Postsurgical Medication Awareness, Recovery, and Tracking Using a Phone-Based App (SMART-APP): A Randomized Clinical Trial

Abstract

Background:

Excess post-operative opioid use is harmful. We tested the hypothesis that a patient-facing educational smartphone app would permit surgical patients to effectively manage their pain while using fewer opioids after discharge.

Methods:

A two-hospital, single-health system trial exposed adult surgical inpatients randomly to a consumer electronic health app (“educational app”) that provided pain education or to a data collection only app (“control app”). During the second half of the trial, a clinician-facing electronic decision-support tool was activated. Linear regression models with log-transformed outcomes estimated the effect of the educational app on the primary outcome of cumulative self-reported opioid use within four weeks after discharge (in morphine milligram equivalents (MME)) and the secondary outcome of opioids prescribed at discharge.

Results:

We enrolled 711 patients from May 2, 2022, to December 10, 2023. Modified intention-to-treat analysis cohort included 606 patients, with 303 randomized to the educational app and 303 to the control app. Median [first quartile, third quartile] discharge opioids prescribed to both groups were 75 [60, 150] MME. Median [Q1, Q3] self-reported cumulative opioid use in the four weeks after discharge was 22 [0, 98] MME with the educational app and 15 [0, 82] MME with the control app (ratio of geometric means: 1.21; 95% CI: 0.87, 1.69; P = 0.264). The use of the clinician-facing electronic decision-support tool also did not modify opioid consumption.

Conclusions:

Neither a patient-facing educational app nor a clinician-facing decision support tool reduced self-reported opioid consumption among surgical patients with low post-discharge opioid requirements.

INTRODUCTION

The management of postoperative pain is a critical pillar of perioperative care, particularly amidst the ongoing opioid crisis. Despite the widespread implementation of opioid-sparing acute pain management protocols in hospitalized patients, adverse effects of opioid analgesia, including new persistent postoperative opioid use, are common in surgical patients.¹ The majority of surgical patients, 67–92%, report having leftover opioids after discharge, and 42% to 71% of all prescribed opioid tablets go unused by patients,² raising concerns about expanding the available reservoir for nonmedical use. Indeed, in a retrospective cohort study of US veterans, patients discharged from the hospital following surgery had 6.7-fold higher odds of suffering an opioid overdose compared to non-surgical discharges.³ Furthermore, the majority of postoperative opioid overdoses (62%) happened in the first 30 days postoperatively,³ thereby highlighting the need for opioid stewardship programs that not only address pain management during a hospitalization but also extend patient education and promotion of non-opioid therapies to the post-discharge period.⁴

At the population level, harm reduction strategies are often fragmented by locale and frequently insufficient in scope. In response to these challenges, electronic health tools, particularly patient-facing apps, have emerged as promising tools for influencing patient behavior. Indeed, apps have improved medication adherence in patients taking long-term medications.⁵ Perioperative care encounters have been successfully leveraged to positively impact adverse habits such as tobacco or alcohol consumption by building on a “teachable moment”, in which heightened health awareness serves as a platform to elicit behavior change.⁶ Health information technology, including medical apps (i.e., mHealth),⁷ offers the opportunity to extend patient education beyond the reach of the typical typed sheet of paper handed to patients at discharge. However, the effect of pain management apps on opioid intake and comprehensive patient-reported outcomes is conflicting.^8–10 Contemporary evidence mainly comes from small studies evaluating a single surgical subpopulation. While many electronic pain management interventions focus on clinician prescribing,¹¹ our objective was to investigate a scalable solution to reduce patient opioid intake after hospital discharge in a large cohort of surgical patients from multiple different surgical specialties.

We hypothesized that a patient-facing educational app would reduce patient opioid intake while maintaining effective postoperative pain control after discharge. Outcomes collected during the first four weeks after discharge primarily included the cumulative self-reported amount of opioids taken, and secondarily, the amount of opioids prescribed at discharge, need for supplementary opioid prescriptions, disposal of leftover opioids, and pain intensity and interference scores.

