Public health concerns regarding opioid medications persist, and health care systems are currently seeking solutions to the ongoing epidemic [1]. In 2017, the rate of drug overdose deaths involving opioids in the United States increased by 12%, totaling 47,600 cases, and prescription opioid medications accounted for over 17,000 of these [2]. Even small amounts of additional daily opioid utilization (10 morphine milliequivalents [MME]) by patients can lead to an increased risk of long-term misuse [3]. In addition, every week of continued opioid utilization represents an increased risk of eventual misuse by patients [3]. Previous studies have found that orthopedic trauma patients use a decreasing amount of opioid medication in the first 2 postoperative weeks, with 6-15 days being the optimal opioid use period [4,5]. In line with these findings, previous studies have used a 2-week postoperative period to assess opioid medication consumption in surgical patients [4].

Patient-reported outcomes (PROs) allow patients to quantify aspects of their orthopedic condition in a standardized fashion [6,7]. These are important tools for determining the efficacy of health care treatments and assessment of clinical research and can be used in determining compensation for health care services provided [6-8]. The National Institutes of Health developed the Patient-Reported Outcome Measure Information System (PROMIS) tools to advance PROs by creating question banks that could be used for many major health issues [9]. The PROMIS Pain Intensity 1A Short form, PROMIS Pain Intensity 3A Short form, PROMIS Pain Interference 8A Short form, and PROMIS Emotional Distress-Anxiety 8A Short form all employ a fixed low number of questions that are highly reliable when compared with their respective domain’s full item bank, making them excellent tools for both patients and clinicians [9,10].

Acceptance and commitment therapy (ACT) is a cognitive contextual behavioral therapy that employs a pragmatic approach to help individuals decrease pain and live according to self-identified personal values [11,12]. The goal of ACT is to augment an individual’s psychological flexibility, thus improving their life according to 6 core cognitive processes: acceptance, defusion, contact with the present moment, self-as-context, values, and committed action (Table 1) [11]. Employing these cognitive processes to increase psychological flexibility allows people to choose their actions based on what they value most, resulting in decreased avoidance behaviors and negative cognitive associations [11]. ACT has proven to be effective across multiple studies and patient populations in the treatment of pain [13]. Several studies report a high value for ACT in the management of chronic pain when compared with standard pharmacological treatment alone [13-16]. Moreover, earlier cessation of pain and opioid utilization in at-risk orthopedic surgery patients receiving office-based ACT interventions has also been reported [17]. However, traditional ACT interventions require a clinic-based, interdisciplinary team approach, which is not always feasible for both patients and health care systems [17,18].

Evolving communication methods, such as automated mobile phone messaging [4,19-21], for health care purposes are increasingly important, as patients prefer these communication methods for delivering and receiving medical information [22]. Software-driven, automated mobile phone messaging robots (also called chatbots or conversational agents) are low-cost tools that can deliver predefined text-based information and receive incoming responses with high reliability when patients either prefer it or it is necessary to communicate at distance [19,23]. This technology demonstrates high efficacy as part of the treatment of conditions ranging from hypertension to substance abuse [24-26], and it has also proved effective in increasing perioperative communication after hip and knee arthroplasty [27] and collecting pain and opioid medication data from patients following orthopedic trauma and hand procedures [4,19]. In addition, automated mobile phone messaging robots have been validated in the collection of PROs in patients undergoing orthopedic hand [20] and hip preservation procedures [23]. Although mobile phone messaging robots (Chat bots) provide the benefit of communicating with patients at distance with no need for human intervention, they do introduce delivery of health care that lacks human interaction with unknown effects [28].

Health care teams caring for patients with traumatic orthopedic injuries have traditionally used opioid medication in the postoperative setting, and these patients are at risk for prolonged opioid utilization in the postoperative period. We theorized that the combination of ACT delivered via automated mobile phone messaging may help to decrease pain and opioid utilization in the acute postoperative setting. The aim of this prospective randomized controlled trial was to evaluate the effectiveness of ACT delivered via an automated mobile messaging robot on (1) decreasing early postoperative opioid utilization and (2) pain-related PROs in the first 2 weeks following surgery for acute traumatic orthopedic injuries.

