Assessing the impact and cost-effectiveness of needle/syringe provision and opiate substitution therapy on hepatitis C transmission among people who inject drugs in the United Kingdom: analysis of pooled datasets and economic modelling

Selection criteria

We included all observational (prospective and retrospective cohorts, cross-sectional surveys and case–control studies) and experimental studies [randomised controlled trials (RCTs)] that measured exposure to either intervention versus no intervention or a reduced exposure, and that reported HCV infection incidence as an outcome.

We included cross-sectional surveys if they included a serological measure of recent infection [e.g. through positive ribonucleic acid (RNA) results on antibody-negative samples]. We excluded cross-sectional studies (including serial cross-sectional studies) reporting HCV infection prevalence only. We excluded studies relying on self-reported data for the outcome.

Data collection and analysis

Two reviewers screened all title and abstracts, and disagreements were resolved following discussion. Full texts were screened by two people to assess eligibility. Data were extracted independently by two people and then checked for consistency. Full-text papers in languages other than English were translated by individuals fluent in those languages.

Meta-analysis was conducted using random-effects models, pooling univariable and multivariable models separately. We examined heterogeneity with the I²-statistic and explored reasons for heterogeneity using univariable random-effects metaregression. We examined heterogeneity with the I² and τ² statistics, and explored reasons for heterogeneity using univariable random-effects metaregression to evaluate the impact of the following covariates on intervention effect: geographical region of study; recruitment setting (community based or treatment); percentage of female participants; main drug injected; type of NSP; frequency of injecting; dose, duration and adherence to NSP/OST (i.e. continuous or interrupted treatment); and study design. There was insufficient information to assess the impact of adherence to NSPs/OST (i.e. continuous or interrupted treatment). We used Stata^® version 14.0 (StataCorp LP, College Station, TX, USA) in all analyses and transferred the data into RevMan software version 5.3 (Cochrane, The Nordic Cochrane Centre, Copenhagen, Denmark).

Grading of evidence

Risk of bias for all studies was assessed using the ACROBAT-NRSI tool (A Cochrane Risk Of Bias Assessment Tool: for Non-Randomized Studies of Interventions), which is in development by the Methods Groups of Cochrane. 28 We assessed the overall quality of the evidence for the primary outcome using the Grading of Recommendation, Assessment, Development and Evaluation (GRADE) system. The GRADE Working Group developed a system for grading the quality of evidence,29–32 which takes into account issues related not only to internal validity but also to external validity, such as the directness of results. In particular, they provide key information concerning the quality of evidence, the magnitude of effect of the interventions examined and the sum of available data on the main outcomes.

The GRADE system uses the following criteria for assigning grades of evidence:

• High – we are very confident that the true effect lies close to that of the estimate of the effect.

• Moderate – we are moderately confident in the effect estimate. The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different.

• Low – our confidence in the effect estimate is limited. The true effect may be substantially different from the estimate of the effect.

• Very low – we have very little confidence in the effect estimate. The true effect is likely to be substantially different from the estimate of effect.

Grading is decreased for the following reasons:

• serious (–1) or very serious (–2) study limitation for risk of bias

• serious (–1) or very serious (–2) inconsistency between study results

• some (–1) or major (–2) uncertainty about directness (the correspondence between the population, the intervention or the outcomes measured in the studies actually found and those under consideration in our systematic review)

• serious (–1) or very serious (–2) imprecision of the pooled estimate (–1)

• publication bias strongly suspected (–1).

Grading is increased for the following reasons:

• Strong evidence of association – significant relative risk of > 2 (< 0.5) based on consistent evidence from two or more observational studies, with no plausible confounders (+1). Very strong evidence of association – significant relative risk of > 5 (< 0.2) based on direct evidence with no major threats to validity (+2).

• Evidence of a dose–response gradient (+1).

• All plausible confounders would have reduced the effect (+1).

Study selection

Figure 1 shows the number of studies identified, reviewed and selected and the reasons for exclusion for both searches.

FIGURE 1.

Flow chart of included studies. CDAG, Cochrane Drug and Alcohol Register; CENTRAL, Cochrane Central Register of Controlled Trials; CINAHL, Cumulative Index to Nursing and Allied Health Literature; CLib, Cochrane Library; DARE, Database of Abstracts of Reviews of Effects; HTA, Health Technology Assessment; NHS EED, NHS Economic Evaluation Database; WOS, Web of Science. Reproduced from Platt et al. ,26,27 which are published open access under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

We identified 21 papers that directly included measures of the impact of exposure to either OST or NSPs on HCV infection transmission. In addition, we identified 11 eligible prospective studies that measured HCV infection incidence and contacted authors of these articles. Of these, unpublished data were obtained from seven cohort studies in Montreal, Canada;33 Baltimore, USA;34 San Francisco, USA;35 London, UK;36 and Sydney and Melbourne, Australia. 37,38 The full text of 141 papers was reviewed. A total of 19 papers were included, and 120 papers were excluded for the following reasons: no HCV infection incidence data (n = 55); no measure of intervention exposure (n = 34); contain no primary data (n = 20); the sample was not PWID (n = 1); all the sample were recruited from the intervention (n = 9); and the article could not be obtained in its original language (Japanese) (n = 1).

In total, we included 21 published studies14,39–58 and seven unpublished studies33–37,59,60 comprising 1827 HCV incident infections and 8789.7 person-years of follow-up. Overall HCV infection incidence ranged between 0.09 and 42 cases per 100 person-years across the studies.

Excluded studies

A total of 101 studies (104 articles) were excluded in which there was no outcome of interest assessed (43 studies); no intervention of interest (32 studies); no comparison of interest (all participants on OST) (nine studies); no outcome and no intervention of interest (11 studies); no outcome and no comparison of interest (four studies); or when the study was an editorial or overview (two studies).

Description of studies

Effects of interventions

Current use of opioid substitution therapy versus no current opioid substitution therapy

Of the 28 studies, we pooled data from a total of 17 studies to assess the impact of current OST use on HCV infection incidence,14,39,40,45–47,49,52,53,55,56,58 including five unpublished estimates. 33,36,37,59,60 Current use of OST was defined as reporting use of OST within the past 6 months (yes or no)33,46,49,58,60 or as reporting use of OST either within 6 months or > 6 months ago,36 use of methadone at the time of survey,39,47,52,59 continuous use of OST throughout the follow-up period,40,45,53 with one study defining continuous use as daily use of methadone (any dosage) in the past 6 months14 or in the past month. 37 One study used a 3-month time frame to measure the use of opioid agonist therapy maintenance treatment. 55 One study measured current use of buprenorphine. 34

All comparison groups were made against no intervention. All the included studies were longitudinal studies, with the exception of one case–control study49 and two cross-sectional surveys. 47,59

The 17 studies included a minority of women (range 3–53%), a high proportion of the samples had experience of prison (range 18–60%) and homelessness (9–70%), and use of stimulants ranged between 1% and 51% across the studies. A total of 1073 HCV infection incident cases were included over 4990.29 person-years of follow-up.

Of these 17 studies, 12 presented adjusted estimates on which the primary analyses were focused. Adjusted estimates controlled for potentially confounding effects of the following factors: duration of injection; frequency of injection;33,36,60 area of residence, homelessness, sharing injecting equipment or needles;2 sex, geographical region, use of condoms, injection of cocaine, duration of injection and sharing injecting equipment;45 duration of injection, frequency of injection and age of whole cohort;34 unstable housing, cocaine, heroin or methamphetamine injection, cohort of recruitment, year of recruitment and follow-up time;46 survey year, homelessness, stimulant injection and duration of injection;47 sex, age, duration of drug use and injection of cocaine;49 age, duration of injection, sex, ethnicity, homelessness or prison in the past 3 months;55 sex, ethnicity, age, frequency of injecting, sharing needles/syringes, not receiving OST while reporting opioid use,58 injected at follow-up, pooled money to buy drugs, injection with used needles and backloading. 53

Random-effects meta-analysis of multivariable estimates shows that OST was associated with a 50% reduction in the risk of HCV infection [rate ratio (RR) 0.50, 95% CI 0.40 to 0.63] with little evidence of heterogeneity between studies (I² = 0; p = 0.889) (Figure 2).

FIGURE 2.

Impact of current use of OST vs. non-OST use on HCV infection incidence from adjusted analyses. Reproduced from Platt et al. ,26,27 which are published open access under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Sensitivity analyses

This effect was increased when excluding estimates from four unpublished data sources33,34,36,53 (RR 0.42, 95% CI 0.31 to 0.58) and little evidence of heterogeneity between studies (I² = 0%; p = 0.96). The effect was maintained when limiting the analysis to exclude all unpublished data sets as well as one study that was judged to be at critical risk of bias49 (RR 0.43, 95% CI 0.31 to 0.59; I² = 0%; p = 0.93). The effect was slightly reduced when limiting the analysis to exclude two studies47,49 that reported baseline measures of effect only (RR 0.51, 95% CI 0.40 to 0.65; I² = 0.0%; p = 0.807) and two studies that reported incident RRs only34,39 (RR 0.51, 95% CI 0.40 to 0.64; I² = 0%; p = 0.853). These data are not shown.

A random-effects meta-analysis of 16 studies that presented univariable estimates suggests that OST was associated with a 40% reduction in the risk of HCV infection (RR 0.60, 95% CI 0.47 to 0.76), with only moderate evidence of heterogeneity between studies (I² = 31.7; p = 0.09) (Figure 3).

FIGURE 3.

