Integrating Nonpharmacological Options Into Perioperative Pain Care

Key Points

Question

Does an electronic health record (EHR)–based intervention promoting nonpharmacologic pain care improve postoperative patient-centered outcomes and reduce opioid exposure?

Findings

In a cluster randomized clinical trial including 68 141 surgical procedures, the intervention did not improve pain interference or physical function but reduced opioid exposure after surgery.

Meaning

These findings suggest that system-level EHR strategies may reduce opioid exposure without worsening patient-reported outcomes.

This cluster randomized clinical trial evaluates whether an electronic health record–embedded intervention promoting nonpharmacologic pain care is associated with improved postoperative pain interference and physical function and reduced opioid exposure.

Abstract

Importance

Postoperative pain is commonly treated with opioids, which may contribute to unnecessary opioid-related harms. Scalable approaches to integrate recommended nonpharmacologic pain care could reduce opioid exposure while improving or at least preserving functional recovery outcomes.

Objective

To evaluate whether an electronic health record (EHR)–embedded intervention promoting nonpharmacologic pain care improves postoperative pain interference and physical function while reducing opioid exposure.

Design, Setting, and Participants

This cluster randomized clinical trial was conducted from October 16, 2020, through April 30, 2024. Participants included patients undergoing surgery at 22 surgical practices in 6 surgical centers in Minnesota, Wisconsin, Florida, and Arizona.

Intervention

A multicomponent EHR-based intervention including a patient-facing Healing After Surgery educational guide, clinical decision support, and nonpharmacologic pain care support materials integrated into routine perioperative touchpoints and workflows.

Main Outcomes and Measures

Co–primary outcomes were Patient-Reported Outcomes Measurement Information System (PROMIS) Pain Interference and Physical Function T scores measured preoperatively and at 1, 2, and 3 months after surgery. Secondary outcomes included opioid prescribing and administration (morphine milligram equivalents [MMEs]), PROMIS Anxiety scores, and health care utilization for 3 months after surgery. Mixed-effects models were used to account for correlation of outcomes within surgery, patient, and cluster, as appropriate for each model.

Results

Among 68 141 included surgical procedures (40 892 male patients [60.0%]; mean [SD] age, 59.27 [16.26] years), 43 053 occurred during the intervention period, and 25 088 occurred during the usual care period. The intervention did not change PROMIS pain interference (adjusted mean difference, 0.09; 95% CI, −0.18 to 0.36; P = .51) or PROMIS physical function (adjusted mean difference, 0.15; 95% CI, −0.11 to 0.41; P = .27) in the intervention compared with control surgical procedures. However, total postoperative opioid exposure was lower during intervention periods (rate ratio, 0.90; 95% CI, 0.82 to 0.99; P = .03), corresponding to an approximately 10% relative reduction and an adjusted mean difference of approximately 260 MMEs per patient, with no changes in anxiety, adverse events, or health care utilization. This reduction was driven primarily by lower opioid administration during hospitalization (adjusted rate ratio, 0.94; 95% CI, 0.90 to 0.97; P = .001), with an adjusted mean difference of approximately 85 MMEs.

Conclusions and Relevance

In this cluster randomized clinical trial of patients undergoing surgery, an EHR-embedded perioperative pain management intervention promoting nonpharmacologic pain care reduced opioid exposure without changes in patient-reported outcomes, suggesting potential for scalable EHR-based strategies to support safer postoperative pain management.

Trial Registration

ClinicalTrials.gov Identifier: NCT04570371

Introduction

Millions of Americans undergo surgery each year, and recovery is often accompanied by new or worsening pain. Opioids remain a common and often default treatment, sometimes necessary to optimize postoperative recovery.^1,2 However, a subset of patients prescribed opioids for surgical pain continue to use opioids long after recovery,^1,2,3,4,5 and some will progress to opioid use disorder or experience opioid-related overdose.^6,7,8 Perioperative opioid use is also associated with impaired surgical recovery, increased postoperative complications and falls, and higher health care utilization and costs.⁹

Clinical guidelines recommend nonpharmacologic pain care (NPPC) modalities including movement-based, behavioral, and physical approaches as part of multimodal postoperative pain management.^{10,11,12,13,14,15,16} Patients report interest in these approaches,^17,18 yet barriers to their routine use persist, including clinician unfamiliarity, limited workflow integration, and uncertainty regarding how to incorporate NPPC into time-constrained perioperative care.¹⁹

Electronic health records (EHRs) can standardize and automate guideline-concordant pain management^20,21 by delivering patient-facing education through portals while also providing point-of-care clinical decision support (CDS) to clinicians. Prior EHR-based pain interventions have primarily targeted clinician prescribing behavior and have not systematically integrated patient education or patient NPPC preferences.²²

As part of a large federal initiative to curb opioid misuse,²³ we conducted the Non-pharmacological Options in Post-Hospital and Rehabilitation Pain Management (NOHARM) pragmatic trial, a study designed to evaluate intervention effectiveness under conditions representative of routine clinical practice. We tested whether a multicomponent, EHR-embedded intervention promoting NPPC could improve postoperative pain interference and physical function and reduce opioid exposure during the first 3 months after surgery, compared with usual care (UC).