METHODS

This study was a two-hospital single-health system (one tertiary hospital and one community hospital part of the same US Midwestern health system) randomized clinical trial in which patients who underwent major inpatient surgery were assigned two separate interventions. The primary intervention, to which individual patients were randomized throughout the trial, was the use of a patient-facing consumer health informatics smartphone-based app (“educational app”). Patients randomized to the control group were provided with an app for data collection only (“control app”). Both groups received usual care, which typically included verbal discharge instructions from the nurse and a written handout. The secondary intervention was a clinician-facing, real-time opioid prescription decision support tool embedded in electronic health records, which was activated halfway through enrollment. Before the prescribing tool intervention, the health system had not implemented formal guidance on opioid prescribing. Given that only the first intervention was randomized, the trial design is referred to as “partial factorial”. The study design followed Pragmatic Explanatory Continuum Indicator Summary guidelines to maximize broad applicability and is reported according to Consolidated Standards of Reporting Trials Extension for Factorial Randomized Trials guidelines.^{12; 13} Institutional Review Board approval was obtained before patient enrollment (UNMC IRB, protocol #0724–21-FB). This trial was registered on January 21, 2022, at ClinicalTrials.gov (NCT05221866; https://clinicaltrials.gov/study/NCT05221866?cond=NCT05221866&rank=1) before the first patient was enrolled on May 2, 2022. Patient consent was obtained before enrollment. Clinician consent was waived due to the minimal risk and because the decision support tool supplemented the current standard of care.

Study design and participants

Patients were eligible for enrollment if they were 19–89 years old, had access to a smartphone, and had inpatient surgery requiring at least an overnight hospitalization with anticipated discharge to home. Patients re-hospitalized within 30 days of a previous hospitalization, pregnant patients, patients unable to read English, patients discharged to a post-acute care facility, and patients with contraindications to opioids, acetaminophen, or nonsteroidal anti-inflammatory agents were excluded. Only patients meeting the Agency for Healthcare Research and Quality definition of long-term opioid therapy (opioid use on most days > 3 months) prior to surgery, not patients reporting any preoperative opioid use, were excluded.¹⁴ Eligibility criteria were not different between factors (patient-facing app and clinician-facing decision support tool). A $10 gift card was offered for completing the survey every week, with a total of $40 in gift cards offered as an incentive for trial completion.

Randomization and masking

Following informed consent, patients were electronically randomized 1:1 to the educational app or a control app with only data collection functions between May 2, 2022, and December 10, 2023. Permuted block randomization (block size 2) was performed using a computer-generated code within a secure central online data management system, Research Electronic Data Capture (REDCap), which was used to store and manage all participant data.¹⁵ The randomization was not done by hospital, but at the overall patient level. No randomization was instituted for the clinician-facing intervention; instead, the decision support tool was embedded into the electronic health record to facilitate optimized opioid prescription upon discharge based on recorded inpatient use.^{11; 16} It was implemented after half of the participants were enrolled in the trial. The principal investigator, coinvestigators, participants, and statisticians remained masked to treatment assignment until the database was locked for analysis.

Procedures

We primarily tested a patient-facing app to reduce opioid intake, which was designed and implemented through a user-centered approach.¹⁷ The educational app provided information on elements of pain management that could be modified by patients, including 1) postoperative pain expectations, 2) over-the-counter pharmacologic alternatives to opioids, 3) opioid safety and appropriate disposal, and 4) non-pharmacological therapy, such as meditation (see Supplemental Information 1 for more details). The control app featured the same interface but only collected patient-reported outcome data. The app was deployed to research team members and pilot patients for beta testing before finalization. After the pilot phase (initiated on March 14, 2022), changes were made to enhance user acceptability, such as additional response choices in the questionnaires, more explicit guidance regarding data entry procedures, and a weekly survey reminder. Overall, user accessibility survey results indicated that the app was feasible and desirable.

Baseline information was obtained from the patient in person during the index hospitalization. During this visit, patients familiarized themselves with the app format, asked questions, and received explanations regarding the app’s functionality. Patients could access the educational content of the educational app immediately after enrollment and continuously throughout the post-discharge follow-up period. Following discharge, patients received four weekly surveys via the app’s notification function, thereby also ensuring active use of the app. The day of discharge was considered day 0, and surveys covering the preceding week’s outcomes were obtained on days 7, 14, 21, and 28. If no responses were obtained within the first day following the survey’s distribution, patients received a text message as a reminder, and, if necessary, a personal phone call from the study team.