Impact of current use of OST vs. no OST on HCV infection incidence findings from unadjusted analyses. Reproduced from Platt et al. ,26,27 which are published open access under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Metaregression

Based on univariable metaregression of unadjusted estimates, we found no evidence that effectiveness varied by other covariates including geographical location. We did find evidence of differential impact in the proportion of female participants in the sample. With each 10% increase of female participants in the sample, the effect of intervention exposure was reduced (RR 1.59, 95% CI 1.13 to 2.29) (see Table 15).

History of opioid substitution therapy

Three studies published unadjusted estimates of history of OST use, comprising 115 HCV infection cases over 511.6 person-years and from three prospective cohorts. 51,56,57 One study did not define the time frame and was coded as past experience of OST. 56

Three studies published unadjusted estimates of interrupted OST use. 40,46,53 Two of these studies were prospective cohorts and one was retrospective, and a total of 200 HCV infection cases were included over 2273.8 person-years. Interrupted OST use was defined as use of OST at baseline but not at follow-up,46 or leaving OST at least once during follow-up. 40,53 One prospective cohort study comprising 149 HCV infection cases over 680 person-years examined OST for detoxification,55 and two studies measured high dosage (≥ 60 mg) or low dosage (1–59 mg) of methadone for daily use14 or use some time in the past 6 months. 33 Both these studies were prospective cohorts and included 148 HCV infection cases over 598.6 person-years.

A random-effects meta-analysis showed no impact among studies measuring historical use of OST (RR 0.81 95% CI 0.52 to 1.27) or among those measuring interrupted use (RR 0.80 95% CI 0.57 to 1.10). 51,56,57 The one study measuring the impact of OST used for detoxification was not associated with reduced HCV infection risk acquisition (RR 1.45, 95% CI 0.79 to 2.66). 55 High dosage with OST was associated with a reduction of HCV infection acquisition (RR 0.52, 95% CI 0.29 to 0.94) but low dosage was not (RR 0.85, 95% CI 0.44 to 1.65) (Figure 4). 14,33

FIGURE 4.

Impact of other modes of OST on HCV infection risk acquisition. Reproduced from Platt et al. 26 This is an open access article under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

High coverage versus non-attendance or lower coverage

We pooled data from seven studies that reported unadjusted measures of high NSP exposure and HCV infection incidence,14,42,44,47,48 including two unpublished data sets. 33,59 High NSP coverage was defined as obtaining 100% of needles and syringes from a safe source33 or reporting ≥ 100% of injections covered by a clean needle or syringe,14,44,59 or ≥ 200% of injections covered by a clean needle/syringe. 47 Other measures of high coverage were defined as regular attendance at least once per week at a NSP48 or obtaining most of all needles/syringes from a NSP in the past 6 months. 42 The seven included studies consisted of four prospective cohorts14,33,42,48 and three cross-sectional surveys,44,47,59 comprising 641 HCV infection cases over 1015.51 person-years.

The primary analysis focused on seven studies that measured high-level uptake of NSP coverage. This primary analysis compared the effect of high-level NSP coverage versus no intervention33,42,48 or lower levels of coverage. 14,44,47,59 These seven studies included a median of 27% of women (range 23–38%), a high proportion of the samples had experience of prison (range 26–60%) and homelessness (range 2–58%), and use of stimulants ranged between 17% and 63% across the studies.

A random-effects meta-analysis suggested no evidence of high coverage of NSPs associated with a reduction in the risk of HCV infection (RR 0.77, 95% CI 0.38 to 1.54) with evidence of high heterogeneity between studies. (I² = 78.8; p < 0.001) (Figure 5).

FIGURE 5.

Impact of high NSP coverage vs. low or no NSP coverage on HCV infection risk acquisition from unadjusted analysis. Reproduced from Platt et al. ,26,27 which are published open access under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Sensitivity analyses

This effect remained the same when excluding the unpublished data sets (RR 0.71, 95% CI 0.23 to 2.19; p < 0.001). 33,59 The effect was also maintained when limiting the analysis to a subset of four studies that excluded three studies assessed to be at critical risk of bias or that were unpublished data sets (RR 0.71, 95% CI 0.17 to 2.98), with evidence of within-study heterogeneity (I² = 89.6%, p < 0.001). 14,33,59 The effect was increased when we excluded studies that reported only incident rate ratios (RR 0.78, 95% CI 0.35 to 1.74; I² = 80.3%; p < 0.001). 14 The effect was further decreased when we excluded studies that reported baseline measures only (RR 1.26, 95% CI 0.55 to 2.93; I² = 87.0%; p < 0.001). 44,47,59 Limiting the analysis to a further subset of four studies that adjusted for confounders, the effect remained the same (RR 0.79, 95% CI 0.39 to 1.61; I² = 77%, p = 0.022; τ² = 0.4482) (data not shown).

Metaregression

Based on univariable metaregression analyses, we found some evidence that the effectiveness of high NSP coverage varied according to geographical region. After removing studies from North America, high NSP coverage in Europe was associated with a 54% reduction in HCV infection acquisition risk (RR 0.44, 95% CI 0.24 to 0.80) with less heterogeneity (I² = 12.3%; p = 0.337), whereas in North America it remained insignificant (RR 1.58, 95% CI 0.57 to 4.42, I² = 89.9%; p < 0.001).

Based on univariable metaregression analyses, the differential impact of NSPs according to geographical region remained, with studies from North America having less impact (ratio of RRs 3.73, 95% CI 0.95 to 14.7; p = 0.06). There was no differential impact by recruitment site (p = 0.89), by proportion of participants reporting stimulant use, homelessness, injection of stimulants or sex. (These data are presented in Appendix 1, Table 16.)

Low-level coverage of needle and syringe programmes versus no needle and syringe programme coverage

Ten studies reported unadjusted measures of low-level NSP coverage and HCV infection incidence. Eight were prospective cohorts14,34,35,42,43,54,58,60 and one was a case–control study. 41 A total of 531 cases were included in the analyses over 1617 person-years. One prospective cohort was dropped because it did not report 95% CIs around the effect estimate or the number of new HCV infection cases in international and comparison groups required to estimate it. 50

A random-effects meta-analysis showed no evidence of an intervention effect of low NSP coverage on HCV infection risk acquisition with moderate levels of heterogeneity (RR 1.41, 95% CI 0.95 to 2.09; I² = 62.3%; p = 0.007; τ² = 0.19) derived from nine studies with a total of 3414 participants (Figure 6).

FIGURE 6.

Impact of low NSP coverage on HCV infection risk acquisition. Reproduced from Platt et al. 26 This is an open access article under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Combined needle and syringe programmes with opioid substitution therapy versus low or no needle and syringe programme coverage and no opioid substitution therapy

A total of four studies reported combined exposure to both NSPs and OST,14,44,47 including one unpublished data set. 33 These categories were defined as high coverage (≥ 100% or ≥ 200%) of injections using clean needles or syringes14,44,47 or obtaining 100% of needles and syringes from a safe source33 plus current use of OST at the point of survey44,47 or within the past 6 month,33 or daily use of methadone during the past 6 months. 14 OST use and low coverage (< 100% of injections covered by a clean needle or syringe) was reported by three studies. 14,44,47 A total of 518 HCV infection incident cases were included in the analysis examining high NSP coverage and 449 for low NSP coverage. Only one study reported the number of person-years. 14 A random-effects meta-analysis showed that combined use of OST and high coverage of NSP was associated with a 71% risk reduction in HCV infection acquisition (RR 0.29, 95% CI 0.13 to 0.65). The effect of exposure to OST and low coverage of NSP was less and non-significant (RR 0.76, 95% CI 0.44 to 1.33) (Figure 7).

FIGURE 7.

Combined OST and high-coverage NSPs. Reproduced from Platt et al. ,26,27 which are published open access under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Publication bias

A funnel plot of 13 estimates (12 studies) suggested no evidence of publication bias in studies of current OST exposure (Figure 8). A funnel plot of nine estimates (eight studies) suggested no evidence of publication bias in studies of high NSP coverage (Figure 9).

FIGURE 8.

Funnel plot with pseudo-95% confidence limits assessing publication bias in 12 studies of current OST exposure. a, Unpublished data sets. IRR, incident rate ratio. Reproduced from Platt et al. ,26,27 which are published open access under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

FIGURE 9.

Funnel plot with pseudo-95% confidence limits assessing publication bias in 15 studies of high NSP exposure. The central line is plotted at the fixed-effect summary effect (RR 0.98, 95% CI 0.0.75 to 1.28). a, Unpublished data sets. IRR, incident rate ratio. Reproduced from Platt et al. ,26,27 which are published open access under the terms of the Creative Commons Attribution-Non-Commercial Licence, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Overall completeness and applicability of evidence

There is a substantial body of observational evidence that reviews the effectiveness of NSPs and OST in reducing HCV infection acquisition among PWID. The majority of evidence was identified in North America and Western Europe. Only one study was identified from China51 and no studies were identified from Eastern Europe or South-East Asia, where the largest populations of PWID are located and where there is a high prevalence of HIV, HCV and HIV/HCV co-infection among PWID. 61–63

Quality of the evidence

Many studies included in the review were assessed as being at severe risk of bias. Of the studies that were assessed, only two were judged as being at moderate overall risk of bias, 17 were judged as being at serious risk and seven were judged as being at critical risk. There is a need to improve transparency and consistency in the reporting of observational studies to facilitate systematic reviews of observational studies. Only a few studies report the effect of exposure to NSPs adjusted for confounders (5/7), which limited the sensitivity analyses that we could conduct. Therefore, efficacy estimates relating to NSP exposure are limited to unadjusted estimates.

