Evidence review for chronic pain
Evidence review B
NICE Guideline, No. 144
Effectiveness of cannabis-based medicinal products for the treatment of chronic pain
Introduction
Chronic pain has recently been defined by the ICD-11 as pain that persists or recurs for longer than 3 months. Chronic primary pain is defined as pain in one or more anatomical regions that persists or recurs for longer than 3 months and is associated with significant emotional distress or functional disability. Chronic secondary pain syndromes are linked to other diseases as the underlying cause, where pain becomes a problem in its own right. In practice, the division between acute and chronic pain can be difficult to establish. This is particularly true in children and young people, and the committee felt that the looser (non-temporal) term ‘persistent pain’ is more commonly used in this group.
According to the British Medical Association briefing paper chronic pain: supporting safer prescribing of analgesics, chronic pain affects about 13% of adults in the UK, and about 8% of children experience severe pain. NICE has published a summary on the evidence base on medicines optimisation in chronic pain. A NICE guideline on chronic pain: assessment and management is in development. This guideline is intended to be used alongside existing NICE guidance for specific conditions that cause pain, including headaches, low back pain and sciatica, rheumatoid arthritis, osteoarthritis, spondyloarthritis, endometriosis and irritable bowel syndrome.
The aim of this review was to find out how effective cannabis-based medicinal products are in managing chronic pain, particularly when conventional treatment options have failed or not been tolerated. The review looked into the safety profile (including complications and contraindications) and examined what individual patient requirements, treatment durations, reviewing and stopping criteria need to be considered when prescribing cannabis-based medicinal products.
Review question
What is the clinical and cost effectiveness of cannabis-based medicinal products for people with chronic pain?
This review question also answered the following as part of the evidence review:
- What is the clinical and cost effectiveness of cannabis-based medicinal products for people with chronic pain?
- What are the adverse effects or complications of cannabis-based medicinal products for people with chronic pain?
- What are the contraindications, potential interactions and risks and cautions for use of cannabis-based medicinal products for people with chronic pain?
- What are the individual patient monitoring requirements, treatment durations, reviewing and stopping criteria, including how should treatment be withdrawn or stopped, for use of cannabis-based medicinal products for people with chronic pain?
The review protocol for this review question is in Appendix A. The PICO table below formed part of the search strategy to identify studies associated with chronic pain.
Table
PICO table.
Methods and process
This evidence review was developed using the methods and process described in Developing NICE guidelines: the manual (2018). Methods specific to this review question are described in the review protocol in Appendix B.
Declarations of interest were recorded according to NICE’s 2018 conflicts of interest policy.
A broad search strategy was used to identify all studies that examined the effectiveness of cannabis-based medicinal products in the treatment of intractable nausea and vomiting, chronic pain, spasticity and severe treatment-resistant epilepsy. The review protocol highlighted in Table 1 and Appendix A was used to identify studies associated with chronic pain.
For the adult population, randomised controlled trials (RCTs) and systematic review of RCTs were considered. The review protocol also specified that in the event of fewer than 5 RCTs being identified, prospective cohort studies would also be considered for inclusion.
For children, RCTs and systematic review of RCTs were considered. The review protocol also specified that in the event of fewer than 5 RCTs being identified, prospective and retrospective cohort studies would also be considered for inclusion. This is because the committee highlighted that there may be fewer studies performed in children.
Additional information on safety concerns and contraindications were obtained from the Summary of Product Characteristics and other relevant sources, such as the U.S Food and Drugs Administration.
Studies were also excluded if they:
- Examined the use of synthetic cannabinoids in schedule 1 of the 2001 regulations,
- Examined the use of smoked cannabis-based products
- Did not clearly report the amount of cannabis-based constituents in the intervention
The review protocol specified that where possible for adults, data would be stratified according to the ICD-11 definition of pain as primary or secondary pain. For primary pain, data was analysed according to whether it was chronic widespread pain, complex regional pain syndrome, chronic primary visceral pain or chronic primary musculoskeletal pain.
For secondary pain, the data was analysed according to whether it is chronic cancer-related pain, chronic postsurgical or posttraumatic pain, chronic neuropathic pain, chronic secondary visceral pain and chronic secondary musculoskeletal pain.
The review protocol also specifies that where possible, subgroup analyses would be conducted to explore the effectiveness of cannabis-based medicinal products in young people, children and babies, pregnant women and women who are breastfeeding, people with existing substance abuse and people with hepatic and renal failure.
The committee agreed that the clinical outcome that matters most is mean change in pain intensity. This is widely used and easily understood. The next most important outcomes were the proportion of patients who experienced pain relief of 30% or 50% or more from baseline. These are also direct measurements of pain. However, the committee felt that they are less descriptive outcomes; information is lost when converting continuous data into dichotomous data.
The next most important outcome is functional impairment caused by pain. This is arguably a more useful measurement compared to pain intensity because it captures the effect that pain has on people’s lives. However, functional impairment caused by pain is not measured often and when measured, it is usually measured in an inconsistent way across studies. Therefore, average pain intensity is more useful for meta-analysing outcomes across studies compared to functional impairment caused by pain.
After these direct measurements of pain, the committee were most interested in opioid sparing with a view to reducing adverse events caused by opioids.
The next most important clinical outcomes are those which are influenced by pain but are also influenced by other factors that may be unrelated to pain, such as mood. These outcomes include Patient Global Impression of Change and measurements of quality of life.
Clinical evidence
A total of 19,491 RCTs and systematic reviews were identified from the search. After removing duplicates, 9,341 references were screened on their titles and abstracts. 292 studies were obtained and reviewed against the inclusion criteria as described in the review protocol for chronic pain (Appendix A). Overall, 20 RCTs (14 parallel and 6 crossover) were included (see Appendix E for evidence tables). 272 references were excluded because they did not meet the eligibility criteria.
Because fewer than 5 RCTs were found for children, an additional search was conducted for observational studies. A total of 5,975 observational studies were identified from the search. After removing duplicates, 4,028 references were screened on their titles and abstracts. No studies were identified as being potentially relevant to chronic pain.
See Appendix E for evidence tables and Appendix J for excluded studies.
There were 20 RCTs, see table 2, summary of included studies.
No studies were identified which included the following subgroups:
- Pregnant women and women who are breastfeeding People with hepatic or renal failure
Quality assessment of clinical studies included in the evidence review
In this review, parallel RCTs and crossover RCTs were identified. The quality of the evidence was initially graded as high.
With regard to crossover studies, the committee identified 1 week as an adequate washout period. It should be noted that this could lead to symptoms of THC withdrawal, including heightened anxiety, which might obscure any potential analgesic effect of the study product.
See Appendix G for full GRADE tables and Appendix F for forest plots in situations where data have been meta-analysed.
Interventions
Of the 20 studies included, 5 looked at treatment of cancer pain. The included studies looked at the following interventions:
- Oromucosal spray containing 2.7 mg THC and 2.5 mg CBD per 100 microlitre actuation. This is abbreviated in this document to THC:CBD spray.
- Oromucosal spray containing 2.7 mg THC only per 100 microlitre actuation
Of the 20 studies included, 7 looked at treatment of neuropathic pain (including multiple sclerosis, peripheral neuropathic pain and neuropathic pain characterised by allodynia). The included studies looked at the following interventions:
- Oromucosal spray containing THC:CBD
- Oral delta-9-THC (dronabinol)
Of the 20 studies included, 3 looked at treatment of musculoskeletal pain (including rheumatoid arthritis, cramps and spasticity). The included studies looked at the following interventions:
- Oromucosal spray containing THC:CBD
- Oral delta-9-THC (dronabinol)
- Oral nabilone
Of the 20 studies included, 3 looked at treatment of visceral pain (including abdominal pain and oesophageal functional chest pain). The included studies looked at the following intervention:
- Oral delta-9-THC (dronabinol)
Of the 20 studies included, 2 looked at treatment of widespread pain (fibromyalgia). The included studies looked at the following interventions:
- Oral nabilone
- Vaporised 22.4 mg THC and <1 mg CBD
- Vaporised 13.4 mg THC and 17.8 mg CBD
- Vaporised <1 mg THC and 18.4 mg CBD
Summary of clinical studies included in the evidence review
Table
Blake 2006 Parallel RCT
See Appendix E for evidence tables and Appendix H for further information on adverse events.
As part of this evidence review, in addition to reviewing efficacy and safety data, studies were reviewed for information about patient monitoring and reviewing and stopping criteria when cannabis-based medicinal products were prescribed.
The interventions, doses, monitoring and stopping criteria are summarised in the table below:
Table
Oromucosal spray 2.7 mg THC with 2.5 mg CBD per 100 microlitre actuation (n=5)
Economic evidence
Included studies
No economic studies were included in this review.
Excluded studies
A global search conducted for this guideline returned 1,863 hits. 1 full paper was ordered for this review question and subsequently excluded. More detail is available in Appendix J.
Summary of studies included in the economic evidence review
No studies included.
Economic model
A de novo economic model was developed to address this review question. The model considered the CBMPs + the Standard of Care (SoC) versus the SoC alone. Subgroup analyses were conducted for specific treatments and for specific types of chronic pain where data were available to do so.
The economic model was comprised of five health states in each arm; on treatment response (OTR), on treatment no response (OTNR), discontinued with response (DR), discontinued with no response (DNR) and dead. In the SoC arm the “on treatment” states were nominal only, simply reflecting different levels of change from baseline observed in the underpinning trials. The model was run in monthly cycles over a lifetime time horizon and costs and QALYs were discounted at 3.5% per year.
