Chronic Opioid Use Is Associated With Early Biologic Discontinuation in Inflammatory Bowel Disease

Christian Rhudy; Courtney L. Perry; Michael Singleton; Jeffery Talbert; Terrence A. Barrett

Disclosures

Aliment Pharmacol Ther. 2021;53(6):704-711. 

In This Article

Materials and Methods

Study Design and Population

This study followed a retrospective cohort design. The study population included individuals diagnosed with IBD who received their first biologic prescription during the years 2011–2016. Individuals with IBD were identified utilising the ICD-9 (555.X, 556.X) or ICD-10 (K50.X, K51.X) diagnosis codes for Crohn's disease and ulcerative colitis. The study index date was identified as the earliest date associated with either inpatient administration or outpatient prescription claims of a biologic agent of interest, including adalimumab, infliximab, certolizumab, golimumab, vedolizumab and ustekinumab. Inpatient administrations were identified by Healthcare Common Procedure Coding System J-codes, and outpatient prescription claims were identified by Generic Product Identifier codes.

At the index date, individuals were included if they were ≤65 years of age and fully enrolled with prescription drug benefits for at least 6 months prior to the index date and 2 years after the index date. Patients >65 were excluded from analysis due to incomplete data in the Truven MarketScan Commercial Database, which does not include Medicare claims data. Relevant covariates and outcomes of interest for each individual were extracted beginning 6 months prior to the index date and ending either at a gap in full enrolment or the study termination date (31 December 2016).

Individuals were then separated into cohorts based upon observed chronic opioid use throughout the entire study period. Outpatient opioid prescription claims were identified utilising GPI codes beginning with '65'. Chronic opioid use was defined as having at least a 90-day supply of outpatient opioid prescriptions in a 6-month period without any 30-day gaps in opioid coverage, a definition that has been utilised in previous studies.[5,14–18] Chronic opioid use was observed both in the 6-month pre-index period and the post-index study period.

Data Source

Insurance claims data utilised in this study were procured from the Truven MarketScan Commercial Claims and Encounters database. The Truven MarketScan Research Databases (www.truvenhealth.com) are administrative claims-based databases representing more than 255 million covered individuals and are considered representative of the privately insured population in the US. More than 2100 peer-reviewed articles have been published utilising MarketScan data.[19]

Outcome Measures

The primary outcome of interest was persistence to biologic therapy. Persistence days were defined as the period from the index date to the first 8-week gap in supply of any biologic of interest. An 8-week gap was chosen as it would be equivalent to missing a full dosing interval of the any of the biologics of interest. Individuals were deemed non-persistent if they had an 8-week gap in supply prior to the end of the study window or their eligible period of enrolment. Secondary outcomes of interest included the number of unique biologics utilised by individuals, and the days persistence to each biologic in chronological sequence. All persistence calculations were carried out in SAS software version 9.4[20] utilising previously validated methods.[21,22]

Demographic and Covariate Measurements

Demographic covariates of interest were collected, including sex, age, geographic region and enrolment dates. Diagnosis of relevant comorbid conditions and Charlson comorbidity index were evaluated and reported.[23] All diagnoses were identified utilising Clinical Classifications Software for ICD-9 and ICD-10 from the Healthcare Cost and Utilization Project (HCUP), as well as previous literature.[24,25] Diagnoses were coded as binary variables for individuals with at least one diagnosis code during the 6-month pre-index period or the post-index study period.

Patterns of healthcare utilisation were also examined, including hospital admissions, emergency department visits, abdominal surgeries, endoscopies, the utilisation of inpatient and outpatient steroids and receipt of opioid agonist therapy. Abdominal surgeries and endoscopies were identified utilising Procedure Classes Software from HCUP.[26] Inpatient medication utilisation was determined by Healthcare Common Procedure Coding System J-codes, and outpatient medication claims were identified by Generic Product Identifier codes. Variations in assessed outcomes and covariates in IBD patients with or without chronic opioid use were initially examined and assessed for statistical significance (P < 0.05) utilising Chi-square and t test methods in SAS software, version 9.4.

Statistical Methods

A survival analysis was conducted with chronic opioid use as the independent variable of interest and days of biologic persistence considered as the endpoint. Patients who were persistent to biologic therapy at the end of their observation period were treated as right-censored. Univariable survival curves were estimated for categorical variables, and the functional form of continuous variables was assessed. Univariable hazard ratios were reported for categorical variables and for those continuous variables that showed evidence of being linear in the log-hazard.

A Cox proportional hazards regression model was developed to assess the relationship between chronic opioid use and the risk of discontinuing therapy, while adjusting for potential confounding influences. A purposeful variable selection strategy was followed in developing the multivariate model.[27] Model discrimination and calibration were quantified by Harrell's c statistic and the May-Hosmer goodness-of-fit test[28] with G = 10 risk score groups respectively. R Core Team[29] was used for the survival data analysis, with the survival package[30] for statistical modelling, and the survminer package[31] for plotting survival curves. Significance level for null hypothesis testing was set at 5%.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....