Opioid intake and quality of life after hospital discharge from major surgery – a health economic evaluation

Abstract

Background:

Earlier studies have not examined the extent to which reducing postoperative opioid use after patient discharge would improve the quality of life of opioid naïve patients, knowledge needed for economic evaluations of drugs and devices (applications) that reduce morphine milligram equivalents. We quantified associations between postoperative opioid use and patient-centered health outcomes using the EuroQol group’s EQ-5D-5L values, recorded longitudinally.

Methods:

Prospective observational data were collected for adult patients, without regular preoperative opioid use, undergoing inpatient surgery at two US hospitals in 2022–2023. Patients were enrolled before hospital discharge. On the 7^th, 14^th, 21^st, and 28^th days after discharge, smartphone-based application notifications requested survey completion, including weekly opioid use and health-related quality of life.

Results:

The 606 patients had 2292 pairwise observations of EQ-5D-5L values and morphine milligram equivalents. Treating each patient as their own control, each one-unit increase in the square root of morphine milligram equivalents was associated with a 0.0108 reduction in the EQ-5D-5L utility value (standard error 0.0011, P <0.0001). If no patient had received opioids after discharge, the predicted improvement would have been a mean of 0.0436 (0.0044) EQ-5D-5L utility values per week. Sensitivity analysis #1 examined the association between mean EQ-5D-5L among weeks and mean morphine milligram equivalents among weeks, while controlling for procedural specialty. Each one-unit increase in the square root of the morphine milligram equivalents was associated with a 0.0166 (0.0026) reduction in EQ-5D-5L values (P <0.0001). Sensitivity analysis #2 examined partial Kendall tau_b between EQ-5D-5L and morphine milligram equivalents, using the 2292 pairwise observations, controlling for baseline observations when estimating the association for week 1, controlling for week 1 observations when estimating the association for week 2, etc. The four Bonferroni adjusted P ≤0.0040 showed that less opioid was associated with greater health-related quality of life.

Conclusions:

Three analyses using different assumptions showed that, among opioid naïve patients undergoing inpatient surgery, greater postoperative opioid consumption among surgical patients was associated with reduced health-related quality of life. However, the effect sizes were small, such that likely close to zero opioid intake after hospital discharge would be needed to exceed the minimal important difference. These results have implications for primary economic endpoints of future randomized trials of improved postoperative analgesia.

INTRODUCTION

Multiple studies have examined the effect of enhanced recovery programs, therapeutic interventions, and patient comorbidities on both postoperative opioid use and patient-centered health outcomes (Supplemental S1). Few have examined the association between postoperative opioid use and the health outcomes.^1,2 Among chronic opioid users undergoing major surgery, complete opioid cessation was associated with significant improvement in pain intensity and pain interference postoperatively as evaluated using the Patient-Reported Outcomes Measurement Information System (PROMIS).¹ In contrast, in a small cohort (N=98) of opioid naïve patients undergoing total hip or knee arthroplasty, multivariable regressions failed to find a significant association between postoperative opioid use and nine patient-centered endpoints,² with each false discovery rate adjusted P ≥0.24. In addition, meta-analysis for the effect of enhanced recovery after surgery for breast reconstruction quantified large reductions in total opioid use and hospital lengths of stay.³ However, no associated benefits were detected for hospital readmission, reoperations, complications, postoperative costs, or pain scores.³ These limited earlier results suggest that perhaps, among opioid naïve patients, reductions or even complete cessation of intake of opioid after hospital discharge are associated with negligible improvements in the patients’ quality of life. If true, the economic implications would be large (e.g., differences in postoperative morphine milligram equivalents would not be expected to have economic value for new analgesics).

Based on these earlier results (Supplemental S1),^1–3 we took a different approach to quantifying the association between postoperative morphine milligram equivalents and quality of life among patients without chronic opioid use. Mean morphine milligram equivalents over four weeks after discharge have a large correlation with the morphine milligram equivalents during the 24 hours before discharge (Spearman rho = 0.60, P <0.001).⁴ Therefore, each patient can be treated as their own control. Accordingly, we analyzed prospective observational data for the association between morphine milligram equivalents after discharge and the EuroQol group’s EQ-5D-5L value⁵ longitudinally over four weeks. This approach tested the association without the effect of patient factors invariant to time (e.g., preexisting psychological disease, education, socioeconomic status, type of surgery, or hospital length of stay).

