Abstract
Background
The risk of persistent postoperative opioid use (PPOU) and its association with the type of surgery are still unclear in Germany.
Methods
We conducted a nationwide retrospective cohort study on the basis of claims data from BARMER, a statutory health insurance carrier in Germany. Opioid-naive adults who did not have cancer and who underwent inpatient surgery in 2018 were included in the study. The operations were divided into 103 categories. PPOU was defined as the prescribing of opioids between postoperative days 1 and 90 and also between postoperative days 91 and 180 after hospital discharge. Patient-associated risk factors in the 12 months before surgery were investigated.
Results
203 327 patients were included. 1.4% had PPOU (95% confidence interval [1.4; 1.5]). There were major differences between operation groups: major amputations and orthopedic procedures carried the greatest risk for the development of PPOU. The type of surgery had a larger effect on the risk of PPOU than pre-existing risk factors (explained variance 22.3% vs. 14.3%). Among such factors, alcohol abuse and pre-existing treatment with antidepressant drugs were associated with the highest risk for PPOU (odds ratios [OR] 1.515 [1.277; 1.797] and 2.131 [1.943; 2.336]).
Conclusion
The incidence of PPOU in Germany is low (1.4%). The type of surgery plays an important role in its development.
Opioid misuse in the United States is still a major problem and leads to high rates of overdose-related deaths (24.2/100 000 population per year between 2018 and 2021) (1). Prescription after surgery is often the patient’s first exposure (2, 3) and may trigger medically non-indicated prolonged postoperative opioid use (PPOU) (4). Knowledge of the influence of type of surgery is limited as most studies included only a single type of surgery or single surgical specialty (5–7). The few studies with a mixed surgical sample focused only on a narrow set of 8–13 pre-selected operations or used a coarse classification by organ system or surgical site (8–11). Therefore, two recent systematic reviews were not able to draw definitive conclusions on the importance of type of surgery for the development of PPOU (5, 7).
Comparisons of incidence of PPOU across countries are challenging, due to different health care systems and to the wide variation in study methodology, especially regarding the many different definitions of PPOU that have been used (6). The only single-center study on PPOU in Germany found that the rate of opioid use 6 months after surgery was four to seven times higher for joint and back surgery than for urological surgery (12).
Therefore, the aims of our study were (a) to assess the incidence of PPOU in a large, population-based sample in Germany and (b) to investigate differences in the risk for PPOU among a large variety of surgical procedures.
Materials and methods
Study design
We conducted a retrospective cohort study based on German administrative health care claims data. Opioid-naïve, adult patients without cancer who underwent inpatient surgery in 2018 were selected for analysis. The year 2018 was chosen so that a follow-up for up to 12 months was completed before the start of the COVID-19 pandemic, which led to severe restrictions on surgery and aftercare. The influence of both pre-existing patient-related factors and type of surgery on PPOU was assessed descriptively and by regression analysis. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of the Jena University Hospital (approval number 2020–1952-Daten) in October 2020. As the claims data were anonymous, informed consent was not required. The description of the study follows the Reporting of studies Conducted using Observational Routinely-collected Data (RECORD) guidelines (13).
Data source
The study is based on anonymized health claims data provided by the statutory medical health insurer BARMER GEK, the second largest German health insurance fund with approximately 9 million members, 10% of the total German population. The data were provided via the BARMER scientific data warehouse and were assessed using controlled remote data processing.
Patient cohort
Adult BARMER insurance holders who received an inpatient surgical procedure (coded according to the German operation and procedure classification system [OPS], chapter 5) in 2018 were included. In the event of multiple hospital stays for surgery, the first stay in 2018 was chosen as the index hospitalization. Opioid naivety was defined as not having received an opioid prescription in the year prior to the index hospitalization. Patients with a cancer diagnosis (ICD-10 code C00-C97) at any time during the period from 12 months before to 12 months after the surgery were excluded, as they might have received opioids primarily due to cancer-related pain. Patients had to be continuously insured with BARMER throughout the observation period (see eMethods for further details).
eSupplement.
eMethods
Selection of the study population
Our aim was to identify an incident onset of opioid use that could be associated with the surgery. For that purpose, it was not so relevant whether patients had used opioids at any time previously, but rather whether they had used opioids during a reasonably long period before surgery. Most studies on PPOU use an opioid-free period of up to 1 year before surgery as definition of opioid naive (7). The same is true for pharmacoepidemiological studies in general (e1). Although we do not exclude every possible previous opioid consumption, we use the term “opioid-naive” for consistency with the conventional terminology in the international literature.
Cancer patients often suffer from tumor-associated pain, which is often treated with opioids. However, the analysis of opioid prescriptions in German health claims data does not allow the reason for prescribing to be identified. Including cancer patients would result in biased comparisons between different types of surgery, which could hardly be controlled based on available data. Therefore, we deemed exclusion of cancer patients from our study to be the best solution from a methodological standpoint. For similar reasons, the most prominent systematic review and meta-analysis on PPOU also excluded cancer patients (7). The relation between surgery and PPOU among cancer patients should be investigated in a separate study.
Definition of variables
Process of variable creation
All variables were defined based on claims records provided in the data warehouse. Therefore, we established a standardized quality routine on the basis of a review of previous literature. Definitions were drafted by an experienced anesthesiologist (JD). A senior clinical pain researcher (WM), and a senior research physician with high experience in claims data (CFS) reviewed the definitions. Variables were checked for plausibility based on descriptive analyses. If necessary, definitions were also discussed with senior clinicians experienced in pain research (CK, TV), medical controllers, and BARMER’s experts on claims data analyses.
