1.3.1 Research Design and Data
Pre-existing data was extracted from medical records at Virginia Commonwealth University’s Massey Cancer Center in Richmond, Virginia (referred to as Massey henceforth) for retrospective analysis. As a large urban medical center and a National Cancer Institute recognized facility, Massey provides inpatient and outpatient services to patients with cancer from a diverse and expansive geographic area [23]. The sole focus of this study was on inpatient (i.e., acute care) occupational therapy services with an investigation into the potential impacts of occupational therapy services on hospital readmissions. Data was extracted for patients who received services at Massey from the dates of January 1st, 2015 through January 1st, 2020. The selected sample dates were chosen to prevent any potential impact from healthcare changes (e.g., COVID-19) or major healthcare regulatory actions, while also ensuring a substantial enough sample size to support a proper power study. In order to create a sample size necessary for the statistical analyses, G*Power [24] was used with a two-tailed logistic regression model with an odds ratio of 1.3 with an err prob (α) set at 0.05, desired power (1-β err prob) set at 0.95, with a sample size of 1188 participants necessary for study inclusion.
1.3.2 Ethics
After receiving institutional review board (IRB) approval from Virginia Commonwealth University (VCU), the data collection location, the data extraction occurred without the inclusion of any patient identifiers. Massey’s health informatics and information technology teams supported data extraction and preservation from Cerner, which was VCU’s electronic health record (EHR) during the intended timeframe.
1.3.3 Participants
Data extraction was based upon pre-selected inclusion and exclusion criteria. To focus the scope of this research, based on national frequency as well as center-specific frequency, six of the most prevalent cancer types were included: breast, blood/hematologic, colorectal/gastrointestinal, gynecologic, lung/respiratory, and prostate/genitourinary. Inclusionary criteria comprised individuals who were aged 18 years or above with a cancer diagnosis who had a qualifying hospital stay (a minimum of a 3-day or longer hospital admission so as to allow for adequate time for the completion of an occupational therapy evaluation and subsequent treatment sessions if needed). Subjects were only included if they had a consistent living environment before their hospital admission with acceptable options including: living at home alone, in the home with others, in long-term supports and services (LTSS), or other (a self-select option chosen by a patient). The term LTSS in the “lives with” grouping was used to include patients who resided at home with caregiver support (non-habitable), in an assisted living facility, or in a nursing home. This allowed for a potential focus on discharge planning and recommendations to assist in returning to a prior living environment. If a patient was admitted for a pre-planned elective surgery (e.g., a total knee replacement), they were excluded from this study.
1.3.4 Outcome Variable
The main outcome of interest was the hospital readmission. Whether or not a patient received occupational therapy services during their hospital stay was the primary variable of effect. Statistical analyses were performed to determine the contributions of OT services to the primary variable of hospital readmission; additional analyses explored this relationship after adjusting for demographic and diagnostic covariates.
1.3.5 Predictor Variables
Variables of interest in the study were analyzed to determine any potential effects on a patient’s chance of a hospital readmission. Our primary predictor of interest was occupational therapy service utilization; a categorical (yes/no) variable indicated by the presence (or absence) at least one occupational therapy code billed in the patient’s account.
Various factors had the ability to influence hospital readmissions and were included in the study as possible covariates. Risk factors in combination with process factors were jointly referred to as covariates within the statistical model and were combined in order to attempt to ascertain any potential relationship to the outcome variable. Cancer type and stage as well as a patient’s discharge location were extracted from Massey’s database in order to determine any potential effect upon the likelihood of readmission. Insurance type was included to ascertain if any insurance types contributed to the likelihood of readmission. Additionally, a patient’s race, ethnicity, age, gender and living environment were included for analysis. In addition, pain levels during sessions were included since uncontrolled pain has been found to have a negative effect upon quality of life and lead to potential readmissions [25]. Lastly, a patient’s admitting diagnosis, which may differ from a patient’s cancer diagnosis, was included to ascertain if any admitting diagnoses had relationships upon patients’ readmissions. A summary of key predictor variables is presented in Table 1.