METHODS

The University of Nebraska Medical Center Institutional Review Board approved study #0724-21-FB, which used EuroQol Research Foundation’s US English version of the EQ-5D-5L over four weeks.⁵ Prospective observational data were collected simultaneously with the randomized trial reported previously;⁶ ClinicalTrials.gov NCT05221866. Those variables are listed in Supplemental Table S2, our study dataset. The University of Iowa Institutional Review Board determined that project #202503256, analyzing the deidentified data, does not meet the regulatory definition of human subjects research. Our report follows the Strengthening the Reporting of Observational studies in Epidemiology (STROBE) format for cohort studies.⁷

Studied cohort

Patient enrollment was from May 2022 through December 2023.⁶ Eligible patients were 19 to 89 years old, underwent surgery at one of two midwestern hospitals (one academic, one community), had access to a smartphone, and were expected to be hospitalized at least one night postoperatively.⁶ Patients not eligible had been hospitalized within 30 days, were not able to read English, reported that they had used opioid on most days >3 months before surgery, had contraindications to opioids, acetaminophen, or nonsteroidal anti-inflammatory drugs, or were currently pregnant.⁶ Patients who had undergone cesarean birth were included as enrollment started postpartum (Supplemental Table S3).⁶ Patients were invited to participate after they had surgery, while in the hospital (“week 0”), usually the first or second postoperative day (Supplemental Table S4). Half of the patients used their smartphone application for data collection only, and the other half also received ongoing education on pain management (Supplemental Table S3). On the 7^th, 14^th, 21^st, and 28^th days after hospital discharge, smartphone notifications were sent reminding the patient to complete the survey questions on the application for the corresponding week. Patients completing all four weekly surveys received a US$40 gift card. Patients reported the opioids they took postoperatively via their smartphone application, which were then converted to morphine milligram equivalents (Table 1).⁸

Table 1.

Morphine milligram equivalents (mme) and EQ-5D-5L value, the analyses’ primary independent and dependent variables

Variable	Week	N	Mean	Standard Deviation	25th percentile	50th percentile	75th percentile	90th percentile
mme	0a	606	141.3	157.0	57.5	98.5	170.5	285.5
mme	1	596	51.4	95.9	0	15.0	63.8	135.0
mme	2	574	19.3	53.7	0	0	15.0	65.1
mme	3	568	9.0	32.7	0	0	0	22.5
mme	4	554	6.4	28.5	0	0	0	10.0c
mme	1–4b	606	23.2	53.5	0	5.6	22.5	67.5
EQ-5D-5L	0a	606	0.604	0.270	0.449	0.636	0.834	0.902
EQ-5D-5L	1	596	0.686	0.247	0.572	0.719	0.872	0.940
EQ-5D-5L	2	575	0.750	0.219	0.630	0.779	0.932	1.000
EQ-5D-5L	3	570	0.793	0.214	0.687	0.845	0.943	1.000
EQ-5D-5L	4	555	0.829	0.208	0.721	0.883	1.000	1.000
EQ-5D-5L	1–4b	606	0.759	0.206	0.660	0.804	0.914	0.970

^aThe morphine milligram equivalents for “week 0” was the total during the hospitalization, median 2 days (Supplemental Table S4). The EQ-5D-5L for week 0 was completed after surgery, soon before discharge.

^bThe means were calculated excluding missing values. There were 535 participants who completed all four weeks, 34 who completed three, 17 completing two, and 20 completing one. The means were calculated excluding missing values (i.e., the means of four, three, two, or one observation). The sample sizes for the means (“1–4”) were ten larger than for the first week (“1”) because there were ten participants who did not use the application during the first week after hospital discharge.

^cThere were 88% (487/554) who consumed no opioid during the fourth week after hospital discharge.