Potential biases in the review process

A potential bias in the review was the heterogeneity across the studies in the use of multiple effect measures. Effect measures were converted into risk ratios in the meta-analysis, but this may have introduced bias into our findings because we had to assume that risk ratios approximated odds ratios (ORs), which may be inappropriate for some sites given the high incidence of HCV seroconversion. We removed cross-sectional study designs that identified serological markers of incidence infection as part of our sensitivity analysis. Effect estimates remained the same for current use of OST versus no intervention, but not for high coverage of NSPs. The majority of studies recruited PWID currently or have done so recently, which may not be representative of all PWID exposed to OST and may lead to an underestimate of the effect of OST on HCV infection transmission. For example, in the Amsterdam cohort, people who reported being on OST and having ceased injecting had a lower risk of HCV infection transmission. 14

Changes from original protocol

We have changed the title of the review to refer to opioids instead of opiates. Opioid encompasses synthetic opiates as well as those derived from opium, whereas opiates includes only drugs derived from opium. The original protocol specified that one sensitivity analysis would be to remove studies that reported only incident rate ratios as effect estimates. We did not do this because only three studies used incident rate ratios. Instead, we removed estimates derived from unpublished data sets as part of our sensitivity analyses because seven estimates were derived in this way, making them a more substantive part of the analysis.

There was insufficient evidence to answer some of the research questions that sought to examine differences in effectiveness in terms of the following factors: duration of treatment, dosage of OST, type of substitution used, NSP modality (fixed vs. mobile site) or setting of OST. This reflects a lack of evaluation of the provision of OST or NSPs in other settings.

Outcomes

The primary outcome was new HCV infection (yes/no) based on the definitions described above. Secondary outcomes were based on self-reported injecting risk behaviours (frequency of injecting, injecting with a used needle or syringe, use of shared spoons or filters for drug preparation, injecting site infection) and HIV/HCV infection testing.

Interventions

We used previously published outcome measures of OST use and NSP coverage. 20 OST was defined as current use of OST for all cross-sectional surveys, whereas in the cohort study it was defined as > 6 months of OST in the past year (yes/no). An internationally used standardised measure of an individual’s NSP coverage was defined as the percentage of injections for which a new needle had been obtained (calculated as the average number of new needles obtained divided by the average number of injections in past four weeks, with the exception of the NESI survey, which measures coverage over 6 months, and the Birmingham study, which uses a 2-week time frame). 5,73,74 The total number of needles or syringes obtained from any source was taken, not limiting data to those needles/syringes obtained from a NSP. We examined the effect of both interventions in two ways. First, we measured the impact of binary measures of NSP and OST to assess their individual effect without considering the influence of the other intervention. Second, we combined these binary measures to form a measure of harm reduction coverage with four categories as used in the original analysis and comparable studies with high coverage defined as ≥ 100% of injections covered by a clean syringe for all sites (Table 5). 14,20,68

Statistical analyses

Of the total 14,734 included participants across the eight studies, a total of 7173 were considered in the primary analysis with an initial HCV antibody-negative result. The analysis involved (1) a meta-analysis of the (unadjusted) effect of OST on new HCV infection incidence limited to 5543 participants; (2) a meta-analysis of the (unadjusted) effect of high NSP coverage limited to 4947 participants; (3) a pooled analysis of the (unadjusted and adjusted) effects of NSP and OST on new HCV incident infections confined to 5280 estimates; and (4) a meta-analysis of the pooled effect estimates alongside estimates from a recent systematic review. 26 A flow chart summarising the numbers of HCV infection cases included in analyses (1) to (3) is summarised in Figure 10. We also conducted a pooled analysis focusing on the effects of NSP and OST on secondary outcomes (see below).

FIGURE 10.

Flow chart summarising the combined HCV antibody and RNA test results among PWID in the UK and Australia. +, positive.

Meta-analysis

We conducted a meta-analysis to test for study heterogeneity in the effects of OST and high NSP coverage on new HCV infection. Separate logistic regression models using fixed-effect meta-analyses were used to estimate the study-specific associations of the (unadjusted) effects of OST and NSP on new HCV infection and to examine levels of heterogeneity between study sites. As published elsewhere, counts of new HCV infection cases were small, resulting in no cases in one intervention group in Birmingham20 and in the 2009 Bristol survey. We added one case and three controls to the intervention group with zero cases. 75,76 We used the I²-statistic to assess between-study heterogeneity in the effects of the interventions on new HCV infection. 77

Pooled analysis of the effects of opioid substitution therapy and needle and syringe programmes on new hepatitis C infection

There was no evidence of heterogeneity between the studies for OST (I² = 13.6%; p = 0.324) or NSP (I² = 0%; p = 0.659) so the data were pooled in subsequent analyses and the augmented data points were removed. We used logistic regression to model the odds of recent infection by NSP and joint effects of OST exposure adjusting for key confounders of HCV infection risk including sex,41,45,51,67 injecting duration,41,44,45,67 injecting crack cocaine45,67 and experience of prison. 55

Secondary outcomes

We also examined the risk of self-reported injecting risk behaviour and access to services according to exposure to different levels of harm reduction (OST and NSP) as defined above. Measures of risk included: (1) the proportion of needle/syringe sharing (n = 6217); (2) the frequency of injecting (n = 11,786); (3) reuse of the same needle/syringe more than once for injecting (n = 1242); (4) used of shared filters or spoons for the preparation of drugs (n = 7223); (5) self-reported symptoms of bacterial infections (n = 5039); (6) receiving testing for HIV (n = 13,435); and (7) receiving testing for HCV infection (n = 14,026). We used logistic regression for binary outcomes (measures 1, 2, 4–7) and linear regression for continuous variables (measure 3). All models were adjusted for key confounders listed in the previous analysis (sex,41,45,51,67 injecting duration,41,44,45,67 injecting crack cocaine45,67 and experience of prison55).

Pooled effects and systematic review findings

We conducted a meta-analysis using fixed-effects models to assess the pooled effects of study-level associations between intervention exposure and new HCV infection incidence cases with findings from the Cochrane systematic review reported in Chapter 1. Estimates from the review of European studies that employed the same definition of high NSP coverage were extracted and combined with our additional estimates from the UAMP and ANSP surveys. We used the I²-statistic to assess between-study heterogeneity in the effects of the interventions on new HCV infection.

Meta-analysis to test for study heterogeneity in the effect of opioid substitution therapy and needle and syringe programmes on the risk of hepatitis C acquisition

The meta-analysis shows that high NSP coverage was associated with a slight reduction in the risk of HCV infection but this was not significant (OR 0.76, 95% CI 0.54 to 1.06). Exposure to OST was associated with a 45% reduction in the risk of HCV infection (OR 0.55, 95% CI 0.40 to 0.77). There was no evidence of heterogeneity between the studies for the effect of either intervention (NSP: I² = 0.0%; p = 0.659; OST: I² = 13.6%; p = 0.324). (These findings are summarised in Figures 11 and 12.)

FIGURE 11.

Meta-analysis summarising study-level estimates of the effect of high NSP coverage (> 100%) on HCV infection incidence.

FIGURE 12.

Meta-analysis summarising study-level estimates of the effect of current OST use on HCV infection incidence.

Pooled analysis of the effect of opioid substitution therapy and needle and syringe programmes on new hepatitis C infection

The impact of OST and NSP coverage on new HCV infections is summarised in Table 6 for the complete case data (n = 5280). In the unadjusted analysis, PWID currently using OST had only a 65% reduced odds of HCV infection (OR 0.35, 95% CI 0.26 to 0.48). High coverage with needle/syringes (≥ 100%) alone was not significantly associated with reduced odds of HCV infection (OR 0.83, 95% CI 0.60 to 1.16). Following adjustment for sex, experience of prison or injecting crack cocaine, the intervention effects of OST alone was reduced [adjusted odds ratio (AOR) 0.61, 95% CI 0.43 to 0.87], but the effect of NSP alone was not altered.

When examining the effects of combined harm reduction interventions, the risk of new HCV infection was halved among those on full harm reduction (defined as receiving OST and ≥ 100% NSP coverage) (AOR 0.44, 95% CI 0.27 to 0.71) compared with those on minimal harm reduction (≤ 100% NSP coverage). There were reduced odds of HCV infection acquisition among those on partial harm reduction exposed to high NSP coverage but not among those on OST (AOR 0.59, 95% CI 0.36 to 0.96) and a higher effect for those on OST but with low NSP coverage (AOR 0.41, 95% CI 0.22 to 0.75).

Pooled analysis of the effect of opioid substitution therapy and needles/syringe programmes on injecting risk behaviours

A total of 15% of the sample reported injecting with a used needle/syringe in the past 4 weeks (958/6217), 20% had shared spoons or filters during the preparation of drugs (3018/14,734) and 30% had injected with the same needle/syringe more than once (377/1242). The mean number of injections in the past month was 44. Across five studies, 39.5% of participants had an abscess or sore at the injection site (1990/5039). The majority (77.5%) reported ever having an HIV test (10,414/13,435) and a test for HCV infection (81%, 11,440/14,026).

Pooling estimates with systematic review findings

Combining estimates of NSP coverage from the systematic review with two data sets69,70 not already represented in the review strengthened the evidence for the effect of high needle syringe programme coverage on reducing the risk of HCV infection acquisition to 39% (RR 0.61, 95% CI 0.43 to 0.87) with moderate heterogeneity (I² = 30%; p = 0.189). The systematic review data comprised a total of 438 new HCV infection cases from a sample of 3990. Similarly, the addition of 12 estimates from a systematic review that examined the effectiveness of current use of OST comprising 998 new HCV infection cases from a sample of 5910 supported the evidence for the effect of OST in reducing HCV infection risk acquisition (RR 0.56, 95% CI 0.45 to 0.69; I² = 0%; p = 0.620). These findings are summarised in Figures 13 and 14.