Methods

Trial Design and Oversight

The NOHARM trial was a preregistered, population-level, stepped-wedge, cluster randomized pragmatic clinical trial conducted at 6 high-volume surgical centers (3 quaternary referral centers and 3 community hospitals) in Minnesota, Wisconsin, Florida, and Arizona. The protocol has been published previously and is also shown in Supplement 1.²⁴ Twenty-two clusters, defined as panels of surgical procedures distinguished by surgical department or division and site, with minimal overlap in clinicians and postoperative care settings, were randomized to 1 of 5 sequences. After an initial 6-month UC period, clusters crossed over to the intervention at 7-month intervals according to sequence assignment.

This study was approved by the Mayo Clinic institutional review board, which noted that NOHARM proposed testing an enhanced standard of care against UC and, on the basis of advice from the National Institutes of Health Systems Collaboratory Ethics and Regulatory Committee, granted a waiver of individual patient consent.²⁵ The trial was overseen by an external Data Safety and Monitoring Board appointed by the National Institutes of Aging. This study follows the Consolidated Standards of Reporting Trials (CONSORT) reporting guidelines.

Participants

Enrollment occurred at the level of the surgical procedure, without additional patient-level eligibility criteria (see eAppendix 1, eFigure 1, eAppendix 2, eTable 1, and eTable 2 in Supplement 2). Procedures were grouped by department or division (colorectal, orthopedic, cardiovascular, thoracic, solid organ transplant, gynecologic, and cesarean delivery) to ensure variation in procedural risk, anatomic region, and expected opioid exposure. A validated EHR algorithm identified and enrolled procedures when a surgical case request was placed or 30 days before the scheduled surgery date, whichever occurred later. Emergent surgical procedures were excluded because the intervention required preoperative patient engagement through the patient portal, except for converted cesarean deliveries and cadaveric organ transplantation.

Intervention

Patients scheduled for a qualifying surgery and assigned to an intervention cluster received a multicomponent, EHR-embedded intervention consisting of (1) the Healing After Surgery guide delivered through the patient portal before surgery, (2) clinician-directed EHR CDS prompting allied health staff to reinforce patients’ NPPC choices, and (3) NPPC education and self-management resources.²⁴ Table 1 illustrates the delivery of intervention components across the perioperative continuum, highlighting how the intervention engaged patients, prompted care teams, and supported NPPC.

Table 1. Major Components of the NOHARM Intervention Across the Perioperative Care Continuum^a.

Intervention component	Preoperative	Hospitalization	Discharge	Postdischarge
Patient engagement	Select preferred NPPC modalities using the HAS guide; learn and practice selected techniques using educational resources	Use preferred NPPC modalities; update preferences as needed	Review pain management plan and recommendations for home NPPC use	Continue preferred NPPC modalities using available educational and community resources
Clinical care	Introduce NPPC; distribute educational materials and workbook; encourage patients to make preoperative selections	Reinforce and facilitate patient-selected NPPC approaches during nursing, physical therapy, and occupational therapy care, when feasible	Review and reinforce individualized pain management plan; support transition to home use	Routine clinical follow-up
EHR support	Deliver HAS guide via patient portal; capture preferred NPPC modalities in Epic flowsheets via HAS	Prompt clinicians through clinical decision support; document and update NPPC preferences	Populate discharge summary with recommendations for home and community use of selected NPPC techniques	Reinforce continued NPPC use through automated portal messaging and targeted outreach

Abbreviations: EHR, electronic health record; HAS, Healing After Surgery; NOHARM, Non-pharmacological Options in Post-Hospital and Rehabilitation Pain Management; NPPC, nonpharmacologic pain care.

^a The NOHARM intervention combined patient engagement, EHR-enabled clinical decision support, and reinforcement by clinical staff across sequential phases of perioperative care.

Developed with input from the Mayo Clinic Patient Family Advisory Committee, the Healing After Surgery guide described opioid risks and harms, encouraged multimodal pain management, and allowed patients to enter their preferred NPPC modalities (yoga, tai chi, walking, guided imagery, music, progressive muscle relaxation, paced breathing, meditation, aromatherapy, transcutaneous electrical nerve stimulation, massage, acupressure, and heat or cold) directly into Epic. These selections were stored in EHR flowsheets and used to drive CDS and portal-based outreach.