Studied outcomes

The primary independent variable was weekly self-reported oral morphine milligram equivalents in each of the first four weeks after discharge. Opioid consumption was calculated based on the patient-reported number of pills consumed. The sample size of patients was chosen for the previously reported randomized trial.⁶ The sample size was based on this outcome, because it was the primary endpoint of the randomized trial.⁶ All N=606 studied patients had at least one postoperative week when they reported their opioid consumption by using the phone application. An earlier prospective study cohort study examined a similar sample size of N=604 patients.⁴

The primary dependent variable for our study was health-related quality of life, measured using the EQ-5D-5L utility values (Table 1).^5,9 The five dimensions are mobility, self-care, usual activities, pain/discomfort, and anxiety/depression. Each of the five items are answered using five ordered responses.⁵ For example, the item on mobility had responses of “I have no problems in walking about” level 1, “I have moderate problems in walking about” level 3, and “I am unable to walk about” level 5.⁵ The index valuations for each of the 25 items were obtained from Pickard,¹⁰ using the Stata code provided by EuroQOL.¹¹ For each of the five dimensions, the valuation of the selected ordered response is used, and then the five dimensions’ valuations are summed to get the primary dependent variable. With the observations having been made weekly, this primary dependent variable (EQ-5D-5L) was the quality-adjusted weeks of life.^5,10

The independent and dependent variables of this prospective observational study are labeled in bold underline in Supplemental Table S2 and available in the publicly posted dataset¹² at https://doi.org/10.25820/data.007798. Supplemental Table S4 lists the other 28 patient assessments from Supplemental Table S2. These other assessments, collected simultaneously, included the EQ visual analog scale and several PROMIS health assessments.

Statistical analyses

The validity of the weekly EQ-5D-5L was evaluated by making comparisons with seven continuous variables for which associations were expected (e.g., EQ visual analog scale), based on each assessing dimensions of patients’ overall health status and well-being during their postoperative recovery. Supplemental Table S4 lists the 25 combinations of these endpoints and the week when measured. Concordance was measured using Kendall’s tau_b. Bonferroni adjusted P <0.05 were treated as statistically significant, the Bonferroni adjustment based on the 25 comparisons. These evaluations of validity are important because the sample size was based on this observational study’s primary independent variable of morphine milligram equivalents taken after discharge, not based on its primary dependent variable of patient quality of life (EQ-5D-5L). Sample sizes differed among weeks because some patients did not complete the survey every week (Tables 1–2).

Table 2.

Evaluating concurrent validity of EQ-5D-5L values using Kendall’s tau_b

Value	Week	N	taub	Adjusted P
Patient Health Questionnaire-8; 0=no depression 24=severe	0	606	−0.21	<0.0001
Days away from work/training due to pain	1	596	−0.20	<0.0001
Days away from work/training due to pain	2	575	−0.27	<0.0001
Days away from work/training due to pain	3	570	−0.24	<0.0001
Days away from work/training due to pain	4	554	−0.27	<0.0001
Partner, relative, or friend’s days caring for patient	1	596	−0.19	<0.0001
Partner, relative, or friend’s days caring for patient	2	575	−0.25	<0.0001
Partner, relative, or friend’s days caring for patient	3	570	−0.15	0.00018
Partner, relative, or friend’s days caring for patient	4	555	−0.19	<0.0001
EQ-5D-5L Visual Analog Scale 0 worst to 100 best	1	590	0.40	<0.0001
EQ-5D-5L Visual Analog Scale 0 worst to 100 best	2	569	0.45	<0.0001
EQ-5D-5L Visual Analog Scale 0 worst to 100 best	3	564	0.47	<0.0001
EQ-5D-5L Visual Analog Scale 0 worst to 100 best	4	551	0.49	<0.0001
PROMIS pain intensity 3a t-score	1	596	−0.47	<0.0001
PROMIS pain intensity 3a t-score	2	575	−0.55	<0.0001
PROMIS pain intensity 3a t-score	3	570	−0.58	<0.0001
PROMIS pain intensity 3a t-score	4	555	−0.62	<0.0001
PROMIS pain interference 8a t-score	1	596	−0.50	<0.0001
PROMIS pain interference 8a t-score	2	575	−0.59	<0.0001
PROMIS pain interference 8a t-score	3	570	−0.64	<0.0001
PROMIS pain interference 8a t-score	4	555	−0.66	<0.0001
PROMIS sleep disturbance 4a t-score	1	596	−0.34	<0.0001
PROMIS sleep disturbance 4a t-score	2	575	−0.38	<0.0001
PROMIS sleep disturbance 4a t-score	3	570	−0.38	<0.0001
PROMIS sleep disturbance 4a t-score	4	555	−0.40	<0.0001