FIGURE 13.

Meta-analysis combining pooled analyses with systematic review findings to measure the impact of high NSP coverage (> 100%) on HCV infection incidence.

FIGURE 14.

Meta-analysis combining pooled analyses with systematic review findings to measure impact of current OST exposure on HCV infection incidence.

Limitations

Our findings suggest a weaker effect of full harm reduction than reported previously, despite having a sample size five times larger and greater study power. One reason for this may be that the majority of new observations were derived from routine surveillance sets (UAMP, ANSP, NESI) that recruit from NSPs and drug treatment centres. First, these data sets cover wide geographic areas for which the effectiveness of interventions is likely to differ, and this variation is not accounted for in the analysis. Second, the reported numbers of needles/syringes obtained from the UAMP were 3–4 times lower than those reported by community surveys, reflecting different inclusion criteria and settings. The community surveys were undertaken in larger urban areas, whereas the UAMP include smaller towns and rural areas. Third, findings clearly point to difficulties in accurately measuring the use of NSP coverage, which relies on correct recall of the frequency of injecting and the numbers of needles/syringes obtained in a time period and does not take into account secondary distribution (people taking additional needle/syringes to give to their peers).

Concluding remarks

This is the largest national UK-wide analysis of the effectiveness of OST and NSP among PWID on HCV infection incidence. Findings clearly underline the importance of OST in combination with NSP as cornerstone interventions to reduce HCV transmission among PWID and the need to prioritise these interventions to prevent HCV. Findings also provide important new evidence of the role of high NSP coverage in reducing not only HCV incidence but also injecting with used needles/syringes and show that increasing the circulation of clean needles/syringes does not increase injecting frequency among PWID. The inclusion of the Australian data set shows that findings are generalisable to other contexts.

Needle and syringe programme costing

Needle and syringe programme outputs

The size of fixed-site services varied substantially across the three cities. The largest fixed site handled > 10,000 visits in the financial year, whereas the smallest fixed site handled only 1756 visits in the same year (see Appendix 3, Table 26). The client mix at fixed sites also varied between sites. The majority of clients attending BDP and CAIR Scotland were users of opioids and other drugs, whereas the majority of clients at Addaction Walsall were IPED users. The total number of clients in 1 year at fixed sites varied from 569 to 1134; however, it is possible that the number of clients was distorted within fixed sites by clients giving different names on different visits.

The majority of clients at all sites were male: 87% of clients visiting BDP, 62% of clients at CAIR Scotland and 91% of clients at Addaction Walsall. At BDP, detailed information on the sex breakdown was available by opioid vs. IPED users. When looking at this breakdown, the sex balance is much more extreme among IPED users, with 98% of IPED users being male. It is possible that the large number of IPED users at Addaction Walsall will have resulted in a greater majority of male clients.

There was considerably less variation in outputs within the pharmacies sampled for the collection of cost data (see Appendix 3, Table 27). The pharmacies for which detailed data were collected reported between 1405 and 2316 visits per year. Pharmacies distributed between 18,518 and 36,100 needles. Two pharmacies distributed needles in ‘pick and mix’ form; the remaining pharmacies distributed needles in pack form only.

Finally, the three ‘other’ modalities distributed a wide range of needles – between 6816 and 101,326 needles per year (see Appendix 3, Table 28). Although detailed information was not available in the out-of-hours pharmacy or the drop-in centre on the breakdown of clients using either opioid and IPEDs, based on feedback from the sites we assumed that 100% of clients at these two sites were opioid users. The outreach service saw a greater mix of opioid/IPED users, with 72% of clients using opioids and 28% of clients using IPEDs. The outreach service saw mostly male clients; this imbalance was more exaggerated among IPED users (at 98% male) than opioid users (84% male).

Needle and syringe programme costs

Total costs

Total annual costs for the distribution of needles to all users (of both opioids and IPEDs) at fixed sites ranged from £25,613 to £65,630 per year, depending on the different sizes and reaches of the different fixed sites (Table 7). Pharmacies saw much less variation in total costs, ranging from £6815 to £10,690 per annum. Total annual costs for the out-of-hours pharmacy in Walsall were slightly less than the total costs for day pharmacies, at £5235. The costs for the distribution of needles using the mobile outreach service in Bristol were £32,887. Total costs for the drop-in centre in Walsall (Hi’s ‘n’ Lows) were lowest, at £2138.

TABLE 7 - Total and unit costs by site

Site	Total annual cost (£)	Total fixed costs (£)	Total variable costs (£)	Total opioid cost (£)	Unit cost per opioid needle (£)	Unit cost per opioid client (£)
Fixed site
BDP	55,854.41	12,229.50	43,624.91	51,860.77	0.21	68.69
CAIR Scotland	65,630.81	5859.49	59,771.32	60,285.41	1.65	86.49
Addaction Walsall	25,613.22	1347.35	24,265.87	23,584.95	0.37	124.13
Other
Out-of-hours pharmacy (Walsall)	5235.09	519.79	4715.30	5235.09	0.77	24.58
Outreach service (Bristol)	32,887.50	536.63	32,350.87	22,361.89	0.24	78.19
Drop-in centre (Walsall)	2138.19	421.68	1716.51	2138.19	0.27	19.01
Pharmacy
Pharmacy 1 (Bristol)	8541.67	860.31	7681.36	9802.99	0.31	48.81
Pharmacy 2 (Bristol)	10,276.01	1262.25	9013.76	11,171.30	0.29	48.36
Pharmacy 3 (Dundee)	6814.59	901.52	5913.07	6814.59	0.31	23.74
Pharmacy 4 (Dundee)	9019.88	2449.12	6570.76	9019.88	0.37	31.15
Pharmacy 5 (Walsall)	10,690.48	650.80	10,039.68	10,690.48	0.58	50.91
Pharmacy 6 (Walsall)	7506.43	443.77	7062.66	7506.43	0.43	36.27

The primary cost driver in most settings was the cost of supplies; this accounted for an average of 60% of total costs across sites (range 23–80%) (Figure 15). Supplies costs for the out-of-hours pharmacy were less than those for day pharmacies, owing to the fact that this pharmacy distributed only one-hit kits, whereas the remaining pharmacies all distributed packs of 5, 10 or 20 needles during the costing period. This was followed in most cases by administrative and overhead costs, which accounted for 6–45% of total costs. Mobile outreach service costs were largely driven by transport costs, making up 47% of total costs.

FIGURE 15.

Cost drivers as a proportion of total costs.

Table 7 shows the fixed and variable costs per site, and the total cost for needle distribution to non-IPED users at each site (methods for this estimation are described previously; see Methods). The total opioid cost varied from £23,584 at Addaction Walsall to £60,285 at CAIR Scotland, and from £6815 at Pharmacy 3 in Dundee to £10,690 at Pharmacy 5, also in Dundee.

The unit cost per opioid needle distributed varied from £0.21 to £1.65, and the unit cost per non-IPED client (annually) varied from £19.01 to £124.13. There was no consistent ranking across sites in terms of unit costs; sites that had high unit costs per client did not necessarily also have high costs per needle, and vice versa.

Uncertainty analysis

Figure 16 provides an illustration of our univariate and multivariate uncertainty analyses for the NSP costing, showing the impact of varying each parameter for which there was uncertainty on the estimated unit cost per opioid needle. The parameter with the greatest effect on costs was that of equipment wastage; this is to be expected, as injecting equipment is the primary cost driver at most sites. Similarly, varying assumptions of the total number of ‘one-hit kits’ distributed per client at the out-of-hours pharmacy impacted the cost per needle substantially. Varying all parameters simultaneously gave a relatively wide estimate of unit costs per needle. The out-of-hours pharmacy was most sensitive to assumptions, with the unit cost per needle varying between –229% and +56% of the base-case cost in the multivariate sensitivity analysis.

FIGURE 16.

Needle and syringe programme costing uncertainty analysis.

Limitations

The primary outcome of interest for this study is needle exchange for the prevention of HCV infection transmission; we therefore include only those costs and benefits that are relevant to this outcome in this costing study. We do not include any costs or benefits from other engagement programmes as delivered by the fixed sites and the peer-led service, such as one-to-one support, sexual health services, legal advocacy, etc. A larger number of visits per person in the injecting population may be necessary for these benefits; further research on the impact of these services is required in order to come to any recommendation on the optimal frequency of visits for injecting populations. These additional services are also likely to bring in more clients to the needle exchange. This may especially be the case for women, who might access sexual health-related services at the sites.

We also do not estimate the costs for the clean-up of drug-related litter, which can also be a substantial investment, and without which some councils would not politically be able to continue the provision of needle exchange. Drug-related litter is a key reason for implementing the distribution of ‘one-hit kits’ in some pharmacies, including the out-of-hours pharmacy, as there have been a number of discarded unused needles found in some study sites. Councils may need to consider the additional costs of litter pick-up schemes; this can vary widely depending on the number of staff employed and the size of the city. In one city, we estimated the salary costs for litter pickup to be approximately £4064 per year, not including transport and equipment costs. In another (larger) city, the overall costs of a dedicated sex- and drug-related litter pickup scheme are reported to be £80,000 per year.

Our analysis also does not consider any potential impact of the prevention of HCV cases in people injecting IPEDs, and excludes variable costs encountered by NSPs in providing needles and syringes for IPED use. Research from Australia indicates that IPED users are at some, albeit reduced, risk of HCV transmission. 85

Finally, we encountered a lack of detailed information on the client mix receiving needle exchange services through pharmacies and we could not obtain any estimates on the relative proportion of IPED to opioid injectors accessing pharmacy needle exchanges. Moreover, we could not find information on male-to-female ratios in pharmacies. Feedback from pharmacies and fixed sites during the costing process indicated that IPED users rarely access pharmacies and that the proportion of opioid visits to pharmacies is likely to be close to 100%.

