Results

Participant recruitment

Figures 4 and 5 detail the patient flow through the study by treatment (i.e. A-P, D-P and P-A) and by chronological treatment pathway (first, second and third). Participating sites collected pre-screening data on all patients who the site had been in contact with regarding the study. Between 2 October 2017 and 31 July 2019, 1004 patients were identified as being potentially eligible for the trial across 18 secondary care hospital centres (a further three centres were activated to recruitment, but did not identify any potential participants). Of these patients, 426 were ineligible. The main reasons for ineligibility at this stage were not having neuropathic pain (n = 202), contraindications for study medications (n = 42), having other painful medical conditions (n = 39) and use of prohibited concomitant treatment (n = 28). A further 314 patients were not interested or unable to continue into the study, with the main reasons being that they were unable or unwilling to attend all study visits (n = 146) or were not wanting to come off current treatment (n = 76). Twelve patients did not continue for other reasons.

FIGURE 4 Patient disposition and study flow chart. a, Switch to second-line monotherapy before week 6; and b, includes switches to second-line monotherapy after week 6. (Figure not reproduced here; see the original article.)

FIGURE 5 Patient disposition and study flow chart by chronological treatment pathway. (Figure not reproduced here; see the original article.)

Between 2 October 2017 and 31 July 2019, 252 patients attended the consent visit at week –2 across 17 trial centres, and 222 of these patients provided informed consent. Of those patients who consented, 40 discontinued from the trial before randomisation and a further 42 were ineligible because of unconfirmed or low levels of neuropathic pain (n = 37) or clinically significant arrhythmia (n = 5). One hundred and forty patients proceeded to be randomised at week 0 across 13 sites. Ten patients were excluded following randomisation and within the first treatment pathway. Six patients provided no post-baseline data (three withdrew citing the trial burden, two were lost to follow-up and one developed a significant comorbidity) and four were randomised in error.

A total of 53 patients withdrew from the study before completing all three treatment pathways. The majority (n = 33) of withdrawals came in the first pathway, with a further 13 withdrawals in pathway 2 and seven withdrawals in pathway 3. The numbers of patients starting first, second and third pathways were 130, 97 and 84, respectively, and the numbers of patients contributing 7-day pain scores at week 16 were 105, 85 and 74, respectively.

Recruitment was originally expected to be completed within 12 months by eight trial centres. Owing to slow recruitment rates, the total number of centres was increased to 21 and a number of changes were made to the eligibility criteria and study processes to improve recruitment (see Chapter 2, Changes to the protocol). In addition, assistance was provided to participating centres via regular teleconferences, recruitment packs and one-to-one discussions with the research fellow based at the lead site in Sheffield.

Protocol non-compliances

A total of 146 (major, n = 73; minor, n = 73) protocol non-compliances were reported during the trial and no serious breaches were reported. Five participants were ineligible and were withdrawn from the trial during the first treatment pathway. One participant did not contribute primary outcome data as a result of a good clinical practice non-compliance in the recording of the data at site.

As the study medication was provided in up to eight blinded medication bottles, the dosing schedule for participants was more complicated than in usual care. Thirty-seven cases of participant non-compliance with treatment were reported. These included the participant taking more/less medication than prescribed or taking medication from incorrect bottles.

Fifty-nine non-compliances were reported in relation to trial treatment or procedural issues and this included 22 cases where a patient was prescribed and/or dispensed an incorrect dose of medication. To our knowledge, the errors in treatment doses did not result in any AEs.

Three non-compliances were reported in relation to on-site visits being missed due to COVID-19 restrictions.

Characteristics of trial participants

Trial participants were of similar age to screened patients (median 61.8 years vs. 61.2 years), but fewer females were enrolled (26% vs. 41%).

