Prescription of benzodiazepines, z-drugs, and gabapentinoids and mortality risk in people receiving opioid agonist treatment: Observational study based on the UK Clinical Practice Research Datalink and Office for National Statistics death records.

BACKGROUND: Patients with opioid dependency prescribed opioid agonist treatment (OAT) may also be prescribed sedative drugs. This may increase mortality risk but may also increase treatment duration, with overall benefit. We hypothesised that prescription of benzodiazepines in patients receiving OAT would increase risk of mortality overall, irrespective of any increased treatment duration. METHODS AND
FINDINGS: Data on 12,118 patients aged 15-64 years prescribed OAT between 1998 and 2014 were extracted from the Clinical Practice Research Datalink. Data from the Office for National Statistics on whether patients had died and, if so, their cause of death were available for 7,016 of these patients. We identified episodes of prescription of benzodiazepines, z-drugs, and gabapentinoids and used linear regression and Cox proportional hazards models to assess the associations of co-prescription (prescribed during OAT and up to 12 months post-treatment) and concurrent prescription (prescribed during OAT) with treatment duration and mortality. We examined all-cause mortality (ACM), drug-related poisoning (DRP) mortality, and mortality not attributable to DRP (non-DRP). Models included potential confounding factors. In 36,126 person-years of follow-up there were 657 deaths and 29,540 OAT episodes, of which 42% involved benzodiazepine co-prescription and 29% concurrent prescription (for z-drugs these respective proportions were 20% and 11%, and for gabapentinoids 8% and 5%). Concurrent prescription of benzodiazepines was associated with increased duration of methadone treatment (adjusted mean duration of treatment episode 466 days [95% CI 450 to 483] compared to 286 days [95% CI 275 to 297]). Benzodiazepine co-prescription was associated with increased risk of DRP (adjusted HR 2.96 [95% CI 1.97 to 4.43], p < 0.001), with evidence of a dose-response effect, but showed little evidence of an association with non-DRP (adjusted HR 0.91 [95% CI 0.66 to 1.25], p = 0.549). Co-prescription of z-drugs showed evidence of an association with increased risk of DRP (adjusted HR 2.75 [95% CI 1.57 to 4.83], p < 0.001) but little evidence of an association with non-DRP (adjusted HR 0.79 [95% CI 0.49 to 1.28], p = 0.342). There was no evidence of an association of gabapentinoid co-prescription with DRP (HR 1.54 [95% CI 0.60 to 3.98], p = 0.373) but evidence of an association with increased non-DRP (HR 1.83 [95% CI 1.28 to 2.62], p = 0.001). Concurrent benzodiazepine prescription also increased mortality risk after consideration of duration of OAT (adjusted HR for DRP with benzodiazepine concurrent prescription 3.34 [95% CI 2.14 to 5.20], p < 0.001). The main limitation of this study is the possibility that unmeasured confounding factors led to an association between benzodiazepine prescription and DRP that is not causal.
CONCLUSIONS: In this study, co-prescription of benzodiazepine was specifically associated with increased risk of DRP in opioid-dependent individuals. Co-prescription of z-drugs and gabapentinoids was also associated with increased mortality risk; however, for z-drugs there was no evidence for a dose-response effect on DRP, and for gabapentinoids the increased mortality risk was not specific to DRP. Concurrent prescription of benzodiazepine was associated with longer treatment but still increased risk of death overall. Clinicians should be cautious about prescribing benzodiazepines to opioid-dependent individuals.

Introduction

Deaths amongst users of illicit opioids are increasing despite the fact that in many jurisdictions, such as the UK, a high proportion of users are in treatment [1]. Evidence suggests that opioid substitution therapy (OAT) is effective in improving health and, in individuals on long-term maintenance therapy, in reducing mortality risk [2-6]. Though mortality rates are highest amongst users not in treatment, deaths during treatment and following the end of treatment are still observed [6,7]. Several aspects of treatment may influence this mortality risk, and clinical guidelines emphasise these [8]. Guidelines typically advise against additional prescription of drugs that may potentiate the respiratory depressive effects of opioids and increase risk of overdose death [9]. These drugs include benzodiazepines, “z-drugs” (zaleplon, zolpidem, and zopiclone), and gabapentinoids (gabapentin and pregabalin).

Despite this advice, these drugs are commonly prescribed to opioid-dependent individuals [10-12]. This practice may reflect both uncertainty around the strength of existing evidence that the practice is harmful and a belief that additional prescription may have benefits through increasing treatment retention that have not been investigated yet [13].

We studied the association between additional prescription and treatment retention and the question of whether additional prescription was associated with overall benefit or harm in opioid-dependent individuals receiving OAT using data from a UK database of electronic patient records from primary care linked to a national mortality registry. We examined associations of concurrent prescription (additional prescribing during OAT) and co-prescription (additional prescribing during OAT and up to 1 year after OAT) of benzodiazepines, z-drugs, and gabapentinoids with treatment duration and mortality. We hypothesised that both concurrent and co-prescription of these drugs would be associated with increased mortality risk, despite any effect on treatment duration. We further hypothesised that a causal basis for such increased risk would be reflected in a specific association of both concurrent and co-prescription with drug-related poisonings (DRPs), a dose–response association of greater risk with higher doses of the additional drug, and an effect estimate robust to adjustment for possible confounding.

Methods

Study protocol and pre-specified analyses

Our study protocol was reviewed by the Independent Scientific Advisory Committee of the Clinical Practice Research Datalink (CPRD) (S1 Protocol). Our full study report including pre-specified analyses is available at https://www.journalslibrary.nihr.ac.uk/hsdr/hsdr07030#/abstract.

Ethical approval

Research studies approved by the CPRD Independent Scientific Advisory Committee do not require additional ethical approval. Scientific approval for this study was given by the CPRD Independent Scientific Advisory Committee as above, and no additional informed consent was required as there was no individual patient involvement The permissions covering the research use of CPRD data are described at https://www.cprd.com/transparency-information.

OAT patients

Data were identified for 49,279 primary care patients within the CPRD (https://www.cprd.com/home/) who had received methadone or buprenorphine between the study dates of 1 January 1998 and 31 July 2014. CPRD is a large database of anonymised patient records from 674 general practices and over 11 million patients in the UK, and currently collects data from approximately 7% of the UK population [14]. As described elsewhere, we used diagnostic and prescription formulation information to exclude 26,324 patients who were prescribed buprenorphine or methadone for pain relief, as well as 9,950 patients who received doses below the minimum expected for OAT (i.e., at least 20 mg methadone or 4 mg buprenorphine per day) [15]. We also excluded patients who were aged <15 or >64 years at the start of the study based on an assumption that their opioid prescription was less likely to be for OAT. This left a total of 667,288 prescriptions related to 12,118 patients; 65% were only prescribed methadone, 22% only buprenorphine, and 13% both medications (S1 Patient Flowchart).

OAT episodes

For each individual, we classified their follow-up time into episodes on and off OAT. These covered the period on treatment and the period off treatment up to 1 year after the prescription ended. An individual could experience several consecutive episodes of OAT. We defined a new episode of OAT where a gap of >28 days existed between the expected completion of one prescription and the start of the next, or, in other words, where the prescription interval exceeded the prescription duration by more than 28 days.

First (observed) episodes were potentially subject to left censoring because of the study start date, date of registration with the current practice, or the CPRD up-to-standard date. All last episodes were right censored because of the study end date, the last collection date by CPRD, the patient leaving the CPRD practice, death of the patient, or censoring 1 year after cessation of the last treatment. The earliest of these 5 events defined the end of follow-up. We censored follow-up at 1 year after the end of the last treatment episode to avoid the dilution of mortality risks where patients may have been at lower risk due to the possibility of recovery.