In addition to the educational app, a clinician-facing decision support tool was operationalized in the second half of the study period. Clinicians, regardless of surgical specialty, who attempted to write for the most commonly chosen discharge opioid prescription were alerted to consider not prescribing any discharge opioids if their patient did not take any opioids in the 24 hours prior. The alert was a soft stop and could be ignored or overridden by clinicians. Although we anticipated the educational app to have the most substantial effect on the primary outcome, patient-reported opioid intake after discharge, we also wanted to account for the potential impacts of the clinician-facing prescribing tool, given that it potentially could lead to fewer opioids being available for patients to consume. Our previous work in three diverse samples of surgical procedures found that among available predictor variables that could be incorporated into an electronic decision support tool, 24-hour pre-discharge opioid intake was most strongly associated with patient-reported post-discharge opioid intake.^{16; 18–20}

Outcomes

The primary outcome was reported by patients weekly and was defined as cumulative opioid intake in oral morphine milligram equivalents (MME) over the first four weeks after discharge. We anticipated that patient exposure to the non-opioid pain management content in the educational app would lead patients to take less of the opioids prescribed to them at discharge. The rationale to secondarily test the effects of the clinician-facing decision support tool on the primary outcome of self-reported cumulative intake was to examine if potentially lower or no opioid prescriptions triggered by the decision support tool after discharge would limit self-reported opioid intake.

The secondary outcomes included the amount of opioid prescribed at discharge (MME), supplementary opioid prescriptions within four weeks after discharge (yes/no), disposal of leftover opioids within four weeks of discharge (yes/no), pain intensity scores during the four weeks after discharge, and pain interference scores during the four weeks after discharge. Non-opioid analgesics, such as ibuprofen, naproxen, and acetaminophen, reported taken within four weeks of discharge (yes/no), were an exploratory outcome. Pain intensity and interference were quantified using the National Institutes of Health Patient-Reported Outcomes Measurement Information System (PROMIS^™) four-item pain intensity scale and the four-item scale pain interference score. All PROMIS^™ scores were collected weekly during the initial four weeks after discharge and were analyzed as standardized T-scores (mean 50, standard deviation 10).²¹

Potential confounders, in the unlikely event that the randomized groups would not be balanced at baseline, included sociodemographic characteristics obtained from the electronic health record and a questionnaire provided to patients during their index hospitalization: age, sex, race, ethnicity, body mass index (kg/m²), and insurance type. Health history variables were obtained via surveys and electronic health records as necessary and included a history of substance use disorder, PROMIS^™ Sleep Disturbance Score,²² Patient Health Questionnaire-8 (PHQ-8) score,²³ preoperative opioid use, preoperative benzodiazepine use, preoperative gabapentinoid use, and preoperative antidepressant use. Clinical characteristics included the primary surgical subspecialty for the requisite hospital stay, the total number of procedures performed during the admission, total in-hospital opioid intake, pain score at discharge, and length of hospital stay.

We considered an adverse event as one requiring more than two additional opioid prescriptions within two weeks of discharge and/or access of confidential information by a non-authorized person. We considered a serious adverse event as all-cause mortality and/or re-hospitalization for any reason.

Statistical analysis

Baseline balance on potential confounding characteristics was assessed using absolute standardized difference (ASD), calculated as the absolute value of the difference in means or proportions divided by the pooled standard deviation. A variable was considered imbalanced if the ASD was greater than 0.159, based on the expected 95^th percentile of the sampling distribution of the standardized difference under the null hypothesis of no difference for the given sample size.²⁴ All regression models were adjusted for imbalanced covariates. Analyses were conducted on a modified intention-to-treat basis, excluding participants who did not engage with the intervention or answer any survey questions.

We assessed the effect of the patient-facing app on the primary outcome of self-reported cumulative opioid use in the four weeks after discharge using a confounder-adjusted linear regression model on the log-transformed outcome. We estimated the treatment effect as the ratio of geometric means. We conducted a sensitivity analysis using quantile regression (on the untransformed outcome) to test the robustness of the primary analysis method (linear regression) to outliers. We also conducted a post-hoc analysis to evaluate whether the treatment effect differed based on the amount of opioids prescribed at discharge, by adding an interaction term between the patient-facing app group and the log-transformed amount of opioids prescribed at discharge to our primary analysis model.