P-values are reported after Bonferroni adjustment for the 25 comparisons. These are the variables in Supplemental Table S4 measured at baseline and then weekly.

The morphine milligram equivalents and the EQ-5D-5L were measured for the five weekly periods for each patient (Table 1). Our primary analysis related EQ-5D-5L utility values (dependent variable) to morphine milligram equivalents (independent variable) using a first-order autoregressive model. We used the Stata v18.5 xtregar command (StataCorp, College Station, Texas). A square root transformation of morphine milligram equivalents was applied to achieve a linear association with EQ-5D-5L values. An autoregressive model was used based on the Wooldridge test showing significant first-order autocorrelation (P <0.00001). The Nagar (i.e., adjusted Durbin-Watson) method was employed for estimating autocorrelation due to its statistical efficiency for small autocorrelation. A fixed effects model was used because patients with lower baseline EQ-5D-5L values had systematically larger morphine milligram equivalents during later weeks (i.e., Hausman and Mundlak tests both P <0.00001). One advantage of the autoregressive model was that each patient served as their own control. Thus, time-invariant factors that could bias the results (e.g., specialty or randomized study group) were collinear with the participant’s fixed effect and dropped out. Another advantage was that the marginal effect of opioid consumption was estimated. Specifically, the mean and standard error of the mean were calculated pairwise by patient for the difference in the EQ-5D-5L values, comparing scenarios where the patient used opioids versus no opioid use. The changes per week in EQ-5D-5L utility values represent differences in quality-adjusted weeks.^5,10 We used the standard errors to address the adequacy of the sample size in the second paragraph of the Discussion below.

The first sensitivity analysis associated the mean EQ-5D-5L among weeks 1–4 with the square root of the mean morphine milligram equivalents over those weeks. Because only one observation was taken for each participant, there could be confounding from covariates associated with both the dependent and independent variables. Supplemental Table S5 shows the evaluation of such potential associations for the categorical (Table S3) and continuous (Table S4) baseline variables. Associations were evaluated using Kruskal-Wallis and Kendall’s tau_b. Both tests’ results are invariant to the square root transformation of morphine milligram equivalents, because they both analyze ranks. No adjustments were made for the multiple comparisons to be deliberately anti-conservative. The one covariate significant for both, specialty, was then included in the linear regression model, with the dependent variable of mean EQ-5D-5L and the independent variables of the square root of the mean morphine milligram equivalents and the specialty category. (Because means among weeks were compared with means among weeks, there was one analyzed observation per subject (N=606) and thus no autocorrelation.) The hc3 bias correction was used for heteroskedastic robust standard errors;¹³ Stata regress command with the var(hc3) option. The mean EQ-5D-5L values were then plotted against the mean morphine milligram equivalents, with the best-fit line presented after subtracting out the estimated effect of each specialty.

The second sensitivity analysis did not provide for a marginal effect (i.e., effect size) but had the advantage of being distribution-free. Partial Kendall tau_b were estimated between week 1 EQ-5D-5L and week 1 morphine milligram equivalents while controlling for the week 0 (baseline) variables. The corresponding process was used for the next three weeks. Jackknife standard errors were estimated using the Stata parttau command. Because this approach analyzed each week separately, unlike the first-order autoregressive model, Bonferroni adjustment was made for four comparisons.