Definition of opioid use
Medications were defined based on Anatomical Therapeutic Chemical Classification (ATC) codes (eTable 1). Postoperative opioid use was defined by analyzing the respective ATC codes (N02A) in health claims data, which include prescriptions by physicians in the outpatient sector. In-hospital perioperative opioid use cannot be obtained from German claims data, as these do not include medication in the inpatient sector. Codes for codeine (mainly used in Germany for its antitussive effect), methadone, and levomethadone (both mainly used in Germany as substitution treatment for opioid dependence) were not included. More than 20 different definitions of PPOU have been reported in the literature (6, 7). A relative majority of studies defined PPOU as at least one prescription of opioids in the period between 91 and 180 days after surgery (6, 7), with some also demanding a perioperative or early post-discharge prescription. Among these studies were two large, population-based, mixed surgical studies from the USA by Brummett et al. and from Canada by Clarke et al. (8, 10). To increase comparability, we therefore followed this approach and used exactly the same definition of PPOU as Clarke et al. (8). This definition includes the criteria of at least one opioid prescription (Anatomical Therapeutic Chemical Classification [ATC] codes N02A) during the first 90 days after discharge and at least one additional opioid prescription between days 91 and 180.
Beside definitions based on prescriptions of opioids in specific time periods, the international literature also reports definitions that demand a minimum number of days of supply (6). In addition, it could be of interest to assess and compare overall opioid dosages during a certain time frame (5). We decided not to rely on one of these definitions for several reasons. First, as explained above, using a definition based on prescription of opioids between day 91 and day 180 had the highest comparability to previous research. Second, German claims data do not provide a valid way to define days of consumption, since the prescription scheme is not documented. Using the pharmaceutical registration number (Pharmazentralnummer, PZN), package size can be determined. Based on this, days of consumption can only be estimated by using the defined daily dose (DDD), which measures the average dosage per day of a certain medication for its primary use among adults. Thus, the assumption is made that every patient takes opioids according to a “typical scheme”, regardless of his/her individual indication. It has been shown that DDD can severely underestimate as well as overestimate the actual opioid consumption in chronic pain patients (e2), which would have caused a bias in our calculations. To assess opioid dosages, different medications need to be standardized to the same metric. This could be done by transformation of the DDD to oral morphine equivalent dosage using a multiplier specific to type of opioid and route of administration. Unfortunately, the respective transformations are not precisely and bindingly defined for all relevant opioids. This would have added a second source of bias in addition to the shortcomings of DDD. In sum, these are the reasons why we did not rely on days of supply or dosages for our study but chose the less complex, and also less biased and more comparable definition given above.
Classification of surgical procedures
Chapter 5 of the German operation and procedure classification system (OPS) lists surgical procedures and includes over 20 000 individual codes with up to six digits. Therefore, it was necessary to define a reduced number of meaningful, clinically homogeneous, and distinct groups. Using only the chapter headings, as done in a comparable Swedish study (11), is insufficient, as this results in groups that contain very different types of minor and major surgery. Therefore, we adapted the most extensive meaningful OPS-based grouping of surgical procedures, which was presented by Gerbershagen et al., who defined 179 surgical groups based on data from a large German pain registry (14). Adaptation was necessary, since while the pain registry contained one major OPS code, claims data usually contain several OPS codes per instance of surgery/hospitalization without ranking them by importance. This led to assignments of many patients to two or more of the original Gerbershagen groups. Therefore, we revised or aggregated empirically overlapping groups until a largely unambiguous allocation was achieved. This led to reduction of the number of groups from 179 to 103. The codes, which define each of the resulting surgical groups, are listed in eTable 2. Patients who were still allocated to more than one of the 103 groups were amalgamated in the additional category “multiple defined surgical groups”.
Definition of pre-existing risk factors
We defined pre-existing risk factors based on claims data from the year before the index hospitalization. Risk factors were selected if they had shown significant effects in at least one of five recent meta-analyses on PPOU and could be operationalized in German health claims data (7, 19, 20, 22, 23). Factors closely related to the indication for specific operations were not included, since this would have led to confusions with the classification of types of surgery (e.g., back pain being closely related to spine surgery). Likewise, any factors that did not precede the surgery—e.g., dosage of postoperative opioids—were not included, since they would themselves be influenced by the types of procedure resulting in biased results.
We included variables from the following domains:
Demographics (age and sex)
Overall comorbidity (number of hospital stays and number of comorbidities according to the Charlson Comorbidity Index [e3])
Specific somatic comorbidities (anemia, pulmonary disease, cardiovascular disease, diabetes)
Psychiatric comorbidities (depression, anxiety, bipolar disorder, other affective disorder, posttraumatic stress disorder, other psychiatric comorbidity)
Substance abuse (alcohol, tobacco, cocaine, other)
Pre-existing pain conditions (chronic pain, fibromyalgia, migraine)
Medication use (antidepressants, benzodiazepines, non-opioid analgesics)
Somatic and psychiatric comorbidities, substance abuse, and pain conditions were defined based on inpatient and outpatient ICD-10 German Modification (GM) codes (eTable 3). Medication use was assessed using ATC codes in outpatient prescriptions (eTable 1).
Stratification of risk of PPOU per type of surgery
The random-intercept logistic regression model allowed us to incorporate far more types of surgery in a model together with pre-existing risk factors than had been possible in previous studies on PPOU (8–11). This comes with a limitation in calculation of adjusted estimates of risk of PPOU per surgery group. From the random-intercept model we can only obtain so-called maximum a posteriori estimates, which are subject to shrinkage effects, which leads to a statistical reduction of the risk for PPOU in smaller groups. At the same time, less frequent surgical procedures, such as amputations, tended to be associated with higher average comorbidity, which leads to a reduction in adjusted risk of PPOU in addition to the shrinkage effect. Thus, individual estimates of adjusted risk of PPOU per surgical group would be misleading and hard to interpret from a clinical point of view.