Patients were categorised into one of the health states after one model cycle by combining the distribution of pain at baseline with the continuous outcomes from the clinical review for this question. Patients with a >30% response were assumed to remain as responders until they discontinued or died. The model calculated costs and QALYs from the distribution of pain scores within each health state, with lower pain scores having higher QoL and lower background management costs. Costs and QALYs associated with adverse events were also included, along with the costs of downstream radiofrequency denervation for the low back pain subgroup.
For all treatment and condition specific subgroups the model produced ICERs far in excess of the usually accepted £20,000–£30,000/QALY range. This was principally due to the modest treatment effects and the high and ongoing cost of treatment with CBMPs. The model had a number of limitations including the lack of long term data on almost all parameters but no plausible variations in any of the model’s input parameters produced ICERs close to £20,000–£30,000/QALY.
Details of the de novo economic model developed for this review question are available in Appendix I.
Summary of evidence
The summary of evidence reflects the evidence on effectiveness of cannabis-based medicinal products. Evidence summarises are stratified by population and reflect evidence that was significant. Further information on adverse events is also provided. The format of the summary of evidence is explained in the methods in Appendix B. Further information on adverse events is provided in Appendix I.
THC:CBD spray vs placebo
Commonly reported adverse events for THC:CBD spray included: dizziness, somnolence, nausea, vertigo and fatigue.
Further details of the quality assessments can be found here in the GRADE tables.
Subgroups were analysed and can been seen here in the forest plots.
Oral delta-9-THC (dronabinol), 7.5 to 16 mg per 24 hours vs placebo
Commonly reported adverse events for oral delta-9-THC (dronabinol) included: dizziness, vertigo, fatigue, nausea and headache.
Further details of the quality assessments can be found here in the GRADE tables.
Subgroups were analysed and can been seen here in the forest plots.
Oral nabilone, 1 to 2 mg per 24 hours vs placebo
Commonly reported adverse events for oral nabilone included: drowsiness, dry mouth, ataxia, confusion and headache.
Further details of the quality assessments can be found here in the GRADE tables.
Subgroups were analysed and can been seen here in the forest plots.
Oromucosal spray 2.7 mg THC only per 100 microlitre actuation, maximum 48 actuations per 24 hours vs placebo
Commonly reported adverse events for THC spray included: somnolence, dizziness, nausea, vomiting and confusion.
Further details of the quality assessments can be found here in the GRADE tables.
Vaporised 13.4 mg THC and 17.8 mg CBD vs placebo
Commonly reported adverse events for vaporised THC:CBD included: drug high, coughing, sore throat, bad taste and nausea.
Further details of the quality assessments can be found here in the GRADE tables.
See Appendix K for further information on the research questions’ PICOs.
The committee’s discussion of the evidence
Interpreting the evidence
The outcomes that matter most
Outcomes were discussed a priori. After reviewing the evidence, the opinion of the committee did not change. Outcomes that matter most are discussed in the Methods and process section.
The quality of the evidence
There was limited evidence of high quality. The main reason for this is that the maintenance dose duration is relatively short in most studies. The committee agreed that a maintenance dose duration of up to 6 weeks is unrealistic for assessing chronic pain treatments. Additionally, many studies did not provide details of methods for randomisation or blinding.
The majority of the RCTs are for CBD in combination with THC. There was only one RCT for THC alone and two for nabilone. There was no evidence for CBD alone and the preparation that had CBD with a small amount of THC (<1 mg) was poor quality.
Benefits and harms
There is evidence to suggest that CBD reduces chronic pain: Nabilone reduced functional impairment caused by pain compared with placebo in a population of 33 participants who had fibromyalgia. THC reduced mean functional impairment caused by pain in a population of 96 participants who had cancer. However, where cannabis-based medicinal products reduced chronic pain, the benefit is small and economic analysis shows that this compares poorly with the high costs of the intervention (see below).
There was high quality evidence which could not differentiate reduction in pain intensity between dronabinol and placebo in a population of 389 participants who had multiple sclerosis, abdominal pain or cramps.
The data could not differentiate THC:CBD for functional pain, change in opioid dose or quality of life. However, the committee considered these are outcomes to be less important compared with mean pain intensity which could not be differentiated between THC:CBD and placebo.
With regard to research recommendation 1, people who have fibromyalgia or persistent treatment-resistant neuropathic pain are often prescribed high doses of analgesia over long periods of time. This can be associated with adverse events including nausea, drowsiness, mood disturbance and fatigue. It is hoped that CBD might have an opioid-sparing effect and therefore reduce the incidence of adverse events such as these. The committee noted that of this significant population with chronic pain, around 15% are referred for specialist pain management. They also noted that this population is usually on many medications, including opioids and treatments for neuropathic pain. In this population, it is unclear whether cannabis-based medicinal products could improve safety by reducing doses of other medicines. Therefore, a research recommendation was made. The committee defined standard treatment as WHO pain ladder step 3: opioids plus adjuvants. RCTs should have at least 6 months follow-up to reflect the chronic nature of these conditions.
With regard to research recommendation 2, there is currently no evidence that explores whether the addition of cannabis-based medicinal products as an adjunct to standard care improves the pain experience in children with rare conditions experiencing persistent pain symptoms, for example children with intractable cancer-related pain or chronic pain associated with specific diseases such as epidermolysis bullosa. The reason for the lack of research so far is probably because there are relatively few children with these conditions. In addition, opioids may not control pain effectively in these conditions and may cause adverse effects. Therefore, a research recommendation was made. The committee defined ‘intractable cancer-related pain’ as cancer-related pain that does not respond to multiple interventions including non- pharmacological and drug therapies sufficiently to enable a reasonable quality of life. The committee defined standard care in this context as tertiary specialist pain/palliative care management. An additional benefit from such research could be a reduction in resource use.
The committee felt that CBD has the potential to be cost effective for all these research populations if they could be robustly demonstrated to improve quality of life and reduce resource use associated with complex conditions requiring standard tertiary specialist pain/palliative care management. For example, if children and young people with chronic pain achieved benefits sufficient for them to be able to receive their care in an outpatient rather than an inpatient setting.
See Appendix K for further information on the research questions’ PICOs.
Cost effectiveness and resource use
No published health economic analyses met the inclusion criteria for this review, but this area was prioritised for de novo economic modelling because the potential eligible population, and therefore potential resource impact, were deemed to be large. The committee considered that outcomes measuring change in pain were the most important in the clinical review and thought it important that the economic model structure should be directly tied to these outcomes. The clinical review provided both continuous (11 studies) and dichotomous (4 studies) data. A continuous model structure was chosen because continuous outcome data were more plentiful, because the continuous data approximated the dichotomous data very well under the assumption that treatment response was normally distributed and because it allowed the model to tie pain to costs and quality of life in a more detailed way.
The intermediate results from the model showed that ~54% of people in the cannabis arm and ~46% of people in the placebo arm achieved a 30% reduction in pain from baseline, while ~31% and ~25% achieved a treatment response of 50% respectively, which was similar to data observed in the clinical trials. Following an initial period of some treatment discontinuation, mean pain scores across the cohorts in both model arms settled into a steady state somewhat lower than the baseline level. The committee noted that adverse events contributed relatively little in terms of costs or quality of life decrements and that there were some savings in pain management costs associated with the treatment effect. These savings were small in comparison to the costs of cannabis based medicinal products (CBMPs), however. Net avoidance of invasive long-term treatments in the cannabis arm also contributed a negligible amount to cost savings.
Using THC:CBD spray, which is the cheapest CBMP with a publicly available price, the model produced an incremental cost-effectiveness ratio (ICER) of over £150,000/QALY gained over the standard of care, a value far higher than the commonly accepted decision threshold of £20,000–£30,000/QALY gained. The committee concluded that this finding was not surprising as CBMPs are not expected to extend life or be fundamentally disease modifying, treatment effects relating to symptom alleviation are modest (about a 0.4 improvement in pain on a 0 to 10 scale on average) and the cost of the treatments is high.
The committee noted the limitations of the model, including that only short term data were available from RCTs, that there were no robust estimates of resource use associated with different pain scores, that data on some parameters were extrapolated from indirect sources, that there were no good data linking either cannabis or pain scores to the downstream treatments that had been included in the model and that good quality data were lacking in some subgroups. These limitations were explored in sensitivity analyses showing that even under extreme assumptions, the model never produced ICERs close to those normally considered cost-effective. Furthermore, the probabilistic sensitivity analysis showed a 0% probability that CBMPs are cost-effective – using NICE’s ‘threshold’ of £20,000 to £30,000 per QALY over which treatments are not likely to be recommended for use in the NHS.