Additionally prescribed medications

Ten benzodiazepine and 3 z-drug medications were listed in British National Formulary, with chapters 4.1 and 4.2 providing information on recommended doses [16]. There were 365,582 prescriptions for the 10 benzodiazepines, 75,926 for the 3 z-drugs, and 23,451 for the 2 gabapentinoids. Treatment episodes were generated in a similar fashion as for OAT except that the minimum gap was reduced to 14 days for benzodiazepine and z-drugs, reflecting their shorter recommended treatment duration.

Confounding factors

Adjustment was made for sex, age, year of treatment, comorbidity, UK region, type of OAT, and OAT period. Comorbidity was calculated based upon 17 chronic illnesses as defined by Khan et al. [17]. The 3,156 READ codes were translated to the current CPRD medcodes. This was possible for 2,856 codes. Occurrence of the medcodes within the clinical notes generated event dates for time-varying covariates. Each occurrence received a weighting from 1 to 6 related to the particular chronic illness. The weights were summed across time and illnesses to create the index. To take account of associations between benzodiazepine, z-drugs, and gabapentinoids, all other drug exposures were included in models examining the effects of each main drug.

Deaths

All-cause mortality (ACM) was derived from dates recorded in the clinical notes and from patient records within CPRD. Dates of death were also available from Office for National Statistics (ONS) records for a subset of the CPRD patients in England. ONS data included details of primary and secondary causes of death, allowing DRP to be defined (S1 Table). The ICD-9/ICD-10 codes assigned in our study are based upon ONS but extended to take account of possible underreporting in the use of codes relating to non-specific or unknown causes [18]. In addition, since deaths from DRP contribute to ACM, we examined deaths not attributable to DRP as a separate category (non-DRP).

Statistical analyses

We investigated the association of concurrent prescription with OAT duration using linear regression. As duration of treatment episodes was highly positively skewed, we repeated this analysis after undertaking a log transformation of duration.

We then used Cox proportional hazards survival analysis to analyse mortality risk associated with co-prescription of benzodiazepines, z-drugs, or gabapentinoids, both on and off OAT (i.e., within a treatment episode) in unadjusted and adjusted analyses. Where a main effect of co-prescription was suggested by these analyses, we investigated whether this effect was modified by an interaction with OAT period (i.e., on or off treatment). For benzodiazepine or z-drug dose effects (2 degrees of freedom), a linear trend was also fitted to examine the effect of dose, with “high” dose defined as a daily dose above the maximum recommended in the British National Formulary (S2 Table). Because of smaller numbers, gabapentinoids were assessed as off/on treatment only.

We then assessed the overall effect of concurrent prescription on mortality risk, using Cox proportional hazards survival analysis. This analysis effectively allows for the longer OAT duration associated with concurrent prescription. As sensitivity analyses, we repeated the main analyses excluding the first treatment episode (to address left censoring).

All analyses were undertaken using Stata version 14.2. We assessed the assumption of constant hazards over time by dividing each episode into weeks 0–4 on treatment, the rest of the period on treatment, weeks 0–4 off treatment, and the rest of the period off treatment. As a further sensitivity analysis, we repeated the analyses using Poisson regression. We also repeated our survival analysis of the effects of co-prescription on risk of DRP considering competing risk of death from other causes.

For clarity we include a summary of our analyses and the relevant Stata commands as S1 Appendix.

Results

Data characteristics

These data contained 29,540 OAT episodes: 17,391 with methadone (median treatment duration, 0.30 years; range, 0.00–23.00; interquartile range, 0.06–1.07 years), 9,180 with buprenorphine (median treatment duration, 0.11 years; range, 0.00–17.00; interquartile range, 0.02–0.38), and 2,969 episodes relating to more than 1 drug. Total follow-up time was 36,126 person-years, during which there were 17,139 episodes of benzodiazepine treatment, 5,957 episodes of z-drug treatment, and 2,822 episodes of gabapentinoid treatment. Benzodiazepines were prescribed during 10,022 OAT episodes in 7,059 patients. Z-drugs were prescribed during 4,279 OAT episodes in 2,822 patients. Gabapentinoids were prescribed during 1,827 OAT episodes in 1,281 patients.

Study participants

Characteristics of study participants are summarised in Table 1. Two-thirds of the sample were male, and mean age of the sample at study exit was 39 years, with a mean follow-up per patient of 3.4 years. During the observation period, 42% of individuals in the sample were co-prescribed benzodiazepines, 20% were co-prescribed z-drugs, and 8% were co-prescribed gabapentinoids.

Table 1

Description of study participants overall and by cause of death.

Characteristic	CPRD		CPRD linked to ONS
	Total	All-cause deaths	Total	Drug-related poisoning deaths	Non-drug/other causes of death
Patients (n)	12,118	657	7,106	113	285
Male (%)	67.3	65.1	68.1	87.6	61.4
Mean (SD) age at exit, years	38.8 (10.4)	47.3 (12.3)	39.3 (10.7)	37.0 (9.6)	51.9 (11.6)
Median (IQR) follow-up, years	2.07 (1.03–4.79)	0.89 (0.29–3.25)	2.09 (1.03–4.82)	1.90 (0.66–4.62)	0.78 (0.20–2.49)
Comorbid at exit (%)	31.4	66.2	32.8	39.8	76.1
Ever prescribed (%)
Benzodiazepine	46.8	58.3	45.0	65.5	52.3
Z-drug	24.0	23.6	26.4	38.9	23.2
Gabapentinoid	8.8	26.5	9.0	8.8	33.0
Concurrent with OAT (%)
Benzodiazepine	42.2	52.5	40.2	61.9	47.0
Z-drug	19.7	20.7	21.8	31.0	22.5
Gabapentinoid	7.6	25.4	7.7	8.8	31.6

CPRD, Clinical Practice Research Datalink; OAT, opioid agonist treatment; ONS, Office for National Statistics.

Unadjusted mortality rates according to participant characteristics other than co-prescription are summarised in S3 Table. DRP was higher in men, not significantly associated with age, higher in the observation period up to 2004, and higher in patients with more comorbidity. Non-drug-related (non-DRP) mortality was higher in women, increased with increasing participant age and comorbidity, and did not show any evidence of trend across the observation period. All mortality rates showed considerable regional variation.

Prevalence of co-prescription over time

We examined prevalence of co-prescription amongst patients on OAT between 1998 and 2014. Benzodiazepine co-prescription decreased over this period from 43% (95% CI 40% to 46%) in 1998 to 28% (95% CI 26% to 29%) in 2014. In contrast, co-prescription of z-drugs remained constant at 13% (95% CI 11% to 15%), while co-prescription of gabapentinoids increased from 0.5% (95% CI 0.1% to 1.1%) in 1998 to 12% (95% CI 11% to 14%) in 2014.

Association of concurrent prescription of benzodiazepines, z-drugs, and gabapentinoids with duration of OAT episodes

In total, 27,598 of 29,540 episodes of OAT were available for this analysis. During 2,969 of these episodes, more than 1 substitute drug was prescribed: We included 1,027 episodes where methadone and buprenorphine were prescribed consecutively but excluded 1,942 episodes where other opioid substitutes such as dihydrocodeine were prescribed. Of these 27,598 OAT episodes, 17,111 (62%) had no concurrent prescription. Buprenorphine episodes were considerably shorter than methadone episodes (Table 2). Concurrent prescription of both benzodiazepines and z-drugs was associated with a similar proportional increase in duration of both methadone and buprenorphine treatment episodes (adjusted mean duration of OAT episode without concurrent prescription of benzodiazepine 244 [95% CI 236 to 252] days, with concurrent prescription of benzodiazepine 416 [95% CI 404 to 429] days, with concurrent prescription of z-drug 441 [95% CI 421 to 461] days, and with concurrent prescription of gabapentinoid 189 [95% CI 159 to 219] days).