Finally, we tested whether the effect of the educational app on the primary outcome depended on whether the clinician-facing decision-support tool was active, by examining the interaction between the two factors in an analogous linear regression model. We also conducted a confounder-adjusted segmented regression analysis to evaluate the effect of the physician-facing electronic decision support tool (secondary intervention) on self-reported cumulative opioid use in the four weeks after discharge (see Supplemental Analysis 1 for more details).

Log-binomial regression was used to assess the effect of the educational app, in terms of relative risk, on supplementary opioid requirements in the four weeks after discharge (any vs. none) and opioid disposal (yes vs. no). The effect of the educational app on pain interference and intensity t-scores was evaluated by fitting linear mixed models with fixed covariates for treatment group and time (categorical), as well as random intercepts for patients. Our use of mixed-effects models handled missing data in pain-related survey responses, assuming that the data were missing at random (MAR). Holm-Bonferroni correction was used to adjust for multiple testing.

Log-binomial regression was used to assess the effect of the educational app on non-opioid analgesic medications taken in the four weeks after discharge (any vs none). No interim analyses were planned because the population studied was not considered high risk for adverse events, and the interventions were generally considered not harmful. The Data and Safety Monitoring Board advised and oversaw patient safety, study progress, data quality, and confidentiality.

Sample Size Justification

We designed the study to have 85% power to detect a ratio of geometric means of 0.8 for the primary outcome of cumulative opioid use in the four weeks after discharge. Assuming a coefficient of variation of 1.20 and an alpha level of 0.05, 300 patients were required per group. The coefficient of variation was estimated based on data combined from our previous observational studies^18–20 on opiates taken (MME) at one week from discharge (where variability in opioid consumption is greatest) and summed across four weeks from discharge.

RESULTS

We enrolled and considered a total of 711 participants between May 2, 2022, and December 10, 2023. After excluding 105 participants without follow-up data, the analysis cohort consisted of 606 patients (Figure 1). Of these, 75% received in-person instruction on how to use the app on their device within one day before discharge, and 89% received it within two days before discharge. In the analysis cohort, 303 patients were randomized to the educational app and 303 to the control app. The median patient age was 50 years, most patients were female (n = 391, 65%), and general surgery was the most common specialty (n = 290, 48%). The median hospital length of stay was two days, and the median in-hospital cumulative opioid use was 98 MME. Imbalance (i.e., ASD > 0.159) was observed on two potential confounders: race and surgery type (Table 1). These variables were adjusted for in all of our analyses.

Figure 1.

Study flow diagram.

Table 1. Patient clinical and demographic characteristics stratified by app randomization.