Overall, the committee considered the economic model to be directly applicable with minor limitations for decision-making. They considered that the CBMPs that they had seen evidence for would have to be around 8 times more effective (accrue 1.22 QALYs compared with 0.162 QALY in the base case) or 6 times less expensive (or some equivalent combination) or associated with very significant pain management savings for the average patient to bring the ICER down to an acceptable level. The committee was aware that given the findings from the clinical evidence, the additional effectiveness is unrealistic. Given how unlikely it would be to observe changes of this scale they concluded that, at current prices, these CBMPs do not represent an effective use of resources in the management of chronic pain. They therefore decided to make a recommendation against their use in the specific populations that were considered in the evidence base for this review. They discussed gaps in the evidence and made a series of recommendations for research into the use of CBMPs in specialised settings. Poor quality evidence on CBD alone was included in this review and the committee were aware of anecdotal evidence that many people with chronic pain are accessing this outside the NHS and reporting benefit. They therefore thought it important to include this intervention in their recommendations for research. They also noted that, although the clinical review had found some evidence showing no difference in opiate use between cannabis and standard of care, the trials were too short in duration to be reliable. As a matter of theory, this outcome might importantly influence decisions to prescribe CBMPs and could also influence their cost-effectiveness in certain populations. They therefore highlighted this is an outcome of interest.
Glossary
Cannabis-based medicinal products
In this guideline cannabis-based medicinal products include:
- cannabis-based products for medicinal use as set out by the UK Government in the 2018 Regulations
- the licensed products delta-9-tetrahydrocannibinol and cannabidiol (Sativex) and nabilone
- plant-derived cannabinoids such as pure cannabidiol (CBD)
- synthetic compounds which are identical in structure to naturally occurring cannabinoids such as delta-tetrahydrocannabinol (THC), for example, dronabinol.
Appendices
Appendix A. Review protocols
Review protocol for effectiveness of cannabis based medicinal products for people with chronic pain
Table
What is the clinical and cost effectiveness of cannabis-based medicinal products for people with chronic pain? What are the adverse effects or complications of cannabis-based medicinal products for people with chronic pain?
Appendix B. Methods
Priority screening
The reviews undertaken for this guideline all made use of the priority screening functionality with the EPPI-reviewer systematic reviewing software. This uses a machine learning algorithm (specifically, an SGD classifier) to take information on features (1, 2 and 3 word blocks) in the titles and abstract of papers marked as being ‘includes’ or ‘excludes’ during the title and abstract screening process, and re-orders the remaining records from most likely to least likely to be an include, based on that algorithm. This re-ordering of the remaining records occurs every time 25 additional records have been screened.
As an additional check to ensure this approach did not miss relevant studies, the included studies’ lists of included systematic reviews were searched to identify any papers not identified through the primary search.
Evidence synthesis and meta-analyses
Where possible, meta-analyses were conducted to combine the results of quantitative studies for each outcome. For continuous outcomes analysed as mean differences, where change from baseline data were reported in the trials and were accompanied by a measure of spread (for example standard deviation), these were extracted and used in the meta-analysis.
Because most of the studies reported odds ratios which could not be converted to risk ratios, all dichotomous outcomes were reported as odds ratios for consistency. Due to the nature of the data reported in the studies, absolute risks could not be calculated for the outcomes.
Evidence of effectiveness of interventions
Quality assessment
Individual RCTs and quasi-randomised controlled trials were quality assessed using the Cochrane Risk of Bias Tool. Each individual study was classified into one of the following three groups:
- Low risk of bias – The true effect size for the study is likely to be close to the estimated effect size.
- Moderate risk of bias – There is a possibility the true effect size for the study is substantially different to the estimated effect size.
- High risk of bias – It is likely the true effect size for the study is substantially different to the estimated effect size.
Each individual study was also classified into one of three groups for directness, based on if there were concerns about the population, intervention, comparator and/or outcomes in the study and how directly these variables could address the specified review question. Studies were rated as follows:
- Direct – No important deviations from the protocol in population, intervention, comparator and/or outcomes.
- Partially indirect – Important deviations from the protocol in one of the population, intervention, comparator and/or outcomes.
- Indirect – Important deviations from the protocol in at least two of the following areas: population, intervention, comparator and/or outcomes.
Methods for combining intervention evidence
Meta-analyses of interventional data were conducted with reference to the Cochrane Handbook for Systematic Reviews of Interventions (Higgins et al. 2011).
Fixed- and random-effects models (der Simonian and Laird) were fitted for all syntheses, with the presented analysis dependent on the degree of heterogeneity in the assembled evidence. Fixed-effects models were deemed to be inappropriate if one or both of the following conditions was met:
- Significant between study heterogeneity in methodology, population, intervention or comparator was identified by the reviewer in advance of data analysis. In analysis which included subgroups from more than 1 ICD classification of pain, random-effects were fitted to account for differences in populations.
- The presence of significant statistical heterogeneity in the meta-analysis, defined as I2≥50%.
Where data for multiple subgroups (i.e. different doses of medicinal cannabis) were combined, this was done in accordance to the advice given in the Cochrane Handbook for Systematic Reviews of Interventions.
Meta-analyses were performed in Cochrane Review Manager V5.3.
Numerical Rating Scale (NRS) is used to score pain intensity on a scale of 0 to 10. Visual Analog Scale (VAS) is used to score pain intensity on a scale of 0 to 100. The committee agreed that if a study only uses VAS, we should transform it to NRS by dividing the score by 10.
Combining groups within studies
When combining the arms of studies, we used Cochrane’s advice and formula for combining groups. In doing so, we assumed equal SD for the placebo and intervention arm.
Minimal clinically important differences (MIDs)
The guideline committee were asked to prospectively specify outcomes where they felt a consensus MID could be defined from their experience. The committee specified a key outcome is participant reported pain relief of 30% or greater. This is in line with the recommended measure of minimal important difference in pain intensity (IMMPACT 2005). For this measure, other dichotomous measures and measures of functional pain, the committee agreed that any statistically significant difference in outcomes would be of interest to them. Therefore, it was decided that the line of no effect was to be used as the MID (OR = 1 and mean or median difference = 0). For mean difference measures between arms reported on the visual analogue scale (VAS) or numerical rating scale (NRS), a clinically important difference of −0.8 was used. This is the same MID used by the Cochrane Pain, Palliative and Supportive Care Group (Mücke 2018) and similar to the median average MID for pain intensity reported in a systematic review of chronic pain intensity MIDs (Olsen 2018).
Therefore, a mean difference and confidence interval below −0.8 would show a clinical benefit to reduction in pain intensity in the treatment arm compared to the comparator.
GRADE for pairwise meta-analyses of interventional evidence
GRADE was used to assess the quality of evidence for the selected outcomes as specified in ‘Developing NICE guidelines: the manual (2018)’. Data from all study designs was initially rated as high quality and the quality of the evidence for each outcome was downgraded or not from this initial point, based on the criteria given in Table 1.
Table 1. Rationale for downgrading quality of evidence for intervention studies
The quality of evidence for each outcome was upgraded if any of the following three conditions were met:
- Data from non-randomised studies showing an effect size sufficiently large that it cannot be explained by confounding alone.
- Data showing a dose-response gradient.
- Data where all plausible residual confounding is likely to increase our confidence in the effect estimate.
Summary of evidence
The evidence is presented in the form of a table because the committee agreed in advance that effect sizes would be an important consideration. Summary of evidence is stratified by comparison and reflects evidence that was statistically significant.
Where the data are only consistent, at a 95% confidence level, with an effect in one direction, and the magnitude of that effect is most likely to meet or exceed the MID (i.e. the point estimate is not in the zone of equivalence). In such cases, we state that the evidence showed that there is an effect. In all other cases, we state that the evidence could not differentiate between the comparators.
Appendix C. Literature search strategies
A single systematic search was conducted for all of the questions within this evidence review between 19th December 2018 and 21st January 2019. The following databases were searched MEDLINE, MEDLINE in Process, MEDLINE e pub Ahead of print, Embase, (all via the Ovid platform), Cochrane Database of Systematic Reviews CENTRAL (all via the Wiley platform), and the HTA and DARE databases (both via the CRD platform). NICE inhouse RCT, systematic review, and observational filters were attached where appropriate.
The MEDLINE strategy is presented below. This was translated for other databases
- Medical Marijuana/
- cannabinoids/ or cannabidiol/ or cannabinol/ or cannabis/
- ((cannabi* or hemp or marijuana or marihuana) adj4 (medicine* or medicinal or medical or oil or oils or product* or extract* or therap* or CBD or vap* or spray* or inhal* or compound* or resin* or derivative*)).tw.
- (epidiolex* or cannabidiol* or cannabinoid*).tw.
- (sativex or THC:CBD spray or tetrabinex or nabidiolex).tw.
- (nabilone or cesamet).tw.
- (tilray* or bedrocan* or bedrobinol* or bedica* or bediol* or bedrolite*).tw.
- Dronabinol/
- (dronabinol* or marinol* or syndros*).tw.
- (9-ene-tetrahydrocannabinol* or 9enetetrahydrocannabinol*).tw.
- (THC or tetrahydrocannabinol*).tw.
- (“delta(1)-thc*” or “delta(1)-tetrahydrocannabinol*” or “delta(9)-thc*” or “delta(9)-tetrahydrocannabinol*”).tw.
- (9-delta-tetra-hydrocannabinol* or “9-delta-THC*” or “9 delta tetra hydrocannabinol*” or “9 delta THC*”).tw.
- (1-delta-tetra-hydrocannabinol* or “1-delta-THC*” or “1 delta tetra hydrocannabinol” or “1 delta thc*”).tw.
- THCa.tw.
- CBDa.tw.
- cannabinol*.tw.
- cannabigerol*.tw.
- cannabichromene*.tw.
- (tetrahydrocannabivarin* or THCV).tw.
- (cannabidivarin* or CBDV).tw.