RESULTS

Table 1 summarizes the morphine milligram equivalents taken by patients and EQ-5D-5L United States utility values over four weeks post-discharge for the 606 surgical patients. Patient and case demographics are given in Supplemental Tables B and C. Study validity was evaluated by comparing EQ-5D-5L values before hospital discharge and for each of the four postoperative weeks with other factors expected to have associations: Patient Health Questionnaire-8 screen for severity of depressive symptoms; days away from work or training; partner, relative, or friend days away from work or training to help; EQ-5D-5L visual analog scale overall assessment of health; PROMIS pain intensity; PROMIS pain interference; and PROMIS sleep disturbance. Comparisons showed significant, expected associations with EQ-5D-5L, supporting concurrent validity (all 25 Bonferroni adjusted P ≤0.0002; Table 2). For example, patients with lower quality of life also reported more days away from work or training due to pain. Concurrent validity means, functionally, that the large observed variability in the measured EQ-5D-5L among patients (Figure 1) represents variability in quality of health among patients rather than measurement error.

FIGURE 1.

Association between the mean EQ-5D-5L values among the 4 postoperative weeks and the patient’s corresponding average morphine milligram equivalent opioid consumption among those 4 weeks. The Figure excludes 7 data (patients), but all were included to produce the predicted line. There were 5 excluded with morphine equivalents > 250 mg and 2 more excluded with EQ-5D-5L less than zero. To produce the line, the predicted values were estimated for each patient. Then, the corresponding estimate for each patient’s specialty was subtracted from the estimate. Thus, while each observation in blue incorporates specialty, the red line has it averaged out. The figure shows graphically the association between EQ-5D-5L values and opioid consumption, P <0.0001, by all three analysis methods in the results. The figure also shows that the magnitude of the effect is small relative to the variation in EQ-5D-5L among patients. However, because the means within subject are shown, unlike as used in our primary longitudinal analysis, the figure gives a biased impression of higher EQ-5D-5L and lower morphine milligram equivalents of opioids used.

Among the 606 patients, there were 2292 pairwise observations of EQ-5D-5L values and morphine milligram equivalents taken, pairwise by combination of subject and week (Table 1). The estimated lag-one autocorrelation among weeks was 0.20. Each one-unit increase in the square root of the morphine milligram equivalents was associated with a 0.0108 reduction in the EQ-5D-5L value (standard error 0.0011, P <0.0001). If no patient had received opioids after discharge, the predicted improvement would have been a mean of 0.0436 (0.0044) EQ-5D-5L utility values per week. This estimate can be compared with the minimally clinically important difference in the EQ-5D-5L value among surgical patients, estimated using external criteria (i.e., anchors).¹⁴ The estimates from polynomial regression and their 95% confidence intervals are given in Table 4 of Reference (14). The estimates depend on the baselines.¹⁴ Our result that zero opioid use after discharge would increase the predicted EQ-5D-5L by a mean of 0.0436 (0.0044) per week means totals of 0.17 (0.02), where 0.17 = (4 weeks) × 0.0436 per week. Substituting our mean EQ-5D-5L (0.759) from the last row of our Table 1 into Reference (14)’s Table 4, the corresponding minimally important difference would be a total of 0.08 (0.04) EQ-5D-5L utility values.¹⁴ Comparing the 0.17 and 0.08, the estimated Z = 2.12, with two-sided P = 0.034; see Discussion.

For each of the 606 patients, the mean EQ-5D-5L among weeks was associated with the mean morphine milligram equivalents taken among weeks while controlling for surgical specialty (Table 1; Figure 1). Each one-unit increase in the square root of the morphine milligram equivalents was associated with a 0.0166 reduction in the EQ-5D-5L value (standard error 0.0026, P <0.0001). (Note the considerably larger standard error, as expected by not relying on the data sequence within participant.) If no patient had taken opioids after discharge, the estimated gain in EQ-5D-5L utility values would have been a mean of 0.052 (0.008) per week.

