Predictors of persistent postoperative opioid use following colectomy: a population‐based cohort study from England

Summary This retrospective cohort study on adults undergoing colectomy from 2010 to 2019 used linked primary (Clinical Practice Research Datalink), and secondary (Hospital Episode Statistics) care data to determine the prevalence of persistent postoperative opioid use following colectomy, stratified by pre‐admission opioid exposure, and identify associated predictors. Based on pre‐admission opioid exposure, patients were categorised as opioid‐naïve, currently exposed (opioid prescription 0–6 months before admission) and previously exposed (opioid prescription within 7–12 months before admission). Persistent postoperative opioid use was defined as requiring an opioid prescription within 90 days of discharge, along with one or more opioid prescriptions 91–180 days after hospital discharge. Multivariable logistic regression analyses were conducted to obtain odds ratios for predictors of persistent postoperative opioid use. Among the 93,262 patients, 15,081 (16.2%) were issued at least one opioid prescription within 90 days of discharge. Of these, 6791 (45.0%) were opioid‐naïve, 7528 (49.9%) were currently exposed and 762 (5.0%) were previously exposed. From the whole cohort, 7540 (8.1%) developed persistent postoperative opioid use. Patients with pre‐operative opioid exposure had the highest persistent use: 5317 (40.4%) from the currently exposed group; 305 (9.8%) from the previously exposed group; and 1918 (2.5%) from the opioid‐naïve group. The odds of developing persistent opioid use were higher among individuals who used long‐acting opioid formulations in the 180 days before colectomy than those who used short‐acting formulations (odds ratio 3.41 (95%CI 3.07–3.77)). Predictors of persistent opioid use included: previous opioid exposure; high deprivation index; multiple comorbidities; use of long‐acting opioids; white race; and open surgery. Minimally invasive surgical approaches were associated with lower odds of persistent opioid use and may represent a modifiable risk factor.


Introduction
Colectomy is a common abdominal surgical procedure, with 300,000 performed annually in the USA [1] and approximately 33,000 performed annually in England [2].
People undergoing colectomy might have diseases that may be associated with pain, such as inflammatory bowel disease, diverticulitis and cancer [3]. Additionally, the procedure itself can lead to significant postoperative pain [3] and opioid analgesia may be indicated.
While short-term opioid use has an established role in managing acute pain [4], it has recently been identified as a risk factor for persistent postoperative opioid use (PPOU) [5][6][7], beyond the expected time frame for complete recovery [8]. Persistent postoperative opioid use is now widely acknowledged as a surgical complication [9], which can be associated with harm, including physical dependence, tolerance and opioid diversion [7,10,11].
Therefore, opioid prescriptions for surgical pain have been recognised as a public health concern and one of the factors implicated in the opioid epidemic in the USA [12].
Accordingly, the UK's Medicines and Healthcare products Regulatory Agency has released recommendations to mitigate the risk of opioid addiction and recommended against extending opioid use for longer than 3 months in the management of acute pain [13]. Hence, it has become a significant focus for opioid-related policy and interventions [14][15][16].
Minor and major surgical procedures are associated with development of PPOU [14,[17][18][19], but there is wide variability around its definition [9]. Studies have used timeto-opioid cessation [20] or presence of repeat prescriptions [21] to define PPOU. According to several studies from the USA, which defined PPOU as having one opioid prescription within the early post-discharge period and another prescription 91-180 days after discharge, 11-17% of opioid na€ ıve patients develop PPOU following colectomy [3,22]. The prevalence of PPOU increases to > 30% for patients previously exposed to opioids [9,19,23], and this might be linked with poor surgical outcomes [6,24] and higher healthcare costs [25].
Despite the risk of PPOU following colectomy being quantified in the USA and Canada, the external validity of these findings is limited and cannot be extrapolated to other populations due to significant variations in prescribing practices. Hence, the extent to which PPOU exists within a subset population from the UK has been hitherto unexplored. We sought to determine the prevalence of PPOU following colectomy, stratified by preadmission opioid exposure and identify associated PPOU predictors using linked electronic healthcare data from England. We hypothesised that the prevalence of PPOU would vary based on the individual's opioid exposure before surgery and that several predictors for PPOU could be identified.

Methods
The study was approved by the Independent Scientific Advisory Committee and performed and reported in accordance with the strengthening the reporting of observational studies in epidemiology (STROBE) guidelines [26].