Table 2

Comparison of median, unadjusted mean, and adjusted mean opioid agonist treatment (OAT) duration with concurrent prescription of benzodiazepines, z-drugs, and gabapentinoids, by OAT type.

OAT prescribed	Concurrent prescription	Number of episodesa	Median duration	Unadjusted mean duration (95% CI)	Adjustedb mean duration (95% CI)
Any OATc	None	17,111	62	231 (223 to 240)	244 (236 to 252)
	Benzodiazepine	7,961	147	423 (410 to 437)	416 (404 to 429)
	Z-drug	3,165	223	448 (427 to 469)	441 (421 to 461)
	Gabapentinoid	1,405	86	270 (239 to 301)	189 (159 to 219)
Methadone	None	10,662	84	272 (261 to 284)	286 (275 to 297)
	Benzodiazepine	5,283	182	471 (453 to 489)	466 (450 to 483)
	Z-drug	1,930	277	491 (462 to 520)	483 (456 to 511)
	Gabapentinoid	745	125	346 (301 to 392)	224 (180 to 268)
Buprenorphine	None	5,983	32	131 (121 to 140)	135 (126 to 144)
	Benzodiazepine	2,250	59	231 (214 to 248)	234 (217 to 250)
	Z-drug	953	93	270 (244 to 296)	266 (240 to 291)
	Gabapentinoid	628	52	180 (149 to 212)	140 (109 to 171)

Duration is reported as days.

aIncludes OAT episodes where more than 1 drug was concurrently prescribed. For both types, this involved 2,044 episodes.

bAdjusted for sex, age, year, comorbidity, region, and, where applicable, OAT type and concurrent prescription of benzodiazepine, z-drug, or gabapentinoid.

cIncludes episodes of only methadone, only buprenorphine, or both medications consecutively prescribed.

Duration of OAT episodes was highly positively skewed; because of this we undertook a log transformation prior to our analysis and also report interquartile ranges (IQRs) of treatment duration in addition to the median (Table 3).

Table 3

Comparison of log opioid agonist treatment (OAT) duration, plus interquartile range (IQR) of duration, for concurrent prescription of benzodiazepines, z-drugs, and gabapentinoids, by OAT type.

OAT prescribed	Concurrent prescription	Unadjusted	Adjusteda	IQR
Any OATb	None	Ref	Ref	13–230
	Benzodiazepine	0.64 (0.56, 0.71)	0.48 (0.42, 0.54)	27–568
	Z-drug	0.88 (0.79, 0.98)	0.77 (0.69, 0.85)	53–688
	Gabapentinoid	0.13 (−0.01, 0.27)	0.23 (0.10, 0.36)	18.5–362.5
Methadone	None	Ref	Ref	18–308
	Benzodiazepine	0.63 (0.56, 0.71)	0.49 (0.42, 0.55)	44–712.5
	Z-drug	0.91 (0.82, 1.00)	0.78 (0.69, 0.86)	84–878
	Gabapentinoid	0.21 (−0.002, 0.43)	0.20 (0.02, 0.38)	29–629
Buprenorphine	None	Ref	Ref	9–139
	Benzodiazepine	0.63 (0.50, 0.76)	0.43 (0.32, 0.54)	14–382.5
	Z-drug	0.97 (0.82, 1.13)	0.64 (0.50, 0.78)	35–524.5
	Gabapentinoid	0.16 (−0.02, 0.34)	0.30 (0.13, 0.47)	13–226

Treatment episodes as in Table 2.

aAdjusted for sex, age, year, comorbidity, region, and, where applicable, OAT type and concurrent prescription of benzodiazepine, z-drug, or gabapentinoid.

bIncludes episodes of only methadone, only buprenorphine, or both medications consecutively prescribed.

Associations of co-prescription with mortality

Benzodiazepines

There was strong evidence of a higher rate of DRP during periods of benzodiazepine co-prescription (adjusted HR 2.96 [95% CI 1.97 to 4.43], p < 0.001), with evidence of a dose–response relationship (adjusted HR on normal dose 2.51 [95% CI 1.57 to 4.01], p < 0.001; on high dose 4.57 [95% CI 2.46 to 8.47], p < 0.001). There was little evidence of a difference in rates of ACM and non-DRP according to whether patients were on or off benzodiazepines (Table 4).

Table 4

Mortality rates and hazard ratios for all-cause mortality, DRP, and other causes of death by co-prescription exposure to benzodiazepines, z-drugs, or gabapentinoids in people with opioid dependency in primary care.

Co-prescription	Deaths	PY	MR (per 100 PY)	Unadjusted		Adjusteda
				HR (95% CI)	p-Value	HR (95% CI)	p-Value
All-cause mortality
Benzodiazepine
Off	513	28,766	1.78	1 (ref)	0.718	1 (ref)	0.105
On	144	7,361	1.96	1.03 (0.86 to 1.25)		1.17 (0.97 to 1.42)
Off	513	28,766	1.78	1 (ref)	0.595	1 (ref)	0.263
On normal dose	116	5,909	1.96	1.00 (0.82 to 1.22)		1.18 (0.96 to 1.46)
On high dose	28	1,452	1.93	1.22 (0.83 to 1.78)		1.12 (0.75 to 1.66)
Linear effect of dose	—	—	—	1.05 (0.91 to 1.22)	0.507	1.11 (0.96 to 1.29)	0.158
Z-drug
Off	606	34,264	1.77	1 (ref)	0.014	1 (ref)	0.124
On	51	1,862	2.74	1.43 (1.08 to 1.91)		1.26 (0.94 to 1.69)
Off	606	34,264	1.77	1 (ref)	0.048	1 (ref)	0.280
On normal dose	29	1,008	2.88	1.42 (0.98 to 2.06)		1.21 (0.83 to 1.78)
On high dose	22	854	2.58	1.45 (0.95 to 2.22)		1.34 (0.86 to 2.11)
Gabapentinoid
Off	574	35,129	1.63	1 (ref)	<0.001	1 (ref)	<0.001
On	83	998	8.32	2.79 (2.20 to 3.54)		1.71 (1.33 to 2.20)
DRP
Benzodiazepine
Off	74	16,270	0.45	1 (ref)	<0.001	1 (ref)	<0.001
On	39	3,679	1.06	2.35 (1.59 to 3.47)		2.96 (1.97 to 4.43)
Off	74	16,270	0.45	1 (ref)	<0.001	1 (ref)	<0.001
On normal dose	25	2,889	0.87	1.93 (1.22 to 3.04)		2.51 (1.57 to 4.01)
On high dose	14	790	1.77	3.83 (2.16 to 6.80)		4.57 (2.46 to 8.47)
Linear effect of dose	—	—	—	1.95 (1.50 to 2.54)	<0.001	2.22 (1.69 to 2.92)	<0.001
Z-drug
Off	98	18,838	0.52	1 (ref)	0.001	1 (ref)	<0.001
On	15	1,110	1.35	2.52 (1.46 to 4.34)		2.75 (1.57 to 4.83)
Off	98	18,838	0.52	1 (ref)	0.001	1 (ref)	0.001
On normal dose	10	593	1.69	3.17 (1.65 to 6.09)		3.66 (1.86 to 7.19)
On high dose	5	517	0.97	1.78 (0.72 to 4.37)		1.55 (0.59 to 4.06)
Gabapentinoid
Off	108	19,410	0.56	1 (ref)	0.212	1 (ref)	0.373
On	5	538	0.93	1.79 (0.72 to 4.46)		1.54 (0.60 to 3.98)
Non-DRP
Benzodiazepine
Off	235	16,628	1.41	1 (ref)	0.302	1 (ref)	0.549
On	50	3,757	1.33	0.85 (0.63 to 1.16)		0.91 (0.66 to 1.25)
Off	235	16,628	1.41	1 (ref)	0.497	1 (ref)	0.711
On normal dose	44	3,007	1.46	0.88 (0.64 to 1.22)		0.94 (0.67 to 1.32)
On high dose	6	751	0.80	0.67 (0.30 to 1.52)		0.72 (0.31 to 1.66)
Linear effect of dose
Z-drug	265	19,252	1.38	1 (ref)	0.266	1 (ref)	0.342
Off	20	1,134	1.76	1.29 (0.82 to 2.04)		0.79 (0.49 to 1.28)
On	265	19,252	1.38	1 (ref)	0.470	1 (ref)	0.669
Off	12	601	2.00	1.42 (0.79 to 2.54)		0.84 (0.45 to 1.54)
On normal dose	8	532	1.50	1.14 (0.57 to 2.32)		0.76 (0.35 to 1.62)
On high dose
Gabapentinoid	239	19,815	1.21	1 (ref)	<0.001	1 (ref)	0.001
Off	46	571	8.06	3.37 (2.44 to 4.65)		1.83 (1.28 to 2.62)

