Ageing opioid users' increased risk of methadone-specific death in the UK

Background: The ﬁ rst evidence that the hazard ratio (HR) for methadone-speci ﬁ c death rises more steeply with age-group than for all drug-related deaths (DRDs) came from Scotland ’ s cohort of 33,000 methadone-pre-scription clients. We aim to examine, for England, whether illicit opioid users ’ risk of methadone-speci ﬁ c death increases with age; and to pool age-related HRs for methadone-speci ﬁ c deaths with those for Scotland ’ s me-thadone-prescription clients. Methods: The setting is all services in England that provide publicly-funded, structured treatment for illicit opioid users, the methodology linkage of the English National Drug Treatment Monitoring System and mortality database, and key measurements are DRDs, methadone-speci ﬁ c DRDs, or heroin-speci ﬁ c DRDs, by age-group and gender, with proportional hazards adjustment for substances used, injecting status and periods in/out of treatment. Results: Linkage was achieved for 129,979 adults receiving (cid:129) By pooling results from the two major UK studies, we showed that opioid-dependent clients ’ hazard ratio for methadone-speci ﬁ c death nearly doubled at 35 – 44 years (compared with 25 – 34 years); and quadrupled at 45+ years.

• Record-linkage studies internationally have shown the value of opioid substitution therapy as a treatment which reduces substantially clients' risk of drug-related death (DRD); also that DRDrates are lower for female opioid users and increase with age beyond 35 years but that the female advantage is much reduced for older clients.
• Despite harm reduction measures, such as opioid substitution therapy, UK's DRDs have increased markedly in the past decade, in a strongly age-related manner.
• One powerful record-linkage study on Scotland's methadone-prescription clients in 2009-2013 has shown that their adjusted hazard ratios for methadone-specific DRD increased sharply by age-group, irrespective of gender.

What this study adds
• By analysing the opioid-specificity of deaths for England's National Drug Treatment Monitoring System (NDTMS) cohort of nearly 130,000 opioid users who started a prescribing treatment modality, predominantly methadone, during 1 April 2005 to 31 March 2009, we confirmed that their hazard ratios for methadone-specific DRDs also increased sharply by age-group.
To assess the role of prescribed methadone in explaining the above demographic influences, Gao et al. (2016) considered age-group and gender, in addition to prescription source (general practitioner, othersource) and quintile for the quantity of prescribed methadone, as being potentially informative about the 361 methadone-specific DRDs experienced by 33,000 methadone-prescription clients in Scotland during 121,000 person-years of follow-up in 2009 to 2013. Their analysis revealed a steeply increased hazard by age-group, irrespective of gender (which was not prognostic) and that the top quintile for the baseline quantity of prescribed methadone conferred additional hazard for methadone-specific DRDs.
The first to demonstrate how steeply the risk of methadone-specific death increases with client-age, Scotland's methadone-prescription cohort had the advantage of a national protocol in Scotland for toxicological reporting at forensic autopsy so that the specified opioids were implicated in, not merely present at, DRD. However, substantial numbers of Scotland's two million methadone prescriptions issued over four years lacked a Community-Health Index (CHI)-number, meaning that they were not readily linkable to individual clients. For that reason, Scottish clients' periods on/off prescribed methadone could not be analysed (Gao et al., 2016). Nonetheless, an estimated 82% of Scotland's methadone clients were linkable to mortality-records because they had at least one CHI-identified prescription (Gao et al., 2016).
Besides calling for a better understanding of methadone's pharmacodynamics in older clients, many with potential confounders such as progressive physical or mental ill-health, Gao et al. (2016) noted that, since 2006, electrocardiograms have been recommended by UK's Medicines and Healthcare products Regulatory Authority for older or persistent methadone clients on higher doses. This recommendation was made to detect prolongation of that part of the heart's normal electrical cycle known as the QT interval because, unlike buprenorphine, methadonealone and with other drugs commonly used for comorbidities, especially mental health conditions: antidepressants, antipsychotics, antibiotics (macrolides, quinolones, azoles), antiarrhythmics, protease inhibitors and the loop diuretic furosemide (Romero et al., 2016) is known to prolong the QT interval; and prolongation increases the risk of torsades de pointes and sudden cardiac death, yet leaves no detectable trace at autopsy. Other risk factors for corrected QT prolongation include age-related co-morbidities such as renal impairment, heart or liver disease; and being female, see Gao et al. (2016).
Before diving too deeply into confounders, as above, in explanation for the strongly age-related increase in Scottish clients' hazard of methadone-specific death, we considered that it was important first to test the Scottish results. England's National Drug Treatment Monitoring System (NDTMS) cohort of nearly 130,000 opioid users who had started a prescribing treatment modality during 1 April 2005 to 31 March 2009 enabled such testing. The calendar period for the NDTMS cohort was immediately preceding Scotland's (Pierce et al. 2016), thereby also preceding the heroin drought of 2010; 47% of the NDTMS cohort's person-years were aged 35+ years, similar to 48% at baseline for Scotland's methadone-prescription cohort. Importantly, information on NDTMS clients' periods in/out of treatment, declared injecting and misuse of alcohol, benzodiazepines and other drugs (Pierce et al. 2016) could be taken into account in addition to age-group and gender. The English cohort's gender and age-specific mortality from causes other than DRDs has been reported elsewhere (Pierce et al., 2015).
In this paper, for NDTMS opioid-user clients who received at least one day of opioid agonist prescribing (OAP), we aimed to: i) document the influence of demographic risk factors (age-group; gender) on OAP clients' HR for DRD, methadone-specific DRD, heroin-specific DRD, having adjusted also for clients' time-dependent declared injecting (ever) and past-month misuse of alcohol, benzodiazepines, and other drugs; ii) repeat the above analysis with adjustment also for periods in/out of treatment; iii) pool age-related HRs for methadone-specific deaths from the Scotland's methadone-prescription cohort and England's OAP cohort.

