Naloxegol for treating opioid‑induced constipation

The main clinical evidence for naloxegol came from the pivotal phase 3 trials KODIAC 4 (n=649) and KODIAC 5 (n=697). These were international, multicentre, randomised, double‑blind, placebo‑controlled trials comparing naloxegol with placebo in adults with non‑cancer pain and opioid‑induced constipation (OIC). Patients included in the trials had a stable maintenance opioid regimen for non‑cancer related pain for a minimum of 4 weeks, and reported less than 3 spontaneous bowel movements (SBM) per week in the 2 weeks before screening. In addition, patients reported at least 1 of the following symptoms: Bristol Stool Scale stool type 1 or 2; moderate severe or very severe straining; incomplete bowel movement (BM), in at least 25% of BMs recorded in the patient's electronic diary during the OIC confirmation period. The 2 trials excluded patients having opioids for cancer‑related pain.

In both trials, patients were randomised in a 1:1:1 ratio to either naloxegol 12.5 mg, naloxegol 25 mg or placebo once daily for 12 weeks. Patients were allowed to continue their baseline opioid pain control regimen with doses adjusted according to clinical need. They were also allowed to have bisacodyl rescue laxative if they had not had a bowel movement in 72 hours or more. The proportion of patients in the naloxegol 25 mg arm who used bisacodyl at least once was 54.7% (KODIAC 4) and 57.3% (KODIAC 5). In the placebo arm, these proportions were 72% and 70.7% respectively. No other laxatives were allowed in the trials.

Before the studies, the company defined several subgroups in terms of response to laxatives at baseline, using the baseline laxative response status questionnaire. The categories defined by the company were as follows:

• Laxative inadequate responder (LIR): people who were taking 1 or more laxative class for at least 4 days before screening and reported moderate, severe or very severe symptoms in at least 1 of the 4 stool symptom domains (that is, incomplete BMs, hard stools, straining or false alarms). Around half of the clinical trial populations (54.6% in KODIAC 4 and 53.2% in KODIAC 5) were classified as laxative inadequate responders. This is the group covered by naloxegol's marketing authorisation.

• Laxative adequate responder (LAR): people whose constipation responded adequately to laxatives taken at least 4 days before screening and who reported mild or no symptoms.

• Laxative unknown responder (LUR): people who had not had laxatives in the last 2 weeks or had taken laxatives for less than 4 days in the last 2 weeks.

An additional subgroup was defined as the 2xLIR population. These were people who met the criteria for LIR but had at least 2 laxatives classes, or reported unsatisfactory relief from 1 or more additional laxative class taken during the 6 months before screening.

The company also conducted a post‑hoc analysis of the LIR+step‑3 opioids subgroup, comprising patients in the LIR population who had step‑3 opioids (classified according to the World Health Organisation analgesic ladder). The company stated that this is a clinically valid subgroup of patients with OIC, because the more severe forms are more likely to be related to the use of step‑3 opioids.

The primary outcome of the KODIAC 4 and 5 studies was response to treatment, defined as the proportion of patients with 3 or more SBMs per week, with improvement from baseline of 1 or more SBM per week for at least 9 of 12 weeks and 3 of the last 4 weeks of the study. SBM was defined as a bowel movement without using laxatives in the last 24 hours). The company stated that SBM frequency is a clinically meaningful measure commonly employed in clinical research to assess the efficacy of a treatment for chronic constipation.

The main secondary outcomes included:

• response to treatment (as defined for the primary outcome) in the LIR population only

• time to first post‑dose SBM without the use of rescue medication in the last 24 hours

• mean number of days per week with at least 1 SBM.

In both KODIAC trials, treatment with naloxegol 25 mg (the recommended dose for all patients except those with renal insufficiency) resulted in significantly higher response rates than placebo in both the overall population (KODIAC 4: 44.4% compared with 29.4%, p=0.001; KODIAC 5: 29.3% compared with 39.7%, p=0.021) and the LIR population (KODIAC 4: 48.7% compared with 28.8%, p=0.002; KODIAC 5: 46.8% compared with 31.4%, p=0.014). In both studies, naloxegol showed consistent improvements in a range of secondary end points, including time to first post‑dose SBM, total SBMs per week, number of days per week with at least 1 SBM and use of rescue medication at least once over the treatment period. The 3 instruments used by the company to measure quality of life (PAC‑SYM, PAC‑QoL and EQ‑5D) also showed advantages with naloxegol compared with placebo.