Unadjusted p-values test for differences in mortality rates by co-prescription treatment. High and normal doses are defined in S2 Table.

aAdjusted for sex, year, comorbidity, region, OAT type, OAT period, and, where applicable, benzodiazepine, z-drug, and gabapentinoid exposure. Linear trend was applied to ln(IRR). IRR for on high dose was estimated as on normal dose squared. The deviation from linearity for adjusted models for benzodiazepine: p = 0.4168 and 0.5352 for all-cause mortality and DRP, respectively.

DRP, drug-related poisoning; HR, hazard ratio; IRR, incidence rate ratio; MR, mortality rate; OAT, opioid agonist treatment; PY, person-years of follow-up.

Z-drugs and gabapentinoids

No significant association between co-prescribed z-drugs and ACM was apparent in adjusted analyses (Table 4). Co-prescribed z-drugs showed an overall significant association with DRP but no dose–response effect (adjusted HR for DRP 3.66 [95% CI 1.86 to 7.19], p = 0.001, for normal dose; 1.55 [95% CI 0.59 to 4.06], p = 0.001, for high dose). Gabapentinoid co-prescription was significantly associated with non-DRP but not with DRP (DRP adjusted HR 1.54 [95% CI 0.60 to 3.98], p = 0.37; non-DRP adjusted HR 1.83 [95% CI 1.28 to 2.62], p = 0.001). It was not possible to examine evidence for a dose–response association of mortality with gabapentinoids.

Interactions between benzodiazepine co-prescription and OAT period

There was a strong and significant association between co-prescribed benzodiazepine and increased risk of death that appeared specific to DRP. We found no evidence of any interaction of this effect with treatment period (Table 5).

Table 5

Survival analysis results for interactions between co-prescribed benzodiazepines and OAT period (on or off treatment) in relation to the effect on risk of drug-related poisoning.

OAT period	Co-prescribed benzodiazepine	Deaths	PY	MR (per 100 PY)	Unadjusted		Adjusteda
					HR (95% CI)	p-Value	HR (95% CI)	p-Value
On	Off	24	10,091	0.24	1 (ref)	0.896	1 (ref)	0.997
	On	20	2,914	0.69	2.87 (1.58 to 5.20)	0.001	2.92 (1.60 to 5.33)	0.001
Off	Off	50	6,179	0.81	1 (ref)		1 (ref)
	On	19	764	2.49	3.02 (1.78 to 5.15)	<0.001	2.92 (1.70 to 5.02)	<0.001

Interaction p-value shown in bold.

aAdjusted for sex, year, comorbidity, region, OAT type, OAT period, and, where applicable, benzodiazepine, z-drug, and gabapentinoid exposure.

HR, hazard ratio; MR, mortality rate; OAT, opioid agonist treatment; PY, person-years of follow-up.

Overall association of concurrent prescription with mortality risk

There was no evidence of any beneficial effect of concurrent prescription such as might arise through increased treatment duration. All concurrent prescription was associated with increased risk of all mortality classes (e.g., HR for DRP with concurrent prescription of benzodiazepine 3.34 [95% CI 2.14 to 5.20], p < 0.001; concurrent prescription of z-drug 1.64 [95% CI 1.02 to 2.64]; p < 0.001) (Table 6).

Table 6

Survival analysis showing overall effect of concurrent exposure to benzodiazepines or z-drugs.

Mortality	Concurrent exposure with OAT	Unadjusted		Adjusteda
		HR (95% CI)	p-Value	HR (95% CI)	p-Value
All cause	None	1 (ref)	0.052	1 (ref)	<0.001
	Benzodiazepine	1.22 (1.04 to 1.42)		1.87 (1.55 to 2.25)
	Z-drug	0.97 (0.79 to 1.19)		1.37 (1.09 to 1.72)
DRP	None	1 (ref)	0.001	1 (ref)	<0.001
	Benzodiazepine	1.98 (1.35 to 2.90)		3.34 (2.14 to 5.20)
	Z-drug	1.24 (0.81 to 1.89)		1.64 (1.02 to 2.64)
Non-DRP	None	1 (ref)	0.570	1 (ref)	<0.001
	Benzodiazepine	1.10 (0.86 to 1.40)		1.73 (1.28 to 2.33)
	Z-drug	1.11 (0.82 to 1.50)		1.37 (0.95 to 1.96)

Results for gabapentinoids are not reported because there was no evidence of an effect of increased treatment duration.

aAdjusted for sex; year; comorbidity; region; OAT type; OAT period; off treatment prescription of benzodiazepine, z-drugs, and gabapentinoids; and, where applicable, concurrent prescription of benzodiazepines, z-drugs, and gabapentinoids.

DRP, drug-related poisoning; OAT, opioid agonist treatment.

Sensitivity analyses

Similar results for all main analyses were seen using Poisson regression instead of Cox proportional hazards regression (S4 Table) and when the first treatment episode was excluded (S5 Table). Consideration of competing risk of death from other causes did not materially change our survival analysis of the effects of co-prescription on risk of DRP (S6 Table).

Discussion

Principal findings

We found that the proportion of opioid-dependent patients prescribed a benzodiazepine fell between 1998 and 2014. Nevertheless, in 2014 almost a third of patients on OAT were co-prescribed a benzodiazepine (mainly diazepam). We found strong evidence of substantially increased mortality risk amongst opioid-dependent patients in primary care who were prescribed benzodiazepines. This increased risk was specific to deaths from DRP, occurred during and after opioid substitution treatment, and was higher with doses of benzodiazepines above those recommended in the British National Formulary. Adjustment for measured possible confounding factors did not reduce these estimates of effect.

OAT reduces mortality risk; however, whether prescribed during OAT or in the 12 months after leaving OAT, benzodiazepines increased mortality risk from DRP by approximately 3-fold. We found no evidence of an interaction between the effect of benzodiazepine co-prescription on mortality risk and treatment period. This is probably because most opioid-dependent individuals continue using illicit opioids in the year following the end of a treatment episode [3]. We also found evidence that concurrent prescription of benzodiazepines was associated with longer OAT episodes. In general, longer OAT duration is associated with lower mortality. However, we found no evidence that the longer treatment duration associated with concurrent benzodiazepine prescription led to an overall reduction in mortality [3,7,19].