- or/1–21
- animals/ not humans/
- 22 not 23
- limit 24 to english language
- Randomized Controlled Trial.pt.
- Controlled Clinical Trial.pt.
- Clinical Trial.pt.
- exp Clinical Trials as Topic/
- Placebos/
- Random Allocation/
- Double-Blind Method/
- Single-Blind Method/
- Cross-Over Studies/
- ((random$ or control$ or clinical$) adj3 (trial$ or stud$)).tw.
- (random$ adj3 allocat$).tw.
- placebo$.tw.
- ((singl$ or doubl$ or trebl$ or tripl$) adj (blind$ or mask$)).tw.
- (crossover$ or (cross adj over$)).tw.
- or/20–33
- Meta-Analysis.pt.
- Network Meta-Analysis/
- Meta-Analysis as Topic/
- Review.pt.
- exp Review Literature as Topic/
- (metaanaly$ or metanaly$ or (meta adj3 analy$)).tw.
- (review$ or overview$).ti.
- (systematic$ adj5 (review$ or overview$)).tw.
- ((quantitative$ or qualitative$) adj5 (review$ or overview$)).tw.
- ((studies or trial$) adj2 (review$ or overview$)).tw.
- (integrat$ adj3 (research or review$ or literature)).tw.
- (pool$ adj2 (analy$ or data)).tw.
- (handsearch$ or (hand adj3 search$)).tw.
- (manual$ adj3 search$).tw.
- or/35–48
- 34 or 49
- 19 and 50
- Observational Studies as Topic/
- Observational Study/
- Epidemiologic Studies/
- exp Case-Control Studies/
- exp Cohort Studies/
- Cross-Sectional Studies/
- Controlled Before-After Studies/
- Historically Controlled Study/
- Interrupted Time Series Analysis/
- Comparative Study.pt.
- case control$.tw.
- case series.tw.
- (cohort adj (study or studies)).tw.
- cohort analy$.tw.
- (follow up adj (study or studies)).tw.
- (observational adj (study or studies)).tw.
- longitudinal.tw.
- prospective.tw.
- retrospective.tw.
- cross sectional.tw.
- or/26–45
- 25 and 46
- 57 or 79
Searches to identify economic evidence were run on 20th December 2018 in MEDLINE, MEDLINE in Process, MEDLINE e pub Ahead of print, Econlit and Embase (all va the Ovid platform), NHS EED and the Health Technology Assessment Database (via the CRD platform. NICE inhouse economic evaluation and Quality of Life filters were attached to lines 1 to 25 of the core strategy (lines 1 to 25 of the Medline version shown above) in the Medline and Embase databases. The Medline version of the filters is displayed below.
Economic evaluations
- Economics/
- exp “Costs and Cost Analysis”/
- Economics, Dental/
- exp Economics, Hospital/
- exp Economics, Medical/
- Economics, Nursing/
- Economics, Pharmaceutical/
- Budgets/
- exp Models, Economic/
- Markov Chains/
- Monte Carlo Method/
- Decision Trees/
- econom$.tw.
- cba.tw.
- cea.tw.
- cua.tw.
- markov$.tw.
- (monte adj carlo).tw.
- (decision adj3 (tree$ or analys$)).tw.
- (cost or costs or costing$ or costly or costed).tw.
- (price$ or pricing$).tw.
- budget$.tw.
- expenditure$.tw.
- (value adj3 (money or monetary)).tw.
- (pharmacoeconomic$ or (pharmaco adj economic$)).tw.
- or/1–25
Quality of Life
- “Quality of Life”/
- quality of life.tw.
- “Value of Life”/
- Quality-Adjusted Life Years/
- quality adjusted life.tw.
- (qaly$ or qald$ or qale$ or qtime$).tw.
- disability adjusted life.tw.
- daly$.tw.
- Health Status Indicators/
- (sf36 or sf 36 or short form 36 or shortform 36 or sf thirtysix or sf thirty six or shortform thirtysix or shortform thirty six or short form thirtysix or short form thirty six).tw.
- (sf6 or sf 6 or short form 6 or shortform 6 or sf six or sfsix or shortform six or short form six).tw.
- (sf12 or sf 12 or short form 12 or shortform 12 or sf twelve or sftwelve or shortform twelve or short form twelve).tw.
- (sf16 or sf 16 or short form 16 or shortform 16 or sf sixteen or sfsixteen or shortform sixteen or short form sixteen).tw.
- (sf20 or sf 20 or short form 20 or shortform 20 or sf twenty or sftwenty or shortform twenty or short form twenty).tw.
- (euroqol or euro qol or eq5d or eq 5d).tw.
- (qol or hql or hqol or hrqol).tw.
- (hye or hyes).tw.
- health$ year$ equivalent$.tw.
- utilit$.tw.
- (hui or hui1 or hui2 or hui3).tw.
- disutili$.tw.
- rosser.tw.
- quality of wellbeing.tw.
- quality of well-being.tw.
- qwb.tw.
- willingness to pay.tw.
- standard gamble$.tw.
- time trade off.tw.
- time tradeoff.tw.
- tto.tw.
- or/1–30
A search of the MHRA was undertaken on the 24th January 2019 to look for safety updates, alerts and recalls. The search terms are displayed below.
- Sativex
- Dronabinol
- Epidiolex
- THC:CBD spray
- Abalone
- Tetrabinex
- Nabidiolex
- Cesamet
- Tilray
- Bedrocan
- Bedrobinol
- Bedica
- Bediol
- Bedrolite
- Marinol
- Syndros
- THC
- Tetrahydrocannabinol
- Cannabinol
- Cannibigerol
- Cannabichromene
- Tetrahydrocannabivarin
- Cannabidivarin
Quality of Life
- “Quality of Life”/
- quality of life.tw.
- “Value of Life”/
- Quality-Adjusted Life Years/
- quality adjusted life.tw.
- (qaly$ or qald$ or qale$ or qtime$).tw.
- disability adjusted life.tw.
- daly$.tw.
- Health Status Indicators/ (22343)
- (sf36 or sf 36 or short form 36 or shortform 36 or sf thirtysix or sf thirty six or shortform thirtysix or shortform thirty six or short form thirtysix or short form thirty six).tw.
- (sf6 or sf 6 or short form 6 or shortform 6 or sf six or sfsix or shortform six or short form six).tw.
- (sf12 or sf 12 or short form 12 or shortform 12 or sf twelve or sftwelve or shortform twelve or short form twelve).tw.
- (sf16 or sf 16 or short form 16 or shortform 16 or sf sixteen or sfsixteen or shortform sixteen or short form sixteen).tw.
- (sf20 or sf 20 or short form 20 or shortform 20 or sf twenty or sftwenty or shortform twenty or short form twenty).tw.
- (euroqol or euro qol or eq5d or eq 5d).tw.
- (qol or hql or hqol or hrqol).tw.
- (hye or hyes).tw.
- health$ year$ equivalent$.tw.
- utilit$.tw.
- (hui or hui1 or hui2 or hui3).tw.
- disutili$.tw.
- rosser.tw.
- quality of wellbeing.tw.
- quality of well-being.tw.
- qwb.tw.
- willingness to pay.tw.
- standard gamble$.tw.
- time trade off.tw.
- time tradeoff.tw.
- tto.tw.
- or/1–30
MHRA search
Table
Alerts and recalls for drugs and medical devices Drug safety update
Search history – observational studies
Search history – observational studies: Medline search
Appendix D. Clinical evidence study selection
RCTs and systematic reviews of RCTs search
Observational studies search for children
Appendix E. Clinical evidence table
E.1. Parallel RCTs
Download PDF (1.6M)
E.2. Crossover RCTs
Download PDF (1.4M)
Appendix F. Forest plots
THC:CBD spray vs placebo
Proportion of people who experienced pain relief of 30% or more from baseline
Proportion of people who experienced pain relief of 50% or more from baseline
Mean average pain intensity: Numerical Rating Scale (0 to 10) or Visual Analogue Scale (0 to 100)/10 converted to NRS
Functional impairment caused by pain: Brief Pain Inventory - Short Form
Change in analgesics: daily change in total dose, morphine equivalents
Change in analgesics: daily change in breakthrough dose, morphine equivalents
Change in analgesics: daily change in breakthrough dose, morphine equivalents
Patient Global Impression of Change (dichotomous)
Patient Global Impression of Change (continuous)
Quality of life: EQ-5D index
People experiencing adverse events, all-causality
People experiencing serious adverse events, all-causality
People experiencing serious adverse events, treatment-related
Withdrawals due to adverse events, all-causality
Adverse events: psychosis, such as hallucinations, delusions, confused and disturbed thoughts, or lack of insight and self-awareness
Oral delta-9-THC (dronabinol), 7.5 to 16 mg per 24 hours vs placebo
Mean average pain intensity: Numerical Rating Scale’ (0 to 10) or Visual Analogue Scale (0 to 100)/10 converted to NRS
People experiencing adverse events, all-causality
People experiencing serious adverse events, all-causality
Withdrawals due to adverse events, all-causality
Oral nabilone, 1 to 2 mg per 24 hours vs placebo
People experiencing adverse events, all-causality
Withdrawals due to adverse events, all-causality
Appendix G. GRADE tables
THC:CBD spray vs placebo
Oral delta-9-THC (dronabinol), 7.5 to 16 mg per 24 hours vs placebo
Oral nabilone 1 to 2 mg per 24 hours vs placebo
Oromucosal spray 2.7 mg THC only per 100 microlitre actuation, maximum 48 actuations per 24 hours vs placebo
Vaporised 22.4 mg THC and <1 mg CBD vs placebo
Vaporised 13.4 mg THC and 17.8 mg CBD vs placebo
Vaporised <1 mg THC and 18.4 mg CBD vs placebo
Appendix H. Adverse events
THC:CBD spray vs placebo
Table
Phase A (standard parallel RCT) Treatment-related: all severities
Oral delta-9-THC (dronabinol), 7.5 to 16 mg per 24 hours vs placebo
Oral nabilone (1 to 2 mg per 24 hours vs placebo)
Table
All-causality Nabilone (13 patients in the arm): one moderate transient weakness of the lower limbs (1), mild drowsiness (2), acute relapse of multiple sclerosis (1)
Oromucosal spray THC only per 100 microlitre actuation, maximum 48 actuations per 24 hours vs placebo
Table
Treatment-related reported by ≥3 patients THC (58 patients in the arm): somnolence (8), dizziness (7), confusion (1), nausea (4), vomiting (4), raised gamma GT (5), hypercalcaemia (0), hypotension (0).