Throughout the perioperative period, nurses, therapists, and other allied health staff reinforced patients’ selected NPPC strategies, incorporated them into care when feasible, and updated preferences in the EHR. Additional NPPC resources were available through a dedicated website, twice-weekly Zoom sessions, printed, video, and audio materials, and individualized discharge instructions. Further intervention details are provided in eFigures 2 and 3 in Supplement 2. The trial used an encouragement design in which exposure to intervention components was encouraged but not required (Supplement 1).^26,27,28,29

Usual Care

UC reflected standard perioperative practices and did not include EHR-based patient education or NPPC preference elicitation, or clinician-directed CDS to increase support for NPPC use. UC is further described in Supplement 1.

Randomization and Blinding

The 22 clusters were randomly assigned to 1 of 5 sequences specifying the timing of crossover from UC to the intervention (eFigure 4 in Supplement 2). Constrained randomization was used to balance postoperative unit clustering, procedural volume, and site.³⁰ EHR-based rule logic assigned individual procedures to clusters. Participants, clinicians, and investigators were not blinded.

Outcomes

Given their validity, responsiveness, and reliability in postoperative populations,^31,32,33 Patient-Reported Outcomes Measurement Information System (PROMIS) Pain Interference and Physical Function T scores measured using computerized adaptive tests (CATs) administered primarily through the Epic patient portal (Supplement 1) were our co–primary outcomes. Assessments were completed preoperatively and at 1, 2, and 3 months after the index surgery. Each patient-reported outcome measure (PROM) was classified according to the intervention status of the cluster at the time of surgery.

To evaluate potential nonresponse bias related to portal nonuse, 6-item printed PROMIS short forms were mailed to patients who did not use the portal and those who did not complete the 1-month assessment. Because these mailed assessments were not prespecified and differed in mode of administration, measurement precision, and item content, they were included in sensitivity analyses rather than the primary analyses.³⁴

Secondary outcomes included PROMIS Anxiety T scores, postoperative opioid exposure, and health care utilization (emergency department visits, hospitalizations, and intensive care unit admissions) (Supplement 1). Opioid exposure was assessed through 90 days after surgery and censored at subsequent surgery or death in accordance with Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials consensus recommendations.³⁵ In-hospital opioid administration and prescribing were comprehensively captured through the EHR. Because patients received follow-up care outside the health system, postdischarge opioid exposure may have been incompletely ascertained; therefore, patient-reported opioid use and prescriptions were also collected at 3 months. Opioid exposure outcomes included prescribed and administered morphine milligram equivalents (MMEs). Patient-reported NPPC and opioid use, including opioid refills, were assessed at 3 months by self-administered survey.³⁶

Adverse events were passively monitored through EHR-derived International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes assigned during trial participation (eTable 3 in Supplement 2). Opioid-related adverse events included new opioid use disorder and opioid poisoning or toxicity. Adverse events potentially related to NPPC included falls, burns (from heat or transcutaneous electrical nerve stimulation), and muscle sprains, strains, or spasms. Episodes of potential failure to rescue, defined as 3 consecutive severe pain scores (numerical rating scale, 7-10 of 10), were also monitored. Deaths were also abstracted from the EHR.

Patient and Surgery Characteristics

For each surgery, we collected patient age, sex, race, ethnicity, and medical comorbidities. Race and ethnicity were identified using self-reported patient tables and were classified as American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White, and other (ie, patients who chose not to disclose, were recorded as unknown, or were unable to provide this information). Data on race and ethnicity are included here to permit assessment of external validity and generalizability. Surgical characteristics included procedure type, duration, laparoscopic vs open approach, same day vs overnight surgery, hospital length of stay, and colorectal or hip or knee procedure status, as these surgical procedures are often associated with more intense pain.^37,38

Sample Size and Statistical Power

Using historical annual volumes for the included clusters, we conducted simulations to estimate detectable clinically meaningful effects with a power of 90%. From a prior study, we calculated the intraclass correlation coefficient for pain interference PROM scores to be r = 0.037. For P = 1, …, 5000 permutations, we randomly assigned the 22 clusters to 5 sequences. For each permutation, we used the closed-form solution of Harrison et al³⁹ for SE(Q), derived from the formula described by Hussey et al,⁴⁰ for true effect Q. For calculations, we used σ² = 1 and τ² = 0.488 so that detectable effects would be in SDs of the PROM score, and intraclass correlation coefficient would be 0.037. Over 5000 permutations, the detectable effect ranged between 0.048 SD and 0.055 SD

Statistical Analysis

All analyses followed the intention-to-treat principle. All methods were appropriate for cluster randomized stepped-wedge trials.⁴⁰ We used mixed-effects models to account for correlation of outcomes within surgery, patient, and cluster, as appropriate for each model. Because cluster randomization may not result in balance across all patient baseline factors, we summarized patient and surgical characteristics for each surgery by intervention status, calculating standardized mean differences (SMDs) for each characteristic; any factors with SMD greater than 0.1 were considered imbalanced and adjusted for in all analyses. Missing baseline characteristics, including baseline PROMIS scores, were imputed using multiple imputation when included in models.