Post-hoc analyses on etiology of amputations
Due to the high risk of PPOU following amputations, we conducted additional post-hoc analyses on the etiology. We defined ischemic etiology and traumatic etiology based on the presence of specific ICD codes (see eTable 3). Codes that were indicative of ischemic etiology were considered if present in the 6 months immediately prior to the index stay or during the index stay in any of the considered sectors (inpatient or outpatient), while codes indicative of traumatic etiology were considered if present in the period of 1 month immediately before the index stay or during the index stay, but only considering the inpatient sector. It was possible that both definitions or none of the definitions applied. We compared the incidence of PPOU between the group with uniquely ischemic and the group with uniquely traumatic etiology.
Results
Results of post-hoc analysis on amputations
The majority of amputations were due to ischemia (81.7%), only a minority to trauma (1.4%, see eTables 6 and 7). The incidence of PPOU was 7.6% (95% CI [6.2; 9.2]) in the ischemic group, and 19% [7.7; 40] in the traumatic group, with no significant difference between the two (p = 0.124).
Discussion
Comparison of risk of PPOU with results from previous international studies
Comparing our results with previous studies in terms of the risk of PPOU as well as the importance of surgery is difficult. The major reason are the large methodological differences between studies. Most important are the different definitions of PPOU. Jivraj et al. showed that risk of PPOU had a range between 0% and 14.4% in the same sample, depending on which of 25 different definitions was used (6). Another major difference is the inclusion of only opioid-naive patients or also opioid-experienced patients. Further variety in methods stems from other aspects such as:
Age range
Inclusion or exclusion of cancer
Measurement of PPOU by self-report or based on claims data
Therefore, systematic reviews and meta-analyses did not arrive at conclusive results. Lawal et al. were not able to show significant differences between minor and major surgery, orthopedic and non-orthopedic surgery, or studies in the USA and elsewhere due to the considerable between-study heterogeneity (7). Likewise, Sitter and Forget concluded that it is not possible to describe persistent opioid use in relation to specific countries or types of surgery in their systematic review of European studies (17).
The best comparability for our results is given for other mixed-surgical studies which include opioid-naive patients and use the same or comparable definitions of PPOU (6, 8, 10). Clarke et al., who used the same definition of PPOU, found a significantly higher incidence of 3.5% in Canadian patients (8). They also reported a higher incidence of PPOU for selected individual operations (open lung resection: 8.5% in Clarke’s study vs 2.8% in our analysis; minimally invasive colorectal surgery: 3.2% in Clarke vs 0.5% for laparoscopic sigmoidectomy in our analysis, open hysterectomy: 2.5% vs 0.9%). Comparability is limited, since Clarke only included patients of 66 years or older, included cancer patients, but excluded palliative use of opioids, which would not be possible based on German claims data. Brummet et al. used a similar definition of PPOU (at least one prescription between 30 days before surgery and 2 weeks after discharge in addition to at least one prescription between day 90 and day 180), but included only patients below 65 years of age (10). They reported incident rates of PPOU between 5.9% and 6.5% after minor and major surgery in a representative cohort from the USA. Jivraj et al. included adult patients from the state of Ontario, Canada, included both outpatient and inpatient surgery, and excluded patients who visited palliative care services or had a prolonged hospital stay (6). They included the same procedures as Brummet et al. and found a risk for PPOU of 3.5% based on both the definition used by Clarke et al. and on that used by Brummet et al. Although comparability is still limited, our results indicate lower rates of PPOU in Germany than in Canada and the USA.
Interpretation of effects of individual pre-existing risk factors
Our analysis found some of the previously reported risk factors to be significant predictors of PPOU, based on the adjusted effects in the logistic regression model, while other risk factors were not significant. There are several reasons for this. First, our aim was to assess the pre-existing risk from patient history as broadly as possible. For this reason, we included all of the risk factors which have been reported in several systematic reviews on the topic and were measurable based on claims data (7, 19, 20, 22, 23, e3). This means that some of the risk factors included had an overlap and therefore substantial correlation with other risk factors. This results in a so-called non-parsimonious model, which aims at explaining as much of the overall variance in outcome from covariates as possible (optimal goodness of fit). The adjusted effect only represents the unique effect of a single covariate, not showing the overlap with the effect of any other covariate in the model. Including correlated risk factors in a regression model can mean that risk factors that have a significant univariate effect no longer have a significant effect in the regression model, which adjusts the effect for all other risk factors. This leads to results that may sometimes seem counterintuitive to clinicians. An example of this is presented by the effects of depression and the prescription of antidepressants. For both, there were strong descriptive differences between the group without PPOU and the group with PPOU (22.5% vs. 34.4% for depression and 14.7% vs. 37.5% for antidepressants, Table 1). Both of these differences were also significant (both p < 0.001, p-values not shown in Table 1). When both were included in the same regression model, the adjusted effect of depression was no longer significant (Table 2, p = 1). Patients with depression have a high chance of also being prescribed antidepressants, so we have overlapping effects in the regression model. The same will be true for other risk factors (e.g., patients with depression have a higher chance of having chronic pain). The p-value of 1 results not only from the described consequence of overlapping effects of correlated predictors, but also from the correction for multiple testing, which we applied due to the large sample size.
Variables
Our primary outcome was PPOU, defined as prescriptions of opioids during days 1–90 and days 91–180 after discharge (8). To define type of surgery in the index stay, we adapted a previously reported classification of surgical procedures for use with claims data, which resulted in 103 distinct surgical groups (14). Pre-existing risk factors were defined based on claims data from the 12 months prior to the index stay. Details are presented in the eMethods.
Statistical analysis
The incidence of PPOU was calculated for the complete sample and for each surgical group with 95% Wilson Score confidence interval (CI). Pre-existing risk factors were compared between cases with and without PPOU. To assess the importance of type of surgery in comparison with pre-existing risk factors for the risk of PPOU, we calculated a random intercept-only logistic regression model including surgical group as random intercept and risk factors as fixed effects. The total contribution of patient history risk factors and the total contribution of type of surgery to prediction of PPOU were estimated based on variance components and residual intraclass correlation (ICC) (15). To assess whether adjustment for pre-existing risk factors causes large changes to the ranking of the surgical procedures with regard to risk of PPOU, we calculated the random intercepts per surgical group both from the model including risk factors and from the model not including risk factors and calculated the Pearson correlation between the two (see eMethods). All statistical analyses were performed using the statistical freeware R (16). Tests were conducted on significance level 0.05.
