of hospitalized days including a PT visit was ≥86%, the time to record the first PT-recorded 6-Click mobility score from admission was ≥9.6 hours, the patient was female, age was <86 years old, the first AM-PAC daily activity score was <38, the total minutes of all completed OT visits were ≥39 minutes, and prehospitalization AM-PAC daily activity score was ≥24.
Fig 3.
Decision tree for predicting minimally detectable changes in daily activity abilities. Primary payor: 1, traditional Medicare; 2, Medicare HMO; 3, Medicaid; 4, commercial; 5, self-pay; 6, other; OT, occupational therapy; PT, physical therapy.
Discussion
This study highlights the importance of the duration and frequency of OT sessions during acute hospital stays for patients with orthopedic conditions. A higher proportion of hospitalized days that included an OT visit was an important predictor for nonreadmission within 30 days after hospital discharge and minimal detectable improvement in daily activities.
The higher likelihood of readmission within 30 days after hospital discharge was associated with a lower proportion of hospitalized days that included an OT visit. Our model predicted that patients with orthopedic conditions who stayed <3 days in the hospital and had a moderate to high risk of mortality (a CCI score of 4 or higher) would be readmitted if <67% of their hospitalized days included an OT visit. Additionally, having 39% or more of their hospitalized days including an OT visit was identified as a readmission cutoff point for patients with orthopedic conditions who stayed <3 days in the hospital, presented with a no to moderate risk of mortality (a CCI score of lower than 4), and had surgery.
However, a higher likelihood of readmission within 30 days after hospital discharge was associated with a higher total number of minutes of completed OT visits. Our results indicated that a subset of patients with 26 or more total minutes of completed OT visits were more likely to be readmitted. This finding aligns with a previous study by Edelstein et al,3 which identified that although patients receiving OT sessions were more likely to be readmitted, those who received a higher frequency of OT sessions had significantly lower odds of readmission. We speculate that this could be because of the triage system within hospitals. The longer OT sessions might have been allocated to patients with a higher likelihood of readmission. Our analysis demonstrated that patients with more severe conditions received longer OT sessions. A careful review of these results and further investigation are needed to develop any practice recommendations.
All 3 OT predictors were identified as important variables for minimal detectable improvements in daily activities. The cutoff points were as follows: (1) 3 or more OT visits; (2) OT visits on 82% or more of hospitalized days; and (3) a total of 39 or more minutes of completed OT visits. These criteria differentiated whether patients with orthopedic conditions experienced minimal detectable improvements. These results are expected because OT often includes training in daily activities. Notably, having 3 or more OT sessions was effective for patients with an initial daily activity score below 30 in isolation from other contributing factors. Similarly, 39 or more total minutes of OT visits were effective for patients with an initial daily activity score below 38 and a prehospital daily activity score of 24 or higher. This suggests that OT sessions were particularly effective for patients with moderate initial functional capacity, allowing for detectable improvement in daily activity functions.
On the other hand, none of the 3 OT predictors were included in the list of predictors for patient home discharge. Instead, having the last physical therapist−recorded AM-PAC 6-Clicks mobility scores below 40 or equal to or higher than 40 was the most critical factor distinguishing between discharge to home or nonhome locations. This finding aligns with expectations because this factor often helps determine eligibility for postacute care transfer.11,12 Environmental factors such as available assistance at home and living conditions were also identified as predictors for patient home discharge. This supports the well-known role of caregivers in helping patients manage medications, postsurgical care, daily activities, and coordinating follow-up appointments.13–15 It also indicates the importance of collaborative discharge planning and caregiver education for home discharge.14,16 Despite OT and PT services (except PT duration) not emerging as predictors for home discharge in our study, additional research is necessary to examine the influence of acute rehabilitation services on discharge disposition. This investigation is particularly important considering that OT and PT interventions are specifically designed to enhance functional independence, which is a critical factor in discharge planning decisions.