All findings were presented to each site mid-way through the project, when results were presented and plausibility was discussed. Permission was sought to present the potentially sensitive nature of the findings. It was agreed not to publish the costings separately but to present these alongside cost-effectiveness analyses in any public dissemination or peer-reviewed output.

Model description

We developed a dynamic ordinary differential equation model of HCV infection transmission and disease progression among PWID and ex-injecting drug users. The model simulates the movement of PWID through different injecting durations, intervention, risk and HCV infection disease states, and also follows them after they have ceased injecting to capture HCV morbidity outcomes.

Stratifications by injecting duration are included in the model to capture higher rates of injecting cessation during the first few years of injecting,91 and the heightened level of HCV infection acquisition risk among recently initiated PWID. 20,92 There are four injecting duration categories: < 3 years, ≥ 3 years to < 10 years, ≥ 10 years and ex-injecting drug users. All PWID enter the model as recently initiated PWID (< 3 years injecting category) at a constant rate θ, and then transition to non-recent PWID (at rate 1/3 year^–1) and then to long-term PWID (at a rate of 1/7 year^–1 from the non-recent PWID). Each of these categories also has specific cessation rates of injecting, which results in users entering the ex-injecting drug user class, as well as death rates that vary by injecting duration, giving a total proportion of PWID leaving the current injector category, µ_i. (Figure 17 shows the schematic for this aspect of the model structure. Ex-injecting drug users are also stratified by the same infection and disease progression states so that morbidity benefits can be tracked.)

FIGURE 17.

Schematic of injecting duration and infection components of model. (a) Injectors cease injecting (cessation or death) at rate µi, where i = 1, 2 or 3 for recent, non-recent and long-term injectors, respectively. Injecting duration is modelled in three categories with progression at rate τi, where i = 1 or 2 for recent and non-recent injectors, respectively; (b) schematic of the intervention component of the model. It is assumed that the recruitment rates β and η are independent of the current intervention state; and (c) schematic of disease progression component of the model. Each of the disease states is stratified by injecting duration, n; risk category, m; OST category, i; and NSP category, j. Progression through the disease states is by a rate determined by the current disease state, as are disease-related death rates. In addition to current injectors, ex-injecting drug users are followed through the model when they cease injecting with an identical pathway, except that reinfection is excluded. All states have a cessation rate from injecting and a background death rate that is not disease related. Reinfection can occur at any stage but these lines are not shown after compensated cirrhosis. C, chronically infected; DC, decompensated cirrhosis; F0, F1, F2, F3, chronic infection states; F4, compensated cirrhosis; HCC, hepatocellular carcinoma; LT, liver transplant; PLT, post liver transplant; S, susceptible.

People who inject drugs are also tracked through different intervention states: no intervention, on OST only, > 100% NSP only, and on both OST and > 100% NSP. When PWID start injecting they are assumed to be in the ‘no intervention’ state. Recruitment on to OST and > 100% NSP occurs at per capita rates β and η, respectively, with both being assumed to be independent of current risk state for simplicity and lack of data. The rates of leaving OST and > 100% NSP are γ and κ, respectively. (See Figure 17b for the schematic for this component of the model structure.)

The model is further stratified by high and low injecting risk, with high-risk behaviours being defined by characteristics that have been shown to increase the risk of HCV infection acquisition among PWID in the UK. High risk is defined as being homeless and/or a crack cocaine injector, and low risk is defined as being neither homeless nor a crack cocaine injector. A proportion of individuals, ϕ, enter the model in the high-risk category. Movement from the low- to the high-risk category and vice versa occurs at per capita rates σ and ζ, respectively.

All PWID enter the model as susceptible individuals and become infected at a per capita rate, (λ^n,m_i,j) or force of infection, which depends on the intervention state, injecting duration category and risk category of the individual, as well as the prevalence of infection. The force of infection for each susceptible state is defined by the relative risk in that state, such that infectivity and susceptibility are multiplied by the following factors: Γ, Π, B if the injecting drug user is on OST, ≥ 100% NSP or both, respectively; X₁ and X₂ if the injecting drug user is a recent injector or non-recent injector, respectively; and Ξ if the injecting drug user is a high-risk individual. Because assortative mixing has been shown to have little effect on transmission when there is movement between high- and low-risk groups, we do not include it here. 86 In addition, evidence from Bristol88 suggests that there is little assortative mixing by injecting duration, so we assumed random mixing between all categories. For this, the chance of an injecting drug user having a transmission event with an injecting drug user in a specific risk category is assumed to be proportional to the overall transmission risk of PWID in that state.

Once infected, a proportion of newly infected PWID, δ, spontaneously clear the infection and remain in the susceptible category, and the remainder enter the chronically infected category (l^n,m_i,j) and remain infected until they die or are treated. The progression of HCV infection for any PWID who enters the chronically infected PWID category is modelled as shown in Figure 17c. The disease states modelled are METAVIR stages F0, F1, F2, F3 (chronically infected), F4 (compensated cirrhosis), decompensated cirrhosis, hepatocellular carcinoma (HCC), liver transplant and post liver transplant. Individuals progress from F0 through each METAVIR stage to compensated cirrhosis, from which they progress either to decompensated cirrhosis or to HCC. Individuals with decompensated cirrhosis can then progress to HCC or can be considered for liver transplant (as can individuals with HCC), which, if successful, results in the individual moving to the post liver-transplant stage. HCV disease-related death occurs from the decompensated cirrhosis, HCC, liver transplant and post liver-transplant stages with disease stage-specific death rates.

Treatment of HCV infection is allowed only in the F0–F3 and compensated cirrhosis (F4) stages, as it is unclear whether or not treatment in later disease stages is beneficial. 93 A number of PWID, Φ, are assumed to be treated each year and on treatment a proportion, α, achieve a sustained virological response (SVR) and re-enter the susceptible non-infectious category. Treatment of ex-injecting drug users occurs at per capita rate r, with the same proportion, α, achieving SVR. Those who do not attain SVR remain in the infected category and are eligible for retreatment. This simplifying assumption was made based on the influx of new treatment regimes becoming available. On successful treatment, individuals enter the respective susceptible category with the same disease stage (see Figure 17c). Disease progression still occurs among these individuals if they have compensated cirrhosis, but at a slower rate. We assume no further disease progression if treatment is successful in the F0–F3 chronically infected non-cirrhotic stages. 94,95 The disease progression of ex-injecting drug users is also tracked to assess the long-term impact that interventions will have on this subpopulation. It is assumed that ex-injecting drug users who are susceptible or who have been successfully treated cannot be reinfected.

Model parameterisation

The model parameters can be found in Table 10. The model was calibrated to the following parameters estimated from survey data: HCV infection prevalence, coverage of OST and NSP, proportion of PWID with high-risk behaviour and injecting duration distribution of PWID over time.

Odds ratio effect estimates for the degree to which the risk of HCV infection transmission is modified when an injecting drug user is on OST or there is ≥ 100% NSPs in the high-risk category or for different injecting durations were taken from a pooled analysis of UK and Australian data (see Chapter 2). The OR combined effect of OST and ≥ 100% NSP was calculated by multiplying together the effects estimates for OST and ≥ 100% NSP alone, to reflect the assumption that both interventions together would be more beneficial than alone, giving a range comparable to that from the systematic review (see Chapter 1).

Hepatitis C treatment was assumed to occur in all settings since 2009 at context-specific rates based on recent data from those settings, except for Walsall where data from Bristol was used because of a lack of data from that setting. Before 2015, we estimated the SVR rate for pegylated interferon and ribavirin from a recent study of PWID in the UK. 99 After 2015, we assumed that treatment would involve using the new direct-acting antiviral (DAA) drugs, and so assumed a high SVR rate of 90% for all genotypes. 103

Instantaneous HCV infection disease stage progression rates were calculated from a recent meta-analysis of PWID-specific progression probabilities to compensated cirrhosis101 and a systematic review of general progression probabilities for other disease stages,101 and, similarly, the spontaneous clearance probability came from a published meta-analysis. 96 Non-HCV-related death and injecting cessation were combined into one rate for each injecting duration strata, with the death rates being estimated from a cohort study of PWID in Scotland,91 and the cessation rates being estimated through fitting the model to the observed distribution of PWID by injecting duration.

The leaving rate from the high-risk category, characterised as homelessness or crack cocaine injection, came from two studies among PWID in the UK. 39,97 The leaving rates from NSP and OST came from a Welsh cohort study. 39 Conversely, the recruitment rates to these states were obtained through calibration.

Model calibration and uncertainty

The model was calibrated to data on PWID population size, historical coverage levels of OST and NSP, prevalence of high-risk behaviours (crack cocaine injection and homelessness), HCV infection prevalence and the distribution of injecting duration. These calibration data were collated from various sources (Appendix 4, Table 29 summarises the data used), including size estimation studies,104–107 three Bristol community surveys in 2004, 2006 and 2009,44,73,88,89 yearly UAMP surveys from 1991 (Bristol) or 2006 (Walsall) to 2014,70 and four surveys from NESI for Dundee from 2008 to 2014. 70 Data on incidence for Dundee (NESI) and Bristol (community survey 2009) were used for model validation as well as for prevalence data post 2006 for Bristol and Walsall (UAMP).