Results
Study cohort
A total of 203 327 opioid-naive patients without cancer were included in at least one of the 103 defined surgery groups (Figure 1). Of these, 193 392 were allocated to only one surgical group, while 9935 were allocated to more than one group, which shows that the defined groups were largely distinct from each other. The baseline characteristics of the patients are shown in Table 1. Females were overrepresented in the sample (63.4%), reflecting the demographic structure of BARMER’s policy holders. Cocaine abuse occurred very rarely in the study sample (0.1%) and was therefore not included in the prediction model.
Figure 1.
Study flow chart
Table 1. Demographics and pre-existing comorbidity/medication of the study sample.
| Variable |
All
(n = 203 327) |
No PPOU
(n = 200,401) |
PPOU
(n = 2926) |
| Female sex | 63.4% (n = 128 843) | 63.3% (n = 126 804) | 69.7% (n = 2039) |
| Age (years) | 54.67 (18.48) | 54.45 (18.43) | 69.77 (14.93) |
| Number of prior hospital stays | 0.39 (0.83) | 0.38 (0.82) | 0.73 (1.18) |
| CCI (unweighted, points) | 0.92 (1.36) | 0.91 (1.35) | 1.93 (1.83) |
| Coagulopathy | 2.9% (n = 5836) | 2.8% (n = 5700) | 4.6% (n = 136) |
| Chronic pulmonary disease | 20.2% (n = 41 043) | 20.1% (n = 40 271) | 26.4% (n = 772) |
| Cardiovascular disease | 50.6% (n = 102 852) | 50.2% (n = 100 552) | 78.6% (n = 2300) |
| Diabetes | 13.9% (n = 28 330) | 13.7% (n = 27 485) | 28.9% (n = 845) |
| Renal disease | 6.7% (n = 13 708) | 6.6% (n = 13 185) | 17.9% (n = 523) |
| Depression | 22.7% (n = 46 176) | 22.5% (n = 45 169) | 34.4% (n = 1007) |
| Anxiety | 7.3% (n = 14 806) | 7.3% (n = 14 549) | 8.8% (n = 257) |
| Bipolar disorder | 0.4% (n = 783) | 0.4% (n = 764) | 0.6% (n = 19) |
| Other mood disorder | 0.3% (n = 682) | 0.3% (n = 666) | 0.5% (n = 16) |
| Posttraumatic stress disorder | 6.4% (n = 12 988) | 6.4% (n = 12 822) | 5.7% (n = 166) |
| Alcohol abuse | 2.5% (n = 5095) | 2.5% (n = 4923) | 5.9% (n = 172) |
| Tobacco abuse | 8% (n = 16 302) | 8% (n = 16 003) | 10.2% (n = 299) |
| Cocaine abuse* | 0.1% (n = 119) | 0.1% (n = 117) | 0.1% (n = 2) |
| Other substance abuse | 1% (n = 2062) | 1% (n = 2003) | 2% (n = 59) |
| Other psychiatric comorbidities | 11.6% (n = 23 595) | 11.6% (n = 23 165) | 14.7% (n = 430) |
| Fibromyalgia | 0.9% (n = 1846) | 0.9% (n = 1783) | 2.2% (n = 63) |
| Migraine | 6.7% (n = 13 677) | 6.7% (n = 13 499) | 6.1% (n = 178) |
| Chronic pain | 12.2% (n = 24 730) | 11.9% (n = 23 893) | 28.6% (n = 837) |
| Prescription of non-opioid analgesics | 48.7% (n = 98 920) | 48.3% (n = 96 778) | 73.2% (n = 2142) |
| Prescription of benzodiazepines | 1.6% (n = 3339) | 1.6% (n = 3216) | 4.2% (n = 123) |
| Prescription of antidepressants | 15% (n = 30 546) | 14.7% (n = 29 450) | 37.5% (n = 1096) |
* Cocaine was not used as a predictor of PPOU in the two-level logistic regression, because its incidence was too low.
Descriptive statistics are given as percentage (N) or mean (SD). CCI, Charlson Comorbidity Index; PPOU, Prolonged postoperative opioid use
Incidence of prolonged postoperative opioid use
The overall incidence of cases with PPOU was 1.4% (95% CI [1.4; 1.5]). Presence of any single pre-existing risk factor was associated with only modest increases in risk (e.g. from 1.06% without to 3.59% with previous prescription of antidepressants, eTable 4). Figure 2 presents PPOU for the 33 surgery groups with n > 50 cases and risk of PPOU = 1% (results for all surgery groups are presented in eTable 5). In general, we found that orthopedic and trauma surgery and vascular surgery were associated with higher risk of PPOU. The highest incidence was found in patients with major amputation (transfemoral amputation: 21.7% [14.5; 31.2], transtibial amputation: 15.3% [8.8; 25.3]) followed by orthopedic interventions (partial shoulder joint replacement: 8.0% [4.0; 15.7], spine surgery: 6.7% [6.2; 7.2], revision of knee joint replacement: 5.3% [3.8; 7.3]). Toe amputation was associated with a risk of 4.1% [3.1; 5.4]. In a post-hoc analysis, we found that the majority of amputations were due to ischemia (81.7%), while only 1.4% were due to trauma (see eMethods, eResults, eTable 6).
eTable 4. Risk of PPOU according to presence of risk factors.