Vaporised 22.4 mg THC and <1 mg CBD vs placebo
Table
All-causality THC and CBD (20 patients in the arm): Drug high (16), coughing (14), sore throat (2), bad taste (5), dyspnoea (0), dizzy (3), headache (1), nausea (3), vomiting (0), sleepy (1)
Vaporised 13.4 mg THC and 17.8 mg CBD vs placebo
Table
All-causality THC and CBD (20 patients in the arm): Drug high (16), coughing (14), sore throat (7), bad taste (6), dyspnoea (1), dizzy (4), headache (2), nausea (6), vomiting (0), sleepy (0)
Vaporised <1 mg THC and 18.4 mg CBD vs placebo
Table
All-causality THC and CBD (20 patients in the arm): Drug high (8), coughing (13), sore throat (1), bad taste (5), dyspnoea (0), dizzy (2), headache (3), nausea (1), vomiting (1), sleepy (1)
Appendix I. Health economic analysis
Introduction
Chronic pain is common in the UK general population but has a heterogeneous aetiology. A recent epidemiological study found that roughly 43.5%, 28 million people in the UK general population were expected to have “severe and chronic pain that is unresponsive to treatment”. Treatment options vary widely depending on the cause of the pain but their effectiveness and side effects vary widely and there is very significant unmet clinical need in the population group whose pain is not adequately controlled by these conventional options. Some chronic pain patients self treat with cannabis based products purchased as health food supplements or online and there is widespread interest in whether Cannabis Based Medicinal Products (CBMPs) should be prescribed on the NHS. However, it is currently very rare for patients with chronic pain to be treated with CBMPs on the NHS.
The CBMPs that are currently on the market could cost several thousand pounds per patient per year, based on publicly available sources for price. This, along with the considerations above meant that the potential resource impact of a positive recommendation in this area could be extremely high. The committee therefore prioritised this question for de novo economic modelling as any positive recommendation would need to be underpinned by robust health economic evaluation.
Methods
Decision problem
The population of interest were people with chronic pain whose pain was not adequately controlled by conventional management. Since no CBMPs have a licence for treating chronic pain, it would not be appropriate to compare them to conventional management. Instead the model has 2 strategies, CBMP + usual care and usual care.
In the base case, the model considers all people with chronic pain as an aggregated group, which is broken down by pain aetiological subgroups in sensitivity analysis. The different aetiologies, decided upon by the committee following review of the available clinical evidence, were neuropathic pain, cancer pain and musculoskeletal pain. Separate analyses were conducted for different CBMPs; THC:CBD spray, oral dronabinol, oral nabilone and oromucosal THC.
Model Structure
The committee indicated that if CBMPs were to be used in the chronic pain population, they would be trialled for a month and discontinued if patients did not achieve a 30% reduction in pain from baseline as this is a well accepted Minimally Clinically Important Difference (MCID) in this population and a threshold that had been reported by studies in the clinical review. They indicated that a small proportion of patients who did not achieve a treatment response of 30% would remain on treatment if they felt they were getting some benefit from it. The 30% improvement threshold is based on the expert opinion at the committee. This parameter is only used to determine the continuation of treatment. The treatment response is based on the absolute NRS changes in the model.
We built a decision analytic model with five Markov states in each model arm; on treatment response (OTR), on treatment no response (OTNR), discontinued with response (DR), discontinued with no response (DNR) and dead. After being initially assigned to a Markov state through treatment effects, patients could transition from OTR to DR and from OTNR to DNR and patients could die from any state but no other transitions were possible (Figure 1). We adopted the same structure as this for the placebo arm because it is logical to operationalise the treatment and placebo effects within the same model structure, but people do not incur treatment cost in the usual care arm of the model despite nominally occupying a nominal “on treatment” state. This structural choice is not expected to have affected any of the results as the total distribution of patients’ pain scores within the usual care arm of the model is unaffected by grouping patients with higher and lower distributions into arbitrary Markov states in this way.
Effect Engine
Distribution of treatment effects
We chose to model treatment effects within our model using continuous (mean changes in pain score) rather than dichotomous (proportion of people achieving a 30% response) data on treatment effects from the clinical review. This decision was made firstly because more trials reported mean changes in pain score rather than proportions of people achieving at least a 30% improvement in pain and secondly because it provided a more detailed breakdown of treatment effects for us to examine the influence of CBMPs on costs and quality of life across the whole distribution of pain scores at different time points in the model.
We assumed that treatment effects would be normally distributed and tested this assumption by conducting simulations. There were two studies from the clinical review that allowed us to test how well normally distributed treatment effects would match empirically observed dichotomous outcomes, (Langford 2013 and Portenoy 2012). These were two of the larger studies included in the systematic review. Using the Langford data, we assumed baseline pain was normally distributed (we did not need to truncate this data to fit between 0 and 10 because of the relatively tight confidence intervals) and simulated 60,000 theoretical patients based on baseline mean and SD pain score. We then added a placebo or active treatment effect to each theoretical patient, randomly assigning values from a normal distribution with mean and SD taken from the change from baseline data in that study. Using the simulated data, we calculated the proportion of patients that received a 30% and 50% improvement in pain and compared them to the data from the trial. The results are in Table 2.
Table 2. Langford 2013 response data compared with normal distribution estimates
We then repeated this process for the Portenoy 2012 data, which provided dichotomous response data at many more levels of response. The data are in Figure 2.
Figure 2. Portenoy response data compared with normal distribution estimates
Based on the fit of these data and their experience, the committee felt that it was clinically plausible that treatment effects are well approximated by a normal distribution.
Application of treatment effects
As detailed in the Natural History section below, the distribution of baseline pain in both model arms was calculated using a beta distribution to assign patients into 200 ‘bins’ representing each 0.05 pain increment from 0.025 to 9.975 on the NRS scale.
We calculated patients’ pain scores after treatment by combining the beta distribution of baseline score and the normal distribution for treatment effects. We were then able to use the normally distributed treatment effects to calculate the proportion of patients moving from each bin to every other bin.
Separate post-treatments distributions were then calculated for patients who achieved a >=30% treatment response and those who did not. This was implemented by calculating whether the difference between 2 bins was >=30% or not when combining distributions. This also allowed calculation of the proportion of the overall cohort achieving a treatment response. Calculating the pain distribution of those who had achieved a >=30% response and those who had not was important for assigning costs and utility values to those continuing treatment with CBMPs beyond the first cycle of the model.
We also calculated distributions for patients who discontinued due to lack of treatment response and patients who did achieve a response but discontinued for other reasons. We made the assumption that patients who did not continue treatment would drop back to baseline in both arms of the model. These two distributions combined are therefore equivalent to the distribution at baseline, but divided into patients who would have achieved a treatment response, and those who would not. The separation of these distributions was necessary because 10% of partial responders are assumed to continue with treatment and because we assumed that patients discontinuing from the OTR health state would transition to the DR health state and patients discontinuing from the OTNR health state would discontinue to the DNR health state.
Cycle Length, Discount Rate and Time Horizon
We adopted a 4-week cycle length as treatment effects were often reported over this time in clinical trials. We adopted a discount rate of 3.5% for both costs and benefits and a lifetime time horizon in line with the NICE reference case. The data available to populate the model were typically short term so we adopted a shorter time horizon in sensitivity analysis.
Input parameters
Natural History
The distribution of baseline pain in both model arms was calculated using a beta distribution to assign patients into 200 ‘bins’ representing each 0.05 pain increment from 0.025 to 9.975 on the NRS scale. We used this distribution because it is not possible for a person to have a pain score below 0 or above 10. Using the ‘method of moments’ formulae, we converted the mean and SD of baseline pain from a large epidemiological study (Farrar 2001) to the alpha and beta parameters necessary for the distribution. We also used the average age and sex from this study to calculate utility values associated with each NRS score (see Utilities section).
In the base case, the assumption was made that pain score does not change over time (unless in response to treatment). This assumption was relaxed in sensitivity analysis by including capacity for increasing or decreasing pain score. Since a linearly changing score would mean that almost all of the cohort would end up with a pain score of either 0 or 10, we modelled the natural history such that pain asymptotically approaches 0 and 10. This was achieved by first specifying a “mean change in pain score per year”. This was used to calculate a “hazard ratio”, by dividing baseline pain score and pain score after 1 year by 10, converting into instantaneous “rates”, and taking the ratio between the two. The resulting value was converted to a HR per cycle of the model. This was then applied to pain scores in each cycle, by converting scores into “rates”, applying the “HR” and then converting back to pain scores.