PROMIS Outcomes Analysis

Baseline and follow-up PROMIS scores were summarized by time point and cluster status at index surgery. Differences in anxiety, pain interference, and physical function between intervention and UC periods were evaluated using mixed-effects linear models with follow-up scores as the dependent variable. Models were restricted to surgical procedures with at least 1 postoperative PROM. Models included random effects for surgery nested within patient and cluster (cross-classified to account for patients undergoing procedures in different clusters). Fixed effects included intervention status, study step, survey time (30, 60, or 90 days), baseline PROMIS score, and covariates with imbalance (SMD >0.1). Sensitivity analyses that included CAT and short-form data were identical. We performed 2 sensitivity analyses: first, we replicated the main analysis including CAT and short-form data; second, we replicated the main analyses using only the first surgery for patients with multiple surgical procedures. For the main analysis, we also performed an ad hoc exploratory noninferiority test, using 3 points as the minimally important difference.^41,42

Opioid Analysis

To reduce bias from correlated outcomes among patients with multiple procedures and avoid contamination, analyses were restricted to each patient’s first qualifying (index) surgery. In contrast, results reported in ClinicalTrials.gov reflect the full cohort, consistent with registry reporting conventions. In-hospital and postdischarge opioid exposure, defined by prescriptions, administrations, or either, was quantified as average daily MMEs, calculated using days hospitalized or days after discharge. Measures were summarized by intervention status, and the proportion of patients with any opioid exposure was calculated for each exposure definition.

Intervention effects were evaluated using mixed-effects models. Any opioid exposure was modeled with a logit link. MME exposure was modeled using generalized linear models with a log link and offset for days of exposure; exponentiated coefficients are presented as rate ratios (RRs). MME distributions were examined and trimmed at the 99th percentile to reduce the influence of extreme outliers. All models included an offset for exposure days (in-hospital or postdischarge) and covariates with imbalance as described above. Results are reported as odds ratios or exponentiated β coefficients with 95% CIs and 2-sided P values. Models that did not converge were assessed for collinearity and simplified as needed, with modifications noted in the Results.

Utilization Analysis

Utilization events were modeled separately using mixed-effects Poisson models with random effects for cluster and offsets for days at risk. Follow-up was censored at a subsequent qualifying surgery or death. We report crude and adjusted event rates and incidence rate ratios (IRRs) with 95% CIs and 2-sided P values.

All analyses were conducted using Stata statistical software version 19 (StataCorp). A 2-sided α level of .05 defined statistical significance. Primary comparisons were based on intervention vs UC periods; for simplicity, these are referred to as intervention and UC periods.

Results

Between October 16, 2020, and April 30, 2024, the NOHARM trial included 68 141 surgical procedures, comprising 62 892 index procedures in unique patients and 5250 subsequent procedures; for additional exclusion information, see eFigure 5 in Supplement 2. The stepped-wedge design included 30 sequence-periods (5 sequences × 6 periods); surgical volume by sequence-period is shown in the CONSORT flow diagram (Figure). Overall, 43 053 surgical procedures occurred during intervention periods and 25 088 during UC periods.

Figure. Flow Diagram for the Non-pharmacological Options in Post-Hospital and Rehabilitation Pain Management (NOHARM) Trial.

Cohort refers to all surgical procedures performed during the study period. A total of 18 surgical procedures were ambiguous with respect to intervention assignments due to surgery dates being moved up ahead of schedule (10 procedures) or those performed by 2 or more teams or at more than 1 location or over multiple dates, with conflicting intervention status. All such cases were allocated to intervention or control on a case-by-case basis.

Postoperative PROM response (≥1 completed primary outcome) was 71.7% (30 890 of 43 053 patients) during intervention periods and 68.3% (17 140 of 25 088 patients) during UC periods. Table 2 summarizes patient and surgical characteristics; model-specific analytic samples varied due to differential PROM completion (eAppendix 3 and eTables 4-6 in Supplement 2).

Table 2. Patient and Surgical Characteristics by Intervention Status.