| Risk of PPOU (%) | |||
| Risk factor | Risk factor not present | Risk factor present | Difference |
| Sex (male vs. female) | 1.19% | 1.58% | 0.39% |
| Coagulopathy (no vs. yes) | 1.41% | 2.33% | 0.92% |
| Chronic pulmonary disease (no vs. yes) | 1.33% | 1.88% | 0.55% |
| Cardiovascular disease (no vs. yes) | 0.62% | 2.24% | 1.61% |
| Diabetes (no vs. yes) | 1.19% | 2.98% | 1.79% |
| Renal disease (no vs. yes) | 1.27% | 3.82% | 2.55% |
| Depression (no vs. yes) | 1.22% | 2.18% | 0.96% |
| Anxiety (no vs. yes) | 1.42% | 1.74% | 0.32% |
| Bipolar disorder (no vs. yes) | 1.44% | 2.43% | 0.99% |
| Other mood disorder (no vs. yes) | 1.44% | 2.35% | 0.91% |
| Posttraumatic stress disorder (no vs. yes) | 1.45% | 1.28% | -0.17% |
| Alcohol abuse (no vs. yes) | 1.39% | 3.38% | 1.99% |
| Tobacco abuse (no vs. yes) | 1.40% | 1.83% | 0.43% |
| Cocaine abuse (no vs. yes) | 1.44% | 1.68% | 0.24% |
| Other substance abuse (no vs. yes) | 1.42% | 2.86% | 1.44% |
| Other psychiatric comorbidities (no vs. yes) | 1.39% | 1.82% | 0.43% |
| Fibromyalgia (no vs. yes) | 1.42% | 3.41% | 1.99% |
| Migraine (no vs. yes) | 1.45% | 1.30% | -0.15% |
| Chronic pain (no vs. yes) | 1.17% | 3.38% | 2.21% |
| Prescription of non-opioid analgesics (no vs. yes) | 0.75% | 2.17% | 1.41% |
| Prescription of benzodiazepines (no vs. yes) | 1.40% | 3.68% | 2.28% |
| Prescription of antidepressants (no vs. yes) | 1.06% | 3.59% | 2.53% |
PPOU, Prolonged postoperative opioid use; vs., versus
Figure 2.
Incidence of PPOU with 95% confidence intervals (CI) by surgical group. Presented are groups with n > 50 cases and incidence of PPOU of at least 1% (see eTable 5 for presentation of all surgical groups). The gray dotted vertical line presents the incidences of PPOU in the overall sample (“overall PPOU”) of 1.44%. The color coding shows if the CI of the PPOU in the respective surgery group overlaps with the overall PPOU, or if the lower confidence limit is above the overall PPOU, which indicates a significant difference. PPOU, Prolonged postoperative opioid use
eTable 6. Subgroups of amputations according to etiology.
| Subgroup | Proportion |
| All amputations | |
| Ischemic | 81.7% (n=1254) |
| Traumatic | 1.4% (n=21) |
| Ischemic and traumatic | 1.2% (n=19) |
| Neither ischemic nor traumatic | 15.6% (n=240) |
| Transfemoral amputation | |
| Ischemic | 89.1% (n=82) |
| Traumatic | 1.1% (n=1) |
| Ischemic and traumatic | 0% (n=0) |
| Neither ischemic nor traumatic | 9.8% (n=9) |
| Transtibial amputation | |
| Ischemic | 87.5% (n=63) |
| Traumatic | 1.4% (n=1) |
| Ischemic and traumatic | 0% (n=0) |
| Neither ischemic nor traumatic | 11.1% (n=8) |
| Amputation of toe | |
| Ischemic | 79% (n=889) |
| Traumatic | 1.2% (n=13) |
| Ischemic and traumatic | 1.2% (n=14) |
| Neither ischemic nor traumatic | 18.6% (n=209) |
Presented are the absolute and relative frequencies for the etiology of amputations.
Influence of type of surgery and pre-existing risk factors on prolonged postsurgical opioid use
Table 2 gives the results of the random-intercept logistic regression model with pre-existing risk factors. Explained variance is a measure of what proportion of the risk of PPOU is predicted by the statistical model. Overall, the model explained 36.65% of the variance of risk of PPOU. Patient history factors explained 14.33%, while the remaining 22.32% was explained by the surgical groups (statistical details given in eTable 7). This indicates that type of surgery might be more important than the considered risk factors for development of PPOU. The ranking of surgical procedures in risk of PPOU was not changed to a relevant extent by controlling for differences in pre-existing risk factors (r = 0.97 between random intercepts from the model not including risk factors and the adjusted random intercepts from the model including risk factors). Prescription of antidepressants, abuse of alcohol, prescription of non-opioid analgesics, and a chronic pain diagnosis were the strongest predictors of PPOU among pre-existing risk factors when controlling for type of surgery and all other risk factors in model (Table 2).
Table 2. Predictors of prolonged postoperative opioid use.
| Predictors |
Odds Ratio
[95% CI] |
p-value* |
| (Intercept) | 0.005 [0.004; 0.006] | <0.001 |
| Female sex | 1.168 [1.070; 1.276] | 0.009 |
| Age (years) | 1.030 [1.027; 1.034] | <0.001 |
| Number of prior hospital stays | 1.117 [1.080; 1.156] | <0.001 |
| CCI (unweighted, points) | 1.103 [1.064; 1.144] | <0.001 |
| Coagulopathy | 1.086 [0.904; 1.304] | 1.000 |
| Chronic pulmonary disease | 0.978 [0.889; 1.076] | 1.000 |
| Cardiovascular disease | 1.030 [0.927; 1.145] | 1.000 |
| Diabetes | 1.047 [0.939; 1.167] | 1.000 |
| Renal disease | 1.076 [0.951; 1.216] | 1.000 |
| Depression | 0.984 [0.892; 1.085] | 1.000 |
| Anxiety | 0.931 [0.809; 1.071] | 1.000 |
| Bipolar disorder | 0.970 [0.604; 1.558] | 1.000 |
| Other mood disorder | 0.955 [0.569; 1.601] | 1.000 |
| Posttraumatic stress disorder | 0.964 [0.817; 1.137] | 1.000 |
| Alcohol abuse | 1.515 [1.277; 1.797] | <0.001 |
| Tobacco abuse | 1.249 [1.095; 1.425] | 0.015 |
| Other substance abuse | 1.154 [0.871; 1.529] | 1.000 |
| Other psychiatric comorbidity | 1.123 [1.001; 1.259] | 0.720 |
| Fibromyalgia | 1.177 [0.902; 1.536] | 1.000 |
| Migraine | 1.096 [0.935; 1.285] | 1.000 |
| Chronic pain | 1.363 [1.248; 1.489] | <0.001 |
| Prescription of non-opioid analgesics | 1.479 [1.357; 1.613] | <0.001 |
| Prescription of benzodiazepines | 1.141 [0.940; 1.383] | 1.000 |
| Prescription of antidepressants | 2.131 [1.943; 2.336] | <0.001 |
Presented are fixed effects (i.e., odds ratios) of the random-intercept logistic regression model predicting prolonged postoperative opioid use from types of surgery (random intercept) and pre-existing risk factors. Random effects are reported in eTable 7.