The full characteristics of the 130 trial participants are presented in Table 5, split according to whether or not the patient completed the three treatment pathways. The majority (82%) of patients had type 2 diabetes, had experienced neuropathic pain for an average of 5 years (median duration 3.4 years, range 4 months to 25 years) and self-rated their pain at 6.6 points out of 10 points (NRS) and 6.1 points out of 10 points (BPI-MSF). Patients more commonly described their pain as relapsing/remitting as opposed to deep or involved (NPSI) and one-third of participants reported having taken each of amitriptyline, pregabalin, duloxetine and gabapentin at some point prior to trial entry.

Trial medication usage

The study treatment use is summarised in Table 6. The uptake of first-line therapy was similar for each arm, with around half of patients remaining on the highest dose at the end of each pathway. Likewise, discontinuation overall was similar across the three arms, but more patients discontinued duloxetine because of AEs or toxicity. The proportion of patients starting second-line therapy was also similar, as were the proportions on the highest dose level at week 16. More participants discontinued or switched from duloxetine during first-line therapy and more participants discontinued intervention during the P-A pathway, but neither was statistically significant.

Numeric Rating Scale pain

None of the pairwise comparisons was statistically significant. At week 16, among participants with outcome data (i.e. complete-case analysis), mean pain in the A-P arm was 0.1 units lower (i.e. better) than in the D-P and P-A arms, but the prespecified clinically important difference of 0.5 units was either excluded from or on the edge of each pairwise CI. For D-P compared with A-P the mean difference was –0.1 (98.3% CI –0.5 to 0.3), for P-A compared with A-P the mean difference was –0.1 (98.3% CI –0.5 to 0.3) and for P-A compared with D-P the mean difference was 0.0 (98.3% CI –0.4 to 0.4) (Table 7). These findings were robust across a range of analyses that assessed missing data under plausible scenarios (Figures 6 and 7), in which the 47 (15%) instances of missing data were imputed by last observation carried forward, multiple imputation or controlled multiple imputation. In all cases, the point effects were within ± 0.1 points.

FIGURE 6 Pairwise comparisons of NRS pain score at week 6. MI, multiple imputation. LOCF, Last observation carried forward. (Figure not reproduced here; see the original article.)

FIGURE 7 Sensitivity analysis of NRS pain score at week 16. MI, multiple imputation. LOCF, Last observation carried forward. (Figure not reproduced here; see the original article.)

The differences at week 6 were larger among patients in the P-A arm, having greater average pain than patients in the D-P and A-P arms; however, again, none of these differences was statistically significant. For the complete-case scenario, the difference between P-A and A-P was 0.3 (98.3% CI –0.1 to 0.8; p = 0.0492), with similar findings in analyses of imputed data, reflecting the similar levels of ‘for cause’ missing data in the two arms. Nevertheless, none of the arms differed on average by more than 0.4 points in any scenario and none was statistically significant at the predefined alpha = 0.0167 level. In total, there were 12 (4%) instances of missing data, meaning that the impact of missing data was small.

Response to treatment

Treatment response is summarised in Table 8. As noted above, duloxetine was the least well-tolerated monotherapy and had the lowest response rate (27% vs. 30% for amitriptyline and 34% for pregabalin) at week 6, although, again, this comparison was not statistically significant. The overall response rates at week 16 were 46% (A-P), 43% (D-P) and 46% (P-A).

Figure 8 shows the trajectory of average self-reported pain by randomised sequence, ranging over the 52 weeks of study (pain scores were not routinely completed for the washout periods and have not been included). Self-rated pain did not return to pre-randomisation levels in the washout periods between pathways 1 and 2 or between pathways 2 and 3. This tempers the conclusions that can be made for the change from baseline outcomes (i.e. 30% or 50% reductions), but the inclusion of period number in the statistical model meant that between-arm comparisons remained valid.

FIGURE 8 Self-reported NRS pain by time in study. (Figure not reproduced here; see the original article.)

