Burden and treatment of chronic obstructive pulmonary disease among people using illicit opioids: matched cohort study in England

Data source

We used data from the Clinical Practice Research Datalink (CPRD) Aurum (version April 2020) and Gold (version January 2019).13 14 These two databases contain pseudo-anonymised data from general practices in England covering about 13% and 8% of the population, respectively. Participants were limited to those with linkage to the national Hospital Episode Statistics and Office for National Statistics mortality databases.

Participants with a history of illicit opioids were defined as those with clinical observations such as heroin dependence or prescriptions of opioid agonist treatment (methadone or buprenorphine). A full list of codes and validation is published, showing that this sample has similar characteristics to other samples of people who use illicit opioids.15 The entry date was the latest of 1 January 2001, 12 months after entry to CPRD, or the first record of illicit opioid use. The 12 month washout period is designed to avoid the unusual time when individuals join primary care research databases, which is often the date of registration at a general practitioner surgery and is associated with poor health and recording of new or pre-existing health conditions. The exit date was the earliest of death and 30 October 2018. A description of rates of all cause mortality and cause specific mortality in this cohort is available elsewhere.16

For each participant, we sampled three patients of the same sex and age (±3 years) and from the same general practice, with no previous records of illicit opioid use. The matched participants were assigned the same cohort entry date as the corresponding participant with a history of opioid use. This design is called exposure density sampling.17 We sampled a second comparison group for the subcohort of COPD cases. Cases with a history of illicit opioid use were each matched to five cases with no history of illicit opioids, by age at diagnosis (±3 years), sex, and date of diagnosis (±12 months).

Definition of chronic obstructive pulmonary disease

New diagnoses of COPD in CPRD Gold were based on a validated code list that has an estimated positive predictive value of 87%.18 For CPRD Aurum, we created a similar case definition by searching diagnostic terms using the keywords "copd," "chronic obstruct*," "bronchitis," "emphysema," and then screening the results for terms that matched the CPRD Gold case definition.

Outcomes for participants with chronic obstructive pulmonary disease

Based on the UK National Institute for Health and Care Excellence's guidance NG115 ("chronic obstructive pulmonary disease in over 16s: diagnosis and management"),19 we defined five evidence based treatments for COPD: seasonal influenza vaccination, pneumococcal vaccine, referral for pulmonary rehabilitation, inhaled drug treatment specific for COPD, and support with smoking cessation. Except for seasonal influenza vaccine, we classified participants as receiving the intervention in the 12 months after diagnosis or not (ie, binary outcomes). For seasonal influenza vaccine, we considered each influenza season (1 September to 31 March) after COPD diagnosis separately. Participants with a diagnosis of COPD during a influenza season who received a vaccine before diagnosis were considered as being vaccinated. Lists of prescriptions and clinical codes for each outcome and covariate are provided in the study protocol.12 We also defined four adverse outcomes after diagnosis of COPD: exacerbations of COPD requiring hospital admission; unplanned hospital admissions with a primary diagnosis of respiratory disease; all cause death; and death with underlying cause of respiratory disease. Definitions and exclusion criteria for each outcome are provided in online supplemental information.

Covariates

For incidence of COPD and death due to COPD in the base cohort, analyses were minimally adjusted (ie, adjusted for age and sex), and then additionally adjusted for smoking status in a mediation analysis. For treatment and adverse outcomes after diagnosis in the subcohort, we chose confounders using a causal model (see diagram in online supplemental information). This model considers the effect of a history of using illicit opioids (or membership of this population) on the outcomes, rather than the direct effects of drug use. An alternative study looking at the direct effects of drugs might consider pathways such as interactions between opioids and COPD drug treatments, or the effect of intoxication on appointment attendance. Instead, this study considers the total effect of a history of illicit opioids on COPD treatment and outcomes, including pathways such as patients’ expectations, stigma among staff, and barriers to healthcare such as homelessness. An example of a question that this study aims to answer is: "when a patient receives a COPD diagnosis from a general practitioner, does the fact the patient uses heroin affect their probability of receiving a pneumococcal vaccine?’"