Characteristic	Education App (N = 303)	Control App (N = 303)	Absolute Standardized Difference	Missing, n (%)
Age, years, median [Q1, Q3]	50 [37, 62]	49 [37, 60]	0.073	0 (0)
Body Mass Index, kg m-2, mean (SD)	36 (10)	34 (9)	0.123	0 (0)
Sex, n (%)			0.132	0 (0)
Male	116 (38)	97 (32)
Female	186 (61)	205 (68)
Other	1 (0.3)	1 (0.3)
Race, n (%)			0.198	5 (0.8)
White	262 (87)	259 (86)
Black or African American	20 (7)	24 (8)
American Indian	4 (1)	1 (0.3)
Asian	2 (0.7)	3 (1.0)
Multiracial	5 (2)	1 (0.3)
Other	8 (3)	12 (4)
Ethnicity: Hispanic or Latino, n (%)	15 (5)	16 (5)	0.014	1 (0.2)
Activated Decision Support Tool,	155 (51)	151 (50)	0.026	0 (0)
Surgical Specialty, n (%)			0.235	0 (0)
General	153 (51)	137 (45)
Thoracic	17 (6)	24 (8)
Colorectal	4 (1)	3 (1)
Obstetric	14 (5)	24 (8)
Gynecologic	20 (7)	19 (6)
Neurological	16 (5)	21 (7)
Orthopedic	35 (12)	33 (11)
Urologic	25 (8)	25 (8)
Vascular	2 (0.7)	1 (0.3)
Ophthalmic, OMF, Plastic ENT or Plastic	11 (4)	13 (4.3)
Not specified	2 (0.7)	0 (0.0)
Multiple	4 (1.3)	3 (1.0)
PROMIS Sleep Disturbance score, mean (SD)	52 (5)	52 (5)	0.114	0 (0)
PHQ-8 score, mean (SD)	5 (4)	5 (4)	0.027	0 (0)
Insurance, n (%)			0.155	1 (0.2)
Medicare	41 (14)	41 (14)
Medicaid	26 (9)	36 (12)
Private	198 (66)	189 (62)
Self-pay	7 (2)	12 (4)
Multiple	30 (10)	25 (8)
Substance use disorder, n (%)	6 (2)	5 (2)	0.025	0 (0)
Opioid use, n (%)	20 (7)	20 (7)	<0.001	0 (0)
Benzodiazepine use, n (%)	28 (9)	19 (6)	0.111	0 (0)
Antidepressant use, n (%)	85 (28)	72 (24)	0.098	0 (0)
Gabapentinoid use, n (%)	78 (26)	66 (22)	0.093	0 (0)
Multiple procedures during admission, n (%)	7 (2)	7 (2)	<0.001	0 (0)
Length of stay, days, median [Q1, Q3]	2 [1, 4]	2 [1, 4]	0.020	0 (0)
Pain score at discharge, mean (SD)	3 (3)	4 (3)	0.017	4 (0.7)
Cumulative opioids in hospital, MME, median [Q1, Q3]	96 [58, 163]	100 [57, 175]	0.005	0

Absolute standardized difference was calculated by dividing the difference in means by SD. Absolute standardized difference greater than 0.159 (bolded) was considered to indicate imbalance. Pain score at discharge was calculated using the visual analog scale. Medication use was identified preoperatively.

Abbreviations: ENT, ear nose and throat; OMF, oral and maxillofacial surgery; PROMIS, Patient-Reported Outcomes Measurement Information System; PHQ-8, Patient Health Questionnaire-8; Q1, First quartile; Q3, Third quartile; SD, standard deviation.

Primary outcome

The overall median [Q1, Q3] opioids consumed in the four weeks after discharge was 22 [0, 90] MME. The median [first quartile (Q1), third quartile (Q3)] self-reported cumulative opioid intake in the four weeks after discharge was 22 [0, 98] MME in the educational app group and 15 [0, 82] MME in the control app group (Figure 2). The ratio of geometric means (intervention app/control app) was estimated to be 1.21 (95% CI: 0.87, 1.69; P = 0.264) (Table 2). In the sensitivity analysis using quantile regression, the median difference between the groups was estimated to be eight MME (95% CI: −4, 20; P = 0.174). In the post-hoc interaction analysis, we did not find a statistically significant interaction between the patient-facing app group and the amount of opioids prescribed at discharge (P = 0.721). Opioid consumption in the four weeks after discharge, stratified by the amounts of prescribed opioids is depicted in Supplemental Figure 1.

Figure 2.

Distribution of the cumulative amount of self-reported opioid consumption in morphine milligram equivalents within four weeks after discharge, stratified by patient-facing app group assignment. Morphine milligram equivalents (MME) are reported on the log scale.

Table 2. Treatment effect on the primary and secondary outcomes.

Outcome	Education App (N = 303)	Control App (N = 303)	Treatment effect (95% CI)	P-value
Primary	Median [Q1, Q3] or Mean (SD)		Ratio of geometric means
Morphine milligram equivalents consumed four weeks after discharge	22 [0, 98]	15 [0, 82]	1.21 (0.87, 1.69)	0.264a
Secondary
Morphine milligram equivalents prescribed at discharge	75 [60, 150]	75 [60, 150]	1.07 (0.85, 1.35)	1.0a,d
Pain Intensity	52.5 (9.7)	51.3 (9.2)	1.34 (0.20, 2.47)	0.084b,d
Pain Interference	55.2 (9.6)	53.8 (9.8)	1.47 (0.31, 2.63)	0.070b,d
	Frequency (%)		Relative risk
Additional opioids after discharge	54 (18)	46 (15)	1.17 (0.82, 1.68)	1.0c,d
Disposal	6/183 (3)	8/170 (5)	0.72 (0.26, 1.98)	1.0c,d
Exploratory
Additional non-opioid analgesic medications taken	35 (12)	30 (10)	1.16 (0.73, 1.84)	0.531c

^aP-value obtained from linear regression model with log-transformed outcome and treatment group, race and surgery type as covariates.