Treatment effects
We obtained treatment effect data from the systematic review for this review question for four separate cannabis based medicinal products (Table 4). These were either derived from single studies or from meta-analyses. Response in the SoC arm of the model was set equal to the control arm from Langford, the largest study in the review, in the base case. Treatment effect data were added to response in the SoC arm to calculate response in the cannabis arm. See appendix E and AppendixF for details.
Table 4. Baseline response and treatment effect data from the clinical review
Discontinuation from Response
No direct, long term data on discontinuation were available for this population so we explored several options in the model. In the base case, discontinuation from response data were obtained from a large, publicly available individual patient dataset (Messina et al. 2017) on patients with advanced MS being treated with THC:CBD spray. These patients were treated for a period of 1 month with responders remaining on treatment and non-responders discontinuing. We selected only the responders, subtracted 28 days from the total time on treatment, converted the time on treatment from days to years and performed survival analysis on these patients where discontinuations were classed as events. Based on AIC/BIC statistics we selected a gompertz parametric curve to use within our economic model.
Table 5. Model fit statistics for discontinuation survival curve
While this dataset relates to the MS population rather than the population with chronic pain, the results indicate that THC:CBD spray is generally well tolerated and that treatment benefit appears to persist, with 80% of responders still being on treatment after 2 years. The most common reasons for discontinuation among those that responded were lack of effectiveness, adverse events or a combination of the two. We set up the model to use an alternative discount rate of 3.1% per cycle, calculated from Hoggart 2015, a study that was specific to cannabis use in chronic pain in scenario analysis.
In the base case we applied the discontinuation curve equally to responders in both the standard of care and active treatment arms of the model but explored no discontinuation and differential discontinuation in sensitivity analysis. For the differential discontinuation we modelled discontinuation by application of a hazard ratio. The hazard ratio was derived by creating an identical dataset to that in Messina 2017 but treating all patients who discontinued for adverse events alone as censors. This dataset was compared with the original using a cox proportional hazards model (HR = 0.482, se=0.06, proportional hazards assumption not rejected). The interpretation of this is that lower discontinuation would be expected in the standard of care arm because people cannot discontinue from pain response through adverse events alone. We included another option which fitted a competing risks model to the data, coding adverse events alone as a separate, competing risk to other discontinuations. We followed the methodology in section 6.3 of the CRAN-R documentation on the flexsurv packagea but used a gompertz model instead of the Weibull example given. The survival curve for the CBMP arm took account of both competing risks whereas the survival curve for the SoC arm included only non-adverse event related discontinuations. The competing risks model produced survival estimates that were very similar to those produced using the hazard ratio method outlined above. There were no deaths recorded in the dataset although there were a number of censoring events with no reason recorded and it is possible that some of these were in fact deaths. By handling deaths separately from discontinuation it is possible that there is a small amount of double counting in the economic model. Given the relatively low average age in the dataset and therefore low mortality rate, and the fact that this issue would apply to both model arms, we assessed this particular limitation as minor.
Clearly there are limitations with all these approaches but in the absence of long term data on changes in response in either the active treatment or standard of care arm the committee acknowledged that they were the best available, noted them as limitations and explored them in sensitivity analysis.
Mortality
There is no data available on whether treatment with CBMPs affect mortality and they are not expected to be fundamentally disease modifying. We therefore did not vary mortality by model arm but modelled overall mortality by applying an SMR of 1.32 (0.08) for people with chronic pain from an epidemiological study (Torrance 2006) to standard population level life tables published by the Office for National Statistics.
Downstream treatments
While CBMPs are not expected to be fundamentally disease modifying, and the clinical review identified no randomised evidence that their use spares other medication, we were interested in whether their potential to reduce or delay invasive treatments would influence the results of the economic model. The committee advised us that the only invasive treatment that was common enough to potentially influence the model’s results was radiofrequency denervation (RFD) for people with chronic low back pain. This section is therefore only relevant when considering the patient population with low back pain. This part of the model was switched off for other subgroups.
Theoretically, any patient with chronic low back pain and an NRS greater than 5 is eligible for RFD and the committee estimated that around 10% of the eligible population might be trialled for RFD per year. The methodology for applying the costs and benefits of RFD was adapted from that employed in the NICE guideline on Low back pain and sciatica in over 16s: assessment and management..
The trial for RFD consists of administration of a diagnostic block which is either positive or negative and to which some patients receive a HRQoL benefit (modelled as the same level of benefit as full RFD, as in the low back pain analysis) for prolonged response for the duration of that response. Negative patients will not receive RFD and 10% of positive patients will decline it. RFD and prolonged response to diagnostic block are implemented as a series of simultaneous tunnel states (i.e. they exist in parallel to the main on treatment/off treatment states). In each cycle, the % of patients with a pain score >5 is calculated using patient distributions, and this is used to determine the number of patients who undergo diagnostic block. Because of the tunnel state structure, the model ensures patients who are already in RFD or prolonged diagnostic block states cannot undergo diagnostic block again.
The QALY gain for RFD is determined by first applying the treatment effects from the NG59 model uniformly (having no specific evidence of non-uniformity) to each level of baseline pain within the initial distribution. The weighted average utility difference between the resulting distribution and the initial distribution is then taken and applied to the proportion of people who respond to diagnostic blocks or are in receipt of RFC benefit in any given cycle, with the treatment effect of RFD lasting two years. 10% of people who receive the full two years of RFD benefit were assumed to undergo repeat RFD after this time. The relevant parameters are in Table 6.
Adverse Events
We obtained adverse event and serious adverse event rates from a systematic review of patients being treated with CBMPs (Wang 2008) and used these data to calculate the events that occurred per cycle in the model. A wide range of non-serious adverse events were reported in this study but for simplicity we assumed them to be distributed among the five most important events selected by the committee; dizziness, dry mouth, fatigue, headache and nausea. We re-scaled the incidence of these five adverse events so that their sum matched the total event rate. Serious adverse events were assumed to be homogenous.
We assumed that all adverse events would be short term in nature, lasting only a few days and sourced temporary health related quality of life decrements from studies that reported the five most important adverse events selected by the committee. A quality of life decrement for grade 2 vomiting was used as a surrogate for serious adverse events because this was the most common non-condition specific adverse event reported in the Wang 2008 study. Grade 2 events are often not classified as serious in papers outside the CBMP field so it is possible that we may have slightly underestimated the QoL decrement associated with treatment related adverse events. 50% of non serious adverse events were assumed to incur a GP appointment and serious adverse events were assumed to incur an A&E visit, with 50% incurring a trip in an ambulance.
To obtain the adverse events for each Markov state in both arms of the model where patients were not on treatment, we multiplied the per cycle event rates for serious and non-serious AEs by the reciprocal of relative risks associated with different forms of CBMP in the clinical review. For patients in the on-treatment states these adverse event rates are unadjusted.
The above assumptions are subject to very serious limitations but, taken together, provide a rough estimate of the scale of the effect that adverse events have on the cost-effectiveness results. Sensitivity analyses including and excluding adverse events and varying input data to the extremes of their confidence intervals were undertaken. Since the only evidence we had indicated that adverse events are relatively rare and short term in nature, they are not expected to materially affect the cost-effectiveness of CBMPs.
Costs
Treatment costs for THC:CBD spray and nabilone were taken from the Drug Tariff with daily doses being taken from representative studies from the clinical review (Langford 2013 and Skrabek 2008 respectively). There are currently no publicly available UK prices for dronabinol or for the various Bedrocan products but the overall cost per patient is expected to be higher than that for THC:CBD spray.
The committee informed us that patients treated with CBMPs might expect to receive four additional outpatient visits within the first year and two outpatient visits in subsequent years to monitor their medication. Outpatient visits were costed at £147 (NHS Reference Costs 2017/18 - non-admitted face-to-face consultant-led attendance, follow-up – pain management).
In line with methodology that had been employed in modelling the spasticity question for this guideline, we assigned resource use to different levels of pain NRS. The committee confirmed that this approach was reasonable and considered that improvement in pain levels might lead to a resource saving in pain management costs. From their clinical experience they estimated the number of community based visits, outpatient clinic visits, A&E visits, hospital admissions and home care visits associated with five broad pain levels, NRS 0–2, NRS 2–4, NRS 4–6, NRS 6–8, NRS 8–10. The overlapping naming is caused by dividing an 11-point scale by five. The overall management cost for a given Markov state in a given cycle is the weighted average of their pain distribution rounded to the nearest fifth of the NRS scale multiplied by these costs. Given the uncertainty inherent in estimating background management costs in this way, these parameters were subject to extreme sensitivity analyses. We adjusted home care costs to account for the proportion that were self funded using data from NICE’s guideline on Parkinson’s disease health economics report.
Adverse event costs are discussed in that section.
Utilities
Utilities associated with each NRS pain level were sourced from a utility study that included 2,719 patients with chronic neuropathic pain (Gu 2012). This study provided dummy variable regression coefficients for each NRS level as well as age, gender and the constant. As well as having been collected from a large and broadly representative sample, the committee agreed that these data had face validity. The per cycle QALYs for each Markov state were the weighted average of the pain distribution and these utility values, with pain scores for the individual bins being rounded to the nearest integer.