We adjusted analyses for smoking status; body mass index, using the most recent data in the 12 months before cohort entry; COPD GOLD stage,20 based on forced expiratory volume in 1 second (FEV1), expressed as % predicted (ie, spirometry); the UK Medical Research Council (MRC) dyspnoea (breathlessness) scale21; and comorbidities, defined as the number of unique ICD-10 (international classification of diseases, 10th revision) chapters from diagnoses made during hospital admissions in the three years before COPD diagnosis, limiting to ICD-10 chapters that represent long-term conditions (2-14 and 17). For descriptive purposes, we also extracted information on the index of multiple deprivation22 (a composite measure of neighbourhood characteristics such as crime and employment) per participant.

Statistical analysis

Firstly, we considered death due to COPD using the base cohort. Following the recommended analysis for studies that use exposure density sampling,17 we deduplicated individuals (as the comparison group is sampled with replacement), assigned the earliest cohort entry, and expanded data so that participants had a new observation period when they entered a different age band (defined as 18-24, 25-34, 35-44, 45-54, 55-64, and ≥65 years), or if participants in the comparison group have a record of illicit opioid use. Of 320 367 participants in the comparison group, 1310 (0.4%) had a record of illicit opioid use during follow-up. At the first record of illicit opioid use, these participants started a new observation period with a different illicit opioid status (ie, illicit opioid use was time varying). We used left-truncated Cox proportional hazards model with times relative to 1 January 2001. We then repeated this process for analysis of incident COPD diagnosis, excluding participants with a diagnosis of COPD before cohort entry. The models were first adjusted for age and sex only, and then additionally adjusted for tobacco smoking status as a mediation analysis (ie, to what extent does smoking explain opioid related inequalities in COPD?).

Secondly, we focused on the subcohort of participants with a new diagnosis of COPD and considered treatment and adverse outcomes. We used Poisson regression to estimate the relative risk for each treatment. Each model had a binary dependent variable showing whether the treatment was recorded within 365 days of diagnosis, and independent variables of history of illicit opioid use, prespecified confounders, and an offset for the follow-up time. Censoring was the earliest of death, 365 days after diagnosis, or the end of CPRD follow-up (ie, the final date when primary care data was available). For seasonal influenza vaccines, participants were eligible for a vaccine in each influenza season after diagnosis and had a new follow-up period starting on 1 September each year. We included each season as a separate observation period, and used a mixed Poisson model with random intercepts for the individual identifier. We then used Cox proportional hazards models to estimate the association between illicit opioid use and each adverse outcome, with censoring at the earliest of death or 30 October 2018. We did a sensitivity analysis restricted to COPD cases with records of current smoking at diagnosis, because never smokers might have different exposures (eg, might have more genetic risk factors for COPD) and disease phenotypes, and might be more common among the comparison group.

In preliminary descriptions of the dataset12 we identified missing data in smoking status, COPD stage, breathlessness symptoms, and body mass index. We used the R package Amelia23 to generate 20 complete datasets by multiple imputation for each analysis. Most analyses used Cox proportional hazards regression, and in these imputation models we included the event indicator and the Nelson-Aalen estimator of the cumulative baseline hazard.24 We then conducted analysis on each complete dataset and combined the estimates (eg, log hazard ratios) and standard errors using Rubin’s rule.25 Analysis was done using R version 3.6.2.

Patient and public involvement

This study was designed after workshops with people who use illicit opioids in treatment at South London and Maudsley NHS Foundation Trust. The workshops were conducted as part of DL’s doctoral research and focused on access to general health services. Participants highlighted difficulties getting respiratory symptoms assessed and treated in primary care. We plan to communicate these findings to relevant patient groups and stakeholders.

Characteristics of participants

The base cohort included 106 789 participants with a history of using illicit opioids and 320 367 matched participants. The median age at study entry was 35.1 years and 73 791 (69.1%) participants were male. Participants with a history of using illicit opioids were more likely to be current smokers (78.2% (83 486/106 789) v 33.7% (107 846/320 367)). Characteristics of the base cohort are shown in the online supplemental information.

Figure 1 shows the derivation of the sample, with more detailed diagrams in the online supplemental information. During follow-up, 4016 participants with a history of using illicit opioids received COPD diagnoses; 113 could not be matched with a COPD case with no history of using illicit opioids, and therefore the subcohort included 3903 COPD cases with a history of illicit opioid use and 19 515 matched cases with no history of illicit opioid use.

