This retrospective study is focused on individuals with opioid use history who accepted virtual PRC services when offered to them during an ED encounter. The study was determined not to meet requirements for human subjects research review by the Indiana University Institutional Review Board (2006108993) since the dataset was limited in nature.
Intervention Description
The intervention of focus is part of a larger telehealth program operated by a single Indiana-based hospital system. Program implementation was supported in part by federal opioid response funding distributed through the Indiana Department of Mental Health and Addiction, with these funds being specifically designated for implementing virtual PRC services for ED patients who use opioids. The telehealth program consists of a centrally located telehealth hub with PRCs and other behavioral health professionals available 24 hours a day to participating EDs: all services of focus in this study were PRC-delivered. ED staff activated telehealth services by calling the telehealth hub for patients with an identified need, based either on the presenting problem (e.g., opioid poisoning, intoxication, withdrawal) or information elicited by ED staff during the course of care. When an opioid use issue is identified, staff bring a cart with a video screen to the bedside where they connect the patient with a PRC. Prior to this point, ED staff have provided minimal, if any, information regarding PRC service. At the initial engagement, the PRC describes the program and inquires whether the patient is interested in services. If the patient expresses interest, the PRC proceeds with a conversation aimed at gathering information on the patient’s current substance use, withdrawal symptoms, previous treatment and recovery attempts/pathways, and current needs/desires for linkage to resources. Additionally, the PRC ensures all information for contacting the patient after discharge is included in the electronic health record. After discharge, the PRC refers the patient to their requested treatment or recovery pathways (e.g., outpatient, inpatient, medication-based treatment, 12-step facilitation, detox, etc.). Post-discharge, a PRC attempts follow-up calls at 48 hours; weeks 1 and 2; months 1, 2, 6, and 9; and 1 year. If unable to reach the patient, the PRC will leave a message, if possible. After 3 consecutive unsuccessful attempts, the PRC will stop trying to contact the patient but, if the patient reinitiates contact with the telehealth hub, services will be continued.
In the Indiana-based hospital system studied in this research, 13 EDs implemented 24-7 virtual PRC services as part of a larger telehealth program. Implementation began in September 2018 with services expanding on a rolling schedule through June 2019. Six of the hospitals were located in cities and 7 were designated rural critical access hospitals. In all participating EDs, the telehealth PRC service was the only program available for patients presenting with opioid use issues.
Data and sample
Measures came from two sources. The first source was a database developed to track telehealth hub services, which included PRC services, resources provided, and outcomes related to baseline and follow-up PRC encounters. Details in this database are recorded by the PRC, with information being pulled from the electronic health record or from the discussion with the patient. The second source was the Indiana Network for Patient Care (INPC), which included electronic health record data from hospitals across the state. This allowed us to follow participating patients across a large number of institutions outside the hospital system of focus [35]. These data broadly included information on patients’ ED encounters and hospital admissions, presenting issues, opioid-related diagnoses based on ICD-10 codes, and discharges.
Our observation window was September 24, 2018 (the date the hub went online at the first ED) through September 2, 2021 (see Table 1 for hospital start dates and total enrollments by hospital and Figure 1 for number of enrollments by month). To be included in the analysis, a patient must have (a) interacted with a virtual PRC during an ED encounter, (b) accepted enrollment into the PRC program, and (c) had a history of opioid use indicated in one or more locations within the available datasets. The final sample comprised 917 patients who collectively engaged in 1,208 baseline (initiating) PRC interactions; some returned to an ED during the follow-up period and re-enrolled in PRC services during the subsequent encounter.
Table 1. Setting, month of first peer telehealth enrollment, and total number of patients by site
Hospital
|
Rural*
|
Month
|
Total number of patients initiating baseline PRC interactions
|
|
Site 1
|
Yes
|
October 2018
|
68
|
Site 2
|
Yes
|
November 2018
|
67
|
Site 3
|
Yes
|
January 2019
|
78
|
Site 4
|
No
|
January 2019
|
17
|
Site 5
|
No
|
March 2019
|
126
|
Site 6
|
No
|
March 2019
|
204
|
Site 7
|
No
|
April 2019
|
148
|
Site 8
|
Yes
|
April 2019
|
37
|
Site 9
|
Yes
|
May 2019
|
18
|
Site 10
|
No
|
May 2019
|
367
|
Site 11
|
Yes
|
May 2019
|
30
|
Site 12
|
No
|
June 2019
|
28
|
Site 13
|
Yes
|
June 2019
|
20
|
*Rural classification is based on hospital designation as a critical access hospital. All other hospitals were located in cities.
|
Variables
The primary outcome—the rate of successful follow-ups recorded in the PRC tracking database—is defined as the number of times a PRC successfully spoke with the patient each month of attempted follow-up. Within the available datasets, this served as the best factor for assessing the PRC program’s success at facilitating collaborative and engaging relationships with patients, a key function of their professional role [33].