* Because of the large sample size, p-values of the regression coefficients were adjusted for multiple comparisons according to Holm (31).
CCI, Charlson Comorbidity Index
eTable 7. Subgroups of amputations according to etiology.
| Random effects | Value |
| Tsurgical groups*1 | 1.159 |
| Residual ICC*2 | 0.261 |
| n surgical groups | 104 |
| n patients | 203 327 |
| Marginal R2 / conditional R2*3 | 0.1433 / 0.3665 |
Presented are random effects (i. e. variances and intraclass correlations [ICC]) from the two-level regression model predicting prolonged postoperative opioid use (PPOU) from types of surgery (random intercept) and pre-existing risk factors. Of relevance for interpretation is the row which shows the importance of pre-existing risk factors and surgical group in predicting PPOU. Conditional R2 of 0.3665 (or 36.65%) represents the overall amount of variance in PPOU explained by both pre-existing risk factors and surgical group. Marginal R2 of 0.1433 (or 14.33%) gives the amount of variance in PPOU explained by pre-existing risk factors. Finally, the difference of 0.2232 (or 22.32%) is the amount of variance in PPOU explained by the surgical group which is not explained by pre-existing risk factors. This result shows that surgical group remains important for predicting the risk of PPOU even after controlling for many pre-existing risk factors, and that the type of surgery is more important for risk of PPOU than the measured pre-existing risk factors taken together.
*1 Tsurgical groups is the variance of the random intercept, i.e., the between-group variance of the surgical groups.
*2 Residual ICC is a standardized measure of the heterogeneity of PPOU among surgical groups under statistical control of the patient-level risk factors.
*3 Marginal R2 is the proportion of explained variance by the fixed part η only (i.e., by pre-existing risk factors). Conditional R2 is the proportion of explained variance by the fixed part η and the random part (i.e., including the unique variance explained by the surgical groups Tsurgical groups).
Discussion
We conducted a retrospective cohort study using German administrative claims data to investigate the incidence of PPOU across a wide range of surgical procedures. In a sample of opioid-naive patients without cancer undergoing inpatient surgery, the overall incidence of PPOU was 1.4%. There was wide variation among the surgical groups studied, with the risk of PPOU being more than ten times higher for procedures identified as high-risk. Special attention should be paid to amputations and spine surgery. The importance of the type of surgery is underlined by the finding that it explained 22.3% of the variance in PPOU, while only 14.3% was explained by pre-existing risk factors.
Only a small number of studies on PPOU have been conducted outside the USA and Canada (7, 17), and only one single-center study has been performed in Germany (12). Due to large differences in the operations included, samples, and definitions of PPOU, two recent meta-analyses reached no definitive conclusions regarding country-specific risks (7, 17). Compared with a risk of 1.4% in our sample, population-based mixed-procedure studies from Canada and the USA found substantially higher incidence of PPOU, between 3.5% and 6.5%, using definitions similar to that in our study (6, 8, 10). Therefore, our study provides some evidence that PPOU in general is less of a problem in Germany.
Previous studies investigated only individual surgical procedures or small sets of preselected interventions using very diverse samples and methods, which biases meta-analytic comparisons across studies. In their meta-analysis of 31 studies, Lawal et al. (7) found no difference between major and minor operations but also reported difficulty in making comparisons due to the methodological heterogeneity of the studies. Likewise, in their systematic review Kent et al. (5) were unable to identify any specific surgeries with an increased risk of PPOU and assumed that surgical characteristics may play only a minor role in assessment of risk. In contrast, based on more than 100 major and minor surgery groups, we found that 22.3% of the variance in the risk of PPOU was explained by the type of surgery. Although we included the significant risk factors identified in previous research (7, 18–20, 22, 23), these explained only the smaller part of the risk variation (14.3%). Therefore, type of surgery plays a major role in risk assessment for PPOU. We found that especially orthopedic and trauma surgery, vascular surgery, and minor and major amputation are associated with increased risk of PPOU. Regarding the association of individual pre-existing risk factors with PPOU, we replicated previous findings from international studies showing an increased risk for female sex, antidepressants, pain medication, chronic pain, alcohol or drug abuse, other psychological comorbidity, and total comorbidity (7, 18–22) (see eDiscussion for further details).
This study presents comprehensive data on the incidence of PPOU in Germany. One of its strengths is that it is based on a large national sample that is fairly representative of the adult population of Germany, although women are slightly overrepresented (24). It is also the first study worldwide to investigate the risk for PPOU across a broad spectrum of more than 100 surgical procedures. The analysis using a random-intercept logistic regression model allowed the effects of type of surgery on the development of PPOU to be distinguished from the effects of patient history.