Table 1
Variable | Coding | Category |
---|
OT services | 0 = Yes 1 = No | Nominal/Categorical |
Readmission within 30 days | 0 = Yes 1 = No | Nominal/Categorical |
Cancer type | 0 = Breast 1 = Blood/Heme 2 = Colorectal/gastrointestinal 3 = Gynecologic 4 = Lung/respiratory 5 = Prostate/genitourinary | Nominal/Categorical |
Lives with | 0 = Home alone 1 = Home with family 2 = LTSS 3 = Other | Nominal/Categorical |
Admitting diagnosis | 0 = Primary admitting diagnosis as cancer 1 = Primary admitting diagnosis not as cancer | Nominal/Categorical |
Cancer stage | 0 = Stage I 1 = Stage II 2 = Stage III 3 = Stage IV | Nominal/Categorical |
Gender | 0 = Female 1 = Male | Nominal/Categorical |
Race | 0 = Asian 1 = Black or African American 2 = Other, American Indian-Alaskan, Multiple, Native Hawaii/Other Pac Island 3 = White | Nominal/Categorical |
Ethnicity | 0 = Hispanic, Latino or Spanish origin 1 = Not Hispanic, or Latino, or Spanish origin | Nominal/Categorical |
Age | Age in years | Scale/Continuous |
Pain at baseline | Numeric | Scale/Continuous |
Pain at discharge | Numeric | Scale/Continuous |
Insurance | 0 = Private/Commercial 1 = Corrections 2 = Indigent 3 = Medicare 4 = Medicaid 5 = Military 6 = Other 7 = Self-pay | Nominal/Categorical |
Discharge location | 0 = Home/Prior living environment without medical assistance (Discharged to home or self-care, including correctional facility) 1 = Home environment with supports (including group home and discharge to home with home health) 2 = Rehabilitation facility 3 = Expired in facility 4 = Hospice 5 = Transfer to psychiatric hospital/psychiatric unit 6 = Additional hospital care 7 = Against Medical Advice (AMA) | Nominal/Categorical |
Note. OT = occupational therapy; LTSS = long-term services and supports; AMA = against medical advice. |
1.3.6 Data Extraction and Cleaning
Data extraction occurred with the assistance of the Massey information technology team and was assessed for missing data. The data was cleaned with variables removed for coding inconsistencies [26]. Complete case analysis was used whereby missing data in any patient category removed the patient from analysis in the study. Descriptive statistics were run on predictor and outcome variables to assess distribution, measures of central tendency, measures of variability, and skewness. No outliers were present within the data.
Before analysis, statistical assumptions were checked and met. In order to have independence of observations, only one admission per patient was able to be included for analysis. In place of a patient’s medical record number (MRN), a patient identification number (ID) was created to protect patient privacy by Massey. The patient’s ID number was then used to assess the number of admissions during the selected time frame. The first hospital admission was retained for statistical analysis and then remaining admissions for a selected individual were removed from the data.
For research question 1, there were 12,159 patients who initially met the inclusion and exclusion criteria for analysis. In order to have independence of events, the number of cases was reduced to 6,614 to encompass only one unique visit (first admission) per individual patient based upon the patient identification number.
For research question two, the data was cleaned and refined according to complete case analysis. Initially there were 12,159 unique admission cases, but 19,113 total observations due to multiple observations that were attributed to each individual patient. As such, further explanation of the refinement process discusses observation numbers and not patient numbers. There were 9,955 observations that were removed from the analysis as a result of not having a cancer stage, and 107 observations who were removed due to missing race specification (unknown-unable to communicate, or unknown-patient refusal). There were 59 observations removed due to missing ethnicity (unknown-patient refusal, unknown-unable to communicate, or N/A-outreach use only) and there were 8 observations removed due to a lack of an appropriate cancer type (designation: All Other). Lastly, there were 3,471 observations without any home information regarding living situation, with 215 without an earliest pain score and 130 observations with no final pain score. After removing those with missing data, the sample was reduced to 5,168 individual patient admissions (including multiple admissions across the timeframe for any particular patient).
The number of participants was further reduced to a final sample size of 1,920 individual patient admissions with the selection of the patient’s first admission within the five-year period to keep independence of observations. The first admission was selected to reduce any potential mitigating effects that could arise from disease progression, such as could occur with a patient’s final admission, during the selected time period.
1.3.7 Statistical Analyses
To determine if receipt of occupational therapy services explained the likelihood of being readmitted within 30 days after discharge (research question one), the independent variable (IV) of receipt of occupational therapy services and the dichotomous dependent variable (DV) of a hospital readmission were analyzed within the context of a crude logistic regression model using SPSS v28. The alpha level was set at 0.05. It was planned that if the significance level was less than 0.05 then the null hypothesis would be rejected.
For question two, an adjusted logistic regression analysis was used to assess if occupational therapy services impacted the odds of readmission after adjusting for demographic and diagnostic factors. Categorical independent variables included: OT services received, cancer type, lives with, admitting diagnosis, cancer stage, gender, race, ethnicity, insurance and discharge location; these categorical variables were dummy coded. Additional independent variables were continuous numeric variables (i.e., age, pain at baseline, and pain at discharge) and were subsequently entered into the model. Alpha was set at 0.05 with the intention that if was p < 0.05 then the null hypothesis would be able to be rejected.