Request permission

Table 1 shows characteristics of participants with a new diagnosis of COPD. Median age at diagnosis was 48.8 years in the opioid group and 49.1 years in the comparison group, and 2507 (64.2%) participants were male. Participants with a history of using illicit opioids were more likely to be underweight (11.1% v 3.8%), less likely to be overweight or obese (43.9% v 60.9%), more likely to be current smokers (86.1% v 65.6%), had more comorbidities (median 2 v 1), and more severe COPD at diagnosis (63.0% v 70.0% median FEV1 as % predicted; and 30.7% v 16.8% had MRC breathlessness scores of 2-4).

Death and diagnosis

Among 106 789 participants with a history of using illicit opioids, 680 (0.6%) died with an underlying cause of COPD over a median 8.7 years of follow-up (table 2). COPD was the underlying cause in 680 (5.1%) of 13 213 all cause deaths in the cohort. In the comparison group (n=320 367), 160 (0.05%) died with an underlying cause of COPD over a median 9.5 years of follow-up. The Cox proportional hazards model showed that participants with a history of illicit opioids had 14.59 times (95% confidence interval 12.28 to 17.33) the hazard of death with an underlying cause of COPD. After adjustment for smoking status at cohort entry, the hazard ratio was 8.81 (7.36 to 10.54).

Among 106 789 patients with a history of using illicit opioids, 2431 (2.3%) had a diagnosis of COPD before joining the cohort, compared with 1368 (0.4%) of 320 367 in the comparison group. During a median 3.3 years of follow-up, 4016 participants with a history of using illicit opioids received a COPD diagnosis, versus 3051 participants in the comparison group over a median of 5.6 years of follow-up. The Cox proportional hazards model showed that participants with a history of illicit opioid use had 5.89 times (95% confidence interval 5.62 to 6.18) the hazard of a new COPD diagnosis. After adjustment for smoking, the hazard was 3.29 (3.13 to 3.46). Kaplan-Meier curves and detailed results from the Cox proportional hazards models are shown in online supplemental information).

Outcomes after a new diagnosis of chronic obstructive pulmonary disease

The proportions of participants with COPD receiving treatment were similar for those using illicit opioids and for those in the comparison group (table 3). The absolute proportion receiving each treatment varied, with higher proportions for drug treatments (bronchodilators and corticosteroids) and lower proportions for pulmonary rehabilitation referral, pneumococcal vaccine, and smoking cessation support.

Poisson regression showed that illicit opioid use was associated with lower probability of seasonal influenza vaccine and pneumococcal vaccine, and higher probability of pulmonary rehabilitation referral and bronchodilators or corticosteroids. However, the associations were small and confidence intervals were close to a ratio of one (ie, no association). Adjusting for smoking status and disease severity in Poisson regression did not substantially change these associations (figure 2).

Request permission

Participants with a history of illicit opioids before COPD diagnosis had about three times the risk of each adverse outcome (exacerbations requiring hospital admission and death). In Cox proportional hazards models adjusting for smoking status and disease severity, these associations were partially reduced, and participants with a history of illicit opioids had about double the adjusted hazard of each outcome. In sensitivity analysis restricted to participants with records of current smoking at diagnosis, associations between illicit opioid use and outcomes after diagnosis were very similar to the main results (online supplemental information).

Principal findings

In this sample of primary care patients in England, individuals with a history of illicit opioid use had about six times the risk of new COPD diagnosis and about 15 times the risk of death with an underlying cause of COPD. Among patients diagnosed with COPD, those with a history of using illicit opioids had more severe disease, potentially reflecting later diagnosis. We found no large inequalities in treatment after diagnosis, although illicit opioids were associated with substantially higher frequency of exacerbations requiring hospital admission and death, even after adjusting for disease severity at diagnosis.

Interpretation of results

Qualitative research has found that people who use illicit opioids have barriers to health services. These include: stigmatising attitudes among healthcare staff, including perceptions that patients who use illicit drugs have poor motivation or are to blame for their health problems26 27; barriers to attending appointments, such as transport costs and competing priorities such as finding enough food or money for the day26 27; clinicians attributing coughing and other symptoms to drug use, sometimes known as diagnostic overshadowing28; and delaying help seeking owing to normalisation of pain and fear of stigma.29 These barriers could cause later diagnosis of COPD among people who use illicit opioids. In the present study, patients with a history of illicit opioid use had considerably worse symptoms at diagnosis than those with no history.