Characteristic	Patients, No. (%)
	Usual care (n = 25 088)	NOHARM (n = 43 053)
Demographics
Age group, y
≤40	3681 (14.7)	7695 (17.9)
41-50	2878 (11.5)	4464 (10.4)
51-60	4507 (18.0)	7039 (16.3)
61-70	7236 (28.8)	12 051 (28.0)
71-80	5271 (21.0)	9302 (21.6)
>80	1515 (6.0)	2502 (5.8)
Gender
Female	10 429 (41.6)	16 818 (39.1)
Male	14 658 (58.4)	26 234 (60.9)
Nonbinary	0	1 (<0.1)
Missing	1 (<0.1)	0
Racea
American Indian or Alaska Native	185 (0.7)	314 (0.7)
Asian	547 (2.2)	1009 (2.3)
Black or African American	1044 (4.2)	1652 (3.8)
Native Hawaiian or Other Pacific Islander	42 (0.2)	71 (0.2)
White	22 682 (90.4)	38 970 (90.5)
Other	588 (2.3)	1037 (2.4)
Ethnicitya
Non-Hispanic	23 316 (92.9)	40 111 (93.2)
All others	458 (1.8)	907 (2.1)
Employment
Employed	11 229 (44.8)	20 549 (47.7)
Retired	10 310 (41.1)	16 528 (38.4)
Not employed or student or military duty	2341 (9.3)	4294 (10.0)
Disabled	1189 (4.7)	1573 (3.7)
Missing	19 (0.1)	109 (0.3)
Marital status
Married	17 015 (67.8)	29 672 (68.9)
Single	3772 (15.0)	6449 (15.0)
Divorced	1936 (7.7)	3263 (7.6)
Widowed	1891 (7.5)	2825 (6.6)
Life partnership	245 (1.0)	458 (1.1)
Separated	163 (0.6)	247 (0.6)
Missing	66 (0.3)	139 (0.3)
Education
Less than high school	4590 (18.3)	7143 (16.6)
High school graduate	737 (2.9)	1230 (2.9)
Some college, vocational, or associate degree	7365 (29.4)	11 660 (27.1)
Bachelor’s degree	5786 (23.1)	9175 (21.3)
Master’s or doctoral degree	4356 (17.4)	7068 (16.4)
Missing	2254 (9.0)	6777 (15.7)
Rurality (Rural Urban Commuting Area Codes)
Urban (1-3)	18 150 (72.3)	29 742 (69.1)
Micropolitan (4-6)	2661 (10.6)	5297 (12.3)
Small town (7-9)	2284 (9.1)	4217 (9.8)
Rural (10)	1835 (7.3)	3555 (8.3)
Missing	158 (0.6)	242 (0.6)
Site
Rochester, Minnesota	9812 (39.1)	23874 (55.5)
Scottsdale, Arizona	6551 (26.1)	5358 (12.4)
Jacksonville, Florida	4363 (17.4)	7290 (16.9)
Eau Claire, Wisconsin	799 (3.2)	4301 (10.0)
La Crosse, Wisconsin	1415 (5.6)	1265 (2.9)
Mankato, Minnesota	2148 (8.6)	965 (2.2)
Patient has surgical procedures in both periods	1786 (7.1)	1835 (4.3)
Surgery characteristics
Surgery: type
Gynecologic	3799 (15.1)	6837 (15.9)
Cesarean delivery	963 (3.8)	3600 (8.4)
Orthopedic	10 398 (41.4)	17964 (41.7)
Colorectal	4049 (16.1)	4244 (9.9)
Cardiac	1760 (7.0)	6205 (14.4)
Transplant	2982 (11.9)	2286 (5.3)
Pulmonary or thoracic	1137 (4.5)	1917 (4.5)
Surgery: same day	5544 (22.1)	10 971 (25.5)
Discharge disposition
Home	21 508 (85.7)	37 335 (86.7)
Died	74 (0.3)	170 (0.4)
Other	3506 (14.0)	5548 (12.9)
Surgery: laparoscopic	9139 (36.4)	12 908 (30.0)
Surgery: gastrointestinal or hip or knee replacement	14 285 (56.9)	21 768 (50.6)
Surgery duration, h
0	2043 (8.1)	5152 (12.0)
1	9852 (39.3)	18 050 (41.9)
2	5862 (23.4)	7622 (17.7)
3	4731 (18.9)	7393 (17.2)
4	2595 (10.3)	4831 (11.2)
Missing	5 (<0.1)	5 (<0.1)
Hospital length of stay, d
1	5544 (22.1)	10 971 (25.5)
2	6376 (25.4)	10 064 (23.4)
3	3321 (13.2)	4462 (10.4)
3	4222 (16.8)	6942 (16.1)
4-5	3799 (15.1)	7286 (16.9)
6-10	1826 (7.3)	3328 (7.7)
History of cancer	7798 (31.1)	11 947 (27.7)
Baseline scores, mean (SD)
Baseline pain interference	59.32 (9.92)	59.96 (9.71)
Missing	6110 (36.3)	11 101 (36.6)
Baseline physical function	40.50 (9.04)	39.50 (8.54)
Missing	6091 (36.8)	10780 (36.0)
Baseline anxious	50.73 (9.19)	51.85 (9.16)
Missing	4754 (42.5)	5661 (37.2)

Abbreviation: NOHARM, Non-pharmacological Options in Post-Hospital and Rehabilitation Pain Management.