Model calibration was carried out in three steps with 1000 parameter sets obtained at each step:

1. population size and injecting duration fitting using a PWID demographic submodel without infection

2. NSP and OST coverage fitting using a submodel that includes HCV infection transmission but no disease progression

3. HCV infection prevalence fitting using the full model with disease progression.

Step 1

In Dundee, survey data70 suggested that the proportion of the PWID population in each injecting duration category was stable from 2008 to 2014, and so we assumed a constant population size estimated from unpublished data from Scotland. In Bristol and Walsall, size estimation data suggest that the PWID population has decreased by between 10% and 30% between 2009 and 2011. 104–106,108 Concurrently, survey data44,70,73,88,89 suggest that the proportion of PWID injecting for > 10 years has increased, whereas the proportion injecting for between 3 and 10 years has decreased (Figure 18). There has been little change in the proportion injecting for < 3 years. It was assumed that these changes were partly attributable to a decrease in the initiation rate of new injectors and a change in the cessation rates of non-recent and long-term injectors. We allowed for uncertainty around these parameters and estimated them by fitting the model to the population size and injecting duration profile (proportion of PWID in each injecting duration category) at two time points for Walsall and Bristol and one time point for Dundee. This fitting was done with a demographic submodel, which had only three injecting duration categories and no other stratification. We assumed that the PWID population size was at equilibrium initially (before 2004, 2006 and 2008 for Bristol, Walsall and Dundee, respectively). We sampled 1000 values for this ‘stable’ initial population size and the cessation rate from the recent injector category for each setting. For each of these 1000 parameter sets, the wide prior distributions for the cessation rates from non-recent and long-term injectors (see Appendix 4, Table 29) were then sampled, and for each sample the model was fit to the initial population size by calculating a suitable PWID recruitment rate, using the steady state equations for the demographic submodel (for more details, see Appendix 4). Parameter sets were retained if the resulting injecting duration profile lay within the ranges suggested from the data, otherwise the cessation rates were resampled. We then sampled 1000 estimates for the later population size in 2011 for Bristol and Walsall, as well as new cessation rates for non-recent and long-term injectors, and the PWID recruitment rate was recalibrated to fit to this new sampled population size for the 2011 data (for Bristol and Walsall only). This refitting of the demographic submodel was done using the Matlab release 2014b (The MathWorks, Inc., Natick, MA, USA) algorithm fzero applied to the analytic solution of the model with initial conditions from the first step of fitting. Parameter sets were retained if the resulting injecting duration profile lay within ranges suggested from data for the years 2004 and 2011 for Bristol and 2008 and 2011 for Walsall, otherwise the new cessation rates for this second step were resampled to obtain a fit to each of the first step parameter sets (1000 each for Bristol and Walsall).

FIGURE 18.

Graphs showing model fitting of the baseline scenarios in each setting. Error bars in black are data points from surveys, error bars in blue are the ranges used for model calibration (for NSP coverage this was the service provision estimate). (a) Bristol; (b) Walsall; and (c) Dundee.

Impact analysis

First, the model was used to estimate the impact of current intervention activities over the next 15 years, from 2016 to 2031. This included the impact of current coverage levels of OST, NSP and HCV infection treatment. To estimate the impact of each intervention, the baseline epidemic projections with these interventions incorporated were compared with what would happen if the effect of these interventions were removed from the start of 2016 (i.e. either the efficacy parameter for a specific intervention was set to one from 2016 or no additional PWID were HCV infection treated). Following this, we also assessed the detrimental impact of current levels of high-risk behaviours. This was evaluated in the same way over 15 years. The impact of each scenario was assessed in terms of the relative change in prevalence and incidence, and the relative change in the number of incident infections and the number of disease-related deaths. Specifically, we investigated how the epidemic in each setting would change from 2016 to 2031 if the following interventions or behaviours had no effect on HCV infection transmission rates:

• > 100% NSP

• OST intervention

• > 100% NSP and OST intervention

• HCV infection treatment of PWID

• all high-risk behaviours.

Following this, we then considered the potential impact to 2031 of increasing the coverage of > 100% NSP from 2016 to 80% or 90% over the next 5 years.

Sensitivity analysis

The model calibration algorithm involved probabilistic sampling over a large number of the model parameters, and we fit the model across the full uncertainty of the calibration data. The different runs from this model calibration exercise were used to assess those parameters that effected the model projections most. We undertook a linear regression analysis of covariance109 to determine those parameter uncertainties that contribute most to uncertainty in the 15-year impact of current NSP coverage levels on the relative change in the number of infections. The proportion of each model outcome’s sum-of-squares contributed by each parameter was calculated to estimate the importance of individual parameters to the overall uncertainty.

Baseline epidemic projections and model validation

In agreement with available recent HCV infection prevalence data up to 2014, we projected a slightly increasing HCV infection prevalence in Bristol and Walsall when fitting prevalence to only the first time point (2004 and 2006, respectively), as shown in Figure 19. The increasing HCV infection prevalence in Dundee was obtained by fitting the prevalence to two time points. The incidence of HCV infection calculated from the model in 2014 was used to validate the model and varies between the three settings, with the highest incidence in Dundee [16.8, 95% credible interval (CrI) 10.7 to 24.7 per 100 person-years], which agrees with incidence estimates from NESI (14.3, 95% CI 4.9 to 25.9 per 100 person-years). In Bristol, we project a lower HCV infection incidence of 6.9 (95% CrI 3.9 to 11.5) per 100 person-years, slightly lower than the most recent HCV infection incidence estimate from 2009 (10.0, 95% CI 9.7 to 14.0 per 100 person-years). 88 In Walsall, we project an even lower HCV infection incidence of 3.4 (95% CrI 1.7 to 6.5) per 100 person-years, corresponding to the low and decreasing prevalence in that setting. Unfortunately, no empirical estimates of HCV infection incidence exist for Walsall against which to compare these estimates.

FIGURE 19.

Impact of each intervention scenario on HCV infection prevalence and incidence in each setting. (a) Bristol incidence; (b) Bristol prevalence; (c) Walsall incidence; (d) Walsall prevalence; (e) Dundee incidence; and (f) Dundee prevalence. The solid line is the median of the baseline model scenario, with the shaded region representing the 95% CrIs around those projections. Error bars in black are data points and those in blue are the ranges used for calibration. py, person-years.

Over the period from 2016 to 2031, the model suggests that HCV infection prevalence will slightly decrease in Bristol and Walsall, and decrease markedly in Dundee from 36.9% (95% CrI 28.7% to 43.5%) to 10.1% (95% CrI 0.0% to 42.3%). The decreases in all settings are attributable to the use of more effective DAA treatment commencing in 2015, with the larger decrease in Dundee being attributable to treatment already being scaled up as part of a trial intervention. Over each year, the models project that 32 (95% CrI 21 to 43), 6 (95% CrI 1 to 11) and 14 (95% CrI 9 to 20) HCV-related deaths will occur among PWID and ex-injecting drug users in Bristol, Dundee and Walsall, respectively, with most (> 85%) of these deaths occurring among ex-injecting drug users. Baseline prevalence, incidence and disease-related mortality can be found in Table 11.

Impact of existing interventions

Regardless of setting, removing either or both of NSP and OST would have a large detrimental impact on both prevalence and incidence in all three settings by 2031, as shown in Figure 19. As expected, removing both interventions has the biggest effect on the epidemics, with the model suggesting at least a 337% (range 337–1525% between settings) relative increase in incidence by 2031 compared with baseline 2031 levels, a 125% (range 125–166%) relative increase in the number of new HCV infections over the period 2016 to 2031 and a 35% (range 65–636%) relative increase in prevalence compared with baseline projections in all settings from 2016 to 2031. Following this, removing OST has the next biggest impact, resulting in at least a 196% (range 196–1034%) relative increase in incidence, an 86% (range 86–125%) increase in the number of HCV infections and a 46% (range 46–562%) relative increase in prevalence in 2031. The next biggest impact is removing NSP, which increases incidence by at least 35% (range 35–372%) and prevalence by at least 17% (range 17–275%). NSP has a smaller impact than OST because it has lower coverage in each setting and lower efficacy (NSP lowers transmission risk by 41% vs. 59% for OST) as suggested by our pooled analysis. Removing NSP still results in a large relative increase in the number of new HCV infections by 2031, with a 30% (95% CrI 7.0% to 67.0%) increase in Bristol, a 22% (95% CrI 6% to 40%) increase in Walsall and 59% (95% CrI 12% to 219%) increase in Dundee, as seen in Figure 20.

FIGURE 20.

Impact of removing NSP, OST, NSP and OST or treatment of PWID on the number of new infections from 2016 to 2031 for each setting. The box plots signify the uncertainty (the middle line is the median, the limits of the box are 25% and 75% percentiles, and the whiskers 2.5% and 97.5% percentiles).

Across the cities, a greater relative impact on prevalence and incidence is seen by removing NSP and/or OST in Dundee, then Walsall and, finally, Bristol. Removing NSP and/or OST has the greatest impact in Dundee because HCV infection prevalence and incidence were otherwise decreasing to low levels by 2031, whereas removing the interventions causes prevalence to increase, as seen in Figure 19.

Removing treatment of PWID in each setting has similar impacts on prevalence and incidence to removing NSP, as seen in Figure 19. However, less impact is seen on the number of new infections, as seen in Figure 20. This is explained by treatments lowering incidence indirectly through reducing prevalence, as opposed to directly reducing the incidence of new infections, which is the case for NSP and OST.