This study used linked primary and secondary care electronic databases previously described and validated [27,28] included the anal canal and rectum were excluded (online Supporting Information Appendix S1). Patients who did not survive the first 90 days following discharge were excluded.
Eligible patients were followed up from the day of discharge to either having the study outcome of PPOU [22], end of follow up (180 days), transfer out of participating practice or date of death, whichever came first (Fig. 1, Online Supporting Information Appendix S1). To have sufficient data on pre-operative opioid exposure, patients were excluded if they did not have a minimum of 12 months of Aurum data before the admission date for surgery.
Baseline characteristics, such as age, sex and race, were obtained from Aurum and Hospital Episode Statistics data.
Race was categorised as white, black, Asian and others [31].
Comorbidities before admission were obtained from Aurum and Hospital Episode Statistics data and classified using the Charlson comorbidity index based on the number of comorbidities into 0, 1 and ≥ 2 [32]. Index of Multiple Deprivation scores [30] were categorised into quintiles from 1 to 5 (least to most deprived, respectively) [27].
A lookback window for 1 year before the date of admission was used to evaluate pre-operative opioid exposure. Patients were considered opioid-na€ ıve if they did not have an opioid prescription issued in the year preceding their date of admission for surgery. They were considered currently exposed´if they were issued an opioid prescription within the 6 months before their admission date and`previously exposed´if an opioid prescription was issued within 7-12 months before their date of admission, thus forming two mutually exclusive pre-operative opioid exposed groups [33]. We categorised prescriptions as long-  Univariable and multivariable logistic regression analyses were used to examine the association of different predictors with the odds of PPOU. The analyses were stratified by pre-operative opioid exposure as opioid-na€ ıve, currently exposed and previously exposed. This decision was made based on additional analysis investigating interactions between pre-operative opioid exposure and surgical approach. The likelihood ratio test was used to check for interaction and compare coefficients between the models. Further stratification of the opioid-na€ ıve group by admission type was performed after detecting significant interaction between admission type and cancer-related surgery. However, when tested on the currently and previously opioid-exposed groups, this interaction was not significant. Age was fitted as a continuous variable; this decision was made by conducting separate models with age fitted as either a continuous or categorical variable.
Then the likelihood ratio test was used to compare model fit in both models, and the variable with the best fit was selected for the final model.
We also analysed potential predictor variables,  (49.9%) were currently exposed and 762 (5.0%) were previously exposed. Among each category of pre-operative opioid exposure in the overall colectomy cohort, 6791  5317 (40.4%) in the currently exposed group; 305 (9.8%) in the previously exposed group; and 1918 (2.5%) in the opioid-na€ ıve group ( were only linked to higher odds of PPOU in the emergency setting (Table 3).
For the currently exposed group, pre-operative use of In the previously exposed group, having two or more comorbidities was the only predictor associated with higher odds of PPOU. Compared with open

Discussion
This nationwide study in patients undergoing colectomy in where only 11% of patients were given post-discharge opioids [38], which is more consistent with our findings.
Half of the patients in our cohort who were discharged from the hospital with a prescription for opioids (8.1% of the overall cohort) and continued to be prescribed opioids for up to 180 days following discharge. This overall finding was lower than the 10% prevalence reported in a prospective study from the USA [3] and the figures determined by a USA database analysis showing PPOU rates ranging between 13.5% and 21.2% following colectomy [39]. Furthermore, among opioid-na€ ıve patients, 2.5% developed PPOU. This finding aligns with that reported by Clarke et al. [17], who used the same definition of PPOU in a study that included different types of abdominopelvic procedures and was not strictly limited to colectomy.
We found that patients with pre-operative opioid exposure accounted for the majority of persistent users. This result is similar to that of previous studies showing that PPOU is more common in patients with a history of opioid exposure before surgery [3,24,40], although pre-operative opioid exposure is not defined consistently in terms of dose, recency, duration and continuity of use. While the definition adopted in the present study did not require evidence of long-term opioid use before surgery, a large proportion of patients in this group continued to use opioids for more than 90 days following discharge.
The odds of persistent opioid use were more than three times higher among individuals who used long-acting opioid formulations in the 180 days before colectomy than those who used short-acting formulations [6,20,41].