^bP-value obtained from linear mixed model with fixed covariates for treatment group (primary covariate of interest), time (categorical), race, and surgery type, and random intercepts for patients.

^cP-value obtained from log-binomial regression model with the treatment group, race, and type of surgery as covariates.

^dHolm-Bonferroni adjusted P value to account for multiple testing.

In the confounder-adjusted segmented regression analysis, the clinician-facing decision support tool (secondary intervention) did not have an immediate (difference in mean log opioid consumption just after activating the intervention) or long-term (the difference in slopes between the postintervention and preintervention periods) effect on opioid consumption (see Supplementary Analysis 1 for more details). Furthermore, the effect of the educational app on self-reported cumulative opioid consumption in the first four weeks after discharge was not found to differ based on whether the clinician-facing decision support tool was active or inactive (interaction P = 0.583). For patients with the decision support tool activated, the ratio of geometric means was estimated to be 1.32 (95% CI: 0.82, 2.12; P = 0.258). For those with the decision support tool inactivated, the ratio of geometric means was estimated to be 1.11 (95% CI: 0.69, 1.79; P = 0.660).

Secondary outcomes

The overall median [Q1, Q3] amount of opioids prescribed at discharge was 75 [60, 150] MME. The median [Q1, Q3] amount of opioids prescribed at discharge (total n = 606) was 75 [60, 150] MME with the educational app and 75 [60, 150] MME with the control app, corresponding to an estimated ratio of geometric means of 1.07 (95% CI: 0.85, 1.35; adjusted P = 1.00). The interaction between the patient-facing educational app and the clinician-facing decision support tool for opioids prescribed at discharge was not statistically significant (P = 0.604). The ratio of geometric means was estimated to be 1.01 (95% CI: 0.73, 1.41; P = 0.953) with the active decision support tool, and 1.08 (95% CI: 0.79, 1.49; P = 0.620) with the inactive decision support tool.

The mean (SD) pain intensity t-score was 52.5 (9.7) with the educational app and 51.3 (9.2) with the control app, with an estimated mean difference between groups of 1.34 (95% CI: 0.20, 2.47; adjusted P = 0.084). The mean (SD) pain interference t-score was 55.2 (9.6) with the educational app and 53.8 (9.8) with the control app, and the estimated mean difference between groups was 1.47 (95% CI: 0.31, 2.63; adjusted P = 0.070). Neither effect was statistically significant.

The fraction of patients requiring supplementary opioids within four weeks of discharge was 54/303 (18%) with the educational app and 46/303 (15%) with the control app, corresponding to an estimated relative risk of 1.17 (95% CI: 0.82, 1.68; adjusted P = 1.00). The fraction of patients who disposed of opioids was 6/183 (3%) with the educational app and 8/170 (5%) with the control app, corresponding to an estimated relative risk of 0.72 (95% CI: 0.26, 1.98; adjusted P = 1.00).

Exploratory outcomes

In an exploratory analysis, the educational app was not found to affect the fraction of patients requiring non-opioid analgesic medications taken within four weeks of discharge, which was 35/303 (12%) with the intervention and 30/303 (10%) without, corresponding to a relative risk of 1.16 (95% CI: 0.73, 1.84; P = 0.531).

No adverse or serious adverse events were reported in relation to the study intervention. However, based on the definition of adverse events, 46 were reported (for control and educational app, respectively): (1, 2) required more than two additional opioid prescriptions within two weeks of discharge, (23, 19) were re-hospitalized for any reason, and (1, 0) died. Participant characteristics stratified by completion and randomization groups are provided in Supplemental Tables 1-4.