This randomized controlled trial was registered with ClinicalTrials.gov (NCT03991546) and reporting is consistent with the Consolidated Standards of Reporting Trials guidelines (Multimedia Appendix 1) [29]. The study was performed at a single center university hospital in Iowa City, Iowa, United States. Ethical approval of this study was provided by the University of Iowa institutional review board, and the study was determined to be Health Insurance Portability and Accountability Act compliant.

Adults presenting to a university hospital level 1 trauma center indicated for operative fixation of a traumatic upper or lower fracture were considered for the study (Table 2). The exclusion criteria are listed in Textbox 1. Eligible patients were approached before surgery by a research assistant in a private room. Individuals not excluded by screening questions and interested in participating underwent the informed consent process (Textbox 1). During consent, all subjects were informed of the outcomes of interests, different study arms, and that no changes would be made to their care in terms of postoperative medication, regardless of study participation.

Participants were randomized to either the control or intervention group using a standard web-based random number generator with a range set from 1 to 10 and a 1:1 ratio by a research assistant. Owing to the nature of this study, subjects and the enrolling research assistant were not blinded to the participant’s study group following randomization.

At the time of consent, subjects were required to complete paper forms comprising a basic demographics questionnaire and baseline PROs consisting of the PROMIS Pain Intensity 1A Short form, PROMIS Pain Intensity 3A Short form, PROMIS Pain Interference 8A Short form, and PROMIS Emotional Distress-Anxiety 8A Short form (Multimedia Appendix 2). Following completion of all PROs, participants were randomized to their study group. Subjects who received an odd number from the 1 to 10 range set on the random number generator were placed in the intervention group, whereas subjects given an even number were placed in the control group.

The intervention group received twice-daily, text-based mobile messages communicating an ACT-based intervention for the first 2 weeks following surgery (Multimedia Appendix 3). Control group subjects did not receive the ACT intervention or any other form of mobile message communication. The mobile messaging ACT protocol consisted of twice per day mobile messages, morning and evening, starting on postoperative day (POD) 1 and ending on POD 14. These mobile phone messages provided participants with an ACT-based intervention that was developed in collaboration with a pain psychologist (VK) specializing in ACT for chronic pain. These messages used all the principles presented in Table 1 with the objective of helping recipients understand and develop better coping skills in relation to their postoperative pain. An example message from day 1 is as follows:

Outside of the mobile messaging intervention, both groups received the same standard postoperative care, health care team communications, and instructions for completing the study follow-up.

A chart review was performed to collect demographic information such as subject age, comorbid conditions, and preoperative outpatient opioid medication prescriptions for treatment of their current traumatic orthopedic injury. All subjects, regardless of group, were seen by a research team member after surgery to review which of their discharge medications was the medication of interest for the study and to confirm that the intervention group subjects received their first mobile phone ACT message. Participants in both groups were instructed to have their opioid medication bottle available at follow-up to confirm their opioid tablet consumption. Owing to the changes in health care teams, staff preferences, and allergies, the opioid pain medications administered at discharge were not standardized between study groups. Following discharge on POD 14, subjects were contacted by phone or seen in the clinic by the research team for follow-up. At this time, the subjects’ opioid pain medication consumption was assessed, and they completed a second set of PROs.

The primary outcome of this study was the amount of opioid pain medication consumed by subjects, and the secondary outcomes analyzed were net changes from baseline PRO scores at the 2-week follow-up.

The method that participants employed to report their opioid medication consumption and how PROs were captured during follow-up were recorded (Table 3). Subjects using their pill bottle to confirm the remaining number of opioid pain medication tablets from their discharge prescription on POD 14 were denoted as reporting a pill count. Cases where subjects or their care facility kept a log of tablet consumption were classified as reporting a daily log. Subjects reporting the number of tablets they used without the use of a log or pill count were designated as providing an estimate. The percentage of opioid pain medication used, total MME, and percentage of available MME consumed were calculated. The mean number of opioid pain medication tablets and MME used by the subjects were compared between groups. The raw scores for PROs were converted to corresponding t-scores using the appropriate PROMIS scoring manual [30]. The changes in PROs from baseline to POD 14 were also calculated by subtracting POD 14 scores from baseline scores, as higher t-scores signify a poorer outcome; thus, lower scores on POD 14 indicate an improvement from baseline PROs. The mean PRO scores and changes were compared between the groups.