The impact of current high-risk behaviours

When the increased transmission risk associated with homelessness and crack cocaine injecting (or homelessness only in Dundee) is removed, a decrease in incidence and prevalence of HCV infection occurs. The biggest relative decrease is seen in Dundee, where incidence decreases by 99% (95% CrI 86% to 100%) and the number of new infections decreases by 58% (95% CrI 29% to 77%) between 2016 and 2031 (Figure 21). Similarly, the number of new infections decreases by 59% (95% CrI 40% to 75%) in Walsall and 64% (95% CrI 45% to 78%) in Bristol. However, despite Bristol having the highest proportion (> 80%) of high-risk PWID, the smallest relative decrease in prevalence (35%, 95% CrI 22% to 53%) occurs here when we remove the elevated risk among high-risk PWID; this is compared with a 99% (95% CrI 69% to 99.7%) decrease in Dundee and a 45% (95% CrI 28% to 64%) decrease in Walsall. This could be due in part to the longer injecting duration predicted by the model in Bristol, with a median of 15 years injecting in Bristol compared with 8 years in Dundee and Walsall.

FIGURE 21.

Impact of removing transmission risk associated with high-risk behaviour on decreasing the number of new infections from 2016 to 2031 for each setting. The box plots signify the uncertainty (the middle line is the median, the limits of the box are the 25% and 75% percentiles, and the whiskers are the 2.5% and 97.5% percentiles).

Impact of scaling up needle and syringe programmes

Scaling up > 100% NSP coverage to 80% or 90% has a substantial impact on the number of new infections by 2031, as shown in Figure 22. For instance, a large impact is seen in Walsall because of the low current coverage of NSP (21–42%), in which the number of new infections is decreased by 25% (95% CrI 7% to 40%) and incidence decreases by 41% (95% CrI 12% to 61%) when NSP is scaled up to 80% coverage. Conversely, increasing NSP coverage in Bristol and Dundee has less impact because of the higher current coverage in both settings (38–82% in Bristol and 34–79% in Dundee), with the number of infections decreasing by 10% (95% CrI 2.0% to 22%) in Bristol and 26% (95% CrI 6.0% to 48%) in Dundee. Less impact is always seen on prevalence, with an absolute drop of approximately 4% in all three settings.

FIGURE 22.

Impact of increasing NSPs in each setting on the number of new infections from 2016 to 2031. The box plots signify the uncertainty (the middle line is the median, the limits of the box are 25% and 75% percentiles, and the whiskers 2.5% and 97.5% percentiles).

Sensitivity analysis

The results of the sensitivity analysis are shown in Figure 23. In Bristol and Walsall, the model parameter with the largest percentage contribution to the variability in the relative number of infections averted from current coverage levels of NSP is the efficacy estimate for NSP (which accounts for 41%, 20% and 48% of variation in Bristol, Dundee and Walsall, respectively). In Bristol and Walsall, the next most important input was the 2014 coverage of NSP (32% of variation in Bristol and 39% in Walsall). Conversely, for Dundee, HCV infection prevalence in 2014 had the largest contribution (36%) to the variability, followed by NSP effectiveness, OST effectiveness then NSP and OST effectiveness (20%, 10% and 6%, respectively). In Bristol and Dundee, the population size estimate contributed 3% and 5%, respectively, to the variability. All other parameters and inputs contributed < 5% to the variability in the impact of NSPs on the number of infections averted. The different patterns seen between Bristol/Walsall and Dundee could be attributable to the differences in the fitting procedures for each setting. Dundee required an extra fitting step to calibrate the model to the increasing prevalence from 2008 to 2014 and had a constant population size, which would account for the impact of those parameters. The presence of the OST and OST + NSP effectiveness parameters can be explained by the impact of these parameters when NSP is removed, which results in the OST + NSP effectiveness becoming the same value as the OST effectiveness parameter.

FIGURE 23.

Sensitivity analysis results: percentage contribution of the uncertainty in different model parameters to the overall variation in the relative increase in number of infections when NSP is removed (only those aspects with > 3% contribution in any of the settings are shown). (a) Bristol; (b) Walsall; and (c) Dundee.

Strengths and limitations

Our detailed modelling of contrasting settings gives a better understanding of how the impact of interventions vary across the UK, with the use of updated empirical efficacy estimates for OST and NSP consistent with the included systematic review, thereby lending added strength to the results. In addition, our model stratified and calibrated PWID by injecting duration, allowing it to incorporate possible important changes in injecting recruitment and cessation over recent years, thus giving more realism to the model projections.

As with all modelling, there are several limitations. The main limitation is the sparsity of data on how injecting cessation and recruitment has changed; we thus had to extrapolate from imperfect data on size estimates and changes in the sample distributions by injecting duration in successive surveys. Importantly, we incorporated the uncertainty in these parameters, and our model projections were informative despite this. However, better data on these hard-to-quantify parameters would improve the accuracy of our model projections. Second, we did not consider any impact of OST on injecting cessation or mortality rates. Long-term OST use has been shown to increase injecting duration and lower mortality rates,91 while being associated with increased incidences of temporary cessation. 110 It is possible that the increasing proportion of long-term injectors seen in Bristol and Walsall may in part be due to the high current levels of OST in these settings. Future modelling could consider temporary cessation as well as permanent cessation, with current OST use impacting on the rates of cessation.

infectionThird, we assumed that the whole population of PWID would be eligible for OST, but there is a growing proportion (from 3.9% in 2004 to 12% in 2014) of PWID in the UK who inject non-opioid substances, such as amphetamines, as their main drug. 70 This is likely to have diluted the effect estimates from the pooled analysis, meaning that we may have underestimated the impact of current OST levels on HCV infection incidence. Fourth, we did not model incarceration, which can be associated with HCV infection transmission risk in the UK. 111 The role of incarceration in driving HCV infection transmission in the UK is the focus of current research. Finally, we did not consider IPED injectors in this model because the HCV infection prevalence in this subpopulation of PWID is much lower than psychoactive drug injectors (3.6% HCV antibody prevalence compared with 50%). 70 It is likely that their contribution to the overall HCV infection epidemic in the UK is small.

Other evidence and implications

Other models have looked at the impact of NSP or OST on HCV infection transmission, but have generally lacked empirical efficacy estimates for the effect of these interventions on HCV infection acquisition risk,79,112–114 or have not considered impact in specific cities parameterised with detailed data. 86 Our modelling extends existing analyses by including detailed context-specific modelling of the impact of OST, NSP and HCV infection treatment, as well as the disabling effect of high-risk behaviours and how it varies across UK settings.

Model calibration

The model was calibrated for each city using survey data from each city (see Chapter 4 for more detail). The model was used to estimate the cost-effectiveness of existing levels of NSP compared with if they were removed from 2016, with the removal of NSP being simulated for 10 years (2016 to end 2025) and then the subsequent transmission effects simulated for a further 40 years (2026 to end 2065). For the baseline intervention scenario, we assumed that OST and > 100% NSP coverage levels remained at current levels and kept other model parameters constant. In the sub-baseline counterfactual scenario, the lower transmission risk associated with > 100% NSP was set to one, and the transmission risk for current OST and > 100% NSP use was set to the transmission risk associated with just current OST use. This was done for 10 years to investigate the detrimental impact of removing current levels of > 100% NSP in each city. No costs of NSP were assumed over this time. After 10 years, the transmission risks associated with > 100% NSP, and current OST and > 100% NSP were set to baseline levels again to simulate the resumption of NSP activities, with the estimated costs of NSP being included for subsequent years. We calculated the costs and quality-adjusted life-years (QALYs) over two time horizons. First, we calculated costs and QALYs for a further 40 years after NSP was reintroduced, giving a total time horizon of 50 years. We also calculated costs and QALYs for a further 90 years after NSP was reintroduced, giving a total time horizon of 100 years. The numbers of individuals in each category of the model were recorded, and QALYs and costs were attached to each category as appropriate. Both time horizons are presented in the results.

Uncertainty in the underlying parameters was accounted for, such that demographic and epidemiological parameters, disease progression rates, costs and health utilities were randomly sampled from appropriate distributions. For each city, we obtained 1000 matched simulations for the costs and QALYs of the baseline intervention scenario and the sub-baseline counterfactual scenario.

Using the sampled simulation sets, we calculate the total costs and the total QALYs gained for the baseline scenario with NSP coverage, and for the scenario removing NSP services for a period of 10 years. These scenarios are presented as ‘NSP’ and ‘no NSP’ respectively for clarity. All future costs and QALYs were discounted using a baseline discount rate of 3.5%.

Cost data

The fixed city-level cost and cost per needle of NSP programmes was estimated over 1000 iterations, simultaneously varying all parameters described in the costing sensitivity analysis, and these 1000 estimates for costs were input into the model. The number of needles distributed each year was calculated by multiplying the total number of PWID with > 100% NSP coverage from the model by the mean injecting frequency per year (see Chapter 4 for details). The total NSP cost per year is then the total number of needles distributed per year multiplied by the cost per needle, added to the fixed city-level cost. (The sampled values of the cost per needle and the fixed city-level cost are shown in Appendix 5, Table 30.)

To enable use of the model for the cost-effectiveness analyses, HCV infection care costs were assigned to different HCV infection stages in the model. These costs were drawn from previously published estimates of the costs of treatment and care for HCV infection in the UK (see Appendix 5, Tables 31 and 32 for data used). Detailed estimates of total cost for each disease stage were drawn from the literature and inflated to 2014 GBP using the hospital and community health services index (see Appendix 5, Table 31 for the cost values used in the cost-effectiveness model). We assumed that 50% of chronic infections (F0–F3 disease stages) in PWID and ex-injecting drug users are diagnosed and incur a cost. We also assumed that treatment costs for active PWID were higher than those for ex-injecting drug users or non-injecting drug users. Costs associated with the supportive care of compensated cirrhosis were applied to all individuals in that disease stage.