DISCUSSION

The objective of this study was to assess the effectiveness of a patient-facing, smartphone-based app in decreasing opioid consumption after discharge while ensuring effective pain management following inpatient surgery. In this randomized controlled two-by-two partial factorial trial of a patient-facing educational app and a clinician-facing decision support tool for responsible opioid prescribing, we found no evidence of an effect on patient-reported cumulative opioid intake, opioids prescribed at discharge, pain intensity or interference, supplementary opioid prescriptions, or opioid disposal in the first four weeks after discharge. The results were consistent among patients whose prescribing clinician had the clinical decision support tool activated. This study demonstrates that a patient-centered smartphone-based educational app, compared to a data collection-only app, did not influence opioid consumption after discharge in a large (n = 606), surgically diverse (11 subspecialties) population across two hospitals. However, the low patient-reported opioid intake amounts (median of 22 MME in the education app and 15 MME in the control app) suggest that stringent opioid prescription practices may be appropriate for most patients.

The median cumulative opioid dose reported taken after discharge was 22 MME in the intervention app group and 15 MME in the control app group, equating to two or three five-milligram oxycodone pills total. Yet, median opioid doses prescribed were still 75 MME in both groups. The low total opioid intake after discharge is analogous to contemporary analyses of patient-reported opioid consumption after general surgery (both laparoscopic and open).²⁵ The Michigan Surgical Quality Collaborative Registry 2024 update reports median patient-reported 30-day opioid consumption of 7.5 to 30 MME in a similar surgical mix as this trial, concordant with our findings.²⁵ Although overdose rates are primarily driven by synthetic opioids such as illicit fentanyl and fentanyl analogs, prescription opioids represent a gateway to opioid use disorder.²⁶ Postsurgical opioids are associated with an elevated risk of developing new persistent opioid use. In a retrospective study of opioid naïve surgical patients, each additional opioid pill taken within 30 days postoperatively was associated with a 0.05% increase in likelihood for developing new persistent opioid use.²⁷ Notably, the overall opioid intake was low in patients who subsequently did and did not develop new persistent opioid use, seven versus four 5mg oxycodone equivalent pills, respectively.²⁷ So, despite discharge opioid prescriptions after surgery being at relatively low levels, their long-term effects remain a clinically significant issue.

Advancements in electronic health digital technologies are well-established and pervasive, given the ubiquitous prevalence of smartphones. Several mobile health programs have been developed to reduce postoperative pain and opioid consumption, though results are conflicting. In an unblinded randomized trial in surgical patients after total knee replacement, the PainCoach app provided information on pain medication use, exercises, warm or cold packs, rest, and when to call the physician’s office.⁸ The PainCoach group used 23.2% fewer opioids, 14.6% more acetaminophen, and experienced improved pain control compared to the control group.⁸ The outcomes of the PainCoach app in the total knee replacement population also informed and motivated our trial design. But since then, three other mobile technology interventions have shown limited effectiveness in reducing patient opioid use. In a single-center randomized controlled trial, the Continuing Precision Medicine mobile app incorporated a behavioral health intervention for patients and a clinical decision support tool for prescribing providers after Cesarean section. Women undergoing Cesarean section who utilized the app were 92% less likely to inappropriately use their prescription opioids after discharge compared to standard care.⁹ However, the number of opioids utilized was not different between groups.⁹ Notably, 75% used four five-milligram oxycodone tablets or fewer.⁹ In another study, real-time opioid use education via text message after outpatient general surgery in 160 patients did not lead to a reduction in patients’ opioid use.¹⁰ Similarly, in a recent cluster-randomized trial after dental procedures, the implementation of a mobile health app did not have a significant impact on pain scores or opioid use at home after discharge from care.²⁸

Mobile applications for postoperative pain and opioid management can be evaluated in relation to the clinical practice guidelines for evidence-based management of postoperative pain established by the American Pain Society, the American Society of Regional Anesthesia and Pain Medicine, and the American Society of Anesthesiologists’ Committee on Regional Anesthesia, Executive Committee, and Administrative Council.²⁹ Relevant recommendations include tailored education, evidence-based nonpharmacological pain management strategies, such as cognitive behavioral therapy and physical therapy, as well as pharmacological intervention when necessary.^{7; 29} Our patient-centered intervention app integrated all these components and employed user-centered feedback, but it lacked the expected results. User-centered design can leverage patient engagement to bridge the gap between healthcare providers and patients in other care contexts,³⁰ but our user engagement varied widely, as evidenced by a 14.8% attrition rate (n = 105). Known sociodemographic factors that have been shown to influence the effectiveness of digital health applications include age, income, education level, and eHealth literacy.^{31; 32}