There were no differences in adverse events between the overall and LIR populations. The most frequently reported adverse events were gastrointestinal in nature (predominantly diarrhoea, abdominal pain, nausea and flatulence); this is to be expected, considering the nature of OIC and naloxegol's pharmacological mechanism of action. Gastrointestinal adverse events were more frequent in the naloxegol 25 mg arms compared with naloxegol 12.5 mg and placebo. There were no notable differences in type or frequency of serious adverse events across the treatment arms of the studies. The incidence of discontinuations because of adverse events was dose‑related, with a higher proportion of patients discontinuing in the naloxegol 25 mg arm compared with those having naloxegol 12.5 mg and placebo. The discontinuation rate with the longer‑term use of naloxegol (52 weeks, as observed in KODIAC 8) was similar to that seen in the 12‑week studies, KODIAC 4 and 5.

The ERG commented that some studies that were potentially relevant may have been omitted from the mixed treatment comparison because of how the company defined and adhered to the criteria in its literature search. For example, the difference in population specification between the scope and the company's submission (that is, limiting the naloxegol studies in the submission to the LIR subgroup) is likely to have reduced the number of studies included.

The ERG stated that insufficient details were presented for comparator study design, quality and data. It stated that the definition of rescue treatment varied between trials and there was not enough information to judge the similarity of the rescue treatments used. It also noted that the company did not present details of the baseline characteristics for the comparator studies, thereby preventing any further assessment of their similarities. The ERG felt that these limitations prevented further analyses based on baseline characteristics (for pain intensity, opioid dose, duration of opioid use, duration of OIC and previous laxative use).

The ERG questioned the reliability of the mixed treatment comparison because it compared the LIR population only from KODIAC 4 and 5 with the overall population from the comparator studies. The ERG conducted its own exploratory analyses using the overall populations from all trials, including KODIAC 4 and 5. The results of these analyses were similar to the company's results, which the ERG felt suggested there were unlikely to be any major differences between the LIR and intention‑to‑treat populations. However, it questioned whether combining the 2 populations in a mixed treatment comparison as the company had, could be clinically justified. The ERG stated that overall there was no robust evidence to distinguish the relative efficacy and safety between naloxegol and the comparators of interest.

The company constructed a decision‑analytic model comparing the cost effectiveness of naloxegol 25 mg with several comparators for OIC that has responded inadequately to laxatives (that is, the population covered by the marketing authorisation). The comparator for the base‑case analysis was placebo (as in the pivotal clinical trials KODIAC 4 and 5), whereas methylnaltrexone was used in a scenario analysis. The company also presented an analysis of the LIR+step‑3 opioids subgroup, in which the comparators were placebo, methylnaltrexone and naloxone‑oxycodone.

The company presented 2 additional analyses which it considered to be the most clinically relevant comparisons (based on clinical guidance from an advisory board and company‑sponsored research):

• Naloxegol compared with placebo plus bisacodyl (where bisacodyl was a proxy for using stimulant laxatives as needed).

• Naloxegol plus bisacodyl compared with placebo plus bisacodyl (to demonstrate the cost effectiveness of naloxegol when used with a stimulant laxative).

The company's economic model comprised a decision‑tree structure for the first 4 weeks of treatment, followed by a Markov structure. All patients entered the model with OIC and were treated with naloxegol or a comparator. Response to treatment was assessed after 4 weeks, with people being classified as responders if they achieved constipation relief and as non‑responders if they did not.

The Markov model consisted of 4 health states: opioid‑induced constipation (OIC), non‑OIC (on treatment), non‑OIC (untreated) and death. OIC and non‑OIC were defined as follows:

• OIC: less than 3 SBMs per week in at least 2 of the last 4 weeks.