People who use illicit opioids in England are ageing,30 and the burden of COPD is likely to increase. Given the large inequality and specific health needs in this population, prevention of COPD will require a dedicated strategy. We discuss three approaches to reducing this inequality.

Strengths and limitations of the study

Other studies have estimated COPD prevalence,3–5 incidence,47 and mortality48 49 among people who use illicit opioids, also finding a high health burden related to COPD. To our knowledge, this is the first study investigating treatment in this population. We used a large, representative database that is likely to reflect clinical reality. We used two well known measures of disease severity (COPD GOLD stage20 and the MRC dyspnoea scale21), which were well recorded with about two thirds of participants having each measurement at the time of diagnosis. The study uses five cost effective and evidence based interventions recommended in UK guidelines (seasonal influenza vaccine, pneumococcal vaccine, pulmonary rehabilitation, inhaled drug treatment, and support with smoking cessation), providing evidence across these interventions that illicit opioid use was not strongly associated with healthcare access after diagnosis.

The study had four key limitations. Firstly, the study could have had selection biases, owing to missing data on COPD cases or opioid exposures. For missing COPD cases, data from Wales suggest that about a third of patients on general practitioners' COPD registers do not have recorded spirometry values50 (consistent with missing data in table 1), and that some patients also might not have the diagnostic codes used in our case definition. Missing COPD diagnoses will mean that absolute incidence of diagnosed COPD is underestimated, although we could not identify a reason why this potential limitation would seriously bias our estimate of the association between illicit opioids and COPD incidence. Missing opioid exposures could have biased our estimates of the association between opioid use and treatment. Disclosure of illicit drug use to a general practitioner might be a marker for good engagement with healthcare, and therefore people with COPD who use illicit opioids but who were not identified in our study might have lower rates of treatment.

Secondly, the measurement of smoking is limited to never smoking, ex-smoking, and current tobacco smoking status at cohort entry. Smokers who use illicit opioids are likely to have longer and heavier smoking histories. We found that the association between opioid use and COPD diagnosis or death was partially explained by tobacco smoking status. However, the limited measurement of smoking meant that we were unable to assess the contribution of smoking to inequality in COPD, and the importance of other exposures. In addition, smoking after diagnosis might vary between the opioid and comparison group, which could contribute to the higher rates of acute exacerbations and death after diagnosis. We also did not have data on the route of opioid administration, and particularly whether participants smoke heroin. Therefore, we were unable to investigate the impact of heroin smoking on COPD risk, or make direct comparisons with studies of COPD among heroin smokers.

Thirdly, there could have been unmeasured differences in disease at diagnosis. The higher rate of acute exacerbations in the opioid group suggests residual differences in severity or exposures during follow-up such as ongoing smoking. We measured severity using the COPD GOLD stage and the MRC breathlessness scale. Although these scales are associated with survival probability,1 51 their prognostic value could differ between populations. Differences might also relate to different types of disease. For example, a study of COPD among people who smoke heroin showed that emphysema is the dominant phenotype,11 so it is possible that patients with COPD and a history of illicit opioid use are more likely to have emphysema than other patients with COPD and no history. The vast majority of patients had generic COPD diagnoses rather than records of a specific phenotype, and we were not able to subclassify COPD phenotypes.

Finally, eligibility and compliance with COPD treatments was difficult to measure. Eligibility was easily determined for support with smoking cessation (current smokers), pneumococcal vaccine (participants without a previous vaccine), and seasonal influenza vaccine (everyone). For COPD drug treatments and pulmonary rehabilitation, the clinical decision to provide treatment is based on additional patient characteristics19 and preferences that were not captured in this study—that is, our eligibility definitions might not capture clinical reality. The proportions of patients actually receiving treatments might also vary between the opioid and comparison groups. Our measures of treatment typically capture the first step taken by a clinician, such as a prescription or referral. Vaccinations are usually given immediately and the rate of prescriptions might closely match the rate of administered vaccines, while patients need to pick up inhalers or attend rehabilitation appointments with some potential dropout (which we could not measure).

Conclusion

Death due to COPD is 15 times more common among people who use or have used illicit opioids than the general population. This inequality does not appear to be explained by differences in treatment after diagnosis, but later diagnosis might contribute. A strategy to prevent and treat COPD among people who use illicit opioids should include more accessible assessment of respiratory problems and prioritisation of smoking cessation by services that support this population.