However, the study also has some limitations. BARMER patients differ from the German population in that a slightly higher proportion are female and they are slightly younger (24). While this plays a minor role when comparing individual operations, it may bias the estimate of overall risk for PPOU. Claims data contain limited clinically relevant information. Opioid use could be assessed only on the basis of prescriptions, not actual consumption – a limitation shared by other international claims-based studies on PPOU (8–11). Claims data are designed for purposes of billing and health care administration; they do not fully reflect clinical reality and may be affected by information biases, particularly regarding ICD coding (25–27). Based on incorrect assignment of ICD codes, estimates of effects of some pre-existing risk factors may be biased. On the other hand, as both surgery coding and opioid prescribing are relevant for reimbursement, we consider both to be highly trustworthy. A major limitation shared with all previous studies is the lack of a common definition of PPOU, which makes it difficult to compare incidence across studies (5, 6). Therefore, we chose the definition that yielded the best international comparability (6–8, 10), although this does not allow differentiation between indicated PPOU and misuse. However, this will not affect the relative differences in PPOU between the surgical interventions.
Conclusions
At 1.4%, the overall incidence of PPOU in Germany seems to be relatively low compared with North America, although direct comparison is difficult because of differences in studied populations and definitions of PPOU. On the other hand, PPOU may pose a relevant problem for a subgroup of these patients, as well as for the German health care system. Especially patients undergoing any of the identified high-risk interventions may deserve more attention to avoidance of the side effects of long-term opioid use or misuse. All clinicians should be aware that initially indicated opioid use may later develop into non-indicated use. Patients with a surgery which has a high risk of PPOU, or with at least one of the identified risk factors should be closely monitored for pain and appropriate postoperative analgesic management, e.g., through post-discharge screening programs, pain specialists, and transitional pain services (28). The national and international guidelines offer helpful practical tools, with numerous recommendations for postoperative pain therapy and long-term opioid treatment (29, 30). Further research is needed to characterize the transition from perioperative to long-term opioid use and to explain the differences in the incidence of PPOU among health systems worldwide.
Footnotes
Funding
The LOPSTER project on which this publication is based was funded by the Innovation Committee of the German Federal Joint Committee (G-BA), Berlin, Germany under grant number 01VSF19019.
Conflict of interest statement
WM has received research grants for his institution from the European Commission, Gemeinsamer Bundesausschuss (G-BA), Medtronic, Pfizer, Mundipharma, Grünenthal, Vertanical and personal honoraria from Merck, Sanofi, MSD, Tafalgie, Kyowa, Mundipharma, Grünenthal and Ethypharm. T.V. reports research grants from Sedana medical, Ratiopharm, Saarland, Pfizer, Infecto Pharm, and Cyto Sorbents Europe GmbH (all paid to the institution) and personal lecture fees from Pajunk and CSL Behring. TV is past president of The European Society of Regional Anaesthesia & Pain Therapy. HLR receives research funding from the DFG, the BMBF and the G-BA. She was honored for lectures by Grünenthal and received consulting fees from Orion.
The remaining authors declare that no conflict of interest exists.
References
- 1.Powell D. Educational attainment and US drug overdose deaths. JAMA Health Forum. 2023;4 doi: 10.1001/jamahealthforum.2023.3274. e233274. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Kaafarani HMA, Han K, El Moheb M, et al. Opioids after surgery in the United States versus the rest of the world: the international patterns of opioid prescribing (iPOP) multicenter study. Ann Surg. 2020;272:879–886. doi: 10.1097/SLA.0000000000004225. [DOI] [PubMed] [Google Scholar]
- 3.Fiore JF, Jr, El-Kefraoui C, Chay MA, et al. Opioid versus opioid-free analgesia after surgical discharge: a systematic review and meta-analysis of randomised trials. Lancet. 2022;399:2280–2293. doi: 10.1016/S0140-6736(22)00582-7. [DOI] [PubMed] [Google Scholar]
- 4.Shadbolt C, Abbott JH, Camacho X, et al. The surgeon‘s role in the opioid crisis: a narrative review and call to action. Front Surg. 2020;7:4. doi: 10.3389/fsurg.2020.00004. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Kent ML, Hurley RW, Oderda GM, et al. American society for enhanced recovery and perioperative quality initiative-4 joint consensus statement on persistent postoperative opioid use: definition, incidence, risk factors, and health care system initiatives. Anesth Analg. 2019;129:543–552. doi: 10.1213/ANE.0000000000003941. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Jivraj NK, Raghavji F, Bethell J, et al. Persistent postoperative opioid use: a systematic literature search of definitions and population-based cohort study. Anesthesiology. 2020;132:1528–1539. doi: 10.1097/ALN.0000000000003265. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Lawal OD, Gold J, Murthy A, et al. Rate and risk factors associated with prolonged opioid use after surgery: a systematic review and meta-analysis. JAMA Netw Open. 2020;3 doi: 10.1001/jamanetworkopen.2020.7367. e207367. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Clarke H, Soneji N, Ko DT, Yun L, Wijeysundera DN. Rates and risk factors for prolonged opioid use after major surgery: population based cohort study. BMJ. 2014;348 doi: 10.1136/bmj.g1251. g1251. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Sun EC, Darnall BD, Baker LC, Mackey S. Incidence of and risk factors for chronic opioid use among opioid-naive patients in the postoperative period. JAMA Intern Med. 2016;176:1286–1293. doi: 10.1001/jamainternmed.2016.3298. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Brummett CM, Waljee JF, Goesling J, et al. New persistent opioid use after minor and major surgical procedures in US adults. JAMA Surg. 2017;152 doi: 10.1001/jamasurg.2017.0504. e170504. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Lindeberg FCB, Bell M, Larsson E, Hallqvist L. New prolonged opioid consumption after major surgery in Sweden: a population-based retrospective cohort study. BMJ Open. 2023;13 doi: 10.1136/bmjopen-2022-071135. e071135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Laufenberg-Feldmann R, Kappis B, Mauff S, Schmidtmann I, Ferner M. Prevalence of pain 6 months after surgery: a prospective observational study. BMC Anesthesiol. 2016;16:91. doi: 10.1186/s12871-016-0261-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Benchimol EI, Smeeth L, Guttmann A, et al. The REporting of studies conducted using Observational Routinely-collected health Data (RECORD) Statement. PLoS Med. 2015;12 doi: 10.1371/journal.pmed.1001885. e1001885. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Gerbershagen HJ, Aduckathil S, van Wijck AJ, Peelen LM, Kalkman CJ, Meissner W. Pain intensity on the first day after surgery: a prospective cohort study comparing 179 surgical procedures. Anesthesiology. 2013;118:934–944. doi: 10.1097/ALN.0b013e31828866b3. [DOI] [PubMed] [Google Scholar]
- 15.Snijders TAB, Bosker RJ. 2nd edition. London: SAGE; 2012. Multilevel analysis: an introduction to basic and advanced multilevel modeling. [Google Scholar]
- 16.R Core Team R. Vienna, Austria: R Foundation for Statistical Computing; 2021. A language and environment for statistical computing. [Google Scholar]
- 17.Sitter T, Forget P. Persistent postoperative opioid use in Europe: a systematic review. Eur J Anaesthesiol. 2021;38:505–511. doi: 10.1097/EJA.0000000000001346. [DOI] [PubMed] [Google Scholar]
- 18.Jann M, Kennedy WK, Lopez G. Benzodiazepines: a major component in unintentional prescription drug overdoses with opioid analgesics. J Pharm Pract. 2014;27:5–16. doi: 10.1177/0897190013515001. [DOI] [PubMed] [Google Scholar]
- 19.Mohamadi A, Chan JJ, Lian J, et al. Risk factors and pooled rate of prolonged opioid use following trauma or surgery: a systematic review and meta-(regression) analysis. J Bone Joint Surg Am. 2018;100:1332–1340. doi: 10.2106/JBJS.17.01239. [DOI] [PubMed] [Google Scholar]
- 20.Pagé MG, Kudrina I, Zomahoun HTV, et al. A systematic review of the relative frequency and risk factors for prolonged opioid prescription following surgery and trauma among adults. Ann Surg. 2020;271:845–854. doi: 10.1097/SLA.0000000000003403. [DOI] [PubMed] [Google Scholar]
- 21.Adeola JO, Urman RD. Psychiatric comorbidities associated with persistent postoperative opioid use. Curr Pain Headache Rep. 2022;26:701–708. doi: 10.1007/s11916-022-01073-3. [DOI] [PubMed] [Google Scholar]
- 22.Liu Z, Karamesinis AD, Plummer M, et al. Epidemiology of persistent postoperative opioid use after cardiac surgery: a systematic review and meta-analysis. Br J Anaesth. 2022;129:366–377. doi: 10.1016/j.bja.2022.05.026. [DOI] [PubMed] [Google Scholar]
- 23.Mohan S, Lynch CP, Cha EDK, et al. Baseline risk factors for prolonged opioid use following spine surgery: systematic review and meta-analysis. World Neurosurg. 2022;159:179–188. doi: 10.1016/j.wneu.2021.12.086. e2. [DOI] [PubMed] [Google Scholar]
- 24.Marschall U, L‘hoest H, Radbruch L, Häuser W. Long-term opioid therapy for chronic non-cancer pain in Germany. Eur J Pain. 2016;20:767–776. doi: 10.1002/ejp.802. [DOI] [PubMed] [Google Scholar]
- 25.Benchimol EI, Manuel DG, To T, Griffiths AM, Rabeneck L, Guttmann A. Development and use of reporting guidelines for assessing the quality of validation studies of health administrative data. J Clin Epidemiol. 2011;64:821–829. doi: 10.1016/j.jclinepi.2010.10.006. [DOI] [PubMed] [Google Scholar]
- 26.Berger K. Obtaining good health-care data: the desire and the reality. Dtsch Arztebl Int. 2024;121:139–140. doi: 10.3238/arztebl.m2024.0031. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Vogelgesang F, Thamm R, Frerk T, et al. The agreement between diagnoses as stated by patients and those contained in routine health insurance data—results of a data linkage study. Dtsch Arztebl Int. 2024;121:141–147. doi: 10.3238/arztebl.m2023.0250. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Mikhaeil J, Ayoo K, Clarke H, Wąsowicz M, Huang A. Review of the transitional pain service as a method of postoperative opioid weaning and a service aimed at minimizing the risk of chronic post-surgical pain. Anaesthesiol Intensive Ther. 2020;52:148–153. doi: 10.5114/ait.2020.96018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Häuser W, Bock F, Hüppe M, et al. Empfehlungen der zweiten Aktualisierung der Leitlinie LONTS. Der Schmerz. 2020;34:204–244. doi: 10.1007/s00482-020-00472-y. [DOI] [PubMed] [Google Scholar]
- 30.Pogatzki-Zahn E, Meißner W. Behandlung akuter perioperativer und posttraumatischer Schmerzen. AWMF S3-Leitlinie. 2021;1:025. [Google Scholar]
- 31.Holm S. A simple sequentially rejective multiple test procedure. Scand J Statist. 1979;6:65–70. [Google Scholar]
- E1.Lee C, Ye M, Weaver O, et al. Defining opioid naïve and implications for monitoring opioid use: a population-based study in Alberta, Canada. Pharmacoepidemiol Drug Saf. 2024;33 doi: 10.1002/pds.5693. e5693. [DOI] [PubMed] [Google Scholar]
- E2.Nielsen S, Gisev N, Bruno R, et al. Defined daily doses (DDD) do not accurately reflect opioid doses used in contemporary chronic pain treatment. Pharmacoepidemiol Drug Saf. 2017;26:587–591. doi: 10.1002/pds.4168. [DOI] [PubMed] [Google Scholar]
- E3.Quan HD, Sundararajan V, Halfon P, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care. 2005;43:1130–1139. doi: 10.1097/01.mlr.0000182534.19832.83. [DOI] [PubMed] [Google Scholar]