• Non‑OIC: 3 or more SBMs per week in at least 3 of the last 4 weeks.
  
  The company justified this change from the clinical definition of OIC because it corresponds with internationally accepted definitions of constipation (although these were not specified) and helped to simplify the model by allowing utility and resource to be estimated as a function of constipation status, rather than a change in that status. People whose constipation responded to treatment after 4 weeks entered the Markov phase of the model in the non‑OIC (on treatment) health state, whereas those who did not respond to treatment by week 4 entered in the OIC health state. People may move between the OIC and non‑OIC state in the model even in the absence of effective treatment. The company stated that clinical expert opinion and trial data suggested that patients in the placebo arm moved between the OIC and non‑OIC states.

The time horizon in the company's base case was 5 years. The company felt this reflected the upper end of a period of opioid use and the model suggested that it reaches a steady state within that period. Scenario analyses were done using several other time horizons to test the effect on cost effectiveness. The cycle length was 4 weeks and a half‑cycle correction was applied. The company applied a discount of 3.5% for costs and benefits and adopted an NHS/personal social services perspective.

The proportion of people in the non‑OIC (on treatment) state of the model was estimated based on response rates derived from the KODIAC 4 and 5 trials. For the comparisons with methylnaltrexone and naloxone‑oxycodone, the outcomes of the mixed treatment comparison were used. To estimate transitions from the non‑OIC (on treatment) state to the OIC state, the company conducted parametric survival analyses based on data from the KODIAC trials; this informed the prediction of how many patients remained in the non‑OIC (treated) health state for 5 years. The exponential distribution was used for both arms in the base case. For the comparisons with methylnaltrexone and naloxone‑oxycodone, the curves were estimated based on the naloxegol curve, assuming proportional hazards and using the hazard ratios estimated from the mixed treatment comparison. The company also did a scenario analysis in which the hazard ratio for methylnaltrexone compared with naloxegol was set to 1 (assuming equal treatment effect for both treatments).

The company's estimates for transitions from the OIC health state to the non‑OIC (untreated) health state and from the non‑OIC (untreated) health state to the OIC health state were generated from an analysis of the LIR population in the placebo arms of KODIAC 4 and 5. The company analysed the placebo data because the model assumed that patients are not on treatment in the OIC and non‑OIC (untreated) health states. This same transition estimate was assumed for all comparators in the model.

For the transition to death, the company applied the same mortality rate to all health states (based on the UK general population). Mortality was calculated based on the UK life table for 2008 to 2010. The company used the exponential function to calculate cycle probability of mortality.

Utility estimates in the economic model were derived from an analysis of KODIAC 4 and 5 EQ‑5D data, collected at 0, 4 and 12 weeks. These data were used with the Dolan algorithm to derive utility scores. Based on the results of a regression analysis, the company applied time‑ and treatment‑specific utilities in the base case for the comparison with placebo. The company also presented scenario analyses applying treatment‑specific utilities only and health state‑specific utilities only. Only health state‑specific utilities were used for the comparisons with methylnaltrexone and naloxone‑oxycodone. No direct estimates of the effect of adverse events on utility were included in the model. The company stated that its clinicians advised that adverse events were unlikely to make a difference to health‑related quality of life.

Costs incorporated in the company's model included drug costs, administration costs for methylnaltrexone, laxative costs, adverse events costs, opioid costs for the naloxegol arm of the comparison with naloxone‑oxycodone, and other costs for managing constipation (including inpatient care, outpatient care, emergency care, GP visit and consultation, nurse visits, rescue therapy and medical tests). The company assumed that patients having naloxone‑oxycodone did not incur additional opioid costs (because of the presence of oxycodone). Therefore, the company presented 2 separate scenarios applying opioid costs to the naloxegol arm of the comparison with naloxone‑oxycodone. The first scenario used morphine (the most commonly prescribed step‑3 opioid) and the second used oxycodone. The company's estimates of resource use associated with managing constipation and adverse events were based on a survey of clinicians. Costs were based on the BNF, NHS reference costs and the Payment by Results tariff.