Data
The National Drug Treatment Monitoring System (NDTMS) provides details on all structured treatment for substance misuse provided in England. The cohort for this national record linkage study was identified from NDTMS records collected over the study period 1st April 2005 to 31st March 2009; with linkage to mortality records provided that the subject's identifier was not in a many-to-one mapping, see Fig. 1 by a given patient. Treatment modalities were categorised, as in Pierce et al. (2016) into: OAP; psychosocial treatment; and residential treatment. A treatment episode is defined as an unbroken or contiguous series of treatment modalities. Data were supplied at the episode-level on: self-reported illicit drug injecting status (past month); and an optional self-report of up to three additional problematic psychoactive substances. In NDTMS, prescribed opioid agonist was not differentiated as to methadone or buprenorphine, but the vast majority was methadone (Cornish et al., 2010).
Patients were included in the OAP cohort if they received at least one day of prescribing treatment over the study period and were aged 18 to 64 at their inclusion-date. The cohort analysed here is a sub-cohort of that described by Pierce et al. (2016) who checked thoroughly on the proportionality of DRD hazards.
The NDTMS cohort was linked to ONS mortality records. For each death record, the underlying cause of death is coded according to the ICD-10 framework with DRDs defined by ICD-10 codes according to ONS's definition (Office for National Statistics, 2017). Additionally, for each death, the ONS created flags to indicate whether 'methadone' and/ or 'heroin' and/or 'buprenorphine' were mentioned on the death certificate.
In England, the opioid-specificity of DRDs is determined by the drugs which are mentioned on the death-certificate. As the majority of DRDs are registered following a coroner's inquest, the text on the coroner's death certificate is used by ONS to code all of the substances involved in the death. Analogous to Gao et al. (2016), we define methadone-specific DRDs as those in which methadone was mentioned but neither heroin nor buprenorphine; and heroin-specific DRDs as those in which heroin was mentioned but neither methadone nor buprenorphine. If opioid was mentioned on a DRD death certificate without further specification, which occurs for around 5% of England's opioidrelated DRDs (Office for National Statistics, 2017), the death was counted as DRD; but neither as methadone-specific DRD nor as heroinspecific DRD. Similarly, any DRD which mentioned buprenorphine was counted as DRD; but neither as methadone-specific DRD nor as heroinspecific DRD. However, buprenorphine is so infrequently mentioned in England and Wales as to be not separately tabulated by ONS even for DRD registrations in 2016.
Treatment and mortality data were linked using a minimal identifier in each database (initials, date of birth, gender) and government region of usual residence. Prior analysis indicated that 22% of minimal identifiers in the treatment population were shared by one or more individuals on the criminal records database which contributed to the cohort's derivational linkages (Pierce et al., 2016). These minimal identifiers were removed from the analysis cohort because the subject's drug treatment record could not be reliably linked to his/her mortality record. The resulting OAP cohort comprised in 129,979 linked individuals, see Fig. 1.
Data were rendered anonymous to the research team, via irreversible encryption of identifying information, prior to their release by source organizations.