Figure 9 shows the trajectory of average NRS score by time within each pathway. The three curves largely overlap throughout the 16 weeks, suggesting that the arms have equivalent efficacy both in the monotherapy phase (i.e. weeks 0–6) and thereafter. Pain scores continued to decrease after week 6 when combination therapy was offered to participants (mean score: week 16 3.3 vs. week 6 3.9), although qualitatively the scores appeared to plateau around week 12.

FIGURE 9 Self-reported NRS pain by treatment pathway. (Figure not reproduced here; see the original article.)

Figure 10 shows the same data as the previous figure, but separately for patients remaining on monotherapy or starting combination therapy on or after week 6. By definition, patients starting combination therapy had higher NRS scores (i.e. worse pain control) at week 6, with average pain levels appearing to plateau in the weeks prior to this. Thereafter, the curves began to converge, with a further mean drop of 1 NRS point among patients on combination therapy (again plateauing over the last weeks of the pathway) compared with 0.2 points among patients on monotherapy (Table 9). This suggests a benefit of combination therapy among patients who had less response to monotherapy alone.

FIGURE 10 Self-reported NRS pain by use of combination therapy and treatment pathway. (Figure not reproduced here; see the original article.)

Subgroup analyses

Further exploratory work looked at whether or not there were subgroups of patients who responded better to some therapies. Figure 11 depicts patient response to monotherapy at week 6 via a Venn diagram. Although most patients responded either to none (n = 37) or all (n = 13) of the therapies, 30 of the 80 patients who provided data had different responses to the treatments.

FIGURE 11 Response to monotherapy among patients starting all three pathways. Response defined as a NRS score of ≤ 3 points and no informative treatment withdrawal. Four patients provided no data to at least one pathway. (Figure not reproduced here; see the original article.)

The results of the subgroup analyses are presented in Appendix 4, Figure 16. Pain was, on average, higher among younger (< 60 years) patients than among the older (≥ 60 years) patients, but the three pathways showed similar efficacy within each of the two subgroups (see Figure 16a). A similar theme was apparent for baseline pain score, with week 16 pain being higher among patients with more severe pain (i.e. a NRS score of ≥ 7 points) but no interaction with treatment (see Figure 16b).

The average week 16 NRS scores were similar across the three pain phenotypes (see Figure 16c). There were, however, some interactions among the individual NPSI components, with a suggestion that D-P was less suited to participants with higher NPSI scores.

There was some evidence of an interaction between treatment and baseline mood. The A-P pathway was notably worse at 16 weeks for patients with high levels of anxiety or depression. Patients with moderate anxiety or depression (scores below 15) had similar responses regardless of treatment arm (see Figure 16j and k).

Pain was also investigated in relation to whether or not the patient had previously used the three study drugs (i.e. amitriptyline, duloxetine and pregabalin) or any opioids (see Figure 16l–o). For all four drug options, patients who had previously been prescribed drugs reported higher pain scores than patients who had not, most notably opioids. Nevertheless, the difference between treatment arms appeared unaffected by previous medication use.

Finally, pain responses during the period of the UK lockdown due to COVID-19 (defined as on/after 20 March 2020) were similar to those measured prior to lockdown, meaning that the treatment comparisons were similar both within lockdown and outside lockdown (see Figure 16p and q). There was no apparent change in weekly NRS scores following imposition of lockdown measures, or any other temporal changes.

Tolerability

Tolerability (rated on a 0–10 NRS, with 10 being least tolerable) was similar between arms at weeks 6 and 16. Scores at weeks 6 and 16 were also similar, suggesting no perceived worsening in side effects with the introduction of combination therapy. In fact, self-reported tolerability was slightly better at week 16 than in week 6, albeit not statistically significantly so (Table 10).