Partial Kendall tau_b values were estimated using the 2292 pairwise observations, controlling for the baseline observations when estimating the association for week 1, controlling for the week 1 observations when estimating the association for week 2, and so forth (Table 3). The four Bonferroni adjusted P ≤0.0040 values showed that less opioid use was associated with greater health-related quality of life. The smallest partial Kendall tau_b was −0.296, statistically reliable (adjusted P <0.0001) but reflecting the large variability in the EQ-5D-5L utility values among patients compared to the functional association with morphine milligram equivalents taken.

Table 3.

Partial Kendall’s tau_b between ED-5D-5L and morphine milligram equivalents for each week while controlling for those observations from the preceding week

Statistic	Week 1	Week 2	Week 3	Week 4
Partial Kendall’s taub, controlling for preceding week	−0.232	−0.085	−0.120	−0.094
98.75% two-sided confidence interval	−0.296 to −0.167	−0.149 to −0.021	−0.186 to −0.053	−0.166 to −0.023
P-value adjusted for the 4 comparisons	<0.0001	0.0035	<0.0001	0.0040
Sample size (N) of patients for preceding three rows	596	564	558	547
Kendall’s taub, for comparison	−0.295	−0.262	−0.289	−0.277
P-value adjusted for the 4 comparisons	<0.0001	<0.0001	<0.0001	<0.0001
Sample size (N) of patients for preceding two rows	596	574	568	554

DISCUSSION

Before the current study, it was known from a similarly sized cohort that opioid use for four weeks postoperatively had the strongest observed correlation with the opioid use during the preceding 24-hours before hospital discharge.⁴ The current study used that knowledge of the importance of each patient’s baseline to design a longitudinal study and analysis that quantified the influence of opioid use after discharge on quality of life over those four postoperative weeks among patients without chronic opioid use. More than two dozen other variables had significant (adjusted P <0.0001) expected association with the EQ-5D-5L values, showing concurrent validity. Three analyses with different assumptions showed that, as expected, lesser opioid use was associated with greater quality of life, including when controlling for the patient’s quality of life and opioid consumption during the preceding week(s) and when including patient-specific covariates. All three analyses showed reliable associations. However, the important finding was that the effect size was small compared to the large variability in quality of life, pain intensity, and pain interference among patients. These results are especially novel because they showed not only that patients who continued to use opioids had lower EQ-5D-5L values (1^st sensitivity analysis) but that each patient’s quality of life was improved little by their individual reduction in opioid consumption (primary and 2^nd sensitivity analyses). These results are especially dependable because even unobserved variables (i.e., unmeasured factors) cannot have influenced our primary results if time-invariant over the four weeks studied. For example, patient characteristics such as educational level and socioeconomic status, surgeons’ practices such as specialty, and hospital length of stay do not affect our primary analysis results, point estimates or confidence intervals, because such terms drop out from the autoregressive model.

There are three implications of our comparison of the estimated reduction in the total EQ-5D-5L from complete cessation of use of opioid postoperatively. First, our study’s sample size was sufficient based on the total EQ-5D-5L exceeding the minimally important difference, P <.05. Second, a perfectly effective program for reducing opioid use postoperatively after major surgery in patients without chronic opioid use would meet the threshold¹⁴ for the minimum clinically important difference. However, for future investigations, the most important finding of our study is likely that achieving this economic goal depends on the use of no opioid after hospital discharge. Such reductions may be impractical after major surgery given that the associated randomized trial of a phone-based application for postsurgical medication awareness and guidance achieved no reduction in mean morphine milligram equivalents.⁶ (Such reductions may also be counter-productive if associated with greater pain.) Third, the current study’s sample size was based on the independent variable of morphine milligram equivalents taken, the primary endpoint of the earlier-reported randomized clinical trial.⁶ Prospectively choosing the sample size of the current study instead based on EQ-5D-5L utility values was not feasible because Cheng et al.’s study was published¹⁴ after the current study’s patient data collection had been completed.