Demographic predictors pulled from the INPC database included: patient age, sex (male versus female), race (White versus non-White; Black versus non-Black), and ethnicity (Hispanic versus non-Hispanic). Additional demographic information came from the PRC database and included: employment at baseline (yes versus no), insurance status (self-pay versus insured), and whether the patient lived in a metro, small, or rural area as defined by applying rural-urban commuting area codes to patients’ home zip codes (United States Department of Agriculture, 2020). Also examined were predictors related to the current ED encounter recorded in the PRC database which included: naloxone administration prior to ED encounter as an indication of an opioid poisoning (yes/no); opioids as the primary reason for the ED encounter (yes/no); PRC-provision of referrals (yes/no) and the specific type of referrals provided; and the duration of the initial virtual PRC interaction. The final two predictors included patients’ self-reported route of drug administration recorded in the PRC database (four binary variables indicating intravenous, oral ingestion, smoking, inhalation/snorting), and the number of ED encounters recorded in the year prior to telehealth PRC enrollment (as drawn from the INPC data).
Analysis
First, we examined the sample composition and pattern of follow-up encounters with PRCs with descriptive statistics and exploratory plotting. Next, to assess the degree to which variables of interest predicted the rate of successful follow-up, we applied a Poisson regression model. The regression was multilevel, with rounds of recovery coaching embedded within patients to account for those patients who enrolled in recovery coaching multiple times. We analyzed follow-up attempts as multilevel by embedding them within patient, as patient ID accounted for significant variance in rates of successful follow-up.[1]
We adjusted the model to account for the large degree of variation in duration of the follow-up period among patients. The duration of the attempted follow-up period depended upon on several factors: (1) patient rehospitalization resulting in a new round of PRC services and ending the previous services and follow-up period; (2) patients reaching the end of the one year telehealth service program; (3) patient death; and (4) the study observation window ending during the patient’s follow-up period. Adjusting for these factors was important for distinguishing between time periods in which patients were not contacted due to eligibility versus time periods when patients were eligible but not contacted. To achieve this, we included an offset to represent the number of months (or fractions of months) when follow-up was possible. With this adjustment, the Poisson regression can be interpreted as predicting the number of successful contacts per month of attempted outreach. We did consider a zero-inflation model to account for the large proportion of participants without any successful follow-ups. This option was ultimately rejected because (a) the data were Poisson distributed despite the high proportion of zeros and (b) the binomial component of zero-inflation models does not allow for the inclusion of the offsets described above, which are important for modeling the specific underlying study design.
Additional analyses was performed to determine whether there were overall changes in rates of successful follow-up due to the COVID-19 pandemic, controlling for changes over time. Corresponding with Indiana’s initial state of emergency order, March 6, 2020 was used as the start date for the pandemic. We also added another interaction term and re-ran the Poisson regressions (described above) to assess whether the impact of variables of interest changed following the pandemic’s onset.
All data processing, modeling, and graphing was performed using the R version 4.1.3 [36], with multilevel regression analyses performed using the lme4 R package version 1.1.27.1 [37].
[1] Observations were not embedded within PRC because patients in this program are not assigned to specific PRCs but are instead outreached by any available PRC at the time the patients present at the ED, thereby making this factor inconsistent for patients who present more than once. Additionally, observations were not embedded within hospital site because site effects were not statistically significant. There was an identified significant effect of a single PRC (out of twelve) whose outreach attempts were associated with substantially lower rates of success. This PRC was the only coach who was exclusively assigned to the night shift; therefore, it seems reasonable to interpret the effect as representing a decreased likelihood of successful follow-up for PRC coaching initiated between the hours of 10 pm and 8 am, rather than the effect of the identity of the PRC.