Statistical analysis
Descriptive life-table analysis (per 10,000 person-years of followup) was undertaken for the outcomes: any DRD; DRD that mentioned methadone on the death certificate but neither heroin nor buprenorphine (methadone-specific DRD); and DRD that mentioned heroin on the death certificate but neither methadone nor buprenorphine (heroinspecific DRD).
Subjects entered the risk-set at their earliest OAP-treatment date during the period 1st April 2005 to 31st March 2009. They left the risk set at the earlier of their date of death, 65th birthday, or the end of data collection (31st March 2009).
Time-dependent covariates for the OAP cohort were time-updated age (categorised as 18-24; 25-34, 35-44, 45-64 years) and patient-reported behavioural covariates, collected at the beginning of each treatment episode and defined from their date of first report, namely: injecting status; problematic use of alcohol, benzodiazepines, crack cocaine, cocaine powder and amphetamines (the latter two substances combined due to a low level of reporting). A time-dependent covariate for being in/out of treatment was also created to indicate whether the patient was enrolled in any structured treatment (prescribing or otherwise) using the start and end dates for treatment modality.
For each outcome, cause-specific Cox proportional hazard ratios were estimated by censoring other deaths and using covariates as above. Interaction between age and gender was anticipated for DRDs (Pierce et al., 2015;King et al., 2013;King et al., 2014).
When pooling the age-related HRs for methadone-specific deaths from the Scottish and English cohorts, age-group specific weights were derived from the information contributed per age-group by each cohort; alternatively, a common weighting was derived from the total information across all three age-groups (< 25 years; 35-44 years; 45+ years). Please see Supplementary Information for details of both calculations. In statistical terms, notice that the information which a cohort contributes about ln HR for a particular age-group (relative to 25-34 year olds) is defined by the reciprocal of the variance of the cohort's estimated ln HR for that age-group.
None of the behavioural risk-factors in Table 1 had a materially different influence on heroin-specific versus methadone-specific DRDs, apart from having ever injected (a more accentuated HR for heroinspecific deaths). Table 2 shows that the doubling of DRD-risk when outof-treatment partitions as a nearly quadrupled risk of heroin-specific DRD but non-significant 20% out-of-treatment reduction in methadonespecific DRD risk. Other covariate influences are unchanged from Table 1.
The weights used for the Scottish methadone-prescription cohort in pooling age-group specific HRs (see Introduction for the Scottish results to be pooled) were 43% for under 25 years of age, 63% for 35-44 year olds, and 55% at 45+ years; alternatively, a common weighting of 58% may be preferred, which is based on the sum of information across agegroups. The alternatives agree inferentially, see Supplementary information.

Summary of main findings and recommendations
The OAP cohort allowed us to validate, for England, the novel record-linkage findings from Scotland that opioid using clients' hazard of methadone-specific death rises steeply with age and is independent of gender. In addition, we have shown that these demographic influences hold up when adjustment is made for a triad of declared behavioural characteristics (injecting, misuse of alcohol and misuse of benzodiazepines) and for periods in/out of opioid substitution therapy, factors not addressed in Scotland's methadone-prescription cohort.
Next, we pooled age-related HRs from the two national cohorts for added precision so that, relative to 25-34 year olds, the OAP clients' pooled HRs for methadone-specific deaths in the UK were doubled at 35-44 years and nearly quadrupled at 45+ years.
We therefore recommend that consideration be given to reviewing older clients' methadone dosage in the context of the patient's cardiovascular, respiratory and liver or renal related co-morbidities (Gao et al., 2016); and prescribed medications, including for mental health problems (Independent Expert Group, 2017). Gao et al. (2016) recommended that a representative sample of 45+ year old methadone clients be offered electrocardiograms (ECG) to establish what proportion of them (1%, 5% or 10%) had prolongation of their corrected QT segment (by 60 milliseconds or to above 500 milliseconds). In the light of our evidence-synthesis, we propose that this surveillance should be extended to 35-44 year olds. We propose surveillance as the initial step because we recognise that methadone treatment centres currently may lack equipment to perform and qualified staff to interpret ECGs and yet may be clients' only contact with health care systems.
Finally, we recommend that methadone clients can safeguard themselves and their older peers by carrying a naloxone-kit (Bird, McAuley, Perry, & Hunter, 2016;Parmar, Strang, Choo, Meade, & Bird, 2017) and ensuring that family and friends know how to intervene in the event of overdose.