Study limitations
It is important to emphasize that the results of this study are based on the current OT allocations with a limited number of OT practitioners from 1 health care system. The findings from this study cannot be generalized to other health care systems or settings. Although patients with orthopedic conditions received 3 PT sessions on average, they received 1.4 OT sessions on average. This indicates that there were fewer available OT sessions for patients than PT sessions. Additionally, more than 70.8% of patients with orthopedic conditions received only 1 session, with 71.1% not receiving follow-up evaluations. Accordingly, our results for minimal detectable improvements represent the remaining 28.9% of patients and may not represent all patients with orthopedic conditions. It should also be noted that the referenced cutoff score for minimal detectable improvements in daily activities was developed for patients with neurologic conditions.10 Different cutoff scores might result in inconsistent outcomes.
Our study may have underestimated readmission rates because we could only track readmissions within our 11-hospital system. Patients readmitted to hospitals outside this network were not captured in our analysis. Additionally, our analysis was subject to practical constraints that limited data collection to patients who received OT. Although this created a more restricted study population, this focus may have strengthened our methodology by eliminating potential confounding variables associated with patients who were ineligible for or declined OT services.
Moreover, the specific content of each session may play a more crucial role in determining outcomes than the total number, duration, or frequency of the sessions.17 Unfortunately, the content of these OT interventions is recorded in clinical text notes, necessitating extensive machine learning techniques for thorough analysis. Further evaluation of the content of OT sessions is needed to provide insights into potential strategies to improve patient outcomes.
Conclusions
Our study found that readmission rates and functional improvements for patients with orthopedic conditions vary with OT service patterns (number, duration, and frequency of sessions). These findings suggest certain patients may benefit from prioritized, more frequent OT to optimize outcomes. This optimization could occur to mitigate broader hospital resource constraints while considering all diagnostic groups requiring OT services. Current orthopedic OT provision may already reflect system-wide prioritization decisions. Further research is needed to determine distribution strategies across diverse patient populations and conditions to maximize benefits within limited OT resources.
Acknowledgments
Supported by the Learning Health System Rehabilitation Research Network (LeaRRn) Scholar Program through the Eunice Kennedy Shriver National Institute of Child Health and Human Development and the National Institute of Nursing Research (grant no. 1P2CHD101895–01) and the K12 program from the National Institutes of Health Eunice Kennedy Shriver National Institute of Child Health and Human Development (grant no. K12 HD055929). Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH.
List of abbreviations:
- AM-PAC
Activity Measure for Postacute Care
- CCI
Charlson comorbidity index
- LOS
length of stay
- OT
occupational therapy
- PT
physical therapy
Appendix 1. Demographic, Environmental, and Clinical Factors of Patients with Orthopedics: Categorical Variables
| Variable | Description | Category | N* | Total* |
|---|---|---|---|---|
| Surgery | Surgery during hospital admission | No | 1571 (4%) | 36,300 (0%) |
| Yes | 34,729 (96%) | |||
| Encounter type† | Patient classification status | Inpatient | 24,898 (69%) | 36,300 (0%) |
| Observation | 11,602 (31%) | |||
| ICU | In the ICU for at least 48 h | No | 35,709 (98%) | 36,300 (0%) |
| Yes | 591 (2%) | |||
| Mortality level | Ordered categorical variable of the mortality risk modifier according to the All Patients Refined Diagnosis Related Groups schema | None or Minor | 15,846 (64%) | 24,893 (31.4%) |
| Moderate | 5691 (23%) | |||
| Major | 2626 (11%) | |||
| Extreme | 730 (3%) | |||
| Severity level | Ordered categorical variable of the illness severity modifier according to the All Patients Refined Diagnosis Related Groups schema | None or Minor | 7994 (32%) | 24,893 (31.4%) |
| Moderate | 12,239 (49%) | |||
| Major | 3707 (15%) | |||
| Extreme | 956 (4%) | |||