Participant characteristics were described using mean (SD) or median (minimum to maximum) for continuous variables and frequencies (percentages) for categorical variables. Visual review of histograms and the results of the Shapiro-Wilk test of continuous variables revealed that only age and BMI were not normally distributed. Between-group differences were evaluated using t tests or Wilcoxon rank-sum tests (age and BMI) for continuous variables and chi-square or exact tests for categorical variables, as appropriate.

To evaluate whether the intervention versus control group had a lower opioid use on average, we determined the number of tablets and MME taken in each group and compared means using t tests. Using a previous study of opioid medication usage in orthopedic trauma patients [4], the sample size estimated to observe a 30% decrease in opioid utilization among 2 groups required a total of 74 subjects to achieve 80% power at an alpha of .05. The percent decrease is calculated using the formula . A separate power analysis was calculated for the PRO portion of the study, and it was determined that a total of 36 subjects would provide 80% power to detect a 10-point difference (1 standard deviation) in t-scores for the PROMIS instruments at an alpha level of .05. Statistical analyses were performed using SAS software version 9.4 (SAS Institute, Inc).

A total of 125 individuals were approached regarding the study over the 5-month enrollment period between February 2019 and June 2019. Of the 125 individuals, 2 patients were excluded at this time, as they were non–English-speaking, and an additional 24 patients were excluded because they did not use mobile phone messaging or did not have a personal mobile phone. This resulted in a total of 99 eligible people who were presented the study, 17 of whom declined participation (Figure 1). Overall, 82 subjects were enrolled, and 6 dropped from the study after providing consent because of various issues: one patient lost to follow-up, one patient withdrew at follow-up, one patient had incomplete follow-up, one patient’s operative plan changed to arthroplasty, and 2 subjects remained inpatient for over 7 days of the study period (Figure 1). This resulted in a final population of 76 subjects (38 per study group). The enrollment period concluded once a powered sample for the primary aim was obtained. A breakdown of the subjects enrolled, reasons for excluding subjects, and subjects removed from the study after consent are presented in Figure 1. Participant demographics for the intervention and control groups are presented in Table 3. The analyses of all collected demographic factors showed no differences between the intervention and control groups in all factors such as subject age (P=.42), current psychiatric diagnosis (including depression, anxiety, bipolar type 1, obsessive-compulsive disorder, posttraumatic stress disorder, panic disorder, and attention-deficit disorder; P=.14), or substance abuse history (P=.10; Table 3). Furthermore, no differences between groups were found for injury type, disposition following discharge, method for reporting opioid medication consumption, preoperative opioid medication prescriptions, or preoperative PROs (Table 3).

No differences between groups were observed in the amount of opioid medication tablets or MME prescribed at discharge (tablets for the ACT group: mean 58.8, SD 27.3 vs tablets for the control group: mean 61.6, SD 22.0). A further breakdown of the medications prescribed to subjects within the study period is presented in Table 4. The mean opioid tablet use for subjects in the ACT-based intervention group was 26.1 (SD 21.4) tablets, whereas the control group used a mean of 41.1 (SD 22.0) tablets, resulting in 36.5% less tablets used by subjects receiving the ACT-based intervention (P=.004; Table 5). Similarly, subjects in the intervention group consumed a mean of 199.9 (SD 163.2) MME, on average, whereas the control group subjects consumed a mean of 307.0 (SD 166.0) MME, indicating 34.9% less MME used by subjects in the intervention group (P=.006; Table 5).

PROMIS instrument t-score values for both ACT and control group subjects are presented in Table 6. At 2-week follow-up, the intervention group subjects reported lower postoperative PROMIS Pain Intensity 3A (mean 45.9, SD 7.2) and Pain Interference 8A (mean 56.6, SD 9.4) scores compared with the control group’s postoperative Pain Intensity 3A (mean 49.7, SD 8.8; P=.04) and Pain Interference 8A scores (mean 60.6, SD 8.2; P=.05; Table 6). No differences were observed between groups at 2-week follow-up in the PROMIS Pain Intensity 1A or PROMIS Emotional Distress-Anxiety 8A forms (Table 6).

This randomized trial delivered an ACT-based intervention via an automated mobile messaging robot to postoperative orthopedic patients. The subjects who received the ACT-based mobile phone intervention used a lower number of opioid tablets and consumed less MME in the first 2 weeks after their injury. We also found that the intervention group reported less pain intensity and pain interference at the 2-week follow-up. These data demonstrate that ACT-based automated mobile messaging protocols may be effective in reducing the amount of opioid medication used and may positively affect postoperative PROs in patients undergoing operative fixation of their acute fractures.