^a Race and ethnicity were identified using self-reported patient tables and categorized as American Indian or Alaska Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White, or other. The other category included patients who chose not to disclose, were recorded as unknown, or were unable to provide this information.

Of the 68 141 surgical procedures, 61 652 (90.5%) occurred in patients identifying as White, 65 364 (95.9%) as non-Hispanic, 2696 (3.9%) as Black or African American, and 40 892 (60.0%) as male; the mean (SD) age was 59.27 (16.26) years. The most common procedures were orthopedic (28 362 procedures [41.6%]), followed by gynecologic (10 636 procedures [15.6%]), colorectal (8293 procedures [12.2%]), and transplant (5268 procedures [7.7%]). Baseline preoperative PROMIS T scores were similar between intervention and UC periods (Table 2). Consistent with the Zelen-type design, 29 278 of 43 053 intervention-period participants (68.0%) reviewed the Healing After Surgery guide preoperatively; 88.7% (38 179 of 43 053 participants) selected NPPC options, 50.4% (21 692 of 43 053 participants) used NPPC during hospitalization, and 52.8% (22 758 of 43 053 participants) reported NPPC use after discharge.²⁹

Co–Primary Outcomes: PROMIS Pain Interference and Physical Function

PROMIS pain interference and physical function scores, our co–primary outcomes, did not differ between intervention periods and UC periods (pain interference adjusted mean difference, 0.09; 95% CI, −0.18 to 0.36; P = .51; physical function adjusted mean difference, 0.15; 95% CI, −0.11 to 0.41; P = .27) (Table 3). Sensitivity analyses incorporating PROMIS short-form and CAT scores yielded similar results (eTables 7-9 in Supplement 2), whereas sensitivity analyses restricted to the first surgery for each patient were also similar to the main results (eTable 10 in Supplement 2). Noninferiority tests using 3 points as the minimally important difference were all significant indicating equivalence.^41,42

Table 3. Intervention Effects on Patient-Reported Outcomes.

Symptom	Adjusted mean difference (intervention vs control), β (95% CI)a	P value
Pain interferenceb	0.09 (−0.18 to 0.36)	.51
Physical functionc	0.15 (−0.11 to 0.41)	.27
Anxietyd	−0.07 (−0.44 to 0.29)	.69

^a All models were adjusted for baseline symptom score, study site, study step, age, rural status, and (if imbalanced with standardized mean difference >0.1 for that symptom cohort) laparoscopic status, gastrointestinal or hip surgery, surgery hours, prior prescription, and study step. Models also include random effects for cluster (care team) and patient (to account for multiple surgical procedures).

^b Data are from a sample of Pain Interference Patient-Reported Outcomes Measurement Information System (PROMIS) computerized adaptive test (CAT) responders (intervention, 25 981 participants; usual care, 15 807 participants).

^c Data are from a sample of Physical Function PROMIS CAT responders (intervention, 25 563 participants; usual care, 15 520 participants).

^d Data are from a sample of Anxiety PROMIS CAT responders (intervention, 15 227 participants; usual care, 11 195 participants).

Opioid and Other Secondary Outcomes

In prespecified secondary analyses (Table 4), patients receiving the NOHARM intervention experienced lower overall opioid exposure than patients receiving UC (adjusted RR, 0.90; 95% CI, 0.82-0.99; P = .03), corresponding to an adjusted mean difference of approximately 260 MMEs per patient. The overall reduction was driven primarily by lower opioid administration during hospitalization (adjusted RR, 0.94; 95% CI, 0.90-0.97; P = .001), an adjusted mean difference of approximately 85 MMEs. Opioid prescribing at hospital discharge was likewise lower (adjusted RR, 0.89; 95% CI, 0.84-0.94; P < .001), whereas opioid administration (adjusted RR, 0.96; 95% CI, 0.90-1.02; P = .15) and prescribing (adjusted RR, 1.05; 95% CI, 0.95-1.17; P = .34) after discharge did not differ significantly between groups.

Table 4. Adjusted Opioid Exposure and Prescribing Outcomes by Intervention Status.