Comparability of cohorts from England and Scotland
The availability of a cohort with well over 100 methadone-specific DRDs is a considerable strength of England's validation study. The OAP cohort of nearly 130,000 clients (1 April 2005-31 March 2009) and Scotland's methadone-prescription cohort of 33,000 clients (1 July 2009-30 June 2013) were broadly similar in terms of age-distribution, number of methadone-specific deaths (271; 361) and average follow-up time. Scottish clients' age-distribution was documented at baseline only (Gao et al., 2016); the English cohort's was time-updated. Scotland's DRD-classification, supported by a national protocol for toxicology at forensic autopsy, is based on drugs implicated in the death. England's used the text on the coroner's death certificate to code all of the substances involved in the death. Scot;1;land's methadone-prescription cohort had an overall DRD rate of 63 per 10,000 pys compared with the OAP cohort's much lower rate of 33 DRDs per 10,000 pys. This difference may be partly accounted for by injection drug use being less prevalent in England (King et al., 2013;King et al., 2014). Under half the OAP clients had a history of injecting (declared or undeclared), a behavioural covariate which, as Table 1 showed, doubles clients' hazard ratio for DRD. Unfortunately, information on past-history of injecting was not directly available for Scotland's methadone-prescription cohort. Period effects on DRD-rates cannot be ruled out between 2005-2009 and 2009-2013, the second of which coincided with a re-focusing of the UK's drug policy away from harm reduction towards accelerated recovery, an emphasis that has since been moderated in the UK Government's 2017 Drugs Strategy (2017), see Her Majesty's Government (2017). Importantly, the earlier period preceded the heroin drought of 2010.

Limitations
For different reasons, the availability of unique identifiers limited linkage to about 80% of clients in the OAP cohort, see Fig. 1; and in Scotland's methadone-client cohort (Gao et al., 2016). As detailed in Methods, we discarded 22% of identifiers from the OAP cohort where there was evidence that two or more individuals might share the same details. Nonetheless, it remains possible that some proportion of the remainder might also have shared details with other individuals in their region. Errors in matching within treatment records and in matching M. Pierce et al. International Journal of Drug Policy 55 (2018) 121-127 these to mortality records might potentially have occurred as a consequence of this but, insofar as these arose due to the random coincidence of the minimal identifying information that was available in the OAP cohort, these errors will have contributed noise to the analysis rather than exerting any systematic effect. England, and many other countries, does not have a national protocol for toxicology and its reporting at forensic autopsy. Thus, England's classification of methadone-specific deaths is on a less firm basis than Scotland's.
Details of opioid agonist prescribing were absent for the English cohort: both which agonist (methadone versus buprenorphine) and the quantity prescribed. We do not know directly what proportion of OAP clients received methadone, although we expect the proportion to have been high, around 90% (Cornish et al., 2010). Scotland's age-related HRs for methadone-specific DRD increased just as steeply whether prescribed quantity of methadone was, or was not, fitted. Hence, the OAP cohort's absence of methadone dose does not undermine our validation. Importantly, the English OAP cohort's additional adjustment for behavioural covariates and for periods in/out of treatment, see also Pierce et al. (2016), did not mitigate the steep age-relatedness in adjusted HRs for methadone-specific deaths either.
Unlike for demography, self-reporting of behavioural covariates could lead to HRs that are closer to 1 than would be the case without mis-classification and so some residual behavoural confounding cannot be ruled out.
The absence from both record-linkage studies of information on ageing clients' co-morbidities and specific other prescriptions which may interact with methadone is a limitation to be addressed in further research. In making a case for the public benefit of linking-in methadone clients' hospitalizations and specified other prescribing records, applicants can now cite, in their support, the UK's evidence-synthesis on ageing opioid users' risk of methadone-specific death.