Quality of life

General health (assessed by SF-36), health utility (assessed by EQ-5D-5L) and anxiety/depression (assessed by HADS) are shown in Table 11. Self-reported QoL was similar across the three arms and no consistent patterns were observed among the QoL inventories. Some of the comparisons were statistically significant. After 6 weeks of treatment, duloxetine was associated with worse limitations due to physical health compared with amitriptyline and pregabalin, but was preferable with regard to general health, as assessed via SF-36 (Figure 12) and EQ-5D-5L. All differences were modest in clinical or psychometric terms. No statistically or clinically significant differences were found at 16 weeks.

FIGURE 12 Short Form questionnaire-36 items domains. Graphs depict the model-based mean and CI for each treatment arm for each of the nine domains at week 16. Higher scores indicate better QoL. (Figure not reproduced here; see the original article.)

Other pain inventories and insomnia

As with the NRS, the NPSI and BPI-MSF were largely similar across arms at 6 and 16 weeks (Table 12). In general, the D-P arm had the highest pain scores at week 16, with the P-A arm having the lowest, but none of the comparisons was statistically significant. By contrast, week 6 pain scores tended to be higher in the P-A arm, with two BPI-MSF scores (worst pain over 24 hours and overall pain severity) being significantly lower in the A-P treatment pathway.

Self-reported insomnia (via the ISI) is summarised in Table 13. Patients rated their insomnia worse in the D-P arm than in the A-P arm, both at week 6 (mean difference = 1.5, 98.3% CI 0.0 to 3.1; p = 0.016) and at week 16 (mean difference = 1.5, 98.3% CI 0.1 to 3.0; p = 0.010).

Patient impression of change and treatment preference

Finally, participants were asked to complete a global impression of change (using the PGIC) at the end of each pathway and their preferred treatment if they had completed the three pathways (Table 14). The treatments were rated positively among this group, with 44%, 43% and 49% of patients reporting ‘much improved’ or ‘very much improved’ for A-P, D-P and P-A, respectively (Kruskal–Wallis test p = 0.70).

On reaching the end of the study, participants were asked to rate their preferred treatment of the study. The most popular choice was P-A (43%), followed by D-P (33%) and A-P (24%), although this comparison was not statistically significant (p = 0.266).

Adverse events

Adverse events are summarised by pathway in Table 15. Fatigue was the most commonly reported AE and was similarly prevalent in each pathway. The most notable difference was in the incidence of dry mouth, which was associated with amitriptyline (A-P, 32%; P-A, 17%; D-P, 8%) (p < 0.001). Dizziness was more common in patients in the P-A arm (24%) than in patients in the A-P (12%) and D-P (16%) arms, and nausea was highest (23%) in the D-P arm (vs. 5% in the A-P arm and 7% in the P-A arm; p = 0.001).

Table 16 shows AEs further split by the time they occurred (i.e. before or after 6 weeks) and SAEs are tabulated in Table 17. One SUSAR occurred [one participant experienced an atrioventricular heart block (Mobitz II) in the P-A arm 11 weeks into the first treatment pathway and while on amitriptyline, having switched from pregabalin at the week 6 visit].

The remaining tables (Tables 18–20) focus on AEs of moderate or severe intensity that are related to one or both study medications. In total, 97 such events occurred, affecting 19% of patients in the A-P arm, 22% of patients in the D-P arm and 21% of patients in the P-A arm (p = 0.719) (see Table 18). A post hoc analysis looked at days affected by moderate or severe related AEs (see Table 19) while in study follow-up, which for the majority of patients was zero but for the groups as a whole resulted in 84, 69 and 73 days per 1000 days of follow-up in the A-P, D-P and P-A arms, respectively (p = 0.858). Most events started during monotherapy but continued to affect patients beyond week 6, by which point combination therapy may have started. The mean days affected for each category is displayed visually in Figure 13 and in Table 20. The AEs with the longest average person-day impacts were dry mouth in the A-P arm (191 days/1000 days of follow-up) and fatigue in all arms (109 days/1000 days of follow-up).

FIGURE 13 Days impacted by adverse effects. (Figure not reproduced here; see the original article.)