Adverse event disutilities were obtained from a utility study which aimed to estimate the disutility associated with a series of common adverse events in patients with breast cancer. The patient group is clearly indirect and, as shown in the table below, several assumptions were necessary to operationalise adverse events in the model but as AEs were typically short term and non-severe, these limitations are not expected to materially influence the model’s results. Please see the adverse events section for the relevant input data.
Table 10. Utility regression model coefficients for chronic pain
Sensitivity and scenario analyses
A large number of one-way and multi-way deterministic sensitivity analyses were conducted in order to test how sensitive the model’s conclusions were to uncertainties in its input parameters. Probabilistic sensitivity analysis, where the model was run thousands of times with input parameters being sampled from appropriate probability distributions was also conducted to test the sensitivity of the model to combined statistical uncertainty. Pre-specified scenario analysis were:-
- Using the costs and effects of nabilone instead of THC:CBD spray
- Using treatment effect data for the neuropathic pain subgroup
- Using treatment effect data for the cancer pain subgroup
- Using treatment effect data for the musculoskeletal pain subgroup
- Using discontinuation data from the Hoggart 2015 chronic pain study instead of the Messina 2017 MS individual patient data
- Using control arm response from Portenoy 2012 instead of Langford 2013
- Excluding the background management costs for chronic pain
- Not allowing a proportion of sub<30% responders to continue with treatment
- Including RFD as a downstream treatment for low back pain
- Declining the treatment effect over time by reducing mean pain to match placebo
- Declining the placebo effect (change from baseline) in both arms so that pain returns to baseline after 2 years
- Allowing differential discontinuation from response in the standard of care arm equalling the hazard ratio
- Assuming no discontinuation from response in either arm
- All adverse events halved
- All adverse events doubled
- All pain management costs halved
- All pain management costs doubled
- All QoL coefficients set to high limits of confidence intervals
- All QoL coefficients set to low limits of confidence intervals
- Competing risks model from Messina for discontinuation
- −0.55 treatment effect to force the model to produce a mean difference equal to the input treatment effect of −0.44
Results
The results section will focus on treatment with THC:CBD spray unless otherwise noted as this is the treatment with the most robust clinical evidence. The patient group will be all patients with chronic pain unless otherwise noted.
Intermediate Results
The model’s intermediate results show that after the initial trial of treatment period, 54% of patients in the cannabis arm and 46% of patients in the standard of care arm achieved a 30% reduction from baseline while 31% and 25% achieved a 50% reduction respectively. These data were similar to data observed in the clinical trials. There is some discontinuation from response and then the model settles into a steady state where about 43% and 37% of patients remain as responders respectively (see Figure 4). Similarly, the graph of mean cohort pain over time shows an initial drop, followed by a slight increase and then a steady state of 4.9 for the cannabis arm and 5.2 for the standard of care with the slight increase being the result of the aforementioned discontinuation (see Figure 4). This is somewhat lower than the mean treatment effect and that is because only patients with a 30% improvement are assumed to continue treatment and carry on receiving the benefits. All other patients drop back to baseline after ending treatment. We set up a sensitivity analysis to increase the mean difference between the arms to match the input effectiveness data.
Figure 4. Intermediate model results
The relatively small difference between the intermediate outcomes in the model reflects the modest effectiveness of cannabis observed in the clinical review.
Figure 5. Total lifetime undiscounted costs by broad area and model arm
It can be seen from Figure 5 that cannabis only results in very small resource savings through reduction in pain scores and small increases in adverse event costs. These values are overwhelmed by the cost of cannabis treatment, however, along with a modest increase in monitoring costs.
Cost-utility Results
In the base case (THC:CBD spray costs and effects, overall chronic pain population, discontinuation data from Messina 2017, SoC response from Langford 2013, no treatment effect or SoC decline over time, no differential discontinuation from response, lifetime time horizon, discounting at 3.5% for both costs and benefits) the model produced incremental costs of £24,474 and incremental QALYs of 0.162 and therefore an ICER of £151,431/QALY gained.
The results for the mean of the probabilistic sensitivity analysis were very similar to this and that analysis found a 100% probability that cannabis is more effective, a 100% probability that it is more expensive and a 0.0% probability that cannabis is cost effective over standard care at the commonly accepted thresholds of £20,000 and £30,000/QALY gained.
Sensitivity and scenario analysis results
The tornado diagram in Figure 7 shows how the ICER changes in response to high and low values in important input parameters. The high and low values are typically limits of confidence intervals or other values selected to represent extreme scenarios. It can be seen from this diagram that no plausible variations in individual model parameters meaningfully affect the ICER.
Figure 7. Tornado Diagram (most influential parameters)
Scenario analyses either involve changing the source data of input parameters, the structural assumptions of the model or groups of input parameters to represent, for example, ‘best’ and ‘worst’ case results. Full descriptions of the scenario analyses are available in that section above.
Table 12. Results of scenario analyses
It can be seen from Table 12 that no scenario analyses bring the ICER close to £20,000–£30,000 per QALY gained. The ICER is lower for nabilone than for THC:CBD spray, but the clinical effectiveness data are much more uncertain and only available for a patient group with ‘widespread pain’.
Discussion
The economic model was characterised by a number of limitations; the clinical input data were of low quality, input parameters were largely drawn from short term trials and extrapolated into the longer term, adverse events and background pain management incorporated several committee assumptions relating to their associated costs and HRQoL effects, costing the standard of care was ignored as cannabis was modelled as an add-on treatment and the only data we had on opioid sparing showed no effect and we had no data on pain progression or the behaviour of the placebo effect over time. Nevertheless, no plausible variations in any of the model parameters or structural assumptions produced ICERs remotely near the commonly accepted cost-effectiveness threshold of £20,000–£30,000 per QALY gained. This is principally because the CBMPs that are currently on the market and for which there is any clinical evidence are quite expensive, costing upwards of £4,000 per patient per year and only provide very modest clinical benefits. Indeed, these products would have to either be around 8 times more effective (accrue 1.22 QALYs compared with 0.162 QALY in the base case) or around 6 times less expensive or some equivalent combination of the 2 for the model to produce ICERs within the range normally accepted by NICE committees. The committee was aware that given the findings from the clinical evidence, the additional effectiveness is unrealistic.
There are a number of products not examined by this analysis because no data were available on their effectiveness or UK price; pure CBD oil, Bedrocan products and dronabinol have been omitted but could be included in an updated version of the model once such data become available.
Appendix J. Excluded studies
Clinical studies
Table
No outcomes of interest [The results are in a format that it’s not possible to data extract: All the data is either given as a narrative account or in form of graphs. This also includes adverse events.]
Economic studies
Appendix K. Research recommendations
- 1.
For adults with fibromyalgia or persistent treatment-resistant neuropathic pain, what is the clinical and cost effectiveness of cannabidiol (CBD containing no or traces of THC) as an add-on to standard treatment?
There are no RCTs that compare CBD (either as a pure product or containing traces of THC) with standard treatment to standard treatment for fibromyalgia or for persistent treatment-resistant neuropathic pain. Cannabis could be a cost-effective treatment for these conditions because it could reduce resource use.
The committee agreed that a follow-up period of 6 months is a realistic duration for assessing chronic pain treatments.
Table
Population: Adults with fibromyalgia or persistent treatment-resistant neuropathic pain being managed by a pain specialist using standard treatment Intervention: CBD (either as a pure product or containing traces of THC)
- 2.
In children and young people with intractable cancer-related pain and pain associated with specific diseases (such as epidermolysis bullosa), what is the clinical and cost effectiveness of cannabis-based medicinal products as an add-on treatment to improve symptoms in comparison to treatment with standard care?
There is currently no evidence that explores whether the addition of cannabis-based medicinal products as an adjunct to standard care improves symptoms for children and young people with intractable cancer-related pain and pain associated with specific diseases, such as epidermolysis bullosa. The reason for the lack of research so far is probably because there are relatively few children and young people with these conditions. In addition, there is concern regarding the use of high dose opioids for children and young people because it often causes adverse events. Therefore, a research recommendation was made. The committee defined ‘intractable cancer-related pain’ as cancer-related pain that does not respond to multiple drugs sufficiently to enable a reasonable quality of life and/or the child to be discharged home. The committee defined standard care as tertiary specialist pain/palliative management. An additional benefit from such research could be a reduction in resource use.