In the company's base‑case analysis (based on KODIAC 4 and 5 data), the incremental cost‑effectiveness ratio (ICER) for naloxegol compared with placebo was £10,849 per quality‑adjusted life year (QALY) gained. However, the company stated that the most clinically relevant comparisons are as follows:

• Naloxegol compared with placebo plus bisacodyl (ICER of £12,639 per QALY gained).

• Naloxegol plus bisacodyl compared with placebo plus bisacodyl (ICER of £11,175 per QALY gained).

The company conducted a number of 1‑way sensitivity analyses to demonstrate the model's robustness to changes in parameters and assumptions. In nearly all of the sensitivity analyses, naloxegol produced an ICER of less than £20,000 per QALY gained. The company also conducted several scenario analyses including: using alternative utility input assumptions; using the burden of illness data for resource use costs of managing constipation; assuming no extrapolation beyond the trial period (that is, using a time horizon of 12 weeks); and using alternative functions to estimate the transitions in the model. Of these, only 2 scenarios produced an ICER for naloxegol that was over £20,000 per QALY gained. Using a 12‑week time horizon resulted in ICERs of £20,020 per QALY gained for naloxegol compared with placebo and £33,708 per QALY gained for naloxegol compared with placebo plus bisacodyl. When a health state‑specific utility input was employed, the ICER for naloxegol compared with placebo increased to £38,921 per QALY gained, and the ICER for naloxegol compared with placebo plus bisacodyl increased to £63,423 per QALY gained. The company did not present results of these 2 scenario analyses for naloxegol plus bisacodyl compared with placebo plus bisacodyl.

For the analysis comparing naloxegol with subcutaneous methylnaltrexone every other day, the results showed that naloxegol dominated methylnaltrexone (that is, naloxegol was more effective and cost less than methylnaltrexone).

For the analysis of the LIR+step‑3 opioids subgroup, naloxegol dominated methylnaltrexone and naloxone‑oxycodone (when oral morphine was added to the naloxegol arm). For the different placebo comparisons (with and without bisacodyl), the ICERs were less than £7,000 per QALY gained. The ICER increased to £34,054 per QALY gained for naloxegol compared with naloxone‑oxycodone when oxycodone was added to the naloxegol arm.

The company did not do a subgroup analysis on patients with cancer who have OIC. However, it stated that there is no scientific rationale to expect the pharmacodynamic properties of naloxegol to differ in this patient population, because the underlying physiology of pain is the same regardless of the underlying cause. The company stated that pain medications act on the same target receptors regardless of whether the trigger for the pain is cancer or non‑cancer. Because of this, extrapolating the available data to the treatment of OIC in patients with cancer pain was justified. The company also stated that patients with cancer and OIC would fit in the acceptable range for the key model variables which resulted in ICERs of less than £20,000 per QALY gained for naloxegol compared with each comparator.

The ERG indicated that the company's model was generally well presented and reported. It noted that the model structure was based on a revised definition of response (3 or more SBMs per week in at least 3 out of the last 4 weeks) compared with that used in the clinical studies (which also included an improvement of 1 or more SBM per week for 9 weeks out of 12). Because of this, the model was able to use absolute health states rather than health states relative to a baseline. The ERG considered the 5‑year time horizon to be acceptable.

The ERG noted that the modelled population was based on the marketing authorisation for naloxegol (that is, people who constipation has had an inadequate response to laxatives). However, it questioned whether the trial definition of inadequate response to laxatives (see section 3.3) reflected clinical practice. It argued that although the effectiveness of some types of laxative can be reasonably assessed after 4 days (for example bisacodyl), others – such as lactulose – may need to be used for slightly longer before their effectiveness can be fully assessed.