Effectiveness

Health utility values (QALY weights) are sourced from previous economic evaluations of HCV infection treatment interventions (see Appendix 5, Table 32). Following previous analyses, we assume that the baseline quality of life for active injectors is lower than that for ex- or non-injectors. As for the model fitting for the impact analyses, the HCV infection disease utility and costs were sampled for each run.

Model outputs

Table 12 shows the total deaths and HCV infections averted through NSP compared with no NSP in each city. Bristol has a median anticipated 21 deaths averted over 50 years and Dundee has a median 23 deaths averted; Walsall has a median anticipated 5.8 deaths averted over the same time horizon. The number of infections averted varies by city, from 93 infections in Walsall to 749 infections in Dundee. Walsall has the smallest prevalence of HCV infection within the injecting population and, therefore, saw the smallest number of deaths and infections averted, despite the fact that the proportion of deaths and infections averted through NSP was similar in each city, as shown in Chapter 4. There was wide uncertainty in the outputs in all cities.

Table 13 shows the total health-related costs incurred over the 50-year time horizon for Bristol, Dundee and Walsall, discounted to reflect their current value. Removing NSPs consistently increases health-related costs across all cities, including costs of health care for early-stage HCV infection, costs for HCV infection treatment among PWID and costs for HCV infection treatment among ex-/non-injecting drug users. These increased costs reflect an increase in HCV infection transmission in the ‘no NSP’ scenario, including to people who then cease injecting. The costs of NSP are higher in the ‘NSP’ scenario. In Dundee, this additional cost for NSPs increases the total cost marginally compared with the ‘no NSP’ scenario; however, in Walsall the total median costs for the ‘NSP’ scenario are lower than those for the ‘no NSP’ scenario, despite additional NSP costs. In Bristol, total median costs are the same in both scenarios.

Cost-effectiveness

Table 14 presents the average total costs, QALYs and incremental cost-effectiveness ratios (ICERs) for the 50-year time horizon. In Bristol and Dundee, providing NSP services is cost-saving; keeping NSP services is estimated to save an average of £137,949 in Bristol and nearly £7.5M over the 50-year time horizon in Dundee. NSPs are also anticipated to contribute 502 and 958 incremental QALYs in comparison to the ‘no NSP’ scenario in Bristol and Dundee, respectively. In Bristol and Dundee, an additional £10M and £26M, respectively, could be spent on NSP without any additional impact, and NSPs would still fall under the commonly cited willingness-to-pay (WTP) threshold of £20,000 per QALY gained.

In Walsall, the mean incremental costs of the NSP scenario compared with the ‘no NSP’ scenario are estimated at £115,250. NSPs are expected to contribute 192 incremental QALYs compared with the scenario removing NSPs. The mean incremental cost-effectiveness of NSPs in Walsall is estimated at £601 per QALY gained. Our central estimate for the ICER in Walsall is well below the commonly cited NICE WTP threshold of £20,000–30,000 per QALY gained and, therefore, can be regarded as highly cost-effective. In fact, an additional £3.7M could be spent with no impact, and NSPs would still fall under the £20,000 per QALY threshold. It has recently been suggested that a more appropriate WTP threshold in a UK setting would be £13,000, as this is the rate at which the NHS is able to turn costs into QALYs. 116 Our central estimate for Walsall also falls well below this threshold, indicating a high return on investment in the current NHS setting.

Table 15 presents the average total costs, QALYs and ICERs for the 100-year time horizon. Over the 100-year time horizon, NSPs are cost-saving in all cities, saving £687,351, £8,800,186 and £177,778 in Bristol, Dundee and Walsall, respectively, and contributing 699, 1326, and 278 QALYs, respectively, in that time. Using the lower WTP threshold of £20,000 per QALY gained, this represents a net monetary benefit of > £14.6M in Bristol, £35M in Dundee and £5.7M in Walsall over the 100-year time horizon.

The ICERs for Bristol, Dundee and Walsall over 1000 baseline runs for the 50-year time horizon are shown in Figures 24–26. All cities show some uncertainty in the ICER; however, across all cities there were no estimates in the bottom-left quadrant (indicating that NSPs are dominated). In Dundee, 99% of iterations are located in the bottom-right quadrant, indicating that NSPs are cost-saving. In Bristol, 45% of iterations are cost-saving, and in Walsall, 40% of iterations are cost-saving.

FIGURE 24.

Incremental cost-effectiveness: Bristol.

FIGURE 25.

Incremental cost-effectiveness: Dundee.

FIGURE 26.

Incremental cost-effectiveness: Walsall.

Sensitivity analysis

Figures 27–29 show cost-effectiveness acceptability curves over a range of sensitivity analysis scenarios for Bristol, Dundee and Walsall, respectively. As indicated in the figures, our model was relatively robust to assumptions, and the likelihood of cost-effectiveness was not substantially altered in the sensitivity analysis scenarios. In the base case, 96% of iterations in Bristol, 100% of iterations in Dundee and 95% of iterations in Walsall are located below the WTP threshold of £13,000 per QALY saved. Using the more commonly cited lower-bound threshold of £20,000 per QALY saved, 98% of iterations from Bristol, 100% of iterations from Dundee and 98% of iterations from Walsall are below the threshold.

FIGURE 27.

Cost-effectiveness sensitivity analysis: Bristol.

FIGURE 28.

Cost-effectiveness sensitivity analysis: Dundee.

FIGURE 29.

Cost-effectiveness sensitivity analysis: Walsall.

The sensitivity analysis scenario with the greatest impact on the ICER assumes a discount rate of 0% for both costs and QALYs. Under this scenario, 86% of iterations in Bristol, 100% of iterations in Dundee and 85% of iterations in Walsall are below a WTP threshold of £0. The scenario reducing HCV infection treatment drugs to half of their listed price returns the lowest proportion of iterations below any given WTP threshold overall. However, even under this scenario, the large majority of iterations are below a WTP threshold of £20,000 (98% in Bristol, 100% in Dundee and 97% in Walsall).

Planned publications

Platt L, Vickerman P, Hope V, Ingle SN, Craine J, Iversen A, et al. The effect of needle/syringe provision and opiate substitution therapy on HCV incidence: update on a pooled analysis from the UK and Australia. Addiction 2016.

Sweeney S, Ward Z, Platt L, Guinness L, Hickman Smith JM, Ayres R, et al. Cost-effectiveness of needle and syringe provision to prevent transmission of HCV.

Ward Z, Platt L, Sweeney S, Hutchinson S, Hickman M, Smith J, et al. Impact of current and scaled up levels of needle and syringe programmes and Opiate substitution therapy in three UK settings.

Policy brief

Preventing Hepatitis C: the role of needle syringe programmes and opiate substitution therapy.

Conference dissemination

Sweeney S, Ward Z, Platt L, Hope V, Maher L, Hutchinson S, et al. The Cost-Effectiveness of Needle and Syringe Provision in Preventing Transmission of Hepatitis C Virus in People who Inject Drugs. Public Health England Research and Applied Epidemiology Conference, Warwick, UK, March 2017.

Ward Z, Platt L, Sweeney S, Hope V, Maher L, Hutchinson S, et al. Impact of Current and Scaled Up Levels of Needle and Syringe Programmes and Opiate Substitution Therapy in Three UK Settings. Fifth International Symposium on Hepatitis Care in Substance Users, Oslo, Norway, September 2016.

Platt L, Reed J, Minozzi S, Hickman M, French C, Hope V, et al. Effectiveness of Needle/Syringe Programmes and Opiate Substitution Therapy in Preventing HCV Transmission Among People who Inject Drugs. International Harm Reduction Conference, Kuala Lumpur, Malaysia, 2015.

Platt L, Reed J, Minozzi S, Hickman M, French C, Hope V, et al. Effectiveness of Needle/Syringe Programmes and Opiate Substitution Therapy in Preventing HCV Transmission Among People who Inject Drugs. Fourth International Symposium on Hepatitis Care in Substance Users, Sydney, NSW, Australia, October 2015.

Platt L. Effectiveness of NSPs and OST on HCV Incidence. National Needle Exchange Forum, High Wycombe, UK, December 2014.

Inflow of injectors

There are only two variables in the model that allow an inflow of new injectors. These are low- and high-risk susceptible individuals in the first disease progression category with no intervention: the number of new low-risk individual per year is θ(1 – ϕ) and the number of new high-risk individuals per year is θϕ.

Injecting duration progression

These terms in the equations are concerned with movement from one injecting duration category to another as well as PWID-related and background mortality. ID^n,m_i,j,k denotes the terms in an ordinary differential equation of injecting duration category n. It occurs for all values of m,i,j,k. Y^n,m_i,j,k is used to describe one of the variables in the model, in which Y = S or C and the subscripts and superscripts are as described previously.

When n = 4 (ex-injecting), the terms have a different form:

Interventions: opioid substitution therapy and needle and syringe programmes

These terms in the equations are concerned with the movement of injectors from one intervention category to another. ITi,j,kn,m denotes the terms in the ordinary differential equation of OST intervention category i and NSP intervention category j. These terms can be found for all values of m,k and current injector categories but not for the ex-injecting drug user category (n = 4):

High and low risk

These terms in the equations are concerned with movement of current injectors between low and high risk. HR^n,m_i,j,k denotes the terms in the ordinary differential equation of risk category m. These terms can be found in the equations for all values of i, j, k and n = 1,2,3.

Disease progression

These terms in the equations are concerned with movement through the disease states. Infection and treatment are described separately. DS^n,m_i,j,k denotes the terms in the ordinary differential equation of disease category k for susceptible individuals and DC^n,m_i,j,k for infected individuals. These terms can be found in the equations for all values of I,j,n,m.