Our study also included limitations. First, the primary outcome is based on self-report, which is an approach that is vulnerable to recall and social desirability biases that may underestimate actual opioid consumption. However, 14-day self-report of opioid use after surgery is reasonably accurate when compared to Bluetooth-enabled smart pill containers.³³ Second, we could have included more measures, such as links to additional health records data, including the Prescription Drug Monitoring Program, or more patient-centered outcomes, such as the level of engagement with the healthcare team. However, the validated patient-reported outcomes (PROMIS^™ measures) were chosen deliberately, as they are validated across various conditions and function optimally in “real-world” clinical settings such as the one encountered in this study.²¹ Third, there were a few extreme outliers in the primary outcome (Figure 1). The ratio of geometric means serves as a point estimate for the primary outcome and is relatively resistant to outliers, although it may not be as intuitive. For that reason, we conducted a sensitivity analysis using quantile regression to ensure robustness to outliers while maintaining readability. We conducted a modified intention-to-treat analysis by including all randomized subjects, even if they received opioid stewardship and multimodal analgesic education from another source. However, given that the primary outcome of this trial was self-reported opioid use after discharge, we excluded 105 randomized patients who opted out or did not respond to the post-discharge surveys. Hence, opioid intake could not be assessed. Among the non-responders, the only significant between-group differences were found in race, ethnicity, and surgical specialty (Supplemental Table 2). None of the randomized participants included in the analysis had incomplete responses for the primary outcome. Lastly, the exclusion of patients unable to read English and those without access to a smartphone limits the generalizability of our study. Future studies warrant exploring how to engage end users bidirectionally with newer technologies, such as artificial intelligence-enabled chatbots, which could help overcome some of the limitations mentioned above.³⁴ This study was focused on improving post-discharge pain management primarily by leveraging a health informatics app to inform patients on pain expectations, over-the-counter alternatives, opioid safety, and non-pharmacological approaches to pain management. Its null results should not be interpreted as indicating that these approaches are not relevant, but rather as suggesting that the chosen eHealth interventions to leverage these concepts are ineffective.

CONCLUSION

In summary, while smartphone-based apps have shown some promise in enhancing postoperative pain management and potentially reducing opioid consumption in certain surgical subpopulations, their effectiveness is contingent upon user engagement, the quality of the application, and the context of its use within broader pain management strategies. Among patients undergoing major surgery, neither a patient-facing educational app nor a clinician-facing prescribing opioid tool reduced self-reported opioid intake in the four weeks following discharge. The low levels of patient-reported opioid intake suggest that stringent opioid prescription practices may be appropriate for most patients.

KEY MESSAGES.

What is already known about this topic –

A significant proportion of surgical patients have unused opioids after discharge, and postdischarge overdose deaths remain a concern. There is a need for opioid stewardship programs that not only address pain management during hospitalization but also extend patient education and promotion of non-opioid therapies to the post-discharge period. The optimal way to implement opioid reduction strategies post-discharge is still being evaluated.

What this study adds –

Among patients undergoing major surgery, neither a patient-facing educational app nor a clinician-facing prescribing opioid tool reduced self-reported opioid intake in the four weeks following discharge.

How this study might affect research, practice or policy –

Given the lack of effect of the educational app and clinician-­ facing decision support tool, future research should explore alternative methods for influencing postoperative opioid use, including approaches beyond digital health tools. These findings also underscore the importance of publishing null result studies to guide resource allocation, inform future research, and optimize quality improvement efforts.

Abbreviations

MME

morphine milligram equivalents

mHealth

Health information technology, including medical apps

REDCap

Research Electronic Data Capture

PROMIS

Patient-Reported Outcomes Measurement Information System

PHQ-8

Patient Health Questionnaire-8

ASD

Absolute standardized difference

ENT

Ear Nose and Throat

OMF

Oral and Maxillofacial Surgery

Q

Quartile

SD

standard deviation

Data availability statement

The datasets used and/or analyzed for the current study are available from the corresponding author upon reasonable request due to the sensitive nature of the data collected.