Metric	MME (95% CI)				RR (95% CI)	P value
	UC	NOHARM	Adjusted UCa	Adjusted NOHARMa
Total opioids MME	398.0 (135.0-1140.0)	400.0 (135.0-1110.0)	2633.8 (1671.9-3595.8)	2373.6 (1521.0-3226.1)	0.90 (0.82-0.99)	.03
Administered opioids
During index hospitalizationb	90.0 (0.0-500.0)	60.0 (0.0-503.0)	1305.8 (977.6-1634.0)	1220.6 (914.0-1527.1)	0.94 (0.90-0.97)	.001
During subsequent hospitalizationc,d	0.0	0	160.2 (132.6-187.8)	153.1 (126.9-179.3)	0.96 (0.90-1.02)	.15
Prescribed opioids
Discharge from index hospitalizatione	113.0 (60.0-225.0)	113.0 (38.0-250.0)	99.87 (51.9-147.8)	88.9 (46.2-131.5)	0.89 (0.84-0.94)	<.001
Following index hospitalizationc,d	0.0	0.0	148.5 (113.2-183.8)	156.4 (119.4-193.4)	1.05 (0.95-1.17)	.34

Abbreviations: MME, morphine milligram equivalents; NOHARM, Non-pharmacological Options in Post-Hospital and Rehabilitation Pain Management; RR, rate ratio; UC, usual care.

^a Estimates were derived from multivariable mixed-effects generalized linear models with a log link. All models adjusted for study site, laparoscopic status, gastrointestinal or hip surgery, surgery hours, prior prescription, and study step, with random effect for cluster (care team).

^b Opioid medications (oral and parenteral) administered during hospitalization for the index surgery were comprehensively captured as documented in the medication administration record.

^c Opioid medications administered or prescribed by a health system clinician (surgical or nonsurgical) were captured, whereas those administered or prescribed outside the health system were not.

^d Dependent variable (MMEs) was trimmed at the 99th percentile due to extreme outlier values.

^e Administered MMEs represent opioid medications received by patients. Prescribed MMEs represent opioid quantities written on prescriptions and do not indicate whether medications were dispensed or consumed.

At 3 months, the odds of patients reporting receiving more than 3 opioid prescriptions were lower during intervention periods (3.32% vs 4.52%; odds ratio, 0.72; 95% CI, 0.57-0.92; P = .003). In contrast, binary self-reported opioid use (any vs none) did not differ between groups.

Adjusted mean PROMIS anxiety T scores were similar between intervention and UC periods (adjusted mean difference, −0.07; 95% CI, −0.44 to 0.29; P = .69) (Table 3). Health services utilization did not differ between intervention and UC periods, including emergency department visits (IRR, 0.98; 95% CI, 0.91 to 1.06; P = .67), hospitalizations (IRR, 0.95; 95% CI, 0.87 to 1.04; P = .26), or intensive care unit admissions (IRR, 0.81; 95% CI, 0.65 to 1.03; P = .08).

Opioid-related adverse events were uncommon and did not differ between groups. New opioid use disorder occurred in 320 participants (0.74%) during intervention periods and 203 participants (0.82%) during UC periods, whereas opioid poisoning or toxic effects occurred in 3 (0.01%) and 4 (0.02%) participants, respectively. Rates of falls, burns, musculoskeletal injuries, episodes meeting criteria for failure to rescue, and all-cause mortality (559 participants [1.30%] vs 320 participants [1.28%]) were likewise similar between groups.

Discussion

In this large cluster randomized pragmatic trial evaluating a multicomponent, EHR-embedded intervention promoting guideline-concordant perioperative pain care, we found no improvement in the co–primary outcomes of pain interference and physical function. However, the intervention reduced total postoperative opioid exposure (MMEs) by approximately 10% over 3 months without evidence of worsening patient-reported outcomes, opioid-related adverse events, or other monitored safety outcomes. These findings suggest that scalable, EHR-enabled strategies integrating patient education, preference elicitation, and CDS may reduce opioid exposure while preserving patient-centered outcomes

The observation of reduced opioid exposure without measurable improvements in pain interference or physical function warrants consideration. NOHARM intentionally used a pragmatic encouragement design that prioritized scalability over prescriptive changes in clinician behavior (Supplement 1). Patients and clinicians retained substantial discretion regarding whether and how NPPC was incorporated into routine care, likely reducing intervention intensity relative to more tightly controlled efficacy trials.²⁹ Implementation also occurred during the COVID-19 pandemic, when sites experienced fluctuating surgical volumes and workforce shortages that challenged routine clinical operations and limited opportunities for nurses and allied health professionals to reinforce NPPC within already demanding workflows. These findings suggest that pragmatic, low-touch implementation strategies may be better suited to influencing proximal care processes, such as opioid exposure, than more complex patient-centered outcomes.