Further research
Similarly large national cohort studies of OAP clients, investigating how age-group and gender influence the opioid-specificity of clientdeaths, are needed for robust generalizations outside of the UK.
Within the UK, a strong public interest case can now be made for more extensive record-linkage studies to include, besides mortality and DRDs' opioid-specificity, OAP clients' or methadone-prescription clients' history of hospitalizations (by which to infer co-morbidities), diagnosed carriage of blood-borne viruses (particularly the most prevalent, Hepatitis C virus), and selected co-prescribing (including for mental health conditions).
Without research, as above, into possible confounders, we can only speculate, as did Gao et al. (2016), on the underlying cause for the nowconfirmed steep age-gradient for methadone-specific DRDs: corrected QT prolongation, renal or liver impairment, or some other aspect of the pharmacology of methadone in older clients that makes for a longer half-life. Indeed, it is the report of one of the authors that some methadone patients have expressed to him the opinion that, as they age, they don't seem to need as high a methadone dose (JRR, personal communication). In short, cardiac and pulmonary co-morbidities, as well as liver and renal disease, may contribute to methadone-specific DRDs.
As too little is known about the gender-and age-specific pharmacology of methadone in older opioid-dependent clients, we suggest a pharmacokinetic study that simultaneously assesses methadone's pharmacokinetic properties in different settings of clients' cardiac, respiratory, renal or liver impairment. Additionally, given the concern for sudden cardiac death linked to QTc prolongation and subsequent arrhythmias, extended monitoring in the form of Holter monitoring or loop recorders could further elucidate older clients' risk when prescribed methadone.

Conclusion
Only formal studies will resolve what can and should be done to moderate opioid dependent clients' increased risk of methadone-specific death as they age. Consistent findings in two major UK cohorts represent sufficient grounds for methadone prescribers, at least in the UK, to alert clients that their risk of methadone-specific death increases as they age − for reasons that we do yet fully understand. Review of older clients' methadone dose in the light of co-morbidities and prescribed other medications may be worthwhile with consideration given, for some clients, to switching their opioid substitution formulation (McCance-Katz, 2011;Jolley et al., 2015).

Funding
Medical Research Council grant G1000021 for the NIQUAD (Nationally Integrated Quantitative Understanding of Drug-harms) Addictions Cluster. The funder had no role in the decision to publish this paper.

Contributions
SMB proposed the validation study for England to MP & TM who are its guarantors; TM holds privacy access permission for linkage of National Drug Data Warehouse (2005)(2006)(2007)(2008)(2009) to Office for National Statistics mortality records, which was conducted in autumn 2011 to allow 2.5 years for the delayed registration of drugs-related deaths. SMB & JRR had instigated inquiries into Scotland's methadone prescribing and a possible link to Scotland's increased number of methadone-specific deaths and been co-authors of the initial study which demonstrated increased hazard for older clients; MP re-programmed previous analyses of the English cohort to focus on methadone-specific deaths; all authors contributed to the writing, referencing and review of the manuscript.

Data sharing statement
The authors are bound by the conditions imposed on their access to the Drugs Data Warehouse but can assist others in how to achieve similar access.

Transparency
The lead author (as the manuscript's guarantor) affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Ethics/privacy access approval
Data were extracted from the Drug Data Warehouse for a cohort of opioid users, aged 18 to 64 years, actively using or being treated for opioid use, in England over the period 1 st April 2005 to 31 st March 2009. Deaths occurring in the cohort were established by case linkage to national mortality records. Data were rendered anonymous to the research team, via irreversible encryption of identifying information, prior to their release by source organisations.
Use of mortality records was approved by the Office for National Statistics Microdata Release Panel. Use of data from the Drug Data Warehouse was authorised by those organisations providing data. The NHS Central Office for Research Ethics Committees and The University of Manchester Research Ethics Committee advised that further approval was not required for a study of this type.

Conflicts of interest
SMB served on Scotland's National Naloxone Advisory Group and was one of three co-principal investigators for the MRC-funded N-ALIVE pilot Trial of naloxone-on-release. SMB holds GSK shares. JRR chaired Scotland's National Forum on Drugs-related Deaths and serves on the committee which is updating UK's opioid prescribing recommendation. SMB and JRR were co-authors of the Scottish study for which this study provides validation.