| Living | Patient’s living environment | Alone | 8,061 (23%) | 35,337 (2.7%) |
| With caregiver | 25,775 (73%) | |||
| In facility | 1501 (4%) | |||
| Stairs | Stairs to enter the home | No | 7247 (21%) | 34,722 (4.3%) |
| Yes | 27,475 (79%) | |||
| Home style | Home setup | First-floor setup | 21,001 (66%) | 31,609 (12.9%) |
| Split level | 1750 (6%) | |||
| Multilevel | 8858 (28%) | |||
| Prior function | Prehospital level of function (according to patient or family report) as recorded in Epic flowsheet pick list | Independent | 27,426 (77%) | 35,505 (2.2%) |
| IADL dependent | 3882 (11%) | |||
| ADL dependent | 4197 (12%) | |||
| PT complexity | PT-designated complexity at time of evaluation | Low | 27,239 (75%) | 36,222 (0.2%) |
| Moderate | 8578 (24%) | |||
| High | 405 (1%) | |||
| Readmission | Readmitted to a CCHS hospital within 30 d of hospital discharge | No | 35,477 (98%) | 36,300 (0%) |
| Yes | 823 (2%) | |||
| Discharge home | Discharged from the hospital to home | No | 9151 (25%) | 36,156 (0.4%) |
| Yes | 27,005 (75%) | |||
| Activity improved | Minimally detectable improvement in daily activities | No | 8348 (80%) | 10,476 (71.1%) |
| Yes | 2128 (20%) | |||
| Mobility improved | Minimally detectable improvement in basic mobility | No | 12,474 (42%) | 29,563 (18.6%) |
| Yes | 17,089 (58%) |
We differentiated between patients admitted under inpatient status versus those classified as observation status. Observation status was not a prerequisite for inpatient admission. Initially, the year was included as a covariate to examine the potential impact of COVID-19. However, during model development and tuning, we observed negligible variance in both predictor variables (proportion, total duration, and frequency of occupational therapy minutes) and outcome measures across different years. Given this limited temporal variability in our target variables, we excluded year as a covariate in the final model to optimize parsimony.
Abbreviations: ADL, activities of daily living. ICU, intensive care unit; IADL, instrumental activities of daily living; CCHS, 11 Cleveland Clinic Hospitals.
Percentage of missing responses.
Patient classification for admission.
Appendix 2. Demographic, Environmental, and Clinical Factors of Patients with Orthopedics: Continuous Variables
| Variable | Description | Mean (SD) | Total* |
|---|---|---|---|
| Patient age | Patient’s age (y) at hospital admission | 66.7 (13.2) | 36,300 (0%) |
| Charlson comorbidity index score | Charlson comorbidity index score | 2 (2.5) | 36,300 (0%) |
| Prehospital daily activity score | Prehospital level of function recorded as a score based on the AM-PAC 6-Clicks daily activity short form | 23.5 (2) | 13,332 (63.3%) |
| Prehospital daily mobility score | Prehospital level of function recorded as a score based on the AM-PAC 6-Clicks basic mobility short form | 23.3 (2.3) | 13,813 (61.9%) |
| Hours from admission to PT evaluation | Duration (h) from admission to first PT-recorded AM-PAC 6-Clicks basic mobility score | 26.2 (28.8) | 36,300 (0%) |
| Hours from PT evaluation to discharge | Duration (h) from last PT-recorded AM-PAC 6-Clicks basic mobility score to discharge | 11.6 (21.6) | 36,300 (0%) |
| First mobility score | First AM-PAC 6-Clicks basic mobility T score recorded by PT | 39.8 (8.1) | 36,262 (0.1%) |
| Last mobility score | Last AM-PAC 6-Clicks basic mobility T score recorded by PT | 44.9 (8.9) | 29,601 (18.5%) |
| Mobility score change | Change in PT-recorded AM-PAC 6-Clicks basic mobility T scores | 6.6 (6.3) | 29,563 (18.6% |
| First daily activity score | First AM-PAC 6-Clicks daily activity T score recorded by OT | 42.4 (8) | 35,482 (2.3%) |
| Last daily activity score | Last AM-PAC 6-Clicks daily activity T score recorded by OT | 40.3 (6.7) | 10,476 (71.1%) |
| Activity score change | Change in OT-recorded AM-PAC 6-Clicks daily activity T scores | 2.6 (4.5) | 10,430 (71.3%) |
| PT proportion | Proportion of hospitalized days with a PT visit | 120.1 (59) | 36,300 (0%) |
| No. of completed PT visits | No. of completed PT visits | 3 (1.9) | 36,300 (0%) |
| Total PT minutes | Total duration (min) of all completed PT visits | 108.6 (73.7) | 36,300 (0%) |
| OT proportion | Proportion of hospitalized days with an OT visit | 59.9 (29.5) | 36,300 (0%) |
| No. of completed OT visits | No. of completed OT visits | 1.4 (0.9) | 36,300 (0%) |
| Total OT minutes | Total duration (min) of all completed OT visits | 48.3 (36.7) | 36,300 (0%) |
Abbreviations: AM-PAC, Activity Measure for Postacute Care; OT, occupational therapy; PT, physical therapy.