Z-drugs when co-prescribed with OAT were also associated with increased risk of mortality. This increased risk was seen both for ACM and DRP but showed no evidence of a dose–response effect. Co-prescription of gabapentinoids was associated with increased risk of ACM. This effect was substantially attenuated on adjustment for possible confounding factors.

Comparison with other studies

Studies from North America have shown increased risk of both mortality from overdose and hospital attendance for overdose amongst patients prescribed both benzodiazepines and opioids [20,21]. These studies did not specifically examine effects in opioid-dependent individuals receiving OAT, and they acknowledged their limited ability to infer causality. A previous study from Scotland did examine the effects of benzodiazepine co-prescription amongst patients receiving methadone [10]. In keeping with our own findings, this study found no strong evidence of increased risk of ACM amongst co-prescribed patients, whereas an increase in DRP (HR 4.35 [95% CI 1.32–14.30], p < 0.05) was apparent in adjusted analyses. Only limited information on other patient characteristics was available in the Scottish study, constraining the ability to address confounding; possible dose–response effects were not assessed, and specific effects of concurrent prescription were not examined.

A recent small study from a single practice in London found that a high proportion of patients on OAT received a concurrent prescription of a benzodiazepine [13]. In keeping with our study, this study showed that concurrent prescription increased treatment duration. The overall effect on mortality of this increased treatment was not studied, though the authors suggested this would be positive. A recent Swedish register-based study examined effects on mortality amongst opioid-dependent individuals receiving OAT with co-prescription of benzodiazepines, z-drugs, and pregabalin [22]. This study included considerably fewer patients and had shorter follow-up than our own. Co-prescription of all these drugs was associated with increased risk of mortality, though for benzodiazepines this risk was stronger in relation to ACM than DRP. Dose–response effects were not examined, and adjustment for possible confounding was limited by a relative lack of information on other patient characteristics. This study did not consider effects on treatment duration. Two recent Canadian case–control studies considered the effects of concomitant gabapentinoid prescription amongst patients prescribed opioids for any reason other than palliative care or cancer and found evidence of increased risk of opioid-related death associated with both gabapentin and pregabalin prescription [23,24]. A recent research letter describes an increase in use of gabapentinoids in the US between 2002 and 2015 and also highlights increases in patients being prescribed gabapentinoids, opioids, and benzodiazepines concomitantly [25]. Pharmacovigilance around gabapentinoid prescribing has been increased in the US in response to concerns of harms associated with these medicines [26]. A further study also based on CPRD data also found evidence that prescribing of gabapentinoids increased in the UK between 1993 and 2017 [27].

The proportion of DRP deaths involving gabapentinoid use is increasing in the UK, and in most of these deaths opioid use is also recorded [28]. There is also animal evidence that gabapentinoids enhance the respiratory depressant effect of opioids [29].

Strengths and limitations of this study

To our knowledge, our study is the largest and most detailed to date to examine the question of the effects of prescription of benzodiazepines, z-drugs, and gabapentinoids amongst opioid-dependent patients receiving OAT in primary care. Prospective observational studies based on administrative data provide the most robust means available to study this important clinical issue. Such studies are inevitably prone to bias, particularly confounding by indication. This is also true of our own study, though we adjusted for a wide range of possible confounding factors. Statistical methods based on propensity scores and the use of instrumental variables such as physician prescribing preference have been used in attempts to reduce this bias [30]. We have previously found that the former provide little benefit, as have others [15,31]. We were unable to identify an instrument suitable to allow the latter. We provide the strongest evidence currently available that benzodiazepine co-prescription in opioid-dependent individuals is of itself a cause of increased risk of death from DRP rather than a marker of other patient characteristics that influence risk of death. This evidence is less strong in relation to the increased mortality risk seen with co-prescription of z-drugs and gabapentinoids. Patients receiving OAT may obtain benzodiazepines, z-drugs, or gabapentinoids from non-prescribed sources, and we were unable to measure this. The resulting exposure misclassification would be expected to dilute any apparent mortality increase seen with co-prescription. Individuals in CPRD are broadly representative of the population of England [14]. As the phenomena we investigated are likely to be pharmacologically mediated, we think our findings may be applicable in other countries.

Conclusions and policy implications

Our findings may have implications for clinical policy. Prescription of benzodiazepines to opioid-dependent patients during OAT or whilst still using illicit opioids should generally be avoided, other than when clinical judgement suggests that the benefits of this are likely to be greater than the harms. Clinical guidelines should emphasise this point. We found some evidence that guidelines may be having an impact, as the prevalence of co-prescription fell over our observation period. In contrast, in North America, though the prevalence of co-prescription is lower overall than in the UK, it appears to be increasing [32]. In the UK, some funding for primary care is linked to treatment quality indicators [33]. Benzodiazepine co-prescription with OAT could be used as a quality indicator in this way. Additional educational support to practitioners and psychological support to patients could also be offered in response to evidence of co-prescription. Periodic urine toxicology is widely used in OAT to monitor treatment compliance and could also be used to detect concomitant use of non-prescribed benzodiazepines, and additional support to reduce this offered. Our data provided some evidence that co-prescription of z-drugs in OAT patients may increase DRP risk such that continued recommendation against their co-prescription in guidelines, alongside the additional measures described above for benzodiazepines, seems appropriate. Our data did not show a significant association between co-prescription of gabapentinoids to patients receiving OAT and increased risk of DRP. These analyses were underpowered. Further analyses on the effects of gabapentinoid co-prescription based on large clinical databases are needed and will soon be feasible given increases in the coverage of clinical databases and apparent increases in the prescribing of gabapentinoids.

Summary of analyses and associated commands in Stata.

(DOCX)

Click here for additional data file.

(DOCX)

Click here for additional data file.

(DOCX)

Click here for additional data file.

(DOCX)

Click here for additional data file.

Definition of drug-related deaths.

(DOCX)

Click here for additional data file.

Benzodiazepine and z-drug dose criteria.

(DOCX)

Click here for additional data file.

All-cause, drug-related, and non-drug-related mortality rates for covariates included in the analyses.

(DOCX)

Click here for additional data file.

Poisson regression analyses showing incidence rate ratios for ACM, DRP mortality, and non-DRP mortality according to co-prescribed medications.

(DOCX)

Click here for additional data file.

Adjusted all-cause, drug-related, and non-drug-related mortality incidence rate ratios for benzodiazepine, z-drug, and gabapentinoid exposure excluding first episode.

(DOCX)

Click here for additional data file.

Mortality rates and hazard ratios for DRP by co-prescription exposure to benzodiazepines considering competing risk of death from other causes, z-drugs, or gabapentinoids in people with opioid dependency in primary care.

(DOCX)

Click here for additional data file.

6 Jul 2019

Dear Dr. MacLeod,

Thank you very much for submitting your manuscript "Prescription of benzodiazepines, z-drugs and gabapentoids and mortality risk in people receiving opioid substitution therapy: observational study based on the UK Clinical Practice Research Datalink and Office of National Statistics death records" (PMEDICINE-D-19-02004) for consideration at PLOS Medicine.

Your paper was discussed among the editorial team, evaluated by the guest editors for the special issue, and sent to independent reviewers, including a statistical reviewer. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to invite you to submit a revised version that fully addresses the reviewers' and editors' comments. You will appreciate that we cannot make a decision about publication until we have seen the revised manuscript and your response, and we expect to seek re-review by one or more of the reviewers.