Appendix L. Reference list of included studies
- Blake, D. R., Robson, P., Ho, M. et al. (2006) Preliminary assessment of the efficacy, tolerability and safety of a cannabis-based medicine (Sativex) in the treatment of pain caused by rheumatoid arthritis. Rheumatology (Oxford, England) 45(1): 50–2 [PubMed: 16282192]
- de Vries, Marjan, Van Rijckevorsel, Dagmar C. M., Vissers, Kris C. P. et al. (2016) Single dose delta-9-tetrahydrocannabinol in chronic pancreatitis patients: analgesic efficacy, pharmacokinetics and tolerability. British journal of clinical pharmacology 81(3): 525–37 [PMC free article: PMC4767190] [PubMed: 26505163]
- de Vries, Marjan, van Rijckevorsel, Dagmar C. M., Vissers, Kris C. P. et al. (2017) Tetrahydrocannabinol Does Not Reduce Pain in Patients with Chronic Abdominal Pain in a Phase 2 Placebo-controlled Study. Clinical gastroenterology and hepatology: the official clinical practice journal of the American Gastroenterological Association 15(7): 1079–1086.e4 [PubMed: 27720917]
- Fallon, Marie T., Albert Lux, Eberhard, McQuade, Robert et al. (2017) Sativex oromucosal spray as adjunctive therapy in advanced cancer patients with chronic pain unalleviated by optimized opioid therapy: two double-blind, randomized, placebo-controlled phase 3 studies. British journal of pain 11(3): 119–133 [PMC free article: PMC5521351] [PubMed: 28785408]
- Johnson, Jeremy R., Burnell-Nugent, Mary, Lossignol, Dominique et al. (2010) Multicenter, double-blind, randomized, placebo-controlled, parallel-group study of the efficacy, safety, and tolerability of THC:CBD extract and THC extract in patients with intractable cancer-related pain. Journal of pain and symptom management 39(2): 167–79 [PubMed: 19896326]
- Langford, R. M., Mares, J., Novotna, A. et al. (2013) A double-blind, randomized, placebo-controlled, parallel-group study of THC/CBD oromucosal spray in combination with the existing treatment regimen, in the relief of central neuropathic pain in patients with multiple sclerosis. Journal of neurology 260(4): 984–97 [PubMed: 23180178]
- Lichtman, Aron H., Lux, Eberhard Albert, McQuade, Robert et al. (2018) Results of a Double-Blind, Randomized, Placebo-Controlled Study of Nabiximols Oromucosal Spray as an Adjunctive Therapy in Advanced Cancer Patients with Chronic Uncontrolled Pain. Journal of pain and symptom management 55(2): 179–188.e1 [PubMed: 28923526]
- Lynch, Mary E.; Cesar-Rittenberg, Paula; Hohmann, Andrea G. (2014) A double-blind, placebo-controlled, crossover pilot trial with extension using an oral mucosal cannabinoid extract for treatment of chemotherapy-induced neuropathic pain. Journal of pain and symptom management 47(1): 166–73 [PubMed: 23742737]
- Malik, Z., Bayman, L., Valestin, J. et al. (2017) Dronabinol increases pain threshold in patients with functional chest pain: a pilot double-blind placebo-controlled trial. Diseases of the esophagus : official journal of the International Society for Diseases of the Esophagus 30(2): 1–8 [PubMed: 26822791]
- Nurmikko, Turo J., Serpell, Mick G., Hoggart, Barbara et al. (2007) Sativex successfully treats neuropathic pain characterised by allodynia: a randomised, double-blind, placebo-controlled clinical trial. Pain 133(13): 210–20 [PubMed: 17997224]
- Portenoy, Russell K., Ganae-Motan, Elena Doina, Allende, Silvia et al. (2012) Nabiximols for opioid-treated cancer patients with poorly controlled chronic pain: a randomized, placebo-controlled, graded-dose trial. The journal of Pain: Official journal of the American Pain Society 13(5): 438–49 [PubMed: 22483680]
- Rog, David J., Nurmikko, Turo J., Friede, Tim et al. (2005) Randomized, controlled trial of cannabis-based medicine in central pain in multiple sclerosis. Neurology 65(6): 812–9 [PubMed: 16186518]
- Schimrigk, Sebastian, Marziniak, Martin, Neubauer, Christine et al. (2017) Dronabinol Is a Safe Long-Term Treatment Option for Neuropathic Pain Patients. European neurology 78(56): 320–329 [PMC free article: PMC5804828] [PubMed: 29073592]
- Serpell, M., Ratcliffe, S., Hovorka, J. et al. (2014) A double-blind, randomized, placebo-controlled, parallel group study of THC/CBD spray in peripheral neuropathic pain treatment. European journal of pain (London, England) 18(7): 999–1012 [PubMed: 24420962]
- Skrabek, Ryan Quinlan, Galimova, Lena, Ethans, Karen et al. (2008) Nabilone for the treatment of pain in fibromyalgia. The journal of Pain: official journal of the American Pain Society 9(2): 164–73 [PubMed: 17974490]
- Svendsen, Kristina B.; Jensen, Troels S.; Bach, Flemming W. (2004) Does the cannabinoid dronabinol reduce central pain in multiple sclerosis? Randomised double-blind placebo-controlled crossover trial. BMJ (Clinical research ed.) 329(7460): 253 [PMC free article: PMC498019] [PubMed: 15258006]
- van de Donk, Tine, Niesters, Marieke, Kowal, Mikael A. et al. (2019) An experimental randomized study on the analgesic effects of pharmaceutical-grade cannabis in chronic pain patients with fibromyalgia. Pain [PMC free article: PMC6430597] [PubMed: 30585986]
- Wade, Derick T., Makela, Petra, Robson, Philip et al. (2004) Do cannabis-based medicinal extracts have general or specific effects on symptoms in multiple sclerosis? A double-blind, randomized, placebo-controlled study on 160 patients. Multiple sclerosis (Houndmills, Basingstoke, England) 10(4): 434–41 [PubMed: 15327042]
- Weber, M.; Goldman, B.; Truniger, S. (2010) Tetrahydrocannabinol (THC) for cramps in amyotrophic lateral sclerosis: a randomised, double-blind crossover trial. Journal of neurology, neurosurgery, and psychiatry 81(10): 1135–40 [PubMed: 20498181]
- Wissel, Jorg, Haydn, Tanja, Muller, Jorg et al. (2006) Low dose treatment with the synthetic cannabinoid Nabilone significantly reduces spasticity-related pain: a double-blind placebo-controlled cross-over trial. Journal of neurology 253(10): 1337–41 [PubMed: 16988792]
- Dworkin, Robert H., et al. (2005) Core outcome measures for chronic pain clinical trials: IMMPACT recommendations.” Pain 113(1): 9–19. [PubMed: 15621359]
- Dworkin, Robert H., et al. (2009) Interpreting the clinical importance of group differences in chronic pain clinical trials: IMMPACT recommendations. Pain 146(3): 238–244. [PubMed: 19836888]
- Farrar, J et al (2001) Clinical importance of changes in chronic pain intensity measured on an 11-point numerical pain rating scale. Pain 94: 149–158 [PubMed: 11690728]
- Messina, S. et al., 2017. Sativex in resistant multiple sclerosis spasticity: Discontinuation study in a large population of Italian patients (SA.FE. study).. PloS one, 12(8), p. e0180651 [PMC free article: PMC5538735] [PubMed: 28763462]
- Hoggart et al. 2015. A multicentre, open-label, follow-on study to assess the long-term maintenance of effect, tolerance and safety of THC/CBD oromucosal spray in the management of neuropathic pain. Journal of Neurology 262(1) [PubMed: 25270679]
- Torrance et al 2006. The Epidemiology of Chronic Pain of Predominantly Neuropathic Origin. Results From a General Population Survey. The Journal of Pain 7(4) [PubMed: 16618472]
- Wang, T., Collet, J.-P., Shapiro, S. & Ware, M. A., 2008. Adverse effects of medical cannabinoids: a systematic review. CMAJ : Canadian Medical Association journal = journal de l’Association medicale canadienne, 6, 178(13), pp. 1669–1678. [PMC free article: PMC2413308] [PubMed: 18559804]
- Gu et al 2012. Estimating Preference-Based EQ-5D Health State Utilities or Item Responses from Neuropathic Pain Scores. Patient 5(3) [PubMed: 22765255]
Other references
References used solely in the economic model
Final
Evidence review underpinning recommendations 1.2.1 to 1.2.3 in the NICE guideline
These evidence reviews were developed by NICE Guideline Updates Team
Disclaimer: The recommendations in this guideline represent the view of NICE, arrived at after careful consideration of the evidence available. When exercising their judgement, professionals are expected to take this guideline fully into account, alongside the individual needs, preferences and values of their patients or service users. The recommendations in this guideline are not mandatory and the guideline does not override the responsibility of healthcare professionals to make decisions appropriate to the circumstances of the individual patient, in consultation with the patient and/or their carer or guardian.
Local commissioners and/or providers have a responsibility to enable the guideline to be applied when individual health professionals and their patients or service users wish to use it. They should do so in the context of local and national priorities for funding and developing services, and in light of their duties to have due regard to the need to eliminate unlawful discrimination, to advance equality of opportunity and to reduce health inequalities. Nothing in this guideline should be interpreted in a way that would be inconsistent with compliance with those duties.
NICE guidelines cover health and care in England. Decisions on how they apply in other UK countries are made by ministers in the Welsh Government, Scottish Government, and Northern Ireland Executive. All NICE guidance is subject to regular review and may be updated or withdrawn.
.&p=BOOKS&id=577083_ch2appif1.jpg "Click on image to zoom")
![Figure 3. Discontinuation from THC:CBD spray in responders and simulated non-responders [the placebo=1 group] (Messina 2017).](/core/lw/2.0/html/tileshop_pmc/tileshop_pmc_inline.html?title=Figure%203.%20Discontinuation%20from%20THC%3ACBD%20spray%20in%20responders%20and%20simulated%20non-responders%20%5Bthe%20placebo%3D1%20group%5D%20(Messina%202017).&p=BOOKS&id=577083_ch2appif3.jpg "Click on image to zoom")
.&p=BOOKS&id=577083_ch2appif7.jpg "Click on image to zoom")