As an economic study, our results were designed for economic guidance, unlikely clinical. For example, our study’s results of significant but (very) small effect sizes provide guidance on the primary economic endpoints of future randomized trials of improved post-operative analgesia. Specifically, our study’s results show that improving patients’ quality of life is an impractical primary economic goal. Most patients are not chronic opioid users, matching the population in our studied cohort. In contrast, enhanced recovery programs aimed at reducing postoperative opioid use while these patients are in the hospital are associated with reduced lengths of stay.³ Managerial epidemiology studies show that hospitals focus on this endpoint.¹⁵ When such reductions substantively increase the percentages of patients having outpatient or overnight stay procedures, throughput increases can be realized.¹⁶ When the percentages exceed some threshold such as 90%, then procedures can be performed outside of inpatient surgical suites, achieving even greater societal benefits because the surgical cases are no longer disrupted by the higher risk urgent surgical cases and elective cases no longer prevent timely care of those sicker patients.^17,18 For the patients undergoing elective surgery and having overnight stay, the length of stay in units of hours are unrelated to cost because discharge times are generally early the next day.¹⁹ Consequently, the clinical trial endpoint to use for future economic studies is literally whether the patient had overnight or ambulatory stay versus longer.^20,21

Our results were consistent with earlier work (Supplemental S1).^1,2,22,23 Chua et al. found that insurer-mandated opioid prescribing limits were associated with only slight decreases in postoperative pain and without significant change in patients’ satisfaction with the surgical experience.²² Debbi et al. reported that among opioid naïve patients undergoing total hip or knee arthroplasty, multivariable regressions for nine patient-centered endpoints failed to find a significant association,² with each false discovery rate adjusted P ≥0.24. Our cohort size completing postoperative patient-reported outcomes was six times larger and we used a longitudinal design.² Debbi et al. did not report confidence intervals for us to make pairwise comparison by endpoint.² Nevertheless, our finding of a statistically significant but small effect size shows qualitative concordance. Gong et al. observed in New Zealand that among opioid naïve patients prescribed opioids upon hospital discharge, 9.1% (23,656/260,726) had ongoing prescription for opioid 91 days after hospital discharge.²³ Matching, our cohort had 12% reporting opioid administration during the fourth week after hospital discharge (Table 1).²³ Finally, Holeman et al. reported that complete opioid cessation after surgery significantly improved patient-reported pain measures, but among chronic opioid users.¹

Regarding limitations, our work is not generalizable to patients with chronic opioid use; see our Supplemental Table S4.^1,24 After emergency surgery, patients with opioid use disorder have significantly greater healthcare spending (P =0.005), emergency department utilization (P =0.001), and post-discharge inpatient days (P <0.001).²⁵ Second, like Debbi et al.,² our study was from the United States. Results may be sensitive to the health-state valuations that we used,¹⁰ suitable for the USA overall, but perhaps not subpopulations. Finally, the primary independent variable of morphine milligram equivalents taken after discharge was based on patient self-reporting of opioid consumption. However, in emergency departments, assessment of 14-day opioid intake using self-report versus estimates made by counting remaining opioid pills revealed between revealed high agreement between both approaches, with estimated intraclass correlation coefficient of 0.992.²⁶

In conclusion, multiple statistical analyses using different assumptions showed that, among opioid naïve patients undergoing inpatient surgery, greater postoperative opioid consumption surgical was associated with reduced health-related quality of life. However, the effect size was small, such that future clinical trials would need to achieve zero opioid intake after hospital discharge to exceed the minimal important difference.

Key points:

Question:

Among opioid naïve patients undergoing inpatient surgery, what are the associations between postoperative opioid use and patient-centered health outcomes using the EuroQol group’s EQ-5D-5L utility values, recorded longitudinally.

Findings:

Based on prospectively collected data from 606 patients, treating each patient as their own control, if no patient had received opioids after discharge, there would have been a statistically significant improvement in their quality of life (P <0.0001), but very small, a mean of 0.0436 (0.0044) EQ-5D-5L utility values per week.

Meaning:

Although greater postoperative opioid consumption among surgical patients was associated with reduced health-related quality of life, the effect sizes were sufficiently small that likely close to zero opioid intake after hospital discharge would be needed to exceed the minimal important difference, a result with substantial implications for primary economic endpoints of future randomized trials of improved postoperative analgesia.