Improved mood symptoms, less pain interference, and faster cessation of opioid pain medication are some of the recognized benefits of using ACT in clinic-based, interdisciplinary approaches to pain management after surgery [17,18]. Previous investigations have used automated mobile phone messaging robots to deliver PROs [20,23], improve communication with patients [27,31], deliver postoperative orthopedic care [32], and inquire about pain and opioid utilization [4,19,33]. In this study, we used ACT and a mobile phone messaging robot to assess whether these tools in combination could decrease opioid utilization and improve individuals’ perception of their early recovery from injury. Prior work has demonstrated a quicker time to opioid cessation and a decrease in postoperative opioid utilization (14% less in the ACT group) when used in office ACT-based treatments [17,18]. Subjects receiving the ACT intervention via automated mobile phone messages reported over 36% less opioid tablets and more than 34% less MME consumed than corresponding control subjects who did not receive ACT. Our findings suggest that software-based communication using ACT through a mobile phone has the potential to have a large impact on the utilization of postoperative pain medication by patients in the first weeks after surgery for fractures. Further study is required to determine if these effects are long lasting and to determine which injuries and patients receive the greatest benefit. In addition, future investigations and trials should consider the effect of software delivery of ACT and other behavioral therapies on different cohorts of patients.

PROs, such as PROMIS, allow patients to quantify aspects of their orthopedic condition in a standardized fashion [6,7]. These are important tools for determining the efficacy of health care treatments and assessment of clinical research and can be used in determining compensation for health care services provided [6-8]. The National Institutes of Health developed the PROMIS tools to advance PROs by creating question banks that could be used for many major health issues [9]. We found that despite less utilization of opioids, subjects in the ACT-based intervention group reported less pain intensity and pain interference at 2 weeks. This most likely does not represent a clinically important difference based on the SD methodology used in prior works with PROMIS tools (Table 6), but it at least suggests that the intervention group did not experience greater pain [34]. There were no other differences between study groups in the other domains at the 2-week follow-up. Previous studies have reported that patients who consume more opioid medication report higher pain at both short- and long-term time points, which is reflected in our findings for both PROMIS pain intensity 3A and the employed pain interference measure [35,36]. Future research efforts may benefit from employing alternative PRO measures to identify the effects of ACT-based interventions, including assessment of psychologic flexibility. Future research may also consider possible modifications of our study protocol to include a longer intervention period and more than one follow-up data point. Future work may also consider designing an ACT-based tool that is more focused on demonstrating an effect on PROs.

Several limitations were present in this study. First, we were limited to a single level 1 trauma center, which may affect the reproducibility of our results across other health care settings. Next, the exclusion criteria for this study were extensive, and thus, the results may not be generalizable to the entire scope of orthopedic trauma patients. We attempted to include a diverse set of injuries and yet excluded patients with a high likelihood of confounding problems from open fractures or prolonged initial hospitalization. Future studies assessing the effects of ACT-based interventions similar to ours should aim for less restrictive exclusion criteria to apply this intervention to a larger, more diverse population. The research assistants were not blinded to the patients’ study group. In addition, patients understood the outcomes of interest in this study, which could be susceptible to reporting bias. In addition, participants were not blinded to their treatment group. The lack of blinding could potentially introduce response or reporting bias, making this a potential area of improvement for studies seeking to follow the present methodology. This could be accomplished through the implementation of a control messaging protocol. Moreover, a retrospective chart review was used to obtain several patient factors, including comorbid conditions and dispensing of preoperative outpatient opioid medication prescriptions. The collection of information in this manner relies on accurate charting and transfer of documents from outside institutions, which may have been incomplete.

In this study, delivering an ACT-based intervention via an automated mobile messaging robot in the acute postoperative period decreased opioid utilization in orthopedic trauma patients in the first 2 weeks after their injury. Subjects in the ACT-based intervention group also reported lower pain intensity and pain interference after 2 weeks, although this likely did not represent a clinically important difference. Future studies may apply this intervention in other patient populations to assess its efficacy on a larger scale and may include assessment of pain and opioid use in a longer time frame after injury.