Design features may also have reduced the ability to detect improvements in pain interference and physical function. Participating departments appropriately sought to include a broad spectrum of procedures, including minimally invasive laparoscopic and thoracoscopic operations from which patients generally recover rapidly but that occasionally result in persistent postoperative pain.^43,44 Although this enhanced the trial’s pragmatic relevance and generalizability, it may have contributed to low average postoperative pain and reduced opportunity to detect clinically meaningful improvements. Because many patients undergoing minimally invasive procedures recover within days to weeks, outcome assessments beginning 1 month after surgery may also have missed the interval during which NPPC exerts its greatest effects. Finally, incomplete outcome ascertainment may have attenuated observed treatment effects if patients experiencing greater symptom burden were less likely to respond.⁴⁵

Although NOHARM did not improve pain interference or physical function, the observed reduction in opioid exposure is clinically meaningful when interpreted within the context of other scalable opioid stewardship interventions. Unlike resource-intensive perioperative pathways, such as Enhanced Recovery After Surgery, or interventions that directly target opioid prescribing,⁴⁶ NOHARM promoted multimodal pain care through routine EHR workflows and existing allied health personnel. Viewed in this context, the approximately 10% reduction in MMEs and 28% reduction in patients receiving more than 3 opioid prescriptions suggest that relatively low-resource health system interventions can produce meaningful population-level reductions in opioid exposure. Although direct comparisons across studies should be interpreted cautiously, these reductions are comparable to those reported for substantially more resource-intensive perioperative opioid stewardship interventions.^47,48

Opioid stewardship is likely to require complementary interventions deployed at different stages along the continuum, from initial opioid exposure to opioid-related morbidity. NOHARM’s scalable EHR-based approach may complement interventions targeting later stages of opioid-related harm, such as the Healing Communities Study,⁴⁹ by reducing opioid exposure through expansion of NPPC rather than by directly restricting prescribing. More broadly, health care systems have an opportunity to design perioperative care that seeks both to optimize patient recovery and reduce downstream societal harms associated with opioid exposure.

The next challenge is not simply broader deployment of NPPC to reduce opioid exposure, but a more empirical understanding of how these strategies should be implemented. NOHARM paired EHR-based CDS with repeated patient engagement, preference elicitation, allied health reinforcement, and readily accessible NPPC resources across routine perioperative care. This implementation strategy may help explain why NOHARM reduced opioid exposure, whereas some prior EHR-based behavioral nudge interventions did not.^50,51 Whether earlier intervention or more individualized implementation strategies can extend the benefits of scalable NPPC interventions from proximal care processes to more distal patient-centered outcomes remains an important question for future research.

Limitations

Several considerations should temper interpretation of these findings. The intervention was implemented within a single integrated health system using a common EHR platform, which may limit generalizability to other health systems. Although the cohort reflected the racial and ethnic diversity of the participating health system, representation of some minoritized racial and ethnic groups was limited, precluding robust subgroup analyses. In addition, although the pragmatic design enhanced external validity, it necessarily included diverse surgical populations and care settings, potentially increasing heterogeneity of treatment effects. Furthermore, because NOHARM evaluated a multicomponent intervention, the relative contribution of individual intervention components could not be determined.

Conclusions

In this cluster randomized pragmatic trial of surgical patients, an EHR-based intervention promoting NPPC did not improve pain interference or physical function but was associated with lower postoperative opioid exposure without evidence of harm. Scalable, EHR-embedded, system-level approaches embedded within routine clinical workflows may contribute to safer postoperative pain management.

Supplement 1.

Trial Protocol and Statistical Analysis Plan

Supplement 2.

eAppendix 1. Additional methods, results, and discussion

eFigure 1. Identification of surgical procedures for trial inclusion

eFigure 2. NOHARM Intervention Conceptual Diagram

eFigure 3. Prototype of Healing After Surgery menu for NPPC selection.

eFigure 4. NOHARM trial stepped wedge design and implementation timeline across 22 clusters at 6 sites

eFigure 5. Reasons for exclusion and surgeries affected

eAppendix 2. Epic ORP procedure codes

eTable 1. Surgical Categories Included in the NOHARM Trial

eTable 2. NOHARM Trial Qualifying Surgeries

eTable 3. ICD-10 Codes Used to Identify New Occurrences of Opioid Use Disorder and Poisoning

eAppendix 3. Additional analyses conducted to evaluate the robustness of the primary study findings

eTable 4. Patient and surgical characteristics of participants included in Pain Interference Analyses

eTable 5. Patient and surgical characteristics of participants included in Physical Function Analyses

eTable 6. Patient and surgical characteristics of participants included in Anxiety Analyses

eTable 7. Number of PROMs completed via printed PROMIS short forms

eTable 8. Unadjusted mean Physical Function and Pain interference scores with and without printed short forms

eTable 9. Adjusted patient-reported outcome results

eTable 10. Sensitivity Analysis Restricted to Each Patient’s First Surgery

eReferences

Supplement 3.

Mayo Clinic NOHARM Research Team Members

Supplement 4.

Data Sharing Statement