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We hope to receive your revised manuscript within two weeks. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see http://journals.plos.org/plosmedicine/s/submission-guidelines#loc-methods.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

Please let me know if you have any questions. Otherwise, we will look forward to receiving your revised manuscript shortly.

Sincerely,

Richard Turner PhD, for Philippa Berman, MBBS

Senior Editor, PLOS Medicine

rturner@plos.org

-----------------------------------------------------------

Requests from the editors:

In the metadata, please briefly mention the restrictions on access to CPRD data that interested readers would encounter.

Please convert your abstract to PLOS Medicine style. The final sentence of the combined "methods and findings" subsection should quote the study's main limitations.

Please include brief demographic and follow-up details in your abstract; and add p values, where available, alongside CI.

Please begin the "conclusions" subsection of your abstract with "In this study ..." or similar, and adopt the past tense when describing findings.

After your abstract, we will need to ask you to add a new and accessible "author summary" section in non-identical prose. You may find it helpful to consult one or two recent research papers published in PLOS Medicine to get a sense of the preferred style.

Early in your methods section, please state whether or not the study had a protocol or prespecified analysis plan, and if so add the relevant document as a supplementary file. Please highlight all analyses that were not prespecified.

Please briefly state the position regarding ethics approval in your methods section.

The first paragraph of the discussion section of your main text should summarize the study's findings, and we suggest beginning this with "We found that ..." or similar. If the subheading "Principle findings" remains, please adapt it to "Principal findings".

The discussion section should also include a discrete paragraph summarizing the study's limitations. Mitigating factors can be discussed in a subsequent paragraph.

Where you make claims such as "Our study is the largest ...", as you do in the discussion section, please add "to our knowledge" or similar.

Please include p values, where appropriate, alongside CI throughout your main text.

Please substitute "sex" for "gender" throughout your paper, where appropriate.

Throughout your article, please call out references with numbers in square brackets preceding punctuation, e.g. "... are in treatment [1].".

Throughout the paper, please substitute "p<0.001" or an exact value for "p<0.0001".

Please ensure that your reference list conforms with journal style. Up to six author names should be listed, followed by "et al." where appropriate. Journal names should be abbreviated consistently, and all italics should be removed.

Please add a completed checklist for the most appropriate reporting guideline, which we suspect will be STROBE or RECORD, as a supplementary file, and refer to this in the methods section. Individual items should be referred to by section and paragraph number rather than by line or page numbers, as the latter generally change upon publication.

Comments from academic editor:

1. The authors' exclusions result in a cohort that is 25% of the original sample. This has implications for generalizability and should be discussed thoughtfully. Second, I am surprised that 50% of the excluded patients were excluded because they were being prescribed buprenorphine or methadone "for pain relief". This proportion seems large. Compared to other large pharmacoepidemiological databases in the US or Europe, is this what one would expect? Or is there just a lot of misclassification going on? This should also be discussed thoughtfully. Sensitivity analyses would also go a long way toward reassuring readers that the estimated associations are robust.

2. Why did treatment episodes include 1 year after the prescription ended? This seems arbitrary and should perhaps be probed with some sensitivity analyses. (The authors select a different cutoff of 14 days for the BZD/z-drug exposure, reasoning that those medications are associated with a "shorter recommended treatment duration". But recommended treatment duration with methadone/buprenorphine is presumably lifelong. So why 28 days?) In addition, I am a little confused about how this worked with patients who had consecutive episodes. If a new episode was defined as a subsequent prescription >28 days after a previous prescription, what if someone had a prescription 45 days after a previous prescription ended? They would still be within the 1-year window, so they would be classified as "in treatment" and this time in treatment would be associated with the first prescription?

3. The authors do not appear to have made any effort to ensure that treatment episodes (of OST or the co-prescribed BZD/z-drug/gabapentinoids) are "new" treatment episodes.

4. Please justify in the text the use of linear regression (rather than, say, Poisson, negative binomial, two-part models, etc) in the analysis of treatment duration. The authors do not appear to have used robust standard errors so the estimates are unlikely to be robust to a wide range of distributional assumptions.

5. After reading through the methods several times, it is still a little unclear to me exactly what models were used here:

(a) "Where a main effect of co-prescription was suggested by these analyses we investigated whether this effect was modified by an interaction with OST treatment period (i.e. on or off treatment)."

(b) "For benzodiazepine or z-drug dose effects (2 degrees of freedom), a linear trend was also fitted to examine the effect of dose, with “high” doses ascribed based on a daily dose above the maximum recommended in the BNF (see Web table 1)."

(c) "We then assessed the overall effect of concurrent prescription (allowing for any effect on treatment duration) on mortality risk, using Cox proportional hazards survival analysis."

It would be helpful if the regression equations corresponding to these sentences were included (eg., in an Appendix).

6. The specificity of the findings is unclear. This would be strengthened by the use of an ancillary analysis in which comparable analyses were done with other sedating medications that are commonly co-prescribed in this setting (eg. antidepressants, antipsychotics) as well with other non-sedating medications that are commonly co-prescribed but would not be expected to increase DRP deaths (eg., NSAIDs). This type of analysis is commonly referred to as a "falsification test" (eg., Prasad & Jena, JAMA 2013;309:241-2) or the use of "negative controls" (eg., Arnold et al. Epidemiology 2016;27:637–641)).

MINOR COMMENTS:

7. I would recommend using the term "opioid agonist treatment", or some other non-stigmatizing term, rather than "opioid substitution therapy". See Wakeman SE. J Addict Med 2017;11: 1–2.

8. The authors indicate they studied 10 benzodiazepines, 3 z-drugs, and 2 gabapentinoids. These should be named.

9. I understand that the CRPD initial cohort received a prescription between 1st January 1998 and 31st July 2014. In the Results section, the authors should provide the respective dates for the final analytic sample.

10. More detail should be provided in the summary statistics paragraph: "These data contained 29,540 treatment episodes: 17,391 methadone (median treatment duration, XXXX; range, XX-YY; interquartile range, XX-YY), 9180 buprenorphine (median treatment duration, XXXX; range, XX-YY; interquartile range, XX-YY) and 2969 episodes relating to more than one drug. Total follow-up time was 36,126 person years during which there were 17,139 episodes of benzodiazepine treatment, XXXX episodes of z-drug treatment, and YYYY episodes of gabapentinoid treatment. Benzodiazepines were prescribed during 10,022 OST episodes in 7059 patients. Z-drugs were prescribed during 4279 OST episodes in 2822 patients. Gabapentoids were prescribed during 1827 OST episodes in 1281 patients."

11. The authors indicate "mean follow-up per patient of 3.4 years" but this should be expressed as "median (IQR; range)".

12. Throughout the manuscript the authors refer to "gabapentoids". These should be replaced with "gabapentinoid".

13. The literature review is incomplete. Several studies of co-prescribing in this area have been published, but the authors neither cite nor interact meaningfully with this literature (eg., discussing the strengths of their study relative to the others).

Gomes T et al. PLoS Med. 2017 Oct 3;14(10):e1002396.

Gomes T et al. Ann Intern Med 2018;169:732-4

Johansen. JAMA Intern Med. 2018 Feb; 178(2): 292–294.

Peckham et al. Drug Saf 2018;41(2):213-228

Montastruc et al. JAMA. 2018;320(20):2149-2151.