The ERG stated that the main weakness of the cost‑effectiveness analysis was the definition of intervention and comparator. It noted that the cost‑effectiveness analysis compared naloxegol with placebo based on SBM in the base case, and naloxegol plus bisacodyl with placebo plus bisacodyl based on bowel movement in a scenario analysis. However, the ERG considered the base‑case analysis (that is, naloxegol without bisacodyl) to be neither clinically relevant nor consistent with the KODIAC 4 and 5 trials in which rescue medication was permitted in all arms. The ERG argued that the use of naloxegol without rescue medication in clinical practice is implausible considering that the rates of SBM in the trial may have been affected by the use of rescue medication.

The ERG stated that EQ‑5D data were available from the clinical studies to inform the utilities used in the model, thus providing good quality evidence for the cost‑effectiveness analysis. However, the ERG stated that it would have been preferable for the company to use health state‑specific utilities in its base case for the comparison of naloxegol with placebo rather than treatment‑specific utilities because there was insufficient evidence to suggest an independent treatment effect of naloxegol on health‑related quality of life. The ERG also commented with regards to the health‑related quality of life analysis that the non‑OIC health state is too broad to be homogeneous. Using the company's definition of response, any patient with at least 9 SBMs over a 28‑day period would be classified as a responder (and so move to the non‑OIC on treatment state). However, patients with 28 or more SBMs over this period would be considered to be in the same health state and have the same quality of life as patients who had only 9 SBMs, which the ERG considered to be unlikely. The ERG stated that the model should have included more discrete health states which were more reflective of patient experience and this would have allowed the company to apply health‑state specific utilities.

The ERG performed an exploratory base‑case analysis comparing naloxegol plus bisacodyl with placebo plus bisacodyl. The outcome was SBM, rather than bowel movement as the company had used in its own analysis. The ERG commented that this was the only accurate comparison that could be made using the data from KODIAC 4 and 5. The analysis increased the ICER for naloxegol compared with placebo to £10,864 per QALY gained.

The ERG conducted an exploratory sensitivity analysis on response rate as a proxy for the 2xLIR population (see section 3.4). The ERG extracted the response rates at 4 weeks for the LIR and 2xLIR populations from KODIAC 4 and 5. Using these data, it calculated a pooled response rate for the 2xLIR population using an adjusted response rate as a proxy for that population. The ERG noted that there was a marked difference in response rates in the naloxegol arm compared with the placebo arm between the LIR and 2xLIR populations. The response rate for the LIR subgroup was 20.4% better in the naloxegol arm than the placebo arm, and the response rate in the 2xLIR population was 13.5 % better in the naloxegol arm than in the placebo arm. The ERG assumed that all other input parameters would be the same as the company's base case, because there were no 2xLIR data to inform the various transition probabilities in the model. This analysis increased the ICER for naloxegol compared with placebo from £10,849 per QALY gained (in the company's base case) to £11,406 per QALY gained. The ERG also conducted a sensitivity analysis exploring the impact of changing the parametric distribution of the time‑to‑event curve used to estimate the transition probability from non‑OIC (on treatment) to opioid‑induced constipation. The ERG noted that in its analyses, the company used the same parametric function for both placebo and naloxegol. Instead, the ERG used different combinations of the following distributions: exponential, Weibull, lognormal and loglogistic. The ERG compared naloxegol plus bisacodyl with placebo plus bisacodyl and most cases, the ICER was similar to or lower than the company's base case ICER. The ICERs only increased beyond £13,000 per QALY gained when the exponential distribution was used for naloxegol and either the lognormal or loglogistic distribution was used for placebo.

The ERG conducted threshold analyses on the hazard ratio for the transition from the non‑OIC (on treatment) state to the OIC state for methylnaltrexone and naloxone‑oxycodone. In the company's model, the hazard ratios for this transition were calculated using the ratio between the 4‑week response rates for methylnaltrexone and naloxone‑oxycodone and the response rate of naloxegol 25 mg. None of the additional economic analysis by the ERG resulted in ICERs that differed from the company's results in any meaningful way. For this reason the ERG considered that the company's cost‑effectiveness results were generally robust. However, given that the company did not perform a full mixed treatment comparison, the ERG cautioned that the results of the comparisons with methylnaltrexone and naloxone‑oxycodone should be interpreted with care.

Full details of all the evidence are in the committee papers.