Percentage of missing responses.
Appendix 3. Methodology and Performance Metrics for Decision Tree Model Selection and Evaluation
Decision tree model
The decision tree model approach is statistically sound and excels in simplifying complex relationships between variables by creating meaningful subgroups, making it easy to interpret and understand. As a nonparametric approach, it does not rely on distributional assumptions and can effectively handle missing values, heavily skewed variables, and outliers without requiring imputation or transformation. These advantages make decision trees particularly valuable for classification, prediction, interpretation, and data manipulation tasks in medical studies.15 This method works like a flowchart, breaking down information into smaller, more manageable parts. To be specific, for categorizing patients based on their likelihood of being readmitted, we started with the most important factor (the lowest Gini Index), which helped us create the most distinct groups. Then, we kept dividing these groups further based on other factors until we could not split them anymore. This created a tree-like structure where each branch represents a decision. To make sure our “tree” was not overly complicated, we used relative cross-validation error for each subtree, from smallest to largest, to determine the optimal complexity parameter for the model. We then validated our model with testing data to ensure our results were both reliable and generalizable. For this process, patient data were randomly partitioned into training (70%) and testing sets (30%). The training set was used to develop the algorithm, whereas the testing set was employed to validate the results. For each decision tree model of our 3 outcomes, we constructed multiple balanced datasets by randomly selecting patients from the majority class to match the number in the minority class. This approach ensured that each outcome had an equal chance of being predicted, preventing the models from being biased toward the more common outcomes.
Decision tree model for readmission within 30 days
Among 36,300 patients with orthopedic conditions, a total of 823 were readmitted within 30 days after hospital discharge. To create a balanced dataset, we randomly selected an equal number of patients (823) from the 35,477 who were not readmitted within the same period, resulting in a total sample size of 1646 patients for our decision tree model (1152 for the training dataset and 494 for the testing dataset). Using the training dataset, we developed an initial model and determined the optimal complexity parameter to be 0.00608. We then validated the pruned model using the testing dataset. The pruned model’s performance metrics were good, with an area under the curve of 0.71, a sensitivity of 0.71, a specificity of 0.66, and an overall accuracy of 0.69 at a threshold of 0.62.
Decision tree model for home discharge
Among 36,300 patients with orthopedic conditions, 9150 were discharged to nonhome locations such as skilled nursing facilities and inpatient rehabilitation facilities. To create a balanced dataset, we randomly selected an equal number of patients (9150) who were discharged home from the remaining 27,150 patients. This resulted in a total sample size of 18,300 patients, which was divided into a training dataset of 12,810 patients and a testing dataset of 5490 patients. Using the training dataset, we developed an initial model and determined the optimal complexity parameter to be 0.001. The pruned model demonstrated excellent performance metrics with an area under the curve of 0.94, a sensitivity of 0.92, a specificity of 0.91, and an overall accuracy of 0.91 at a threshold of 0.51.
Decision tree model for minimal detectable improvement in daily activities
Among 36,300 patients with orthopedic conditions, 2218 exhibited minimal detectable improvement in daily activities. To create a balanced dataset, we randomly selected 2218 patients from the 8348 who did not show such improvements. The remaining 25,824 patients lacked this specific information. We allocated 2979 patients to the training dataset and 1277 to the testing dataset. Our initial model for predicting minimal detectable improvement in daily activities determined that the optimal complexity parameter was 0.0021. However, this parameter led to 48 splits, complicating the model’s interpretation. Consequently, we opted for the second-best complexity parameter, which resulted in fewer than 20 splits (0.004). The pruned model achieved good performance metrics, with an area under the curve of 0.73, a sensitivity of 0.67, a specificity of 0.72, and an overall accuracy of 0.70, using a threshold of 0.52.