Comments from the reviewers:

*** Reviewer #1:

The authors present a cohort study using administrative data from 1998 - 2014 from the United Kingdom examining over 12,000 patients receiving opioid substitution treatment (OST) and their duration of OST use and mortality in relation to use of benzodiazepines, z-drugs (zaleplon, zolpidem, and zopiclone), and gabapentoids. The findings of the study most strongly suggest an increased duration of OST associated with concurrent benzodiazepine use but also a significantly increased risk of mortality relative to non-use. A few comments and questions are noted below:

1. It's unclear whether the cohort was a 'new user' cohort (i.e. relatively new use of OST). If it was a new user cohort, please include the definition of 'new user'. If not, would it be possible to provide a sense of how long they had being OST prior to the cohort entry / start date? As the authors may be aware, prevalent user designs may be subject to various sources of bias (https://www.ncbi.nlm.nih.gov/pubmed/14585769). On a related note, how was the cohort entry date / start date defined? Was it the first instance of OST use observed during the study period? Please explicitly define.

2. For the exposures of interest, these drugs are often used on an as needed (prn) basis and 14 days was assumed to be the minimum gap between prescriptions for purposes of estimating use (page 8). Was this an arbitrary decision? How was occasional use handled (e.g. some people may spread the use of 30 pills of benzodiazepines over 90 days)?

3. Was there a single primary endpoint of interest? There were numerous exposures, definitions of exposure duration, and outcomes. As a result, quite a few statistical tests were conducted. How was the issue of multiple hypothesis testing handled?

4. Were comorbidities assessed prior to cohort entry (i.e. the start date) or were 'new' comorbidities identified during follow-up included (page 8)? The use of time varying covariates in this regard is a bit confusing. Please explain.

5. The definition of death involves multiple data sources (page 9). Has the outlined approach to defining death been validated? If so, how accurate is death using the approaches taken? Also, have there been any validation studies of the outcome of drug-related death poisoning?

6. Linear regression was used to examine the association between concurrent use of an exposure of interest and OST. Was a time-to-event analysis such as a survival model considered for this assessment? Were competing risks considered?

7. It would be helpful to more fully describe how selection bias was handled in this study. For example, patients not using a benzodiazepine may be considerably different than those who use a benzodiazepine. Such information has not been provided. Use of time-varying covariates in a survival model does not obviate selection bias. Further, timing of as needed drugs such as benzodiazepines may reflect critical events in a patient's life that lead to the use of such medications. The critical events may have a stronger effect on the risk of mortality than the actual exposure being assessed (i.e. the exposure to a drug such as a benzodiazepine at a particular time may mask a critical event that may not be captured by the available data). It would be helpful for the authors to highlight how such temporal confounding was handled, if at all.

*** Reviewer #2:

This is a well-conducted epidemiological study on the associations between prescription of benzodiazepines, z-drugs and gabapentoids and mortality risk in people receiving opioid substitution therapy based on UK's CPRD primary care database and death records from the Office of National Statistics. The study design, datasets, statistical methods and analyses, presentations (tables and figures) and interpretations of results are mostly adequate and of a good standard. However, there are still a few important statistical issues needing attention.

1) The key conclusion of the study is that Co-prescription of benzodiazepine is associated with a three-fold increased risk of DRP (drug-related poisoning mortality) as shown in table 3. This was done by the Cox model. However, in this case all cause mortality becomes a competing risk that could potentially hinder or modify the chance of DRP, therefore it would be appropriate to estimate the risk of DRP within the competing risk framework. Unfortunately the authors didn't address this issue in the paper.

2) Table 2 on treatment durations. In the second column, median duration was presented which suggested that the distributions of the duration were not normally distributed. Is this right? If so, then these durations as outcome should be transformed (e.g., log-transformation) into normal variables before they can be used in linear regressions. Can authors check, report and correct as appropriate?

3) Table 1. Follow-ups (years) should be summarised as median and IQR rather than mean and SD as they are skewed data.

*** Reviewer #3:

This paper is important as it addresses a question that has, and should be, high on the agenda for policy makers and clinical practitioners managing people who have used drugs.

It has been difficult in the past to unravel the interactions of drugs and their, often fatal, effects. Using administrative data is an innovative way to provide some clarity and although the authors acknowledge the incomplete nature of their enquiry into this complex area they have opened up some very important leads and articulated the research questions that need to be answered.

Are benzodiazepines prescribed because methadone doses are too low? This is a common problem in clinical practice and diazepam particularly, but also other sedatives such as Z drugs, are prescribed or obtained illegally to supplement the effects of inadequate opiate substitute treatment.

People on <20mg methadone excluded. This clearly is a dose way below the recognized therapeutic level but isn't 30 or 40mg in the same range?

Are these patients using non prescribed BDZs. How is it known that the BDZ present at death isn't non prescribed?

Mortality showed regional variation, is there any explanation for this variation and is it associated with coprescribing?

With evidence of a dose response effect can it ne made clearer how strong this was? Is it not true that there would be and expectation that it would be significant?

Why not include other drugs with an equal potential for additional respiratory depression such as SSRIs and those with QRS complex impairment? The former are widely used to supplement opiate substitute treatment and to complement the management of the symptomatology around a lifestyle of opiate use.

What is the mechanism for association of co prescription of benzodiazepines and methadone and DRP? Is it possible to comment or even speculate on whether or not this is cumulative respiratory depression or if, as had been suggested, that co prescribed drugs has a behavioural effect in inducing confusion/recklessness or misjudgement of doses?

We already know that deaths are associated with multiple drugs at toxicology but the causation is complex and less well understood.

Research is needed into the reasons why overdose can occur when it hasn't before and the role of other factors such as COPD and GI (liver primarily) disorders, reference for example;

Lu Gao J. Roy Robertson Sheila M. Bird Non drug-related and opioid-specific causes of 3262 deaths in Scotland's methadone-prescription clients, 2009-2015 Drug and Alcohol Dependence Volume 197, 1 April 2019, Pages 262-270

Is this subset a higher risk group with different characteristics? It is recognised that drug users are not a homogeneous group with varying behaviour patterns suggesting an increased risk

Mortality risk during and after opioid substitution treatment: systematic review and meta-analysis of cohort studies

BMJ 2017; 357 doi: https://doi.org/10.1136/bmj.j1550 (Published 26 April 2017)

Cite this as: BMJ 2017;357:j1550

The conclusion that benzodiazepines should not be coprescribed is a sweeping statement based on results that don't necessarily take into account the best intentions of clinical services that are trying to balance the needs for symptomatic treatment of a distressed caseload.

Avoiding one drug (such as diazepam in this case) often leads prescribers to look for an alternative, which may be equally toxic.

The other, perhaps more recent, problem is the escalation in availability of synthetic or illegally sourced benzodiazepines. Supplies of etizolam and alprazolam have distorted the marked and caused enormous self medication problems leading to increased risk of death. One response to this has been to accommodate these needs by prescribing benzodiazepines.

Perhaps the underlying problem is inadequate dosing of methadone or buprenorphine and it might be useful to say this.

***

Any attachments provided with reviews can be seen via the following link:

[LINK]

30 Aug 2019

Submitted filename: PLOS Med_response to reviewers.docx

Click here for additional data file.

30 Sep 2019

Dear Dr. MacLeod,

Thank you very much for re-submitting your manuscript "Prescription of benzodiazepines, z-drugs and gabapentoids and mortality risk in people receiving opioid substitution therapy: observational study based on the UK Clinical Practice Research Datalink and Office of National Statistics death records" (PMEDICINE-D-19-02004R1) for review by PLOS Medicine.

I have discussed the paper with my colleagues and the academic editor and it was also seen again by reviewers. I am pleased to say that provided the remaining editorial and production issues are dealt with we are planning to accept the paper for publication in the journal.