This finding contributes to the growing body of evidence suggesting that long-acting and modified-release formulations are a modifiable risk factor for PPOU [42].
The association between PPOU and pre-operative opioid exposure is likely to be multifactorial. One possible explanation is that patients with previous opioid exposure can develop tolerance or hyperalgesia, which may make the management of their postoperative pain more challenging and lead to persistent use [43,44]. Another possible explanation is that patients who were taking opioids preoperatively had already adjusted to opioid-related adverse effects such as nausea, vomiting and constipation, while these may have discouraged their opioid-na€ ıve counterparts from continuing their opioids. In addition to the currently exposed group, we also included patients with previous opioid exposure. This is a distinct group with a potentially different trajectory of PPOU that is often overlooked. We found that despite their remote exposure to opioids before surgery, these patients were still at greater risk of PPOU than those in the na€ ıve group. Although we did not find an association between the use of long-acting opioid formulations and PPOU in this group of patients, it is  [50]. Further prospective studies are needed to assess the possible benefits of minimally invasive approaches on PPOU in specific surgical populations and pre-operative opioid use groups.
Having two or more comorbidities increased the odds of PPOU among all groups, while those in the most deprived quintiles had increased odds of PPOU in opioid-na€ ıve patients and current users. These results align with previous studies that have evaluated these factors in major abdominal surgical procedures [17,22].
Variation was observed between racial groups. Opioid-na€ ıve patients of black race had significantly lower odds of developing PPOU, when compared with patients of white race, while current opioid users of white race were at higher risk of becoming persistent users compared with all other races. Previous research has identified racial disparities in pain diagnosis and treatment [51], and white patients are more likely to be prescribed opioids than black patients [52].
In light of this evidence, we have to consider that the present study's findings may have been confounded by clinicians' implicit bias in the assessment of pain severity and choice of treatment, implicit bias related to repeat opioid prescriptions [53], hospital-level factors and surgical setting.
Over the 10-year study period, there were several changes to clinical practice that may have impacted the prescription of opioids and the incidence of PPOU. These changes include: widespread implementation of enhanced recovery programmes [54]; increased use of multimodal and opioid-sparing analgesia [55]; regional and neuraxial anaesthesia; and increased uptake of minimally invasive surgery. Additionally, there has been an increased awareness of the potential problems associated with opioids, which may have led to more responsible prescribing and stewardship practices.
Our study has several limitations. First, although Aurum has longitudinal data on opioid prescription records before and after surgery, limiting the possibility of recall bias, it lacks clinical details such as in-hospital drug therapy, patientreported outcome measures, and some complications (such as persistent postsurgical pain). Moreover, the assessment of PPOU using electronic health record data is limited by the inability to measure whether drugs in prescriptions dispensed were subsequently taken by the patients. Nevertheless, despite these limitations, the use of prescription data as a proxy for confirmed drug consumption is widespread in drug utilisation research [19].
Additionally, data obtained from Hospital Episode Statistics lack information on hospital-level factors such as pre-operative preparation, use of regional anaesthesia and availability of enhanced recovery protocols. It is unknown whether opioid prescribing guidance and discharge opioid tapering instructions were available for patients. While Daliya et al. [35] previously acknowledged the lack of these resources within hospitals in England, implementing these services along with opioid stewardship programmes may be effective for minimising post-discharge opioid prescribing [56,57].
Another limitation is the lack of information on drugs prescribed privately or obtained via other sources. In addition, during the study period, some`weak´opioids, such as dihydrocodeine and codeine, were available without a prescription, which may have led to the underrepresentation of the prevalence of PPOU related to these opioids. We did not study all the risk factors for PPOU reported in the literature such as history of depression, anxiety and pre-operative benzodiazepine and antidepressant use [6,7]. The dose, duration and type of opioids used before surgery may also be associated with the development of PPOU [7], but these factors were not tested in the current analysis. Additionally, we could not control for factors affecting the choice of surgical approach or admission type. However, we identified several patientand surgery-specific predictors associated with long-term opioid use that had not been identified previously in a population from the UK.
After undergoing colectomy in hospitals across England, 8.1% of patients continued to receive opioid prescriptions more than 3 months after discharge.
Persistent postoperative opioid use was more common in patients with pre-operative opioid exposure. Importantly, a minimally invasive surgical approach was associated with lower odds of PPOU in opioid na€ ıve and previously exposed