Footnotes
Disclosures: none.
References
- 1.U.S. Bureau of Labor Statistics. Division of Occupational Employment Wages Statistics. Occupational employment and wages, May 2021 - Occupational therapists. Available at: www.bls.gov/oes/current/oes291122.htm; 2022. Accessed March 13, 2023. [Google Scholar]
- 2.Mozafaripour S The role of occupational therapy in acute care | USAHS. .St. Augustine, FL: University of St. Augustine for Health Sciences; 2020. Available at: https://www.usa.edu/blog/occupational-therapy-in-acute-care/ Accessed July 5, 2023. [Google Scholar]
- 3.Edelstein J, Walker R, Middleton A, Reistetter T, Williams Gary K, Reynolds S. Higher frequency of acute occupational therapy services is associated with reduced hospital readmissions. Am J Occup Ther 2022;76:7601180090. [DOI] [PubMed] [Google Scholar]
- 4.Rogers AT, Bai G, Lavin RA, Anderson GF. Higher hospital spending on occupational therapy is associated with lower readmission rates. Med Care Res Rev 2017;74:668–86. [DOI] [PubMed] [Google Scholar]
- 5.Álvarez EA, Garrido MA, Tobar EA, et al. Occupational therapy for delirium management in elderly patients without mechanical ventilation in an intensive care unit: a pilot randomized clinical trial. J Crit Care 2017;37:85–90. [DOI] [PubMed] [Google Scholar]
- 6.Song YY, Lu Y. Decision tree methods: applications for classification and prediction. Shanghai Arch Psychiatry 2015;27:130–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Loh W. Classification and regression trees. WIREs Data Min Knowl 2011;1:14–23. [Google Scholar]
- 8.Molnar C Interpretable machine learning: a guide for making black box models explainable. 2nd ed. Munich: Christoph Molnar; 2022. [Google Scholar]
- 9.CDC. Arthritis [Internet]. Chronic disease indicators. .Atlanta, GA: CDC; 2024. Available at: https://www.cdc.gov/cdi/indicator-definitions/arthritis.html Accessed July 2, 2024. [Google Scholar]
- 10.Jette DU, Stilphen M, Ranganathan VK, Passek SD, Frost FS, Jette AM. Validity of the AM-PAC “6-Clicks” inpatient daily activity and basic mobility short forms. Phys Ther 2014;94:379–91. [DOI] [PubMed] [Google Scholar]
- 11.Mo KC, Schmerler J, Olson J, et al. AM-PAC mobility scores predict non-home discharge following adult spinal deformity surgery. Spine J 2022;22:1884–92. [DOI] [PubMed] [Google Scholar]
- 12.Hadad MJ, Orr MN, Emara AK, Klika AK, Johnson JK, Piuzzi NS. PLAN and AM-PAC “6-Clicks” scores to predict discharge disposition after primary total hip and knee arthroplasty. J Bone Joint Surg 2022;104:326–35. [DOI] [PubMed] [Google Scholar]
- 13.Hahn-Goldberg S, Jeffs L, Troup A, Kubba R, Okrainec K. We are doing it together”; the integral role of caregivers in a patients’ transition home from the medicine unit. PLoS One 2018;13:e0197831. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Topham EW, Bristol A, Luther B, Elmore CE, Johnson E, Wallace AS. Caregiver inclusion in ideal discharge teaching: implications for transitions from hospital to home. Prof Case Manage 2022;27:181–93. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Dossa A, Bokhour B, Hoenig H. Care transitions from the hospital to home for patients with mobility impairments: patient and family caregiver experiences. Rehabil Nurs J 2012;37:277–85. [DOI] [PubMed] [Google Scholar]
- 16.Geng F, McGarry BE, Rosenthal MB, Zubizarreta JR, Resch SC, Grabowski DC. Preferences for postacute care at home vs facilities. JAMA Health Forum 2024;5:e240678. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Johnson JK, Rothberg MB, Adams K, et al. Association of physical therapy treatment frequency in the acute care hospital with improving functional status and discharging home. Med Care 2022;60:444–52. [DOI] [PMC free article] [PubMed] [Google Scholar]