The remaining issues that need to be addressed are listed at the end of this email. Any accompanying reviewer attachments can be seen via the link below. Please take these into account before resubmitting your manuscript:

[LINK]

Our publications team (plosmedicine@plos.org) will be in touch shortly about the production requirements for your paper, and the link and deadline for resubmission. DO NOT RESUBMIT BEFORE YOU'VE RECEIVED THE PRODUCTION REQUIREMENTS.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

In revising the manuscript for further consideration here, please ensure you address the specific points made by each reviewer and the editors. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We expect to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

If you have any questions in the meantime, please contact me or the journal staff on plosmedicine@plos.org.

We look forward to receiving the revised manuscript by Oct 07 2019 11:59PM.

Sincerely,

Richard Turner,PhD

Senior Editor

PLOS Medicine

plosmedicine.org

------------------------------------------------------------

Requests from Editors:

Abstract – where there are 95% values for quantifiable data, please also add p values (also in main text and tables, where necessary). Even though the abstract does not exceed the requested 500 words, that doesn’t need to be a target and I find it quite long and wonder if it could be trimmed for conciseness?

Abstract – as this isn’t a trial, please be cautious to avoid causal language…for example in the abstract “In this study, co-prescription of benzodiazepine was associated with increased risk of DRP in opioid dependent individuals both on and off OAT. Co-prescription of z-drugs and gabapentinoids was also associated with increased mortality risk, however evidence for causality was less strong.” Please avoid the word evidence and also vague language…an association is either significant or not and if not, remove – or say no association. Also line 312 (causal effect).

Author summary – thanks you for adding this; it needs to be in a bullet point format.

Line 259 – “was not strongly associated with age” – please be specific in terms of significance of association – it is, or isn’t.

Line 423 (should) please say ‘may be ‘ and would bit be more accurate to say applicable in other countries instead of externally valid?

Line 428 (Our findings have clear implications for clinical policy) – please say ‘may have’ and remove ‘clear’

Line 442 (Our data did not provide strong evidence that co-prescription of gabapentinoids to patients receiving OAT causes increased risk of DRP) – please be clear – it did, or didn’t…

STROBE – thanks for providing. Instead of METHODS THIRD TO FIFTH PARAGRAPH (for example), please say Methods Paragraphs 3-5 and adjust throughout. Also please remove the word ‘Done’.

Comments from Reviewers:

Reviewer #1: While the authors have dutifully responded to the many comments of the editors and reviewers, I am not convinced that the validity of the study findings are high enough to warrant publication in PLOS Medicine. Several outstanding issues have either not been addressed adequately or simply cannot be addressed well given the nature of the study design:

1. Generalizability: The academic editor raised the issue of a very high exclusion rate (i.e. a study cohort that is 25% of the original sample). While the authors provide some explanations for the exclusions, the level of misclassification is unclear and a validated approach to arriving at the study sample would have been extremely helpful. Simply accepting a possibly high level of misclassification doesn't seem adequate. Some more 'homework' needs to be done in this regard.

2. Establishing a 'new user' cohort: New user designs are known to minimize bias, however the authors acknowledge that this study design is virtually impossible with the data sources available. It may be that 'chronic' OAT users may be more likely to use one of the study drugs than another. The use of time-dependent exposure assessment may help to mitigate some of the potential biases here, but it still may be problematic.

3. Temporal confounding and specificity of findings: Perhaps most concerning is the possibility of temporal confounding as it relates to protopathic bias. The issue here is that the initiation of a drug (e.g. a benzodiazepine) may 'unmask' an underlying problem (either related or unrelated to the drug being initiated) that then leads to an outcome such as death or overdose. The use of 'falsification tests' or 'negative controls', as suggested by the academic editor may help understand this issue better. The use of antidepressants / antipsychotics and NSAIDs was suggested by the academic reviewer. The authors did not examine the effects of antidepressants / antipsychotics since they felt there may be considerable confounding by indication. I suppose, perhaps to a lesser extent, the same argument could be made for benzodiazepines. The authors did, however conduct such analyses using antimicrobials and NSAIDs. The findings highlight the issue of bias as statistically significant associations between antimicrobials and several of the outcomes of interest were observed. The risk estimates for the association between NSAIDs and the outcomes of interest were also elevated but not statistically significant. These findings raise serious concerns about the validity of the observed association between benzodiazepines and the selected outcomes. The authors seem to dismiss the associations between antimicrobials and the selected outcomes by suggesting these are driven by confounding by indication and fail to consider that this may also be what's driving the observed associations between benzodiazepines and the selected outcomes.

For the above key reasons, I question the validity of the findings of this study.

Reviewer #2: Thanks authors for their effort to improve the manuscript. The response and revision are mostly satisfactory. However, still one minor issue remaining. In the revised table 1, could authors please remove the row with Mean follow-up years (SD)? For skewed data, it doesn't make sense and is inadequate to use mean and SD. Median and IQR are sufficient and adequate in this case.

Reviewer #3: I am happy that the authors have considered and responded to my suggestions and concerns. I understand that many of my comments were outwit the range of the study and my expectations were, for that reason, unrealistic. I do think that more research is required to illuminate some of the unanswered questions but that is clearly beyond the scope of this study.

I am still worried about the statement in the abstract and text that prescribers should avoid prescribing benzodiazepines completely. This is an unrealistic expectation in an era of enhanced Harm Reduction when availability of illegal benzodiazepines and self medication is at an all time high. I accept that BDZs and other coprescribed drugs may increase the risk of death but there must be a balance of risks present (as there is with many drugs that are commonly prescribed in patients with serious disease). For an influential group to make a uncompromising statement is likely to cause unintended consequences. Guidelines will cite it and conclude that prescribing any BDZ is outwit normal practice and any strategy to reduce harm by prescribing a (relatively safer) benzodiazepine will be inhibited.

Could the authors draw attention to the risk they have identified and suggest less binary solution?

Any attachments provided with reviews can be seen via the following link:

[LINK]

21 Oct 2019

Submitted filename: response to reviewers PLOS MED 101019.docx

Click here for additional data file.

24 Oct 2019

Dear Dr. MacLeod,

On behalf of my colleagues and the academic editor, Dr. Alexander Tsai, I am delighted to inform you that your manuscript entitled "Prescription of benzodiazepines, z-drugs and gabapentoids and mortality risk in people receiving opioid substitution therapy: observational study based on the UK Clinical Practice Research Datalink and Office of National Statistics death records" (PMEDICINE-D-19-02004R2) has been accepted for publication in PLOS Medicine.

PRODUCTION PROCESS

Before publication you will see the copyedited word document (in around 1-2 weeks from now) and a PDF galley proof shortly after that. The copyeditor will be in touch shortly before sending you the copyedited Word document. We will make some revisions at the copyediting stage to conform to our general style, and for clarification. When you receive this version you should check and revise it very carefully, including figures, tables, references, and supporting information, because corrections at the next stage (proofs) will be strictly limited to (1) errors in author names or affiliations, (2) errors of scientific fact that would cause misunderstandings to readers, and (3) printer's (introduced) errors.

If you are likely to be away when either this document or the proof is sent, please ensure we have contact information of a second person, as we will need you to respond quickly at each point.

PRESS

A selection of our articles each week are press released by the journal. You will be contacted nearer the time if we are press releasing your article in order to approve the content and check the contact information for journalists is correct. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact.

PROFILE INFORMATION

Now that your manuscript has been accepted, please log into EM and update your profile. Go to https://www.editorialmanager.com/pmedicine, log in, and click on the "Update My Information" link at the top of the page. Please update your user information to ensure an efficient production and billing process.

Thank you again for submitting the manuscript to PLOS Medicine. We look forward to publishing it.

Best wishes,

Richard Turner, PhD

Senior Editor

PLOS Medicine

plosmedicine.org